Upload
ngohanh
View
222
Download
0
Embed Size (px)
Citation preview
Hiatus Resolving In Chitumbuka
Lee Yong-Sung lt
1 Introduction
Chitumbuka has a very simple syllabic structure as shown in (1)
(1) Chitumbuka syllable structure
a
~ (C) (C) v I J I
A vowel can be preceded by maximum two consonants and if there are two consonants in
the same syllable the second one must be a glide Notice also that all the syllables are V-final
There is no coda element in this language Since all the syllables end with vowels and consonants
optionally appear before vowel we may expect to see that there may be some cases where two
vowels can get together However general observations about the Chitumbuka pronunciation seem
to reveal different characteristic of its sound system
(2) Each vowel should be separated by one or more consonants
This is not because there is no example of two consecutive vowels in Chitumbuka
accidentally Consider the types of Chitumbuka morpheme structures
(3) a All the prefixes are V-final
-325shy
b All the suffixes are V-initial c Some of the stems are V-initial andor V-final
We can immediately see that vowels can come together through prefixation or suffixation to
stems Given the two observations (2) and (3) we are expect to see that there may be certain
phonological precesses in Chitumbuka that take two consecutive adjacent vowels as inputs and
produce outputs that confirm to the observation in (2)
I will call them hiatus resolving processes In this paper different strategies of hiatus resolving in Chitumbuka will be discussed Generally three different operations are noticeable
(4) a The first of two consecutive vowels becomes a glide b The first vowel is lost or deleted c A glide appears between the two vowels
In section 2 data of vowel clashes observed in various morpheme concatenation processes
will be presented and strategies of resolving hiatus will be discussed employing linear segmental
approach It will be seen that there are some problems in this analysis and the nature of the
problems will also be discussed
Section 3 presents another analysis of the same data from a different angle using moraic
syllable structure It will also be shown that the syllable-based approach can provide solution to
some of the problems posed in section 2
The data used in this paper have two different sources one from the transcriptions that I got
with the help of the native Chitumbuka speaker Mr Wongani Katundu And some of them are
from Tumbuka-Tonga- English dictionary I got the necessary words and checked them with the
informant and screen them to use only those which are in his vocabulary
-326shy
3 Hiatus Resolving in Chitumbuka
2 Vowel clashes
In Chitumbuka the sequence of two vowels in a word is not allowed as shown in (5)
(5) a mu (AGR) - ova (beer) --gt muova b ka (past) - imba (sing) --gt kaimba
As a first approximation we will just assume that there is principled prohibition of vowel
clashes which can be expressible as in (6)
(6) Well-formedness condition No two vowels can be adjacent without intervening consonant(s)
Though there may be problems with this condition this may suffice to begin the discussion
And more constrained condition will be introduced in 31 In spite of this condition we can
immediately see that through morpheme concatenation or even in the base form or underlying
representation vowels can come together Three different types of changes are witnessed in
Chitumbuka to conform to (6) First we can see the glide formation (henceforth GF) as in (7)
(7) a ku (V-marker) -Ii (eat) - eNge (future) --gt kulyeNge (will eat immediately) b mu (AGR) - ana (female) --gtmwana (a girl)
As shown in (7) the first of the two consequent vowels becomes a glide And since the
glide is not a vowel the hiatus is resolved
Secondly we can see the case of vowel deletion (henceforth VD) operation as in (8)
(8) a ka (past) - imb (sing) - a (FV) --gt kimba (to sing) b ku (V-marker) - och (burn) - a (FV) --gt kocha (to burn)
Finally a glide may be inserted (GI glide insertion) between two vowels as in (9)
1 )IlN the capitaln in the transcription represents the velar nasal [Q ] 2)FV means final vowel The morphological meaning of this affix is not fully investigated
In general the -a means indicative mood since it can be used with the past tense marker Ika-I or la-I and the future-tense marker Izamu-I but not compatible with subjunctive mood ending I-el or next-time point (future) marker I-eNgel
-327shy
(9) a ku (V-marker) - end (walk) - a (FV) --gt kuyenda (to walk) b ka (past) - amb (start) - a (FV) --gt kayamba (to start)
Some of the forms may take more than one resolving processes resulting in variations of
different forms as in (10)
(10) a ku (V-marker) - op (fear) - a (FV) --gt kopa (VD) kuwopa (GI) b ka (past) - im (stand) +a (FV) --gt kima (VD) kayima (GI)
In this section data of two consecutive vowels are presented and the changes of different
types will be discussed
21 Morpheme-internal and between stems and suffixes
First we will begin with morpheme internal vowel clashes and also the case where a stem
ends in a vowel and the following suffix begins with another vowel
(11)Morpheme internal vowel sequence ex~mple GF VD GI gloss
a imue imwe (you-pI) b niama nyama (meat) c ku-puan-a ku-pwan-a (to shatter) d uoko woko (arm) e nai nayi (four) f chi-mai chimayi (knife) g ku-iouoi-a ku-yowoy-a (to speak)
(12)V-final stem and V-initial suffix a ku-Ii - esx+a kulyesxa (to eat) b c
ku-temu - a Ii - eNge
kutemwa lyeNge
(to love) (will eat)
d ku-mani - eka kumanyeka (to know)
The interesting thing readily noticeable in above data is the lack of variation Each of the
examples shows either GF or GI but not both and no example of VD is witnessed Ivith (11) I
think that the absence of vowel deletion process is only natural because if vowel deletion is applied
-328shy
Hiatus Resolving in Chitumbuka 5
within a morpheme it is usually unrecoverable And therefore the native speakers will take the
form with vowel deletion change if any as the underlying form
This explanation gives rise to a possibility that there may not be any process at all be it GF
or GI in (11) For example we may say that the underlying form of (11b) is just nyumba or
Ijumba for that matter My assumption that the underlying base forms have vowels is based on
the following considerations Positing underlying vowels instead of glides does not make wrong
predictions In other words it never produces wrong phonetic forms and it does not fail to
produces all the correct forms as will be clearly illustrated in the course of the discussion In the
mean time this assumption gets rid of three possible phonemes Iy Iwl and IJ1 without losing any
explanatory power and thereby achieves the economy of phonological system
Of course the cost here is to posit GF rule The phonemic inventory is simplified at the
cost of complicating the rule component There is no clear judgement criterion But as will be
discussed in 31 GF rule may not be a costly rule to explain the specific cases as in (11) but a
natural consequence of the syllabification Further this reveals some characteristics of
syllabification in this language
In (12) we can see that there are restrictions in the range of data First we can see that all
the stem final vowels are high vowels and all the suffix-initial vowels are non-round vowels
Further there are suffixes with unspecified vowels such as IVkl (a passive marker) IVrl (an
applicative marker) IVsxl (a causative or an efferential marker) The unspecified vowels are realized
as lei if the preceding vowel is [-high -low] and as Iii if the preceding vowel is [ahigh -alow]
But if the stem ends in a vowel the unspecified vowel is realized as lei in spite of the fact that stem
3) I ran through Diocese of Mzuzu (1970) and all the data I have but there are no examples of morphemes of these kinds
-329shy
final vowels are always [s-high -low] as shown in (13)
(13) a ku (V-marker) -Ii (eat) - Vk (passive) - a (FV) --gt kulyeka (to be eaten) b ku(V-marker) - mu(drink) - Vsxtcausative) - a(FV) --gt kumwesxa (to make drink)
(13) ensures that the initial vowel of the suffix is always of greater sonority Therefore
stem final vowels are always less sonorous than the suffix initial vowels Now we can find one
thing in common between the data of (11) and (12) except (12f) and (12g)
We may therefore posit the following informal glide formation rule
(14) Glide formation (GF) Vl --gt [-syllabic]_V2
(condition V2 is not less sonorous than V1 )41
Now lets turn to (12f) and (12g) In these cases V1 is more sonorous than V2 Given
condition (6) and rule (14) on the one hand and the absence of vowel deletion on the other the
only way to resolve the ill-formed form seems to insert a glide in between two vowels At this
point it is not clear whether the inserted glide is y or w Though we only have the examples of
y-insertion the rule will be formulated to accomodate w so that we can use GI for discussions in
the following subsections
(15) Glide insertion (GI)
o --gt [~~~~~~~J
VI V2 [a round]
a round (Condition V1 is more sonorous than V2)
This rule inserts y before e and a and w before 051 One problem that should be
4)1 simply followed Jesperson (1932 191) to say that open vowels are more sonorous than closed vowels
5) Chitumbuka has five vowel system Its phonemic vowels are le iI lui lei 101
-330shy
7 Hiatus Resolving in Chitumbuka
pointed out now is the nature of the inserted segment y or WI These are not in the phonemic
inventory of Chitumbuka as discussed earlier Insertion of sounds which are not phonemic is the
characteristic of the phonetic rules that do not participate in the phonological derivation This may
mean that there should be glides in the phonemic inventory contrary to the basic assumption here
or that GI is formulated in the wrong way We will postpone the discussion to 31 and stick to GI
rule for the purpose of discussion in the present section
The GF in (14) and GI in (15) can take care of all the data of morpheme-internal vowel
clashes as well as the vowel sequence cause by the morpheme concatenation of V-final stems and V-
initial affixes
22 Between V-final affixes and V-initial stems
Another very productive source of vowel clashes is the concatenation of agreement markers
with V-initial stems In Chitumbuka nouns adjectives and verbs carry agreement markers and all
the agreement markers are V-final
(16) Agreement marker and V-initial nouns a mn shy ini mwini (owner) b mu - anice mwanice (child) c vi - aro vyaro (lands) d In - izsa Iwizxa (economy) e chi - aro charo (land) f va - ana vana (kids) g mn - ono mono mnwono (creel) h mn shy ova mova (beer)
(17) Agreement marker and possessive pronouns a vi - ane vyane (mine) b u - ane wane (mine) c tu - ako twako (yours-sg) d li - ake lake (hishers) e zi - awo zawo (theirs) f va - ithu vithu (ours) g va - ako vako (hishers)
-331shy
h li - inu inu (yours-pl)
Before going into any detail I should say that (16g) and (16h) may be problematic GF can
be applied but (16h) does not show the form with a glide Vail (197240) observed that mu-
alternates with m or syllabic m I will simply assume that this morphemic alternation is
responsible for the variations in (16g)61 It is not the application of different rules but the
coexistence of mu- and m- form that results in the two different surface forms in (16g)
We can make very interesting observations here the given data except for (16g) are similar
to (11) and (12) in that there is no variant form and in that there is a glide formation process One
difference is that no case of VD is witnessed in (11) and (12) while the data in (16) and (17)
presents very productive use of VD and do not show any instances of GI
All the examples presented show that when VI and V2 get together it is always the VI that
is deleted V2 remains undeleted This may not necessary mean that V2 triggers vowel deletion To
go deeper into VD phenomenon lets first consider environments for vowel deletion
(18) Vowel deletion environments a VI is less sonorous than V2 (16e) (16h) (l7d) (17e) b VI is more sonorous than V2 (17f) c VI andVj are identical (16f) (17g) (17h)
Apparently (18) covers almost all the possible combinations except i-u and u-i
sequences First of all we can see that (18a) shares similar environments with GF in (14) And
actually GF applies to such examples as in (16c) (16d) (17b) and (17e) Just saying that both GF
and VD are optional (cf Mphande (1989121)) and that they share same the same environments
6) I think that m- might have come from mu- through u-deletion This may be a historical change since it is very difficult to find the synchronic evidence that m- is related to mu- as in the case of mpaka (until) Then we may conclude that both forms are available for (16g) but only mshy is available for (16h)
-332~
9 Hiatus Resolving in Chitumbuka
does not explain why there is no variation
One may say that the difference in boundary type results in the application of the different
rules and add the boundary in the rule-description accordingly This can be illustrated as in (19)
(19) a ku - Ii - a --gt kulya b Ii - ake --gt lake
Though (19a) and (19b) shows the same sequence of i-a IIi in (19a) is a stem while it is
a prefix in (19b) And lal in (19a) is a grammatical suffix and lal in (19b) is a part of a lexical
morpheme In other words (19a) is suffixation and (19b) is prefixation By assigning different
boundaries we can explain why we have GI only in (11) and (12) and not in (16) and (17)
One example of the boundary functioning in conditioning rules may be found in English g-
deletion as in Hyman (1976197) According to them there are two different I-erl suffixes in
English The first one is introduced with word boundary as in singer Another is introduced
with morpheme boundary + as in long-er and English has a g-deletion rule that goes as under
(20) g-deletion in English g --gt 0N_ (N a velar nasal)
This rule with the boundary differentiation and with the specification of the boundary type
deletes IgI in singer but not in longer
The same strategy can be employed here We may assume that the stems are introduced
with one type of boundary say and the suffixes with another + we may represent (19) as
(21)7
7)1 will simply assume that suffixes are introduced with + boundary and prefixes are introduced with boundary But I will continue to use - to mark the morpheme break The boundary differentiation according to Aronoff (1976) means that suffixation preceded prefixation
-- 333shy
(21) aku li + a b li aka
Now we can formulate the following VD rule
(22) Vowel deletion Vi --gt 0 I Vz
This seems to be a reasonable approach I believe this may be partly true for (17) cases
But we need something else This does not help explaining (16) because we have to explain why
GF instead of VD is applied to such examples as (16a) to (16d) and (17a) to (17c)
One way out from this dilemma is to stick to glide formation for (16) then make a rule that
deletes the glide What is meant by this is that it is not VD but GF followed by GD in (16) as
shown in (23)
(23) li - ake --(GF)--gt lyake --(GD)--gt lake
This process is motivated by the fact that in all the cases of VD in GF environments Iii is
preceded by a coronal consonant as shown in (16e) (17d) and (17e) Notice also that the recent
works on features such as Sagey (1986) McCarthy (1988) and many others claim that palatal
glides have [coronal] feature Given this we may formulate the following GD rule
(24) Glide Deletion [+coronal] --gt 0 I [+ coronal]
The rule is maximally generalized We dont have to specify whether the sound deleted is a
glide The syllable structure given in section 1 will naturally predict that it is a glide since in C1C2
consequence the Cz must be a glide But still the boundary is necessary to prevent GD to such
words as ku li + a
Now lets take a look at (17) cases Here again we find that the environment of VD is
-334~
Hiatus Resolving in Chitumbuka 11
very similar to GI in (15) We will have to resort to the boundary distinction as proposed in
discussion with the sonority condition as in (25)
(25) Vowel deletion V1--gt0_V2 (conditionV1 is not less sonorous than V2)
The condition does not necessarily require a higher sonority for the first vowel This
reflects an effort to explain (17g) (17h) cases together with (17f) This approach however has
one immediate problem of explaining (16a) and (16d) VD as in (25) can apply to these cases but
mini and Tizxa are ill-formed One major observational difference of the examples to which
VD is applied is the roundness feature of the two vowels When two vowels have the same
roundness feature and the same height the first one deletes And GF applies when the two vowels
are of the same height (and therefore the same sonority) but differ in roundness This problem can
be solved by adding roundness agreement condition to VD as in (26)81
(26) Vowel deletion - revised VI __ gt 0 __ V2
[a round] [a round] (conditionVIis not less sonorous than V2)
23 Between V-final affixes and V-initial Verbs
There are five different types of verb prefixes They are Iku-I Iliku-I Ngu Izamu-I Ika-I
and le-I I called these prefixes not agreement markers The reason is two-fold first they have their
own meanings These prefixes express tenses or aspects of the verb And unlike agreement
8 lAnother possibility that is imposing roundness disagreement condition on the GF rule makes a wrong prediction that lyenge in (12c) and kumanyeka in (12d) should not undergo GF which is contradictory with the given data
9 )The followings are the general meaning of these prefixes a ku- Present tense present progressive or habitual aspect of the present
-335shy
12 7j-iilfHl ~1lyenlt(1993 2)
markers in 21 or 22 they do not carry grammatical information Further we find that agreement
markers are added before these prefixes in the sentences
Lets begin the discussion by looking at the following data
(27) Prefixes with final lui and V-initial verbs a ku shy amba kwamba kuyamba (to start) b ku - ambuka kwambuka kuyambuka (to cross) c ku - ata kwata kuyata (to create) d ku shy enda kwenda kuyenda (to walk) e ku shy eNga kweNga kuyeNga (to melt) f ku - imba kwimba kuyimba (to sing) g ku - ima kwima kuyima (to stand) h ku - iba kwiba (to steal) I ku - iza kw iza (to come) J ku shy ocha kocha kuwocha (to burn) k l
ku - opa ku - ona
kopa ku wopaku tJY(1Y~- Hu
(to fear) in ~f~Ofmiddot
ffi ku -ma KU wura (to fear) n ku - umba kuwurnba (to mold something) o ku - undira kuwundira (to bury)
(28)Tense or aspect marker and V-initial verbs a ka - imba kimba kayimba (sang) b ka - iba kiba (stole) c ka - enda kenda kayenda (walked) d ka - opa kawopa (feared) e ka - ona kawona (saw) f ka - undira kawundira (buried) g ka - amba kamba kayamba (started)
The first thing readily noticeable in the given data is the existence of different alternatives in
b liku- Perfective the state of verb or the result of activity in the past is carried on to the present
c zamu- Future tense d ka- distant past (activity in the past) e kaNgu-habitual activity in the past f a- Activity in the immediate past g aNgu- Duration of activity that ended in the immediate past
lO)Though all the examples given here have Iku-I prefix the exactly same phonological changes are observed with all other prefixes that end in lui such as INgu-l liku-j and zamu-z All the V-initial verbs elicited so far are given here
-336shy
Hiatus Resolving in Chitumbuka 13
most of the cases Lets first begin with the apparently exceptional cases that do not have
variations Our Chitumbuka informant was not sure whether the stems of (27h) and (27i) begin
with Iii or not On a couple of occasions he said that to steal is kuba And even when he
pronounced kwiba he still maintained that the stem is Ibal instead of iba Though without any
sufficient supporting data I take this as the evidence that the root form of to steal has lost the
alternations and fossilized into kwiba and the same might be true for (27i)
Next it is very strange that only (271) shows different behavior in comparison with the two
preceding examples (27j) and (27k) It may be extremely difficult to assume that (271) has lost the
alternations and fossilized into kuwona form Intuitively if it is to be fossilized the form should
have been kona a phonologically simpler form I suggest that the base form of to see is uon shy
a GF is applied to this to produce Iwonal and then there is no vowel clash when Iku-I is added
failing to meet the VD environment
(27m) (27n) and (270) look quite exceptional in that VD does not apply to theses examples
though they are in VD environments I do not have systematic explanation for these There may he
some historical change in these words III or the underlying forms may begin with luul and the GF
changes the first lui into glide before tense or aspect markers are added 121 or they are just
exceptional At this point I will leave this as an open question
The GF and VD are quite straight forward There is no need to revise the rules Now we
will have to think about GI cases GI rule given in (15) does not explain the examples given in this
11)For example their friend in Chitumbuka is munyawo or rnunyalo Here we sec the alternations hetween Iwland iraquo] The same thing might have happened in (27m) (27n) and (270) words we may speculate that historically those stems begin with lraquo] And the sound weakened into the glide Iw
12)50 far no two same segments have been found to be adjacent morpheme internally However given the GF rule environment I think this idea may not be totally speculative
-- 337shy
subsection the sonority condition is not satisfied
One possible linear phonology approach may be to resort to the rule ordering We may
eliminate the sonority condition of GI and simply order GI after GF Further GI should not be
applied in (16) and (17) and therefore VD should be placed before GI hence the following order
(29) Rule ordering 1 Glide formation 2 Vowel deletion 3 Glide insertion
The major problem with this approach is the nature of the rule The data (27) (28) indicate
that all the three rules are optional But even though they are optional there is no option of not
applying any of the rules and therefore the last rule (GI) should be obligatory Further the data
(11) tells us a different story GF seems to be obligatory Otherwise we may have to say that
(1Ia) can be realized either as imwe (GF) or as imuye (GI) Moreover (17) shows that VD is
obligatory to prevent the operation of GI Here is the paradox A rule is optional for certain group
of words and obligatory for the rest This may suggest that rule based linear approach may not be
suitable for the explanation of hiatus resolving in Chitumbuka Further theoretical investigation will
be made in 32
24 Subject agreement marker and past imperfective marker la-I
All the prefixes introduced in 21 and some other personal agreement markers are used in
the sentence to satisfy the subject agreement Meanwhile there is one V-initial tense marker in
chitumbuka a- And therefore here again we expect to find vowel clashes The representative
data are given in (30)
13) If any of the rule is obligatory then we should not see the variations in the data given in (27) and (28)
-338shy
Hiatus Resolving in Chitumbuka 15
(30) Subject agreement marker and past imperfective marker a ni - aluta naluta (I walked) b u - aluta waluta (you walked) c wa - aluta waluta (he walked) d ti - aluta taluta (we went) e li - aluta laluta (It went) f vi - aluta (vyaluta) valuta (They went)
Superficially the data given here looks like those in (17) There is no GI example And the
choice between GF and VD seems to be determined by the relative sonority of the vowels
However (30f) shows that there is something different here (30) may not be the extension of
examples given in (17) The reason is twofold first the examples given here are not lexical The
agreement marker assignment should refer to the syntactic category the subject Secondly there is
an independent process of i-deletion in chitumbuka Consider following examples
(31) i-deletion and complex segment formation a ni (copula) -litali (big) --gt nilitali or nditali laquoit) is big) b ni (subject agreement)- kulya --gt nikulya or Nkhulya laquol) eat)
Here again i is deleted though it is not followed by another vowel I will tentatively
formulate the i-deletion rule as in (32)
(32) i-deletion i --gt 0_ [ (where [ is a morphological word domain)
This rule as (31a) (31 b) and (30f) shows is optional Then what if the option is not taken
in (30) examples We might expect to see nyaluta in case of example (30a) Notice that the
14 )The irrelevant morpheme breaks are not shown here aluta can be further divided into a (past imperfective) -Iut (go)- a (FV)
15 )The form in the parenthesis means that it may be used by some But the other form is far more widely used and the informant prefers to use valuta
16 )Further investigation of phonological word derivation based on the syntactic information and morphological words should be made Here only the difference of phonological and morphological words is used without mentioning how different they are
-339shy
morphological word domain contains boundary and that the agreement markers begin with
coronal sounds If GF applies then GD in (24) will eliminate nyaluta
25 Unresolved vowel clashes
Not all the vowel sequences are resolved in Chitumbuka Some representative examples are
given in (33) and (34)
(33) V-final nouns and determiners a khuni Ili --gt khuni (y)ili khunyili khunili
tree this b buku ilo --gt buku (y)ilo bukwilo
book that c chipinda ico --gt chipinda (y)ico chipindico
room this
(34) Pronominal objects and V-initial verbs 171
a u-ku-mu-at-a --gt wakumu(y)ata wakumwata (you created him)
b wa-ku-ti-at-a --gt wakuti (y)ata wakutyata (He created you)
c ni-ku-tu-och-a --gt nikuku(w)ocha nikukocha nikukwocha (I roasted them)
d mu-ku-ka-eNg-a --gt mukuka(y)eNga muku (I melted it)
This seems to tell us that the main domain of hiatus resolving is within phonological words
GF and VD do not operate on this level and only G1 optionally shows up For further discussion
the phonological word boundaries are represented with parentheses in (35)
17 )The detailed breakdowns of the strings are as under
(yOU) ~u (him) tt (create)tni (me) ku + 11 (~e) + och (bake) + a (FV) wa (heher ka (It) Ng (melt)
u (them) verb stem
subject verb pronominal agreement tense objects
-340shy
Hiatus Resolving in Chitumbuka 17
(35) a (khuni) (ili) b (niku) (tu) (ocha) c (nikuwocha) (tuyuni)
One very interesting observation here is the change in the phonological word domains
(35b) and (35c) are semantically same The object in (35b) is pronominal while (35c) has a noun
object As a result of movement two phonological words are broken into three phonological
words This is not an arbitrary boundary just to explain the lack of hiatus resolving We may find
a very similar situation in English Intonational Phrase parsing
(36) a (As you know) (the tall boy is a genius) b (The tall boy) (as you know) (is a genius)
Nespor and Vogels (1986189) intonational phrase formation rule produces (36a) as well
as (36b) The syntax dependency of the phonological word formation is partly responsible for the
difference between (35b) and (35c)
Further hiatus is not the only rule sensitive to boundary difference in Chitumbuka We can
find that i-deletion and complex segment formation discussed in the preceding subsection shows
the same phenomenon Consider the following example (parentheses indicate the phonological
word domain)
(37) a (niku) (mu) (temwa) I love him Nkhu
b (uku) (ni)(temwa) You love me nthe
c (khunilClllno) the tree here mpha
One final observation is that our rules so far can not explain why glides can be inserted in
the examples of (33) and (34) We have already pointed out the problem of GI in 21 The
additional complication is that even if we posit glides as a phoneme we still have to revise GIor
-341shy
make another one to explain post-lexical GL This will be addressed in 32
3 Syllable phonology of hiatus resol ving
In the discussion in section 2 some of the problems in the linear analysis were detected
(What are they ) In this section I will go over them and reanalyze the data with the help
of syllable structure to see whether we can answer some of the questions left open McCarthy and
Prince (1986) style of moraic syllable structure will be employed for our discussion in this section
The basic syllable structure of Chitumbuka is given in (38)
(38) Syllable structure in Chitumbuka a a b a c a
~~((r r1T ( (I I 1 All the syllables are mora final 2 All the syllables are mono-moraic
This is not the template that is required in the underlying form as will be discussed
Therefore the non-moraic tier may not have to be satisfied In other words though the structure
shows two pre-moraic consonant slots these can be left empty The two branches simple mean the
maximum number of consonants that can come before a vowel
18)1n the post lexical level this structural condition may not hold In Chitumbuka vowels in the stressed syllable is lengthened The long vowel is represented with bimoraic syllable structure It is not uncommon that some of the rules or conditions are applied only on the lexical level but not on the post lexical level such as extrametricality stress assignment and even VD and GF we talked about in the preceding section
- 342-shy
Hiatus Resolving in Chitumbuka 19
31 Syllabification
Contrary to the claim that all the segments are already syllabified in the base form (cf
Selkirk (1982raquo I assume that syllable structure is assigned through the syllabification process
The syllabification principles in Chitumbuka are given under
(39) Syllabification in Chitumbuka i) Morafication A segment projects a mora if it is more sonorous than the following
segrnent ii) Syllable projection Each mora projects a syllable node iii) Segment incorporation Unsyllabified segments are incorporated into syllables
Now lets take a look at the following sample syllabification
(40) a
0 0 00
I (iii)I(ii) --(i) f1 f1 f1 f1 ~ rlrOI I
~
I I n i u mb a n 1 u mb a n 1 u mb a n i u mb a
b
0 o 0 0 o 0
I I I (iW (~Of1 f1 f1 f1 ~ ~ I I r~r I I I I I ch i m a i ch i m a ch i m a ch i m a i
Notice that we do not need GF as a rule the same result of GF application is achieved as a
19)1 suppose that morafication principle is universally dependent upon the sonority hierarchy There are three different strategies in using sonority in syllabifications
i) y is moraic in xy if it is more sonorous than x ii) y is moraic in yz if it is more sonorous than z iii) y is moraic if its sonority is greater than N
The language in the world employ one or two all all of these in the syllabification and language may slightly vary in setting up the value of N in (iii)
-- 343shy
natural consequence of syllabification In 21 we obtain the simplicity of phonemic inventory at
the cost of introducing GF thereby complicating the rule component The environment of GF is
reflected in the syllabification principle Of course syllabification can not be an additional burden
the language needs syllable structure anyhow for example for the application of many phonological
rules and even for utterance
The nature of glide sounds are clearly reflected in the syllable structure If a vowel is
dominated by a mora it will be interpreted as a full vowel But if it is dominated by the syllable
node directly it will surface as a glide
32 Syllabification domain
The syllabification principles introduced in 31 have presupposed that the string of segments
are not syllabified in the underlying representation Suppose every morpheme is syllabified in the
deep structure and consider the following example of word formation
(41)
a aa a
~ (I (noa k 0 V 1 a k 0
Now we will have to make a rule for mora deletion which will be almost the repetition of
the syllabification principle Syllabification process in the course of derivation can get rid of the
redundant repetition of syllable structural description as well as save the later-on application of mora
deletion
I assume that the syllabification takes place in the first derived domain In other words
segments are syllabified after the initial word formation process Before going into the detailed
-344shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
b All the suffixes are V-initial c Some of the stems are V-initial andor V-final
We can immediately see that vowels can come together through prefixation or suffixation to
stems Given the two observations (2) and (3) we are expect to see that there may be certain
phonological precesses in Chitumbuka that take two consecutive adjacent vowels as inputs and
produce outputs that confirm to the observation in (2)
I will call them hiatus resolving processes In this paper different strategies of hiatus resolving in Chitumbuka will be discussed Generally three different operations are noticeable
(4) a The first of two consecutive vowels becomes a glide b The first vowel is lost or deleted c A glide appears between the two vowels
In section 2 data of vowel clashes observed in various morpheme concatenation processes
will be presented and strategies of resolving hiatus will be discussed employing linear segmental
approach It will be seen that there are some problems in this analysis and the nature of the
problems will also be discussed
Section 3 presents another analysis of the same data from a different angle using moraic
syllable structure It will also be shown that the syllable-based approach can provide solution to
some of the problems posed in section 2
The data used in this paper have two different sources one from the transcriptions that I got
with the help of the native Chitumbuka speaker Mr Wongani Katundu And some of them are
from Tumbuka-Tonga- English dictionary I got the necessary words and checked them with the
informant and screen them to use only those which are in his vocabulary
-326shy
3 Hiatus Resolving in Chitumbuka
2 Vowel clashes
In Chitumbuka the sequence of two vowels in a word is not allowed as shown in (5)
(5) a mu (AGR) - ova (beer) --gt muova b ka (past) - imba (sing) --gt kaimba
As a first approximation we will just assume that there is principled prohibition of vowel
clashes which can be expressible as in (6)
(6) Well-formedness condition No two vowels can be adjacent without intervening consonant(s)
Though there may be problems with this condition this may suffice to begin the discussion
And more constrained condition will be introduced in 31 In spite of this condition we can
immediately see that through morpheme concatenation or even in the base form or underlying
representation vowels can come together Three different types of changes are witnessed in
Chitumbuka to conform to (6) First we can see the glide formation (henceforth GF) as in (7)
(7) a ku (V-marker) -Ii (eat) - eNge (future) --gt kulyeNge (will eat immediately) b mu (AGR) - ana (female) --gtmwana (a girl)
As shown in (7) the first of the two consequent vowels becomes a glide And since the
glide is not a vowel the hiatus is resolved
Secondly we can see the case of vowel deletion (henceforth VD) operation as in (8)
(8) a ka (past) - imb (sing) - a (FV) --gt kimba (to sing) b ku (V-marker) - och (burn) - a (FV) --gt kocha (to burn)
Finally a glide may be inserted (GI glide insertion) between two vowels as in (9)
1 )IlN the capitaln in the transcription represents the velar nasal [Q ] 2)FV means final vowel The morphological meaning of this affix is not fully investigated
In general the -a means indicative mood since it can be used with the past tense marker Ika-I or la-I and the future-tense marker Izamu-I but not compatible with subjunctive mood ending I-el or next-time point (future) marker I-eNgel
-327shy
(9) a ku (V-marker) - end (walk) - a (FV) --gt kuyenda (to walk) b ka (past) - amb (start) - a (FV) --gt kayamba (to start)
Some of the forms may take more than one resolving processes resulting in variations of
different forms as in (10)
(10) a ku (V-marker) - op (fear) - a (FV) --gt kopa (VD) kuwopa (GI) b ka (past) - im (stand) +a (FV) --gt kima (VD) kayima (GI)
In this section data of two consecutive vowels are presented and the changes of different
types will be discussed
21 Morpheme-internal and between stems and suffixes
First we will begin with morpheme internal vowel clashes and also the case where a stem
ends in a vowel and the following suffix begins with another vowel
(11)Morpheme internal vowel sequence ex~mple GF VD GI gloss
a imue imwe (you-pI) b niama nyama (meat) c ku-puan-a ku-pwan-a (to shatter) d uoko woko (arm) e nai nayi (four) f chi-mai chimayi (knife) g ku-iouoi-a ku-yowoy-a (to speak)
(12)V-final stem and V-initial suffix a ku-Ii - esx+a kulyesxa (to eat) b c
ku-temu - a Ii - eNge
kutemwa lyeNge
(to love) (will eat)
d ku-mani - eka kumanyeka (to know)
The interesting thing readily noticeable in above data is the lack of variation Each of the
examples shows either GF or GI but not both and no example of VD is witnessed Ivith (11) I
think that the absence of vowel deletion process is only natural because if vowel deletion is applied
-328shy
Hiatus Resolving in Chitumbuka 5
within a morpheme it is usually unrecoverable And therefore the native speakers will take the
form with vowel deletion change if any as the underlying form
This explanation gives rise to a possibility that there may not be any process at all be it GF
or GI in (11) For example we may say that the underlying form of (11b) is just nyumba or
Ijumba for that matter My assumption that the underlying base forms have vowels is based on
the following considerations Positing underlying vowels instead of glides does not make wrong
predictions In other words it never produces wrong phonetic forms and it does not fail to
produces all the correct forms as will be clearly illustrated in the course of the discussion In the
mean time this assumption gets rid of three possible phonemes Iy Iwl and IJ1 without losing any
explanatory power and thereby achieves the economy of phonological system
Of course the cost here is to posit GF rule The phonemic inventory is simplified at the
cost of complicating the rule component There is no clear judgement criterion But as will be
discussed in 31 GF rule may not be a costly rule to explain the specific cases as in (11) but a
natural consequence of the syllabification Further this reveals some characteristics of
syllabification in this language
In (12) we can see that there are restrictions in the range of data First we can see that all
the stem final vowels are high vowels and all the suffix-initial vowels are non-round vowels
Further there are suffixes with unspecified vowels such as IVkl (a passive marker) IVrl (an
applicative marker) IVsxl (a causative or an efferential marker) The unspecified vowels are realized
as lei if the preceding vowel is [-high -low] and as Iii if the preceding vowel is [ahigh -alow]
But if the stem ends in a vowel the unspecified vowel is realized as lei in spite of the fact that stem
3) I ran through Diocese of Mzuzu (1970) and all the data I have but there are no examples of morphemes of these kinds
-329shy
final vowels are always [s-high -low] as shown in (13)
(13) a ku (V-marker) -Ii (eat) - Vk (passive) - a (FV) --gt kulyeka (to be eaten) b ku(V-marker) - mu(drink) - Vsxtcausative) - a(FV) --gt kumwesxa (to make drink)
(13) ensures that the initial vowel of the suffix is always of greater sonority Therefore
stem final vowels are always less sonorous than the suffix initial vowels Now we can find one
thing in common between the data of (11) and (12) except (12f) and (12g)
We may therefore posit the following informal glide formation rule
(14) Glide formation (GF) Vl --gt [-syllabic]_V2
(condition V2 is not less sonorous than V1 )41
Now lets turn to (12f) and (12g) In these cases V1 is more sonorous than V2 Given
condition (6) and rule (14) on the one hand and the absence of vowel deletion on the other the
only way to resolve the ill-formed form seems to insert a glide in between two vowels At this
point it is not clear whether the inserted glide is y or w Though we only have the examples of
y-insertion the rule will be formulated to accomodate w so that we can use GI for discussions in
the following subsections
(15) Glide insertion (GI)
o --gt [~~~~~~~J
VI V2 [a round]
a round (Condition V1 is more sonorous than V2)
This rule inserts y before e and a and w before 051 One problem that should be
4)1 simply followed Jesperson (1932 191) to say that open vowels are more sonorous than closed vowels
5) Chitumbuka has five vowel system Its phonemic vowels are le iI lui lei 101
-330shy
7 Hiatus Resolving in Chitumbuka
pointed out now is the nature of the inserted segment y or WI These are not in the phonemic
inventory of Chitumbuka as discussed earlier Insertion of sounds which are not phonemic is the
characteristic of the phonetic rules that do not participate in the phonological derivation This may
mean that there should be glides in the phonemic inventory contrary to the basic assumption here
or that GI is formulated in the wrong way We will postpone the discussion to 31 and stick to GI
rule for the purpose of discussion in the present section
The GF in (14) and GI in (15) can take care of all the data of morpheme-internal vowel
clashes as well as the vowel sequence cause by the morpheme concatenation of V-final stems and V-
initial affixes
22 Between V-final affixes and V-initial stems
Another very productive source of vowel clashes is the concatenation of agreement markers
with V-initial stems In Chitumbuka nouns adjectives and verbs carry agreement markers and all
the agreement markers are V-final
(16) Agreement marker and V-initial nouns a mn shy ini mwini (owner) b mu - anice mwanice (child) c vi - aro vyaro (lands) d In - izsa Iwizxa (economy) e chi - aro charo (land) f va - ana vana (kids) g mn - ono mono mnwono (creel) h mn shy ova mova (beer)
(17) Agreement marker and possessive pronouns a vi - ane vyane (mine) b u - ane wane (mine) c tu - ako twako (yours-sg) d li - ake lake (hishers) e zi - awo zawo (theirs) f va - ithu vithu (ours) g va - ako vako (hishers)
-331shy
h li - inu inu (yours-pl)
Before going into any detail I should say that (16g) and (16h) may be problematic GF can
be applied but (16h) does not show the form with a glide Vail (197240) observed that mu-
alternates with m or syllabic m I will simply assume that this morphemic alternation is
responsible for the variations in (16g)61 It is not the application of different rules but the
coexistence of mu- and m- form that results in the two different surface forms in (16g)
We can make very interesting observations here the given data except for (16g) are similar
to (11) and (12) in that there is no variant form and in that there is a glide formation process One
difference is that no case of VD is witnessed in (11) and (12) while the data in (16) and (17)
presents very productive use of VD and do not show any instances of GI
All the examples presented show that when VI and V2 get together it is always the VI that
is deleted V2 remains undeleted This may not necessary mean that V2 triggers vowel deletion To
go deeper into VD phenomenon lets first consider environments for vowel deletion
(18) Vowel deletion environments a VI is less sonorous than V2 (16e) (16h) (l7d) (17e) b VI is more sonorous than V2 (17f) c VI andVj are identical (16f) (17g) (17h)
Apparently (18) covers almost all the possible combinations except i-u and u-i
sequences First of all we can see that (18a) shares similar environments with GF in (14) And
actually GF applies to such examples as in (16c) (16d) (17b) and (17e) Just saying that both GF
and VD are optional (cf Mphande (1989121)) and that they share same the same environments
6) I think that m- might have come from mu- through u-deletion This may be a historical change since it is very difficult to find the synchronic evidence that m- is related to mu- as in the case of mpaka (until) Then we may conclude that both forms are available for (16g) but only mshy is available for (16h)
-332~
9 Hiatus Resolving in Chitumbuka
does not explain why there is no variation
One may say that the difference in boundary type results in the application of the different
rules and add the boundary in the rule-description accordingly This can be illustrated as in (19)
(19) a ku - Ii - a --gt kulya b Ii - ake --gt lake
Though (19a) and (19b) shows the same sequence of i-a IIi in (19a) is a stem while it is
a prefix in (19b) And lal in (19a) is a grammatical suffix and lal in (19b) is a part of a lexical
morpheme In other words (19a) is suffixation and (19b) is prefixation By assigning different
boundaries we can explain why we have GI only in (11) and (12) and not in (16) and (17)
One example of the boundary functioning in conditioning rules may be found in English g-
deletion as in Hyman (1976197) According to them there are two different I-erl suffixes in
English The first one is introduced with word boundary as in singer Another is introduced
with morpheme boundary + as in long-er and English has a g-deletion rule that goes as under
(20) g-deletion in English g --gt 0N_ (N a velar nasal)
This rule with the boundary differentiation and with the specification of the boundary type
deletes IgI in singer but not in longer
The same strategy can be employed here We may assume that the stems are introduced
with one type of boundary say and the suffixes with another + we may represent (19) as
(21)7
7)1 will simply assume that suffixes are introduced with + boundary and prefixes are introduced with boundary But I will continue to use - to mark the morpheme break The boundary differentiation according to Aronoff (1976) means that suffixation preceded prefixation
-- 333shy
(21) aku li + a b li aka
Now we can formulate the following VD rule
(22) Vowel deletion Vi --gt 0 I Vz
This seems to be a reasonable approach I believe this may be partly true for (17) cases
But we need something else This does not help explaining (16) because we have to explain why
GF instead of VD is applied to such examples as (16a) to (16d) and (17a) to (17c)
One way out from this dilemma is to stick to glide formation for (16) then make a rule that
deletes the glide What is meant by this is that it is not VD but GF followed by GD in (16) as
shown in (23)
(23) li - ake --(GF)--gt lyake --(GD)--gt lake
This process is motivated by the fact that in all the cases of VD in GF environments Iii is
preceded by a coronal consonant as shown in (16e) (17d) and (17e) Notice also that the recent
works on features such as Sagey (1986) McCarthy (1988) and many others claim that palatal
glides have [coronal] feature Given this we may formulate the following GD rule
(24) Glide Deletion [+coronal] --gt 0 I [+ coronal]
The rule is maximally generalized We dont have to specify whether the sound deleted is a
glide The syllable structure given in section 1 will naturally predict that it is a glide since in C1C2
consequence the Cz must be a glide But still the boundary is necessary to prevent GD to such
words as ku li + a
Now lets take a look at (17) cases Here again we find that the environment of VD is
-334~
Hiatus Resolving in Chitumbuka 11
very similar to GI in (15) We will have to resort to the boundary distinction as proposed in
discussion with the sonority condition as in (25)
(25) Vowel deletion V1--gt0_V2 (conditionV1 is not less sonorous than V2)
The condition does not necessarily require a higher sonority for the first vowel This
reflects an effort to explain (17g) (17h) cases together with (17f) This approach however has
one immediate problem of explaining (16a) and (16d) VD as in (25) can apply to these cases but
mini and Tizxa are ill-formed One major observational difference of the examples to which
VD is applied is the roundness feature of the two vowels When two vowels have the same
roundness feature and the same height the first one deletes And GF applies when the two vowels
are of the same height (and therefore the same sonority) but differ in roundness This problem can
be solved by adding roundness agreement condition to VD as in (26)81
(26) Vowel deletion - revised VI __ gt 0 __ V2
[a round] [a round] (conditionVIis not less sonorous than V2)
23 Between V-final affixes and V-initial Verbs
There are five different types of verb prefixes They are Iku-I Iliku-I Ngu Izamu-I Ika-I
and le-I I called these prefixes not agreement markers The reason is two-fold first they have their
own meanings These prefixes express tenses or aspects of the verb And unlike agreement
8 lAnother possibility that is imposing roundness disagreement condition on the GF rule makes a wrong prediction that lyenge in (12c) and kumanyeka in (12d) should not undergo GF which is contradictory with the given data
9 )The followings are the general meaning of these prefixes a ku- Present tense present progressive or habitual aspect of the present
-335shy
12 7j-iilfHl ~1lyenlt(1993 2)
markers in 21 or 22 they do not carry grammatical information Further we find that agreement
markers are added before these prefixes in the sentences
Lets begin the discussion by looking at the following data
(27) Prefixes with final lui and V-initial verbs a ku shy amba kwamba kuyamba (to start) b ku - ambuka kwambuka kuyambuka (to cross) c ku - ata kwata kuyata (to create) d ku shy enda kwenda kuyenda (to walk) e ku shy eNga kweNga kuyeNga (to melt) f ku - imba kwimba kuyimba (to sing) g ku - ima kwima kuyima (to stand) h ku - iba kwiba (to steal) I ku - iza kw iza (to come) J ku shy ocha kocha kuwocha (to burn) k l
ku - opa ku - ona
kopa ku wopaku tJY(1Y~- Hu
(to fear) in ~f~Ofmiddot
ffi ku -ma KU wura (to fear) n ku - umba kuwurnba (to mold something) o ku - undira kuwundira (to bury)
(28)Tense or aspect marker and V-initial verbs a ka - imba kimba kayimba (sang) b ka - iba kiba (stole) c ka - enda kenda kayenda (walked) d ka - opa kawopa (feared) e ka - ona kawona (saw) f ka - undira kawundira (buried) g ka - amba kamba kayamba (started)
The first thing readily noticeable in the given data is the existence of different alternatives in
b liku- Perfective the state of verb or the result of activity in the past is carried on to the present
c zamu- Future tense d ka- distant past (activity in the past) e kaNgu-habitual activity in the past f a- Activity in the immediate past g aNgu- Duration of activity that ended in the immediate past
lO)Though all the examples given here have Iku-I prefix the exactly same phonological changes are observed with all other prefixes that end in lui such as INgu-l liku-j and zamu-z All the V-initial verbs elicited so far are given here
-336shy
Hiatus Resolving in Chitumbuka 13
most of the cases Lets first begin with the apparently exceptional cases that do not have
variations Our Chitumbuka informant was not sure whether the stems of (27h) and (27i) begin
with Iii or not On a couple of occasions he said that to steal is kuba And even when he
pronounced kwiba he still maintained that the stem is Ibal instead of iba Though without any
sufficient supporting data I take this as the evidence that the root form of to steal has lost the
alternations and fossilized into kwiba and the same might be true for (27i)
Next it is very strange that only (271) shows different behavior in comparison with the two
preceding examples (27j) and (27k) It may be extremely difficult to assume that (271) has lost the
alternations and fossilized into kuwona form Intuitively if it is to be fossilized the form should
have been kona a phonologically simpler form I suggest that the base form of to see is uon shy
a GF is applied to this to produce Iwonal and then there is no vowel clash when Iku-I is added
failing to meet the VD environment
(27m) (27n) and (270) look quite exceptional in that VD does not apply to theses examples
though they are in VD environments I do not have systematic explanation for these There may he
some historical change in these words III or the underlying forms may begin with luul and the GF
changes the first lui into glide before tense or aspect markers are added 121 or they are just
exceptional At this point I will leave this as an open question
The GF and VD are quite straight forward There is no need to revise the rules Now we
will have to think about GI cases GI rule given in (15) does not explain the examples given in this
11)For example their friend in Chitumbuka is munyawo or rnunyalo Here we sec the alternations hetween Iwland iraquo] The same thing might have happened in (27m) (27n) and (270) words we may speculate that historically those stems begin with lraquo] And the sound weakened into the glide Iw
12)50 far no two same segments have been found to be adjacent morpheme internally However given the GF rule environment I think this idea may not be totally speculative
-- 337shy
subsection the sonority condition is not satisfied
One possible linear phonology approach may be to resort to the rule ordering We may
eliminate the sonority condition of GI and simply order GI after GF Further GI should not be
applied in (16) and (17) and therefore VD should be placed before GI hence the following order
(29) Rule ordering 1 Glide formation 2 Vowel deletion 3 Glide insertion
The major problem with this approach is the nature of the rule The data (27) (28) indicate
that all the three rules are optional But even though they are optional there is no option of not
applying any of the rules and therefore the last rule (GI) should be obligatory Further the data
(11) tells us a different story GF seems to be obligatory Otherwise we may have to say that
(1Ia) can be realized either as imwe (GF) or as imuye (GI) Moreover (17) shows that VD is
obligatory to prevent the operation of GI Here is the paradox A rule is optional for certain group
of words and obligatory for the rest This may suggest that rule based linear approach may not be
suitable for the explanation of hiatus resolving in Chitumbuka Further theoretical investigation will
be made in 32
24 Subject agreement marker and past imperfective marker la-I
All the prefixes introduced in 21 and some other personal agreement markers are used in
the sentence to satisfy the subject agreement Meanwhile there is one V-initial tense marker in
chitumbuka a- And therefore here again we expect to find vowel clashes The representative
data are given in (30)
13) If any of the rule is obligatory then we should not see the variations in the data given in (27) and (28)
-338shy
Hiatus Resolving in Chitumbuka 15
(30) Subject agreement marker and past imperfective marker a ni - aluta naluta (I walked) b u - aluta waluta (you walked) c wa - aluta waluta (he walked) d ti - aluta taluta (we went) e li - aluta laluta (It went) f vi - aluta (vyaluta) valuta (They went)
Superficially the data given here looks like those in (17) There is no GI example And the
choice between GF and VD seems to be determined by the relative sonority of the vowels
However (30f) shows that there is something different here (30) may not be the extension of
examples given in (17) The reason is twofold first the examples given here are not lexical The
agreement marker assignment should refer to the syntactic category the subject Secondly there is
an independent process of i-deletion in chitumbuka Consider following examples
(31) i-deletion and complex segment formation a ni (copula) -litali (big) --gt nilitali or nditali laquoit) is big) b ni (subject agreement)- kulya --gt nikulya or Nkhulya laquol) eat)
Here again i is deleted though it is not followed by another vowel I will tentatively
formulate the i-deletion rule as in (32)
(32) i-deletion i --gt 0_ [ (where [ is a morphological word domain)
This rule as (31a) (31 b) and (30f) shows is optional Then what if the option is not taken
in (30) examples We might expect to see nyaluta in case of example (30a) Notice that the
14 )The irrelevant morpheme breaks are not shown here aluta can be further divided into a (past imperfective) -Iut (go)- a (FV)
15 )The form in the parenthesis means that it may be used by some But the other form is far more widely used and the informant prefers to use valuta
16 )Further investigation of phonological word derivation based on the syntactic information and morphological words should be made Here only the difference of phonological and morphological words is used without mentioning how different they are
-339shy
morphological word domain contains boundary and that the agreement markers begin with
coronal sounds If GF applies then GD in (24) will eliminate nyaluta
25 Unresolved vowel clashes
Not all the vowel sequences are resolved in Chitumbuka Some representative examples are
given in (33) and (34)
(33) V-final nouns and determiners a khuni Ili --gt khuni (y)ili khunyili khunili
tree this b buku ilo --gt buku (y)ilo bukwilo
book that c chipinda ico --gt chipinda (y)ico chipindico
room this
(34) Pronominal objects and V-initial verbs 171
a u-ku-mu-at-a --gt wakumu(y)ata wakumwata (you created him)
b wa-ku-ti-at-a --gt wakuti (y)ata wakutyata (He created you)
c ni-ku-tu-och-a --gt nikuku(w)ocha nikukocha nikukwocha (I roasted them)
d mu-ku-ka-eNg-a --gt mukuka(y)eNga muku (I melted it)
This seems to tell us that the main domain of hiatus resolving is within phonological words
GF and VD do not operate on this level and only G1 optionally shows up For further discussion
the phonological word boundaries are represented with parentheses in (35)
17 )The detailed breakdowns of the strings are as under
(yOU) ~u (him) tt (create)tni (me) ku + 11 (~e) + och (bake) + a (FV) wa (heher ka (It) Ng (melt)
u (them) verb stem
subject verb pronominal agreement tense objects
-340shy
Hiatus Resolving in Chitumbuka 17
(35) a (khuni) (ili) b (niku) (tu) (ocha) c (nikuwocha) (tuyuni)
One very interesting observation here is the change in the phonological word domains
(35b) and (35c) are semantically same The object in (35b) is pronominal while (35c) has a noun
object As a result of movement two phonological words are broken into three phonological
words This is not an arbitrary boundary just to explain the lack of hiatus resolving We may find
a very similar situation in English Intonational Phrase parsing
(36) a (As you know) (the tall boy is a genius) b (The tall boy) (as you know) (is a genius)
Nespor and Vogels (1986189) intonational phrase formation rule produces (36a) as well
as (36b) The syntax dependency of the phonological word formation is partly responsible for the
difference between (35b) and (35c)
Further hiatus is not the only rule sensitive to boundary difference in Chitumbuka We can
find that i-deletion and complex segment formation discussed in the preceding subsection shows
the same phenomenon Consider the following example (parentheses indicate the phonological
word domain)
(37) a (niku) (mu) (temwa) I love him Nkhu
b (uku) (ni)(temwa) You love me nthe
c (khunilClllno) the tree here mpha
One final observation is that our rules so far can not explain why glides can be inserted in
the examples of (33) and (34) We have already pointed out the problem of GI in 21 The
additional complication is that even if we posit glides as a phoneme we still have to revise GIor
-341shy
make another one to explain post-lexical GL This will be addressed in 32
3 Syllable phonology of hiatus resol ving
In the discussion in section 2 some of the problems in the linear analysis were detected
(What are they ) In this section I will go over them and reanalyze the data with the help
of syllable structure to see whether we can answer some of the questions left open McCarthy and
Prince (1986) style of moraic syllable structure will be employed for our discussion in this section
The basic syllable structure of Chitumbuka is given in (38)
(38) Syllable structure in Chitumbuka a a b a c a
~~((r r1T ( (I I 1 All the syllables are mora final 2 All the syllables are mono-moraic
This is not the template that is required in the underlying form as will be discussed
Therefore the non-moraic tier may not have to be satisfied In other words though the structure
shows two pre-moraic consonant slots these can be left empty The two branches simple mean the
maximum number of consonants that can come before a vowel
18)1n the post lexical level this structural condition may not hold In Chitumbuka vowels in the stressed syllable is lengthened The long vowel is represented with bimoraic syllable structure It is not uncommon that some of the rules or conditions are applied only on the lexical level but not on the post lexical level such as extrametricality stress assignment and even VD and GF we talked about in the preceding section
- 342-shy
Hiatus Resolving in Chitumbuka 19
31 Syllabification
Contrary to the claim that all the segments are already syllabified in the base form (cf
Selkirk (1982raquo I assume that syllable structure is assigned through the syllabification process
The syllabification principles in Chitumbuka are given under
(39) Syllabification in Chitumbuka i) Morafication A segment projects a mora if it is more sonorous than the following
segrnent ii) Syllable projection Each mora projects a syllable node iii) Segment incorporation Unsyllabified segments are incorporated into syllables
Now lets take a look at the following sample syllabification
(40) a
0 0 00
I (iii)I(ii) --(i) f1 f1 f1 f1 ~ rlrOI I
~
I I n i u mb a n 1 u mb a n 1 u mb a n i u mb a
b
0 o 0 0 o 0
I I I (iW (~Of1 f1 f1 f1 ~ ~ I I r~r I I I I I ch i m a i ch i m a ch i m a ch i m a i
Notice that we do not need GF as a rule the same result of GF application is achieved as a
19)1 suppose that morafication principle is universally dependent upon the sonority hierarchy There are three different strategies in using sonority in syllabifications
i) y is moraic in xy if it is more sonorous than x ii) y is moraic in yz if it is more sonorous than z iii) y is moraic if its sonority is greater than N
The language in the world employ one or two all all of these in the syllabification and language may slightly vary in setting up the value of N in (iii)
-- 343shy
natural consequence of syllabification In 21 we obtain the simplicity of phonemic inventory at
the cost of introducing GF thereby complicating the rule component The environment of GF is
reflected in the syllabification principle Of course syllabification can not be an additional burden
the language needs syllable structure anyhow for example for the application of many phonological
rules and even for utterance
The nature of glide sounds are clearly reflected in the syllable structure If a vowel is
dominated by a mora it will be interpreted as a full vowel But if it is dominated by the syllable
node directly it will surface as a glide
32 Syllabification domain
The syllabification principles introduced in 31 have presupposed that the string of segments
are not syllabified in the underlying representation Suppose every morpheme is syllabified in the
deep structure and consider the following example of word formation
(41)
a aa a
~ (I (noa k 0 V 1 a k 0
Now we will have to make a rule for mora deletion which will be almost the repetition of
the syllabification principle Syllabification process in the course of derivation can get rid of the
redundant repetition of syllable structural description as well as save the later-on application of mora
deletion
I assume that the syllabification takes place in the first derived domain In other words
segments are syllabified after the initial word formation process Before going into the detailed
-344shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
3 Hiatus Resolving in Chitumbuka
2 Vowel clashes
In Chitumbuka the sequence of two vowels in a word is not allowed as shown in (5)
(5) a mu (AGR) - ova (beer) --gt muova b ka (past) - imba (sing) --gt kaimba
As a first approximation we will just assume that there is principled prohibition of vowel
clashes which can be expressible as in (6)
(6) Well-formedness condition No two vowels can be adjacent without intervening consonant(s)
Though there may be problems with this condition this may suffice to begin the discussion
And more constrained condition will be introduced in 31 In spite of this condition we can
immediately see that through morpheme concatenation or even in the base form or underlying
representation vowels can come together Three different types of changes are witnessed in
Chitumbuka to conform to (6) First we can see the glide formation (henceforth GF) as in (7)
(7) a ku (V-marker) -Ii (eat) - eNge (future) --gt kulyeNge (will eat immediately) b mu (AGR) - ana (female) --gtmwana (a girl)
As shown in (7) the first of the two consequent vowels becomes a glide And since the
glide is not a vowel the hiatus is resolved
Secondly we can see the case of vowel deletion (henceforth VD) operation as in (8)
(8) a ka (past) - imb (sing) - a (FV) --gt kimba (to sing) b ku (V-marker) - och (burn) - a (FV) --gt kocha (to burn)
Finally a glide may be inserted (GI glide insertion) between two vowels as in (9)
1 )IlN the capitaln in the transcription represents the velar nasal [Q ] 2)FV means final vowel The morphological meaning of this affix is not fully investigated
In general the -a means indicative mood since it can be used with the past tense marker Ika-I or la-I and the future-tense marker Izamu-I but not compatible with subjunctive mood ending I-el or next-time point (future) marker I-eNgel
-327shy
(9) a ku (V-marker) - end (walk) - a (FV) --gt kuyenda (to walk) b ka (past) - amb (start) - a (FV) --gt kayamba (to start)
Some of the forms may take more than one resolving processes resulting in variations of
different forms as in (10)
(10) a ku (V-marker) - op (fear) - a (FV) --gt kopa (VD) kuwopa (GI) b ka (past) - im (stand) +a (FV) --gt kima (VD) kayima (GI)
In this section data of two consecutive vowels are presented and the changes of different
types will be discussed
21 Morpheme-internal and between stems and suffixes
First we will begin with morpheme internal vowel clashes and also the case where a stem
ends in a vowel and the following suffix begins with another vowel
(11)Morpheme internal vowel sequence ex~mple GF VD GI gloss
a imue imwe (you-pI) b niama nyama (meat) c ku-puan-a ku-pwan-a (to shatter) d uoko woko (arm) e nai nayi (four) f chi-mai chimayi (knife) g ku-iouoi-a ku-yowoy-a (to speak)
(12)V-final stem and V-initial suffix a ku-Ii - esx+a kulyesxa (to eat) b c
ku-temu - a Ii - eNge
kutemwa lyeNge
(to love) (will eat)
d ku-mani - eka kumanyeka (to know)
The interesting thing readily noticeable in above data is the lack of variation Each of the
examples shows either GF or GI but not both and no example of VD is witnessed Ivith (11) I
think that the absence of vowel deletion process is only natural because if vowel deletion is applied
-328shy
Hiatus Resolving in Chitumbuka 5
within a morpheme it is usually unrecoverable And therefore the native speakers will take the
form with vowel deletion change if any as the underlying form
This explanation gives rise to a possibility that there may not be any process at all be it GF
or GI in (11) For example we may say that the underlying form of (11b) is just nyumba or
Ijumba for that matter My assumption that the underlying base forms have vowels is based on
the following considerations Positing underlying vowels instead of glides does not make wrong
predictions In other words it never produces wrong phonetic forms and it does not fail to
produces all the correct forms as will be clearly illustrated in the course of the discussion In the
mean time this assumption gets rid of three possible phonemes Iy Iwl and IJ1 without losing any
explanatory power and thereby achieves the economy of phonological system
Of course the cost here is to posit GF rule The phonemic inventory is simplified at the
cost of complicating the rule component There is no clear judgement criterion But as will be
discussed in 31 GF rule may not be a costly rule to explain the specific cases as in (11) but a
natural consequence of the syllabification Further this reveals some characteristics of
syllabification in this language
In (12) we can see that there are restrictions in the range of data First we can see that all
the stem final vowels are high vowels and all the suffix-initial vowels are non-round vowels
Further there are suffixes with unspecified vowels such as IVkl (a passive marker) IVrl (an
applicative marker) IVsxl (a causative or an efferential marker) The unspecified vowels are realized
as lei if the preceding vowel is [-high -low] and as Iii if the preceding vowel is [ahigh -alow]
But if the stem ends in a vowel the unspecified vowel is realized as lei in spite of the fact that stem
3) I ran through Diocese of Mzuzu (1970) and all the data I have but there are no examples of morphemes of these kinds
-329shy
final vowels are always [s-high -low] as shown in (13)
(13) a ku (V-marker) -Ii (eat) - Vk (passive) - a (FV) --gt kulyeka (to be eaten) b ku(V-marker) - mu(drink) - Vsxtcausative) - a(FV) --gt kumwesxa (to make drink)
(13) ensures that the initial vowel of the suffix is always of greater sonority Therefore
stem final vowels are always less sonorous than the suffix initial vowels Now we can find one
thing in common between the data of (11) and (12) except (12f) and (12g)
We may therefore posit the following informal glide formation rule
(14) Glide formation (GF) Vl --gt [-syllabic]_V2
(condition V2 is not less sonorous than V1 )41
Now lets turn to (12f) and (12g) In these cases V1 is more sonorous than V2 Given
condition (6) and rule (14) on the one hand and the absence of vowel deletion on the other the
only way to resolve the ill-formed form seems to insert a glide in between two vowels At this
point it is not clear whether the inserted glide is y or w Though we only have the examples of
y-insertion the rule will be formulated to accomodate w so that we can use GI for discussions in
the following subsections
(15) Glide insertion (GI)
o --gt [~~~~~~~J
VI V2 [a round]
a round (Condition V1 is more sonorous than V2)
This rule inserts y before e and a and w before 051 One problem that should be
4)1 simply followed Jesperson (1932 191) to say that open vowels are more sonorous than closed vowels
5) Chitumbuka has five vowel system Its phonemic vowels are le iI lui lei 101
-330shy
7 Hiatus Resolving in Chitumbuka
pointed out now is the nature of the inserted segment y or WI These are not in the phonemic
inventory of Chitumbuka as discussed earlier Insertion of sounds which are not phonemic is the
characteristic of the phonetic rules that do not participate in the phonological derivation This may
mean that there should be glides in the phonemic inventory contrary to the basic assumption here
or that GI is formulated in the wrong way We will postpone the discussion to 31 and stick to GI
rule for the purpose of discussion in the present section
The GF in (14) and GI in (15) can take care of all the data of morpheme-internal vowel
clashes as well as the vowel sequence cause by the morpheme concatenation of V-final stems and V-
initial affixes
22 Between V-final affixes and V-initial stems
Another very productive source of vowel clashes is the concatenation of agreement markers
with V-initial stems In Chitumbuka nouns adjectives and verbs carry agreement markers and all
the agreement markers are V-final
(16) Agreement marker and V-initial nouns a mn shy ini mwini (owner) b mu - anice mwanice (child) c vi - aro vyaro (lands) d In - izsa Iwizxa (economy) e chi - aro charo (land) f va - ana vana (kids) g mn - ono mono mnwono (creel) h mn shy ova mova (beer)
(17) Agreement marker and possessive pronouns a vi - ane vyane (mine) b u - ane wane (mine) c tu - ako twako (yours-sg) d li - ake lake (hishers) e zi - awo zawo (theirs) f va - ithu vithu (ours) g va - ako vako (hishers)
-331shy
h li - inu inu (yours-pl)
Before going into any detail I should say that (16g) and (16h) may be problematic GF can
be applied but (16h) does not show the form with a glide Vail (197240) observed that mu-
alternates with m or syllabic m I will simply assume that this morphemic alternation is
responsible for the variations in (16g)61 It is not the application of different rules but the
coexistence of mu- and m- form that results in the two different surface forms in (16g)
We can make very interesting observations here the given data except for (16g) are similar
to (11) and (12) in that there is no variant form and in that there is a glide formation process One
difference is that no case of VD is witnessed in (11) and (12) while the data in (16) and (17)
presents very productive use of VD and do not show any instances of GI
All the examples presented show that when VI and V2 get together it is always the VI that
is deleted V2 remains undeleted This may not necessary mean that V2 triggers vowel deletion To
go deeper into VD phenomenon lets first consider environments for vowel deletion
(18) Vowel deletion environments a VI is less sonorous than V2 (16e) (16h) (l7d) (17e) b VI is more sonorous than V2 (17f) c VI andVj are identical (16f) (17g) (17h)
Apparently (18) covers almost all the possible combinations except i-u and u-i
sequences First of all we can see that (18a) shares similar environments with GF in (14) And
actually GF applies to such examples as in (16c) (16d) (17b) and (17e) Just saying that both GF
and VD are optional (cf Mphande (1989121)) and that they share same the same environments
6) I think that m- might have come from mu- through u-deletion This may be a historical change since it is very difficult to find the synchronic evidence that m- is related to mu- as in the case of mpaka (until) Then we may conclude that both forms are available for (16g) but only mshy is available for (16h)
-332~
9 Hiatus Resolving in Chitumbuka
does not explain why there is no variation
One may say that the difference in boundary type results in the application of the different
rules and add the boundary in the rule-description accordingly This can be illustrated as in (19)
(19) a ku - Ii - a --gt kulya b Ii - ake --gt lake
Though (19a) and (19b) shows the same sequence of i-a IIi in (19a) is a stem while it is
a prefix in (19b) And lal in (19a) is a grammatical suffix and lal in (19b) is a part of a lexical
morpheme In other words (19a) is suffixation and (19b) is prefixation By assigning different
boundaries we can explain why we have GI only in (11) and (12) and not in (16) and (17)
One example of the boundary functioning in conditioning rules may be found in English g-
deletion as in Hyman (1976197) According to them there are two different I-erl suffixes in
English The first one is introduced with word boundary as in singer Another is introduced
with morpheme boundary + as in long-er and English has a g-deletion rule that goes as under
(20) g-deletion in English g --gt 0N_ (N a velar nasal)
This rule with the boundary differentiation and with the specification of the boundary type
deletes IgI in singer but not in longer
The same strategy can be employed here We may assume that the stems are introduced
with one type of boundary say and the suffixes with another + we may represent (19) as
(21)7
7)1 will simply assume that suffixes are introduced with + boundary and prefixes are introduced with boundary But I will continue to use - to mark the morpheme break The boundary differentiation according to Aronoff (1976) means that suffixation preceded prefixation
-- 333shy
(21) aku li + a b li aka
Now we can formulate the following VD rule
(22) Vowel deletion Vi --gt 0 I Vz
This seems to be a reasonable approach I believe this may be partly true for (17) cases
But we need something else This does not help explaining (16) because we have to explain why
GF instead of VD is applied to such examples as (16a) to (16d) and (17a) to (17c)
One way out from this dilemma is to stick to glide formation for (16) then make a rule that
deletes the glide What is meant by this is that it is not VD but GF followed by GD in (16) as
shown in (23)
(23) li - ake --(GF)--gt lyake --(GD)--gt lake
This process is motivated by the fact that in all the cases of VD in GF environments Iii is
preceded by a coronal consonant as shown in (16e) (17d) and (17e) Notice also that the recent
works on features such as Sagey (1986) McCarthy (1988) and many others claim that palatal
glides have [coronal] feature Given this we may formulate the following GD rule
(24) Glide Deletion [+coronal] --gt 0 I [+ coronal]
The rule is maximally generalized We dont have to specify whether the sound deleted is a
glide The syllable structure given in section 1 will naturally predict that it is a glide since in C1C2
consequence the Cz must be a glide But still the boundary is necessary to prevent GD to such
words as ku li + a
Now lets take a look at (17) cases Here again we find that the environment of VD is
-334~
Hiatus Resolving in Chitumbuka 11
very similar to GI in (15) We will have to resort to the boundary distinction as proposed in
discussion with the sonority condition as in (25)
(25) Vowel deletion V1--gt0_V2 (conditionV1 is not less sonorous than V2)
The condition does not necessarily require a higher sonority for the first vowel This
reflects an effort to explain (17g) (17h) cases together with (17f) This approach however has
one immediate problem of explaining (16a) and (16d) VD as in (25) can apply to these cases but
mini and Tizxa are ill-formed One major observational difference of the examples to which
VD is applied is the roundness feature of the two vowels When two vowels have the same
roundness feature and the same height the first one deletes And GF applies when the two vowels
are of the same height (and therefore the same sonority) but differ in roundness This problem can
be solved by adding roundness agreement condition to VD as in (26)81
(26) Vowel deletion - revised VI __ gt 0 __ V2
[a round] [a round] (conditionVIis not less sonorous than V2)
23 Between V-final affixes and V-initial Verbs
There are five different types of verb prefixes They are Iku-I Iliku-I Ngu Izamu-I Ika-I
and le-I I called these prefixes not agreement markers The reason is two-fold first they have their
own meanings These prefixes express tenses or aspects of the verb And unlike agreement
8 lAnother possibility that is imposing roundness disagreement condition on the GF rule makes a wrong prediction that lyenge in (12c) and kumanyeka in (12d) should not undergo GF which is contradictory with the given data
9 )The followings are the general meaning of these prefixes a ku- Present tense present progressive or habitual aspect of the present
-335shy
12 7j-iilfHl ~1lyenlt(1993 2)
markers in 21 or 22 they do not carry grammatical information Further we find that agreement
markers are added before these prefixes in the sentences
Lets begin the discussion by looking at the following data
(27) Prefixes with final lui and V-initial verbs a ku shy amba kwamba kuyamba (to start) b ku - ambuka kwambuka kuyambuka (to cross) c ku - ata kwata kuyata (to create) d ku shy enda kwenda kuyenda (to walk) e ku shy eNga kweNga kuyeNga (to melt) f ku - imba kwimba kuyimba (to sing) g ku - ima kwima kuyima (to stand) h ku - iba kwiba (to steal) I ku - iza kw iza (to come) J ku shy ocha kocha kuwocha (to burn) k l
ku - opa ku - ona
kopa ku wopaku tJY(1Y~- Hu
(to fear) in ~f~Ofmiddot
ffi ku -ma KU wura (to fear) n ku - umba kuwurnba (to mold something) o ku - undira kuwundira (to bury)
(28)Tense or aspect marker and V-initial verbs a ka - imba kimba kayimba (sang) b ka - iba kiba (stole) c ka - enda kenda kayenda (walked) d ka - opa kawopa (feared) e ka - ona kawona (saw) f ka - undira kawundira (buried) g ka - amba kamba kayamba (started)
The first thing readily noticeable in the given data is the existence of different alternatives in
b liku- Perfective the state of verb or the result of activity in the past is carried on to the present
c zamu- Future tense d ka- distant past (activity in the past) e kaNgu-habitual activity in the past f a- Activity in the immediate past g aNgu- Duration of activity that ended in the immediate past
lO)Though all the examples given here have Iku-I prefix the exactly same phonological changes are observed with all other prefixes that end in lui such as INgu-l liku-j and zamu-z All the V-initial verbs elicited so far are given here
-336shy
Hiatus Resolving in Chitumbuka 13
most of the cases Lets first begin with the apparently exceptional cases that do not have
variations Our Chitumbuka informant was not sure whether the stems of (27h) and (27i) begin
with Iii or not On a couple of occasions he said that to steal is kuba And even when he
pronounced kwiba he still maintained that the stem is Ibal instead of iba Though without any
sufficient supporting data I take this as the evidence that the root form of to steal has lost the
alternations and fossilized into kwiba and the same might be true for (27i)
Next it is very strange that only (271) shows different behavior in comparison with the two
preceding examples (27j) and (27k) It may be extremely difficult to assume that (271) has lost the
alternations and fossilized into kuwona form Intuitively if it is to be fossilized the form should
have been kona a phonologically simpler form I suggest that the base form of to see is uon shy
a GF is applied to this to produce Iwonal and then there is no vowel clash when Iku-I is added
failing to meet the VD environment
(27m) (27n) and (270) look quite exceptional in that VD does not apply to theses examples
though they are in VD environments I do not have systematic explanation for these There may he
some historical change in these words III or the underlying forms may begin with luul and the GF
changes the first lui into glide before tense or aspect markers are added 121 or they are just
exceptional At this point I will leave this as an open question
The GF and VD are quite straight forward There is no need to revise the rules Now we
will have to think about GI cases GI rule given in (15) does not explain the examples given in this
11)For example their friend in Chitumbuka is munyawo or rnunyalo Here we sec the alternations hetween Iwland iraquo] The same thing might have happened in (27m) (27n) and (270) words we may speculate that historically those stems begin with lraquo] And the sound weakened into the glide Iw
12)50 far no two same segments have been found to be adjacent morpheme internally However given the GF rule environment I think this idea may not be totally speculative
-- 337shy
subsection the sonority condition is not satisfied
One possible linear phonology approach may be to resort to the rule ordering We may
eliminate the sonority condition of GI and simply order GI after GF Further GI should not be
applied in (16) and (17) and therefore VD should be placed before GI hence the following order
(29) Rule ordering 1 Glide formation 2 Vowel deletion 3 Glide insertion
The major problem with this approach is the nature of the rule The data (27) (28) indicate
that all the three rules are optional But even though they are optional there is no option of not
applying any of the rules and therefore the last rule (GI) should be obligatory Further the data
(11) tells us a different story GF seems to be obligatory Otherwise we may have to say that
(1Ia) can be realized either as imwe (GF) or as imuye (GI) Moreover (17) shows that VD is
obligatory to prevent the operation of GI Here is the paradox A rule is optional for certain group
of words and obligatory for the rest This may suggest that rule based linear approach may not be
suitable for the explanation of hiatus resolving in Chitumbuka Further theoretical investigation will
be made in 32
24 Subject agreement marker and past imperfective marker la-I
All the prefixes introduced in 21 and some other personal agreement markers are used in
the sentence to satisfy the subject agreement Meanwhile there is one V-initial tense marker in
chitumbuka a- And therefore here again we expect to find vowel clashes The representative
data are given in (30)
13) If any of the rule is obligatory then we should not see the variations in the data given in (27) and (28)
-338shy
Hiatus Resolving in Chitumbuka 15
(30) Subject agreement marker and past imperfective marker a ni - aluta naluta (I walked) b u - aluta waluta (you walked) c wa - aluta waluta (he walked) d ti - aluta taluta (we went) e li - aluta laluta (It went) f vi - aluta (vyaluta) valuta (They went)
Superficially the data given here looks like those in (17) There is no GI example And the
choice between GF and VD seems to be determined by the relative sonority of the vowels
However (30f) shows that there is something different here (30) may not be the extension of
examples given in (17) The reason is twofold first the examples given here are not lexical The
agreement marker assignment should refer to the syntactic category the subject Secondly there is
an independent process of i-deletion in chitumbuka Consider following examples
(31) i-deletion and complex segment formation a ni (copula) -litali (big) --gt nilitali or nditali laquoit) is big) b ni (subject agreement)- kulya --gt nikulya or Nkhulya laquol) eat)
Here again i is deleted though it is not followed by another vowel I will tentatively
formulate the i-deletion rule as in (32)
(32) i-deletion i --gt 0_ [ (where [ is a morphological word domain)
This rule as (31a) (31 b) and (30f) shows is optional Then what if the option is not taken
in (30) examples We might expect to see nyaluta in case of example (30a) Notice that the
14 )The irrelevant morpheme breaks are not shown here aluta can be further divided into a (past imperfective) -Iut (go)- a (FV)
15 )The form in the parenthesis means that it may be used by some But the other form is far more widely used and the informant prefers to use valuta
16 )Further investigation of phonological word derivation based on the syntactic information and morphological words should be made Here only the difference of phonological and morphological words is used without mentioning how different they are
-339shy
morphological word domain contains boundary and that the agreement markers begin with
coronal sounds If GF applies then GD in (24) will eliminate nyaluta
25 Unresolved vowel clashes
Not all the vowel sequences are resolved in Chitumbuka Some representative examples are
given in (33) and (34)
(33) V-final nouns and determiners a khuni Ili --gt khuni (y)ili khunyili khunili
tree this b buku ilo --gt buku (y)ilo bukwilo
book that c chipinda ico --gt chipinda (y)ico chipindico
room this
(34) Pronominal objects and V-initial verbs 171
a u-ku-mu-at-a --gt wakumu(y)ata wakumwata (you created him)
b wa-ku-ti-at-a --gt wakuti (y)ata wakutyata (He created you)
c ni-ku-tu-och-a --gt nikuku(w)ocha nikukocha nikukwocha (I roasted them)
d mu-ku-ka-eNg-a --gt mukuka(y)eNga muku (I melted it)
This seems to tell us that the main domain of hiatus resolving is within phonological words
GF and VD do not operate on this level and only G1 optionally shows up For further discussion
the phonological word boundaries are represented with parentheses in (35)
17 )The detailed breakdowns of the strings are as under
(yOU) ~u (him) tt (create)tni (me) ku + 11 (~e) + och (bake) + a (FV) wa (heher ka (It) Ng (melt)
u (them) verb stem
subject verb pronominal agreement tense objects
-340shy
Hiatus Resolving in Chitumbuka 17
(35) a (khuni) (ili) b (niku) (tu) (ocha) c (nikuwocha) (tuyuni)
One very interesting observation here is the change in the phonological word domains
(35b) and (35c) are semantically same The object in (35b) is pronominal while (35c) has a noun
object As a result of movement two phonological words are broken into three phonological
words This is not an arbitrary boundary just to explain the lack of hiatus resolving We may find
a very similar situation in English Intonational Phrase parsing
(36) a (As you know) (the tall boy is a genius) b (The tall boy) (as you know) (is a genius)
Nespor and Vogels (1986189) intonational phrase formation rule produces (36a) as well
as (36b) The syntax dependency of the phonological word formation is partly responsible for the
difference between (35b) and (35c)
Further hiatus is not the only rule sensitive to boundary difference in Chitumbuka We can
find that i-deletion and complex segment formation discussed in the preceding subsection shows
the same phenomenon Consider the following example (parentheses indicate the phonological
word domain)
(37) a (niku) (mu) (temwa) I love him Nkhu
b (uku) (ni)(temwa) You love me nthe
c (khunilClllno) the tree here mpha
One final observation is that our rules so far can not explain why glides can be inserted in
the examples of (33) and (34) We have already pointed out the problem of GI in 21 The
additional complication is that even if we posit glides as a phoneme we still have to revise GIor
-341shy
make another one to explain post-lexical GL This will be addressed in 32
3 Syllable phonology of hiatus resol ving
In the discussion in section 2 some of the problems in the linear analysis were detected
(What are they ) In this section I will go over them and reanalyze the data with the help
of syllable structure to see whether we can answer some of the questions left open McCarthy and
Prince (1986) style of moraic syllable structure will be employed for our discussion in this section
The basic syllable structure of Chitumbuka is given in (38)
(38) Syllable structure in Chitumbuka a a b a c a
~~((r r1T ( (I I 1 All the syllables are mora final 2 All the syllables are mono-moraic
This is not the template that is required in the underlying form as will be discussed
Therefore the non-moraic tier may not have to be satisfied In other words though the structure
shows two pre-moraic consonant slots these can be left empty The two branches simple mean the
maximum number of consonants that can come before a vowel
18)1n the post lexical level this structural condition may not hold In Chitumbuka vowels in the stressed syllable is lengthened The long vowel is represented with bimoraic syllable structure It is not uncommon that some of the rules or conditions are applied only on the lexical level but not on the post lexical level such as extrametricality stress assignment and even VD and GF we talked about in the preceding section
- 342-shy
Hiatus Resolving in Chitumbuka 19
31 Syllabification
Contrary to the claim that all the segments are already syllabified in the base form (cf
Selkirk (1982raquo I assume that syllable structure is assigned through the syllabification process
The syllabification principles in Chitumbuka are given under
(39) Syllabification in Chitumbuka i) Morafication A segment projects a mora if it is more sonorous than the following
segrnent ii) Syllable projection Each mora projects a syllable node iii) Segment incorporation Unsyllabified segments are incorporated into syllables
Now lets take a look at the following sample syllabification
(40) a
0 0 00
I (iii)I(ii) --(i) f1 f1 f1 f1 ~ rlrOI I
~
I I n i u mb a n 1 u mb a n 1 u mb a n i u mb a
b
0 o 0 0 o 0
I I I (iW (~Of1 f1 f1 f1 ~ ~ I I r~r I I I I I ch i m a i ch i m a ch i m a ch i m a i
Notice that we do not need GF as a rule the same result of GF application is achieved as a
19)1 suppose that morafication principle is universally dependent upon the sonority hierarchy There are three different strategies in using sonority in syllabifications
i) y is moraic in xy if it is more sonorous than x ii) y is moraic in yz if it is more sonorous than z iii) y is moraic if its sonority is greater than N
The language in the world employ one or two all all of these in the syllabification and language may slightly vary in setting up the value of N in (iii)
-- 343shy
natural consequence of syllabification In 21 we obtain the simplicity of phonemic inventory at
the cost of introducing GF thereby complicating the rule component The environment of GF is
reflected in the syllabification principle Of course syllabification can not be an additional burden
the language needs syllable structure anyhow for example for the application of many phonological
rules and even for utterance
The nature of glide sounds are clearly reflected in the syllable structure If a vowel is
dominated by a mora it will be interpreted as a full vowel But if it is dominated by the syllable
node directly it will surface as a glide
32 Syllabification domain
The syllabification principles introduced in 31 have presupposed that the string of segments
are not syllabified in the underlying representation Suppose every morpheme is syllabified in the
deep structure and consider the following example of word formation
(41)
a aa a
~ (I (noa k 0 V 1 a k 0
Now we will have to make a rule for mora deletion which will be almost the repetition of
the syllabification principle Syllabification process in the course of derivation can get rid of the
redundant repetition of syllable structural description as well as save the later-on application of mora
deletion
I assume that the syllabification takes place in the first derived domain In other words
segments are syllabified after the initial word formation process Before going into the detailed
-344shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
(9) a ku (V-marker) - end (walk) - a (FV) --gt kuyenda (to walk) b ka (past) - amb (start) - a (FV) --gt kayamba (to start)
Some of the forms may take more than one resolving processes resulting in variations of
different forms as in (10)
(10) a ku (V-marker) - op (fear) - a (FV) --gt kopa (VD) kuwopa (GI) b ka (past) - im (stand) +a (FV) --gt kima (VD) kayima (GI)
In this section data of two consecutive vowels are presented and the changes of different
types will be discussed
21 Morpheme-internal and between stems and suffixes
First we will begin with morpheme internal vowel clashes and also the case where a stem
ends in a vowel and the following suffix begins with another vowel
(11)Morpheme internal vowel sequence ex~mple GF VD GI gloss
a imue imwe (you-pI) b niama nyama (meat) c ku-puan-a ku-pwan-a (to shatter) d uoko woko (arm) e nai nayi (four) f chi-mai chimayi (knife) g ku-iouoi-a ku-yowoy-a (to speak)
(12)V-final stem and V-initial suffix a ku-Ii - esx+a kulyesxa (to eat) b c
ku-temu - a Ii - eNge
kutemwa lyeNge
(to love) (will eat)
d ku-mani - eka kumanyeka (to know)
The interesting thing readily noticeable in above data is the lack of variation Each of the
examples shows either GF or GI but not both and no example of VD is witnessed Ivith (11) I
think that the absence of vowel deletion process is only natural because if vowel deletion is applied
-328shy
Hiatus Resolving in Chitumbuka 5
within a morpheme it is usually unrecoverable And therefore the native speakers will take the
form with vowel deletion change if any as the underlying form
This explanation gives rise to a possibility that there may not be any process at all be it GF
or GI in (11) For example we may say that the underlying form of (11b) is just nyumba or
Ijumba for that matter My assumption that the underlying base forms have vowels is based on
the following considerations Positing underlying vowels instead of glides does not make wrong
predictions In other words it never produces wrong phonetic forms and it does not fail to
produces all the correct forms as will be clearly illustrated in the course of the discussion In the
mean time this assumption gets rid of three possible phonemes Iy Iwl and IJ1 without losing any
explanatory power and thereby achieves the economy of phonological system
Of course the cost here is to posit GF rule The phonemic inventory is simplified at the
cost of complicating the rule component There is no clear judgement criterion But as will be
discussed in 31 GF rule may not be a costly rule to explain the specific cases as in (11) but a
natural consequence of the syllabification Further this reveals some characteristics of
syllabification in this language
In (12) we can see that there are restrictions in the range of data First we can see that all
the stem final vowels are high vowels and all the suffix-initial vowels are non-round vowels
Further there are suffixes with unspecified vowels such as IVkl (a passive marker) IVrl (an
applicative marker) IVsxl (a causative or an efferential marker) The unspecified vowels are realized
as lei if the preceding vowel is [-high -low] and as Iii if the preceding vowel is [ahigh -alow]
But if the stem ends in a vowel the unspecified vowel is realized as lei in spite of the fact that stem
3) I ran through Diocese of Mzuzu (1970) and all the data I have but there are no examples of morphemes of these kinds
-329shy
final vowels are always [s-high -low] as shown in (13)
(13) a ku (V-marker) -Ii (eat) - Vk (passive) - a (FV) --gt kulyeka (to be eaten) b ku(V-marker) - mu(drink) - Vsxtcausative) - a(FV) --gt kumwesxa (to make drink)
(13) ensures that the initial vowel of the suffix is always of greater sonority Therefore
stem final vowels are always less sonorous than the suffix initial vowels Now we can find one
thing in common between the data of (11) and (12) except (12f) and (12g)
We may therefore posit the following informal glide formation rule
(14) Glide formation (GF) Vl --gt [-syllabic]_V2
(condition V2 is not less sonorous than V1 )41
Now lets turn to (12f) and (12g) In these cases V1 is more sonorous than V2 Given
condition (6) and rule (14) on the one hand and the absence of vowel deletion on the other the
only way to resolve the ill-formed form seems to insert a glide in between two vowels At this
point it is not clear whether the inserted glide is y or w Though we only have the examples of
y-insertion the rule will be formulated to accomodate w so that we can use GI for discussions in
the following subsections
(15) Glide insertion (GI)
o --gt [~~~~~~~J
VI V2 [a round]
a round (Condition V1 is more sonorous than V2)
This rule inserts y before e and a and w before 051 One problem that should be
4)1 simply followed Jesperson (1932 191) to say that open vowels are more sonorous than closed vowels
5) Chitumbuka has five vowel system Its phonemic vowels are le iI lui lei 101
-330shy
7 Hiatus Resolving in Chitumbuka
pointed out now is the nature of the inserted segment y or WI These are not in the phonemic
inventory of Chitumbuka as discussed earlier Insertion of sounds which are not phonemic is the
characteristic of the phonetic rules that do not participate in the phonological derivation This may
mean that there should be glides in the phonemic inventory contrary to the basic assumption here
or that GI is formulated in the wrong way We will postpone the discussion to 31 and stick to GI
rule for the purpose of discussion in the present section
The GF in (14) and GI in (15) can take care of all the data of morpheme-internal vowel
clashes as well as the vowel sequence cause by the morpheme concatenation of V-final stems and V-
initial affixes
22 Between V-final affixes and V-initial stems
Another very productive source of vowel clashes is the concatenation of agreement markers
with V-initial stems In Chitumbuka nouns adjectives and verbs carry agreement markers and all
the agreement markers are V-final
(16) Agreement marker and V-initial nouns a mn shy ini mwini (owner) b mu - anice mwanice (child) c vi - aro vyaro (lands) d In - izsa Iwizxa (economy) e chi - aro charo (land) f va - ana vana (kids) g mn - ono mono mnwono (creel) h mn shy ova mova (beer)
(17) Agreement marker and possessive pronouns a vi - ane vyane (mine) b u - ane wane (mine) c tu - ako twako (yours-sg) d li - ake lake (hishers) e zi - awo zawo (theirs) f va - ithu vithu (ours) g va - ako vako (hishers)
-331shy
h li - inu inu (yours-pl)
Before going into any detail I should say that (16g) and (16h) may be problematic GF can
be applied but (16h) does not show the form with a glide Vail (197240) observed that mu-
alternates with m or syllabic m I will simply assume that this morphemic alternation is
responsible for the variations in (16g)61 It is not the application of different rules but the
coexistence of mu- and m- form that results in the two different surface forms in (16g)
We can make very interesting observations here the given data except for (16g) are similar
to (11) and (12) in that there is no variant form and in that there is a glide formation process One
difference is that no case of VD is witnessed in (11) and (12) while the data in (16) and (17)
presents very productive use of VD and do not show any instances of GI
All the examples presented show that when VI and V2 get together it is always the VI that
is deleted V2 remains undeleted This may not necessary mean that V2 triggers vowel deletion To
go deeper into VD phenomenon lets first consider environments for vowel deletion
(18) Vowel deletion environments a VI is less sonorous than V2 (16e) (16h) (l7d) (17e) b VI is more sonorous than V2 (17f) c VI andVj are identical (16f) (17g) (17h)
Apparently (18) covers almost all the possible combinations except i-u and u-i
sequences First of all we can see that (18a) shares similar environments with GF in (14) And
actually GF applies to such examples as in (16c) (16d) (17b) and (17e) Just saying that both GF
and VD are optional (cf Mphande (1989121)) and that they share same the same environments
6) I think that m- might have come from mu- through u-deletion This may be a historical change since it is very difficult to find the synchronic evidence that m- is related to mu- as in the case of mpaka (until) Then we may conclude that both forms are available for (16g) but only mshy is available for (16h)
-332~
9 Hiatus Resolving in Chitumbuka
does not explain why there is no variation
One may say that the difference in boundary type results in the application of the different
rules and add the boundary in the rule-description accordingly This can be illustrated as in (19)
(19) a ku - Ii - a --gt kulya b Ii - ake --gt lake
Though (19a) and (19b) shows the same sequence of i-a IIi in (19a) is a stem while it is
a prefix in (19b) And lal in (19a) is a grammatical suffix and lal in (19b) is a part of a lexical
morpheme In other words (19a) is suffixation and (19b) is prefixation By assigning different
boundaries we can explain why we have GI only in (11) and (12) and not in (16) and (17)
One example of the boundary functioning in conditioning rules may be found in English g-
deletion as in Hyman (1976197) According to them there are two different I-erl suffixes in
English The first one is introduced with word boundary as in singer Another is introduced
with morpheme boundary + as in long-er and English has a g-deletion rule that goes as under
(20) g-deletion in English g --gt 0N_ (N a velar nasal)
This rule with the boundary differentiation and with the specification of the boundary type
deletes IgI in singer but not in longer
The same strategy can be employed here We may assume that the stems are introduced
with one type of boundary say and the suffixes with another + we may represent (19) as
(21)7
7)1 will simply assume that suffixes are introduced with + boundary and prefixes are introduced with boundary But I will continue to use - to mark the morpheme break The boundary differentiation according to Aronoff (1976) means that suffixation preceded prefixation
-- 333shy
(21) aku li + a b li aka
Now we can formulate the following VD rule
(22) Vowel deletion Vi --gt 0 I Vz
This seems to be a reasonable approach I believe this may be partly true for (17) cases
But we need something else This does not help explaining (16) because we have to explain why
GF instead of VD is applied to such examples as (16a) to (16d) and (17a) to (17c)
One way out from this dilemma is to stick to glide formation for (16) then make a rule that
deletes the glide What is meant by this is that it is not VD but GF followed by GD in (16) as
shown in (23)
(23) li - ake --(GF)--gt lyake --(GD)--gt lake
This process is motivated by the fact that in all the cases of VD in GF environments Iii is
preceded by a coronal consonant as shown in (16e) (17d) and (17e) Notice also that the recent
works on features such as Sagey (1986) McCarthy (1988) and many others claim that palatal
glides have [coronal] feature Given this we may formulate the following GD rule
(24) Glide Deletion [+coronal] --gt 0 I [+ coronal]
The rule is maximally generalized We dont have to specify whether the sound deleted is a
glide The syllable structure given in section 1 will naturally predict that it is a glide since in C1C2
consequence the Cz must be a glide But still the boundary is necessary to prevent GD to such
words as ku li + a
Now lets take a look at (17) cases Here again we find that the environment of VD is
-334~
Hiatus Resolving in Chitumbuka 11
very similar to GI in (15) We will have to resort to the boundary distinction as proposed in
discussion with the sonority condition as in (25)
(25) Vowel deletion V1--gt0_V2 (conditionV1 is not less sonorous than V2)
The condition does not necessarily require a higher sonority for the first vowel This
reflects an effort to explain (17g) (17h) cases together with (17f) This approach however has
one immediate problem of explaining (16a) and (16d) VD as in (25) can apply to these cases but
mini and Tizxa are ill-formed One major observational difference of the examples to which
VD is applied is the roundness feature of the two vowels When two vowels have the same
roundness feature and the same height the first one deletes And GF applies when the two vowels
are of the same height (and therefore the same sonority) but differ in roundness This problem can
be solved by adding roundness agreement condition to VD as in (26)81
(26) Vowel deletion - revised VI __ gt 0 __ V2
[a round] [a round] (conditionVIis not less sonorous than V2)
23 Between V-final affixes and V-initial Verbs
There are five different types of verb prefixes They are Iku-I Iliku-I Ngu Izamu-I Ika-I
and le-I I called these prefixes not agreement markers The reason is two-fold first they have their
own meanings These prefixes express tenses or aspects of the verb And unlike agreement
8 lAnother possibility that is imposing roundness disagreement condition on the GF rule makes a wrong prediction that lyenge in (12c) and kumanyeka in (12d) should not undergo GF which is contradictory with the given data
9 )The followings are the general meaning of these prefixes a ku- Present tense present progressive or habitual aspect of the present
-335shy
12 7j-iilfHl ~1lyenlt(1993 2)
markers in 21 or 22 they do not carry grammatical information Further we find that agreement
markers are added before these prefixes in the sentences
Lets begin the discussion by looking at the following data
(27) Prefixes with final lui and V-initial verbs a ku shy amba kwamba kuyamba (to start) b ku - ambuka kwambuka kuyambuka (to cross) c ku - ata kwata kuyata (to create) d ku shy enda kwenda kuyenda (to walk) e ku shy eNga kweNga kuyeNga (to melt) f ku - imba kwimba kuyimba (to sing) g ku - ima kwima kuyima (to stand) h ku - iba kwiba (to steal) I ku - iza kw iza (to come) J ku shy ocha kocha kuwocha (to burn) k l
ku - opa ku - ona
kopa ku wopaku tJY(1Y~- Hu
(to fear) in ~f~Ofmiddot
ffi ku -ma KU wura (to fear) n ku - umba kuwurnba (to mold something) o ku - undira kuwundira (to bury)
(28)Tense or aspect marker and V-initial verbs a ka - imba kimba kayimba (sang) b ka - iba kiba (stole) c ka - enda kenda kayenda (walked) d ka - opa kawopa (feared) e ka - ona kawona (saw) f ka - undira kawundira (buried) g ka - amba kamba kayamba (started)
The first thing readily noticeable in the given data is the existence of different alternatives in
b liku- Perfective the state of verb or the result of activity in the past is carried on to the present
c zamu- Future tense d ka- distant past (activity in the past) e kaNgu-habitual activity in the past f a- Activity in the immediate past g aNgu- Duration of activity that ended in the immediate past
lO)Though all the examples given here have Iku-I prefix the exactly same phonological changes are observed with all other prefixes that end in lui such as INgu-l liku-j and zamu-z All the V-initial verbs elicited so far are given here
-336shy
Hiatus Resolving in Chitumbuka 13
most of the cases Lets first begin with the apparently exceptional cases that do not have
variations Our Chitumbuka informant was not sure whether the stems of (27h) and (27i) begin
with Iii or not On a couple of occasions he said that to steal is kuba And even when he
pronounced kwiba he still maintained that the stem is Ibal instead of iba Though without any
sufficient supporting data I take this as the evidence that the root form of to steal has lost the
alternations and fossilized into kwiba and the same might be true for (27i)
Next it is very strange that only (271) shows different behavior in comparison with the two
preceding examples (27j) and (27k) It may be extremely difficult to assume that (271) has lost the
alternations and fossilized into kuwona form Intuitively if it is to be fossilized the form should
have been kona a phonologically simpler form I suggest that the base form of to see is uon shy
a GF is applied to this to produce Iwonal and then there is no vowel clash when Iku-I is added
failing to meet the VD environment
(27m) (27n) and (270) look quite exceptional in that VD does not apply to theses examples
though they are in VD environments I do not have systematic explanation for these There may he
some historical change in these words III or the underlying forms may begin with luul and the GF
changes the first lui into glide before tense or aspect markers are added 121 or they are just
exceptional At this point I will leave this as an open question
The GF and VD are quite straight forward There is no need to revise the rules Now we
will have to think about GI cases GI rule given in (15) does not explain the examples given in this
11)For example their friend in Chitumbuka is munyawo or rnunyalo Here we sec the alternations hetween Iwland iraquo] The same thing might have happened in (27m) (27n) and (270) words we may speculate that historically those stems begin with lraquo] And the sound weakened into the glide Iw
12)50 far no two same segments have been found to be adjacent morpheme internally However given the GF rule environment I think this idea may not be totally speculative
-- 337shy
subsection the sonority condition is not satisfied
One possible linear phonology approach may be to resort to the rule ordering We may
eliminate the sonority condition of GI and simply order GI after GF Further GI should not be
applied in (16) and (17) and therefore VD should be placed before GI hence the following order
(29) Rule ordering 1 Glide formation 2 Vowel deletion 3 Glide insertion
The major problem with this approach is the nature of the rule The data (27) (28) indicate
that all the three rules are optional But even though they are optional there is no option of not
applying any of the rules and therefore the last rule (GI) should be obligatory Further the data
(11) tells us a different story GF seems to be obligatory Otherwise we may have to say that
(1Ia) can be realized either as imwe (GF) or as imuye (GI) Moreover (17) shows that VD is
obligatory to prevent the operation of GI Here is the paradox A rule is optional for certain group
of words and obligatory for the rest This may suggest that rule based linear approach may not be
suitable for the explanation of hiatus resolving in Chitumbuka Further theoretical investigation will
be made in 32
24 Subject agreement marker and past imperfective marker la-I
All the prefixes introduced in 21 and some other personal agreement markers are used in
the sentence to satisfy the subject agreement Meanwhile there is one V-initial tense marker in
chitumbuka a- And therefore here again we expect to find vowel clashes The representative
data are given in (30)
13) If any of the rule is obligatory then we should not see the variations in the data given in (27) and (28)
-338shy
Hiatus Resolving in Chitumbuka 15
(30) Subject agreement marker and past imperfective marker a ni - aluta naluta (I walked) b u - aluta waluta (you walked) c wa - aluta waluta (he walked) d ti - aluta taluta (we went) e li - aluta laluta (It went) f vi - aluta (vyaluta) valuta (They went)
Superficially the data given here looks like those in (17) There is no GI example And the
choice between GF and VD seems to be determined by the relative sonority of the vowels
However (30f) shows that there is something different here (30) may not be the extension of
examples given in (17) The reason is twofold first the examples given here are not lexical The
agreement marker assignment should refer to the syntactic category the subject Secondly there is
an independent process of i-deletion in chitumbuka Consider following examples
(31) i-deletion and complex segment formation a ni (copula) -litali (big) --gt nilitali or nditali laquoit) is big) b ni (subject agreement)- kulya --gt nikulya or Nkhulya laquol) eat)
Here again i is deleted though it is not followed by another vowel I will tentatively
formulate the i-deletion rule as in (32)
(32) i-deletion i --gt 0_ [ (where [ is a morphological word domain)
This rule as (31a) (31 b) and (30f) shows is optional Then what if the option is not taken
in (30) examples We might expect to see nyaluta in case of example (30a) Notice that the
14 )The irrelevant morpheme breaks are not shown here aluta can be further divided into a (past imperfective) -Iut (go)- a (FV)
15 )The form in the parenthesis means that it may be used by some But the other form is far more widely used and the informant prefers to use valuta
16 )Further investigation of phonological word derivation based on the syntactic information and morphological words should be made Here only the difference of phonological and morphological words is used without mentioning how different they are
-339shy
morphological word domain contains boundary and that the agreement markers begin with
coronal sounds If GF applies then GD in (24) will eliminate nyaluta
25 Unresolved vowel clashes
Not all the vowel sequences are resolved in Chitumbuka Some representative examples are
given in (33) and (34)
(33) V-final nouns and determiners a khuni Ili --gt khuni (y)ili khunyili khunili
tree this b buku ilo --gt buku (y)ilo bukwilo
book that c chipinda ico --gt chipinda (y)ico chipindico
room this
(34) Pronominal objects and V-initial verbs 171
a u-ku-mu-at-a --gt wakumu(y)ata wakumwata (you created him)
b wa-ku-ti-at-a --gt wakuti (y)ata wakutyata (He created you)
c ni-ku-tu-och-a --gt nikuku(w)ocha nikukocha nikukwocha (I roasted them)
d mu-ku-ka-eNg-a --gt mukuka(y)eNga muku (I melted it)
This seems to tell us that the main domain of hiatus resolving is within phonological words
GF and VD do not operate on this level and only G1 optionally shows up For further discussion
the phonological word boundaries are represented with parentheses in (35)
17 )The detailed breakdowns of the strings are as under
(yOU) ~u (him) tt (create)tni (me) ku + 11 (~e) + och (bake) + a (FV) wa (heher ka (It) Ng (melt)
u (them) verb stem
subject verb pronominal agreement tense objects
-340shy
Hiatus Resolving in Chitumbuka 17
(35) a (khuni) (ili) b (niku) (tu) (ocha) c (nikuwocha) (tuyuni)
One very interesting observation here is the change in the phonological word domains
(35b) and (35c) are semantically same The object in (35b) is pronominal while (35c) has a noun
object As a result of movement two phonological words are broken into three phonological
words This is not an arbitrary boundary just to explain the lack of hiatus resolving We may find
a very similar situation in English Intonational Phrase parsing
(36) a (As you know) (the tall boy is a genius) b (The tall boy) (as you know) (is a genius)
Nespor and Vogels (1986189) intonational phrase formation rule produces (36a) as well
as (36b) The syntax dependency of the phonological word formation is partly responsible for the
difference between (35b) and (35c)
Further hiatus is not the only rule sensitive to boundary difference in Chitumbuka We can
find that i-deletion and complex segment formation discussed in the preceding subsection shows
the same phenomenon Consider the following example (parentheses indicate the phonological
word domain)
(37) a (niku) (mu) (temwa) I love him Nkhu
b (uku) (ni)(temwa) You love me nthe
c (khunilClllno) the tree here mpha
One final observation is that our rules so far can not explain why glides can be inserted in
the examples of (33) and (34) We have already pointed out the problem of GI in 21 The
additional complication is that even if we posit glides as a phoneme we still have to revise GIor
-341shy
make another one to explain post-lexical GL This will be addressed in 32
3 Syllable phonology of hiatus resol ving
In the discussion in section 2 some of the problems in the linear analysis were detected
(What are they ) In this section I will go over them and reanalyze the data with the help
of syllable structure to see whether we can answer some of the questions left open McCarthy and
Prince (1986) style of moraic syllable structure will be employed for our discussion in this section
The basic syllable structure of Chitumbuka is given in (38)
(38) Syllable structure in Chitumbuka a a b a c a
~~((r r1T ( (I I 1 All the syllables are mora final 2 All the syllables are mono-moraic
This is not the template that is required in the underlying form as will be discussed
Therefore the non-moraic tier may not have to be satisfied In other words though the structure
shows two pre-moraic consonant slots these can be left empty The two branches simple mean the
maximum number of consonants that can come before a vowel
18)1n the post lexical level this structural condition may not hold In Chitumbuka vowels in the stressed syllable is lengthened The long vowel is represented with bimoraic syllable structure It is not uncommon that some of the rules or conditions are applied only on the lexical level but not on the post lexical level such as extrametricality stress assignment and even VD and GF we talked about in the preceding section
- 342-shy
Hiatus Resolving in Chitumbuka 19
31 Syllabification
Contrary to the claim that all the segments are already syllabified in the base form (cf
Selkirk (1982raquo I assume that syllable structure is assigned through the syllabification process
The syllabification principles in Chitumbuka are given under
(39) Syllabification in Chitumbuka i) Morafication A segment projects a mora if it is more sonorous than the following
segrnent ii) Syllable projection Each mora projects a syllable node iii) Segment incorporation Unsyllabified segments are incorporated into syllables
Now lets take a look at the following sample syllabification
(40) a
0 0 00
I (iii)I(ii) --(i) f1 f1 f1 f1 ~ rlrOI I
~
I I n i u mb a n 1 u mb a n 1 u mb a n i u mb a
b
0 o 0 0 o 0
I I I (iW (~Of1 f1 f1 f1 ~ ~ I I r~r I I I I I ch i m a i ch i m a ch i m a ch i m a i
Notice that we do not need GF as a rule the same result of GF application is achieved as a
19)1 suppose that morafication principle is universally dependent upon the sonority hierarchy There are three different strategies in using sonority in syllabifications
i) y is moraic in xy if it is more sonorous than x ii) y is moraic in yz if it is more sonorous than z iii) y is moraic if its sonority is greater than N
The language in the world employ one or two all all of these in the syllabification and language may slightly vary in setting up the value of N in (iii)
-- 343shy
natural consequence of syllabification In 21 we obtain the simplicity of phonemic inventory at
the cost of introducing GF thereby complicating the rule component The environment of GF is
reflected in the syllabification principle Of course syllabification can not be an additional burden
the language needs syllable structure anyhow for example for the application of many phonological
rules and even for utterance
The nature of glide sounds are clearly reflected in the syllable structure If a vowel is
dominated by a mora it will be interpreted as a full vowel But if it is dominated by the syllable
node directly it will surface as a glide
32 Syllabification domain
The syllabification principles introduced in 31 have presupposed that the string of segments
are not syllabified in the underlying representation Suppose every morpheme is syllabified in the
deep structure and consider the following example of word formation
(41)
a aa a
~ (I (noa k 0 V 1 a k 0
Now we will have to make a rule for mora deletion which will be almost the repetition of
the syllabification principle Syllabification process in the course of derivation can get rid of the
redundant repetition of syllable structural description as well as save the later-on application of mora
deletion
I assume that the syllabification takes place in the first derived domain In other words
segments are syllabified after the initial word formation process Before going into the detailed
-344shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
Hiatus Resolving in Chitumbuka 5
within a morpheme it is usually unrecoverable And therefore the native speakers will take the
form with vowel deletion change if any as the underlying form
This explanation gives rise to a possibility that there may not be any process at all be it GF
or GI in (11) For example we may say that the underlying form of (11b) is just nyumba or
Ijumba for that matter My assumption that the underlying base forms have vowels is based on
the following considerations Positing underlying vowels instead of glides does not make wrong
predictions In other words it never produces wrong phonetic forms and it does not fail to
produces all the correct forms as will be clearly illustrated in the course of the discussion In the
mean time this assumption gets rid of three possible phonemes Iy Iwl and IJ1 without losing any
explanatory power and thereby achieves the economy of phonological system
Of course the cost here is to posit GF rule The phonemic inventory is simplified at the
cost of complicating the rule component There is no clear judgement criterion But as will be
discussed in 31 GF rule may not be a costly rule to explain the specific cases as in (11) but a
natural consequence of the syllabification Further this reveals some characteristics of
syllabification in this language
In (12) we can see that there are restrictions in the range of data First we can see that all
the stem final vowels are high vowels and all the suffix-initial vowels are non-round vowels
Further there are suffixes with unspecified vowels such as IVkl (a passive marker) IVrl (an
applicative marker) IVsxl (a causative or an efferential marker) The unspecified vowels are realized
as lei if the preceding vowel is [-high -low] and as Iii if the preceding vowel is [ahigh -alow]
But if the stem ends in a vowel the unspecified vowel is realized as lei in spite of the fact that stem
3) I ran through Diocese of Mzuzu (1970) and all the data I have but there are no examples of morphemes of these kinds
-329shy
final vowels are always [s-high -low] as shown in (13)
(13) a ku (V-marker) -Ii (eat) - Vk (passive) - a (FV) --gt kulyeka (to be eaten) b ku(V-marker) - mu(drink) - Vsxtcausative) - a(FV) --gt kumwesxa (to make drink)
(13) ensures that the initial vowel of the suffix is always of greater sonority Therefore
stem final vowels are always less sonorous than the suffix initial vowels Now we can find one
thing in common between the data of (11) and (12) except (12f) and (12g)
We may therefore posit the following informal glide formation rule
(14) Glide formation (GF) Vl --gt [-syllabic]_V2
(condition V2 is not less sonorous than V1 )41
Now lets turn to (12f) and (12g) In these cases V1 is more sonorous than V2 Given
condition (6) and rule (14) on the one hand and the absence of vowel deletion on the other the
only way to resolve the ill-formed form seems to insert a glide in between two vowels At this
point it is not clear whether the inserted glide is y or w Though we only have the examples of
y-insertion the rule will be formulated to accomodate w so that we can use GI for discussions in
the following subsections
(15) Glide insertion (GI)
o --gt [~~~~~~~J
VI V2 [a round]
a round (Condition V1 is more sonorous than V2)
This rule inserts y before e and a and w before 051 One problem that should be
4)1 simply followed Jesperson (1932 191) to say that open vowels are more sonorous than closed vowels
5) Chitumbuka has five vowel system Its phonemic vowels are le iI lui lei 101
-330shy
7 Hiatus Resolving in Chitumbuka
pointed out now is the nature of the inserted segment y or WI These are not in the phonemic
inventory of Chitumbuka as discussed earlier Insertion of sounds which are not phonemic is the
characteristic of the phonetic rules that do not participate in the phonological derivation This may
mean that there should be glides in the phonemic inventory contrary to the basic assumption here
or that GI is formulated in the wrong way We will postpone the discussion to 31 and stick to GI
rule for the purpose of discussion in the present section
The GF in (14) and GI in (15) can take care of all the data of morpheme-internal vowel
clashes as well as the vowel sequence cause by the morpheme concatenation of V-final stems and V-
initial affixes
22 Between V-final affixes and V-initial stems
Another very productive source of vowel clashes is the concatenation of agreement markers
with V-initial stems In Chitumbuka nouns adjectives and verbs carry agreement markers and all
the agreement markers are V-final
(16) Agreement marker and V-initial nouns a mn shy ini mwini (owner) b mu - anice mwanice (child) c vi - aro vyaro (lands) d In - izsa Iwizxa (economy) e chi - aro charo (land) f va - ana vana (kids) g mn - ono mono mnwono (creel) h mn shy ova mova (beer)
(17) Agreement marker and possessive pronouns a vi - ane vyane (mine) b u - ane wane (mine) c tu - ako twako (yours-sg) d li - ake lake (hishers) e zi - awo zawo (theirs) f va - ithu vithu (ours) g va - ako vako (hishers)
-331shy
h li - inu inu (yours-pl)
Before going into any detail I should say that (16g) and (16h) may be problematic GF can
be applied but (16h) does not show the form with a glide Vail (197240) observed that mu-
alternates with m or syllabic m I will simply assume that this morphemic alternation is
responsible for the variations in (16g)61 It is not the application of different rules but the
coexistence of mu- and m- form that results in the two different surface forms in (16g)
We can make very interesting observations here the given data except for (16g) are similar
to (11) and (12) in that there is no variant form and in that there is a glide formation process One
difference is that no case of VD is witnessed in (11) and (12) while the data in (16) and (17)
presents very productive use of VD and do not show any instances of GI
All the examples presented show that when VI and V2 get together it is always the VI that
is deleted V2 remains undeleted This may not necessary mean that V2 triggers vowel deletion To
go deeper into VD phenomenon lets first consider environments for vowel deletion
(18) Vowel deletion environments a VI is less sonorous than V2 (16e) (16h) (l7d) (17e) b VI is more sonorous than V2 (17f) c VI andVj are identical (16f) (17g) (17h)
Apparently (18) covers almost all the possible combinations except i-u and u-i
sequences First of all we can see that (18a) shares similar environments with GF in (14) And
actually GF applies to such examples as in (16c) (16d) (17b) and (17e) Just saying that both GF
and VD are optional (cf Mphande (1989121)) and that they share same the same environments
6) I think that m- might have come from mu- through u-deletion This may be a historical change since it is very difficult to find the synchronic evidence that m- is related to mu- as in the case of mpaka (until) Then we may conclude that both forms are available for (16g) but only mshy is available for (16h)
-332~
9 Hiatus Resolving in Chitumbuka
does not explain why there is no variation
One may say that the difference in boundary type results in the application of the different
rules and add the boundary in the rule-description accordingly This can be illustrated as in (19)
(19) a ku - Ii - a --gt kulya b Ii - ake --gt lake
Though (19a) and (19b) shows the same sequence of i-a IIi in (19a) is a stem while it is
a prefix in (19b) And lal in (19a) is a grammatical suffix and lal in (19b) is a part of a lexical
morpheme In other words (19a) is suffixation and (19b) is prefixation By assigning different
boundaries we can explain why we have GI only in (11) and (12) and not in (16) and (17)
One example of the boundary functioning in conditioning rules may be found in English g-
deletion as in Hyman (1976197) According to them there are two different I-erl suffixes in
English The first one is introduced with word boundary as in singer Another is introduced
with morpheme boundary + as in long-er and English has a g-deletion rule that goes as under
(20) g-deletion in English g --gt 0N_ (N a velar nasal)
This rule with the boundary differentiation and with the specification of the boundary type
deletes IgI in singer but not in longer
The same strategy can be employed here We may assume that the stems are introduced
with one type of boundary say and the suffixes with another + we may represent (19) as
(21)7
7)1 will simply assume that suffixes are introduced with + boundary and prefixes are introduced with boundary But I will continue to use - to mark the morpheme break The boundary differentiation according to Aronoff (1976) means that suffixation preceded prefixation
-- 333shy
(21) aku li + a b li aka
Now we can formulate the following VD rule
(22) Vowel deletion Vi --gt 0 I Vz
This seems to be a reasonable approach I believe this may be partly true for (17) cases
But we need something else This does not help explaining (16) because we have to explain why
GF instead of VD is applied to such examples as (16a) to (16d) and (17a) to (17c)
One way out from this dilemma is to stick to glide formation for (16) then make a rule that
deletes the glide What is meant by this is that it is not VD but GF followed by GD in (16) as
shown in (23)
(23) li - ake --(GF)--gt lyake --(GD)--gt lake
This process is motivated by the fact that in all the cases of VD in GF environments Iii is
preceded by a coronal consonant as shown in (16e) (17d) and (17e) Notice also that the recent
works on features such as Sagey (1986) McCarthy (1988) and many others claim that palatal
glides have [coronal] feature Given this we may formulate the following GD rule
(24) Glide Deletion [+coronal] --gt 0 I [+ coronal]
The rule is maximally generalized We dont have to specify whether the sound deleted is a
glide The syllable structure given in section 1 will naturally predict that it is a glide since in C1C2
consequence the Cz must be a glide But still the boundary is necessary to prevent GD to such
words as ku li + a
Now lets take a look at (17) cases Here again we find that the environment of VD is
-334~
Hiatus Resolving in Chitumbuka 11
very similar to GI in (15) We will have to resort to the boundary distinction as proposed in
discussion with the sonority condition as in (25)
(25) Vowel deletion V1--gt0_V2 (conditionV1 is not less sonorous than V2)
The condition does not necessarily require a higher sonority for the first vowel This
reflects an effort to explain (17g) (17h) cases together with (17f) This approach however has
one immediate problem of explaining (16a) and (16d) VD as in (25) can apply to these cases but
mini and Tizxa are ill-formed One major observational difference of the examples to which
VD is applied is the roundness feature of the two vowels When two vowels have the same
roundness feature and the same height the first one deletes And GF applies when the two vowels
are of the same height (and therefore the same sonority) but differ in roundness This problem can
be solved by adding roundness agreement condition to VD as in (26)81
(26) Vowel deletion - revised VI __ gt 0 __ V2
[a round] [a round] (conditionVIis not less sonorous than V2)
23 Between V-final affixes and V-initial Verbs
There are five different types of verb prefixes They are Iku-I Iliku-I Ngu Izamu-I Ika-I
and le-I I called these prefixes not agreement markers The reason is two-fold first they have their
own meanings These prefixes express tenses or aspects of the verb And unlike agreement
8 lAnother possibility that is imposing roundness disagreement condition on the GF rule makes a wrong prediction that lyenge in (12c) and kumanyeka in (12d) should not undergo GF which is contradictory with the given data
9 )The followings are the general meaning of these prefixes a ku- Present tense present progressive or habitual aspect of the present
-335shy
12 7j-iilfHl ~1lyenlt(1993 2)
markers in 21 or 22 they do not carry grammatical information Further we find that agreement
markers are added before these prefixes in the sentences
Lets begin the discussion by looking at the following data
(27) Prefixes with final lui and V-initial verbs a ku shy amba kwamba kuyamba (to start) b ku - ambuka kwambuka kuyambuka (to cross) c ku - ata kwata kuyata (to create) d ku shy enda kwenda kuyenda (to walk) e ku shy eNga kweNga kuyeNga (to melt) f ku - imba kwimba kuyimba (to sing) g ku - ima kwima kuyima (to stand) h ku - iba kwiba (to steal) I ku - iza kw iza (to come) J ku shy ocha kocha kuwocha (to burn) k l
ku - opa ku - ona
kopa ku wopaku tJY(1Y~- Hu
(to fear) in ~f~Ofmiddot
ffi ku -ma KU wura (to fear) n ku - umba kuwurnba (to mold something) o ku - undira kuwundira (to bury)
(28)Tense or aspect marker and V-initial verbs a ka - imba kimba kayimba (sang) b ka - iba kiba (stole) c ka - enda kenda kayenda (walked) d ka - opa kawopa (feared) e ka - ona kawona (saw) f ka - undira kawundira (buried) g ka - amba kamba kayamba (started)
The first thing readily noticeable in the given data is the existence of different alternatives in
b liku- Perfective the state of verb or the result of activity in the past is carried on to the present
c zamu- Future tense d ka- distant past (activity in the past) e kaNgu-habitual activity in the past f a- Activity in the immediate past g aNgu- Duration of activity that ended in the immediate past
lO)Though all the examples given here have Iku-I prefix the exactly same phonological changes are observed with all other prefixes that end in lui such as INgu-l liku-j and zamu-z All the V-initial verbs elicited so far are given here
-336shy
Hiatus Resolving in Chitumbuka 13
most of the cases Lets first begin with the apparently exceptional cases that do not have
variations Our Chitumbuka informant was not sure whether the stems of (27h) and (27i) begin
with Iii or not On a couple of occasions he said that to steal is kuba And even when he
pronounced kwiba he still maintained that the stem is Ibal instead of iba Though without any
sufficient supporting data I take this as the evidence that the root form of to steal has lost the
alternations and fossilized into kwiba and the same might be true for (27i)
Next it is very strange that only (271) shows different behavior in comparison with the two
preceding examples (27j) and (27k) It may be extremely difficult to assume that (271) has lost the
alternations and fossilized into kuwona form Intuitively if it is to be fossilized the form should
have been kona a phonologically simpler form I suggest that the base form of to see is uon shy
a GF is applied to this to produce Iwonal and then there is no vowel clash when Iku-I is added
failing to meet the VD environment
(27m) (27n) and (270) look quite exceptional in that VD does not apply to theses examples
though they are in VD environments I do not have systematic explanation for these There may he
some historical change in these words III or the underlying forms may begin with luul and the GF
changes the first lui into glide before tense or aspect markers are added 121 or they are just
exceptional At this point I will leave this as an open question
The GF and VD are quite straight forward There is no need to revise the rules Now we
will have to think about GI cases GI rule given in (15) does not explain the examples given in this
11)For example their friend in Chitumbuka is munyawo or rnunyalo Here we sec the alternations hetween Iwland iraquo] The same thing might have happened in (27m) (27n) and (270) words we may speculate that historically those stems begin with lraquo] And the sound weakened into the glide Iw
12)50 far no two same segments have been found to be adjacent morpheme internally However given the GF rule environment I think this idea may not be totally speculative
-- 337shy
subsection the sonority condition is not satisfied
One possible linear phonology approach may be to resort to the rule ordering We may
eliminate the sonority condition of GI and simply order GI after GF Further GI should not be
applied in (16) and (17) and therefore VD should be placed before GI hence the following order
(29) Rule ordering 1 Glide formation 2 Vowel deletion 3 Glide insertion
The major problem with this approach is the nature of the rule The data (27) (28) indicate
that all the three rules are optional But even though they are optional there is no option of not
applying any of the rules and therefore the last rule (GI) should be obligatory Further the data
(11) tells us a different story GF seems to be obligatory Otherwise we may have to say that
(1Ia) can be realized either as imwe (GF) or as imuye (GI) Moreover (17) shows that VD is
obligatory to prevent the operation of GI Here is the paradox A rule is optional for certain group
of words and obligatory for the rest This may suggest that rule based linear approach may not be
suitable for the explanation of hiatus resolving in Chitumbuka Further theoretical investigation will
be made in 32
24 Subject agreement marker and past imperfective marker la-I
All the prefixes introduced in 21 and some other personal agreement markers are used in
the sentence to satisfy the subject agreement Meanwhile there is one V-initial tense marker in
chitumbuka a- And therefore here again we expect to find vowel clashes The representative
data are given in (30)
13) If any of the rule is obligatory then we should not see the variations in the data given in (27) and (28)
-338shy
Hiatus Resolving in Chitumbuka 15
(30) Subject agreement marker and past imperfective marker a ni - aluta naluta (I walked) b u - aluta waluta (you walked) c wa - aluta waluta (he walked) d ti - aluta taluta (we went) e li - aluta laluta (It went) f vi - aluta (vyaluta) valuta (They went)
Superficially the data given here looks like those in (17) There is no GI example And the
choice between GF and VD seems to be determined by the relative sonority of the vowels
However (30f) shows that there is something different here (30) may not be the extension of
examples given in (17) The reason is twofold first the examples given here are not lexical The
agreement marker assignment should refer to the syntactic category the subject Secondly there is
an independent process of i-deletion in chitumbuka Consider following examples
(31) i-deletion and complex segment formation a ni (copula) -litali (big) --gt nilitali or nditali laquoit) is big) b ni (subject agreement)- kulya --gt nikulya or Nkhulya laquol) eat)
Here again i is deleted though it is not followed by another vowel I will tentatively
formulate the i-deletion rule as in (32)
(32) i-deletion i --gt 0_ [ (where [ is a morphological word domain)
This rule as (31a) (31 b) and (30f) shows is optional Then what if the option is not taken
in (30) examples We might expect to see nyaluta in case of example (30a) Notice that the
14 )The irrelevant morpheme breaks are not shown here aluta can be further divided into a (past imperfective) -Iut (go)- a (FV)
15 )The form in the parenthesis means that it may be used by some But the other form is far more widely used and the informant prefers to use valuta
16 )Further investigation of phonological word derivation based on the syntactic information and morphological words should be made Here only the difference of phonological and morphological words is used without mentioning how different they are
-339shy
morphological word domain contains boundary and that the agreement markers begin with
coronal sounds If GF applies then GD in (24) will eliminate nyaluta
25 Unresolved vowel clashes
Not all the vowel sequences are resolved in Chitumbuka Some representative examples are
given in (33) and (34)
(33) V-final nouns and determiners a khuni Ili --gt khuni (y)ili khunyili khunili
tree this b buku ilo --gt buku (y)ilo bukwilo
book that c chipinda ico --gt chipinda (y)ico chipindico
room this
(34) Pronominal objects and V-initial verbs 171
a u-ku-mu-at-a --gt wakumu(y)ata wakumwata (you created him)
b wa-ku-ti-at-a --gt wakuti (y)ata wakutyata (He created you)
c ni-ku-tu-och-a --gt nikuku(w)ocha nikukocha nikukwocha (I roasted them)
d mu-ku-ka-eNg-a --gt mukuka(y)eNga muku (I melted it)
This seems to tell us that the main domain of hiatus resolving is within phonological words
GF and VD do not operate on this level and only G1 optionally shows up For further discussion
the phonological word boundaries are represented with parentheses in (35)
17 )The detailed breakdowns of the strings are as under
(yOU) ~u (him) tt (create)tni (me) ku + 11 (~e) + och (bake) + a (FV) wa (heher ka (It) Ng (melt)
u (them) verb stem
subject verb pronominal agreement tense objects
-340shy
Hiatus Resolving in Chitumbuka 17
(35) a (khuni) (ili) b (niku) (tu) (ocha) c (nikuwocha) (tuyuni)
One very interesting observation here is the change in the phonological word domains
(35b) and (35c) are semantically same The object in (35b) is pronominal while (35c) has a noun
object As a result of movement two phonological words are broken into three phonological
words This is not an arbitrary boundary just to explain the lack of hiatus resolving We may find
a very similar situation in English Intonational Phrase parsing
(36) a (As you know) (the tall boy is a genius) b (The tall boy) (as you know) (is a genius)
Nespor and Vogels (1986189) intonational phrase formation rule produces (36a) as well
as (36b) The syntax dependency of the phonological word formation is partly responsible for the
difference between (35b) and (35c)
Further hiatus is not the only rule sensitive to boundary difference in Chitumbuka We can
find that i-deletion and complex segment formation discussed in the preceding subsection shows
the same phenomenon Consider the following example (parentheses indicate the phonological
word domain)
(37) a (niku) (mu) (temwa) I love him Nkhu
b (uku) (ni)(temwa) You love me nthe
c (khunilClllno) the tree here mpha
One final observation is that our rules so far can not explain why glides can be inserted in
the examples of (33) and (34) We have already pointed out the problem of GI in 21 The
additional complication is that even if we posit glides as a phoneme we still have to revise GIor
-341shy
make another one to explain post-lexical GL This will be addressed in 32
3 Syllable phonology of hiatus resol ving
In the discussion in section 2 some of the problems in the linear analysis were detected
(What are they ) In this section I will go over them and reanalyze the data with the help
of syllable structure to see whether we can answer some of the questions left open McCarthy and
Prince (1986) style of moraic syllable structure will be employed for our discussion in this section
The basic syllable structure of Chitumbuka is given in (38)
(38) Syllable structure in Chitumbuka a a b a c a
~~((r r1T ( (I I 1 All the syllables are mora final 2 All the syllables are mono-moraic
This is not the template that is required in the underlying form as will be discussed
Therefore the non-moraic tier may not have to be satisfied In other words though the structure
shows two pre-moraic consonant slots these can be left empty The two branches simple mean the
maximum number of consonants that can come before a vowel
18)1n the post lexical level this structural condition may not hold In Chitumbuka vowels in the stressed syllable is lengthened The long vowel is represented with bimoraic syllable structure It is not uncommon that some of the rules or conditions are applied only on the lexical level but not on the post lexical level such as extrametricality stress assignment and even VD and GF we talked about in the preceding section
- 342-shy
Hiatus Resolving in Chitumbuka 19
31 Syllabification
Contrary to the claim that all the segments are already syllabified in the base form (cf
Selkirk (1982raquo I assume that syllable structure is assigned through the syllabification process
The syllabification principles in Chitumbuka are given under
(39) Syllabification in Chitumbuka i) Morafication A segment projects a mora if it is more sonorous than the following
segrnent ii) Syllable projection Each mora projects a syllable node iii) Segment incorporation Unsyllabified segments are incorporated into syllables
Now lets take a look at the following sample syllabification
(40) a
0 0 00
I (iii)I(ii) --(i) f1 f1 f1 f1 ~ rlrOI I
~
I I n i u mb a n 1 u mb a n 1 u mb a n i u mb a
b
0 o 0 0 o 0
I I I (iW (~Of1 f1 f1 f1 ~ ~ I I r~r I I I I I ch i m a i ch i m a ch i m a ch i m a i
Notice that we do not need GF as a rule the same result of GF application is achieved as a
19)1 suppose that morafication principle is universally dependent upon the sonority hierarchy There are three different strategies in using sonority in syllabifications
i) y is moraic in xy if it is more sonorous than x ii) y is moraic in yz if it is more sonorous than z iii) y is moraic if its sonority is greater than N
The language in the world employ one or two all all of these in the syllabification and language may slightly vary in setting up the value of N in (iii)
-- 343shy
natural consequence of syllabification In 21 we obtain the simplicity of phonemic inventory at
the cost of introducing GF thereby complicating the rule component The environment of GF is
reflected in the syllabification principle Of course syllabification can not be an additional burden
the language needs syllable structure anyhow for example for the application of many phonological
rules and even for utterance
The nature of glide sounds are clearly reflected in the syllable structure If a vowel is
dominated by a mora it will be interpreted as a full vowel But if it is dominated by the syllable
node directly it will surface as a glide
32 Syllabification domain
The syllabification principles introduced in 31 have presupposed that the string of segments
are not syllabified in the underlying representation Suppose every morpheme is syllabified in the
deep structure and consider the following example of word formation
(41)
a aa a
~ (I (noa k 0 V 1 a k 0
Now we will have to make a rule for mora deletion which will be almost the repetition of
the syllabification principle Syllabification process in the course of derivation can get rid of the
redundant repetition of syllable structural description as well as save the later-on application of mora
deletion
I assume that the syllabification takes place in the first derived domain In other words
segments are syllabified after the initial word formation process Before going into the detailed
-344shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
final vowels are always [s-high -low] as shown in (13)
(13) a ku (V-marker) -Ii (eat) - Vk (passive) - a (FV) --gt kulyeka (to be eaten) b ku(V-marker) - mu(drink) - Vsxtcausative) - a(FV) --gt kumwesxa (to make drink)
(13) ensures that the initial vowel of the suffix is always of greater sonority Therefore
stem final vowels are always less sonorous than the suffix initial vowels Now we can find one
thing in common between the data of (11) and (12) except (12f) and (12g)
We may therefore posit the following informal glide formation rule
(14) Glide formation (GF) Vl --gt [-syllabic]_V2
(condition V2 is not less sonorous than V1 )41
Now lets turn to (12f) and (12g) In these cases V1 is more sonorous than V2 Given
condition (6) and rule (14) on the one hand and the absence of vowel deletion on the other the
only way to resolve the ill-formed form seems to insert a glide in between two vowels At this
point it is not clear whether the inserted glide is y or w Though we only have the examples of
y-insertion the rule will be formulated to accomodate w so that we can use GI for discussions in
the following subsections
(15) Glide insertion (GI)
o --gt [~~~~~~~J
VI V2 [a round]
a round (Condition V1 is more sonorous than V2)
This rule inserts y before e and a and w before 051 One problem that should be
4)1 simply followed Jesperson (1932 191) to say that open vowels are more sonorous than closed vowels
5) Chitumbuka has five vowel system Its phonemic vowels are le iI lui lei 101
-330shy
7 Hiatus Resolving in Chitumbuka
pointed out now is the nature of the inserted segment y or WI These are not in the phonemic
inventory of Chitumbuka as discussed earlier Insertion of sounds which are not phonemic is the
characteristic of the phonetic rules that do not participate in the phonological derivation This may
mean that there should be glides in the phonemic inventory contrary to the basic assumption here
or that GI is formulated in the wrong way We will postpone the discussion to 31 and stick to GI
rule for the purpose of discussion in the present section
The GF in (14) and GI in (15) can take care of all the data of morpheme-internal vowel
clashes as well as the vowel sequence cause by the morpheme concatenation of V-final stems and V-
initial affixes
22 Between V-final affixes and V-initial stems
Another very productive source of vowel clashes is the concatenation of agreement markers
with V-initial stems In Chitumbuka nouns adjectives and verbs carry agreement markers and all
the agreement markers are V-final
(16) Agreement marker and V-initial nouns a mn shy ini mwini (owner) b mu - anice mwanice (child) c vi - aro vyaro (lands) d In - izsa Iwizxa (economy) e chi - aro charo (land) f va - ana vana (kids) g mn - ono mono mnwono (creel) h mn shy ova mova (beer)
(17) Agreement marker and possessive pronouns a vi - ane vyane (mine) b u - ane wane (mine) c tu - ako twako (yours-sg) d li - ake lake (hishers) e zi - awo zawo (theirs) f va - ithu vithu (ours) g va - ako vako (hishers)
-331shy
h li - inu inu (yours-pl)
Before going into any detail I should say that (16g) and (16h) may be problematic GF can
be applied but (16h) does not show the form with a glide Vail (197240) observed that mu-
alternates with m or syllabic m I will simply assume that this morphemic alternation is
responsible for the variations in (16g)61 It is not the application of different rules but the
coexistence of mu- and m- form that results in the two different surface forms in (16g)
We can make very interesting observations here the given data except for (16g) are similar
to (11) and (12) in that there is no variant form and in that there is a glide formation process One
difference is that no case of VD is witnessed in (11) and (12) while the data in (16) and (17)
presents very productive use of VD and do not show any instances of GI
All the examples presented show that when VI and V2 get together it is always the VI that
is deleted V2 remains undeleted This may not necessary mean that V2 triggers vowel deletion To
go deeper into VD phenomenon lets first consider environments for vowel deletion
(18) Vowel deletion environments a VI is less sonorous than V2 (16e) (16h) (l7d) (17e) b VI is more sonorous than V2 (17f) c VI andVj are identical (16f) (17g) (17h)
Apparently (18) covers almost all the possible combinations except i-u and u-i
sequences First of all we can see that (18a) shares similar environments with GF in (14) And
actually GF applies to such examples as in (16c) (16d) (17b) and (17e) Just saying that both GF
and VD are optional (cf Mphande (1989121)) and that they share same the same environments
6) I think that m- might have come from mu- through u-deletion This may be a historical change since it is very difficult to find the synchronic evidence that m- is related to mu- as in the case of mpaka (until) Then we may conclude that both forms are available for (16g) but only mshy is available for (16h)
-332~
9 Hiatus Resolving in Chitumbuka
does not explain why there is no variation
One may say that the difference in boundary type results in the application of the different
rules and add the boundary in the rule-description accordingly This can be illustrated as in (19)
(19) a ku - Ii - a --gt kulya b Ii - ake --gt lake
Though (19a) and (19b) shows the same sequence of i-a IIi in (19a) is a stem while it is
a prefix in (19b) And lal in (19a) is a grammatical suffix and lal in (19b) is a part of a lexical
morpheme In other words (19a) is suffixation and (19b) is prefixation By assigning different
boundaries we can explain why we have GI only in (11) and (12) and not in (16) and (17)
One example of the boundary functioning in conditioning rules may be found in English g-
deletion as in Hyman (1976197) According to them there are two different I-erl suffixes in
English The first one is introduced with word boundary as in singer Another is introduced
with morpheme boundary + as in long-er and English has a g-deletion rule that goes as under
(20) g-deletion in English g --gt 0N_ (N a velar nasal)
This rule with the boundary differentiation and with the specification of the boundary type
deletes IgI in singer but not in longer
The same strategy can be employed here We may assume that the stems are introduced
with one type of boundary say and the suffixes with another + we may represent (19) as
(21)7
7)1 will simply assume that suffixes are introduced with + boundary and prefixes are introduced with boundary But I will continue to use - to mark the morpheme break The boundary differentiation according to Aronoff (1976) means that suffixation preceded prefixation
-- 333shy
(21) aku li + a b li aka
Now we can formulate the following VD rule
(22) Vowel deletion Vi --gt 0 I Vz
This seems to be a reasonable approach I believe this may be partly true for (17) cases
But we need something else This does not help explaining (16) because we have to explain why
GF instead of VD is applied to such examples as (16a) to (16d) and (17a) to (17c)
One way out from this dilemma is to stick to glide formation for (16) then make a rule that
deletes the glide What is meant by this is that it is not VD but GF followed by GD in (16) as
shown in (23)
(23) li - ake --(GF)--gt lyake --(GD)--gt lake
This process is motivated by the fact that in all the cases of VD in GF environments Iii is
preceded by a coronal consonant as shown in (16e) (17d) and (17e) Notice also that the recent
works on features such as Sagey (1986) McCarthy (1988) and many others claim that palatal
glides have [coronal] feature Given this we may formulate the following GD rule
(24) Glide Deletion [+coronal] --gt 0 I [+ coronal]
The rule is maximally generalized We dont have to specify whether the sound deleted is a
glide The syllable structure given in section 1 will naturally predict that it is a glide since in C1C2
consequence the Cz must be a glide But still the boundary is necessary to prevent GD to such
words as ku li + a
Now lets take a look at (17) cases Here again we find that the environment of VD is
-334~
Hiatus Resolving in Chitumbuka 11
very similar to GI in (15) We will have to resort to the boundary distinction as proposed in
discussion with the sonority condition as in (25)
(25) Vowel deletion V1--gt0_V2 (conditionV1 is not less sonorous than V2)
The condition does not necessarily require a higher sonority for the first vowel This
reflects an effort to explain (17g) (17h) cases together with (17f) This approach however has
one immediate problem of explaining (16a) and (16d) VD as in (25) can apply to these cases but
mini and Tizxa are ill-formed One major observational difference of the examples to which
VD is applied is the roundness feature of the two vowels When two vowels have the same
roundness feature and the same height the first one deletes And GF applies when the two vowels
are of the same height (and therefore the same sonority) but differ in roundness This problem can
be solved by adding roundness agreement condition to VD as in (26)81
(26) Vowel deletion - revised VI __ gt 0 __ V2
[a round] [a round] (conditionVIis not less sonorous than V2)
23 Between V-final affixes and V-initial Verbs
There are five different types of verb prefixes They are Iku-I Iliku-I Ngu Izamu-I Ika-I
and le-I I called these prefixes not agreement markers The reason is two-fold first they have their
own meanings These prefixes express tenses or aspects of the verb And unlike agreement
8 lAnother possibility that is imposing roundness disagreement condition on the GF rule makes a wrong prediction that lyenge in (12c) and kumanyeka in (12d) should not undergo GF which is contradictory with the given data
9 )The followings are the general meaning of these prefixes a ku- Present tense present progressive or habitual aspect of the present
-335shy
12 7j-iilfHl ~1lyenlt(1993 2)
markers in 21 or 22 they do not carry grammatical information Further we find that agreement
markers are added before these prefixes in the sentences
Lets begin the discussion by looking at the following data
(27) Prefixes with final lui and V-initial verbs a ku shy amba kwamba kuyamba (to start) b ku - ambuka kwambuka kuyambuka (to cross) c ku - ata kwata kuyata (to create) d ku shy enda kwenda kuyenda (to walk) e ku shy eNga kweNga kuyeNga (to melt) f ku - imba kwimba kuyimba (to sing) g ku - ima kwima kuyima (to stand) h ku - iba kwiba (to steal) I ku - iza kw iza (to come) J ku shy ocha kocha kuwocha (to burn) k l
ku - opa ku - ona
kopa ku wopaku tJY(1Y~- Hu
(to fear) in ~f~Ofmiddot
ffi ku -ma KU wura (to fear) n ku - umba kuwurnba (to mold something) o ku - undira kuwundira (to bury)
(28)Tense or aspect marker and V-initial verbs a ka - imba kimba kayimba (sang) b ka - iba kiba (stole) c ka - enda kenda kayenda (walked) d ka - opa kawopa (feared) e ka - ona kawona (saw) f ka - undira kawundira (buried) g ka - amba kamba kayamba (started)
The first thing readily noticeable in the given data is the existence of different alternatives in
b liku- Perfective the state of verb or the result of activity in the past is carried on to the present
c zamu- Future tense d ka- distant past (activity in the past) e kaNgu-habitual activity in the past f a- Activity in the immediate past g aNgu- Duration of activity that ended in the immediate past
lO)Though all the examples given here have Iku-I prefix the exactly same phonological changes are observed with all other prefixes that end in lui such as INgu-l liku-j and zamu-z All the V-initial verbs elicited so far are given here
-336shy
Hiatus Resolving in Chitumbuka 13
most of the cases Lets first begin with the apparently exceptional cases that do not have
variations Our Chitumbuka informant was not sure whether the stems of (27h) and (27i) begin
with Iii or not On a couple of occasions he said that to steal is kuba And even when he
pronounced kwiba he still maintained that the stem is Ibal instead of iba Though without any
sufficient supporting data I take this as the evidence that the root form of to steal has lost the
alternations and fossilized into kwiba and the same might be true for (27i)
Next it is very strange that only (271) shows different behavior in comparison with the two
preceding examples (27j) and (27k) It may be extremely difficult to assume that (271) has lost the
alternations and fossilized into kuwona form Intuitively if it is to be fossilized the form should
have been kona a phonologically simpler form I suggest that the base form of to see is uon shy
a GF is applied to this to produce Iwonal and then there is no vowel clash when Iku-I is added
failing to meet the VD environment
(27m) (27n) and (270) look quite exceptional in that VD does not apply to theses examples
though they are in VD environments I do not have systematic explanation for these There may he
some historical change in these words III or the underlying forms may begin with luul and the GF
changes the first lui into glide before tense or aspect markers are added 121 or they are just
exceptional At this point I will leave this as an open question
The GF and VD are quite straight forward There is no need to revise the rules Now we
will have to think about GI cases GI rule given in (15) does not explain the examples given in this
11)For example their friend in Chitumbuka is munyawo or rnunyalo Here we sec the alternations hetween Iwland iraquo] The same thing might have happened in (27m) (27n) and (270) words we may speculate that historically those stems begin with lraquo] And the sound weakened into the glide Iw
12)50 far no two same segments have been found to be adjacent morpheme internally However given the GF rule environment I think this idea may not be totally speculative
-- 337shy
subsection the sonority condition is not satisfied
One possible linear phonology approach may be to resort to the rule ordering We may
eliminate the sonority condition of GI and simply order GI after GF Further GI should not be
applied in (16) and (17) and therefore VD should be placed before GI hence the following order
(29) Rule ordering 1 Glide formation 2 Vowel deletion 3 Glide insertion
The major problem with this approach is the nature of the rule The data (27) (28) indicate
that all the three rules are optional But even though they are optional there is no option of not
applying any of the rules and therefore the last rule (GI) should be obligatory Further the data
(11) tells us a different story GF seems to be obligatory Otherwise we may have to say that
(1Ia) can be realized either as imwe (GF) or as imuye (GI) Moreover (17) shows that VD is
obligatory to prevent the operation of GI Here is the paradox A rule is optional for certain group
of words and obligatory for the rest This may suggest that rule based linear approach may not be
suitable for the explanation of hiatus resolving in Chitumbuka Further theoretical investigation will
be made in 32
24 Subject agreement marker and past imperfective marker la-I
All the prefixes introduced in 21 and some other personal agreement markers are used in
the sentence to satisfy the subject agreement Meanwhile there is one V-initial tense marker in
chitumbuka a- And therefore here again we expect to find vowel clashes The representative
data are given in (30)
13) If any of the rule is obligatory then we should not see the variations in the data given in (27) and (28)
-338shy
Hiatus Resolving in Chitumbuka 15
(30) Subject agreement marker and past imperfective marker a ni - aluta naluta (I walked) b u - aluta waluta (you walked) c wa - aluta waluta (he walked) d ti - aluta taluta (we went) e li - aluta laluta (It went) f vi - aluta (vyaluta) valuta (They went)
Superficially the data given here looks like those in (17) There is no GI example And the
choice between GF and VD seems to be determined by the relative sonority of the vowels
However (30f) shows that there is something different here (30) may not be the extension of
examples given in (17) The reason is twofold first the examples given here are not lexical The
agreement marker assignment should refer to the syntactic category the subject Secondly there is
an independent process of i-deletion in chitumbuka Consider following examples
(31) i-deletion and complex segment formation a ni (copula) -litali (big) --gt nilitali or nditali laquoit) is big) b ni (subject agreement)- kulya --gt nikulya or Nkhulya laquol) eat)
Here again i is deleted though it is not followed by another vowel I will tentatively
formulate the i-deletion rule as in (32)
(32) i-deletion i --gt 0_ [ (where [ is a morphological word domain)
This rule as (31a) (31 b) and (30f) shows is optional Then what if the option is not taken
in (30) examples We might expect to see nyaluta in case of example (30a) Notice that the
14 )The irrelevant morpheme breaks are not shown here aluta can be further divided into a (past imperfective) -Iut (go)- a (FV)
15 )The form in the parenthesis means that it may be used by some But the other form is far more widely used and the informant prefers to use valuta
16 )Further investigation of phonological word derivation based on the syntactic information and morphological words should be made Here only the difference of phonological and morphological words is used without mentioning how different they are
-339shy
morphological word domain contains boundary and that the agreement markers begin with
coronal sounds If GF applies then GD in (24) will eliminate nyaluta
25 Unresolved vowel clashes
Not all the vowel sequences are resolved in Chitumbuka Some representative examples are
given in (33) and (34)
(33) V-final nouns and determiners a khuni Ili --gt khuni (y)ili khunyili khunili
tree this b buku ilo --gt buku (y)ilo bukwilo
book that c chipinda ico --gt chipinda (y)ico chipindico
room this
(34) Pronominal objects and V-initial verbs 171
a u-ku-mu-at-a --gt wakumu(y)ata wakumwata (you created him)
b wa-ku-ti-at-a --gt wakuti (y)ata wakutyata (He created you)
c ni-ku-tu-och-a --gt nikuku(w)ocha nikukocha nikukwocha (I roasted them)
d mu-ku-ka-eNg-a --gt mukuka(y)eNga muku (I melted it)
This seems to tell us that the main domain of hiatus resolving is within phonological words
GF and VD do not operate on this level and only G1 optionally shows up For further discussion
the phonological word boundaries are represented with parentheses in (35)
17 )The detailed breakdowns of the strings are as under
(yOU) ~u (him) tt (create)tni (me) ku + 11 (~e) + och (bake) + a (FV) wa (heher ka (It) Ng (melt)
u (them) verb stem
subject verb pronominal agreement tense objects
-340shy
Hiatus Resolving in Chitumbuka 17
(35) a (khuni) (ili) b (niku) (tu) (ocha) c (nikuwocha) (tuyuni)
One very interesting observation here is the change in the phonological word domains
(35b) and (35c) are semantically same The object in (35b) is pronominal while (35c) has a noun
object As a result of movement two phonological words are broken into three phonological
words This is not an arbitrary boundary just to explain the lack of hiatus resolving We may find
a very similar situation in English Intonational Phrase parsing
(36) a (As you know) (the tall boy is a genius) b (The tall boy) (as you know) (is a genius)
Nespor and Vogels (1986189) intonational phrase formation rule produces (36a) as well
as (36b) The syntax dependency of the phonological word formation is partly responsible for the
difference between (35b) and (35c)
Further hiatus is not the only rule sensitive to boundary difference in Chitumbuka We can
find that i-deletion and complex segment formation discussed in the preceding subsection shows
the same phenomenon Consider the following example (parentheses indicate the phonological
word domain)
(37) a (niku) (mu) (temwa) I love him Nkhu
b (uku) (ni)(temwa) You love me nthe
c (khunilClllno) the tree here mpha
One final observation is that our rules so far can not explain why glides can be inserted in
the examples of (33) and (34) We have already pointed out the problem of GI in 21 The
additional complication is that even if we posit glides as a phoneme we still have to revise GIor
-341shy
make another one to explain post-lexical GL This will be addressed in 32
3 Syllable phonology of hiatus resol ving
In the discussion in section 2 some of the problems in the linear analysis were detected
(What are they ) In this section I will go over them and reanalyze the data with the help
of syllable structure to see whether we can answer some of the questions left open McCarthy and
Prince (1986) style of moraic syllable structure will be employed for our discussion in this section
The basic syllable structure of Chitumbuka is given in (38)
(38) Syllable structure in Chitumbuka a a b a c a
~~((r r1T ( (I I 1 All the syllables are mora final 2 All the syllables are mono-moraic
This is not the template that is required in the underlying form as will be discussed
Therefore the non-moraic tier may not have to be satisfied In other words though the structure
shows two pre-moraic consonant slots these can be left empty The two branches simple mean the
maximum number of consonants that can come before a vowel
18)1n the post lexical level this structural condition may not hold In Chitumbuka vowels in the stressed syllable is lengthened The long vowel is represented with bimoraic syllable structure It is not uncommon that some of the rules or conditions are applied only on the lexical level but not on the post lexical level such as extrametricality stress assignment and even VD and GF we talked about in the preceding section
- 342-shy
Hiatus Resolving in Chitumbuka 19
31 Syllabification
Contrary to the claim that all the segments are already syllabified in the base form (cf
Selkirk (1982raquo I assume that syllable structure is assigned through the syllabification process
The syllabification principles in Chitumbuka are given under
(39) Syllabification in Chitumbuka i) Morafication A segment projects a mora if it is more sonorous than the following
segrnent ii) Syllable projection Each mora projects a syllable node iii) Segment incorporation Unsyllabified segments are incorporated into syllables
Now lets take a look at the following sample syllabification
(40) a
0 0 00
I (iii)I(ii) --(i) f1 f1 f1 f1 ~ rlrOI I
~
I I n i u mb a n 1 u mb a n 1 u mb a n i u mb a
b
0 o 0 0 o 0
I I I (iW (~Of1 f1 f1 f1 ~ ~ I I r~r I I I I I ch i m a i ch i m a ch i m a ch i m a i
Notice that we do not need GF as a rule the same result of GF application is achieved as a
19)1 suppose that morafication principle is universally dependent upon the sonority hierarchy There are three different strategies in using sonority in syllabifications
i) y is moraic in xy if it is more sonorous than x ii) y is moraic in yz if it is more sonorous than z iii) y is moraic if its sonority is greater than N
The language in the world employ one or two all all of these in the syllabification and language may slightly vary in setting up the value of N in (iii)
-- 343shy
natural consequence of syllabification In 21 we obtain the simplicity of phonemic inventory at
the cost of introducing GF thereby complicating the rule component The environment of GF is
reflected in the syllabification principle Of course syllabification can not be an additional burden
the language needs syllable structure anyhow for example for the application of many phonological
rules and even for utterance
The nature of glide sounds are clearly reflected in the syllable structure If a vowel is
dominated by a mora it will be interpreted as a full vowel But if it is dominated by the syllable
node directly it will surface as a glide
32 Syllabification domain
The syllabification principles introduced in 31 have presupposed that the string of segments
are not syllabified in the underlying representation Suppose every morpheme is syllabified in the
deep structure and consider the following example of word formation
(41)
a aa a
~ (I (noa k 0 V 1 a k 0
Now we will have to make a rule for mora deletion which will be almost the repetition of
the syllabification principle Syllabification process in the course of derivation can get rid of the
redundant repetition of syllable structural description as well as save the later-on application of mora
deletion
I assume that the syllabification takes place in the first derived domain In other words
segments are syllabified after the initial word formation process Before going into the detailed
-344shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
7 Hiatus Resolving in Chitumbuka
pointed out now is the nature of the inserted segment y or WI These are not in the phonemic
inventory of Chitumbuka as discussed earlier Insertion of sounds which are not phonemic is the
characteristic of the phonetic rules that do not participate in the phonological derivation This may
mean that there should be glides in the phonemic inventory contrary to the basic assumption here
or that GI is formulated in the wrong way We will postpone the discussion to 31 and stick to GI
rule for the purpose of discussion in the present section
The GF in (14) and GI in (15) can take care of all the data of morpheme-internal vowel
clashes as well as the vowel sequence cause by the morpheme concatenation of V-final stems and V-
initial affixes
22 Between V-final affixes and V-initial stems
Another very productive source of vowel clashes is the concatenation of agreement markers
with V-initial stems In Chitumbuka nouns adjectives and verbs carry agreement markers and all
the agreement markers are V-final
(16) Agreement marker and V-initial nouns a mn shy ini mwini (owner) b mu - anice mwanice (child) c vi - aro vyaro (lands) d In - izsa Iwizxa (economy) e chi - aro charo (land) f va - ana vana (kids) g mn - ono mono mnwono (creel) h mn shy ova mova (beer)
(17) Agreement marker and possessive pronouns a vi - ane vyane (mine) b u - ane wane (mine) c tu - ako twako (yours-sg) d li - ake lake (hishers) e zi - awo zawo (theirs) f va - ithu vithu (ours) g va - ako vako (hishers)
-331shy
h li - inu inu (yours-pl)
Before going into any detail I should say that (16g) and (16h) may be problematic GF can
be applied but (16h) does not show the form with a glide Vail (197240) observed that mu-
alternates with m or syllabic m I will simply assume that this morphemic alternation is
responsible for the variations in (16g)61 It is not the application of different rules but the
coexistence of mu- and m- form that results in the two different surface forms in (16g)
We can make very interesting observations here the given data except for (16g) are similar
to (11) and (12) in that there is no variant form and in that there is a glide formation process One
difference is that no case of VD is witnessed in (11) and (12) while the data in (16) and (17)
presents very productive use of VD and do not show any instances of GI
All the examples presented show that when VI and V2 get together it is always the VI that
is deleted V2 remains undeleted This may not necessary mean that V2 triggers vowel deletion To
go deeper into VD phenomenon lets first consider environments for vowel deletion
(18) Vowel deletion environments a VI is less sonorous than V2 (16e) (16h) (l7d) (17e) b VI is more sonorous than V2 (17f) c VI andVj are identical (16f) (17g) (17h)
Apparently (18) covers almost all the possible combinations except i-u and u-i
sequences First of all we can see that (18a) shares similar environments with GF in (14) And
actually GF applies to such examples as in (16c) (16d) (17b) and (17e) Just saying that both GF
and VD are optional (cf Mphande (1989121)) and that they share same the same environments
6) I think that m- might have come from mu- through u-deletion This may be a historical change since it is very difficult to find the synchronic evidence that m- is related to mu- as in the case of mpaka (until) Then we may conclude that both forms are available for (16g) but only mshy is available for (16h)
-332~
9 Hiatus Resolving in Chitumbuka
does not explain why there is no variation
One may say that the difference in boundary type results in the application of the different
rules and add the boundary in the rule-description accordingly This can be illustrated as in (19)
(19) a ku - Ii - a --gt kulya b Ii - ake --gt lake
Though (19a) and (19b) shows the same sequence of i-a IIi in (19a) is a stem while it is
a prefix in (19b) And lal in (19a) is a grammatical suffix and lal in (19b) is a part of a lexical
morpheme In other words (19a) is suffixation and (19b) is prefixation By assigning different
boundaries we can explain why we have GI only in (11) and (12) and not in (16) and (17)
One example of the boundary functioning in conditioning rules may be found in English g-
deletion as in Hyman (1976197) According to them there are two different I-erl suffixes in
English The first one is introduced with word boundary as in singer Another is introduced
with morpheme boundary + as in long-er and English has a g-deletion rule that goes as under
(20) g-deletion in English g --gt 0N_ (N a velar nasal)
This rule with the boundary differentiation and with the specification of the boundary type
deletes IgI in singer but not in longer
The same strategy can be employed here We may assume that the stems are introduced
with one type of boundary say and the suffixes with another + we may represent (19) as
(21)7
7)1 will simply assume that suffixes are introduced with + boundary and prefixes are introduced with boundary But I will continue to use - to mark the morpheme break The boundary differentiation according to Aronoff (1976) means that suffixation preceded prefixation
-- 333shy
(21) aku li + a b li aka
Now we can formulate the following VD rule
(22) Vowel deletion Vi --gt 0 I Vz
This seems to be a reasonable approach I believe this may be partly true for (17) cases
But we need something else This does not help explaining (16) because we have to explain why
GF instead of VD is applied to such examples as (16a) to (16d) and (17a) to (17c)
One way out from this dilemma is to stick to glide formation for (16) then make a rule that
deletes the glide What is meant by this is that it is not VD but GF followed by GD in (16) as
shown in (23)
(23) li - ake --(GF)--gt lyake --(GD)--gt lake
This process is motivated by the fact that in all the cases of VD in GF environments Iii is
preceded by a coronal consonant as shown in (16e) (17d) and (17e) Notice also that the recent
works on features such as Sagey (1986) McCarthy (1988) and many others claim that palatal
glides have [coronal] feature Given this we may formulate the following GD rule
(24) Glide Deletion [+coronal] --gt 0 I [+ coronal]
The rule is maximally generalized We dont have to specify whether the sound deleted is a
glide The syllable structure given in section 1 will naturally predict that it is a glide since in C1C2
consequence the Cz must be a glide But still the boundary is necessary to prevent GD to such
words as ku li + a
Now lets take a look at (17) cases Here again we find that the environment of VD is
-334~
Hiatus Resolving in Chitumbuka 11
very similar to GI in (15) We will have to resort to the boundary distinction as proposed in
discussion with the sonority condition as in (25)
(25) Vowel deletion V1--gt0_V2 (conditionV1 is not less sonorous than V2)
The condition does not necessarily require a higher sonority for the first vowel This
reflects an effort to explain (17g) (17h) cases together with (17f) This approach however has
one immediate problem of explaining (16a) and (16d) VD as in (25) can apply to these cases but
mini and Tizxa are ill-formed One major observational difference of the examples to which
VD is applied is the roundness feature of the two vowels When two vowels have the same
roundness feature and the same height the first one deletes And GF applies when the two vowels
are of the same height (and therefore the same sonority) but differ in roundness This problem can
be solved by adding roundness agreement condition to VD as in (26)81
(26) Vowel deletion - revised VI __ gt 0 __ V2
[a round] [a round] (conditionVIis not less sonorous than V2)
23 Between V-final affixes and V-initial Verbs
There are five different types of verb prefixes They are Iku-I Iliku-I Ngu Izamu-I Ika-I
and le-I I called these prefixes not agreement markers The reason is two-fold first they have their
own meanings These prefixes express tenses or aspects of the verb And unlike agreement
8 lAnother possibility that is imposing roundness disagreement condition on the GF rule makes a wrong prediction that lyenge in (12c) and kumanyeka in (12d) should not undergo GF which is contradictory with the given data
9 )The followings are the general meaning of these prefixes a ku- Present tense present progressive or habitual aspect of the present
-335shy
12 7j-iilfHl ~1lyenlt(1993 2)
markers in 21 or 22 they do not carry grammatical information Further we find that agreement
markers are added before these prefixes in the sentences
Lets begin the discussion by looking at the following data
(27) Prefixes with final lui and V-initial verbs a ku shy amba kwamba kuyamba (to start) b ku - ambuka kwambuka kuyambuka (to cross) c ku - ata kwata kuyata (to create) d ku shy enda kwenda kuyenda (to walk) e ku shy eNga kweNga kuyeNga (to melt) f ku - imba kwimba kuyimba (to sing) g ku - ima kwima kuyima (to stand) h ku - iba kwiba (to steal) I ku - iza kw iza (to come) J ku shy ocha kocha kuwocha (to burn) k l
ku - opa ku - ona
kopa ku wopaku tJY(1Y~- Hu
(to fear) in ~f~Ofmiddot
ffi ku -ma KU wura (to fear) n ku - umba kuwurnba (to mold something) o ku - undira kuwundira (to bury)
(28)Tense or aspect marker and V-initial verbs a ka - imba kimba kayimba (sang) b ka - iba kiba (stole) c ka - enda kenda kayenda (walked) d ka - opa kawopa (feared) e ka - ona kawona (saw) f ka - undira kawundira (buried) g ka - amba kamba kayamba (started)
The first thing readily noticeable in the given data is the existence of different alternatives in
b liku- Perfective the state of verb or the result of activity in the past is carried on to the present
c zamu- Future tense d ka- distant past (activity in the past) e kaNgu-habitual activity in the past f a- Activity in the immediate past g aNgu- Duration of activity that ended in the immediate past
lO)Though all the examples given here have Iku-I prefix the exactly same phonological changes are observed with all other prefixes that end in lui such as INgu-l liku-j and zamu-z All the V-initial verbs elicited so far are given here
-336shy
Hiatus Resolving in Chitumbuka 13
most of the cases Lets first begin with the apparently exceptional cases that do not have
variations Our Chitumbuka informant was not sure whether the stems of (27h) and (27i) begin
with Iii or not On a couple of occasions he said that to steal is kuba And even when he
pronounced kwiba he still maintained that the stem is Ibal instead of iba Though without any
sufficient supporting data I take this as the evidence that the root form of to steal has lost the
alternations and fossilized into kwiba and the same might be true for (27i)
Next it is very strange that only (271) shows different behavior in comparison with the two
preceding examples (27j) and (27k) It may be extremely difficult to assume that (271) has lost the
alternations and fossilized into kuwona form Intuitively if it is to be fossilized the form should
have been kona a phonologically simpler form I suggest that the base form of to see is uon shy
a GF is applied to this to produce Iwonal and then there is no vowel clash when Iku-I is added
failing to meet the VD environment
(27m) (27n) and (270) look quite exceptional in that VD does not apply to theses examples
though they are in VD environments I do not have systematic explanation for these There may he
some historical change in these words III or the underlying forms may begin with luul and the GF
changes the first lui into glide before tense or aspect markers are added 121 or they are just
exceptional At this point I will leave this as an open question
The GF and VD are quite straight forward There is no need to revise the rules Now we
will have to think about GI cases GI rule given in (15) does not explain the examples given in this
11)For example their friend in Chitumbuka is munyawo or rnunyalo Here we sec the alternations hetween Iwland iraquo] The same thing might have happened in (27m) (27n) and (270) words we may speculate that historically those stems begin with lraquo] And the sound weakened into the glide Iw
12)50 far no two same segments have been found to be adjacent morpheme internally However given the GF rule environment I think this idea may not be totally speculative
-- 337shy
subsection the sonority condition is not satisfied
One possible linear phonology approach may be to resort to the rule ordering We may
eliminate the sonority condition of GI and simply order GI after GF Further GI should not be
applied in (16) and (17) and therefore VD should be placed before GI hence the following order
(29) Rule ordering 1 Glide formation 2 Vowel deletion 3 Glide insertion
The major problem with this approach is the nature of the rule The data (27) (28) indicate
that all the three rules are optional But even though they are optional there is no option of not
applying any of the rules and therefore the last rule (GI) should be obligatory Further the data
(11) tells us a different story GF seems to be obligatory Otherwise we may have to say that
(1Ia) can be realized either as imwe (GF) or as imuye (GI) Moreover (17) shows that VD is
obligatory to prevent the operation of GI Here is the paradox A rule is optional for certain group
of words and obligatory for the rest This may suggest that rule based linear approach may not be
suitable for the explanation of hiatus resolving in Chitumbuka Further theoretical investigation will
be made in 32
24 Subject agreement marker and past imperfective marker la-I
All the prefixes introduced in 21 and some other personal agreement markers are used in
the sentence to satisfy the subject agreement Meanwhile there is one V-initial tense marker in
chitumbuka a- And therefore here again we expect to find vowel clashes The representative
data are given in (30)
13) If any of the rule is obligatory then we should not see the variations in the data given in (27) and (28)
-338shy
Hiatus Resolving in Chitumbuka 15
(30) Subject agreement marker and past imperfective marker a ni - aluta naluta (I walked) b u - aluta waluta (you walked) c wa - aluta waluta (he walked) d ti - aluta taluta (we went) e li - aluta laluta (It went) f vi - aluta (vyaluta) valuta (They went)
Superficially the data given here looks like those in (17) There is no GI example And the
choice between GF and VD seems to be determined by the relative sonority of the vowels
However (30f) shows that there is something different here (30) may not be the extension of
examples given in (17) The reason is twofold first the examples given here are not lexical The
agreement marker assignment should refer to the syntactic category the subject Secondly there is
an independent process of i-deletion in chitumbuka Consider following examples
(31) i-deletion and complex segment formation a ni (copula) -litali (big) --gt nilitali or nditali laquoit) is big) b ni (subject agreement)- kulya --gt nikulya or Nkhulya laquol) eat)
Here again i is deleted though it is not followed by another vowel I will tentatively
formulate the i-deletion rule as in (32)
(32) i-deletion i --gt 0_ [ (where [ is a morphological word domain)
This rule as (31a) (31 b) and (30f) shows is optional Then what if the option is not taken
in (30) examples We might expect to see nyaluta in case of example (30a) Notice that the
14 )The irrelevant morpheme breaks are not shown here aluta can be further divided into a (past imperfective) -Iut (go)- a (FV)
15 )The form in the parenthesis means that it may be used by some But the other form is far more widely used and the informant prefers to use valuta
16 )Further investigation of phonological word derivation based on the syntactic information and morphological words should be made Here only the difference of phonological and morphological words is used without mentioning how different they are
-339shy
morphological word domain contains boundary and that the agreement markers begin with
coronal sounds If GF applies then GD in (24) will eliminate nyaluta
25 Unresolved vowel clashes
Not all the vowel sequences are resolved in Chitumbuka Some representative examples are
given in (33) and (34)
(33) V-final nouns and determiners a khuni Ili --gt khuni (y)ili khunyili khunili
tree this b buku ilo --gt buku (y)ilo bukwilo
book that c chipinda ico --gt chipinda (y)ico chipindico
room this
(34) Pronominal objects and V-initial verbs 171
a u-ku-mu-at-a --gt wakumu(y)ata wakumwata (you created him)
b wa-ku-ti-at-a --gt wakuti (y)ata wakutyata (He created you)
c ni-ku-tu-och-a --gt nikuku(w)ocha nikukocha nikukwocha (I roasted them)
d mu-ku-ka-eNg-a --gt mukuka(y)eNga muku (I melted it)
This seems to tell us that the main domain of hiatus resolving is within phonological words
GF and VD do not operate on this level and only G1 optionally shows up For further discussion
the phonological word boundaries are represented with parentheses in (35)
17 )The detailed breakdowns of the strings are as under
(yOU) ~u (him) tt (create)tni (me) ku + 11 (~e) + och (bake) + a (FV) wa (heher ka (It) Ng (melt)
u (them) verb stem
subject verb pronominal agreement tense objects
-340shy
Hiatus Resolving in Chitumbuka 17
(35) a (khuni) (ili) b (niku) (tu) (ocha) c (nikuwocha) (tuyuni)
One very interesting observation here is the change in the phonological word domains
(35b) and (35c) are semantically same The object in (35b) is pronominal while (35c) has a noun
object As a result of movement two phonological words are broken into three phonological
words This is not an arbitrary boundary just to explain the lack of hiatus resolving We may find
a very similar situation in English Intonational Phrase parsing
(36) a (As you know) (the tall boy is a genius) b (The tall boy) (as you know) (is a genius)
Nespor and Vogels (1986189) intonational phrase formation rule produces (36a) as well
as (36b) The syntax dependency of the phonological word formation is partly responsible for the
difference between (35b) and (35c)
Further hiatus is not the only rule sensitive to boundary difference in Chitumbuka We can
find that i-deletion and complex segment formation discussed in the preceding subsection shows
the same phenomenon Consider the following example (parentheses indicate the phonological
word domain)
(37) a (niku) (mu) (temwa) I love him Nkhu
b (uku) (ni)(temwa) You love me nthe
c (khunilClllno) the tree here mpha
One final observation is that our rules so far can not explain why glides can be inserted in
the examples of (33) and (34) We have already pointed out the problem of GI in 21 The
additional complication is that even if we posit glides as a phoneme we still have to revise GIor
-341shy
make another one to explain post-lexical GL This will be addressed in 32
3 Syllable phonology of hiatus resol ving
In the discussion in section 2 some of the problems in the linear analysis were detected
(What are they ) In this section I will go over them and reanalyze the data with the help
of syllable structure to see whether we can answer some of the questions left open McCarthy and
Prince (1986) style of moraic syllable structure will be employed for our discussion in this section
The basic syllable structure of Chitumbuka is given in (38)
(38) Syllable structure in Chitumbuka a a b a c a
~~((r r1T ( (I I 1 All the syllables are mora final 2 All the syllables are mono-moraic
This is not the template that is required in the underlying form as will be discussed
Therefore the non-moraic tier may not have to be satisfied In other words though the structure
shows two pre-moraic consonant slots these can be left empty The two branches simple mean the
maximum number of consonants that can come before a vowel
18)1n the post lexical level this structural condition may not hold In Chitumbuka vowels in the stressed syllable is lengthened The long vowel is represented with bimoraic syllable structure It is not uncommon that some of the rules or conditions are applied only on the lexical level but not on the post lexical level such as extrametricality stress assignment and even VD and GF we talked about in the preceding section
- 342-shy
Hiatus Resolving in Chitumbuka 19
31 Syllabification
Contrary to the claim that all the segments are already syllabified in the base form (cf
Selkirk (1982raquo I assume that syllable structure is assigned through the syllabification process
The syllabification principles in Chitumbuka are given under
(39) Syllabification in Chitumbuka i) Morafication A segment projects a mora if it is more sonorous than the following
segrnent ii) Syllable projection Each mora projects a syllable node iii) Segment incorporation Unsyllabified segments are incorporated into syllables
Now lets take a look at the following sample syllabification
(40) a
0 0 00
I (iii)I(ii) --(i) f1 f1 f1 f1 ~ rlrOI I
~
I I n i u mb a n 1 u mb a n 1 u mb a n i u mb a
b
0 o 0 0 o 0
I I I (iW (~Of1 f1 f1 f1 ~ ~ I I r~r I I I I I ch i m a i ch i m a ch i m a ch i m a i
Notice that we do not need GF as a rule the same result of GF application is achieved as a
19)1 suppose that morafication principle is universally dependent upon the sonority hierarchy There are three different strategies in using sonority in syllabifications
i) y is moraic in xy if it is more sonorous than x ii) y is moraic in yz if it is more sonorous than z iii) y is moraic if its sonority is greater than N
The language in the world employ one or two all all of these in the syllabification and language may slightly vary in setting up the value of N in (iii)
-- 343shy
natural consequence of syllabification In 21 we obtain the simplicity of phonemic inventory at
the cost of introducing GF thereby complicating the rule component The environment of GF is
reflected in the syllabification principle Of course syllabification can not be an additional burden
the language needs syllable structure anyhow for example for the application of many phonological
rules and even for utterance
The nature of glide sounds are clearly reflected in the syllable structure If a vowel is
dominated by a mora it will be interpreted as a full vowel But if it is dominated by the syllable
node directly it will surface as a glide
32 Syllabification domain
The syllabification principles introduced in 31 have presupposed that the string of segments
are not syllabified in the underlying representation Suppose every morpheme is syllabified in the
deep structure and consider the following example of word formation
(41)
a aa a
~ (I (noa k 0 V 1 a k 0
Now we will have to make a rule for mora deletion which will be almost the repetition of
the syllabification principle Syllabification process in the course of derivation can get rid of the
redundant repetition of syllable structural description as well as save the later-on application of mora
deletion
I assume that the syllabification takes place in the first derived domain In other words
segments are syllabified after the initial word formation process Before going into the detailed
-344shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
h li - inu inu (yours-pl)
Before going into any detail I should say that (16g) and (16h) may be problematic GF can
be applied but (16h) does not show the form with a glide Vail (197240) observed that mu-
alternates with m or syllabic m I will simply assume that this morphemic alternation is
responsible for the variations in (16g)61 It is not the application of different rules but the
coexistence of mu- and m- form that results in the two different surface forms in (16g)
We can make very interesting observations here the given data except for (16g) are similar
to (11) and (12) in that there is no variant form and in that there is a glide formation process One
difference is that no case of VD is witnessed in (11) and (12) while the data in (16) and (17)
presents very productive use of VD and do not show any instances of GI
All the examples presented show that when VI and V2 get together it is always the VI that
is deleted V2 remains undeleted This may not necessary mean that V2 triggers vowel deletion To
go deeper into VD phenomenon lets first consider environments for vowel deletion
(18) Vowel deletion environments a VI is less sonorous than V2 (16e) (16h) (l7d) (17e) b VI is more sonorous than V2 (17f) c VI andVj are identical (16f) (17g) (17h)
Apparently (18) covers almost all the possible combinations except i-u and u-i
sequences First of all we can see that (18a) shares similar environments with GF in (14) And
actually GF applies to such examples as in (16c) (16d) (17b) and (17e) Just saying that both GF
and VD are optional (cf Mphande (1989121)) and that they share same the same environments
6) I think that m- might have come from mu- through u-deletion This may be a historical change since it is very difficult to find the synchronic evidence that m- is related to mu- as in the case of mpaka (until) Then we may conclude that both forms are available for (16g) but only mshy is available for (16h)
-332~
9 Hiatus Resolving in Chitumbuka
does not explain why there is no variation
One may say that the difference in boundary type results in the application of the different
rules and add the boundary in the rule-description accordingly This can be illustrated as in (19)
(19) a ku - Ii - a --gt kulya b Ii - ake --gt lake
Though (19a) and (19b) shows the same sequence of i-a IIi in (19a) is a stem while it is
a prefix in (19b) And lal in (19a) is a grammatical suffix and lal in (19b) is a part of a lexical
morpheme In other words (19a) is suffixation and (19b) is prefixation By assigning different
boundaries we can explain why we have GI only in (11) and (12) and not in (16) and (17)
One example of the boundary functioning in conditioning rules may be found in English g-
deletion as in Hyman (1976197) According to them there are two different I-erl suffixes in
English The first one is introduced with word boundary as in singer Another is introduced
with morpheme boundary + as in long-er and English has a g-deletion rule that goes as under
(20) g-deletion in English g --gt 0N_ (N a velar nasal)
This rule with the boundary differentiation and with the specification of the boundary type
deletes IgI in singer but not in longer
The same strategy can be employed here We may assume that the stems are introduced
with one type of boundary say and the suffixes with another + we may represent (19) as
(21)7
7)1 will simply assume that suffixes are introduced with + boundary and prefixes are introduced with boundary But I will continue to use - to mark the morpheme break The boundary differentiation according to Aronoff (1976) means that suffixation preceded prefixation
-- 333shy
(21) aku li + a b li aka
Now we can formulate the following VD rule
(22) Vowel deletion Vi --gt 0 I Vz
This seems to be a reasonable approach I believe this may be partly true for (17) cases
But we need something else This does not help explaining (16) because we have to explain why
GF instead of VD is applied to such examples as (16a) to (16d) and (17a) to (17c)
One way out from this dilemma is to stick to glide formation for (16) then make a rule that
deletes the glide What is meant by this is that it is not VD but GF followed by GD in (16) as
shown in (23)
(23) li - ake --(GF)--gt lyake --(GD)--gt lake
This process is motivated by the fact that in all the cases of VD in GF environments Iii is
preceded by a coronal consonant as shown in (16e) (17d) and (17e) Notice also that the recent
works on features such as Sagey (1986) McCarthy (1988) and many others claim that palatal
glides have [coronal] feature Given this we may formulate the following GD rule
(24) Glide Deletion [+coronal] --gt 0 I [+ coronal]
The rule is maximally generalized We dont have to specify whether the sound deleted is a
glide The syllable structure given in section 1 will naturally predict that it is a glide since in C1C2
consequence the Cz must be a glide But still the boundary is necessary to prevent GD to such
words as ku li + a
Now lets take a look at (17) cases Here again we find that the environment of VD is
-334~
Hiatus Resolving in Chitumbuka 11
very similar to GI in (15) We will have to resort to the boundary distinction as proposed in
discussion with the sonority condition as in (25)
(25) Vowel deletion V1--gt0_V2 (conditionV1 is not less sonorous than V2)
The condition does not necessarily require a higher sonority for the first vowel This
reflects an effort to explain (17g) (17h) cases together with (17f) This approach however has
one immediate problem of explaining (16a) and (16d) VD as in (25) can apply to these cases but
mini and Tizxa are ill-formed One major observational difference of the examples to which
VD is applied is the roundness feature of the two vowels When two vowels have the same
roundness feature and the same height the first one deletes And GF applies when the two vowels
are of the same height (and therefore the same sonority) but differ in roundness This problem can
be solved by adding roundness agreement condition to VD as in (26)81
(26) Vowel deletion - revised VI __ gt 0 __ V2
[a round] [a round] (conditionVIis not less sonorous than V2)
23 Between V-final affixes and V-initial Verbs
There are five different types of verb prefixes They are Iku-I Iliku-I Ngu Izamu-I Ika-I
and le-I I called these prefixes not agreement markers The reason is two-fold first they have their
own meanings These prefixes express tenses or aspects of the verb And unlike agreement
8 lAnother possibility that is imposing roundness disagreement condition on the GF rule makes a wrong prediction that lyenge in (12c) and kumanyeka in (12d) should not undergo GF which is contradictory with the given data
9 )The followings are the general meaning of these prefixes a ku- Present tense present progressive or habitual aspect of the present
-335shy
12 7j-iilfHl ~1lyenlt(1993 2)
markers in 21 or 22 they do not carry grammatical information Further we find that agreement
markers are added before these prefixes in the sentences
Lets begin the discussion by looking at the following data
(27) Prefixes with final lui and V-initial verbs a ku shy amba kwamba kuyamba (to start) b ku - ambuka kwambuka kuyambuka (to cross) c ku - ata kwata kuyata (to create) d ku shy enda kwenda kuyenda (to walk) e ku shy eNga kweNga kuyeNga (to melt) f ku - imba kwimba kuyimba (to sing) g ku - ima kwima kuyima (to stand) h ku - iba kwiba (to steal) I ku - iza kw iza (to come) J ku shy ocha kocha kuwocha (to burn) k l
ku - opa ku - ona
kopa ku wopaku tJY(1Y~- Hu
(to fear) in ~f~Ofmiddot
ffi ku -ma KU wura (to fear) n ku - umba kuwurnba (to mold something) o ku - undira kuwundira (to bury)
(28)Tense or aspect marker and V-initial verbs a ka - imba kimba kayimba (sang) b ka - iba kiba (stole) c ka - enda kenda kayenda (walked) d ka - opa kawopa (feared) e ka - ona kawona (saw) f ka - undira kawundira (buried) g ka - amba kamba kayamba (started)
The first thing readily noticeable in the given data is the existence of different alternatives in
b liku- Perfective the state of verb or the result of activity in the past is carried on to the present
c zamu- Future tense d ka- distant past (activity in the past) e kaNgu-habitual activity in the past f a- Activity in the immediate past g aNgu- Duration of activity that ended in the immediate past
lO)Though all the examples given here have Iku-I prefix the exactly same phonological changes are observed with all other prefixes that end in lui such as INgu-l liku-j and zamu-z All the V-initial verbs elicited so far are given here
-336shy
Hiatus Resolving in Chitumbuka 13
most of the cases Lets first begin with the apparently exceptional cases that do not have
variations Our Chitumbuka informant was not sure whether the stems of (27h) and (27i) begin
with Iii or not On a couple of occasions he said that to steal is kuba And even when he
pronounced kwiba he still maintained that the stem is Ibal instead of iba Though without any
sufficient supporting data I take this as the evidence that the root form of to steal has lost the
alternations and fossilized into kwiba and the same might be true for (27i)
Next it is very strange that only (271) shows different behavior in comparison with the two
preceding examples (27j) and (27k) It may be extremely difficult to assume that (271) has lost the
alternations and fossilized into kuwona form Intuitively if it is to be fossilized the form should
have been kona a phonologically simpler form I suggest that the base form of to see is uon shy
a GF is applied to this to produce Iwonal and then there is no vowel clash when Iku-I is added
failing to meet the VD environment
(27m) (27n) and (270) look quite exceptional in that VD does not apply to theses examples
though they are in VD environments I do not have systematic explanation for these There may he
some historical change in these words III or the underlying forms may begin with luul and the GF
changes the first lui into glide before tense or aspect markers are added 121 or they are just
exceptional At this point I will leave this as an open question
The GF and VD are quite straight forward There is no need to revise the rules Now we
will have to think about GI cases GI rule given in (15) does not explain the examples given in this
11)For example their friend in Chitumbuka is munyawo or rnunyalo Here we sec the alternations hetween Iwland iraquo] The same thing might have happened in (27m) (27n) and (270) words we may speculate that historically those stems begin with lraquo] And the sound weakened into the glide Iw
12)50 far no two same segments have been found to be adjacent morpheme internally However given the GF rule environment I think this idea may not be totally speculative
-- 337shy
subsection the sonority condition is not satisfied
One possible linear phonology approach may be to resort to the rule ordering We may
eliminate the sonority condition of GI and simply order GI after GF Further GI should not be
applied in (16) and (17) and therefore VD should be placed before GI hence the following order
(29) Rule ordering 1 Glide formation 2 Vowel deletion 3 Glide insertion
The major problem with this approach is the nature of the rule The data (27) (28) indicate
that all the three rules are optional But even though they are optional there is no option of not
applying any of the rules and therefore the last rule (GI) should be obligatory Further the data
(11) tells us a different story GF seems to be obligatory Otherwise we may have to say that
(1Ia) can be realized either as imwe (GF) or as imuye (GI) Moreover (17) shows that VD is
obligatory to prevent the operation of GI Here is the paradox A rule is optional for certain group
of words and obligatory for the rest This may suggest that rule based linear approach may not be
suitable for the explanation of hiatus resolving in Chitumbuka Further theoretical investigation will
be made in 32
24 Subject agreement marker and past imperfective marker la-I
All the prefixes introduced in 21 and some other personal agreement markers are used in
the sentence to satisfy the subject agreement Meanwhile there is one V-initial tense marker in
chitumbuka a- And therefore here again we expect to find vowel clashes The representative
data are given in (30)
13) If any of the rule is obligatory then we should not see the variations in the data given in (27) and (28)
-338shy
Hiatus Resolving in Chitumbuka 15
(30) Subject agreement marker and past imperfective marker a ni - aluta naluta (I walked) b u - aluta waluta (you walked) c wa - aluta waluta (he walked) d ti - aluta taluta (we went) e li - aluta laluta (It went) f vi - aluta (vyaluta) valuta (They went)
Superficially the data given here looks like those in (17) There is no GI example And the
choice between GF and VD seems to be determined by the relative sonority of the vowels
However (30f) shows that there is something different here (30) may not be the extension of
examples given in (17) The reason is twofold first the examples given here are not lexical The
agreement marker assignment should refer to the syntactic category the subject Secondly there is
an independent process of i-deletion in chitumbuka Consider following examples
(31) i-deletion and complex segment formation a ni (copula) -litali (big) --gt nilitali or nditali laquoit) is big) b ni (subject agreement)- kulya --gt nikulya or Nkhulya laquol) eat)
Here again i is deleted though it is not followed by another vowel I will tentatively
formulate the i-deletion rule as in (32)
(32) i-deletion i --gt 0_ [ (where [ is a morphological word domain)
This rule as (31a) (31 b) and (30f) shows is optional Then what if the option is not taken
in (30) examples We might expect to see nyaluta in case of example (30a) Notice that the
14 )The irrelevant morpheme breaks are not shown here aluta can be further divided into a (past imperfective) -Iut (go)- a (FV)
15 )The form in the parenthesis means that it may be used by some But the other form is far more widely used and the informant prefers to use valuta
16 )Further investigation of phonological word derivation based on the syntactic information and morphological words should be made Here only the difference of phonological and morphological words is used without mentioning how different they are
-339shy
morphological word domain contains boundary and that the agreement markers begin with
coronal sounds If GF applies then GD in (24) will eliminate nyaluta
25 Unresolved vowel clashes
Not all the vowel sequences are resolved in Chitumbuka Some representative examples are
given in (33) and (34)
(33) V-final nouns and determiners a khuni Ili --gt khuni (y)ili khunyili khunili
tree this b buku ilo --gt buku (y)ilo bukwilo
book that c chipinda ico --gt chipinda (y)ico chipindico
room this
(34) Pronominal objects and V-initial verbs 171
a u-ku-mu-at-a --gt wakumu(y)ata wakumwata (you created him)
b wa-ku-ti-at-a --gt wakuti (y)ata wakutyata (He created you)
c ni-ku-tu-och-a --gt nikuku(w)ocha nikukocha nikukwocha (I roasted them)
d mu-ku-ka-eNg-a --gt mukuka(y)eNga muku (I melted it)
This seems to tell us that the main domain of hiatus resolving is within phonological words
GF and VD do not operate on this level and only G1 optionally shows up For further discussion
the phonological word boundaries are represented with parentheses in (35)
17 )The detailed breakdowns of the strings are as under
(yOU) ~u (him) tt (create)tni (me) ku + 11 (~e) + och (bake) + a (FV) wa (heher ka (It) Ng (melt)
u (them) verb stem
subject verb pronominal agreement tense objects
-340shy
Hiatus Resolving in Chitumbuka 17
(35) a (khuni) (ili) b (niku) (tu) (ocha) c (nikuwocha) (tuyuni)
One very interesting observation here is the change in the phonological word domains
(35b) and (35c) are semantically same The object in (35b) is pronominal while (35c) has a noun
object As a result of movement two phonological words are broken into three phonological
words This is not an arbitrary boundary just to explain the lack of hiatus resolving We may find
a very similar situation in English Intonational Phrase parsing
(36) a (As you know) (the tall boy is a genius) b (The tall boy) (as you know) (is a genius)
Nespor and Vogels (1986189) intonational phrase formation rule produces (36a) as well
as (36b) The syntax dependency of the phonological word formation is partly responsible for the
difference between (35b) and (35c)
Further hiatus is not the only rule sensitive to boundary difference in Chitumbuka We can
find that i-deletion and complex segment formation discussed in the preceding subsection shows
the same phenomenon Consider the following example (parentheses indicate the phonological
word domain)
(37) a (niku) (mu) (temwa) I love him Nkhu
b (uku) (ni)(temwa) You love me nthe
c (khunilClllno) the tree here mpha
One final observation is that our rules so far can not explain why glides can be inserted in
the examples of (33) and (34) We have already pointed out the problem of GI in 21 The
additional complication is that even if we posit glides as a phoneme we still have to revise GIor
-341shy
make another one to explain post-lexical GL This will be addressed in 32
3 Syllable phonology of hiatus resol ving
In the discussion in section 2 some of the problems in the linear analysis were detected
(What are they ) In this section I will go over them and reanalyze the data with the help
of syllable structure to see whether we can answer some of the questions left open McCarthy and
Prince (1986) style of moraic syllable structure will be employed for our discussion in this section
The basic syllable structure of Chitumbuka is given in (38)
(38) Syllable structure in Chitumbuka a a b a c a
~~((r r1T ( (I I 1 All the syllables are mora final 2 All the syllables are mono-moraic
This is not the template that is required in the underlying form as will be discussed
Therefore the non-moraic tier may not have to be satisfied In other words though the structure
shows two pre-moraic consonant slots these can be left empty The two branches simple mean the
maximum number of consonants that can come before a vowel
18)1n the post lexical level this structural condition may not hold In Chitumbuka vowels in the stressed syllable is lengthened The long vowel is represented with bimoraic syllable structure It is not uncommon that some of the rules or conditions are applied only on the lexical level but not on the post lexical level such as extrametricality stress assignment and even VD and GF we talked about in the preceding section
- 342-shy
Hiatus Resolving in Chitumbuka 19
31 Syllabification
Contrary to the claim that all the segments are already syllabified in the base form (cf
Selkirk (1982raquo I assume that syllable structure is assigned through the syllabification process
The syllabification principles in Chitumbuka are given under
(39) Syllabification in Chitumbuka i) Morafication A segment projects a mora if it is more sonorous than the following
segrnent ii) Syllable projection Each mora projects a syllable node iii) Segment incorporation Unsyllabified segments are incorporated into syllables
Now lets take a look at the following sample syllabification
(40) a
0 0 00
I (iii)I(ii) --(i) f1 f1 f1 f1 ~ rlrOI I
~
I I n i u mb a n 1 u mb a n 1 u mb a n i u mb a
b
0 o 0 0 o 0
I I I (iW (~Of1 f1 f1 f1 ~ ~ I I r~r I I I I I ch i m a i ch i m a ch i m a ch i m a i
Notice that we do not need GF as a rule the same result of GF application is achieved as a
19)1 suppose that morafication principle is universally dependent upon the sonority hierarchy There are three different strategies in using sonority in syllabifications
i) y is moraic in xy if it is more sonorous than x ii) y is moraic in yz if it is more sonorous than z iii) y is moraic if its sonority is greater than N
The language in the world employ one or two all all of these in the syllabification and language may slightly vary in setting up the value of N in (iii)
-- 343shy
natural consequence of syllabification In 21 we obtain the simplicity of phonemic inventory at
the cost of introducing GF thereby complicating the rule component The environment of GF is
reflected in the syllabification principle Of course syllabification can not be an additional burden
the language needs syllable structure anyhow for example for the application of many phonological
rules and even for utterance
The nature of glide sounds are clearly reflected in the syllable structure If a vowel is
dominated by a mora it will be interpreted as a full vowel But if it is dominated by the syllable
node directly it will surface as a glide
32 Syllabification domain
The syllabification principles introduced in 31 have presupposed that the string of segments
are not syllabified in the underlying representation Suppose every morpheme is syllabified in the
deep structure and consider the following example of word formation
(41)
a aa a
~ (I (noa k 0 V 1 a k 0
Now we will have to make a rule for mora deletion which will be almost the repetition of
the syllabification principle Syllabification process in the course of derivation can get rid of the
redundant repetition of syllable structural description as well as save the later-on application of mora
deletion
I assume that the syllabification takes place in the first derived domain In other words
segments are syllabified after the initial word formation process Before going into the detailed
-344shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
9 Hiatus Resolving in Chitumbuka
does not explain why there is no variation
One may say that the difference in boundary type results in the application of the different
rules and add the boundary in the rule-description accordingly This can be illustrated as in (19)
(19) a ku - Ii - a --gt kulya b Ii - ake --gt lake
Though (19a) and (19b) shows the same sequence of i-a IIi in (19a) is a stem while it is
a prefix in (19b) And lal in (19a) is a grammatical suffix and lal in (19b) is a part of a lexical
morpheme In other words (19a) is suffixation and (19b) is prefixation By assigning different
boundaries we can explain why we have GI only in (11) and (12) and not in (16) and (17)
One example of the boundary functioning in conditioning rules may be found in English g-
deletion as in Hyman (1976197) According to them there are two different I-erl suffixes in
English The first one is introduced with word boundary as in singer Another is introduced
with morpheme boundary + as in long-er and English has a g-deletion rule that goes as under
(20) g-deletion in English g --gt 0N_ (N a velar nasal)
This rule with the boundary differentiation and with the specification of the boundary type
deletes IgI in singer but not in longer
The same strategy can be employed here We may assume that the stems are introduced
with one type of boundary say and the suffixes with another + we may represent (19) as
(21)7
7)1 will simply assume that suffixes are introduced with + boundary and prefixes are introduced with boundary But I will continue to use - to mark the morpheme break The boundary differentiation according to Aronoff (1976) means that suffixation preceded prefixation
-- 333shy
(21) aku li + a b li aka
Now we can formulate the following VD rule
(22) Vowel deletion Vi --gt 0 I Vz
This seems to be a reasonable approach I believe this may be partly true for (17) cases
But we need something else This does not help explaining (16) because we have to explain why
GF instead of VD is applied to such examples as (16a) to (16d) and (17a) to (17c)
One way out from this dilemma is to stick to glide formation for (16) then make a rule that
deletes the glide What is meant by this is that it is not VD but GF followed by GD in (16) as
shown in (23)
(23) li - ake --(GF)--gt lyake --(GD)--gt lake
This process is motivated by the fact that in all the cases of VD in GF environments Iii is
preceded by a coronal consonant as shown in (16e) (17d) and (17e) Notice also that the recent
works on features such as Sagey (1986) McCarthy (1988) and many others claim that palatal
glides have [coronal] feature Given this we may formulate the following GD rule
(24) Glide Deletion [+coronal] --gt 0 I [+ coronal]
The rule is maximally generalized We dont have to specify whether the sound deleted is a
glide The syllable structure given in section 1 will naturally predict that it is a glide since in C1C2
consequence the Cz must be a glide But still the boundary is necessary to prevent GD to such
words as ku li + a
Now lets take a look at (17) cases Here again we find that the environment of VD is
-334~
Hiatus Resolving in Chitumbuka 11
very similar to GI in (15) We will have to resort to the boundary distinction as proposed in
discussion with the sonority condition as in (25)
(25) Vowel deletion V1--gt0_V2 (conditionV1 is not less sonorous than V2)
The condition does not necessarily require a higher sonority for the first vowel This
reflects an effort to explain (17g) (17h) cases together with (17f) This approach however has
one immediate problem of explaining (16a) and (16d) VD as in (25) can apply to these cases but
mini and Tizxa are ill-formed One major observational difference of the examples to which
VD is applied is the roundness feature of the two vowels When two vowels have the same
roundness feature and the same height the first one deletes And GF applies when the two vowels
are of the same height (and therefore the same sonority) but differ in roundness This problem can
be solved by adding roundness agreement condition to VD as in (26)81
(26) Vowel deletion - revised VI __ gt 0 __ V2
[a round] [a round] (conditionVIis not less sonorous than V2)
23 Between V-final affixes and V-initial Verbs
There are five different types of verb prefixes They are Iku-I Iliku-I Ngu Izamu-I Ika-I
and le-I I called these prefixes not agreement markers The reason is two-fold first they have their
own meanings These prefixes express tenses or aspects of the verb And unlike agreement
8 lAnother possibility that is imposing roundness disagreement condition on the GF rule makes a wrong prediction that lyenge in (12c) and kumanyeka in (12d) should not undergo GF which is contradictory with the given data
9 )The followings are the general meaning of these prefixes a ku- Present tense present progressive or habitual aspect of the present
-335shy
12 7j-iilfHl ~1lyenlt(1993 2)
markers in 21 or 22 they do not carry grammatical information Further we find that agreement
markers are added before these prefixes in the sentences
Lets begin the discussion by looking at the following data
(27) Prefixes with final lui and V-initial verbs a ku shy amba kwamba kuyamba (to start) b ku - ambuka kwambuka kuyambuka (to cross) c ku - ata kwata kuyata (to create) d ku shy enda kwenda kuyenda (to walk) e ku shy eNga kweNga kuyeNga (to melt) f ku - imba kwimba kuyimba (to sing) g ku - ima kwima kuyima (to stand) h ku - iba kwiba (to steal) I ku - iza kw iza (to come) J ku shy ocha kocha kuwocha (to burn) k l
ku - opa ku - ona
kopa ku wopaku tJY(1Y~- Hu
(to fear) in ~f~Ofmiddot
ffi ku -ma KU wura (to fear) n ku - umba kuwurnba (to mold something) o ku - undira kuwundira (to bury)
(28)Tense or aspect marker and V-initial verbs a ka - imba kimba kayimba (sang) b ka - iba kiba (stole) c ka - enda kenda kayenda (walked) d ka - opa kawopa (feared) e ka - ona kawona (saw) f ka - undira kawundira (buried) g ka - amba kamba kayamba (started)
The first thing readily noticeable in the given data is the existence of different alternatives in
b liku- Perfective the state of verb or the result of activity in the past is carried on to the present
c zamu- Future tense d ka- distant past (activity in the past) e kaNgu-habitual activity in the past f a- Activity in the immediate past g aNgu- Duration of activity that ended in the immediate past
lO)Though all the examples given here have Iku-I prefix the exactly same phonological changes are observed with all other prefixes that end in lui such as INgu-l liku-j and zamu-z All the V-initial verbs elicited so far are given here
-336shy
Hiatus Resolving in Chitumbuka 13
most of the cases Lets first begin with the apparently exceptional cases that do not have
variations Our Chitumbuka informant was not sure whether the stems of (27h) and (27i) begin
with Iii or not On a couple of occasions he said that to steal is kuba And even when he
pronounced kwiba he still maintained that the stem is Ibal instead of iba Though without any
sufficient supporting data I take this as the evidence that the root form of to steal has lost the
alternations and fossilized into kwiba and the same might be true for (27i)
Next it is very strange that only (271) shows different behavior in comparison with the two
preceding examples (27j) and (27k) It may be extremely difficult to assume that (271) has lost the
alternations and fossilized into kuwona form Intuitively if it is to be fossilized the form should
have been kona a phonologically simpler form I suggest that the base form of to see is uon shy
a GF is applied to this to produce Iwonal and then there is no vowel clash when Iku-I is added
failing to meet the VD environment
(27m) (27n) and (270) look quite exceptional in that VD does not apply to theses examples
though they are in VD environments I do not have systematic explanation for these There may he
some historical change in these words III or the underlying forms may begin with luul and the GF
changes the first lui into glide before tense or aspect markers are added 121 or they are just
exceptional At this point I will leave this as an open question
The GF and VD are quite straight forward There is no need to revise the rules Now we
will have to think about GI cases GI rule given in (15) does not explain the examples given in this
11)For example their friend in Chitumbuka is munyawo or rnunyalo Here we sec the alternations hetween Iwland iraquo] The same thing might have happened in (27m) (27n) and (270) words we may speculate that historically those stems begin with lraquo] And the sound weakened into the glide Iw
12)50 far no two same segments have been found to be adjacent morpheme internally However given the GF rule environment I think this idea may not be totally speculative
-- 337shy
subsection the sonority condition is not satisfied
One possible linear phonology approach may be to resort to the rule ordering We may
eliminate the sonority condition of GI and simply order GI after GF Further GI should not be
applied in (16) and (17) and therefore VD should be placed before GI hence the following order
(29) Rule ordering 1 Glide formation 2 Vowel deletion 3 Glide insertion
The major problem with this approach is the nature of the rule The data (27) (28) indicate
that all the three rules are optional But even though they are optional there is no option of not
applying any of the rules and therefore the last rule (GI) should be obligatory Further the data
(11) tells us a different story GF seems to be obligatory Otherwise we may have to say that
(1Ia) can be realized either as imwe (GF) or as imuye (GI) Moreover (17) shows that VD is
obligatory to prevent the operation of GI Here is the paradox A rule is optional for certain group
of words and obligatory for the rest This may suggest that rule based linear approach may not be
suitable for the explanation of hiatus resolving in Chitumbuka Further theoretical investigation will
be made in 32
24 Subject agreement marker and past imperfective marker la-I
All the prefixes introduced in 21 and some other personal agreement markers are used in
the sentence to satisfy the subject agreement Meanwhile there is one V-initial tense marker in
chitumbuka a- And therefore here again we expect to find vowel clashes The representative
data are given in (30)
13) If any of the rule is obligatory then we should not see the variations in the data given in (27) and (28)
-338shy
Hiatus Resolving in Chitumbuka 15
(30) Subject agreement marker and past imperfective marker a ni - aluta naluta (I walked) b u - aluta waluta (you walked) c wa - aluta waluta (he walked) d ti - aluta taluta (we went) e li - aluta laluta (It went) f vi - aluta (vyaluta) valuta (They went)
Superficially the data given here looks like those in (17) There is no GI example And the
choice between GF and VD seems to be determined by the relative sonority of the vowels
However (30f) shows that there is something different here (30) may not be the extension of
examples given in (17) The reason is twofold first the examples given here are not lexical The
agreement marker assignment should refer to the syntactic category the subject Secondly there is
an independent process of i-deletion in chitumbuka Consider following examples
(31) i-deletion and complex segment formation a ni (copula) -litali (big) --gt nilitali or nditali laquoit) is big) b ni (subject agreement)- kulya --gt nikulya or Nkhulya laquol) eat)
Here again i is deleted though it is not followed by another vowel I will tentatively
formulate the i-deletion rule as in (32)
(32) i-deletion i --gt 0_ [ (where [ is a morphological word domain)
This rule as (31a) (31 b) and (30f) shows is optional Then what if the option is not taken
in (30) examples We might expect to see nyaluta in case of example (30a) Notice that the
14 )The irrelevant morpheme breaks are not shown here aluta can be further divided into a (past imperfective) -Iut (go)- a (FV)
15 )The form in the parenthesis means that it may be used by some But the other form is far more widely used and the informant prefers to use valuta
16 )Further investigation of phonological word derivation based on the syntactic information and morphological words should be made Here only the difference of phonological and morphological words is used without mentioning how different they are
-339shy
morphological word domain contains boundary and that the agreement markers begin with
coronal sounds If GF applies then GD in (24) will eliminate nyaluta
25 Unresolved vowel clashes
Not all the vowel sequences are resolved in Chitumbuka Some representative examples are
given in (33) and (34)
(33) V-final nouns and determiners a khuni Ili --gt khuni (y)ili khunyili khunili
tree this b buku ilo --gt buku (y)ilo bukwilo
book that c chipinda ico --gt chipinda (y)ico chipindico
room this
(34) Pronominal objects and V-initial verbs 171
a u-ku-mu-at-a --gt wakumu(y)ata wakumwata (you created him)
b wa-ku-ti-at-a --gt wakuti (y)ata wakutyata (He created you)
c ni-ku-tu-och-a --gt nikuku(w)ocha nikukocha nikukwocha (I roasted them)
d mu-ku-ka-eNg-a --gt mukuka(y)eNga muku (I melted it)
This seems to tell us that the main domain of hiatus resolving is within phonological words
GF and VD do not operate on this level and only G1 optionally shows up For further discussion
the phonological word boundaries are represented with parentheses in (35)
17 )The detailed breakdowns of the strings are as under
(yOU) ~u (him) tt (create)tni (me) ku + 11 (~e) + och (bake) + a (FV) wa (heher ka (It) Ng (melt)
u (them) verb stem
subject verb pronominal agreement tense objects
-340shy
Hiatus Resolving in Chitumbuka 17
(35) a (khuni) (ili) b (niku) (tu) (ocha) c (nikuwocha) (tuyuni)
One very interesting observation here is the change in the phonological word domains
(35b) and (35c) are semantically same The object in (35b) is pronominal while (35c) has a noun
object As a result of movement two phonological words are broken into three phonological
words This is not an arbitrary boundary just to explain the lack of hiatus resolving We may find
a very similar situation in English Intonational Phrase parsing
(36) a (As you know) (the tall boy is a genius) b (The tall boy) (as you know) (is a genius)
Nespor and Vogels (1986189) intonational phrase formation rule produces (36a) as well
as (36b) The syntax dependency of the phonological word formation is partly responsible for the
difference between (35b) and (35c)
Further hiatus is not the only rule sensitive to boundary difference in Chitumbuka We can
find that i-deletion and complex segment formation discussed in the preceding subsection shows
the same phenomenon Consider the following example (parentheses indicate the phonological
word domain)
(37) a (niku) (mu) (temwa) I love him Nkhu
b (uku) (ni)(temwa) You love me nthe
c (khunilClllno) the tree here mpha
One final observation is that our rules so far can not explain why glides can be inserted in
the examples of (33) and (34) We have already pointed out the problem of GI in 21 The
additional complication is that even if we posit glides as a phoneme we still have to revise GIor
-341shy
make another one to explain post-lexical GL This will be addressed in 32
3 Syllable phonology of hiatus resol ving
In the discussion in section 2 some of the problems in the linear analysis were detected
(What are they ) In this section I will go over them and reanalyze the data with the help
of syllable structure to see whether we can answer some of the questions left open McCarthy and
Prince (1986) style of moraic syllable structure will be employed for our discussion in this section
The basic syllable structure of Chitumbuka is given in (38)
(38) Syllable structure in Chitumbuka a a b a c a
~~((r r1T ( (I I 1 All the syllables are mora final 2 All the syllables are mono-moraic
This is not the template that is required in the underlying form as will be discussed
Therefore the non-moraic tier may not have to be satisfied In other words though the structure
shows two pre-moraic consonant slots these can be left empty The two branches simple mean the
maximum number of consonants that can come before a vowel
18)1n the post lexical level this structural condition may not hold In Chitumbuka vowels in the stressed syllable is lengthened The long vowel is represented with bimoraic syllable structure It is not uncommon that some of the rules or conditions are applied only on the lexical level but not on the post lexical level such as extrametricality stress assignment and even VD and GF we talked about in the preceding section
- 342-shy
Hiatus Resolving in Chitumbuka 19
31 Syllabification
Contrary to the claim that all the segments are already syllabified in the base form (cf
Selkirk (1982raquo I assume that syllable structure is assigned through the syllabification process
The syllabification principles in Chitumbuka are given under
(39) Syllabification in Chitumbuka i) Morafication A segment projects a mora if it is more sonorous than the following
segrnent ii) Syllable projection Each mora projects a syllable node iii) Segment incorporation Unsyllabified segments are incorporated into syllables
Now lets take a look at the following sample syllabification
(40) a
0 0 00
I (iii)I(ii) --(i) f1 f1 f1 f1 ~ rlrOI I
~
I I n i u mb a n 1 u mb a n 1 u mb a n i u mb a
b
0 o 0 0 o 0
I I I (iW (~Of1 f1 f1 f1 ~ ~ I I r~r I I I I I ch i m a i ch i m a ch i m a ch i m a i
Notice that we do not need GF as a rule the same result of GF application is achieved as a
19)1 suppose that morafication principle is universally dependent upon the sonority hierarchy There are three different strategies in using sonority in syllabifications
i) y is moraic in xy if it is more sonorous than x ii) y is moraic in yz if it is more sonorous than z iii) y is moraic if its sonority is greater than N
The language in the world employ one or two all all of these in the syllabification and language may slightly vary in setting up the value of N in (iii)
-- 343shy
natural consequence of syllabification In 21 we obtain the simplicity of phonemic inventory at
the cost of introducing GF thereby complicating the rule component The environment of GF is
reflected in the syllabification principle Of course syllabification can not be an additional burden
the language needs syllable structure anyhow for example for the application of many phonological
rules and even for utterance
The nature of glide sounds are clearly reflected in the syllable structure If a vowel is
dominated by a mora it will be interpreted as a full vowel But if it is dominated by the syllable
node directly it will surface as a glide
32 Syllabification domain
The syllabification principles introduced in 31 have presupposed that the string of segments
are not syllabified in the underlying representation Suppose every morpheme is syllabified in the
deep structure and consider the following example of word formation
(41)
a aa a
~ (I (noa k 0 V 1 a k 0
Now we will have to make a rule for mora deletion which will be almost the repetition of
the syllabification principle Syllabification process in the course of derivation can get rid of the
redundant repetition of syllable structural description as well as save the later-on application of mora
deletion
I assume that the syllabification takes place in the first derived domain In other words
segments are syllabified after the initial word formation process Before going into the detailed
-344shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
(21) aku li + a b li aka
Now we can formulate the following VD rule
(22) Vowel deletion Vi --gt 0 I Vz
This seems to be a reasonable approach I believe this may be partly true for (17) cases
But we need something else This does not help explaining (16) because we have to explain why
GF instead of VD is applied to such examples as (16a) to (16d) and (17a) to (17c)
One way out from this dilemma is to stick to glide formation for (16) then make a rule that
deletes the glide What is meant by this is that it is not VD but GF followed by GD in (16) as
shown in (23)
(23) li - ake --(GF)--gt lyake --(GD)--gt lake
This process is motivated by the fact that in all the cases of VD in GF environments Iii is
preceded by a coronal consonant as shown in (16e) (17d) and (17e) Notice also that the recent
works on features such as Sagey (1986) McCarthy (1988) and many others claim that palatal
glides have [coronal] feature Given this we may formulate the following GD rule
(24) Glide Deletion [+coronal] --gt 0 I [+ coronal]
The rule is maximally generalized We dont have to specify whether the sound deleted is a
glide The syllable structure given in section 1 will naturally predict that it is a glide since in C1C2
consequence the Cz must be a glide But still the boundary is necessary to prevent GD to such
words as ku li + a
Now lets take a look at (17) cases Here again we find that the environment of VD is
-334~
Hiatus Resolving in Chitumbuka 11
very similar to GI in (15) We will have to resort to the boundary distinction as proposed in
discussion with the sonority condition as in (25)
(25) Vowel deletion V1--gt0_V2 (conditionV1 is not less sonorous than V2)
The condition does not necessarily require a higher sonority for the first vowel This
reflects an effort to explain (17g) (17h) cases together with (17f) This approach however has
one immediate problem of explaining (16a) and (16d) VD as in (25) can apply to these cases but
mini and Tizxa are ill-formed One major observational difference of the examples to which
VD is applied is the roundness feature of the two vowels When two vowels have the same
roundness feature and the same height the first one deletes And GF applies when the two vowels
are of the same height (and therefore the same sonority) but differ in roundness This problem can
be solved by adding roundness agreement condition to VD as in (26)81
(26) Vowel deletion - revised VI __ gt 0 __ V2
[a round] [a round] (conditionVIis not less sonorous than V2)
23 Between V-final affixes and V-initial Verbs
There are five different types of verb prefixes They are Iku-I Iliku-I Ngu Izamu-I Ika-I
and le-I I called these prefixes not agreement markers The reason is two-fold first they have their
own meanings These prefixes express tenses or aspects of the verb And unlike agreement
8 lAnother possibility that is imposing roundness disagreement condition on the GF rule makes a wrong prediction that lyenge in (12c) and kumanyeka in (12d) should not undergo GF which is contradictory with the given data
9 )The followings are the general meaning of these prefixes a ku- Present tense present progressive or habitual aspect of the present
-335shy
12 7j-iilfHl ~1lyenlt(1993 2)
markers in 21 or 22 they do not carry grammatical information Further we find that agreement
markers are added before these prefixes in the sentences
Lets begin the discussion by looking at the following data
(27) Prefixes with final lui and V-initial verbs a ku shy amba kwamba kuyamba (to start) b ku - ambuka kwambuka kuyambuka (to cross) c ku - ata kwata kuyata (to create) d ku shy enda kwenda kuyenda (to walk) e ku shy eNga kweNga kuyeNga (to melt) f ku - imba kwimba kuyimba (to sing) g ku - ima kwima kuyima (to stand) h ku - iba kwiba (to steal) I ku - iza kw iza (to come) J ku shy ocha kocha kuwocha (to burn) k l
ku - opa ku - ona
kopa ku wopaku tJY(1Y~- Hu
(to fear) in ~f~Ofmiddot
ffi ku -ma KU wura (to fear) n ku - umba kuwurnba (to mold something) o ku - undira kuwundira (to bury)
(28)Tense or aspect marker and V-initial verbs a ka - imba kimba kayimba (sang) b ka - iba kiba (stole) c ka - enda kenda kayenda (walked) d ka - opa kawopa (feared) e ka - ona kawona (saw) f ka - undira kawundira (buried) g ka - amba kamba kayamba (started)
The first thing readily noticeable in the given data is the existence of different alternatives in
b liku- Perfective the state of verb or the result of activity in the past is carried on to the present
c zamu- Future tense d ka- distant past (activity in the past) e kaNgu-habitual activity in the past f a- Activity in the immediate past g aNgu- Duration of activity that ended in the immediate past
lO)Though all the examples given here have Iku-I prefix the exactly same phonological changes are observed with all other prefixes that end in lui such as INgu-l liku-j and zamu-z All the V-initial verbs elicited so far are given here
-336shy
Hiatus Resolving in Chitumbuka 13
most of the cases Lets first begin with the apparently exceptional cases that do not have
variations Our Chitumbuka informant was not sure whether the stems of (27h) and (27i) begin
with Iii or not On a couple of occasions he said that to steal is kuba And even when he
pronounced kwiba he still maintained that the stem is Ibal instead of iba Though without any
sufficient supporting data I take this as the evidence that the root form of to steal has lost the
alternations and fossilized into kwiba and the same might be true for (27i)
Next it is very strange that only (271) shows different behavior in comparison with the two
preceding examples (27j) and (27k) It may be extremely difficult to assume that (271) has lost the
alternations and fossilized into kuwona form Intuitively if it is to be fossilized the form should
have been kona a phonologically simpler form I suggest that the base form of to see is uon shy
a GF is applied to this to produce Iwonal and then there is no vowel clash when Iku-I is added
failing to meet the VD environment
(27m) (27n) and (270) look quite exceptional in that VD does not apply to theses examples
though they are in VD environments I do not have systematic explanation for these There may he
some historical change in these words III or the underlying forms may begin with luul and the GF
changes the first lui into glide before tense or aspect markers are added 121 or they are just
exceptional At this point I will leave this as an open question
The GF and VD are quite straight forward There is no need to revise the rules Now we
will have to think about GI cases GI rule given in (15) does not explain the examples given in this
11)For example their friend in Chitumbuka is munyawo or rnunyalo Here we sec the alternations hetween Iwland iraquo] The same thing might have happened in (27m) (27n) and (270) words we may speculate that historically those stems begin with lraquo] And the sound weakened into the glide Iw
12)50 far no two same segments have been found to be adjacent morpheme internally However given the GF rule environment I think this idea may not be totally speculative
-- 337shy
subsection the sonority condition is not satisfied
One possible linear phonology approach may be to resort to the rule ordering We may
eliminate the sonority condition of GI and simply order GI after GF Further GI should not be
applied in (16) and (17) and therefore VD should be placed before GI hence the following order
(29) Rule ordering 1 Glide formation 2 Vowel deletion 3 Glide insertion
The major problem with this approach is the nature of the rule The data (27) (28) indicate
that all the three rules are optional But even though they are optional there is no option of not
applying any of the rules and therefore the last rule (GI) should be obligatory Further the data
(11) tells us a different story GF seems to be obligatory Otherwise we may have to say that
(1Ia) can be realized either as imwe (GF) or as imuye (GI) Moreover (17) shows that VD is
obligatory to prevent the operation of GI Here is the paradox A rule is optional for certain group
of words and obligatory for the rest This may suggest that rule based linear approach may not be
suitable for the explanation of hiatus resolving in Chitumbuka Further theoretical investigation will
be made in 32
24 Subject agreement marker and past imperfective marker la-I
All the prefixes introduced in 21 and some other personal agreement markers are used in
the sentence to satisfy the subject agreement Meanwhile there is one V-initial tense marker in
chitumbuka a- And therefore here again we expect to find vowel clashes The representative
data are given in (30)
13) If any of the rule is obligatory then we should not see the variations in the data given in (27) and (28)
-338shy
Hiatus Resolving in Chitumbuka 15
(30) Subject agreement marker and past imperfective marker a ni - aluta naluta (I walked) b u - aluta waluta (you walked) c wa - aluta waluta (he walked) d ti - aluta taluta (we went) e li - aluta laluta (It went) f vi - aluta (vyaluta) valuta (They went)
Superficially the data given here looks like those in (17) There is no GI example And the
choice between GF and VD seems to be determined by the relative sonority of the vowels
However (30f) shows that there is something different here (30) may not be the extension of
examples given in (17) The reason is twofold first the examples given here are not lexical The
agreement marker assignment should refer to the syntactic category the subject Secondly there is
an independent process of i-deletion in chitumbuka Consider following examples
(31) i-deletion and complex segment formation a ni (copula) -litali (big) --gt nilitali or nditali laquoit) is big) b ni (subject agreement)- kulya --gt nikulya or Nkhulya laquol) eat)
Here again i is deleted though it is not followed by another vowel I will tentatively
formulate the i-deletion rule as in (32)
(32) i-deletion i --gt 0_ [ (where [ is a morphological word domain)
This rule as (31a) (31 b) and (30f) shows is optional Then what if the option is not taken
in (30) examples We might expect to see nyaluta in case of example (30a) Notice that the
14 )The irrelevant morpheme breaks are not shown here aluta can be further divided into a (past imperfective) -Iut (go)- a (FV)
15 )The form in the parenthesis means that it may be used by some But the other form is far more widely used and the informant prefers to use valuta
16 )Further investigation of phonological word derivation based on the syntactic information and morphological words should be made Here only the difference of phonological and morphological words is used without mentioning how different they are
-339shy
morphological word domain contains boundary and that the agreement markers begin with
coronal sounds If GF applies then GD in (24) will eliminate nyaluta
25 Unresolved vowel clashes
Not all the vowel sequences are resolved in Chitumbuka Some representative examples are
given in (33) and (34)
(33) V-final nouns and determiners a khuni Ili --gt khuni (y)ili khunyili khunili
tree this b buku ilo --gt buku (y)ilo bukwilo
book that c chipinda ico --gt chipinda (y)ico chipindico
room this
(34) Pronominal objects and V-initial verbs 171
a u-ku-mu-at-a --gt wakumu(y)ata wakumwata (you created him)
b wa-ku-ti-at-a --gt wakuti (y)ata wakutyata (He created you)
c ni-ku-tu-och-a --gt nikuku(w)ocha nikukocha nikukwocha (I roasted them)
d mu-ku-ka-eNg-a --gt mukuka(y)eNga muku (I melted it)
This seems to tell us that the main domain of hiatus resolving is within phonological words
GF and VD do not operate on this level and only G1 optionally shows up For further discussion
the phonological word boundaries are represented with parentheses in (35)
17 )The detailed breakdowns of the strings are as under
(yOU) ~u (him) tt (create)tni (me) ku + 11 (~e) + och (bake) + a (FV) wa (heher ka (It) Ng (melt)
u (them) verb stem
subject verb pronominal agreement tense objects
-340shy
Hiatus Resolving in Chitumbuka 17
(35) a (khuni) (ili) b (niku) (tu) (ocha) c (nikuwocha) (tuyuni)
One very interesting observation here is the change in the phonological word domains
(35b) and (35c) are semantically same The object in (35b) is pronominal while (35c) has a noun
object As a result of movement two phonological words are broken into three phonological
words This is not an arbitrary boundary just to explain the lack of hiatus resolving We may find
a very similar situation in English Intonational Phrase parsing
(36) a (As you know) (the tall boy is a genius) b (The tall boy) (as you know) (is a genius)
Nespor and Vogels (1986189) intonational phrase formation rule produces (36a) as well
as (36b) The syntax dependency of the phonological word formation is partly responsible for the
difference between (35b) and (35c)
Further hiatus is not the only rule sensitive to boundary difference in Chitumbuka We can
find that i-deletion and complex segment formation discussed in the preceding subsection shows
the same phenomenon Consider the following example (parentheses indicate the phonological
word domain)
(37) a (niku) (mu) (temwa) I love him Nkhu
b (uku) (ni)(temwa) You love me nthe
c (khunilClllno) the tree here mpha
One final observation is that our rules so far can not explain why glides can be inserted in
the examples of (33) and (34) We have already pointed out the problem of GI in 21 The
additional complication is that even if we posit glides as a phoneme we still have to revise GIor
-341shy
make another one to explain post-lexical GL This will be addressed in 32
3 Syllable phonology of hiatus resol ving
In the discussion in section 2 some of the problems in the linear analysis were detected
(What are they ) In this section I will go over them and reanalyze the data with the help
of syllable structure to see whether we can answer some of the questions left open McCarthy and
Prince (1986) style of moraic syllable structure will be employed for our discussion in this section
The basic syllable structure of Chitumbuka is given in (38)
(38) Syllable structure in Chitumbuka a a b a c a
~~((r r1T ( (I I 1 All the syllables are mora final 2 All the syllables are mono-moraic
This is not the template that is required in the underlying form as will be discussed
Therefore the non-moraic tier may not have to be satisfied In other words though the structure
shows two pre-moraic consonant slots these can be left empty The two branches simple mean the
maximum number of consonants that can come before a vowel
18)1n the post lexical level this structural condition may not hold In Chitumbuka vowels in the stressed syllable is lengthened The long vowel is represented with bimoraic syllable structure It is not uncommon that some of the rules or conditions are applied only on the lexical level but not on the post lexical level such as extrametricality stress assignment and even VD and GF we talked about in the preceding section
- 342-shy
Hiatus Resolving in Chitumbuka 19
31 Syllabification
Contrary to the claim that all the segments are already syllabified in the base form (cf
Selkirk (1982raquo I assume that syllable structure is assigned through the syllabification process
The syllabification principles in Chitumbuka are given under
(39) Syllabification in Chitumbuka i) Morafication A segment projects a mora if it is more sonorous than the following
segrnent ii) Syllable projection Each mora projects a syllable node iii) Segment incorporation Unsyllabified segments are incorporated into syllables
Now lets take a look at the following sample syllabification
(40) a
0 0 00
I (iii)I(ii) --(i) f1 f1 f1 f1 ~ rlrOI I
~
I I n i u mb a n 1 u mb a n 1 u mb a n i u mb a
b
0 o 0 0 o 0
I I I (iW (~Of1 f1 f1 f1 ~ ~ I I r~r I I I I I ch i m a i ch i m a ch i m a ch i m a i
Notice that we do not need GF as a rule the same result of GF application is achieved as a
19)1 suppose that morafication principle is universally dependent upon the sonority hierarchy There are three different strategies in using sonority in syllabifications
i) y is moraic in xy if it is more sonorous than x ii) y is moraic in yz if it is more sonorous than z iii) y is moraic if its sonority is greater than N
The language in the world employ one or two all all of these in the syllabification and language may slightly vary in setting up the value of N in (iii)
-- 343shy
natural consequence of syllabification In 21 we obtain the simplicity of phonemic inventory at
the cost of introducing GF thereby complicating the rule component The environment of GF is
reflected in the syllabification principle Of course syllabification can not be an additional burden
the language needs syllable structure anyhow for example for the application of many phonological
rules and even for utterance
The nature of glide sounds are clearly reflected in the syllable structure If a vowel is
dominated by a mora it will be interpreted as a full vowel But if it is dominated by the syllable
node directly it will surface as a glide
32 Syllabification domain
The syllabification principles introduced in 31 have presupposed that the string of segments
are not syllabified in the underlying representation Suppose every morpheme is syllabified in the
deep structure and consider the following example of word formation
(41)
a aa a
~ (I (noa k 0 V 1 a k 0
Now we will have to make a rule for mora deletion which will be almost the repetition of
the syllabification principle Syllabification process in the course of derivation can get rid of the
redundant repetition of syllable structural description as well as save the later-on application of mora
deletion
I assume that the syllabification takes place in the first derived domain In other words
segments are syllabified after the initial word formation process Before going into the detailed
-344shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
Hiatus Resolving in Chitumbuka 11
very similar to GI in (15) We will have to resort to the boundary distinction as proposed in
discussion with the sonority condition as in (25)
(25) Vowel deletion V1--gt0_V2 (conditionV1 is not less sonorous than V2)
The condition does not necessarily require a higher sonority for the first vowel This
reflects an effort to explain (17g) (17h) cases together with (17f) This approach however has
one immediate problem of explaining (16a) and (16d) VD as in (25) can apply to these cases but
mini and Tizxa are ill-formed One major observational difference of the examples to which
VD is applied is the roundness feature of the two vowels When two vowels have the same
roundness feature and the same height the first one deletes And GF applies when the two vowels
are of the same height (and therefore the same sonority) but differ in roundness This problem can
be solved by adding roundness agreement condition to VD as in (26)81
(26) Vowel deletion - revised VI __ gt 0 __ V2
[a round] [a round] (conditionVIis not less sonorous than V2)
23 Between V-final affixes and V-initial Verbs
There are five different types of verb prefixes They are Iku-I Iliku-I Ngu Izamu-I Ika-I
and le-I I called these prefixes not agreement markers The reason is two-fold first they have their
own meanings These prefixes express tenses or aspects of the verb And unlike agreement
8 lAnother possibility that is imposing roundness disagreement condition on the GF rule makes a wrong prediction that lyenge in (12c) and kumanyeka in (12d) should not undergo GF which is contradictory with the given data
9 )The followings are the general meaning of these prefixes a ku- Present tense present progressive or habitual aspect of the present
-335shy
12 7j-iilfHl ~1lyenlt(1993 2)
markers in 21 or 22 they do not carry grammatical information Further we find that agreement
markers are added before these prefixes in the sentences
Lets begin the discussion by looking at the following data
(27) Prefixes with final lui and V-initial verbs a ku shy amba kwamba kuyamba (to start) b ku - ambuka kwambuka kuyambuka (to cross) c ku - ata kwata kuyata (to create) d ku shy enda kwenda kuyenda (to walk) e ku shy eNga kweNga kuyeNga (to melt) f ku - imba kwimba kuyimba (to sing) g ku - ima kwima kuyima (to stand) h ku - iba kwiba (to steal) I ku - iza kw iza (to come) J ku shy ocha kocha kuwocha (to burn) k l
ku - opa ku - ona
kopa ku wopaku tJY(1Y~- Hu
(to fear) in ~f~Ofmiddot
ffi ku -ma KU wura (to fear) n ku - umba kuwurnba (to mold something) o ku - undira kuwundira (to bury)
(28)Tense or aspect marker and V-initial verbs a ka - imba kimba kayimba (sang) b ka - iba kiba (stole) c ka - enda kenda kayenda (walked) d ka - opa kawopa (feared) e ka - ona kawona (saw) f ka - undira kawundira (buried) g ka - amba kamba kayamba (started)
The first thing readily noticeable in the given data is the existence of different alternatives in
b liku- Perfective the state of verb or the result of activity in the past is carried on to the present
c zamu- Future tense d ka- distant past (activity in the past) e kaNgu-habitual activity in the past f a- Activity in the immediate past g aNgu- Duration of activity that ended in the immediate past
lO)Though all the examples given here have Iku-I prefix the exactly same phonological changes are observed with all other prefixes that end in lui such as INgu-l liku-j and zamu-z All the V-initial verbs elicited so far are given here
-336shy
Hiatus Resolving in Chitumbuka 13
most of the cases Lets first begin with the apparently exceptional cases that do not have
variations Our Chitumbuka informant was not sure whether the stems of (27h) and (27i) begin
with Iii or not On a couple of occasions he said that to steal is kuba And even when he
pronounced kwiba he still maintained that the stem is Ibal instead of iba Though without any
sufficient supporting data I take this as the evidence that the root form of to steal has lost the
alternations and fossilized into kwiba and the same might be true for (27i)
Next it is very strange that only (271) shows different behavior in comparison with the two
preceding examples (27j) and (27k) It may be extremely difficult to assume that (271) has lost the
alternations and fossilized into kuwona form Intuitively if it is to be fossilized the form should
have been kona a phonologically simpler form I suggest that the base form of to see is uon shy
a GF is applied to this to produce Iwonal and then there is no vowel clash when Iku-I is added
failing to meet the VD environment
(27m) (27n) and (270) look quite exceptional in that VD does not apply to theses examples
though they are in VD environments I do not have systematic explanation for these There may he
some historical change in these words III or the underlying forms may begin with luul and the GF
changes the first lui into glide before tense or aspect markers are added 121 or they are just
exceptional At this point I will leave this as an open question
The GF and VD are quite straight forward There is no need to revise the rules Now we
will have to think about GI cases GI rule given in (15) does not explain the examples given in this
11)For example their friend in Chitumbuka is munyawo or rnunyalo Here we sec the alternations hetween Iwland iraquo] The same thing might have happened in (27m) (27n) and (270) words we may speculate that historically those stems begin with lraquo] And the sound weakened into the glide Iw
12)50 far no two same segments have been found to be adjacent morpheme internally However given the GF rule environment I think this idea may not be totally speculative
-- 337shy
subsection the sonority condition is not satisfied
One possible linear phonology approach may be to resort to the rule ordering We may
eliminate the sonority condition of GI and simply order GI after GF Further GI should not be
applied in (16) and (17) and therefore VD should be placed before GI hence the following order
(29) Rule ordering 1 Glide formation 2 Vowel deletion 3 Glide insertion
The major problem with this approach is the nature of the rule The data (27) (28) indicate
that all the three rules are optional But even though they are optional there is no option of not
applying any of the rules and therefore the last rule (GI) should be obligatory Further the data
(11) tells us a different story GF seems to be obligatory Otherwise we may have to say that
(1Ia) can be realized either as imwe (GF) or as imuye (GI) Moreover (17) shows that VD is
obligatory to prevent the operation of GI Here is the paradox A rule is optional for certain group
of words and obligatory for the rest This may suggest that rule based linear approach may not be
suitable for the explanation of hiatus resolving in Chitumbuka Further theoretical investigation will
be made in 32
24 Subject agreement marker and past imperfective marker la-I
All the prefixes introduced in 21 and some other personal agreement markers are used in
the sentence to satisfy the subject agreement Meanwhile there is one V-initial tense marker in
chitumbuka a- And therefore here again we expect to find vowel clashes The representative
data are given in (30)
13) If any of the rule is obligatory then we should not see the variations in the data given in (27) and (28)
-338shy
Hiatus Resolving in Chitumbuka 15
(30) Subject agreement marker and past imperfective marker a ni - aluta naluta (I walked) b u - aluta waluta (you walked) c wa - aluta waluta (he walked) d ti - aluta taluta (we went) e li - aluta laluta (It went) f vi - aluta (vyaluta) valuta (They went)
Superficially the data given here looks like those in (17) There is no GI example And the
choice between GF and VD seems to be determined by the relative sonority of the vowels
However (30f) shows that there is something different here (30) may not be the extension of
examples given in (17) The reason is twofold first the examples given here are not lexical The
agreement marker assignment should refer to the syntactic category the subject Secondly there is
an independent process of i-deletion in chitumbuka Consider following examples
(31) i-deletion and complex segment formation a ni (copula) -litali (big) --gt nilitali or nditali laquoit) is big) b ni (subject agreement)- kulya --gt nikulya or Nkhulya laquol) eat)
Here again i is deleted though it is not followed by another vowel I will tentatively
formulate the i-deletion rule as in (32)
(32) i-deletion i --gt 0_ [ (where [ is a morphological word domain)
This rule as (31a) (31 b) and (30f) shows is optional Then what if the option is not taken
in (30) examples We might expect to see nyaluta in case of example (30a) Notice that the
14 )The irrelevant morpheme breaks are not shown here aluta can be further divided into a (past imperfective) -Iut (go)- a (FV)
15 )The form in the parenthesis means that it may be used by some But the other form is far more widely used and the informant prefers to use valuta
16 )Further investigation of phonological word derivation based on the syntactic information and morphological words should be made Here only the difference of phonological and morphological words is used without mentioning how different they are
-339shy
morphological word domain contains boundary and that the agreement markers begin with
coronal sounds If GF applies then GD in (24) will eliminate nyaluta
25 Unresolved vowel clashes
Not all the vowel sequences are resolved in Chitumbuka Some representative examples are
given in (33) and (34)
(33) V-final nouns and determiners a khuni Ili --gt khuni (y)ili khunyili khunili
tree this b buku ilo --gt buku (y)ilo bukwilo
book that c chipinda ico --gt chipinda (y)ico chipindico
room this
(34) Pronominal objects and V-initial verbs 171
a u-ku-mu-at-a --gt wakumu(y)ata wakumwata (you created him)
b wa-ku-ti-at-a --gt wakuti (y)ata wakutyata (He created you)
c ni-ku-tu-och-a --gt nikuku(w)ocha nikukocha nikukwocha (I roasted them)
d mu-ku-ka-eNg-a --gt mukuka(y)eNga muku (I melted it)
This seems to tell us that the main domain of hiatus resolving is within phonological words
GF and VD do not operate on this level and only G1 optionally shows up For further discussion
the phonological word boundaries are represented with parentheses in (35)
17 )The detailed breakdowns of the strings are as under
(yOU) ~u (him) tt (create)tni (me) ku + 11 (~e) + och (bake) + a (FV) wa (heher ka (It) Ng (melt)
u (them) verb stem
subject verb pronominal agreement tense objects
-340shy
Hiatus Resolving in Chitumbuka 17
(35) a (khuni) (ili) b (niku) (tu) (ocha) c (nikuwocha) (tuyuni)
One very interesting observation here is the change in the phonological word domains
(35b) and (35c) are semantically same The object in (35b) is pronominal while (35c) has a noun
object As a result of movement two phonological words are broken into three phonological
words This is not an arbitrary boundary just to explain the lack of hiatus resolving We may find
a very similar situation in English Intonational Phrase parsing
(36) a (As you know) (the tall boy is a genius) b (The tall boy) (as you know) (is a genius)
Nespor and Vogels (1986189) intonational phrase formation rule produces (36a) as well
as (36b) The syntax dependency of the phonological word formation is partly responsible for the
difference between (35b) and (35c)
Further hiatus is not the only rule sensitive to boundary difference in Chitumbuka We can
find that i-deletion and complex segment formation discussed in the preceding subsection shows
the same phenomenon Consider the following example (parentheses indicate the phonological
word domain)
(37) a (niku) (mu) (temwa) I love him Nkhu
b (uku) (ni)(temwa) You love me nthe
c (khunilClllno) the tree here mpha
One final observation is that our rules so far can not explain why glides can be inserted in
the examples of (33) and (34) We have already pointed out the problem of GI in 21 The
additional complication is that even if we posit glides as a phoneme we still have to revise GIor
-341shy
make another one to explain post-lexical GL This will be addressed in 32
3 Syllable phonology of hiatus resol ving
In the discussion in section 2 some of the problems in the linear analysis were detected
(What are they ) In this section I will go over them and reanalyze the data with the help
of syllable structure to see whether we can answer some of the questions left open McCarthy and
Prince (1986) style of moraic syllable structure will be employed for our discussion in this section
The basic syllable structure of Chitumbuka is given in (38)
(38) Syllable structure in Chitumbuka a a b a c a
~~((r r1T ( (I I 1 All the syllables are mora final 2 All the syllables are mono-moraic
This is not the template that is required in the underlying form as will be discussed
Therefore the non-moraic tier may not have to be satisfied In other words though the structure
shows two pre-moraic consonant slots these can be left empty The two branches simple mean the
maximum number of consonants that can come before a vowel
18)1n the post lexical level this structural condition may not hold In Chitumbuka vowels in the stressed syllable is lengthened The long vowel is represented with bimoraic syllable structure It is not uncommon that some of the rules or conditions are applied only on the lexical level but not on the post lexical level such as extrametricality stress assignment and even VD and GF we talked about in the preceding section
- 342-shy
Hiatus Resolving in Chitumbuka 19
31 Syllabification
Contrary to the claim that all the segments are already syllabified in the base form (cf
Selkirk (1982raquo I assume that syllable structure is assigned through the syllabification process
The syllabification principles in Chitumbuka are given under
(39) Syllabification in Chitumbuka i) Morafication A segment projects a mora if it is more sonorous than the following
segrnent ii) Syllable projection Each mora projects a syllable node iii) Segment incorporation Unsyllabified segments are incorporated into syllables
Now lets take a look at the following sample syllabification
(40) a
0 0 00
I (iii)I(ii) --(i) f1 f1 f1 f1 ~ rlrOI I
~
I I n i u mb a n 1 u mb a n 1 u mb a n i u mb a
b
0 o 0 0 o 0
I I I (iW (~Of1 f1 f1 f1 ~ ~ I I r~r I I I I I ch i m a i ch i m a ch i m a ch i m a i
Notice that we do not need GF as a rule the same result of GF application is achieved as a
19)1 suppose that morafication principle is universally dependent upon the sonority hierarchy There are three different strategies in using sonority in syllabifications
i) y is moraic in xy if it is more sonorous than x ii) y is moraic in yz if it is more sonorous than z iii) y is moraic if its sonority is greater than N
The language in the world employ one or two all all of these in the syllabification and language may slightly vary in setting up the value of N in (iii)
-- 343shy
natural consequence of syllabification In 21 we obtain the simplicity of phonemic inventory at
the cost of introducing GF thereby complicating the rule component The environment of GF is
reflected in the syllabification principle Of course syllabification can not be an additional burden
the language needs syllable structure anyhow for example for the application of many phonological
rules and even for utterance
The nature of glide sounds are clearly reflected in the syllable structure If a vowel is
dominated by a mora it will be interpreted as a full vowel But if it is dominated by the syllable
node directly it will surface as a glide
32 Syllabification domain
The syllabification principles introduced in 31 have presupposed that the string of segments
are not syllabified in the underlying representation Suppose every morpheme is syllabified in the
deep structure and consider the following example of word formation
(41)
a aa a
~ (I (noa k 0 V 1 a k 0
Now we will have to make a rule for mora deletion which will be almost the repetition of
the syllabification principle Syllabification process in the course of derivation can get rid of the
redundant repetition of syllable structural description as well as save the later-on application of mora
deletion
I assume that the syllabification takes place in the first derived domain In other words
segments are syllabified after the initial word formation process Before going into the detailed
-344shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
12 7j-iilfHl ~1lyenlt(1993 2)
markers in 21 or 22 they do not carry grammatical information Further we find that agreement
markers are added before these prefixes in the sentences
Lets begin the discussion by looking at the following data
(27) Prefixes with final lui and V-initial verbs a ku shy amba kwamba kuyamba (to start) b ku - ambuka kwambuka kuyambuka (to cross) c ku - ata kwata kuyata (to create) d ku shy enda kwenda kuyenda (to walk) e ku shy eNga kweNga kuyeNga (to melt) f ku - imba kwimba kuyimba (to sing) g ku - ima kwima kuyima (to stand) h ku - iba kwiba (to steal) I ku - iza kw iza (to come) J ku shy ocha kocha kuwocha (to burn) k l
ku - opa ku - ona
kopa ku wopaku tJY(1Y~- Hu
(to fear) in ~f~Ofmiddot
ffi ku -ma KU wura (to fear) n ku - umba kuwurnba (to mold something) o ku - undira kuwundira (to bury)
(28)Tense or aspect marker and V-initial verbs a ka - imba kimba kayimba (sang) b ka - iba kiba (stole) c ka - enda kenda kayenda (walked) d ka - opa kawopa (feared) e ka - ona kawona (saw) f ka - undira kawundira (buried) g ka - amba kamba kayamba (started)
The first thing readily noticeable in the given data is the existence of different alternatives in
b liku- Perfective the state of verb or the result of activity in the past is carried on to the present
c zamu- Future tense d ka- distant past (activity in the past) e kaNgu-habitual activity in the past f a- Activity in the immediate past g aNgu- Duration of activity that ended in the immediate past
lO)Though all the examples given here have Iku-I prefix the exactly same phonological changes are observed with all other prefixes that end in lui such as INgu-l liku-j and zamu-z All the V-initial verbs elicited so far are given here
-336shy
Hiatus Resolving in Chitumbuka 13
most of the cases Lets first begin with the apparently exceptional cases that do not have
variations Our Chitumbuka informant was not sure whether the stems of (27h) and (27i) begin
with Iii or not On a couple of occasions he said that to steal is kuba And even when he
pronounced kwiba he still maintained that the stem is Ibal instead of iba Though without any
sufficient supporting data I take this as the evidence that the root form of to steal has lost the
alternations and fossilized into kwiba and the same might be true for (27i)
Next it is very strange that only (271) shows different behavior in comparison with the two
preceding examples (27j) and (27k) It may be extremely difficult to assume that (271) has lost the
alternations and fossilized into kuwona form Intuitively if it is to be fossilized the form should
have been kona a phonologically simpler form I suggest that the base form of to see is uon shy
a GF is applied to this to produce Iwonal and then there is no vowel clash when Iku-I is added
failing to meet the VD environment
(27m) (27n) and (270) look quite exceptional in that VD does not apply to theses examples
though they are in VD environments I do not have systematic explanation for these There may he
some historical change in these words III or the underlying forms may begin with luul and the GF
changes the first lui into glide before tense or aspect markers are added 121 or they are just
exceptional At this point I will leave this as an open question
The GF and VD are quite straight forward There is no need to revise the rules Now we
will have to think about GI cases GI rule given in (15) does not explain the examples given in this
11)For example their friend in Chitumbuka is munyawo or rnunyalo Here we sec the alternations hetween Iwland iraquo] The same thing might have happened in (27m) (27n) and (270) words we may speculate that historically those stems begin with lraquo] And the sound weakened into the glide Iw
12)50 far no two same segments have been found to be adjacent morpheme internally However given the GF rule environment I think this idea may not be totally speculative
-- 337shy
subsection the sonority condition is not satisfied
One possible linear phonology approach may be to resort to the rule ordering We may
eliminate the sonority condition of GI and simply order GI after GF Further GI should not be
applied in (16) and (17) and therefore VD should be placed before GI hence the following order
(29) Rule ordering 1 Glide formation 2 Vowel deletion 3 Glide insertion
The major problem with this approach is the nature of the rule The data (27) (28) indicate
that all the three rules are optional But even though they are optional there is no option of not
applying any of the rules and therefore the last rule (GI) should be obligatory Further the data
(11) tells us a different story GF seems to be obligatory Otherwise we may have to say that
(1Ia) can be realized either as imwe (GF) or as imuye (GI) Moreover (17) shows that VD is
obligatory to prevent the operation of GI Here is the paradox A rule is optional for certain group
of words and obligatory for the rest This may suggest that rule based linear approach may not be
suitable for the explanation of hiatus resolving in Chitumbuka Further theoretical investigation will
be made in 32
24 Subject agreement marker and past imperfective marker la-I
All the prefixes introduced in 21 and some other personal agreement markers are used in
the sentence to satisfy the subject agreement Meanwhile there is one V-initial tense marker in
chitumbuka a- And therefore here again we expect to find vowel clashes The representative
data are given in (30)
13) If any of the rule is obligatory then we should not see the variations in the data given in (27) and (28)
-338shy
Hiatus Resolving in Chitumbuka 15
(30) Subject agreement marker and past imperfective marker a ni - aluta naluta (I walked) b u - aluta waluta (you walked) c wa - aluta waluta (he walked) d ti - aluta taluta (we went) e li - aluta laluta (It went) f vi - aluta (vyaluta) valuta (They went)
Superficially the data given here looks like those in (17) There is no GI example And the
choice between GF and VD seems to be determined by the relative sonority of the vowels
However (30f) shows that there is something different here (30) may not be the extension of
examples given in (17) The reason is twofold first the examples given here are not lexical The
agreement marker assignment should refer to the syntactic category the subject Secondly there is
an independent process of i-deletion in chitumbuka Consider following examples
(31) i-deletion and complex segment formation a ni (copula) -litali (big) --gt nilitali or nditali laquoit) is big) b ni (subject agreement)- kulya --gt nikulya or Nkhulya laquol) eat)
Here again i is deleted though it is not followed by another vowel I will tentatively
formulate the i-deletion rule as in (32)
(32) i-deletion i --gt 0_ [ (where [ is a morphological word domain)
This rule as (31a) (31 b) and (30f) shows is optional Then what if the option is not taken
in (30) examples We might expect to see nyaluta in case of example (30a) Notice that the
14 )The irrelevant morpheme breaks are not shown here aluta can be further divided into a (past imperfective) -Iut (go)- a (FV)
15 )The form in the parenthesis means that it may be used by some But the other form is far more widely used and the informant prefers to use valuta
16 )Further investigation of phonological word derivation based on the syntactic information and morphological words should be made Here only the difference of phonological and morphological words is used without mentioning how different they are
-339shy
morphological word domain contains boundary and that the agreement markers begin with
coronal sounds If GF applies then GD in (24) will eliminate nyaluta
25 Unresolved vowel clashes
Not all the vowel sequences are resolved in Chitumbuka Some representative examples are
given in (33) and (34)
(33) V-final nouns and determiners a khuni Ili --gt khuni (y)ili khunyili khunili
tree this b buku ilo --gt buku (y)ilo bukwilo
book that c chipinda ico --gt chipinda (y)ico chipindico
room this
(34) Pronominal objects and V-initial verbs 171
a u-ku-mu-at-a --gt wakumu(y)ata wakumwata (you created him)
b wa-ku-ti-at-a --gt wakuti (y)ata wakutyata (He created you)
c ni-ku-tu-och-a --gt nikuku(w)ocha nikukocha nikukwocha (I roasted them)
d mu-ku-ka-eNg-a --gt mukuka(y)eNga muku (I melted it)
This seems to tell us that the main domain of hiatus resolving is within phonological words
GF and VD do not operate on this level and only G1 optionally shows up For further discussion
the phonological word boundaries are represented with parentheses in (35)
17 )The detailed breakdowns of the strings are as under
(yOU) ~u (him) tt (create)tni (me) ku + 11 (~e) + och (bake) + a (FV) wa (heher ka (It) Ng (melt)
u (them) verb stem
subject verb pronominal agreement tense objects
-340shy
Hiatus Resolving in Chitumbuka 17
(35) a (khuni) (ili) b (niku) (tu) (ocha) c (nikuwocha) (tuyuni)
One very interesting observation here is the change in the phonological word domains
(35b) and (35c) are semantically same The object in (35b) is pronominal while (35c) has a noun
object As a result of movement two phonological words are broken into three phonological
words This is not an arbitrary boundary just to explain the lack of hiatus resolving We may find
a very similar situation in English Intonational Phrase parsing
(36) a (As you know) (the tall boy is a genius) b (The tall boy) (as you know) (is a genius)
Nespor and Vogels (1986189) intonational phrase formation rule produces (36a) as well
as (36b) The syntax dependency of the phonological word formation is partly responsible for the
difference between (35b) and (35c)
Further hiatus is not the only rule sensitive to boundary difference in Chitumbuka We can
find that i-deletion and complex segment formation discussed in the preceding subsection shows
the same phenomenon Consider the following example (parentheses indicate the phonological
word domain)
(37) a (niku) (mu) (temwa) I love him Nkhu
b (uku) (ni)(temwa) You love me nthe
c (khunilClllno) the tree here mpha
One final observation is that our rules so far can not explain why glides can be inserted in
the examples of (33) and (34) We have already pointed out the problem of GI in 21 The
additional complication is that even if we posit glides as a phoneme we still have to revise GIor
-341shy
make another one to explain post-lexical GL This will be addressed in 32
3 Syllable phonology of hiatus resol ving
In the discussion in section 2 some of the problems in the linear analysis were detected
(What are they ) In this section I will go over them and reanalyze the data with the help
of syllable structure to see whether we can answer some of the questions left open McCarthy and
Prince (1986) style of moraic syllable structure will be employed for our discussion in this section
The basic syllable structure of Chitumbuka is given in (38)
(38) Syllable structure in Chitumbuka a a b a c a
~~((r r1T ( (I I 1 All the syllables are mora final 2 All the syllables are mono-moraic
This is not the template that is required in the underlying form as will be discussed
Therefore the non-moraic tier may not have to be satisfied In other words though the structure
shows two pre-moraic consonant slots these can be left empty The two branches simple mean the
maximum number of consonants that can come before a vowel
18)1n the post lexical level this structural condition may not hold In Chitumbuka vowels in the stressed syllable is lengthened The long vowel is represented with bimoraic syllable structure It is not uncommon that some of the rules or conditions are applied only on the lexical level but not on the post lexical level such as extrametricality stress assignment and even VD and GF we talked about in the preceding section
- 342-shy
Hiatus Resolving in Chitumbuka 19
31 Syllabification
Contrary to the claim that all the segments are already syllabified in the base form (cf
Selkirk (1982raquo I assume that syllable structure is assigned through the syllabification process
The syllabification principles in Chitumbuka are given under
(39) Syllabification in Chitumbuka i) Morafication A segment projects a mora if it is more sonorous than the following
segrnent ii) Syllable projection Each mora projects a syllable node iii) Segment incorporation Unsyllabified segments are incorporated into syllables
Now lets take a look at the following sample syllabification
(40) a
0 0 00
I (iii)I(ii) --(i) f1 f1 f1 f1 ~ rlrOI I
~
I I n i u mb a n 1 u mb a n 1 u mb a n i u mb a
b
0 o 0 0 o 0
I I I (iW (~Of1 f1 f1 f1 ~ ~ I I r~r I I I I I ch i m a i ch i m a ch i m a ch i m a i
Notice that we do not need GF as a rule the same result of GF application is achieved as a
19)1 suppose that morafication principle is universally dependent upon the sonority hierarchy There are three different strategies in using sonority in syllabifications
i) y is moraic in xy if it is more sonorous than x ii) y is moraic in yz if it is more sonorous than z iii) y is moraic if its sonority is greater than N
The language in the world employ one or two all all of these in the syllabification and language may slightly vary in setting up the value of N in (iii)
-- 343shy
natural consequence of syllabification In 21 we obtain the simplicity of phonemic inventory at
the cost of introducing GF thereby complicating the rule component The environment of GF is
reflected in the syllabification principle Of course syllabification can not be an additional burden
the language needs syllable structure anyhow for example for the application of many phonological
rules and even for utterance
The nature of glide sounds are clearly reflected in the syllable structure If a vowel is
dominated by a mora it will be interpreted as a full vowel But if it is dominated by the syllable
node directly it will surface as a glide
32 Syllabification domain
The syllabification principles introduced in 31 have presupposed that the string of segments
are not syllabified in the underlying representation Suppose every morpheme is syllabified in the
deep structure and consider the following example of word formation
(41)
a aa a
~ (I (noa k 0 V 1 a k 0
Now we will have to make a rule for mora deletion which will be almost the repetition of
the syllabification principle Syllabification process in the course of derivation can get rid of the
redundant repetition of syllable structural description as well as save the later-on application of mora
deletion
I assume that the syllabification takes place in the first derived domain In other words
segments are syllabified after the initial word formation process Before going into the detailed
-344shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
Hiatus Resolving in Chitumbuka 13
most of the cases Lets first begin with the apparently exceptional cases that do not have
variations Our Chitumbuka informant was not sure whether the stems of (27h) and (27i) begin
with Iii or not On a couple of occasions he said that to steal is kuba And even when he
pronounced kwiba he still maintained that the stem is Ibal instead of iba Though without any
sufficient supporting data I take this as the evidence that the root form of to steal has lost the
alternations and fossilized into kwiba and the same might be true for (27i)
Next it is very strange that only (271) shows different behavior in comparison with the two
preceding examples (27j) and (27k) It may be extremely difficult to assume that (271) has lost the
alternations and fossilized into kuwona form Intuitively if it is to be fossilized the form should
have been kona a phonologically simpler form I suggest that the base form of to see is uon shy
a GF is applied to this to produce Iwonal and then there is no vowel clash when Iku-I is added
failing to meet the VD environment
(27m) (27n) and (270) look quite exceptional in that VD does not apply to theses examples
though they are in VD environments I do not have systematic explanation for these There may he
some historical change in these words III or the underlying forms may begin with luul and the GF
changes the first lui into glide before tense or aspect markers are added 121 or they are just
exceptional At this point I will leave this as an open question
The GF and VD are quite straight forward There is no need to revise the rules Now we
will have to think about GI cases GI rule given in (15) does not explain the examples given in this
11)For example their friend in Chitumbuka is munyawo or rnunyalo Here we sec the alternations hetween Iwland iraquo] The same thing might have happened in (27m) (27n) and (270) words we may speculate that historically those stems begin with lraquo] And the sound weakened into the glide Iw
12)50 far no two same segments have been found to be adjacent morpheme internally However given the GF rule environment I think this idea may not be totally speculative
-- 337shy
subsection the sonority condition is not satisfied
One possible linear phonology approach may be to resort to the rule ordering We may
eliminate the sonority condition of GI and simply order GI after GF Further GI should not be
applied in (16) and (17) and therefore VD should be placed before GI hence the following order
(29) Rule ordering 1 Glide formation 2 Vowel deletion 3 Glide insertion
The major problem with this approach is the nature of the rule The data (27) (28) indicate
that all the three rules are optional But even though they are optional there is no option of not
applying any of the rules and therefore the last rule (GI) should be obligatory Further the data
(11) tells us a different story GF seems to be obligatory Otherwise we may have to say that
(1Ia) can be realized either as imwe (GF) or as imuye (GI) Moreover (17) shows that VD is
obligatory to prevent the operation of GI Here is the paradox A rule is optional for certain group
of words and obligatory for the rest This may suggest that rule based linear approach may not be
suitable for the explanation of hiatus resolving in Chitumbuka Further theoretical investigation will
be made in 32
24 Subject agreement marker and past imperfective marker la-I
All the prefixes introduced in 21 and some other personal agreement markers are used in
the sentence to satisfy the subject agreement Meanwhile there is one V-initial tense marker in
chitumbuka a- And therefore here again we expect to find vowel clashes The representative
data are given in (30)
13) If any of the rule is obligatory then we should not see the variations in the data given in (27) and (28)
-338shy
Hiatus Resolving in Chitumbuka 15
(30) Subject agreement marker and past imperfective marker a ni - aluta naluta (I walked) b u - aluta waluta (you walked) c wa - aluta waluta (he walked) d ti - aluta taluta (we went) e li - aluta laluta (It went) f vi - aluta (vyaluta) valuta (They went)
Superficially the data given here looks like those in (17) There is no GI example And the
choice between GF and VD seems to be determined by the relative sonority of the vowels
However (30f) shows that there is something different here (30) may not be the extension of
examples given in (17) The reason is twofold first the examples given here are not lexical The
agreement marker assignment should refer to the syntactic category the subject Secondly there is
an independent process of i-deletion in chitumbuka Consider following examples
(31) i-deletion and complex segment formation a ni (copula) -litali (big) --gt nilitali or nditali laquoit) is big) b ni (subject agreement)- kulya --gt nikulya or Nkhulya laquol) eat)
Here again i is deleted though it is not followed by another vowel I will tentatively
formulate the i-deletion rule as in (32)
(32) i-deletion i --gt 0_ [ (where [ is a morphological word domain)
This rule as (31a) (31 b) and (30f) shows is optional Then what if the option is not taken
in (30) examples We might expect to see nyaluta in case of example (30a) Notice that the
14 )The irrelevant morpheme breaks are not shown here aluta can be further divided into a (past imperfective) -Iut (go)- a (FV)
15 )The form in the parenthesis means that it may be used by some But the other form is far more widely used and the informant prefers to use valuta
16 )Further investigation of phonological word derivation based on the syntactic information and morphological words should be made Here only the difference of phonological and morphological words is used without mentioning how different they are
-339shy
morphological word domain contains boundary and that the agreement markers begin with
coronal sounds If GF applies then GD in (24) will eliminate nyaluta
25 Unresolved vowel clashes
Not all the vowel sequences are resolved in Chitumbuka Some representative examples are
given in (33) and (34)
(33) V-final nouns and determiners a khuni Ili --gt khuni (y)ili khunyili khunili
tree this b buku ilo --gt buku (y)ilo bukwilo
book that c chipinda ico --gt chipinda (y)ico chipindico
room this
(34) Pronominal objects and V-initial verbs 171
a u-ku-mu-at-a --gt wakumu(y)ata wakumwata (you created him)
b wa-ku-ti-at-a --gt wakuti (y)ata wakutyata (He created you)
c ni-ku-tu-och-a --gt nikuku(w)ocha nikukocha nikukwocha (I roasted them)
d mu-ku-ka-eNg-a --gt mukuka(y)eNga muku (I melted it)
This seems to tell us that the main domain of hiatus resolving is within phonological words
GF and VD do not operate on this level and only G1 optionally shows up For further discussion
the phonological word boundaries are represented with parentheses in (35)
17 )The detailed breakdowns of the strings are as under
(yOU) ~u (him) tt (create)tni (me) ku + 11 (~e) + och (bake) + a (FV) wa (heher ka (It) Ng (melt)
u (them) verb stem
subject verb pronominal agreement tense objects
-340shy
Hiatus Resolving in Chitumbuka 17
(35) a (khuni) (ili) b (niku) (tu) (ocha) c (nikuwocha) (tuyuni)
One very interesting observation here is the change in the phonological word domains
(35b) and (35c) are semantically same The object in (35b) is pronominal while (35c) has a noun
object As a result of movement two phonological words are broken into three phonological
words This is not an arbitrary boundary just to explain the lack of hiatus resolving We may find
a very similar situation in English Intonational Phrase parsing
(36) a (As you know) (the tall boy is a genius) b (The tall boy) (as you know) (is a genius)
Nespor and Vogels (1986189) intonational phrase formation rule produces (36a) as well
as (36b) The syntax dependency of the phonological word formation is partly responsible for the
difference between (35b) and (35c)
Further hiatus is not the only rule sensitive to boundary difference in Chitumbuka We can
find that i-deletion and complex segment formation discussed in the preceding subsection shows
the same phenomenon Consider the following example (parentheses indicate the phonological
word domain)
(37) a (niku) (mu) (temwa) I love him Nkhu
b (uku) (ni)(temwa) You love me nthe
c (khunilClllno) the tree here mpha
One final observation is that our rules so far can not explain why glides can be inserted in
the examples of (33) and (34) We have already pointed out the problem of GI in 21 The
additional complication is that even if we posit glides as a phoneme we still have to revise GIor
-341shy
make another one to explain post-lexical GL This will be addressed in 32
3 Syllable phonology of hiatus resol ving
In the discussion in section 2 some of the problems in the linear analysis were detected
(What are they ) In this section I will go over them and reanalyze the data with the help
of syllable structure to see whether we can answer some of the questions left open McCarthy and
Prince (1986) style of moraic syllable structure will be employed for our discussion in this section
The basic syllable structure of Chitumbuka is given in (38)
(38) Syllable structure in Chitumbuka a a b a c a
~~((r r1T ( (I I 1 All the syllables are mora final 2 All the syllables are mono-moraic
This is not the template that is required in the underlying form as will be discussed
Therefore the non-moraic tier may not have to be satisfied In other words though the structure
shows two pre-moraic consonant slots these can be left empty The two branches simple mean the
maximum number of consonants that can come before a vowel
18)1n the post lexical level this structural condition may not hold In Chitumbuka vowels in the stressed syllable is lengthened The long vowel is represented with bimoraic syllable structure It is not uncommon that some of the rules or conditions are applied only on the lexical level but not on the post lexical level such as extrametricality stress assignment and even VD and GF we talked about in the preceding section
- 342-shy
Hiatus Resolving in Chitumbuka 19
31 Syllabification
Contrary to the claim that all the segments are already syllabified in the base form (cf
Selkirk (1982raquo I assume that syllable structure is assigned through the syllabification process
The syllabification principles in Chitumbuka are given under
(39) Syllabification in Chitumbuka i) Morafication A segment projects a mora if it is more sonorous than the following
segrnent ii) Syllable projection Each mora projects a syllable node iii) Segment incorporation Unsyllabified segments are incorporated into syllables
Now lets take a look at the following sample syllabification
(40) a
0 0 00
I (iii)I(ii) --(i) f1 f1 f1 f1 ~ rlrOI I
~
I I n i u mb a n 1 u mb a n 1 u mb a n i u mb a
b
0 o 0 0 o 0
I I I (iW (~Of1 f1 f1 f1 ~ ~ I I r~r I I I I I ch i m a i ch i m a ch i m a ch i m a i
Notice that we do not need GF as a rule the same result of GF application is achieved as a
19)1 suppose that morafication principle is universally dependent upon the sonority hierarchy There are three different strategies in using sonority in syllabifications
i) y is moraic in xy if it is more sonorous than x ii) y is moraic in yz if it is more sonorous than z iii) y is moraic if its sonority is greater than N
The language in the world employ one or two all all of these in the syllabification and language may slightly vary in setting up the value of N in (iii)
-- 343shy
natural consequence of syllabification In 21 we obtain the simplicity of phonemic inventory at
the cost of introducing GF thereby complicating the rule component The environment of GF is
reflected in the syllabification principle Of course syllabification can not be an additional burden
the language needs syllable structure anyhow for example for the application of many phonological
rules and even for utterance
The nature of glide sounds are clearly reflected in the syllable structure If a vowel is
dominated by a mora it will be interpreted as a full vowel But if it is dominated by the syllable
node directly it will surface as a glide
32 Syllabification domain
The syllabification principles introduced in 31 have presupposed that the string of segments
are not syllabified in the underlying representation Suppose every morpheme is syllabified in the
deep structure and consider the following example of word formation
(41)
a aa a
~ (I (noa k 0 V 1 a k 0
Now we will have to make a rule for mora deletion which will be almost the repetition of
the syllabification principle Syllabification process in the course of derivation can get rid of the
redundant repetition of syllable structural description as well as save the later-on application of mora
deletion
I assume that the syllabification takes place in the first derived domain In other words
segments are syllabified after the initial word formation process Before going into the detailed
-344shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
subsection the sonority condition is not satisfied
One possible linear phonology approach may be to resort to the rule ordering We may
eliminate the sonority condition of GI and simply order GI after GF Further GI should not be
applied in (16) and (17) and therefore VD should be placed before GI hence the following order
(29) Rule ordering 1 Glide formation 2 Vowel deletion 3 Glide insertion
The major problem with this approach is the nature of the rule The data (27) (28) indicate
that all the three rules are optional But even though they are optional there is no option of not
applying any of the rules and therefore the last rule (GI) should be obligatory Further the data
(11) tells us a different story GF seems to be obligatory Otherwise we may have to say that
(1Ia) can be realized either as imwe (GF) or as imuye (GI) Moreover (17) shows that VD is
obligatory to prevent the operation of GI Here is the paradox A rule is optional for certain group
of words and obligatory for the rest This may suggest that rule based linear approach may not be
suitable for the explanation of hiatus resolving in Chitumbuka Further theoretical investigation will
be made in 32
24 Subject agreement marker and past imperfective marker la-I
All the prefixes introduced in 21 and some other personal agreement markers are used in
the sentence to satisfy the subject agreement Meanwhile there is one V-initial tense marker in
chitumbuka a- And therefore here again we expect to find vowel clashes The representative
data are given in (30)
13) If any of the rule is obligatory then we should not see the variations in the data given in (27) and (28)
-338shy
Hiatus Resolving in Chitumbuka 15
(30) Subject agreement marker and past imperfective marker a ni - aluta naluta (I walked) b u - aluta waluta (you walked) c wa - aluta waluta (he walked) d ti - aluta taluta (we went) e li - aluta laluta (It went) f vi - aluta (vyaluta) valuta (They went)
Superficially the data given here looks like those in (17) There is no GI example And the
choice between GF and VD seems to be determined by the relative sonority of the vowels
However (30f) shows that there is something different here (30) may not be the extension of
examples given in (17) The reason is twofold first the examples given here are not lexical The
agreement marker assignment should refer to the syntactic category the subject Secondly there is
an independent process of i-deletion in chitumbuka Consider following examples
(31) i-deletion and complex segment formation a ni (copula) -litali (big) --gt nilitali or nditali laquoit) is big) b ni (subject agreement)- kulya --gt nikulya or Nkhulya laquol) eat)
Here again i is deleted though it is not followed by another vowel I will tentatively
formulate the i-deletion rule as in (32)
(32) i-deletion i --gt 0_ [ (where [ is a morphological word domain)
This rule as (31a) (31 b) and (30f) shows is optional Then what if the option is not taken
in (30) examples We might expect to see nyaluta in case of example (30a) Notice that the
14 )The irrelevant morpheme breaks are not shown here aluta can be further divided into a (past imperfective) -Iut (go)- a (FV)
15 )The form in the parenthesis means that it may be used by some But the other form is far more widely used and the informant prefers to use valuta
16 )Further investigation of phonological word derivation based on the syntactic information and morphological words should be made Here only the difference of phonological and morphological words is used without mentioning how different they are
-339shy
morphological word domain contains boundary and that the agreement markers begin with
coronal sounds If GF applies then GD in (24) will eliminate nyaluta
25 Unresolved vowel clashes
Not all the vowel sequences are resolved in Chitumbuka Some representative examples are
given in (33) and (34)
(33) V-final nouns and determiners a khuni Ili --gt khuni (y)ili khunyili khunili
tree this b buku ilo --gt buku (y)ilo bukwilo
book that c chipinda ico --gt chipinda (y)ico chipindico
room this
(34) Pronominal objects and V-initial verbs 171
a u-ku-mu-at-a --gt wakumu(y)ata wakumwata (you created him)
b wa-ku-ti-at-a --gt wakuti (y)ata wakutyata (He created you)
c ni-ku-tu-och-a --gt nikuku(w)ocha nikukocha nikukwocha (I roasted them)
d mu-ku-ka-eNg-a --gt mukuka(y)eNga muku (I melted it)
This seems to tell us that the main domain of hiatus resolving is within phonological words
GF and VD do not operate on this level and only G1 optionally shows up For further discussion
the phonological word boundaries are represented with parentheses in (35)
17 )The detailed breakdowns of the strings are as under
(yOU) ~u (him) tt (create)tni (me) ku + 11 (~e) + och (bake) + a (FV) wa (heher ka (It) Ng (melt)
u (them) verb stem
subject verb pronominal agreement tense objects
-340shy
Hiatus Resolving in Chitumbuka 17
(35) a (khuni) (ili) b (niku) (tu) (ocha) c (nikuwocha) (tuyuni)
One very interesting observation here is the change in the phonological word domains
(35b) and (35c) are semantically same The object in (35b) is pronominal while (35c) has a noun
object As a result of movement two phonological words are broken into three phonological
words This is not an arbitrary boundary just to explain the lack of hiatus resolving We may find
a very similar situation in English Intonational Phrase parsing
(36) a (As you know) (the tall boy is a genius) b (The tall boy) (as you know) (is a genius)
Nespor and Vogels (1986189) intonational phrase formation rule produces (36a) as well
as (36b) The syntax dependency of the phonological word formation is partly responsible for the
difference between (35b) and (35c)
Further hiatus is not the only rule sensitive to boundary difference in Chitumbuka We can
find that i-deletion and complex segment formation discussed in the preceding subsection shows
the same phenomenon Consider the following example (parentheses indicate the phonological
word domain)
(37) a (niku) (mu) (temwa) I love him Nkhu
b (uku) (ni)(temwa) You love me nthe
c (khunilClllno) the tree here mpha
One final observation is that our rules so far can not explain why glides can be inserted in
the examples of (33) and (34) We have already pointed out the problem of GI in 21 The
additional complication is that even if we posit glides as a phoneme we still have to revise GIor
-341shy
make another one to explain post-lexical GL This will be addressed in 32
3 Syllable phonology of hiatus resol ving
In the discussion in section 2 some of the problems in the linear analysis were detected
(What are they ) In this section I will go over them and reanalyze the data with the help
of syllable structure to see whether we can answer some of the questions left open McCarthy and
Prince (1986) style of moraic syllable structure will be employed for our discussion in this section
The basic syllable structure of Chitumbuka is given in (38)
(38) Syllable structure in Chitumbuka a a b a c a
~~((r r1T ( (I I 1 All the syllables are mora final 2 All the syllables are mono-moraic
This is not the template that is required in the underlying form as will be discussed
Therefore the non-moraic tier may not have to be satisfied In other words though the structure
shows two pre-moraic consonant slots these can be left empty The two branches simple mean the
maximum number of consonants that can come before a vowel
18)1n the post lexical level this structural condition may not hold In Chitumbuka vowels in the stressed syllable is lengthened The long vowel is represented with bimoraic syllable structure It is not uncommon that some of the rules or conditions are applied only on the lexical level but not on the post lexical level such as extrametricality stress assignment and even VD and GF we talked about in the preceding section
- 342-shy
Hiatus Resolving in Chitumbuka 19
31 Syllabification
Contrary to the claim that all the segments are already syllabified in the base form (cf
Selkirk (1982raquo I assume that syllable structure is assigned through the syllabification process
The syllabification principles in Chitumbuka are given under
(39) Syllabification in Chitumbuka i) Morafication A segment projects a mora if it is more sonorous than the following
segrnent ii) Syllable projection Each mora projects a syllable node iii) Segment incorporation Unsyllabified segments are incorporated into syllables
Now lets take a look at the following sample syllabification
(40) a
0 0 00
I (iii)I(ii) --(i) f1 f1 f1 f1 ~ rlrOI I
~
I I n i u mb a n 1 u mb a n 1 u mb a n i u mb a
b
0 o 0 0 o 0
I I I (iW (~Of1 f1 f1 f1 ~ ~ I I r~r I I I I I ch i m a i ch i m a ch i m a ch i m a i
Notice that we do not need GF as a rule the same result of GF application is achieved as a
19)1 suppose that morafication principle is universally dependent upon the sonority hierarchy There are three different strategies in using sonority in syllabifications
i) y is moraic in xy if it is more sonorous than x ii) y is moraic in yz if it is more sonorous than z iii) y is moraic if its sonority is greater than N
The language in the world employ one or two all all of these in the syllabification and language may slightly vary in setting up the value of N in (iii)
-- 343shy
natural consequence of syllabification In 21 we obtain the simplicity of phonemic inventory at
the cost of introducing GF thereby complicating the rule component The environment of GF is
reflected in the syllabification principle Of course syllabification can not be an additional burden
the language needs syllable structure anyhow for example for the application of many phonological
rules and even for utterance
The nature of glide sounds are clearly reflected in the syllable structure If a vowel is
dominated by a mora it will be interpreted as a full vowel But if it is dominated by the syllable
node directly it will surface as a glide
32 Syllabification domain
The syllabification principles introduced in 31 have presupposed that the string of segments
are not syllabified in the underlying representation Suppose every morpheme is syllabified in the
deep structure and consider the following example of word formation
(41)
a aa a
~ (I (noa k 0 V 1 a k 0
Now we will have to make a rule for mora deletion which will be almost the repetition of
the syllabification principle Syllabification process in the course of derivation can get rid of the
redundant repetition of syllable structural description as well as save the later-on application of mora
deletion
I assume that the syllabification takes place in the first derived domain In other words
segments are syllabified after the initial word formation process Before going into the detailed
-344shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
Hiatus Resolving in Chitumbuka 15
(30) Subject agreement marker and past imperfective marker a ni - aluta naluta (I walked) b u - aluta waluta (you walked) c wa - aluta waluta (he walked) d ti - aluta taluta (we went) e li - aluta laluta (It went) f vi - aluta (vyaluta) valuta (They went)
Superficially the data given here looks like those in (17) There is no GI example And the
choice between GF and VD seems to be determined by the relative sonority of the vowels
However (30f) shows that there is something different here (30) may not be the extension of
examples given in (17) The reason is twofold first the examples given here are not lexical The
agreement marker assignment should refer to the syntactic category the subject Secondly there is
an independent process of i-deletion in chitumbuka Consider following examples
(31) i-deletion and complex segment formation a ni (copula) -litali (big) --gt nilitali or nditali laquoit) is big) b ni (subject agreement)- kulya --gt nikulya or Nkhulya laquol) eat)
Here again i is deleted though it is not followed by another vowel I will tentatively
formulate the i-deletion rule as in (32)
(32) i-deletion i --gt 0_ [ (where [ is a morphological word domain)
This rule as (31a) (31 b) and (30f) shows is optional Then what if the option is not taken
in (30) examples We might expect to see nyaluta in case of example (30a) Notice that the
14 )The irrelevant morpheme breaks are not shown here aluta can be further divided into a (past imperfective) -Iut (go)- a (FV)
15 )The form in the parenthesis means that it may be used by some But the other form is far more widely used and the informant prefers to use valuta
16 )Further investigation of phonological word derivation based on the syntactic information and morphological words should be made Here only the difference of phonological and morphological words is used without mentioning how different they are
-339shy
morphological word domain contains boundary and that the agreement markers begin with
coronal sounds If GF applies then GD in (24) will eliminate nyaluta
25 Unresolved vowel clashes
Not all the vowel sequences are resolved in Chitumbuka Some representative examples are
given in (33) and (34)
(33) V-final nouns and determiners a khuni Ili --gt khuni (y)ili khunyili khunili
tree this b buku ilo --gt buku (y)ilo bukwilo
book that c chipinda ico --gt chipinda (y)ico chipindico
room this
(34) Pronominal objects and V-initial verbs 171
a u-ku-mu-at-a --gt wakumu(y)ata wakumwata (you created him)
b wa-ku-ti-at-a --gt wakuti (y)ata wakutyata (He created you)
c ni-ku-tu-och-a --gt nikuku(w)ocha nikukocha nikukwocha (I roasted them)
d mu-ku-ka-eNg-a --gt mukuka(y)eNga muku (I melted it)
This seems to tell us that the main domain of hiatus resolving is within phonological words
GF and VD do not operate on this level and only G1 optionally shows up For further discussion
the phonological word boundaries are represented with parentheses in (35)
17 )The detailed breakdowns of the strings are as under
(yOU) ~u (him) tt (create)tni (me) ku + 11 (~e) + och (bake) + a (FV) wa (heher ka (It) Ng (melt)
u (them) verb stem
subject verb pronominal agreement tense objects
-340shy
Hiatus Resolving in Chitumbuka 17
(35) a (khuni) (ili) b (niku) (tu) (ocha) c (nikuwocha) (tuyuni)
One very interesting observation here is the change in the phonological word domains
(35b) and (35c) are semantically same The object in (35b) is pronominal while (35c) has a noun
object As a result of movement two phonological words are broken into three phonological
words This is not an arbitrary boundary just to explain the lack of hiatus resolving We may find
a very similar situation in English Intonational Phrase parsing
(36) a (As you know) (the tall boy is a genius) b (The tall boy) (as you know) (is a genius)
Nespor and Vogels (1986189) intonational phrase formation rule produces (36a) as well
as (36b) The syntax dependency of the phonological word formation is partly responsible for the
difference between (35b) and (35c)
Further hiatus is not the only rule sensitive to boundary difference in Chitumbuka We can
find that i-deletion and complex segment formation discussed in the preceding subsection shows
the same phenomenon Consider the following example (parentheses indicate the phonological
word domain)
(37) a (niku) (mu) (temwa) I love him Nkhu
b (uku) (ni)(temwa) You love me nthe
c (khunilClllno) the tree here mpha
One final observation is that our rules so far can not explain why glides can be inserted in
the examples of (33) and (34) We have already pointed out the problem of GI in 21 The
additional complication is that even if we posit glides as a phoneme we still have to revise GIor
-341shy
make another one to explain post-lexical GL This will be addressed in 32
3 Syllable phonology of hiatus resol ving
In the discussion in section 2 some of the problems in the linear analysis were detected
(What are they ) In this section I will go over them and reanalyze the data with the help
of syllable structure to see whether we can answer some of the questions left open McCarthy and
Prince (1986) style of moraic syllable structure will be employed for our discussion in this section
The basic syllable structure of Chitumbuka is given in (38)
(38) Syllable structure in Chitumbuka a a b a c a
~~((r r1T ( (I I 1 All the syllables are mora final 2 All the syllables are mono-moraic
This is not the template that is required in the underlying form as will be discussed
Therefore the non-moraic tier may not have to be satisfied In other words though the structure
shows two pre-moraic consonant slots these can be left empty The two branches simple mean the
maximum number of consonants that can come before a vowel
18)1n the post lexical level this structural condition may not hold In Chitumbuka vowels in the stressed syllable is lengthened The long vowel is represented with bimoraic syllable structure It is not uncommon that some of the rules or conditions are applied only on the lexical level but not on the post lexical level such as extrametricality stress assignment and even VD and GF we talked about in the preceding section
- 342-shy
Hiatus Resolving in Chitumbuka 19
31 Syllabification
Contrary to the claim that all the segments are already syllabified in the base form (cf
Selkirk (1982raquo I assume that syllable structure is assigned through the syllabification process
The syllabification principles in Chitumbuka are given under
(39) Syllabification in Chitumbuka i) Morafication A segment projects a mora if it is more sonorous than the following
segrnent ii) Syllable projection Each mora projects a syllable node iii) Segment incorporation Unsyllabified segments are incorporated into syllables
Now lets take a look at the following sample syllabification
(40) a
0 0 00
I (iii)I(ii) --(i) f1 f1 f1 f1 ~ rlrOI I
~
I I n i u mb a n 1 u mb a n 1 u mb a n i u mb a
b
0 o 0 0 o 0
I I I (iW (~Of1 f1 f1 f1 ~ ~ I I r~r I I I I I ch i m a i ch i m a ch i m a ch i m a i
Notice that we do not need GF as a rule the same result of GF application is achieved as a
19)1 suppose that morafication principle is universally dependent upon the sonority hierarchy There are three different strategies in using sonority in syllabifications
i) y is moraic in xy if it is more sonorous than x ii) y is moraic in yz if it is more sonorous than z iii) y is moraic if its sonority is greater than N
The language in the world employ one or two all all of these in the syllabification and language may slightly vary in setting up the value of N in (iii)
-- 343shy
natural consequence of syllabification In 21 we obtain the simplicity of phonemic inventory at
the cost of introducing GF thereby complicating the rule component The environment of GF is
reflected in the syllabification principle Of course syllabification can not be an additional burden
the language needs syllable structure anyhow for example for the application of many phonological
rules and even for utterance
The nature of glide sounds are clearly reflected in the syllable structure If a vowel is
dominated by a mora it will be interpreted as a full vowel But if it is dominated by the syllable
node directly it will surface as a glide
32 Syllabification domain
The syllabification principles introduced in 31 have presupposed that the string of segments
are not syllabified in the underlying representation Suppose every morpheme is syllabified in the
deep structure and consider the following example of word formation
(41)
a aa a
~ (I (noa k 0 V 1 a k 0
Now we will have to make a rule for mora deletion which will be almost the repetition of
the syllabification principle Syllabification process in the course of derivation can get rid of the
redundant repetition of syllable structural description as well as save the later-on application of mora
deletion
I assume that the syllabification takes place in the first derived domain In other words
segments are syllabified after the initial word formation process Before going into the detailed
-344shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
morphological word domain contains boundary and that the agreement markers begin with
coronal sounds If GF applies then GD in (24) will eliminate nyaluta
25 Unresolved vowel clashes
Not all the vowel sequences are resolved in Chitumbuka Some representative examples are
given in (33) and (34)
(33) V-final nouns and determiners a khuni Ili --gt khuni (y)ili khunyili khunili
tree this b buku ilo --gt buku (y)ilo bukwilo
book that c chipinda ico --gt chipinda (y)ico chipindico
room this
(34) Pronominal objects and V-initial verbs 171
a u-ku-mu-at-a --gt wakumu(y)ata wakumwata (you created him)
b wa-ku-ti-at-a --gt wakuti (y)ata wakutyata (He created you)
c ni-ku-tu-och-a --gt nikuku(w)ocha nikukocha nikukwocha (I roasted them)
d mu-ku-ka-eNg-a --gt mukuka(y)eNga muku (I melted it)
This seems to tell us that the main domain of hiatus resolving is within phonological words
GF and VD do not operate on this level and only G1 optionally shows up For further discussion
the phonological word boundaries are represented with parentheses in (35)
17 )The detailed breakdowns of the strings are as under
(yOU) ~u (him) tt (create)tni (me) ku + 11 (~e) + och (bake) + a (FV) wa (heher ka (It) Ng (melt)
u (them) verb stem
subject verb pronominal agreement tense objects
-340shy
Hiatus Resolving in Chitumbuka 17
(35) a (khuni) (ili) b (niku) (tu) (ocha) c (nikuwocha) (tuyuni)
One very interesting observation here is the change in the phonological word domains
(35b) and (35c) are semantically same The object in (35b) is pronominal while (35c) has a noun
object As a result of movement two phonological words are broken into three phonological
words This is not an arbitrary boundary just to explain the lack of hiatus resolving We may find
a very similar situation in English Intonational Phrase parsing
(36) a (As you know) (the tall boy is a genius) b (The tall boy) (as you know) (is a genius)
Nespor and Vogels (1986189) intonational phrase formation rule produces (36a) as well
as (36b) The syntax dependency of the phonological word formation is partly responsible for the
difference between (35b) and (35c)
Further hiatus is not the only rule sensitive to boundary difference in Chitumbuka We can
find that i-deletion and complex segment formation discussed in the preceding subsection shows
the same phenomenon Consider the following example (parentheses indicate the phonological
word domain)
(37) a (niku) (mu) (temwa) I love him Nkhu
b (uku) (ni)(temwa) You love me nthe
c (khunilClllno) the tree here mpha
One final observation is that our rules so far can not explain why glides can be inserted in
the examples of (33) and (34) We have already pointed out the problem of GI in 21 The
additional complication is that even if we posit glides as a phoneme we still have to revise GIor
-341shy
make another one to explain post-lexical GL This will be addressed in 32
3 Syllable phonology of hiatus resol ving
In the discussion in section 2 some of the problems in the linear analysis were detected
(What are they ) In this section I will go over them and reanalyze the data with the help
of syllable structure to see whether we can answer some of the questions left open McCarthy and
Prince (1986) style of moraic syllable structure will be employed for our discussion in this section
The basic syllable structure of Chitumbuka is given in (38)
(38) Syllable structure in Chitumbuka a a b a c a
~~((r r1T ( (I I 1 All the syllables are mora final 2 All the syllables are mono-moraic
This is not the template that is required in the underlying form as will be discussed
Therefore the non-moraic tier may not have to be satisfied In other words though the structure
shows two pre-moraic consonant slots these can be left empty The two branches simple mean the
maximum number of consonants that can come before a vowel
18)1n the post lexical level this structural condition may not hold In Chitumbuka vowels in the stressed syllable is lengthened The long vowel is represented with bimoraic syllable structure It is not uncommon that some of the rules or conditions are applied only on the lexical level but not on the post lexical level such as extrametricality stress assignment and even VD and GF we talked about in the preceding section
- 342-shy
Hiatus Resolving in Chitumbuka 19
31 Syllabification
Contrary to the claim that all the segments are already syllabified in the base form (cf
Selkirk (1982raquo I assume that syllable structure is assigned through the syllabification process
The syllabification principles in Chitumbuka are given under
(39) Syllabification in Chitumbuka i) Morafication A segment projects a mora if it is more sonorous than the following
segrnent ii) Syllable projection Each mora projects a syllable node iii) Segment incorporation Unsyllabified segments are incorporated into syllables
Now lets take a look at the following sample syllabification
(40) a
0 0 00
I (iii)I(ii) --(i) f1 f1 f1 f1 ~ rlrOI I
~
I I n i u mb a n 1 u mb a n 1 u mb a n i u mb a
b
0 o 0 0 o 0
I I I (iW (~Of1 f1 f1 f1 ~ ~ I I r~r I I I I I ch i m a i ch i m a ch i m a ch i m a i
Notice that we do not need GF as a rule the same result of GF application is achieved as a
19)1 suppose that morafication principle is universally dependent upon the sonority hierarchy There are three different strategies in using sonority in syllabifications
i) y is moraic in xy if it is more sonorous than x ii) y is moraic in yz if it is more sonorous than z iii) y is moraic if its sonority is greater than N
The language in the world employ one or two all all of these in the syllabification and language may slightly vary in setting up the value of N in (iii)
-- 343shy
natural consequence of syllabification In 21 we obtain the simplicity of phonemic inventory at
the cost of introducing GF thereby complicating the rule component The environment of GF is
reflected in the syllabification principle Of course syllabification can not be an additional burden
the language needs syllable structure anyhow for example for the application of many phonological
rules and even for utterance
The nature of glide sounds are clearly reflected in the syllable structure If a vowel is
dominated by a mora it will be interpreted as a full vowel But if it is dominated by the syllable
node directly it will surface as a glide
32 Syllabification domain
The syllabification principles introduced in 31 have presupposed that the string of segments
are not syllabified in the underlying representation Suppose every morpheme is syllabified in the
deep structure and consider the following example of word formation
(41)
a aa a
~ (I (noa k 0 V 1 a k 0
Now we will have to make a rule for mora deletion which will be almost the repetition of
the syllabification principle Syllabification process in the course of derivation can get rid of the
redundant repetition of syllable structural description as well as save the later-on application of mora
deletion
I assume that the syllabification takes place in the first derived domain In other words
segments are syllabified after the initial word formation process Before going into the detailed
-344shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
Hiatus Resolving in Chitumbuka 17
(35) a (khuni) (ili) b (niku) (tu) (ocha) c (nikuwocha) (tuyuni)
One very interesting observation here is the change in the phonological word domains
(35b) and (35c) are semantically same The object in (35b) is pronominal while (35c) has a noun
object As a result of movement two phonological words are broken into three phonological
words This is not an arbitrary boundary just to explain the lack of hiatus resolving We may find
a very similar situation in English Intonational Phrase parsing
(36) a (As you know) (the tall boy is a genius) b (The tall boy) (as you know) (is a genius)
Nespor and Vogels (1986189) intonational phrase formation rule produces (36a) as well
as (36b) The syntax dependency of the phonological word formation is partly responsible for the
difference between (35b) and (35c)
Further hiatus is not the only rule sensitive to boundary difference in Chitumbuka We can
find that i-deletion and complex segment formation discussed in the preceding subsection shows
the same phenomenon Consider the following example (parentheses indicate the phonological
word domain)
(37) a (niku) (mu) (temwa) I love him Nkhu
b (uku) (ni)(temwa) You love me nthe
c (khunilClllno) the tree here mpha
One final observation is that our rules so far can not explain why glides can be inserted in
the examples of (33) and (34) We have already pointed out the problem of GI in 21 The
additional complication is that even if we posit glides as a phoneme we still have to revise GIor
-341shy
make another one to explain post-lexical GL This will be addressed in 32
3 Syllable phonology of hiatus resol ving
In the discussion in section 2 some of the problems in the linear analysis were detected
(What are they ) In this section I will go over them and reanalyze the data with the help
of syllable structure to see whether we can answer some of the questions left open McCarthy and
Prince (1986) style of moraic syllable structure will be employed for our discussion in this section
The basic syllable structure of Chitumbuka is given in (38)
(38) Syllable structure in Chitumbuka a a b a c a
~~((r r1T ( (I I 1 All the syllables are mora final 2 All the syllables are mono-moraic
This is not the template that is required in the underlying form as will be discussed
Therefore the non-moraic tier may not have to be satisfied In other words though the structure
shows two pre-moraic consonant slots these can be left empty The two branches simple mean the
maximum number of consonants that can come before a vowel
18)1n the post lexical level this structural condition may not hold In Chitumbuka vowels in the stressed syllable is lengthened The long vowel is represented with bimoraic syllable structure It is not uncommon that some of the rules or conditions are applied only on the lexical level but not on the post lexical level such as extrametricality stress assignment and even VD and GF we talked about in the preceding section
- 342-shy
Hiatus Resolving in Chitumbuka 19
31 Syllabification
Contrary to the claim that all the segments are already syllabified in the base form (cf
Selkirk (1982raquo I assume that syllable structure is assigned through the syllabification process
The syllabification principles in Chitumbuka are given under
(39) Syllabification in Chitumbuka i) Morafication A segment projects a mora if it is more sonorous than the following
segrnent ii) Syllable projection Each mora projects a syllable node iii) Segment incorporation Unsyllabified segments are incorporated into syllables
Now lets take a look at the following sample syllabification
(40) a
0 0 00
I (iii)I(ii) --(i) f1 f1 f1 f1 ~ rlrOI I
~
I I n i u mb a n 1 u mb a n 1 u mb a n i u mb a
b
0 o 0 0 o 0
I I I (iW (~Of1 f1 f1 f1 ~ ~ I I r~r I I I I I ch i m a i ch i m a ch i m a ch i m a i
Notice that we do not need GF as a rule the same result of GF application is achieved as a
19)1 suppose that morafication principle is universally dependent upon the sonority hierarchy There are three different strategies in using sonority in syllabifications
i) y is moraic in xy if it is more sonorous than x ii) y is moraic in yz if it is more sonorous than z iii) y is moraic if its sonority is greater than N
The language in the world employ one or two all all of these in the syllabification and language may slightly vary in setting up the value of N in (iii)
-- 343shy
natural consequence of syllabification In 21 we obtain the simplicity of phonemic inventory at
the cost of introducing GF thereby complicating the rule component The environment of GF is
reflected in the syllabification principle Of course syllabification can not be an additional burden
the language needs syllable structure anyhow for example for the application of many phonological
rules and even for utterance
The nature of glide sounds are clearly reflected in the syllable structure If a vowel is
dominated by a mora it will be interpreted as a full vowel But if it is dominated by the syllable
node directly it will surface as a glide
32 Syllabification domain
The syllabification principles introduced in 31 have presupposed that the string of segments
are not syllabified in the underlying representation Suppose every morpheme is syllabified in the
deep structure and consider the following example of word formation
(41)
a aa a
~ (I (noa k 0 V 1 a k 0
Now we will have to make a rule for mora deletion which will be almost the repetition of
the syllabification principle Syllabification process in the course of derivation can get rid of the
redundant repetition of syllable structural description as well as save the later-on application of mora
deletion
I assume that the syllabification takes place in the first derived domain In other words
segments are syllabified after the initial word formation process Before going into the detailed
-344shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
make another one to explain post-lexical GL This will be addressed in 32
3 Syllable phonology of hiatus resol ving
In the discussion in section 2 some of the problems in the linear analysis were detected
(What are they ) In this section I will go over them and reanalyze the data with the help
of syllable structure to see whether we can answer some of the questions left open McCarthy and
Prince (1986) style of moraic syllable structure will be employed for our discussion in this section
The basic syllable structure of Chitumbuka is given in (38)
(38) Syllable structure in Chitumbuka a a b a c a
~~((r r1T ( (I I 1 All the syllables are mora final 2 All the syllables are mono-moraic
This is not the template that is required in the underlying form as will be discussed
Therefore the non-moraic tier may not have to be satisfied In other words though the structure
shows two pre-moraic consonant slots these can be left empty The two branches simple mean the
maximum number of consonants that can come before a vowel
18)1n the post lexical level this structural condition may not hold In Chitumbuka vowels in the stressed syllable is lengthened The long vowel is represented with bimoraic syllable structure It is not uncommon that some of the rules or conditions are applied only on the lexical level but not on the post lexical level such as extrametricality stress assignment and even VD and GF we talked about in the preceding section
- 342-shy
Hiatus Resolving in Chitumbuka 19
31 Syllabification
Contrary to the claim that all the segments are already syllabified in the base form (cf
Selkirk (1982raquo I assume that syllable structure is assigned through the syllabification process
The syllabification principles in Chitumbuka are given under
(39) Syllabification in Chitumbuka i) Morafication A segment projects a mora if it is more sonorous than the following
segrnent ii) Syllable projection Each mora projects a syllable node iii) Segment incorporation Unsyllabified segments are incorporated into syllables
Now lets take a look at the following sample syllabification
(40) a
0 0 00
I (iii)I(ii) --(i) f1 f1 f1 f1 ~ rlrOI I
~
I I n i u mb a n 1 u mb a n 1 u mb a n i u mb a
b
0 o 0 0 o 0
I I I (iW (~Of1 f1 f1 f1 ~ ~ I I r~r I I I I I ch i m a i ch i m a ch i m a ch i m a i
Notice that we do not need GF as a rule the same result of GF application is achieved as a
19)1 suppose that morafication principle is universally dependent upon the sonority hierarchy There are three different strategies in using sonority in syllabifications
i) y is moraic in xy if it is more sonorous than x ii) y is moraic in yz if it is more sonorous than z iii) y is moraic if its sonority is greater than N
The language in the world employ one or two all all of these in the syllabification and language may slightly vary in setting up the value of N in (iii)
-- 343shy
natural consequence of syllabification In 21 we obtain the simplicity of phonemic inventory at
the cost of introducing GF thereby complicating the rule component The environment of GF is
reflected in the syllabification principle Of course syllabification can not be an additional burden
the language needs syllable structure anyhow for example for the application of many phonological
rules and even for utterance
The nature of glide sounds are clearly reflected in the syllable structure If a vowel is
dominated by a mora it will be interpreted as a full vowel But if it is dominated by the syllable
node directly it will surface as a glide
32 Syllabification domain
The syllabification principles introduced in 31 have presupposed that the string of segments
are not syllabified in the underlying representation Suppose every morpheme is syllabified in the
deep structure and consider the following example of word formation
(41)
a aa a
~ (I (noa k 0 V 1 a k 0
Now we will have to make a rule for mora deletion which will be almost the repetition of
the syllabification principle Syllabification process in the course of derivation can get rid of the
redundant repetition of syllable structural description as well as save the later-on application of mora
deletion
I assume that the syllabification takes place in the first derived domain In other words
segments are syllabified after the initial word formation process Before going into the detailed
-344shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
Hiatus Resolving in Chitumbuka 19
31 Syllabification
Contrary to the claim that all the segments are already syllabified in the base form (cf
Selkirk (1982raquo I assume that syllable structure is assigned through the syllabification process
The syllabification principles in Chitumbuka are given under
(39) Syllabification in Chitumbuka i) Morafication A segment projects a mora if it is more sonorous than the following
segrnent ii) Syllable projection Each mora projects a syllable node iii) Segment incorporation Unsyllabified segments are incorporated into syllables
Now lets take a look at the following sample syllabification
(40) a
0 0 00
I (iii)I(ii) --(i) f1 f1 f1 f1 ~ rlrOI I
~
I I n i u mb a n 1 u mb a n 1 u mb a n i u mb a
b
0 o 0 0 o 0
I I I (iW (~Of1 f1 f1 f1 ~ ~ I I r~r I I I I I ch i m a i ch i m a ch i m a ch i m a i
Notice that we do not need GF as a rule the same result of GF application is achieved as a
19)1 suppose that morafication principle is universally dependent upon the sonority hierarchy There are three different strategies in using sonority in syllabifications
i) y is moraic in xy if it is more sonorous than x ii) y is moraic in yz if it is more sonorous than z iii) y is moraic if its sonority is greater than N
The language in the world employ one or two all all of these in the syllabification and language may slightly vary in setting up the value of N in (iii)
-- 343shy
natural consequence of syllabification In 21 we obtain the simplicity of phonemic inventory at
the cost of introducing GF thereby complicating the rule component The environment of GF is
reflected in the syllabification principle Of course syllabification can not be an additional burden
the language needs syllable structure anyhow for example for the application of many phonological
rules and even for utterance
The nature of glide sounds are clearly reflected in the syllable structure If a vowel is
dominated by a mora it will be interpreted as a full vowel But if it is dominated by the syllable
node directly it will surface as a glide
32 Syllabification domain
The syllabification principles introduced in 31 have presupposed that the string of segments
are not syllabified in the underlying representation Suppose every morpheme is syllabified in the
deep structure and consider the following example of word formation
(41)
a aa a
~ (I (noa k 0 V 1 a k 0
Now we will have to make a rule for mora deletion which will be almost the repetition of
the syllabification principle Syllabification process in the course of derivation can get rid of the
redundant repetition of syllable structural description as well as save the later-on application of mora
deletion
I assume that the syllabification takes place in the first derived domain In other words
segments are syllabified after the initial word formation process Before going into the detailed
-344shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
natural consequence of syllabification In 21 we obtain the simplicity of phonemic inventory at
the cost of introducing GF thereby complicating the rule component The environment of GF is
reflected in the syllabification principle Of course syllabification can not be an additional burden
the language needs syllable structure anyhow for example for the application of many phonological
rules and even for utterance
The nature of glide sounds are clearly reflected in the syllable structure If a vowel is
dominated by a mora it will be interpreted as a full vowel But if it is dominated by the syllable
node directly it will surface as a glide
32 Syllabification domain
The syllabification principles introduced in 31 have presupposed that the string of segments
are not syllabified in the underlying representation Suppose every morpheme is syllabified in the
deep structure and consider the following example of word formation
(41)
a aa a
~ (I (noa k 0 V 1 a k 0
Now we will have to make a rule for mora deletion which will be almost the repetition of
the syllabification principle Syllabification process in the course of derivation can get rid of the
redundant repetition of syllable structural description as well as save the later-on application of mora
deletion
I assume that the syllabification takes place in the first derived domain In other words
segments are syllabified after the initial word formation process Before going into the detailed
-344shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
Hiatus Resolving in Chitumbuka 21
discussion consider the following English stress data (from Inkelas (1989132))
(42)a parent parental b nation national origin original federal federalist miracle miraculize national nationalize diplomat diplomacy intimate intimacy
We can easily see that (42a) words show stress changes when suffixes are added while
(42b) words do not have such alternation Inkelas (1989 132) attributes the difference of (42a) and
(42b) words to the difference of the morphological structure In short the left column of (42a)
words are underived (one morpheme) while the words in the left column of (42b) can be broken
into the combination of morphemes
This indicates that English stress domain is a derived word domain When the affixes are
added to (42a) words the stems are without stress and the stress is assigned taking the whole
string into account along with other necessary conditions such as heavyflight distinction and
extrametricality Though not explicitly explained in Inkelas(1989) it may be safe to assume that
underived words also should go through stress assignment in some later stage
Now returning to Chitumbuka we can make the similar claim underived words are
syllabified later and the first syllabification takes place when the morphemes are combined together
through word formation process for the first time Therefore the syllabification domain correspond
to (16) and (17)
After VD and syllabification in the relevant domain there is one optional operation called
Onset creation (OC)
(43) Onset Creation (OC)
a a I~
I1 ~ ~ I (R segmental root) I
R R R
~~ 345~
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
This rule creates onset to the mora-initial syllable The created segment is thought to be
maximally underspecified and the relevant features are filled by spreading from the following
vowel This segment will therefore be realized either as [w] or as [y] according to the feature
contents of the vowel as shown in (44)
(44) Feature spreading
rr a
R RI [labial]
Without going into details of the feature geometry I will simply assume that the [round] or
[labial] feature is spread onto the newly created root to make a round glide [w] If there is no
feature like that in the vowel (if it is underspecified) the default feature filling rule will assign [shy
round] and therefore the segment will appear as [y]
OC is the replacement of GI Notice that it does not require intervocalic position in the
structural description of the rule environment nor the sonority condition Now we have three rules
to explain the hiatus as in (45)
(45) a Vowel Deletion (=VD) b Syllabification (=Syll) c Onset Creation (=OC)
OC is intrinsically ordered after Syll since it operates on the syllable structure And VD is
thought to be ordered before syllabification because it needs specific morphological boundary
information
Now lets see how this words with sample derivations Consider ku och + a A
-346shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
Hiatus Resolving in Chitumbuka 23
assumed in 22 (cf fn 7) suffixation takes place earlier and loch + al is the first derived domain
VD does not apply And syllabification produces (0)0 (chaj OC mayor may not be applied to
this structure If applied we get (wo)o(cha)o if not (0)0 (chaj On the next level of word
formation either of the two forms can be the input First if the input is (wo (chaj the
syllabification on the next level will result in (ku)o (wo)o(cha)o and if (o)o(cha)o is taken VD
applies to produce (ko)a(cha)a
But in case of the form mu opal OC can not be applied Since it is the first affixation
the entire string is the domain of sy llabification The outcome of the syllabification is (moj Cpa)o
With VD within segmental framework we have to say that VD is optional in ku och + al and
obligatory in mu opal Again we were unable to explain why GI is obligatory in mu opal but
not in ku och +a Notice also that resorting to boundary in VD and GI does not help at all The
two examples have same boundary between two same vowel sequence u - 0
33 Word-level well-formedness condition
We have seen that syllable-based approach to hiatus resolving in Chitumbuka solves the
majority of the problems discussed in the preceding section One final question is what happens to
underived words As implicitly assumed by Inkelas (1989) I will posit a word-level default
syllabification The reason is simple all the lexical words must be syllabified before they leave the
lexicon It6 (1986 3) expressed this idea in her Prosodic Licensing condition I suppose that there
are two different types of word-level well-formedness condition
(46) Well-formedness condition a Prosodic Licencing
All the segments should be incorporated into syllables b Non-moraic Adjacency
No two moraic segments should be segmentally adjacent
~347-
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
(46a) is universally important without this condition we can easily see that all kinds of
imaginable abnormal phenomena may happen thereby making phonological theories virtually
useless Therefore this condition is an absolute prerequisite of (46b) (46b) is the simple repetition
of well-formedness condition in (6) This underlies all our discussion so far
Now we will pick up one minor point left unexplained If OC is optional we might expect
that Inail (four) may have two different syllable structure (na)o (i)o or (na)o (yi)o nai does not
have boundary between al and Iii Therefore it does not undergo VD and clearly (naj is
wrong We can not make OC obligatory because if we do so we can not explain the two different
forms of verbs that have V-initial stems
The well-formedness conditions given in (46) is the answer Figuratively speaking these
conditions are janitors on duty at the exit of lexicon Both (naj (i)o and (na)o (yi)o will be
generated But the conditions allow the second form to go out from the lexicon onto the postshy
lexical level But the ill-formed structure (na)o (i)o is not released from the lexicon and will be
discarded
4 Conclusion
So far we have seen many different types of vowel clashes and discussed what kinds of
processes are used to resolve the hiatus to meet the well-formedness condition
The segmental analyses given in section 2 pose some problems in such area as GF rule
formulation the rule nature of GF GI and VD and the nature of the segment inserted by GL GI
inserts glide but there is no glide in the phonemic inventory of Chitumbuka Of course there arc
non-phonemic sound insertions and changes into non-phonemic sounds in natural languages But
--348shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
Hiatus Resolving in Chitumbuka 25
all these rules which Chomsky and Halle (1968) called phonetic implementation rules are quite
automatic in nature and ordered after all the phonological rules Since GI interacts with other
phonological rules We can not say that it is a phonetic implementation rule
We may say that the inserted sounds are vowels This may give birth to an unlimited
sequence of vowels it is iterative Though we claim that vowel insertion is non-iterative the
inserted vowel should undergo GF rule This GI - GF ordering contradicts with the ordering
proposed in (29)
The same data were reanalyzed again using moraic syllable structure This syllable-based
approach provides answers to all the problems that segmental level analyses suffered from
Three major rule VD SyII and OC are introduced Within this framework GF does not
participate as a rule It is shown that syllabification produces the same results as GF GI is
replaced by OC which is an optional rule on the syllable domain One of the problems of GI was
that this optional rule should be obligatorily applied if other optional rule is not applied Here VD is
always obligatory and the optional nature of VD is explained by DC that bleeds VD on the next level
of (re-)syllabification
The created onset root is assumed to be maximally underspecified and the spreading or lack
of spreading of relevant feature from the vowel in the same syllable correctly produces glides
without digressing into the problem of the phonemic status of the glide Truly in the moraic
syllable structure as given in the section 3 does not make any distinction between vowels and
consonants only their relative sonority is thought to be important
It was also seen that in some cases the vowel clashes are tolerated Vowel clashes across
the phonological boundaries are left unresolved while they are resolved it happened across
morphological word boundary within a phonological word
Still there are some problems which were not addressed in this paper First apparently
-349shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
26 7~~ ~1lyenJjC1993 2)
irregular behavior ofu-initial vowel stems (cf (27m) (27n) and (270)) was not explained though
a couple of speculations were made Secondly the distinction between morphological and
phonological word was not fully investigated It is simply assumed that there might be differences
But since the difference is of crucial importance Further findings of rules that makes use of this
distinction may help clarify the nature of the difference In addition the environements of GD or
the GO rule itself may also be problematic This rule needs boundary information and clearly is
lexical Coronal sounds are thought to be less marked and the feature [coronal] is usually
underspecified in many languages Still GD as proposed in this paper needs specified [coronal]
feature in the relatively early level of phonological derivation This might be an indication that
something is wrong with the rule formulation
-350shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
Hiatus Resolving in Chitumbuka 27
References
Aronoff Mark (1976) Word Formation in Generative Grammar Linguistic Inquiry monograph 1 Cambridge Mass MIT press
Chomsky N amp Morris Halle (1968) The Sound Pattern ofEnglish New York Harper amp row
Diocese of Mzuzu (1970) Tumbuka Grammar
Hyman Larry (1976) Phonology Theory andAnalysis New York Holt Rinehart and Winston
Jesperson Otto (1932) Lehrbuch der Phonetik (5th edition) Leipzig and Berlin
Inkelas Sharon (1989) Prosodic Constituency in the Lexicon PhD dissertation Stanford University Stanford California
It6 Junko (1986) Syllable Theories in Prosodic Phonology PhD dissertation University of Massachusetts Amherst Also published by Garland (1988) New York amp London
Lupenga Mphande (1989) A Phonological Analysis ofIdeophones in Chitumbuka PhD dissertation University of Texas Austin
McCarthy John J (1988) Feature Geometry and dependency A review Phonetica 43 84-108
McCarthy 1 1 and A Prince (1986) Prosodic Morphology manuscript University of Massachusetts Amherst and Brandeis University
Nespor M and I Vogel (1986) Prosodic phonology DordrechtForis
Sagey Elizabeth C (1986) The Representation ofFeatures and Relations in Non-linear Phonology PhD dissertation MIT
Selkirk E O (1982) The syllable In van der Hulst and Smith (eds) The Stucture of Phonological Representations Part II Dordrecht Foris
Turner William Y (1952) Tumbuka - Tonga - English Dictionary Blantyre Malawi Church of Scotland Mission
Vail H Leroy (1971) Noun classes of Tumbuka African Studies 3035 - 59
-351shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy
~-yen-7o-j(Chitumbuka) u~Jf- ~~ J TL~ 7gt1JL 3J~y Jif TLojIA~ 2il 2g cl
ltj~iic~ --L 7SJf-~ 71II1ii BJ-f-o-j(Bantu languagesr s] ~iJJl 11~ 7lt1JL 3J4 lt3] 0]
ltdo-j o]l1 cJ ~2] ~7](Glide formation) 2g ~i2f(Vowel deletion) sr= olncJ~2J AJOJ
(Glide insertion) ~ E1 Tflt~l ltgt1] E1 ii9 2 E1 oj Sl1 ~ ~ degB -8- ~H-c] 3J q
~JL iA~ 2ltJ-ojIA~ 0~t-i7lt] _2g~ ~AJ-~ ~~ii~ ~ -8-~(Segmental phonology)
~oJ AJ~ AV]iiJL olt-1~ -n-~E1 ~H)-~5 2g ~~ AJ~ii )01 poundJEtTLojIA~ o]ncJ
~2J~ ~5 AJ7jlt5~O]ii -i-AJ-AjE1 -rAil -n-~ ~--il -n-~ rA~ u~7Joj1 31o-jA~E1 AJ-~2r
51 -i[-Ai] rJdeg] 3Jg ~ -ti14 3ltJ-ojlA~ 2~oj1A~ -tdeg~T 2g ~ sectjgt11E1 Al1 ~AJ- g ~ -8-~
(Syllable phonolohyr-sl JHJo11A~ 2t-(mora)~ pounddegJiic~ lt5~~~ A]poundiii1q o]ncJ ~2] ~71
~AJ-~ g ltJsect)- Jl7jE1 AoL~cj 1Jl0J~ -tc]JL o]ncJ ~2J AJOJ ElAl Jsect)-E1 Tg ~ltj Jl
ltj~5 i2J iii14 2 ~i2flt sr=~ -T- lt]ltjJlltjoJ gt~--51A1 ii~ ~Jlltgt11 ut~- o~Jl(Filter)5
12J ~ T- 3l g -ti1 q
olt-i~ 11~ Jlltjoj1A~ U-2g~5 Al~~-i- o-j7JE1 01~ ~A-~ AJ~iiJ~1ltgt11 3Jo-jA~E1-i[shy
(PrJ L 2JJL g -8-~lt~ ~o-j (Phonological word) 9+ lt5J Ell ~~ ~o-j (Morphological word) E1 Aj E17
~71 ii~5 ii9 J-]~i- -i[- Ai17 ~gt] ii~y i-JLE1 0j ~ 8-iio~ -yen-7o-jE1 2~
-352shy