Language Comprehension and Word-Order Variation - PowerPoint PPT Presentation

1 / 205
About This Presentation
Title:

Language Comprehension and Word-Order Variation

Description:

Title: PowerPoint Presentation Author: Colin Phillips Last modified by: Colin Phillips Created Date: 12/11/2002 6:45:01 PM Document presentation format – PowerPoint PPT presentation

Number of Views:825
Avg rating:3.0/5.0
Slides: 206
Provided by: ColinPh1
Learn more at: http://ling.umd.edu
Category:

less

Transcript and Presenter's Notes

Title: Language Comprehension and Word-Order Variation


1
Language Comprehension andWord-Order Variation
  • Colin PhillipsCognitive Neuroscience of Language
    LaboratoryDepartment of LinguisticsUniversity
    of Maryland

2
Overview of Talks
  • 1. The Unification Problem
  • 2. Building Syntactic Relations
  • 3. Abstraction Sounds to Symbols
  • 4. Linguistics and Learning

?????
3
Outline
  • Long-distance Dependencies The Issues
  • Active Structure Building
  • Pieces of Long-Distance Dependencies
  • Grammatical Accuracy
  • Timing and Electrophysiology
  • Conclusion

4
Long-distance Dependencies
  • Many different types of long-distance
    relationship
  • wh-movement, topicalization, relative clauses
  • pronoun-antecedent relations (referential)
  • ellipsis-antecedent relations
  • Require active computation, hard to reduce to
    retrieval of stored templates
  • what are the computational operations
  • what motivates them
  • when do they occur
  • are they the product of the grammar, or of
    something else

5
Outline
  • Long-distance Dependencies The Issues
  • Active Structure Building
  • Pieces of Long-Distance Dependencies
  • Grammatical Accuracy
  • Timing and Electrophysiology
  • Conclusion

6
Processing a fronted wh-phrase
How do readers interpret a fronted wh-phrase
online? What drives formation of
wh-dependencies?
what
did
you
say
that
Mary
t
read
7
Active Filler Effect
Processing English wh-questions
what
gap
8
Active Filler Effect
Processing English wh-questions
what
did
gap
9
Active Filler Effect
Processing English wh-questions
what
did
you
gap
10
Active Filler Effect
Processing English wh-questions
what
did
you
say
gap
11
Active Filler Effect
Processing English long-distance wh-questions
what
did
you
say
that
gap
12
Active Filler Effect
Processing English long-distance wh-questions
what
did
you
say
that
Mary
gap
13
Active Filler Effect
Processing English long-distance wh-questions
what
did
you
say
that
Mary
gap
read
14
Active Filler Effect
Generalization Gap for a wh-phrase is initially
posited in the first/highest available position.
what
did
you
say
gap
15
Filled Gap Effect
  • My brother wanted to know who Ruth will bring us
    home to at Christmas
  • My brother wanted to know if Ruth will bring us
    home to at Christmas

Readers slow down upon encountering an NP where a
gap was expected, relative to a control where no
gap was expected. Filled Gap Effect
Stowe 1986
16
Argument Structure
V NP
remind
V NP IP
(Boland et al. 1995)
17
Argument Structure
V NP
remind
V NP IP
  • Samuel asked whether Mark reminded them to watch
    the child.
  • Which child did Mark remind them to
    watch ___?
  • Which movie did Mark remind them to
    watch ___?

(Boland et al. 1995)
18
Argument Structure
V NP
remind
V NP IP
  • Samuel asked whether Mark reminded them to watch
    the child.
  • Which child did Mark remind them to
    watch ___?
  • Which movie did Mark remind them to
    watch ___?

(Boland et al. 1995)
19
Argument Structure
V NP
remind
V NP IP
  • Samuel asked whether Mark reminded them to watch
    the child.
  • Which child did Mark remind them to
    watch ___?
  • Which movie did Mark remind them to
    watch ___?

(Boland et al. 1995)
20
Argument Structure
  • Construction of gaps is guided by verb
    argument-structure information

21
Typing Mismatch Effects
  • Edson MiyamotoShoichi Takahashi

22
Question Formation
  • Japanese uses question particles (Q-particles) to
    mark questions.
  • John-nom the book-acc read.
  • John-nom the book-acc read-Q yes/no
    question
  • Sally-top John-nom what-acc read-declC said-Q
    root question
  • What did Sally say that John read?
  • Sally-top John-nom what-acc read-Q said embedded
    question
  • Sally said what John read.

23
Q-Particles
yonda-to (Declarative)
John-ga hon-o
yonda-ka (Q-Particle)
John-nom book-acc read
24
Q-Particles
yonda-to (Declarative)
Normally, a Q-particle is unexpected relative to
the high frequency declarative marker.
John-ga hon-o
yonda-ka (Q-Particle)
John-nom book-acc read
25
Q-Particles
yonda-to (Declarative)
Normally, a Q-particle is unexpected relative to
the high frequency declarative marker.
John-ga hon-o
yonda-ka (Q-Particle)
John-nom book-acc read
yonda-to (Declarative)
John-ga nani-o
yonda-ka (Q-Particle)
John-nom what-acc read
26
Q-Particles
yonda-to (Declarative)
Normally, a Q-particle is unexpected relative to
the high frequency declarative marker.
John-ga hon-o
yonda-ka (Q-Particle)
John-nom book-acc read
yonda-to (Declarative)
In a clause in which a wh-phrase is interpreted,
the expectations are reversed.
John-ga nani-o
yonda-ka (Q-Particle)
John-nom what-acc read
27
Q-Particles
  • Experimental diagnostic for clause where
    wh-phrase is interpreted in Japanese
    declarative marker -to is read more slowly than
    question marker -ka.
  • Miyamoto Takahashi 2001

28
Typing Mismatch Effect
  • a. NP-nom Wh-acc Verb-Q ... Verb.
  • b. NP-nom Wh-acc Verb-DeclC Verb-Q


  • Miyamoto
    Takahashi 2001

Slowdown TME
29
Parsing Japanese Wh-Questions
  • Sachiko Aoshima, Colin PhillipsAmy Weinberg

30
  • Japanese speakers appear to process questions in
    exactly the opposite way from English speakers

31
Two approaches for processing wh-phrases
  • Strategy-based Approach
  • When a wh-phrase has been identified, rank the
    option of assigning it to a gap above all other
    options.
  • (Crain Fodor 1985, Frazier Clifton 1989,
    among others)

32
Two approaches for processing wh-phrases
  • Strategy-based Approach
  • Active Filler Strategy
  • When a wh-phrase has been identified, rank the
    option of assigning it to a gap above all other
    options.
  • (Crain Fodor 1985, Frazier Clifton 1989,
    among others)
  • Grammatical principle-based Approach
  • Online interpretation of wh-phrases is driven by
    independently motivated grammatical requirements,
    e.g. thematic role assignment.
  • (Gibson 1991, Pritchett 1992, among others)

33
Two approaches for processing wh-phrases
head-initial languages
  • Strategy-based

Grammatical principle-based
CP
CP
WH
WH
IP
C
IP
C
NP

NP

VP
VP
V
gap
V
gap
the first possible gap position complement of
the first verb
the first possible gap position complement of
the first verb
34
Two approaches for processing wh-phrases
head-final languages
  • Strategy-based
  • Grammatical principle-based

CP
CP
WH
WH
IP
C
IP
C

NP

VP
NP
VP
CP
The first opportunity to satisfy thematic
requirements occurs at the embedded clause.
V
gap

CP
NP
V
VP
gap
V
the first possible gap position
35
Long-distance Wh-scrambling
  • Japanese wh-phrases are canonically in-situ, but
    they can be fronted by means of scrambling.
  • Dare-ni Taro-wa Jiro-ga t atta-ka
    itta.
  • Who-dat Taro-top Jiro-nom met-Q said
  • Taro said who Jiro met.

36
Experiment 1Stimuli
  • In-situ conditions
  • NP-top
  • NP-nom WH-dat NP-acc Verb-DeclC/QP
  • AdvP NP-dat Verb-QP/.

37
Experiment 1Stimuli
  • Scrambled conditions
  • NP-top
  • NP-nom WH-dat NP-acc Verb-DeclC/QP
  • AdvP NP-dat
    Verb-QP/.

WH-dat
38
Experiment 1 Conditions
  • a. ltSCRAMDECLCgt
  • Wh-dat NP-top NP-nom NP-acc V-DeclC AdvP NP-dat
    V-Q
  • b. ltINSITDECLCgt
  • NP-top NP-nom Wh-dat NP-acc V-DeclC AdvP NP-dat
    V-Q
  • c. ltSCRAMQgt
  • Wh-dat NP-top NP-nom NP-acc V-Q  AdvP NP-dat
    V   
  • d. ltINSITQgt
  • NP-top NP-nom Wh-dat NP-acc V-Q AdvP NP-dat V   

39
Experiment 1 Examples
  • a. ltSCRAMDECLCgt
  • ????? ??? ??? ?? ???????? ??? ???????
  • 'Which student did the class teacher tell the
    librarian at the library that the principal read
    a book for?'
  • b. ltINSITDECLCgt
  • ??? ??? ????? ?? ???? ???? ??? ???????
  • c. ltSCRAMQgt
  • ????? ??? ??? ?? ???? ???? ??? ??????
  • 'The class teacher told the librarian at the
    library which student the principal read a book
    for.'
  • d. ltINSITQgt
  • ??? ??? ????? ?? ???? ???? ??? ??????

40
Experiment 1Self-paced reading task
  • ----- --- --- ---- ---- --- ------ -------

41
Experiment 1Self-paced reading task
  • ????? --- --- ---- ---- --- ------ -------

42
Experiment 1Self-paced reading task
  • ----- ??? --- ---- ---- --- ------ -------

43
Experiment 1Self-paced reading task
  • ----- --- ??? ---- ---- --- ------ -------

44
Experiment 1Self-paced reading task
  • ----- --- --- ???? ---- --- ------ -------

45
Experiment 1Self-paced reading task
  • ----- --- --- ---- ???? --- ------ -------

46
Experiment 1Self-paced reading task
  • ----- --- --- ---- ---- ??? ------ -------

47
Experiment 1Self-paced reading task
  • ----- --- --- ---- ---- --- ??????? -------

48
Experiment 1Self-paced reading task
  • ----- --- --- ---- ---- --- ------ ????????

49
Experiment 1Design Procedure
  • 2 X 2 factorial design
  • 4 lists were created by distributing 24 items in
    a Latin Square design
  • 48 filler sentences
  • Comprehension questions matching a subject with
    a predicate
  • Self-paced reading task -Moving Window -
  • 48 native speakers of Japanese

50
Experiment 1 Results In-situ Condition
  • b. ltINSITDECLCgt
  • NP-top NP-nom Wh-dat NP-acc V-DeclC Verb-Q
  • d. ltINSITQgt
  • NP-top NP-nom Wh-dat NP-acc V-Q Verb

51
F1 (1, 47) 5.5, p lt.01 F2 (1, 18) 2.8, p
0.09
V-DeclC/Q
Miyamoto Takahashis observation is replicated.
52
Experiment 1 Results Scrambled Condition
  • a. ltSCRAMDECLCgt
  • Wh-dat NP-top NP-nom NP-acc V-DeclC Verb-Q
  • c. ltSCRAMQgt
  • Wh-dat NP-top NP-nom NP-acc V-Q  Verb.  

53
F1 (1, 47) 6.1, p lt.01 F2 (1, 18) 5.6, p lt.01
V-DeclC/Q
Readers also exhibit Typing Mismatch effect in
Scrambled Condition.
54
Experiment 1 Results Scrambled Condition
  • Readers create a gap position in the embedded
    clause.
  • Wh-phrase is associated with the first verb that
    readers encounter.
  • This finding is expected under the grammatical
    principle-based approach.

WH-dat
NP-top
CP
Verb
NP-nom
VP
gap
Verb
55
Experiment 2Question
Is this gap posited before the embedded verb is
encountered?

WH-dat
NP-top
CP
NP-nom
VP
gap
Verb
56
Experiment 2Question
Is this gap posited before the embedded verb is
encountered?

WH-dat
WH-dat
NP-top
NP-top
CP
CP
NP-nom
VP
NP-nom
VP
gap
gap
Verb
Verb
Or, is this gap posited after the embedded verb
is encountered?
57
Outline
  • Long-distance Dependencies The Issues
  • Active Structure Building
  • Pieces of Long-Distance Dependencies
  • Grammatical Accuracy
  • Timing and Electrophysiology
  • Conclusion

58
Grammatical Derivations
Wh-Questions
Englishmen cook wonderful dinners.
59
Grammatical Derivations
Wh-Questions
Englishmen cook wonderful dinners.
60
Grammatical Derivations
Wh-Questions
Englishmen cook what
61
Grammatical Derivations
Wh-Questions
Englishmen cook what
62
Grammatical Derivations
Wh-Questions
What do Englishmen cook
63
Grammatical Derivations
Wh-Questions
What do Englishmen cook gap/trace

64
Grammatical Derivations
Wh-Questions
?
What do Englishmen cook gap/trace
65
Competing Theories
Indirect AssociationTransformational
Grammar(--gt Projection Principle)
What do Englishmen cook gap/trace/copy
Direct AssociationHPSG/GPSGCategorial
GrammarDependency Grammaretc.
What do Englishmen cook
66
Competing Theories
Indirect AssociationTransformational
Grammar(--gt Projection Principle)
What do Englishmen cook gap/trace/copy
Direct AssociationHPSG/GPSGCategorial
GrammarDependency Grammaretc.
What do Englishmen cook
Attempts to distinguish between these
theoriesusing evidence from language processing
67
1. English Filled-Gap Effect
  • My brother wanted to know who Ruth will bring
  • us home to at Christmas
  • My brother wanted to know if Ruth will bring
  • us home to Mom at Christmas

(Stowe 1986)
68
1. English Filled-Gap Effect
  • My brother wanted to know who Ruth will bring
  • us home to at Christmas
  • My brother wanted to know if Ruth will bring
  • us home to Mom at Christmas

Surprise at pronoun following verb iscompatible
with both theories!
(Stowe 1986)
69
2. Trace Reactivation Studies
Which boy did the old man from Osaka meet at the
station?
(e.g. Nicol Swinney 1989, Bever McElree 1988,
MacDonald 1989)
70
2. Trace Reactivation Studies
Which boy did the old man from Osaka meet at the
station?
boy
faster decision
boy
same speed
girl
girl
(e.g. Nicol Swinney 1989, Bever McElree 1988,
MacDonald 1989)
71
2. Trace Reactivation Studies
Which boy did the old man from Osaka meet at the
station?
boy
faster decision
boy
same speed
girl
girl
Both theories can account for reactivation at or
after the verb!
(e.g. Nicol Swinney 1989, Bever McElree 1988,
MacDonald 1989)
72
3. Verb Position vs. Trace Position
give
NP
PP
(Pickering Barry 1991)
73
3. Verb Position vs. Trace Position
give
NP
PP
To which child did the teacher give a long
speech about the importance of honesty ___?
(Pickering Barry 1991)
74
3. Verb Position vs. Trace Position
give
NP
PP
To which child did the teacher give a long
speech about the importance of honesty ___?
Various diagnostics indicate that the
dependencyis formed at the verb, not at the
trace position.
(Pickering Barry 1991)
75
3. Verb Position vs. Trace Position
give
NP
PP
To which child did the teacher give a long
speech about the importance of honesty ___?
Various diagnostics indicate that the
dependencyis formed at the verb, not at the
trace position.
Still compatible with both theories!
(Pickering Barry 1991)
76
Effects at Verb Position


Direct Association
Gap-based Approach
CP
CP
WH
WH
IP
C
IP
C


NP
NP
VP
VP
gap
V
V
1
2
1
77
Pre-Verbal Gap Effects
start
  • The two theories could be distinguished by
    effects of dependency formation associated with
    argument positions that precede the verb of a
    clause.
  • Filled-gap effect expected at pre-verbal position
    only under indirect association/gap-based theory.

78
Parsing Japanese Wh-Questions(continued)
  • Sachiko AoshimaColin PhillipsAmy Weinberg

79
English Filled-Gap Effect
  • My brother wanted to know who Ruth will bring
  • us home to at Christmas
  • My brother wanted to know if Ruth will bring
  • us home to Mom at Christmas

Readers slow down upon encountering an NP where a
gap was expected, relative to a control where no
gap was expected.
Stowe 1986
80
Japanese Filled-Gap Effect
  • Slowdown upon encountering an NP where a gap was
    expected.

WH-dat
NP-top
  • Second NP-dat is unexpected if the first NP-dat
    has already been interpreted in embedded clause.

CP
NP-nom
VP
  • Position of the unexpected NP is before the verb

gap
NP-dat
Verb
81
Experiment 2Conditions

control
target
WH-dat
WH-nom
NP-top
NP-dat
CP
CP
NP-nom
VP
NP-nom
VP
gap
Verb
NP-dat
NP-dat
Verb
Slowdown
82
Experiment 2Conditions
  • a. Filled
  • WH-dat NP-top NP-nom Adv NP-dat NP-acc
  • Verb-DeclC Verb-Q
  • b. Non-Filled
  • WH-nom NP-dat NP-nom Adv NP-dat NP-acc
  • Verb-DeclC Verb-Q

83
Experiment 2Design Procedure
  • 2 conditions
  • 2 lists were created by distributing 20 paired
    items in a Latin Square design
  • 60 filler sentences
  • Comprehension questions matching a subject with
    a predicate
  • Self-paced reading task -Moving Window -
  • 34 native speakers of Japanese

84
Experiment 2Examples
  • a. Dono-syain-ni senmu-wa
    syacyoo-ga kaigi-de
  • Which employee-dat managing director-top
    president-nom meeting-at
  • kacyoo-ni syookyuu-o
    yakusokusita-to iimasita-ka?
  • assistant manager-dat raise-acc
    promised-DeclC told-Q
  • To which employee did the managing director
    tell that the president promised a raise to the
    assistant manager at the meeting?
  • b. Dono-syain-ga senmu-ni
    syacyoo-ga kaigi-de
  • Which employee-nom managing director-dat
    president-nom meeting-at
  • kacyoo-ni syookyuu-o
    yakusokusita-to iimasita-ka?
  • assistant manager-dat raise-acc
    promised-DeclC told-Q
  • Which employee told the managing director
    that the president promised a raise to the
    assistant manager at the meeting?

85
F1 (1, 33) 11.9, p lt.01 F2 (1, 19) 6.4, p
lt.05
NP-dat
Verb
Japanese readers exhibit Filled Gap effect.
Confirms that theyinterpret a sentence-initial
wh-phrase in the embedded clause,before reaching
the embedded verb.
86
From Experiment 1 2,
  • Readers create a gap position in the embedded
    clause.
  • Gap creation takes place before the verb is
    processed.
  • This finding is expected under the grammatical
    principle-based approach.

WH-dat
NP-top
CP
Verb
NP-nom
VP
gap
Verb
87
Summary

These findings are unexpected if objective is to
posit gap as early as possible.
who
? Strategy-based account does not properly
predict the time course of gap creation.
Taro
gap
CP
Verb
88
Summary
start
These findings are expected if objective is to
satisfy thematic role assignment as early as
possible.

who
Taro
? This supports the grammatical principle-based
approach
CP
Verb
Jiro
VP
gap
Verb
89
Summary Experiment 2 and 3

WH
HIS-WH
NP-top
NP-nom
CP
gap
Verb
Verb
NP-nom
VP
gap
Structures are built incrementally even when
critical lexical heads (i.e. verbs) are delayed.
Verb
90
Processing wh-interrogatives

WH
WH
NP-top
NP-nom
CP
gap
Verb
Verb
NP-nom
VP
The gap is posited in a simple clause.
gap
Verb
Successive attempts to posit a gap
The gap is posited in the mostly deep embedded
clause.
91
Processing wh-interrogatives
Dare-ni Taro-wa Jiro-ga atta-ka itta. Who-dat
Taro-top Jiro-nom met-Q said Taro said who Jiro
met.
Assumption Wh-phrases Two types of features need
to be satisfied as soon as possible. 1.
?-feature (Theta role assignment) 2.
Wh-feature (Wh-scope licensing)
92
Processing wh-interrogatives
Dare-ni Taro-wa Jiro-ga atta-ka itta. Who-dat
Taro-top Jiro-nom met-Q said Taro said who Jiro
met.
who
q wh
93
Processing wh-interrogatives
Dare-ni Taro-wa Jiro-ga atta-ka itta. Who-dat
Taro-top Jiro-nom met-Q said Taro said who Jiro
met.
who
q wh
Taro
gap
q wh
?
?
Verb
94
start
Processing wh-interrogatives
Dare-ni Taro-wa Jiro-ga atta-ka itta. Who-dat
Taro-top Jiro-nom met-Q said Taro said who Jiro
met.
who
Taro

gap
q wh
?
CP
?
Verb
Jiro
VP
gap
Reanalysis is allowed .
Verb
q wh
95
Processing Long-Distance Dependencies in Two
Varieties of Spanish
  • Leticia PablosColin Phillips

96
Two Varieties of Spanish
  • A estas chicas, mi hermana dijo que las conoce.
  • A estas chicas, mi hermana dijo que conoce.
  • to these girls my sister said that (them)
    knows
  • These girls, my sister said that she knows
    them.

97
Two Varieties of Spanish
  • A estas chicas, mi hermana dijo que las conoce.
  • A estas chicas, mi hermana dijo que conoce.
  • to these girls my sister said that (them)
    knows
  • These girls, my sister said that she knows them.

98
Two Varieties of Spanish
  • A estas chicas, mi hermana dijo que las conoce.
  • A estas chicas, mi hermana dijo que conoce.
  • to these girls my sister said that (them)
    knows
  • These girls, my sister said that she knows them.

?
Spain
Latin Am.
?
Spain
?
?
Latin Am.
99
Two Varieties of Spanish
  • Castilian subjects have a more restricted dialect
    where, if a subject intervenes, the prediction of
    a clitic pronoun after encountering the fronted
    phrase is obligatory.
  • A estas chicas, mi hermana dijo que las conoce.

Clitic predicted
100
Two Varieties of Spanish
  • Latin subjects have the two options available and
    even if there is an intervening subject, they
    dont necessarily predict the clitic.
  • A estas chicas, mi hermana dijo que las conoce.

A estas chicas, mi hermana dijo que conoce.
Clitic not necessarily predicted
101
Topic-Clitic Dependencies
? Clitic mismatch
  • A estas chicas, mi hermana mayor más tarde
    dijo que ya lo cree que las conoce desde hace
    tiempo
  • These girls, my old sister later said that she
    indeed thinks she knows them for a long time
  • A estas chicas, mi hermana mayor más tarde les
    dijo que ya lo cree que las conoce desde hace
    tiempo
  • These girls, my old sister later told them that
    she indeed thinks she knows them for a long time

?
102
Pre-verbal Gap Effects in Spanish

Direct Association
Indirect Association
CP
CP
WH
WH
IP
C
IP
C


NP
NP
VP
VP
V
V
clitic
V
clitic
V
103
Pre-verbal Gap Effects in Spanish Topic-Clitic
Dependencies
  • Design
  • Manipulation of the clitic pronoun at the matrix
    verb to control for the expected effect at the
    clitic in the embedded verb.
  • Predictions
  • There should be a longer reading time at the
    first embedded clitic if theres a mismatch.
  • The interpretation of the long-distance
    dependency should be mediated by the clitic.

104
Pre-verbal Gap Effects in Spanish Design and
Procedure
  • 2 conditions
  • 2 lists were created by distributing 24 paired
    items in a Latin Square design
  • 72 filler sentences
  • Context sentence preceded its item(identical in
    both conditions)
  • Comprehension questions
  • Self-paced reading task-Moving Window-
  • 24 Spanish subjects (16 Latin 8 Castilian)

105
Pre-verbal Gap Effects in SpanishResults
Castilian Subjects
First embedded verb
Clitic
Longer RT at no-clitic condition (spillover
effect from region 12)
F1(1,7)6.36,plt.05 F2(1,23)1.51,p.2
106
Pre-verbal Gap Effects in SpanishResults Latin
Subjects
Flt 1
First embedded verb
Clitic
No effect
107
Interim Conclusion
  • In both Japanese and Spanish long-distance
    dependencies formed by constructing argument
    positions before the verb.
  • Supports
  • (i) very active structure-building (-gt
    analysis-by-synthesis)
  • (ii) gap-based approaches

108
Outline
  • Long-distance Dependencies The Issues
  • Active Structure Building
  • Pieces of Long-Distance Dependencies
  • Grammatical Accuracy
  • Timing and Electrophysiology
  • Conclusion

109
Island Constraints and a Look-Ahead Problem
  • Colin PhillipsBeth RabbinKaia WongLeticia
    Pablos

110
  • If there is no distinction between real-time
    processors and the grammar, then the real-time
    processors should show immediate grammatical
    precision delay not possible
  • Good testing ground constraints on wh-movement
  • described in great detail by linguistic theories
  • real-time processing literature presents
    apparently conflicting results

111
Grammatical Accuracy in Parsing
Subject
?
What did the plan to remove gap
ultimately destroy the building
Island Constraint A wh-phrase cannot be moved out
of a subject.
112
Island Constraints in Real-Time
  • A number of studies report that island
    constraints are respected in real-time
  • Stowe (1986) subject islands (PP)
  • Traxler Pickering (1996) subject islands
    (finite RC)
  • McElree Griffith (1998)
  • A number of studies suggest that island
    constraints are violated in real-time
  • Freedman Forster (1985) complex NP islands
  • Kluender Kutas (1993) wh-islands
  • Previous discussions attribute differing results
    to technique used - alternative is that the
    structures are critical

113
Island Constraints in Real-Time
  • It is an over-simplification to say that the
    grammar excludes creation of dependencies into
    islands

114
Parasitic Gaps
which school did the proposal to expand the
school ultimately overburdened the teachers.
(Phillips, Rabbin, et al. 2002)
115
Parasitic Gaps
which school did the proposal to expand the
school ultimately overburdened the teachers.
(Phillips, Rabbin, et al. 2002)
116
Parasitic Gaps
which people did the proposal to expand the
school ultimately overburdened the teachers.
(Phillips, Rabbin, et al. 2002)
117
Parasitic Gaps
which school did the proposal to expand the
school ultimately overburdened the teachers.
(Phillips, Rabbin, et al. 2002)
118
Parasitic Gaps
which school did the proposal to expand the
school ultimately overburdened the teachers.
Generalization (Subject Island Constraint) No
long-distance dependencies across subject
boundaries
(Phillips, Rabbin, et al. 2002)
119
Parasitic Gaps
which school did the proposal to expand the
school ultimately overburdened the teachers.
Generalization (informal) Violations can be
rescued by subsequent well-formed gaps.
(Phillips, Rabbin, et al. 2002)
120
Parasitic Gaps
which school did the proposal to expand the
school ultimately overburdened the teachers.
which school did the proposal that expanded the
school ultimately overburdened the teachers.
Updated Generalization (informal) A subclass of
violations can be rescued by subsequent gaps.
(Phillips, Rabbin, et al. 2002)
121
Parasitic Gaps
Acceptability Ratings (n32)
(Phillips, Rabbin, et al. 2002)
122
Parasitic Gaps
which school did the proposal to expand the
school ultimately overburdened the teachers.
which school did the proposal that expanded the
school ultimately overburdened the teachers.
The challenge for incremental parsing Which
verbs can the wh-phrase be interpreted with?
Lookahead problem.
(Phillips, Rabbin, et al. 2002)
123
Parasitic Gaps
which school did the proposal to expand the
school ultimately overburdened the teachers.
which school did the proposal that expanded the
school ultimately overburdened the teachers.
  • The challenge for incremental parsing
  • Which verbs can the wh-phrase be interpreted
    with? Lookahead problem.
  • avoid islands conservative, but non-incremental
  • infinitival subject islands only incremental,
    but risky
  • all subject islands reckless

(Phillips, Rabbin, et al. 2002)
124
Parasitic Gaps
which school did the proposal to expand the
school ultimately overburdened the teachers.
which students
which school did the proposal that expanded the
school ultimately overburdened the teachers.
which students
Technique plausibility manipulation (Garnsey
1989, Boland et al. 1991, Traxler Pickering
1996)
(Phillips, Rabbin, et al. 2002)
125
Parasitic Gaps
plausible at expandplausible at overburden
which school did the proposal to expand the
school ultimately overburdened the teachers.
which students
implausible at expandplausible at overburden
which school did the proposal that expanded the
school ultimately overburdened the teachers.
which students
Technique plausibility manipulation (Garnsey
1989, Boland et al. 1991, Traxler Pickering
1996)
(Phillips, Rabbin, et al. 2002)
126
Materials 4 conditions (a) The chief executive
knew which building the plan to expand upon the
architects design would improve. (infinitival,
plausible) ? PG possible (b) The chief executive
knew which carpenters the plan to expand upon the
architects design would require. (infinitival,
implausible) ? PG possible (c)The chief executive
knew which building the plan that expanded upon
the architects design would require. (finite,
plausible) ? PG impossible (d) The chief
executive knew which carpenters the plan that
expanded upon the architects design would
require. (finite, implausible) ? PG impossible
(Phillips, Rabbin, et al. 2002)
127
embedded verb
128
(No Transcript)
129
Parasitic Gaps
  • Results
  • plausibility effect inside infinitival subject
    island, i.e., when the island violation can be
    rescued
  • no plausibility effect inside finite subject
    island
  • indicates extreme grammatical precision of
    on-line interpretive processes
  • A problem for attempts to treat island
    constraints as epiphenomena of parsing, memory,
    etc. (many different incarnations in both formal
    and functional literature)

(Phillips, Rabbin, et al. 2002)
130
Outline
  • Long-distance Dependencies The Issues
  • Active Structure Building
  • Pieces of Long-Distance Dependencies
  • Grammatical Accuracy
  • Timing and Electrophysiology
  • Conclusion

elan
wh
131
Electrophysiology
  • Can use electrophysiological techniques to
    investigate different aspects of syntactic
    dependency formation
  • incomplete dependencies (e.g. wh )
  • completion of dependencies (e.g. wh verb)

132
Event-Related Potentials (ERPs)
  • John is laughing.

s1 s2 s3
133
ERP Sentence Processing
N400
  • Developing understanding of N400 is informative
  • Response to violations

I drink my coffee with cream and sugar I drink my
coffee with cream and socks
Kutas Hillyard (1980)
134
Morpho-Syntactic violations
  • Every Monday he mows the lawn.
  • Every Monday he mow the lawn.
  • The plane brought us to paradise.
  • The plane brought we to paradise.
  • (Coulson et al., 1998)

(Slide from Kaan (2001)
135
Left Anterior Negativity (LAN)

P600
he mows he mow
(Slide from Kaan (2001)
136
ERP Sentence Processing
LAN, P600
Sie bereist dasneuter Landneuter Sie
bereist denmasculine Landneuter she travels the
land ...
Gunter et al. (2000)
137
ERP Sentence Processing
P600
Emily wondered who the performer in the concert
had imitated for the audiences amusement.Emily
wondered whether the performer in the concert had
imitated a pop star for the audiences amusement.
  • P600 reflects normal structure-building processes.

Kaan et al. (2000)
138
How fast is Structural Computation?
  • Silke UrbanColin Phillips

139
Background
  • Early Left Anterior Negativity
  • (Angela Friederici, Anja Hahne, et al.)

140
Neville et al., 1991
  • The scientist criticized a proof of the theorem.
  • The scientist criticized Maxs of proof the
    theorem.

141
500ms/word
142
500ms/word
143
Hahne Friederici, 1999
  • ? Das Baby wurde gefüttert
  • The baby was fed
  • ? Das Baby wurde im gefüttert
  • The baby was in-the fedQuestion are the
    brain responses to violations automatic?

144
Hahne Friederici, 1999
P600
145
P600
146
Hahne Friederici, 1999
ELAN
147
ELAN
148
ELAN
  • Very fast 150-250ms
  • Automatic
  • Left anterior ( frontal) scalp distribution
  • Elicited by a subclass of syntactic
    violationsPhase 1 (100300 ms) represents the
    time window in which the initial syntactic
    structure is formed on the basis of information
    about the word category. (Friederici
    2002)

149
Questions about ELAN
  • How plausible is it that ELAN reflects syntactic
    structure building?
  • Speed 150ms is faster than lexical access!
  • Generality ELAN is not elicited by most
    violations almost all studies on ELAN involve
    one construction (for each of German, English)
  • Localization

150
Brodmann Areas
151
Magnetoencephalography (MEG)
pickup coil SQUIDassembly
160 SQUIDwhole-headarray
152
(Friederici et al. 2000)
153
Two regions of interest
(Friederici et al. 2000)
154
(Friederici et al. 2000)
155
Anterior Temporal Lobe?
  • Why is anterior temporal lobe so important in
    ELAN?
  • How does it differ from Brocas area (BA 44 etc.)
    that are implicated so often in other studies?
  • Friederici both responsible for structure
    building BA44 also responsible for syntactic
    working memory the inferior portion of BA44 is
    selectively activated when syntactic processes
    are in focus.
  • Anterior temporal lobe associated with
  • lexical information
  • activated in fMRI by comparisons of sentences
    with word lists

156
Alternative Interpretation
  • ELAN reflects violation/suppression of automatic
    lexical prediction
  • accounts for localization to anterior temporal
    lobe
  • accounts for very early timing
  • might account for automaticity
  • accounts for very limited distribution

157
Neville et al., 1991
  • The scientist criticized a proof of the theorem.
  • The scientist criticized Maxs of proof the
    theorem.

NP
Maxs
N
158
Hahne Friederici, 1999
  • ? Das Baby wurde gefüttert
  • The baby was fed
  • ? Das Baby wurde im gefüttert
  • The baby was in-the fed

PP
in
NP
the
N
159
Prediction
  • If ELAN reflects violation of lexical prediction,
    rather than syntactic structure-building, then
  • change lexical predictions
  • keep syntactic violation the same
  • should turn off ELAN brain response

160
Neville et al., 1991
  • The scientist criticized a proof of the theorem.
  • The scientist criticized Maxs of proof the
    theorem.

NP
Possible to block the automatic prediction of an
N following a possessor ellipsis
Maxs
N
161
Ellipsis
  • Possessors may appear alone in ellipsis
    contextsAlthough I like Marys theory, I dont
    like Johns.

162
Experimental Conditions
  • Although Erica kissed Marys mother, she did not
    kiss the daughter of the bride.
  • Although Erica kissed Marys mother, she did not
    kiss Danas of the bride.
  • Although the bridesmaid kissed Mary, she did not
    kiss the daughter of the bride.
  • Although the bridesmaid kissed Mary, she did not
    kiss Danas of the bride.

163
Experimental Conditions
  • Although Erica kissed Marys mother, she did not
    kiss the daughter of the bride.
  • Although Erica kissed Marys mother, she did not
    kiss Danas of the bride.
  • Although the bridesmaid kissed Mary, she did not
    kiss the daughter of the bride.
  • Although the bridesmaid kissed Mary, she did not
    kiss Danas of the bride.

ellipsis possible
?
ellipsis impossible
164
Experimental Design
  • 384 sentences per session
  • 128 targets (drawn from 128 sets of 4 conditions)
  • 64 items designed to elicit agreement violation
    LAN
  • 192 filler items, designed to hide violations and
    promote ellipsis
  • Procedure
  • RSVP (Rapid Serial Visual Presentation),
    500ms/word
  • Grammaticality judgment task
  • Recording 32-electrode montage
  • 22 subjects (so far)

165

166
Although
167
Erica
168
kissed
169
Marys
170
mother,
171
she
172
did
173
not
174
kiss
175
Danas
176
of
177
the
178
bride.
179
???
good
bad
180
Preliminary Results
b
a
a. Although Marys mother Danas of b.
Although Mary Danas of
181
Interim Conclusion
start
  • Preliminary results lend support to our
    interpretation of the (E)LAN - the anterior
    negativity is reduced in an ellipsis context
  • structural violation is identical in both
    conditions
  • obligatory lexical prediction of N following
    possessor (e.g. Marys) is absent in ellipsis
    context
  • Structure-building may begin 250-300ms after a
    word is presented (see next study)

182
Electrophysiology of Wh-movement
start
  • Colin Phillips, Nina Kazanina, Shani Abada,
    Daniel Garcia-Pedrosa
  • revision of work done at UDel in 2000

183
Electrophysiology of Wh-movement
  • Objective is to investigate electrophysiological
    indices of different sub-processes involved in
    forming long-distance dependencies
  • Manipulation of resource demands of long-distance
    dependency formation

184
Experiment Design
Materials
a. The actress wished that the producers knew
that the witty host would tell the jokes during
the party. b. The actress wished that the
producers knew which jokes the witty host would
tell __ during the party. c. The producers knew
that the actress wished that the witty host would
tell the jokes during the party. d. The producers
knew which jokes the actress wished that the
witty host would tell __ during the party.
185
Experiment Design
Materials
a. The actress wished that the producers knew
that the witty host would tell the jokes during
the party. b. The actress wished that the
producers knew which jokes the witty host would
tell during the party. c. The
producers knew that the actress wished that the
witty host would tell the jokes during the
party. d. The producers knew which jokes the
actress wished that the witty host would tell
during the party.
Short wh-dependency
186
Experiment Design
Materials
a. The actress wished that the producers knew
that the witty host would tell the jokes during
the party. b. The actress wished that the
producers knew which jokes the witty host would
tell during the party. c. The
producers knew that the actress wished that the
witty host would tell the jokes during the
party. d. The producers knew which jokes the
actress wished that the witty host would tell
during the party.
Long wh-dependency
187
Experiment Design
Materials
a. The actress wished that the producers knew
that the witty host would tell the jokes during
the party. b. The actress wished that the
producers knew which jokes the witty host would
tell during the party. c. The
producers knew that the actress wished that the
witty host would tell the jokes during the
party. d. The producers knew which jokes the
actress wished that the witty host would tell
during the party.
Controls No wh-dependency
188
Experiment Design
Materials
  • Design as close as possible to regular sentence
    processing studies just bigger
  • 160 sets of 4 conditions Latin Square design
  • 320 fillers (21 ratio) --gt total of 480
    sentences per subject
  • Yes/No comprehension questions after all trials
  • 2 x 2 1/2 hour sessions (2 x 75 mins. reading)
  • RSVP presentation 300ms on, 200ms off
  • 32-electrode montage DC-recordings

189
Electrode PZ
Short Conditions
n20
-wh
Verb
wh
Effect of wh-movement significant (plt.01) from
300-400ms onwards
190
Electrode PZ
Long Conditions
n20
-wh
Verb
wh
Effect of wh-movement significant (plt.01) from
300-400ms onwards
191
Electrode PZ
All Conditions
n20
-wh
Verb
wh
Effect of wh-movement significant (plt.01) from
300-400ms onwards
192
ERPs Long-Distance Dependencies
  • P600 due to completion of wh-dependency, with
    onset around 300ms
  • Positivity associated with wh-dependencies
    previously observed by Kaan et al. (2000),
    Fiebach et al. (2002)
  • Identical amplitude timing for shorter and
    longer wh-dependency
  • Conclusion
  • The ERP response indexes resource-independent
    aspects of building wh-dependencies,
    OR
  • 1 vs. 2 clause dependencies incur identical
    resource demands

193
Subprocesses in the Formation of Long-Distance
Dependencies
  • Ana GouveaDavid Poeppel
  • Colin PhillipsNina Kazanina

194
  • which student case, thematic role ambiguous
  • predicate unknown
  • to which student case, thematic role clear
  • predicate unknown

195
  • (1) The patient met the doctor to whom the nurse
    with the white dress showed the problem in the
    meeting.
  • (2) The patient met the doctor to whom the nurse
    with the white dress show the problem in the
    meeting.
  • (3) The patient met the doctor while the nurse
    with the white dress showed the problem in the
    meeting.
  • (4) The patient met the doctor while the nurse
    with the white dress show the problem in the
    meeting.
  • (5) The patient met the doctor and the nurse with
    the white dress showed the problem in the
    meeting.

196
Wh-dependency formation (dative wh-phrase)
anterior positivity (300-500ms)
Agreement violation posterior positivity (600ms
onwards)
197
Possible Implication
  • There may be a difference between the following
    aspects of wh-dependency completion
  • specification of predicate
  • resolution of ambiguous case/thematic information

198
Outline
  • Long-distance Dependencies The Issues
  • Active Structure Building
  • Pieces of Long-Distance Dependencies
  • Grammatical Accuracy
  • Timing and Electrophysiology
  • Conclusion

199
Conclusion
  • Active construction of long-distance
    dependencies driven by satisfaction of
    grammatical requirements
  • Dependencies completed before verb in some
    languages - indirect association
  • Real-time structure building is highly accurate,
    sensitive to detailed constraints on islands,
    parasitic gaps
  • Electrophysiology allows investigation of timing
    of different aspects of dependency formation
  • We can learn a lot about real-time computatation
    - what kind of computational system is this?

200
Standard View
specialized algorithm
specialized algorithm
understanding
speaking
grammaticalknowledge,competence
language
201
with help from ...
  • University of MarylandShani AbadaSachiko
    AoshimaDaniel Garcia-PedrosaAna GouveaNina
    KazaninaMoti LiebermanLeticia PablosDavid
    PoeppelBeth RabbinSilke UrbanCarol Whitney
  • University of DelawareEvniki EdgarBowen Hui
  • Baris KabakTom PellathyDave SchneiderKaia
    WongAlec Marantz, MITElron Yellin, MIT

National Science FoundationJames S. McDonnell
FoundationHuman Frontiers Science ProgramJapan
Science Technology ProgramKanazawa Institute
of Technology
202
(No Transcript)
203
Experiment 1 Examples
  • a. ltSCRAMDECLCgt
  • Dono kodomo-ni tannin-wa koocyoo-ga hon-o
    yonda-to tosyokan-de sisyo-ni iimasita-ka.
  • 'Which student did the class teacher tell the
    librarian at the library that the principal read
    a book for?'
  • b. ltINSITDECLCgt
  • Tannin-wa koocyoo-ga dono kodomo-ni hon-o
    yonda-to tosyokan-de sisyo-ni iimasita-ka.
  • c. ltSCRAMQgt
  • Dono kodomo-ni tannin-wa koocyoo-ga hon-o
    yonda-ka  tosyokan-de sisyo-ni iimasita.
  • 'The class teacher told the librarian at the
    library which student the principal read a book
    for.'
  • d. ltINSITQgt
  • Tannin-wakoocyoo-ga dono kodomo-ni hon-o
    yonda-ka tosyokan-de sisyo-ni iimasita-ka.

204
Experime5t on Pre-verbal Gap Effects in Spanish
(3) Topic-clitic Dependencies
Target Condition Slowdown expected (due
to mismatch) if the parser is trying to complete
the dependency before the first embedded verb
Clitic Verb
205
Experiment on Pre-verbal Gap Effects in Spanish
(5) Topic-clitic Dependencies
Control Condition Clitic les licenses to
close the dependency at the matrix verb.
Write a Comment
User Comments (0)
About PowerShow.com