Grammatical Illusions: Encoding and navigating linguistic ...

9/15/10

1

Grammatical Illusions: Encoding and navigating linguistic

structures in real time

Parts #1 & #2 Background and

Unbounded Dependencies

Colin Phillips Dept of Linguistics

Cogn. Neurosci. of Language Lab University of Maryland

From Cells to Syntax

Perceptual Illusions

Illusory Comparative

“More people have been to Russia than I have.”

(Montalbetti 1984, Townsend & Bever 2001, Wellwood et al. 2009)

Outline of Talks 1.  Introduction

o  grammars o  illusions & non-illusions o  memory access

2.  Unbounded dependencies (wh-mvt, relativization, etc.) o  ‘active’ processes o  on-line constraint application o  standard experimental approaches

3.  Anaphora o  prospective vs. retrospective processes

4.  Agreement 5.  Semantic constraints

o  Negative Polarity o  Comparatives o  Thematic role binding

9/15/10

2

http://www.ling.umd.edu/colin

[email protected]

1. Some general questions

A.  What is knowledge of grammar? B.  Motivation for testing on-line constraint application

C.  Selective Fallibility/Grammatical Illusions D.  Memory and access mechanisms

A. What is grammatical knowledge?

Chomsky (1965) •  “Linguistic theory is concerned primarily with an ideal speaker-listener

[…] who knows its language perfectly, and is unaffected by such grammatically irrelevant conditions as memory limitations, distractions, shifts of attention, and interest, and errors (random or characteristic) in applying his knowledge of language in actual performance. […]

We thus make a fundamental distinction between competence (the speaker-hearer’s knowledge of his language) and performance (the actual use of language in concrete situations). Only under the idealization set forth in the preceding paragraph is performance a direct reflection of competence.” (pp. 3-4)

•  “When we say that a sentence has a certain derivation with respect to a particular generative grammar, we say nothing about how the speaker or hearer might proceed, in some practical or efficient way, to construct such a derivation. These questions belong to the theory of language use - the theory of performance.” (p. 9)

•  Says what knowledge is not … but little about what gr. knowledge is

Standard View

324 697+ ?

217 x 32 = ?

arithmetic

Standard View

324 697+ ?

217 x 32 = ?

specialized algorithm specialized algorithm

‘number��sense’?

something deeper

arithmetic

9/15/10

3

Standard View

specialized algorithm specialized algorithm

recursive characterization of��well-formed expressions

speaking understanding

grammatical��knowledge,��competence language

What is a (mental) grammar? •  Mentalistic: grammars describe ‘knowledge of language’

They are task-independent

•  But how to interpret the mental status of specific grammatical mechanisms?

•  A. Literalist view: mechanisms/derivations are real, one among multiple systems used by speakers (Townsend & Bever, 2001)

•  B. Formalist view (of derivational theories): all representations in grammatical derivation are constructed at some point; derivational order is a formal relation, not a timing relation

•  C. Extensionalist view: grammatical descriptions simply meet the goal of predicting (un)acceptable sentences

•  My view: grammars are real-time computational systems (Phillips & Lewis, in press)

Implementation (In)dependence •  An abstract system is implementation independent if the same

system can be realized in different ways by multiple lower-level implementations

•  Arithmetical operations, e.g., 48 ÷ 8 = 6

Can be implemented in digital computer, abacus, human brain, etc.

•  An abstraction is implementation dependent if it is only ever realized in one way at a less abstract level

•  Applied to generative grammars: are there multiple ways of generating the same representation in different tasks?

(Phillips & Lewis, in press)

Implementation (In)dependence •  Generative grammars typically assumed to be implementation

independent – but there is very little evidence for this (… and very little investigation of the issue, to be honest)

•  Sentence representations consistently constructed in same order, regardless of task

(one cannot construct sentences ‘backwards’)

•  When sentence comprehension fails, repair appears to simply reprocess the same sentence in the original order

(Phillips & Lewis, in press)

B. Why study on-line constraint application?

Real-time status of grammatical constraints

•  In theoretical linguistics, the existence of richly articulated grammatical constraints are used to justify richly articulated representations

•  The same reasoning applies when looking at what speakers do in real time: on-line application of grammatical constraints implicates: (i) existence of suitably rich representations (ii) ability to use those representations successfully and quickly

•  … much harder to get facts about on-line constraint application

•  … and some people think it’s not terribly relevant

9/15/10

4

“It has sometimes been argued that linguistic theory must meet the empirical condition that it account for the ease and rapidity of parsing. But parsing does not, in fact, have these properties. […] In general, it is not the case that language is readily usable or ‘designed for use.’” (Chomsky & Lasnik, 1993, p. 18)

“…the language comprehension system creates representations that are ‘good enough’ (GE) given the task that the comprehender needs to perform. GE representations contrast with ones that are detailed, complete, and accurate with respect to the input.” (Ferreira & Patson, 2007, p. 71) “we understand everything twice” (Townsend & Bever, 2001)

Noam Chomsky, MIT

Howard Lasnik, Maryland

Fernanda Ferreira, S. Carolina

Tom Bever, Arizona

Two Views •  Real-time processes only indirectly related to grammatical knowledge

predicts on-line insensitivity to many grammatical constraints

(Psycholinguistics: Townsend & Bever 2001; Ferreira & Patson 2007; Lewis et al. 2006; Theoretical linguistics: standard position since Chomsky 1965)

•  Real-time processes are all there is

predicts on-line grammatical “infallibility”

(Grammatical models: Kempen & Harbusch 2002; Phillips 2004; Cann et al. 2005 Psycholinguistics: implicit in most ambiguity resolution models since 1970s)

C. Selective Fallibility & Grammatical Illusions

A quick tour

On-line Interpretation of Backwards Anaphora

Kazanina, Lau, Lieberman, Yoshida, & Phillips (2007, J. Mem. Lang.)

Ellen Lau Harvard Med Sch./Maryland

Nina Kazanina U. of Bristol, Psych.

Pronoun Interpretation •  A pronoun may precede its antecedent…

–  While hei was reading the book, Johni ate an apple.

•  But not always…

–  *Hei ate the apple while Johni was reading the book.

•  Binding Condition C

–  A pronoun may not c-command its antecedent

(i.e., the antecedent can’t occur in the scope of the pronoun)

A Constraint on Interpretation

S

NP VP

V NP

John

ate the apple

S’

S

while S

NP VP

Comp

he

was reading the book

While he was reading the book, John ate the apple

9/15/10

5

A Constraint on Interpretation S

NP VP

V NP

he

ate the apple

S’ VP

while S

NP VP

Comp

John

was reading the book

He ate the apple while John was reading the book

Pronoun Interpretation •  A pronoun may precede its antecedent…

–  While hei was reading the book, Johni ate an apple.

•  But not always…

–  *Hei ate the apple while Johni was reading the book.

•  Binding Condition C

–  A pronoun may not c-command its antecedent

(i.e., the antecedent can’t occur in the scope of the pronoun)

Can antecedent search ignore NPs in structurally inappropriate positions?

----- -- --- ------- --- ----- ---- --- -- ------!

Self Paced Reading

While -- --- ------- --- ----- ---- --- -- ------!

Self Paced Reading

----- he --- ------- --- ----- ---- --- -- ------!

Self Paced Reading

----- -- was ------- --- ----- ---- --- -- ------!

Self Paced Reading

9/15/10

6

----- -- --- reading --- ----- ---- --- -- ------!

Self Paced Reading

----- -- --- ------- the ----- ---- --- -- ------!

Self Paced Reading

----- -- --- ------- --- book, ---- --- -- ------!

Self Paced Reading

----- -- --- ------- --- ----- John --- -- ------!

Self Paced Reading

----- -- --- ------- --- ----- ---- ate -- ------!

Self Paced Reading

----- -- --- ------- --- ----- ---- --- an ------!

Self Paced Reading

9/15/10

7

----- -- --- ------- --- ----- ---- --- -- apple.!

Self Paced Reading Immediate Constraint Application

While she was taking classes full-time, Jessica was working two jobs to pay the bills. While she was taking classes full-time, Russell was working two jobs to pay the bills.

While she … Jessica …��

Russell …

Self-Paced Reading, Gender Mismatch Paradigm

(Kazanina et al., 2007)

Immediate Constraint Application

While she was taking classes full-time, Jessica was working two jobs to pay the bills. While she was taking classes full-time, Russell was working two jobs to pay the bills.

She was taking classes full-time while Jessica was working two jobs to pay the bills. She was taking classes full-time while Russell was working two jobs to pay the bills.

While she …

She …

Jessica …��

Russell …

while Jessica …��

while Russell …

Self-Paced Reading, Gender Mismatch Paradigm


-60

-40

-20

0

20

40

60

80

100

120

because lastsemester

while-cd SHE wastaking

classes while-ab NAME wasworking

full-time to…

Resi

dual

Rea

ding

Tim

es

nonPrC GM

nonPrc GMM

PrC GM

PrC GMM

Results

GME at the 2nd NP in non-PrC pair

while while Jessica

Russell


-60

-40

-20

0

20

40

60

80

100

120

because lastsemester

while-cd SHE wastaking

classes while-ab NAME wasworking

full-time to…

Resi

dual

Rea

ding

Tim

es

nonPrC GM

nonPrc GMM

PrC GM

PrC GMM

Results

GME at the 2nd NP in non-PrC pair NO GME at the 2nd NP in PrC pair

Condition C – immediate

while while Jessica

Russell


Principle C •  Gender mismatch effect reflects search for antecedent for a pronoun

No gender mismatch effect in Principle C conditions – constraint is immediate

•  Contrast is robust across structural environments in English

–  Experiment 2: It seemed worrisome to {him, his family} that John/Ruth … –  Experiment 3: {He, His managers} chatted amiably with some fans while

the talented young quarterback …

•  Similar contrasts across languages

–  Japanese: backwards anaphora in surface vs. scrambled word orders (Aoshima, Yoshida, & Phillips, 2009, Syntax)

–  Russian: backwards anaphora with variation in conjunctions & aspect (Kazanina & Phillips, 2010, QJEP)

9/15/10

8

Illusory NPI Licensing “The bills that no democratic senators supported will ever become law.”

“Ein Mann, der kein Bart hatte, war jemals glücklich.” a man who no beard has was ever happy

(Vasishth et al., 2008; Xiang, Dillon, & Phillips, 2009)

Illusory Agreement “The key to the cells unsurprisingly were rusty …”

“The musicians who the reviewer praise so highly …”

(Bock & Miller 1991; Pearlmutter et al. 1999; Deevy et al. 1998; Staub 2009; Wagers, Lau, & Phillips 2009; Eberhard et al. 2005)

Illusory Comparative

“More people have been to Russia than I have.”

(Montalbetti 1984, Townsend & Bever 2001, Wellwood et al. 2009)

I’m not going to solely blame all of man’s activities on changes in climate

I’m not one to attribute every activity of man to climate change

9/30/08

10/02/08

Kim & Osterhout 2005, J. Mem. Lang.

__ The hearty meal was devoured … … The hearty meal was devouring …

“Thema'c P600s”

For breakfast the boys would only eat toast and jam. For breakfast the eggs would only eat toast and jam.

Kuperberg et al., 2003, Cogn. Br. Res.; see also Kolk et al. 2003, Hoeks et al. 2004 Reviews: Kolk 2006; Kuperberg 2007; Bornkessel-‐Schlesewsky & Schlesewsky 2008

P600

Evidence for independent semantic composition?

D. Memory and access mechanisms

9/15/10

9

Success & Failure •  Successful on-line constraint application requires:

(i) structured representations in memory (ii) mechanisms that can use those structures to access relevant material

•  Therefore, failure of on-line constraint application could reflect:

(i) representations (ii) access mechanisms

Is there a green square?

Is there a green square? Is there a green square?


9/15/10

10



Is there a green square?

Visual Search Mechanisms

Feature search is (i) fast, set-size invariant (ii) susceptible to interference and “illusory conjunction”

Conjunction search is slow, serial

(Treisman & Gelade 1980 etc.; but cf. McElree & Carrasco, 1999)

9/15/10

11

Two ways to search structures

•  Ordered search path through a structure (as in a standard address-based memory architecture)

–  Easy to implement structure-sensitive search –  Predicts timing advantage for local material –  Requires complete and accurate encoding of structure

•  Parallel search in content-addressable memory (cf. McElree 2000; Lewis & Vasishth, 2006; Gordon & Hendrick 2004)

–  Efficient retrieval of feature-matching items –  Risk of retrieving structurally inappropriate items –  Does not require full encoding of hierarchy in memory –  Not clear how to use relational notions in retrieval (e.g., c-command)

[+wh]

0 0 0 0 0 0 1

Two ways to search structures

•  Ordered search path through a structure (as in a standard address-based memory architecture)

–  Easy to implement structure-sensitive search –  Predicts timing advantage for local material –  Requires complete and accurate encoding of structure

•  Parallel search in content-addressable memory (cf. McElree 2000; Lewis & Vasishth, 2006; Gordon & Hendrick 2004)

–  Efficient retrieval of feature-matching items –  Risk of retrieving structurally inappropriate items –  Does not require full encoding of hierarchy in memory –  Not clear how to use relational notions in retrieval (e.g., c-command)

Evidence for Parallel Search •  #1: Non-structural interference effects

–  Retrieval of feature-matching but structurally inappropriate items

Negative Polarity Items (Drenhaus et al. 2005) Agreement (Wagers et al. 2009; Badecker & Lewis 2008)

•  #2: Timing non-effects

–  Processing dynamics identical across all dependency lengths

Wh-movement (McElree et al. 2003) Ellipsis (Martin & McElree 2008)

•  Stronger claim (Lewis, McElree, …)

–  Parallel cue-based retrieval is pervasive, even exclusive –  Sentences encoded as ‘chunks’ in memory –  No encoding of order

McElree et al. 2003

9/15/10

12

Speed-Accuracy Tradeoff (SAT) Possible Outcomes

Fig. 3 presents hypothetical SAT functions illustrat-ing how different SAT timecourse patterns can dis-criminate between alternative retrieval processes.Consider first the expected result that interpolating morematerial between the filler and gap position decreases theaccuracy of responding. Recall that this could be be-cause there is a lower probability that a representationof the filler is available when the verb is processed and/or

because there is a higher probability of misanalyzingmaterial up to and including the final verb. If additionalmaterial decreases only the overall accuracy of re-sponding, the corresponding functions will differ in as-ymptotic level alone. Panel A depicts two hypotheticalconditions that differ in this manner.

The pre-asymptotic portion of the SAT functionmeasures processing speed or dynamics, jointly specifiedby the intercept of the function (when accuracy departsfrom chance, d 0 ¼ 0) and the rate at which accuracygrows from intercept to asymptote. The intercept mea-sures the minimum time needed to form an interpreta-tion that would serve to discriminate acceptable fromunacceptable forms. The rate of the SAT function re-flects either the rate of continuous information accrual ifprocessing is continuous or the distribution of finishingtimes if processing is discrete or quantal (Dosher, 1976,1979, 1981, 1982, 1984; Meyer, Irwin, Osman, & Kou-nois, 1988). In either case, differences in intercept or rateimplicate underlying differences in the speed of pro-cessing. This situation is depicted in Panel B of Fig. 3,where the functions are associated with different inter-cepts and rates of rise to a common asymptote.

If access to the filler!s representation requires a searchprocess when the matrix verb is encountered, then theSAT intercept and/or rate of will systematically slow asmore material is interpolated between the filler and gap.Rate or intercept differences can arise from factors otherthan retrieval speed; for example, they might arise fromdifferences in the inherent complexity of computing

Fig. 2. Sample trial sequence illustrating the speed-accuracy tradeoff (SAT) variant of the acceptability judgment task.

Fig. 3. Hypothetical SAT functions illustrating two conditionsthat differ by SAT asymptote only (A) or by SAT intercept andrate (B).

B. McElree et al. / Journal of Memory and Language 48 (2003) 67–91 73

Asymptotic difference Reflects the strength of the representation or the likelihood of completing a parse/process.

Rate/intercept difference Reflects the speed of processing: how quickly information accumulates continuously, or the differences in an underlying discrete finishing time distribution.

Evidence from memory dynamics

ACCESSING SEMANTIC AND PHONOLOGICAL INFORMATION 179

position functions within each judgment. After identifyingthe best fit for each judgment, all 15 conditions (3 judg-ments ! 5 serial positions) were fit simultaneously inorder to isolate differences among the three judgments. Inall cases, the fits were performed on individual subjectdata. Fits of the average (over subjects) data were used tosummarize consistent patterns across subjects.

To quantify the impact of serial position on the retrievalfunctions for each judgment, sets of competitive fits were

performed that systematically varied the three parametersof Equation 1. These fits ranged from a null model, in whichthe 5 serial position functions were fit with a single ! (as-ymptote), " (rate), and # (intercept) parameter, to a fullysaturated 15-parameter model, in which each serial posi-tion was allotted a unique !, ", and # parameter. This analy-sis yielded a clear and consistent pattern across the threejudgments. In each case, the best fit to the data was a 5!-1"-2# model. With one exception (noted below), models

Figure 3. Average d $ accuracy (symbols) as a function of processing time for item(A), rhyme (B), and synonym (C) judgments. Smooth curves in each panel show thebest fits of Equation 1 with (the average) parameters listed in Table 1.

talk – yard – boat – store -‐ tales

Probe recognition – SAT response-signal task Wickelgren, et al., 1980, McElree & Dosher, 1989

FAST SLOOOOOWWWW

Bi-partite architecture of memory

Passive

Stringent limitations on the scope of information that can be concurrently processed

Broadbent 1958; Wickelgren et al., 1980; Garavan, 1998; Cowan, 2001; McElree, 2006; Verhaegen & Basak, 2007; Jonides et al., 2008

Active Passive “Focal attention” 1-item capacity

McElree & Griffith (1995)

Thematic *… alarm books

Subcategorization *… agree books

Category *… rarely books

McElree & Griffith (1995)

Detection of thematic violations delayed relative to category and subcategorization violations

9/15/10

13

Distance affects asymptote, but not temporal dynamics Interpretation: whole sentence accessed in parallel – no serial search Concern: wh-constructions are not a good test of distance effects

Part #2

Long-distance Dependencies Basic Paradigms and Generalizations

How filler-gap relations are built

Generalization 1

Wh-movement is Local

Long-distance Wh-Questions

Few people think that anybody realizes that Englishmen cook wonderful dinners

9/15/10

14


Few people think that anybody realizes that Englishmen cook what


What do few people think that anybody realizes that Englishmen cook gap

✓

Island Constraints

What do Few people believe anybody who claims that Englishmen cook what

Island Constraints

What do Few people believe anybody who claims that Englishmen cook what

Relative Clause

Island Constraints

What do few people believe anybody who claims that Englishmen cook gap

Relative Clause

Island Constraints


✘

Relative Clause

9/15/10

15

Island Constraints * Who did the candidate read a book that praised ___?

[relative clause island] * Who did the candidate read The Times’ article about ___?

[complex NP island] * Who did the candidate wonder whether the press would denounce ___?

[wh-island] * Why did you remember that the senator supported the bill ___?

[factive island] * Who did the fact that the candidate supported ___ upset voters in Florida?

[subject island] * Who did the candidate raise two million dollars by talking to ___?

[adjunct island]

Standard Conclusion: wh-movement must be local

Generalization 2a

Longer is ‘harder’

(Phillips, Kazanina, & Abada, 2005)

Length Matters

Generalization 2b

‘Active’ Gap Finding

Active Gap Creation

My brother wanted to know who

(Stowe 1986, Crain & Fodor 1985) �

Active Gap Creation

My brother wanted to know who Ruth


9/15/10

16

Active Gap Creation

My brother wanted to know who Ruth will


Active Gap Creation

My brother wanted to know who Ruth will bring


Active Gap Creation

My brother wanted to know who Ruth will bring gap


Active Gap Creation

My brother wanted to know who Ruth will bring us


Active Gap Creation



home to at Christmas. �

Readers slow down upon encountering an NP where a gap was expected.�

My brother wanted to know if Ruth will bring us home to Mom at Christmas. �

Slowdown

970 ms

755 ms

Stowe results •  My brother wanted to know…

…if Ruth will bring us home to Mom at Christmas. …who __ will bring us home to Mom at Christmas. …who Ruth will bring __ home to Mom at Christmas. …who Ruth will bring us home to __ at Christmas.

•  Ruth us Mom IF 661 755 755 Wh-S -- 801 812 Wh-O 680 -- 833 Wh-P 689 970 --

9/15/10

17

‘Active’ gap finding •  Highly robust generalization

–  Filled gap effects (Crain & Fodor 1985; Stowe 1986; et seq.) –  Plausibility manipulations (Boland et al. 1990; Traxler &

Pickering 1996) –  Cross-modal priming (Nicol & Swinney 1989) –  Eye-movements in visual world paradigm (Sussman & Sedivy

2003) –  Electrophysiological indices (Garnsey et al. 1989; Kaan et al.

2000; Phillips et al. 2005 etc.)

•  Wh-dependency formation occurs at least as soon as an appropriate verb is encountered

… but this does not bear on issues of transformational vs. non-transformational approaches

Head-mounted eye-tracker

Traxler & Pickering 1996 •  Plausibility manipulation - eye-tracking

–  That’s the {pistol/garage} with which the heartless killer shot the hapless man yesterday afternoon.

–  That’s the {garage/pistol} in which the heartless killer shot the hapless man yesterday afternoon.

Head-mounted Eye-Tracking

Sussman & Sedivy (2003)

shoe

spider

Jody

milk

Y/N: “Did Jody squash the spider with her shoe?”

WH: “What did Jody squash the spider with __?”

WH

Y/N verb

Sussman & Sedivy, 2003, LCP

Visual world measure of comple_ng filler-‐gap dependencies

9/15/10

18

Scenario with multiple events

Q: Can you tell me…

Wh: … what Emily was ea_ng the cake with ___ ? YN: … if Emily was ea_ng the cake with the fork?

1. eat

2. wash

Omaki, Trock, Wagers, Lidz & Phillips, CUNY 2009

Steps for Interpretation

“… what Emily was ea_ng the cake with __?”

1.  Recognize the verb; lexical retrieval

2.  Integrate wh-‐filler into the structure; build seman_c representa_on for VP “EAT WHAT”

3.  Recall events from story (“It is the cake that was eaten.”)

4.  Program and execute eye movement

Fixations on Cake

*

Verb Onset

WH

YN

“Can you tell me what/if Emily was ea_ng …”

400 (+200) ms

Omaki, Trock, Wagers, Lidz & Phillips, CUNY 2009

Electrophysiology of Sentence Comprehension

•  Semantic anomaly

N400

I drink my coffee with cream and sugar I drink my coffee with cream and socks

Kutas & Hillyard (1980)

N400

ERP Sentence Processing

•  Developing understanding of N400 is informa_ve

•  Response to normal sentences

N400"

Fully CongruentMost new drugs are tested on"white lab rats."

Van Petten & Kutas (1991)"

Amy bought the napkins that the café manager diligently folded in the booth."Amy bought the napkins that the café manager diligently baked in the booth."

(Yeung et al., 2004)"

MEG counterpart of N400 atsuccessive word positions in sentence comprehension."

9/15/10

19

Lau, Phillips, & Poeppel, Nat. Revs. Neurosci. 2008 Ellen Lau, Tufts/MGH

The N400 probably reflects lexical access rather than combinatorial semantic processing

(Garnsey, Tanenhaus, & Chapman, 1989)

N400 as measure of filler-gap dependency completion.

… called the customer … … called the article …

… which customer … called … … which article … called …

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)

he mows he *mow

P600

(Slide from Kaan (2001)

he mows he *mow

P600

Left Anterior Negativity (LAN)

(Slide from Kaan (2001) (Kaan, Harris, Gibson, & Holcomb, 2000)"

Long-Distance Dependencies

WH "Emily wondered who the performer in the concert had imitated "for the audienceʼs amusement.

Control "Emily wondered whether the performer in the concert had imitated "a pop star for the audienceʼs amusement."

P600 reflects normal structure-building processes."

“P600 amplitude is an index of syntactic integration difficulty.”"

P600 amplitude should covary with integration difficulty."

9/15/10

20

Long-Distance Dependencies a. Short control The actress wished that the producers knew that the witty host would tell the jokes during the party. b. Short WH The actress wished that the producers knew which jokes the witty host would tell __ during the party.

c. Long control The producers knew that the actress wished that the witty host would tell the jokes during the party. d. Long WH The producers knew which jokes the actress wished that the witty host would tell __ during the party.

(Phillips, Kazanina & Abada, Cog. Br. Res., 2005)"

Embedded Verb

The actress wished that the producers knew that the witty host would tell …"The actress wished that the producers knew which jokes the witty host would tell…"The producers knew that the actress wished that the witty host would tell …"The producers knew which jokes the actress wished that the witty host would tell…"

Embedded Verb


Embedded Verb


Interim Summary •  ‘Active’ formation of filler-gap dependencies

–  Diverse measures converge on similar generalization

–  Most measures give timing information - some just pinpoint ‘difficulty’, others show more specific measures

–  ‘Active’ mechanism need not be specific to wh-movement

Can timing evidence resolve theoretical disputes?

9/15/10

21

Competing Theories

What do Englishmen cook gap/trace/copy

What do Englishmen cook Direct Association ��HPSG/GPSG ��Categorial Grammar��Dependency Grammar��etc.

Indirect Association ��Transformational Grammar��(--> Projection Principle)

Competing Theories

What do Englishmen cook gap/trace/copy

What do Englishmen cook Direct Association ��HPSG/GPSG ��Categorial Grammar��Dependency Grammar��etc.

Indirect Association ��Transformational Grammar��(--> Projection Principle)

Attempts to distinguish between these theories ��using evidence from language processing…

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)

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)

Surprise at pronoun following verb is ��compatible with both theories!

Cross-modal Priming

Cross-Modal Priming

The guests drank vodka, sherry and port at the reception

(Swinney 1979, Seidenberg et al. 1979)

9/15/10

22

Cross-Modal Priming


WINE

SHIP


Cross-Modal Priming


WINE

SHIP


Cross-Modal Priming


WINE

SHIP


Cross-Modal Priming


WINE

SHIP


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)



boy

girl

boy

girl

faster decision

same speed


9/15/10

23



boy

girl

boy

girl

faster decision

same speed


Both theories can account for reactivation at or after the verb!

Pickering & Barry 1991

3. Verb Position vs. Trace Position

(Pickering & Barry 1991)

give NP PP



give NP PP

To which child did the teacher give [a long speech about the importance of honesty] ___?



give NP PP


Various diagnostics indicate that the dependency��is formed at the verb, not at the trace position.



give NP PP


Various diagnostics indicate that the dependency��is formed at the verb, not at the trace position.

Still compatible with both theories!

9/15/10

24

WH

CP

C IP

VP

NP

V

…

WH

CP

C IP

VP

NP

V

…

Direct Association Gap-based Approach

gap

Effects at Verb Position

#1

#1

#2

Timing evidence as theoretical arbitration

…works when theories make timing predictions

see Phillips & Wagers 2007 (Oxford Hbk of Psycholing.) On course readings page

Pre-Verbal Gap Effects

•  Suggestion… Perhaps the two theories could be distinguished by effects of dependency formation at argument positions that precede the verb of a clause.

•  Rationale Filled-gap effect expected at pre-verbal position only under indirect association/gap-based theory.

Motivations

What is driving gap creation?

Locality in Japanese Wh-Questions

Aoshima, Phillips & Weinberg. J. Mem. Lang. 51, 23-54 (2004)

Japanese Wh-Question Formation

Japanese wh-phrases are canonically in-situ.

Japanese uses question particles (Q-particles) to mark questions. �Yes/No Question

-nom book-acc read-Q

Wh-Question

‘What did Mary read?’

‘Did Mary read the book?’

Mary-ga hon-o yonda-no.

Mary-ga nani-o yonda-no. -nom what-acc read-Q

9/15/10

25

Japanese Wh-Question Formation Position of Q-particle indicates the scope of wh-questions.

Direct Wh-question

‘What did Tom say that Mary read?’


Direct Wh-question Tom-wa [Mary-ga nani-o yonda-to] itta-no? -top -nom what-acc read-DeclC said-Q ‘What did Tom say that Mary read?’


Indirect Wh-question

‘Tom said what Mary read.’



Indirect Wh-question Tom-wa [Mary-ga nani-o yonda-ka] itta. -top Mary-nom what-acc read-Q said

Japanese: Types of wh-questions depend on the positions of the question particle.

‘Tom said what Mary read.’


Processing Japanese Wh-Questions

Tom-wa

-top


Tom-wa [Mary-ga

-top -nom

This nominative NP indicates the start of the embedded clause.

9/15/10

26


Tom-wa [Mary-ga nani-wo

Due to verb-final property, wh-scope is temporarily ambiguous prior to the verb.

-top -nom what-acc


Tom-wa [Mary-ga nani-wo

-top -nom what-acc read

yonda-to (Declarative)

yonda-ka (Q-Particle)

Due to verb-final property, wh-scope is temporarily ambiguous prior to a verb.

Experiment: Processing In-situ Wh-Question

Tom-wa [Mary-ga nani-wo yonda-to (Declarative)


Miyamoto and Takahashi (2003); Aoshima, Phillips and Weinberg (2004)

Experiment: Self-paced reading task �

----- --- --- ---- ---- --- --------

先生は --- --- ---- ---- --- --------

teacher-top

Experiment: Self-paced reading task

----- 生徒が --- ---- ---- --- --------

student-nom


9/15/10

27

----- --- 教室で ---- ---- --- --------

classroom-at


----- --- --- だれに ---- ---- --- --------

who-dat


----- --- --- ---- まんがを ---- --------

comics-acc


----- --- --- ---- ---- あげたか --------

gave-Q


----- --- --- ---- ---- --- 知っています。

knows

Experiment: Self-paced reading task Experiment: Processing Japanese Wh-Question

Readers slow down when they see a declarative complementizer –to ‘that’ at the first verb.

……….. Mary-ga nani-wo yonda-to (Declarative)


Miyamoto and Takahashi (2003); Aoshima, Phillips and Weinberg (2004)

Slowdown 889 ms

787 ms

9/15/10

28

Parallelism between English and Japanese

Tom-top Mary-nom what-acc read-Q … .

My brother wanted to know who Ruth will bring gap ….

English speakers want to interpret the thematic role of wh-phrase.

Japanese speakers want to interpret the wh-scope of wh-phrase.

Parallelism between English and Japanese:

Native speakers of English and Japanese both show a reliable locality bias in processing wh-questions.

Who-dat

who-dat

Scrambled wh-question

In-situ wh-question

Tom-top [Mary-nom book-acc gave-Q] said.

Tom-top [Mary-nom book-acc gave-Q] said.

who-dat

‘Tom asked who Mary gave a book to.’

‘Tom asked who Mary gave a book to.’

Long-distance Wh-scrambling formation

Who-dat Tom-top [Mary-nom book-acc gave-Q] said. gap

Long-distance Wh-scrambling formation

  Fronted wh-phrases – Wh-gap dependencies must be formed, just like English wh-questions.

Ambiguous structures

Who-dat Tom-top [Mary-nom book-acc



Verb-final property provides temporarily ambiguous structures.

read-Q] said.

read-that] said-Q.



read-that] said-Q.

gap


‘To whom did Tom say that Mary read the book?’

9/15/10

29




read-Q] said. gap

‘Tom said who Mary read the book to.’


Experiment

read-Q]

read-that]

Readers slow down when they see a declarative complementizer –to ‘that’ at the first verb.

Slowdown

Aoshima, Phillips and Weinberg (2004)

835 ms

723 ms

Long-distance scrambling is preferred!

English Filled Gap Effect

who

Ruth

will

bring

us

My brother wanted to know

home to at Christmas Slowdown

Readers slow down upon encountering an NP where a gap was expected.

(Stowe 1986, Crain & Fodor 1985)

Japanese Filled-Gap Effect

Position of the unexpected NP is before the verb.

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

WH-dat

NP-top

CP

gap

NP-nom

Verb

VP

NP-dat

Slowdown

Verb

Japanese Filled-Gap Effect

WH-dat

NP-top

CP

NP-nom VP

WH-nom

NP-dat

CP

NP-nom

Verb

VP

NP-dat

target control

gap

Verb NP-dat

Slowdown

Verb Verb

9/15/10

30

Low-tech Replication •  Pencil-and-paper sentence fragment completion

–  Wh-dat NP-nom NP-nom …

•  Fragments with dative wh-phrases were completed with embedded gap in 61% of trials

Implications •  Japanese speakers show preference to interpret fronted wh-

phrase as long-distance scrambled

•  This does follow naturally in terms of local satisfaction of grammatical constraints

•  … but it does not correspond to the notion of locality that syntactic theory typically aims to capture

•  Syntactic locality does not reduce to ambiguity resolution constraints

Traces (again)

WH

CP

C IP

VP

NP

V

…

WH

CP

C IP

VP

NP

V

…

Direct Association Gap-based Approach

gap

Effects at Verb Position

#1

#1

#2

Traces (again) •  Does pre-verbal dependency formation implicate

gaps/traces?

–  Yes! If direct association to verb requires presence of verb

–  No! If verb position is built in advance of overt verb

9/15/10

31

Constraints on Unbounded Dependencies

Stowe 1986 •  Experiment 1

My brother wanted to know …

…if Ruth will bring us home to Mom at Christmas …who Ruth will bring us home to at Christmas

•  Experiment 2

The teacher asked …

…if [the silly story about Greg’s older brother] was supposed to mean anything. …what [the silly story about Greg’s older brother] was supposed to mean.

Stowe 1986 •  The teacher asked …

…if [the silly story about Greg’s older brother]… …what [the silly story about Greg’s older brother]…

…if the team laughed about Greg’s older brother… …what the team laughed about Greg’s older brother…

•  the silly story about Greg’s if-S 611 677 752 750 798 wh-S 616 698 760 880 800 if-V 613 735 754 678 782 wh-V 608 698 736 755 1063

Traxler & Pickering 1996 •  Plausibility manipulation, subject islands

–  WAITING FOR A PUBLISHING CONTRACT The big city was a fascinating subject for the new book.

–  We like the book that the author wrote unceasingly and with great dedication about while waiting for a contract.

–  We like the city that the author wrote unceasingly and with great dedication about while waiting for a contract.

–  We like the book that the author who wrote unceasingly and with great dedication saw while waiting for a contract.

–  We like the city that the author who wrote unceasingly and with great dedication saw while waiting for a contract.

Islands Non-Islands

The Real-Time Status of Island Constraints

Phillips (2006), Language, 82, 795-823

9/15/10

32

Island Constraints


✘

Relative Clause

Island Constraints * Who did the candidate read a book that praised ___?

[relative clause island] * Who did the candidate read The Times’ article about ___?

[complex NP island] * Who did the candidate wonder whether the press would denounce ___?

[wh-island] * Why did you remember that the senator supported the bill ___?

[factive island] * Who did the fact that the candidate supported ___ upset voters in Florida?

[subject island] * Who did the candidate raise two million dollars by talking to ___?

[adjunct island]

Standard Conclusion: wh-movement must be local

(Phillips, Kazanina, & Abada, 2005)

Length Matters

Active Gap Creation



home to at Christmas. �

Readers slow down upon encountering an NP where a gap was expected.�

My brother wanted to know if Ruth will bring us home to Mom at Christmas. �

Slowdown

970 ms

755 ms

‘Active’ gap finding •  Highly robust generalization

–  Filled gap effects (Crain & Fodor 1985; Stowe 1986; et seq.) –  Plausibility manipulations (Boland et al. 1990; Traxler &

Pickering 1996) –  Cross-modal priming (Nicol & Swinney 1989) –  Eye-movements in visual world paradigm (Sussman & Sedivy

2003) –  Electrophysiological indices (Garnsey et al. 1989; Kaan et al.

2000; Phillips et al. 2005 etc.)

•  Wh-dependency formation occurs at least as soon as an appropriate verb is encountered

… but this does not bear on issues of transformational vs. non-transformational approaches

A Common Inference •  Grammatical generalization: wh-dependencies are local •  Parsing generalization(s): local wh-dependencies are easier/

preferred •  ERGO… perhaps the grammatical generalization derives from

the parsing generalization (Fodor 1978; Berwick & Weinberg 1984; Deane 1991; Pritchett 1991; Kluender & Kutas 1993; Hawkins 1999; Sag et al. 2005; Maratsos & Kowalsky 2005 –  Variant I: locality constraints are nevertheless grammaticized –  Variant II: locality constraints in grammar are epiphenomenal

Fodor Weinberg Berwick Kluender Hawkins Sag Maratsos

9/15/10

33

Real-time Status of Island Constraints

•  Are island constraints respected in real-time syntactic computation?

•  Many studies - conflicting results (various techniques, various island-types, etc.)

•  …but, it is not even true of the grammar that it disallows long-distance dependencies that cross islands…

Parasitic Gaps

which school did the proposal to expand the school ultimately overburdened the teachers.

Parasitic Gaps


Parasitic Gaps

which people did the proposal to expand the school ultimately overburdened the teachers.

Parasitic Gaps


Parasitic Gaps


Generalization (Subject Island Constraint) No long-distance dependencies across subject boundaries

9/15/10

34

Parasitic Gaps


Generalization (informal) Violations can be rescued by subsequent well-formed gaps.

Parasitic Gaps


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.

Grammaticality Ratings

1

1.5

2

2.5

3

3.5

4

Good Bad Both

Gap Type

Acc

ep

tab

ilit

y R

ati

ng

INFFIN

Parasitic Gaps


which school did the proposal that expanded the school ultimately overburdened the teachers.

which students…

which students…

implausible at ‘expand’��plausible at ‘overburden’

plausible at ‘expand’��plausible at ‘overburden’

Materials a)  The school superintendent learned which schools the proposal to

expand drastically and innovatively upon the current curriculum would overburden during the following semester. [INF, Plaus]

b)  The school superintendent learned which high school students the proposal to expand … [INF, Implaus]

c)  The school superintendent learned which schools the proposal that expanded … [FIN, Plaus]

d)  The school superintendent learned which high school students the proposal that expanded … [Fin, Implaus]

Materials

9/15/10

35

… which schools/students the proposal to expand …

*

… which schools/students the proposal that expanded …

n.s

Implications - Previous Findings

•  This experiment showed violation of one type of island, and non-violation of another type of island: same task, same participants

•  Suggests that variability in previous results cannot just be attributed to methodological artifacts

•  Incrementality and accuracy preserved

•  Can variability in previous results be due to choice of islands tested, and to possibility of parasitic gaps?

Implications - ‘Parsing Accounts’

•  Repeated attempts to reduced movement constraints to artifacts of ‘processing constraints’ (working memory, etc.)

•  The existence of parasitic gaps shows that it’s not true that dependencies that cross islands are always impossible.

•  If subject parasitic gaps were only marginally acceptable, or were processed non-incrementally, this would be compatible with ‘parsing accounts’ of islands

•  But since parasitic gaps are constructed immediately, this is more problematic for ‘processing accounts’ of islands

Implications - ‘Parsing Accounts’



Any ‘processing based’ account of why this is bad…

…will fail to explain why the first gap can be created here…

(cf. Deane, 1991; Pritchett, 1991)

Therefore… •  The notion that long-distance dependencies cannot cross

islands is an over-simplification

•  The parser appears to be well aware of this

•  Creates a challenge for attempts to ‘explain away’ island phenomena as artifacts of processing

•  Further evidence that a good deal of what we know about language is deployed immediately in language processing

9/15/10

36

Islands and Resource Limitations

Jon Sprouse (UC Irvine)

Matt Wagers (UC Santa Cruz)

Sprouse, Wagers, & Phillips, submitted

The logic of resource limitation theories

an even greater processing disruption

processing load 1

processing load 2 + + = limited

resources

John thought that you knew who the president would pardon __ ?

John knew who you thought that the president would pardon __ ?

processing load 1: Longer wh-dependencies are harder to process than shorter dependencies (LENGTH)

Crucially, longer dependencies lead to lower acceptability

There are many potential reasons for this: memory encoding, or maintenance, or retrieval...

The logic of resource limitation theories

an even greater processing disruption

processing load 1

processing load 2 + + = limited

resources

processing load 2: Island structures are harder to process than non- island structures (STRUCTURE)

Who __ thinks [CP that John bought a car ]? Who __ wonders [CP whether John bought a car ]?

Crucially, island structures without island violations lead to lower acceptability

There are many potential reasons for this: referential processing, syntactic complexity, semantic complexity...

(Kluender and Kutas 1993b, Sprouse 2007)

Does Capacity Affect Island Acceptability?

Participants:

N = 144 Age Range: 18-25 Mean Age: 21.13 (1.67)

Expected mean: 4 (1) (Cowan 2000)

Actual mean: 4.17 (.98)

To avoid rehearsal: Repeat the word the while listening To avoid mnemonics: Repeat the task 10x with the same 8 words in a

different order

p < .001

high low

Basic Idea: if island effects reflect capacity constraints, then varying capacity should modulate island judgments.

Verbal memory span test (Cowan 2000 for review) Listen to 8 words spoken at .5s intervals Recall as many as possible in the correct order

5.38 n=36

2.98 n=36

Adjunct Island - comparisons 1. Baseline: Who __ thinks that John forgot his briefcase at the office? 2. Wh-only: What do you think that John forgot __ at the office? 3. Island-only: Who __ laughs if John forgets his briefcase at the office? 4. Both: What do you laugh if John forgets __ at the office?

None of the paired comparisons were significant: the low and high groups rated these sentences equally, and both show the same interaction.

Paired comparisons: low versus high

p<.44 p<.35

p<.52 p<.92

1

3

4

2

NP Island - comparisons 1. Baseline: Who __ claimed that John bought a car? 2. Wh-only: What did you claim that John bought __? 3. Island-only: Who __ made the claim that John bought a car? 4. Both: What did you make the claim that John bought__?

None of the paired comparisons were significant: the low and high groups rated these sentences equally, and both show the same interaction.


p<.18 p<.38

p<.50 p<.23

1

3

4

2

9/15/10

37

1

2 3

4

It looks like the high capacity group rates grammatical structures that involve additional processing resources as less acceptable


p<.51 p<.01

p<.004 p<.88

This looks like calibration of the scale: structures that require additional processing are pushed lower to better equalize the scale distribution

But there is no effect for island violations, and the super-additive interaction survives this calibration

Whether Island - comparisons 1. Baseline: Who __ thinks that John bought a car? 2. Wh-only: What did you think that John bought __? 3. Island-only: Who __ wonders whether John bought a car? 4. Both: What did you wonder whether John bought __?

1 2

3

4

It looks like the high capacity group rates grammatical structures that involve additional processing resources as less acceptable


p<.31 p<.004

p<.003 p<.51

This looks like calibration of the scale: structures that require additional processing are pushed lower to better equalize the scale distribution

But there is no effect for island violations, and the super-additive interaction survives this calibration

Subject Island - comparisons 1. What do you think the speech interrupted __ ? 2. What do you think __ interrupted the TV show? 3. What do you think the speech by the president interrupted the TV show about __? 4. Who do you think the speech by __ interrupted the TV show about surgery?

Expt 2

MagEst n=173

2 memory tasks: serial recall n-back

Early Warning Signals for Japanese Islands

Masaya Yoshida Sachiko Aoshima

Colin Phillips

John-ga …

John-nom …

(Mazuka & Itoh 1995)

John-ga Mary-ni …

John-nom Mary-dat …


9/15/10

38

John-ga Mary-ni ringo-o …

John-nom Mary-dat apple-acc …


John-ga Mary-ni ringo-o tabeta …

John-nom Mary-dat apple-acc ate …


John-ga Mary-ni ringo-o tabeta inu-o ageta

John-nom Mary-dat apple-acc ate dog-acc gave


John-ga Mary-ni [[ti ringo-o tabeta] inu-oi] ageta

John-nom Mary-dat [apple-acc ate dog-acc] gave ‘John gave Mary the dog that ate the apple


Japanese Relative Clauses •  Notorious garden paths arise because relative

clauses are head final in Japanese.

•  But: overt movement/scrambling in Japanese is subject to (roughly) the same island constraints as English

Time-course of gap creation

  Gap-creation takes place before the verb is processed. Structures are built incrementally.

  Gap is posited in the most deeply embedded clause.

  Embedded clause could be an island (e.g. relative clause)

  How could island violations ever be avoided in real-time computation?

  What evidence could allow a speaker to learn about avoiding islands?

NP-subj

Verb CP

gap

NP-subj

Verb

VP

WH-dat

gap

9/15/10

39

Early Warning •  Japanese numeral classifiers

–  san-satsu hon 3-cl book

–  san-nin gakusei 3-cl students

•  Numeral classifiers and Relative Clauses

–  John-ga san-satsu-no [RC … ] hon-o yonda John-nom 3-cl [RC … ] book-acc read

Early Warning •  Japanese numeral classifiers

–  san-satsu hon 3-cl book

–  san-nin gakusei 3-cl students

•  Numeral classifiers and Relative Clauses

–  John-ga san-satsu-no [RC gakusei-ga … ] hon-o yonda John-nom 3-cl [RC student-nom… ] book-acc read

Early Warning •  Can numeral classifiers be used to detect relative

clauses?

–  John-ga san-nin-no gakusei-ga … John-nom 3-clhuman student-nom …

–  John-ga san-satsu-no gakusei-ga … John-nom 3-clbooks etc. student-nom …

Early Warning •  Can numeral classifiers be used to detect relative

clauses?

–  John-ga [san-nin-no gakusei-ga … John-nom [3-clhuman student-nom …

–  John-ga san-satsu-no [gakusei-ga … John-nom 3-clbooks etc. [student-nom …

complement��clause

relative��clause

Early Warning •  Can numeral classifiers be used to detect relative clauses?

–  John-ga [san-nin-no gakusei-ga … John-nom [3-clhuman student-nom …

–  John-ga san-satsu-no [gakusei-ga … John-nom 3-clbooks etc. [student-nom …

•  Experiment #1: sentence fragment completion (n = 64) rel. clause other

classifier match 1 566 classifier mismatch 483 91

complement��clause

relative��clause

Early Warning •  Can numeral classifiers be used to detect relative clauses?

–  John-ga san-nin-no [gakusei-ga … V] NP-o … V John-nom 3-clhuman [student-nom …

–  John-ga san-satsu-no [gakusei-ga … V] NP-o … V John-nom 3-clbooks etc. [student-nom …

•  Experiment #2: reading-times for relative clauses (n = 32)

–  are relative clauses processed more easily following a mismatching classifier-noun sequence?

classifier match

classifier��mismatch

9/15/10

40

550

650

750

850

950

1050

1150

1 2 3 4 5 6 7 8 9 10 11 12Regions

Mea

n RT

(ms.)

GNC MathingGNC Mismatching

Early Warning

mismatch"

RC verb + head"

Direct signals of relative clause processed more easily in��classifier-mismatch (‘indicator’) condition.

Early Warning •  Experiment #3: Filled-gap Effect and Relative Clauses (n = 80)

–  WH-DAT John-ga san-nin-no [gakusei-ga … NP-DAT John-nom 3-clhuman [student-nom …

–  WH-DAT John-ga san-satsu-no [gakusei-ga … NP-DAT John-nom 3-clbooks etc. [student-nom …

GNC Mismatching Conditions

550

600

650

700

750

800

850

900

950

1000

1 2 3 4 5 6 7 8 9 10 11 12Regions

Mea

n RT

s (m

s.)

Scr/GNC MismatchingNonScr/GNC Mismatching

GNC Matching Conditions

500550600650700750800850900950

1000

1 2 3 4 5 6 7 8 9 10 11 12Regions

Mea

n RT

s (m

s.)

Scr/GNC MatchingNonScr/GNC Matching

Matching��Classifier

Mismatching��Classifier


550

600

650

700

750

800

850

900

950

1000

1 2 3 4 5 6 7 8 9 10 11 12Regions

Mea

n RT

s (m

s.)



500550600650700750800850900950

1000

1 2 3 4 5 6 7 8 9 10 11 12Regions

Mea

n RT

s (m

s.)


NP-nom ±match




550

600

650

700

750

800

850

900

950

1000

1 2 3 4 5 6 7 8 9 10 11 12Regions

Mea

n RT

s (m

s.)



500550600650700750800850900950

1000

1 2 3 4 5 6 7 8 9 10 11 12Regions

Mea

n RT

s (m

s.)


NP-nom ±match



NP-dat

Filled-gap��Effect

Early Warning •  Yes - Japanese speakers can use numeral classifiers

to

–  pre-emptively construct relative clauses –  avoid island constraint violations

9/15/10

41

Next to anaphora …

Overflow …

Argument Structure

remind V NP

V NP IP

(Boland et al. 1995)

Argument Structure

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 ___?

remind V NP

V NP IP


Argument Structure




remind V NP

V NP IP


9/15/10

42

Argument Structure




remind V NP

V NP IP


Boland et al., 1995 1a. Which client did the salesman visit while in the city? b. Which prize did the salesman visit while in the city? 2a. Which child did your brother remind to watch the show? b. Which movie did your brother remind to watch the show?

Sentence Matching •  HOUSE

HOUSE •  HSEUO

HSEUO

•  HOUSE HORSE

•  HSEUO HSERO

(Freedman & Forster 1985)

Sentence Matching •  DOGS GROWL

DOGS GROWL

•  GROWL DOGS GROWL DOGS


Sentence Matching •  Specificity constraint violations

–  Who did the duchess sell a portrait of? –  *Who did the duchess sell Turner’s portrait of?

•  Other violations

–  Mary were writing a letter to her husband. –  Where does bears usually hibernate?

–  The baby ate his cereal up all. –  Lesley’s parents are chemical engineers both.


Grammatical Illusions: Encoding and navigating linguistic ...

Documents