Phases, Strong Islands and Computational Nesting
Valentina Bianchi & Cristiano Chesi
University of Siena
The 28th GLOW Colloquium 2005 Genève, 31 March - 2 April, 2005
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
Outline
Data: Left branch islands and the connectedness effect
Kayne’s Connectedness Condition
The computational model (Chesi 2004)
Left-branch islands as computationally nested phases
Right-hand adjunct islands
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
(Kayne 1983; Pollard & Sag 1994, 182 ff.)
(1) a. * [Which famous playwright]i did [close friends of ei] become famous ?
b. ? [Which famous playwright]i did [close friends of ei] admire ei ?
(2) a. * Who did [my talking to ei] bother Hilary ? (Pollard & Sag 1994)
b. √ Who did [my talking to ei] bother ei ?
(3) a. * Whoi did you consider [friends of ei] angry at Sandy ?
b. √ Whoi did you consider [friends of ei] angry at ei ? (Pollard & Sag 1994)
■□□□□ Data: Left branch islands and the connectedness effect
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
• Left branch constituents are islands for extraction
• A legitimate gap on a right branch can “rescue” an illegitimate gap inside a left branch
X
eX
eXeX
■□□□□ Data: Left branch islands and the connectedness effect
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
(5) Y is a g-projection of X iff
i. Y is an ( X' ) projection of X or of a g-projection of X, or
ii. X is a structural governor and Y immediately dominates W and Z, where Z is a maximal projection of a g-projection of X, and W and Z are in a canonical government configuration:
(6) W and Z (Z a maximal projection, and W and Z immediately dominated by
some Y) are in a canonical government configuration iffa. V governs NP to its right in the grammar of the language and W precedes Zb. V governs NP to its left in the grammar of the language and Z precedes W
(7) The g-projection set G of a category is defined as follows (where governs ):
a. , = a g-projection of G b. G andb'. dominates and does not dominate G
□■□□□ Kayne’s Connectedness Condition(Kayne 1983)
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
(8) Connectedness Condition
Let 1 ... j, j+1 ... n be a maximal set of empty categories in a tree T such that
j, j is locally bound by . Then {} ( Gj) must constitute a subtree of T.
n
□■□□□ Kayne’s Connectedness Condition
nj1
1 - all the maximal projections in the path between the gap and its binder are on a right branch or
2 - a path terminating in a left branch is connected to a legitimate path of right branches
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
Which famousplaywright
become
did
close
friends
of e
famous
1
G1
(1) a. *
□■□□□ Kayne’s Connectedness Condition
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
Which famousplaywright
become
did1
close
friends
of
1
1
e
famous
1 1
(1) a. *
1
□■□□□ Kayne’s Connectedness Condition
G1
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
Which famousplaywright
admire
did1
close
friends
of
1
1
e
e
1 1
2
(1) b.
1
□■□□□ Kayne’s Connectedness Condition
G1
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
2Which famousplaywright
2
admire
2
did1
close
friends
of
1
1
e
e
2
1 1
G2
22
(1) b.
1
□■□□□ Kayne’s Connectedness Condition
2
G1
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
who
youbecause
(9) *a person who you admire e because [close friends of e] became famous
admire e
becamefamous
1
close
friends
1
G1 of
1
e
1
1 1
□■□□□ Kayne’s Connectedness Condition
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
who
2
youbecause
(9) *a person who you admire e because [close friends of e] became famous
admire
2
e
becamefamous
1
close
friends
1
G1 of
1
e
1
1 1
□■□□□ Kayne’s Connectedness Condition
2
G1
2
22
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
1 - Kayne’s Connectedness Condition does not subsume right hand islands
2 - Nature of the parasitic gap: is it like any ordinary gap (as in HPSG), or is it an empty resumptive pronoun (Cinque 1990, Postal 1994)?
Parasitic gaps has been claimed to differ from ordinary gaps w.r.t.
• restriction to the NP category
• incompatibility with antipronominal contexts
• lack of reconstruction effects
(see Culicover & Postal (2001) and Levine & Sag (2003), for various positions).
We remain neutral w.r.t. this question. For simplicity, we will assimilate the parasitic gap-antecedent dependency to the usual antecedent-gap dependency, and treat both in terms of copy-remerging.
□■□□□ Kayne’s Connectedness Condition
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
Generalization on legitimate recursion and gap licensing
Legitimate gaps lie on the main recursive branch of the tree, whereas illegitimate gaps lie on “secondary” branches, which do not allow for unlimited recursion (in that such a secondary branch cannot be the lowest one in a tree).
□■□□□ Kayne’s Connectedness Condition
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
Competence
Features Structures(semantic + syntactic/abstract + phonetic features → lexicon )
Structure Building Operations (merge, move, phase)
Performance tasks
Parsing
Generation
Flexibility
Universals
Parameterization
Economy conditions
□□■□□ The computational model(Chesi 2004)
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
Structure Building Operations (merge, move, phase)
□□■□□ The computational model
Structure Building Operations
Structure Building Operations
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
□□■□□ The computational model
the flexibility requirement implies a Top-to-bottom orientation of Structure Building Operations
theoretical arguments:
• Phillips’ (1996) temporary constituency;
• Teleological movement in a bottom-to-top perspective;
psycholinguistic evidence:
• incremental parsing: garden paths;
• incremental generation: false starts.
Structure Building Operations
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
MERGE
binary function (sensitive to temporal order) taking two features structures and unifying them.
PHASE PROJECTION
is the minimal set of dominance relations introduced in the SD based on the expectations triggered by each select feature of the currently processed lexical items
MOVE
top-down oriented function which stores an un-selected element in a memory buffer and re-merges it at the point of the computation where the element is selected
□□■□□ The computational model
Structure Building Operations
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
MOVELinearization Principle (inspired by Kayne’s LCA) if A immediately dominates B, then either a. <A, B> if A selects B as an argument, or
b. <B, A> if B is in a functional specification of A
e.g. “the boy kissed the girl”
PHASE
the boy
<the boy>kissed [=o kiss]
[=s =o kiss]
[+T kiss]
[=s =o kiss]
Memory Bufferthe boyMemory Buffer
□□■□□ The computational model
Vhead
V
V
V
V
V
V
V
Selected Phase(s)
(select features)
...(left
periphery)
...
F1
FnFunctionalSequence
(licensor features)
the girl
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
Sequential Phase Nested PhaseVs.
Fn
Slast
head
Memory Buffer
F1
S1
Memory Buffer
Memory Buffer
FnSlast
head
F1
S1
Memory Buffer
Success Condition: the memory buffer must be empty at the end of the phase orelse its content is inherited by the memory buffer of the next sequential phase (if any)
□□■□□ The computational model
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
To summarize:
1. Every computation is a top-down process divided into phases.
2. A phase gets closed when the last selected complement of its head is processed; this last projected complement constitutes the next sequential phase.
3. All unselected constituents are instead nested phases: they are processed while the superordinate phase has not been closed yet.
4. The Move operation stores an unselected element found before (i.e. on the left of) the head position in the local memory buffer of the current phase, and discharges it in a selected position if possible; if not, when the phase is closed the content of the memory buffer is inherited by that of the next sequential phase.
□□■□□ The computational model
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
□□□■□ Strong islands as computationally nested phases
(10) Whoi do you believe [twho that everybody admires twho]?
Who
believe
do
you
you = 2nd Nested Phase (DP)
Matrix Phase (CP)
Memory Buffer (Matrix Phase, CP)
who = 1st Nested Phase (DP)
who
you V
Sel.
Lic.
<you><who>
that
everybody admires
who
<who>
that = Selected Phase (CP)
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
Who
become
did
close friends of _
G1 = 2nd Nested Phase (DP)
V
Matrix Phase (CP)
Sel.
Lic.
Memory Buffer (Matrix Phase, CP)
who = 1st Nested Phase (DP)
whofamous
□□□■□ Strong islands as computationally nested phases
(1.a) *Whoi did [close friends of ei] become famous ?
e
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
Who
admire
did
close friends of e
G1 = 2nd Nested Phase (DP)
V
G = Matrix Phase (CP)
Sel.
Lic.
Memory Buffer (Matrix Phase, CP)
who = 1st Nested Phase (DP)
who
G1 <who><G1>who
□□□■□ Strong islands as computationally nested phases
(1.b) ?Whoi did [close friends of ei] admire ei ?
<who>
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
Who
become
did
close friends of e
G1 = 2nd Nested Phase (DP)
V
Matrix Phase (CP)
Sel.
Lic.
Memory Buffer (Matrix Phase, CP)
who = 1st Nested Phase (DP)
who
G1
<who>
<G1>who
<who>
□□□■□ Strong islands as computationally nested phases
famous
(1.a) *Whoi did [close friends of ei] become famous ?
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
Summary of the proposed analysis
We have recast the Connectedness Condition in derivational terms, by assuming:
(a) a top-to-bottom derivation divided in phases
(b) a “storage” conception of the Move operation
(c) a distinction between sequential and nested phases (corresponding to branches on the recursive vs. non-recursive side of the tree).
(d) The content of the memory buffer of a phase can only be inherited by the next sequential phase, and not by a nested phase.
(e) Parasitic gaps exploit the possibility of “parasitically” copying the content of the buffer of a matrix phase into the buffer of a nested phase.
(f) Parasitic copying, however, cannot empty the matrix memory buffer, whence the necessity of another (“legitimate”) gap within the matrix phase itself (or within a phase that is sequential to the matrix one).
□□□■□ Strong islands as computationally nested phases
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
□□□□■ Right hand adjunct islands as nested phases
(11) a. ??[Those boring old reports]i , Kim went to lunch [without reading ei].
b. √ [Those boring old reports]i , Kim filed ei [without reading ei].
(12) ?[A person]i that they spoke to ei [because they admire ei]Longobardi (1985) strenghtens the notion of g-projection, by adding
a proper government requirement: a non properly governed maximal projection is a boundary to the extension of g-projections.
By definition, subjects and adjuncts are not properly governed: thus, the adjunct island is assimilated to the subject island, much as in Huang’s (1982) Condition on Extraction Domains.
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
Those boring old reports
Kim
1
without
PRO
1
G
(11.a) ??[Those boring old reports]i , Kim went to lunch [without reading ei].
went
tolunch
reading
1
e
1
1 1
□□□□■ Right hand adjunct islands as nested phases
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
2
Those boring old reports
2
2
Kim
1
without
PRO
1e
G
(11.b) [Those boring old reports]i , Kim filed ei [without reading ei].
filed
2
2
reading
1
e
1
1 12
□□□□■ Right hand adjunct islands as nested phases
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
Problem 1:
Not all right-hand adjuncts are equally strong islands (cf. a.o. Pollard & Sag 1994, 191 and Haider 2003):
(13) a. Who did you go to Girona [in order to meet e]?
b.This is the blanket that Rebecca refuses to sleep [without e].
c. How many of the book reports did the teacher smile [after reading e]?
(Pollard & Sag 1994)
(14) a. the car that he left his coat [in e]
b. the day that she was born [on e]
c. * the day that she was born in England [on e] (Haider 2003, 3)
Then it is not obvious that “adjunct islands” should be assimilated to “left branch islands” (Pollard & Sag 1994); minimally we should distinguish “true adjuncts” from “oblique complements”
□□□□■ Right hand adjunct islands as nested phases
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
Problem 2:
Right-hand relative clauses are, prima facie, another instance of a non properly governed maximal projection (cf. Complex NP Island Constraint):
(15) ?* Which book did John meet [NP a child [CP who read t]]
But a subject complex NP allows for the extension of g-projections in a connectedness configuration:
(16)a. * A person who [people that talk to ei ] usually have money in mind
b. ? A person who [people that talk to ei ] usually end up fascinated with ei
□□□□■ Right hand adjunct islands as nested phases
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
A person
with
who
1
people
that
e
1
1
e
G 1
(16) b. ?
usually
end up
fascinated
to
1
e
1 1talk
1
2
□■□□□ Kayne’s Connectedness Condition
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
2A person
2
with
2
who
1
people
that
e
1
1
e
2
G 1
(16) b. ?
usually
2
end up
2
fascinated
2
to
1
e
1 1talk
2
2
□■□□□ Kayne’s Connectedness Condition
1
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
Longobardi (1985) must modify his definition of proper government so that the relative clause counts as properly governed; but then, the Complex NP Island Constraint must be stipulated as a separate constraint on extraction. (The Complex NP Island Constraint also did not follow from Kayne’s original Connectedness Condition, since it is a right branch: cf. Kayne 1984, n. 5.)
The variable strenght of adjunct islands and the unresolved status of the (relative clause) Complex NP Island in the connectedness approach cast some doubt on the idea that right-hand adjuncts must be completely assimilated to left branch islands, as in Longobardi’s approach (cf. Pollard & Sag 1994, Levine & Sag 2003 for a similar conclusion).
□□□□■ Right hand adjunct islands as nested phases
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
Memory Buffer
FnClast
head
F1
C1
Memory Buffer
□□□□■ Right hand adjunct islands as nested phases
Nested Phase
Memory Buffer
Fn
Clast
head
F1
C1
Memory Buffer
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
Memory Buffer
XClast
head
F1
C1
Memory Buffer
□□□□■ Right hand adjunct islands as nested phases
Nested Phase
Memory Buffer
Fn
Clast
head
F1
C1
Memory Buffer
[=x Fn ]
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
Memory Buffer
X
[=x head]
F1
C1
□□□□■ Right hand adjunct islands as nested phases
Nested Phase
Memory Buffer
Fn
Clast
head
F1
C1
Memory Buffer
Fn
C2
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
Who
admire
you
you = Nested Phase
V
G = Matrix Phase
Memory Buffer (Matrix Phase, CP)
who = Nested Phase
who
you <you> <who> because
close friends of _ became
famous
because = Nested PhaseG1 = Doubly-nested
Phase
1
(9) *... Who you admire e because [close friends of e] became famous
□□□□■ Right hand adjunct islands as nested phases
who
who
<who>
Bianchi, Chesi - Phases, Strong Islands, and Computational
Nesting
1. Novel account of recursive/transparent phases (at least in VO languages) depending on select features of the previous phase-head, in particular:
a. Left branch islands are computationally nested phases (selected phases are on the right of the head, cf. Linearization Principle)
b. Right hand adverbials too can be analyzed as computationally nested phases, depending on the structure of the relevant licensor feature
2. Extending the Top-to-Bottom orientation (Phillips 1996) to Move and to Phase Projection allows us to capture (a subset of) Strong Islands effects and the related connectedness effects in a derivational way
3. These results directly follow from a conception of the competence that includes Structure Building Operations fulfilling the flexibility requirement
Conclusions
Phases, Strong Islands and Computational Nesting
Valentina Bianchi
Cristiano Chesi
http://www.ciscl.unisi.it