Upload
andrew-gilmore
View
213
Download
0
Embed Size (px)
Citation preview
A quantitative model of word order and movement in English, Dutch and German complement constructions
Karin Harbusch & Gerard Kempen
Computer Science Psychology Dept.,
Dept., University Leiden University
of Koblenz- & Max Planck
Landau Institute Nijmegen
Preview
1. Introduction2. Essentials of Performance Grammar
Hierarchical component Linearization component: topologies Topology sharing in the three target languages Linear order and typed feature unification
3. Examples in the three target languages4. Conclusions
1. Introduction
Linear order in English, Dutch and German complement constructions varies considerably w.r.t.: Wh-extraction, clause union, extraposition, verb clustering, particle movement, etc.
We show that both the within- and between-language variations of these phenomena reduce to differences between a few numerical parameters.
2. Essentials of Performance Grammar
Performance Grammar (PG) is a psycho-linguistically motivated formalism. Psycholinguistic phenomena suggest separate hierarchical and linear grammar components.
We focus on the linear component and describe it in declarative terms based on feature unification.
The parametrization scheme we propose belongs to the linear component.
S
HD
v
S
SUBJ
NP
S
DOBJ
NP
S
IOBJ
NP|PP
S
PRED
NP|ADJP
HD
v
SUBJ
NP
S
DOBJ
NP
MOD*
ADVP|PP | S
Segments such as in clauses ...
fool
S
CMPR
CP
S
PRT
prep|adv
S
CMP
S|PP| PP
S
PINF
PP
…
... combine intolexical , e.g. clausal, frames
Example
“Slim snijdertje fopte dertig zeerovers”
“Clever tailor fooled thirty pirates”
(Title of Dutch children’s story by Annie M.G. Schmidt)
HD
crd
dertig/thirty
ADJP
MOD*
ADVP
HD
adj
slim/clever
NP
DET
DP
Q MOD*
ADJP|PP
HD
n
snijdertje/tailor
NP
DET
DP
Q MOD*
ADJP|PP
HD
n
zeerover/pirate
S
SUBJ
NP
HD
v
fopte/fooled
DOBJ
NP
MOD*
ADVP |PP|S
CNP
CNP
CNP
Lexical frames from the mental lexicon
HD
crd
dertig/thirty
ADJP
MOD*
ADVP
HD
adj
slim/clever
NP
DET
DP
Q MOD*
ADJP|PP
HD
n
snijdertje/tailor
NP
DET
DP
Q MOD*
ADJP|PP
hd
n
zeerover/pirate
S
SUBJ
NP
HD
v
fopte/fooled
DOBJ
NP
MOD*
ADVP|PP|S
CNP CNP
CNP
Substitution (feature structures are omitted)
HD
crd
dertig/thirty
ADJP
HD
adj
slim/clever
MOD HD
n
snijdertje/tailor
q HD
n
zeerover/pirate
S
SUBJ
NP
DOBJ
NP
HD
v
fopte/fooled
CNP
Reduced dominance structure
HD
v
fopte/fooled
HD
v
fopte/fooled
Forefield Midfield Endfield
Linearization ComponentData structure: topology
A topology is associated with the foot node layer of every lexical frame
Function: reservation of work/storage space for frame constituents
Topologies with nine slots for clauses:
English
Dutch/German
F1 F2 F3 M1 M2 M3 M4 E1 E2
F1 M1 M2 M3 M4 M5 M6 E1 E2
Assigning slot positions (English)
Slot Filler
F1 Declarative main clause: Topic, Focus (one constituent only) Interrogative main clause: Wh-constituent Complement clause: Wh-constituent
F2 Complement clause: CoMPLementizeR that
F3 Subject (iff non-Wh)
M1 Pre-INFinitive to < HeaD verb (oblig.) < PaRTicle
M2 Direct Object (iff personal pronoun) Interrogative main cl.: Subject (iff non-Wh); SUBJ < DOBJ
M3 Indirect OBJect < Direct OBJect (non-Wh)
M4 PaRTicle
E1 Non-finite Complement of 'Verb Raiser‘ (in particular Auxiliaries)
E2 Non-finite Complement of 'VP Extraposition verb‘ Finite Complement clause
Slot positions for Dutch & German Slot Filler
F1 Declarative main cl.: SUBJect, Topic or Focus (one constituent only) Interrogative main clause: Wh-constituent Complement clause: Wh-constituent
M1 Main clause: HeaD verb Complement clause: CoMPLementizer dat/om (Du.), dass (Ger.)
M2 Subject NP (iff non-Wh), Direct OBJect (iff personal pronoun)
M3 Direct OBJect < Indirect OBJect (iff non-Wh)
M4 PaRTicle (Du. only)
M5 Non-finite CoMPlement of Verb Raiser
M6 Subordinate clause: Du.: Pre-INFinitive te < HeaD verb Ger.: PaRTicle < Pre-INFinitive zu < HeaD verb
E1 Non-finite Complement of 'Verb Raiser‘ (Du. only)
E2 Non-finite Complement of 'VP Extraposition verb‘
Finite Complement clause
HD
crd
thirty
ADJP
HD
adj
clever
MOD HD
n
tailor
Q HD
n
pirate
S
SUBJ HD DOBJ
CNP
English clausal topology
F1 F2 F3 M1 M2 M3 M4 E1 E2v
fooled
NP NP
HD
crd
dertig
ADJP
HD
adj
slim
MOD HD
n
snijdertje
Q HD
n
zeerover
S
SUBJ HD DOBJ
CNP
Dutch clausal topology
F1 M1 M2 M3 M4 M5 M6 E1 E2v
fopte
NP NP
Topology sharing
If a sentence consists of a main clause plus one or more complement clauses, each of the clauses (i.e. verb frames) instantiates its own topology. In such cases, topologies are allowed to share slots, conditionally upon several restrictions.
After two slots have been shared, they are no longer distinguishable; in fact, they are the same object.
This operation may cause upward movement of constituents: "promotion".
General constraints on topology sharing: Only between adjacent clausal topologies
Only between identically labeled slots
HeaD slot never participate in sharing
Only left- and/or right-peripheral
left-peripheral central non-shared right-peripheral shared area (LS) area shared area (RS)
… HD …
S
HD
v
did
CMP
S
SUBJ
NP
John HD
v
try
CMP
S
DOBJ
NP
who
PINF
PP
to
HD
v
call
F1 M1 M2 E1
E1M1
M1 M1
Example (cont.)
F1
F1
S
HD
v
did
CMP
S
SUBJ
NP
John HD
v
try
CMP
S
PINF
PP
to
HD
v
call
F1 M1 M2 E1
E1M1
M1 M1
Example (cont.)
F1
DOBJ
NP
who
F1
S
HD
v
did
CMP
S
SUBJ
NP
John HD
v
try
CMP
S
DOBJ
NP
Who
PINF
PP
to
HD
v
call
F1 M1 M2 E1
E1M1
M1 M1
Example (cont.)
F1
F1
S
HD
v
did
SUBJ
NP
Poirot
CMP
S
HD
v
make
DOBJ
NP
DET
DP
the
HD
n
claim
CMP
S
DOBJ
NP
who
CMPR
CP
that
SUBJ
NP
he
HD
v
saw
MOD
NP
last week
Who did Poirot claim that he saw last week? *Who did Poirot make the claim that he saw last week?
Unsuccessful attempt at sharing
x
F1
F1
Language-spec. values for LS/RSClause type English Dutch German
Interrogative LS=0 RS=0
LS=0 RS=1
LS=0 RS=1
Declarative & Finite LS=1 RS=0
LS=1 RS=1
LS=1 RS=1
Decl. & Non-Finite, VP Extraposition
LS=3 RS=0
LS=1 RS=1
LS=1 RS=1
Decl. & Non-Finite, Verb Raising
LS=3 RS=0
LS=4:6 RS=1
LS=5 RS=1
Decl. & Non-Finite, Third Construction
n.a. LS=1:6 RS=1
LS=1:6 RS=1
Within-language parametrization: Slot assignment (Engl.)
M1 … M4 E1 E2F1 F2 F3
HD
Decl. & Non-Finite, VP Extraposition
SUBJ
Parameters (cont.): English sharing left-peripheral central non-shared RS shared area (LS) area
M1 M2 … E2F1 F2 F3
Decl. & Non-Finite, VP Extraposition
HD SUBJ
Between-language parametrization left-peripheral central non-shared RS shared area (LS) area
M1 M2 … E2F1 F2 F3
HD verb in subclause
M6 E1F1 M1 M2 M3 M4 M5
English Decl. & Non-Finite, VP Extraposition
Dutch/German Decl. & Non- Finite, VP Extraposition
SUBJ
E2
Comparison: Engl./Du./Ger. sharing left-peripheral central non-shared RS shared area (LS) area
M1 M2 … E2F1 F2 F3
M6 E1
F1 M1 M2 M3 M4 M5
English Decl. & Non-Finite, VP Extraposition
Dutch/German Decl. & Non- Finite, VP Extraposition
E2
Specification of topologies in terms of typed feature unification
S [tpl p(1)t, p(2)t, ... p(9)t] where p(i)t denotes the type of the ith member of the list. For each of the target languages 9 slot types are defined (e.g., F1t). Slots are attributes that take a non-branching list of lemmas or constituents (e.g. SUBJect-NP, CoMPlement-S or HeaD-v) as their value.
Slots are initialized with the value empty list, denoted by "" (e.g., [F1t F1 ]. Lists of segments can be combined by the append operation, represented by the symbol ”O". A slot type may impose a constraint on the cardinality (the number of members) of the list serving as its value. Cardinality constraints are expressed as subscripts of the value list. E.g., the subscript "c=1" in [F1t F1 c=1] states that the list serving as F1's value should contain exactly one member.
Specification of topologies (cont.) Depending on the values of sharing parameters LS and RS,
the list is divided into a left area, the central area, and the right area. LS and RS are set to zero by default; this applies to the root S of main clauses and adverbial subordinate clauses. The root S of a complement clause obtains its sharing parameter values from the foot of the S-CMP-S segment belonging to the lexical frame of its governing verb.
Sharing (see 1 ) simply means unifying the slots in the two laterally shared areas according to the LS and RS parameters.
The contents of non-shared (central) slots are appended to the contents of the receiving slot
(see 2 ).
S
CMP
S
[tpl 1 F1, ..., E2 2 ]
2 tpl 1 F1, ... ctype decl-fin
3. English question formationWho do I have to call?
F1 F2 F3 M1 M2 M3 M4 E1 E2
do
have
Who to call
The non-finite complements of both do and have are declarative. (Cf. the paraphrase "For which person x is it the case that I have to call x", which highlights the scope of who.) It follows that LS=3 in both complements. Do is a Verb Raiser, have (in have to) is a VP Extraposition verb.
English question formation (cont.)Who did you say John saw?
F1 F2 F3 M1 M2 M3 M4 E1 E2
did you
say
Who John saw
The lower clause is finite and declarative (LS = 1) — cf. the paraphrase “For which person x is it the case that you said that John saw x”. (The scope of who exceeds its ‘own’ clause and includes the matrix clause.) LS = 3 in the middle topology.
English question formation (cont.)
I know who John sawF1 F2 F3 M1 M2 M3 M4 E1 E2
I know
Here, the scope of the interrogative pronoun does not include the main clause (“I know for which person x it is the case that John saw x”). Therefore, the complement is interrogative and does not share its F1 slot with that of the main clause (LS = 0).
who John saw
I know who John saw
a. Who did you claim that you saw last week?
English question formation (cont.)
S
HD
v
did
SUBJ
NP
you
CMP
S
HD
v
claim
CMP
DOBJ
NP
who
CMPR
CP
that
SUBJ
NP
you
HD
v
saw
MOD
NP
last week
S
F1
F1
F1
a. Who did you claim that you saw last week?
b.*Who did you make the claim that you saw last week?
Island Effects in English
S
HD
v
did
SUBJ
NP
you
CMP
S
HD
v
make
DOBJ
NP
DET
DP
the
HD
n
claim
CMP
S
DOBJ
NP
who
CMPR
CP
that
SUBJ
NP
you
HD
v
saw
MOD
NP
last week
x
F1
F1
a. Zag je dat? saw you that ‘Did you see that?’
Dutch question formation
F1 M1 M2 M3 M4 M5 M6 E1 E2
Zag je dat
Dutch interrogative main clauses feature Subject-Verb inversion without the equivalent of do-insertion:
a´. Je zag dat?
F1 M1 M2 M3 M4 M5 M6 E1 E2
Je zag dat
b. Wie zag dat? who saw that ‘Who saw that?’
c. Wat zagen ze? ‘What did they see?’
Dutch question formation (cont.)
F1 M1 M2 M3 M4 M5 M6 E1 E2
Wat zagen ze
F1 M1 M2 M3 M4 M5 M6 E1 E2
Wie zag dat
Dutch question formation (cont.)
Zij vroeg of ik Jan kende She asked whether I John knew‘She asked whether I knew John’
of ik Jan kende
Because the complement is interrogative here, the sharing rule prohibits left-peripheral sharing: LS=0.
F1 M1 M2 M3 M4 M5 M6 E1 E2
Zij vroeg
... dat ik Jan zal bellen that I John will phone'... that I will phone John
The subordinate clause features clause union, causing the auxiliary zal to intervene between the Direct OBJect Jan and its governor bellen. The left-peripheral sharing area may vary between 4 and 6 slots (LS=4:6). Because Jan lands in M3, i.e. in the shared area, it is promoted. The remainder of the lower topology, including the HeaD bellen itself, occupies E1 — one of the options of the complement of a Verb Raiser.
F1 M1 M2 M3 M4 M5 M6 E1 E2
dat ik zal
Jan bellen
Clause Union in Dutch
... dat ik Jan bellen zal that I John phone will '... that I will phone John'
F1 M1 M2 M3 M4 M5 M6 E1 E2
dat ik zal
Jan bellen
Clause Union in Dutch (cont.)
F1 M1 M2 M3 M4 M5 M6 E1 E2
dat ik zou
The positions marked by "" are grammatical alternatives to the particle (op) position mentioned in the example; no other positions are allowed. Given LS=4:6 for complements of Verb Raisers, it follows that Jan is obligatorily promoted into the higher topology. However, sharing of the fifth slot (M4) is optional.
... dat ik Jan zou hebben op gebeld that I John would have up called'... that I would have called John up'
hebben
Jan op gebeld
Dutch Particle Hopping
... dat ik Jan zou hebben op gebeld that I John would have up called'... that I would have called John up'
Jan op gebeld
Dutch Particle Hopping (cont.)
E2M5M3
zouikdat
E1M6M4M2M1F1
hebben
Second level: LS=4:6, third level: LS=4
... dat ik Jan zou op hebben gebeld that I John would up have called'... that I would have called John up'
hebben
Jan gebeldop
Dutch Particle Hopping (cont.)
E2M5M3
zouikdat
E1M6M4M2M1F1
Second level: LS=4, third level: LS=4:6
... dat ik Jan op zou hebben gebeld that I John up would have called'... that I would have called John up'
E2M5M3
zouikdat
E1M6M4M2M1F1
hebben
Jan gebeldop
Dutch Particle Hopping (cont.)
Second level: LS=4:6, third level: LS=4:6
Assumption about the order of constituents that land in the same slot but originate from different levels in the clause hierarchy: We stipulate that constituents from more deeply embedded clauses follow constituents belonging to higher clauses.
... dat ik Jan de fiets wil helpen maken that I John the bike want-to help repair'... that I want to help John to repair the bike'
F1 M1 M2 M3 M4 M5 M6 E1 E2
dat ik wil
Jan helpen
de fiets maken
Dutch Cross-serial Dependency
German VP Extraction... dass er uns zwingt es zu tun that he us (Akk.) forces it to do'... that he forces us to do it'
F1 M1 M2 M3 M4 M5 M6 E1 E2
dass er uns zwingt
es zu tun
Parametrization for German VP Extraposion verbs: shared areas: LS=1, RS=1 slot assignment for complement clause: E2
German Third Constructiona. ... dass er uns verspricht es zu tun that he us (Dat.) promises it to do '... that he promises us to do it'
uns
E2M5M3
versprichterdass
E1M6M4M2M1F1
es zu tun
Parametrization for German Third Construction verbs: shared areas: LS=1:6, RS=1 slot assignment for complement clause: M5 or E2
German Third Construction (cont.)b. ... dass er uns es zu tun verspricht
c. ... dass er es uns zu tun verspricht
F1 M1 M2 M3 M4 M5 M6 E1 E2
dass er uns verspricht
es zu tun
F1 M1 M2 M3 M4 M5 M6 E1 E2
dass er uns verspricht
es zu tun
Third Construction (cont.)d. ... dass er es uns verspricht zu tun
e. ? ... dass er uns es verspricht zu tun
F1 M1 M2 M3 M4 M5 M6 E1 E2
dass er uns verspricht
es zu tun
F1 M1 M2 M3 M4 M5 M6 E1 E2
dass er uns verspricht
es zu tun
4. Conclusions We have shown that the introduction of topologies with a
fixed number of slots, in conjunction with cross-clause lateral topology sharing enables a simple treatment of word order and movement (promotion) in complement structures of the three target languages. The great amount of within- and between-language variation typical of these constructions could be analyzed as resulting from different settings of a small number of quantitative parameters (size of shared areas; slot number of landing site targeted by the complement clause, by head verb, and by other major constituents).
Due to space limitations we could not go into much detail. Elsewhere we have provided a more fine-grained discussion of our approach and its psycholinguistic motivation. Future study is needed to find out whether the PG approach generalizes to other languages.
Thank you!
Sources of the examples: Haegeman, 1994 Kathol, 2000 Rambow,1994 Sag & Wasow, 1999
For more PG details see
http://www.uni-koblenz.de/~harbusch/pg.html
NP
HD
n|pro
NP
DET
DP
NP
Q
CNP
NP
MOD
ADJP|PP|S
NP
CMP
PP|S
HD
n
DET
DP
NP
Q
CNP
MOD*
ADJP|PP | S
Nominal Segments ...
pirate
... form nominal lexical frames
Example
Hierarchical structure
Simplified lexical frames underlying the sentences We know Dana hates Kim and Kim we know Dana hates.
HD
pro
we
SUBJ HD
v
DOBJ
NP
Dana
S
SUBJ
NP
HD
v
know
CMP
S
NP
Kim
hates
Example (cont.)
Topology slot assignment
The focused Direct OBJect Kim may go to M3, producing:
We know Dana hates Kim
HD
pro
we
SUBJ HD
v
DOBJ
NP
Dana
S
SUBJ
NP
HD
v
know
CMP
S
NP
Kim
F1 F2 F3 M1 M2 M3 M4 E1 E2
F1 F2 F3 M1 M2 M3 M4 E1 E2
hates
Example (cont.)
Topology slot assignment
The Direct OBJect Kim may also go to F1 if the constituent is focused
HD
pro
we
SUBJ HD
v
DOBJ
NP
Dana
S
SUBJ
NP
HD
v
know
CMP
S
NP
Kim
F1 F2 F3 M1 M2 M3 M4 E1 E2
F1 F2 F3 M1 M2 M3 M4 E1 E2
hates
Example (cont.)
Sharing produces:Kim we know Dana hates
HD
pro
we
SUBJ HD
v
DOBJ
NP
Dana
S
SUBJ
NP
HD
v
know
CMP
S
NP
Kim
F1 F2 F3 M1 M2 M3 M4 E1 E2
F1 F2 F3 M1 M2 M3 M4 E1 E2
hates
ctype MainCl
tpl 1 F1, F3 o 2 , M1 o 3 , B2 o 4
1 F1,F3o 5 ,M1o 6 ,M3o 7
Example (cont.)
Topology slot assignment in terms of feature structuresBoth placement options of the focused direct object are
specified in the disjunctive alternatives of the TPL feature of the complement S node (gray rectangle).
HD
pro
we
SUBJ HD
v
DOBJ
NP
Dana
S
SUBJ
NP
HD
v
know
CMP
S
NP
Kim
hates
[tpl 2 ] [lemma 3 ] 1 F1o 7 ,F3o 5 ,M1o 6
[tpl 5 ]
[lemma 6 ]
foc +tpl 7
tpl 4
a. ... dass niemand verspricht zu versuchen das Fahrrad zu reparieren
… that nobody promises to try the
bike to repair
‘… that nobody promises to try to repair the bike’
Scrambling in German
E2M5M3
verspr.niemand
E1M6M4M2M1F1
zu vers.
d.Fahrr. zu rep.
dass
b. ... dass niemand das Fahrrad verspricht zu versuchen zu reparieren
Scrambling in German (cont.)
E2M5M3
verspr.niemand
E1M6M4M2M1F1
zu vers.
d.Fahrr. zu rep.
dass
c. ?... dass das Fahrrad niemand verspricht zu versuchen zu reparieren
Scrambling in German (cont.)
E2M5M3
verspr. niem.
E1M6M4M2M1F1
zu vers.
zu rep.
dass
d.Fahrr.