The Psycholinguistics of Ellipsis Colin Phillips & Dan Parker Department of Linguistics Neuroscience & Cognitive Science Program University of Maryland

The Psycholinguistics of Ellipsis

Colin Phillips & Dan Parker

Department of LinguisticsNeuroscience & Cognitive Science Program

University of Maryland

languagescience.umd.edu

Overview• Grammatical alternatives

• Experimental investigations of ellipsis: 3 leading themes• Spoiler: these themes don’t resolve today’s debate• Theme #1: mismatching antecedents

• Interlude: theories and experiments on wh-movement and anaphora• Theme #2: accessing information about antecedents• Theme #3: does size matter (or distance)?

Slides available at http://www.ling.umd.edu/colin, under downloadable papers

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

A. Nature of antecedent

semantic/discourse

syntactic(roughly)

B. Content of ellipsis site

pointer/anaphor

detailedstructure

C. Derivational status ofstructure at ellipsis site

only prespell-out only post

spell-out

throughout

Baltin MerchantLi

Dalrymple et al. 1991Hardt 1993, etc.

Alternatives

Many!

Sag 1976, Williams 1977,Fiengo & May 1994, etc.

Experiments on Ellipsis: Three Themes

1. Mismatching antecedents

What to make of mismatches between antecedent and ellipsis sitee.g., Seeing the comet was nearly impossible, but John did __.

2. Accessing information about antecedents

3. Does size (or distance) matter?


semantic/discourse

syntactic(roughly)


pointer/anaphor

detailedstructure



spell-out

throughout

Baltin MerchantLi


Alternatives

Mismatch Studies

Mostly acceptability ratingsExploring grammatical status

of mismatches

e.g., Arregui, Clifton, Frazier, & Moulton, 2006; Kim, Kobele, Runner, & Hale, 2010

Consistent judgments/RTsInteresting storiesEarly days in testing what’sinside/outside grammar





Does processing of the ellipsis site involve accessing words from the antecedent (semantics or phonology)? Are binding relations between elided and non-elided material rapidly computed?



semantic/discourse

syntactic(roughly)


pointer/anaphor

detailedstructure



spell-out

throughout

Baltin MerchantLi


Alternatives

Accessing Antec. Features

Cross-modal lexical decisionVisual world eye-trackingSelf-paced reading, etc.

e.g., Shapiro & Hestvik 1995; Snider & Runner 2010; Yoshida, Dickey, & Sturt, 2011; Kaan, Wijnen, & Swaab, 2004

Various interesting effectsShows rapid access toantecedent … but does thisshow what’s in ellipsis site?





Does processing of the ellipsis site involve accessing words from the antecedent (semantics or phonology)? Are binding relations between elided and non-elided material rapidly computed?


Does resolution of ellipsis become slower/harder for larger/distant antecedents?


semantic/discourse

syntactic(roughly)


pointer/anaphor

detailedstructure



spell-out

throughout

Baltin MerchantLi


Alternatives

Size/Distance Studies

“Got it” semantic judgmentsSelf-paced readingSpeed-accuracy tradeoff (SAT)

e.g., Murphy 1985; Frazier & Clifton 2001; Martin & McElree 2008

A number of studies, andemerging consensus that sizedoes not matter (different morals drawn) … but are thefinings conclusive?


semantic/discourse

syntactic(roughly)


pointer/anaphor

detailedstructure



spell-out

throughout

Baltin MerchantLi


Alternatives



spell-out

throughout

Baltin MerchantLi

They agree: detailed structural representation in ellipsis site

They disagree: derivational “timing”

They don’t tell us: how their derivations relate to real-time mechanisms

So we’ll make something up for them …

A. Very little relation (cf. Townsend & Bever, 2001) no predictionsB. Real-time mechanism incrementally builds a representation that includes sound (PF), meaning (LF), plus mediating structure identical predictionsAny evidence of building structure at ellipsis site fits all 3 proposals

MoralFor psycholinguists to help with your grammatical disputes, it helps to come clean about your mentalistic commitments

(for menu of options, cf. Phillips & Lewis 2010)

Theme #1: Mismatches

“This information could have been released by Gorbachev, but he chose not to ___.”

attributed to Daniel Schorr, NPRcited in Hardt 1993

release this information

Theme #1: Mismatches• ‘Acceptability cline’ across various forms of antecedent-ellipsis mismatches

(Arregui, Clifton, Frazier, & Moulton 2006; Kim, Kobele, Runner, & Hale 2010)

• Various judgment studies fine-tune generalizationsAppeal to parser properties to account for mismatches

• Active-Passive > Passive-Activea. The advisor praised the student, and the old school-master was.b. The student was praised by the old school-master, and the

advisor did too.

Verbal Gerunds > Nominal Gerundsa. Singing the arias tomorrow night will be difficult, but Maria will.b. Tomorrow night’s singing of the arias will be difficult, but Maria will.

Category N-VP > Adj-VPa. The criticism of Roy was harsh, but Kate didn’t.b. The report was critical of Roy, but Kate didn’t.

Theme #1: Mismatches

Arregui, Clifton, Frazier, & Moulton 2006• VPE requires syntactic identity

… standard notion• VPE mismatches are ungrammatical• Partial acceptability reflects repair

• VP-repair (‘recycling’)– Transform mismatching antecedent– Rules guide repair process

– Cline reflects amount of repair work

Kim, Kobele, Runner, & Hale 2010• VPE requires syntactic identity

… in a novel gram. analysis• VPE mismatches are grammatical• Partial acceptability reflects search

• Search heuristics– Search for matching antecedent– Constraints guide search for

matching antecedent– Cline reflects amount of search

work

Interlude

Experimental findings in other theoretically contentious domains

Competing accounts of ellipsis

He ate something but I don’t know what he ate __.

He ate something but I don’t know what. No/minimal null structureAnaphor/‘pointer’ account

Null structure at foot of dep.All of today’s speakers

Competing accounts of wh-dependencies

What do Englishmen cook gap/trace/copy

What do Englishmen cookDirect AssociationHPSG/GPSGCategorial GrammarDependency Grammaretc.

Null structure at foot of dep.Transformational Grammar(--> Projection Principle)

Competing accounts of anaphora

John thinks that Mary hates him.Standard viewAnaphor points to content elsewhere in syntax/discourse

Pronominalization (Postal)Movement theory of control/reflexives (Hornstein et al.)

John thinks that Mary hates him John.

Experiments as Theory Arbitrators

Many of the themes raised in experiments on ellipsis have been investigated in experiments on wh-dependencies and anaphora.

Wh-dependencies: much discussion of whether expt. findings are decisive regarding gaps/traces vs. direct association.

Conclusion: the timing evidence probably isn’t decisive (yet).Gibson & Hickok 1993; Phillips & Wagers 2007; Kempen, LSA 2011

So it’s interesting to see parallel arguments being presented as theoretically decisive in the case of ellipsis.

Anaphora: nobody thinks they’re testing pronominalization etc.

Cross-modal Priming

The policeman saw the boy that the crowd at the party accused __ of the crime.

Semantic Associate PrimingSwinney, Ford, Frauenfelder, & Bresnan 1988; Nicol & Swinney 1989; McKoon, Ratcliff & Ward 1994; Nicol, Fodor, & Swinney 1994

The man was surprised at which beer/wine the judges awarded the first prize to __.fear

girl

group

Rhyme PrimingTanenhaus, Carlson, & Seidenberg 1985

Eye-fixations in visual world

Verb Onset

WH

YN

400 ms

Fixations on picture of cake

1. eat

2. wash

Q: Can you tell me…Wh: … what Emily was eating the cake with ___ ?YN: … if Emily was eating the cake with the fork?

Omaki, Trock, Wagers, Lidz & Phillips, 2009

Pronouns & Lexical Properties• In pronoun generation (production) phonological properties of the

antecedent are accessed

(Schmitt, Meyer, & Levelt 1999)

• In pronoun comprehension, effects of lexical frequency of the antecedent

(van Gompel & Majid 2004; Lago, Chow, & Phillips in prep.)

• These effects show how antecedents are accessed

Few would consider them as evidence for Postal-style pronominalization

Point of the Interlude …• In wh-dependencies and anaphora, many interesting experiments show

how antecedent information is accessed, and when.

This is all very useful for building an account of real-time computation.

• But evidence on access to antecedent properties does not show whether there’s unpronounced structure at the foot of the dependency

For any argument for full structure at an ellipsis site, ask this question:

Would the same argument convince us of the need for (i) traces/gaps for wh-movement, or (ii) a transformational analysis of anaphora?

Theme #2

Accessing Antecedents

Accessing Words in Antecedent

Cross-modal lexical decision shows semantic priming of noun inside antecedent (Shapiro, Hestvik, Lesan, & Garcia, 2003)

The old professor [VP loved the ocean], and the teenager […] did __ too …

Syntactically-defined antecedent accessed at ellipsis site.

PACIFIC

TEACHER

Accessing Antecedent Properties

Snider & Runner, AMLaP 2010 (& Sat. 10:30am)

The customer dropped the lock,and the manager did too.

The security guard opened the lock,and the night manager did too.

Experiment 1: semantic associates

Experiment 2: phonological associates

lock

key

log

lock

“Only if syntactic structure is present in ellipsis site should phonological information be reactivated.”

These arguments for syntactic structure in ellipsis parallel earlier findings on wh-dependencies and pronoun processing.

Fast Use of Syntactic Constraints• Rapid building of binding relations in sluicing (Yoshida, Dickey, & Sturt 2011)

Jane’s {grandfather|grandmother} told some stories at the family reunion, but we couldn’t remember which story about himself … [sentence continues]

Gender mismatch effect at reflexive when sluicing is a viable option(No corresponding effect at reflexive when pied-piped wh-PP blocks sluicing)

• Rapid sensitivity to islands in sluicing vs. sprouting (Yoshida et al. 2010)

Nick’s father was startled … because he smoked secretly in the garden

because he smoked something in the garden… but it wasn’t clear what …

Evidence of immediate sensitivity to islands for sprouting vs. sluicing

• Clever contrasts – but they motivate structure at ellipsis site to the same extent that connectivity effects in wh-movement motivate traces.

Fast resolution of gapping• ERPs suggest rapid detection of implausibility in gapped sentences (Kaan,

Wijnen, & Swaab 2004)

Ron took the planks for the bookcase, and Bill __ the hammer …Ron sanded the planks for the bookcase, and Bill __ the hammer …

N400 to implausible verb-noun combination

“the sentence processor […] reconstructs the verb information at the earliest possible occasion” (p. 584)

Any mechanism that gets the meaning can capture this.

Theme #3: Does size matter?• “The canonical interpretation of a literal copy mechanism is that copying

more information should take more time. One could simply assert that ‘copying’ does not require time, but we suggest that in that case, the notion ‘copy’ is no longer explanatory.” (Martin & McElree 2008, p. 894)

• A number of studies have tested whether size/complexity affects the time needed for ellipsis resolution.

Mixed results.But most currently assume that the evidence shows no size cost.

• Size effect ≠ copy mechanismNo size effect = no copy mechanism

Reason: merely accessing a complex antecedent could take a while

Yes – size matters!• Shorter antecedents yield shorter response times in an end-of-sentence “got

it” task (Murphy 1985)

Short Antecedent: Jimmy swept the tile floor behind the chair

Long Antecedent : Jimmy swept the tile floor behind the chair free of hair and cigarettes.

Ellipsis: Later, his uncle did too.

• Size effect only holds for nearby antecedents. It disappears when distance between antecedent and ellipsis is increased by adding an intervening sentence.

Evidence criticized by Tanenhaus & Carlson (1990) based on poss. confounds

long antecedent: 244ms slower RTs

No – it doesn’t!• Widely cited lack of antecedent size effects in VPE (Frazier & Clifton 2000,

2001)

Short Antecedent: Sarah left her boyfriend last May.Long Antecedent : Sarah got up the courage to leave her boyfriend last May.

Ellipsis: Tina did too.

• F&C conclude cost free copying … what Martin & McElree call non-explanatory

• Although F&C’s paper reports multiple studies, this is the only specific test of the antecedent size effect.

Measure: reading time to final region in self-paced reading – not best practice.Small study: half the size of a regular study ( reduced power), intermittent comprehension questions.No effect? Numerical slowdown in some comparisons (50ms), not reliable.

Speed-Accuracy Tradeoff (SAT)Memory Access

No – it doesn’t!

Version 2 – Martin & McElree 2008

SAT: Possible Outcomes

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

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

VP-complexity manipulation (Expt 3)

The history professor understood Roman mythology …The history professor understood Rome’s swift and brutal destruction of Carthage …

… but the principal was displeased to learn that the overworked students […] did not.… but the principal was displeased to learn that the overly worn books […] did not

Martin & McElree 2008

Antecedent distance effect (Expt 1)

no effect of complexityon dynamics or asymptote

distance affects asymptote,but not dynamics

VP-complexity manipulation (Expt 3)

The history professor understood Roman mythology …The history professor understood Rome’s swift and brutal destruction of Carthage …

… but the principal was displeased to learn that the overworked students […] did not.… but the principal was displeased to learn that the overly worn books […] did not

Martin & McElree 2008

no effect of complexityon dynamics or asymptote

But …

The time course profile measures the sensicality judgment task.Task requires only matching of subject with antecedent verb. Added complexity isn’t relevant.

Needed: a version of this study where entire VP is task relevant.

Cautionary Note from Wh-studies• Larger antecedents sometimes correspond to shorter reading times at foot of

wh-dependency (Hofmeister 2007)

What did the reporter that Scooter avoided discuss …Which poll did the reporter that Scooter avoided discuss …Which political poll did the reporter that Scooter avoided discuss …

It was a communist who the members of the club banned …It was an alleged communist who the members of the club banned …It was an alleged Venezuelan communist who the members of the club banned …

• Hofmeister attributes effects to elaborationor depth of encoding in memory.

Moral: bigger antecedents aren’t all harder


semantic/discourse

syntactic(roughly)


pointer/anaphor

detailedstructure



spell-out

throughout

Baltin MerchantLi


Conclusions1. Psycholinguists are helping

with the overgeneration problem that syntactic theories of ellipsis face.

2. Many interesting findings about rapid access to information in ellipsis resolution.But this is different than showing structure in the ellipsis site.

3. No experiments yet resolve the differences between today’s speakers.

A linking hypothesis from syntactic derivations to real-time computations would be a good start.

Thanks to …• NSF DGE-0801465 IGERT training program in language science

NSF BCS-0948554, Structure Generation in Language Comprehension

• Masaya Yoshida, Matt Wagers, Roumi Pancheva for filling many holes in our knowledge

Annotated Bibliography of Experimental Studies on Ellipsis

Bibliography - EllipsisArregui, A., Clifton Jr., C. L. Frazier & K. Moulton. (2006). Processing elided VPs with flawed antecedents. Journal of Memory & Language 55:232–246. [Uses acceptability and self-paced reading studies to measure mismatch effects. Argues for a repair strategy to create an appropriate antecedent in cases of syntactic mismatch]

Frazier, L. & Clifton Jr., C. (1998). Comprehension of sluiced sentences. Language and Cognitive Processes 13: 499-520. [Investigates a variety of factors that affect the processing of sluiced sentences (e.g. focus, overt/covert antecedents) using self-paced reading and eye-tracking measures.]

Frazier, L., & Clifton Jr., C. (2001). Parsing coordinates and ellipsis: Copy alpha. Syntax: 4(1), 1–22. [Reports a self-paced reading study to measure antecedent size effects. Evidence for the absence of size effects, and argues for a “Cost-free” copy mechanism. That study is reported in (somewhat) more detail in Frazier et al. 2000, J. Psycholing. Res.]

Frazier, L. & Clifton Jr., C. (2005). The syntax-discourse divide: Processing ellipsis. Syntax: 8(1), 121–174. [Acceptability and self-paced reading studies testing for structure at ellipsis site. Argues for structure at the ellipsis site. See also Frazier & Clifton (2006)]

Bibliography - EllipsisKaan, E., Wijnen, F., & Swaab, T. Y. (2004). Gapping: Electrophysiological evidence for immediate processing of “missing” verbs in sentence comprehension. Brain and Language: 89(3), 584–592. [ERP study investigating the time-course of gapping resolution. N400 effects suggest rapid integration of the elided verb at the gap site.]

Gregory, H. and Lappin, S. (1997). A computational model of ellipsis resolution. In Geert-Jan Kruijff, Glyn V. Morrill, and Richard T. Oehrle, editors, Formal Grammar: proceedings of the conference. [Offers an implemented algorithm to capture syntactic reconstruction in ellipsis]

Kim, C., Kobele, G., Runner, J. & Hale, J. (to appear) The Acceptability Cline in VP Ellipsis. Syntax. [Uses magnitude estimation acceptability judgments to measure syntactic mismatch effects. Outlines a computational model in which mismatching VPE effects result from violating parsing heuristics]

Kim, C. & Runner, J. (in press). Discourse parallelism and VP ellipsis. UMass Occasional Papers in Linguistics: Ellipsis. [Several magnitude estimation studies investigating mismatch effects in VPE]

Bibliography - EllipsisMartin, A. E., & McElree, B. (2008). A content-addressable pointer mechanism underlies comprehension of verb-phrase ellipsis. Journal of Memory and Language: 58, 879–906. [Several speed-accuracy tradeoff experiments investigating the mechanisms of antecedent access and retrieval. Argues for a pointer mechanism]

Martin, A. E., & McElree, B. (2009). Memory operations that support language comprehension: Evidence from verb-phrase ellipsis. Journal of Experimental Psychology: Learning Memory & Cognition: 35, 1231-1239. [Speed-accuracy tradeoff experiment investigating the mechanisms of retrieval. Measures the effects of proactive and retroactive interference using ellipsis constructions]

Murphy, G. (1985). Processes of understanding anaphora. Journal of Memory and Language: 24:290–303. [Uses a “Got It” test of comprehension to measure size and distance effects. Shows that size matters for nearby antecedents, but not distant antecedents]

Poirier, J., Wolfinger, K., Spellman, L. & Shapiro, L. (2010). The Real-Time Processing of Sluiced Sentences. Journal of Psycholinguistic Research. 39:411–427. [Cross-modal lexical decision study investigating antecedent access in sluicing. Shows that the antecedent object NP is accessed, but not the subject NP]

Bibliography - EllipsisShapiro, L. P., Hestvik, A., Lesan, L., & Garcia, A. R. (2003). Charting the time-course of VP-ellipsis sentence comprehension: Evidence for an initial and independent structural analysis. Journal of Memory and Language: 49(1), 1–19. [Cross-modal lexical decision study investigating antecedent access in VPE. Shows that the syntactic VP antecedent is accessed, and argues for syntactic reconstruction in ellipsis. See also Shapiro & Hestvik (1995)]

Snider, N. & Runner, J. (2010). "Structural Parallelism Aids Ellipsis and Anaphor Resolution: Evidence from Eye Movements to Semantic and Phonological Neighbors," 16th Annual Conference on Architectures and Mechanisms for Language Processing, York, UK. [Slides retrieved from Snider’s home page. Examines antecedent retrieval using visual world eye-tracking. Argues that semantic and phonological neighbors are activated as a consequence of antecedent retrieval in VPE. Argues for structure at ellipsis site]

Bibliography - EllipsisStreb, J., E. Hennighausen. and Rosler, F. (2004). Different anaphoric expressions are investigated by event-related brain potentials, Journal of Psycholinguistic Research: 33(3), 175-201. [ERP study investigating distance effects. Shows that comprehension times increase with distance to the antecedent. Also, a LAN-like effect suggests that ellipsis is resolved during syntactic parsing steps]

Tanenhaus, M. K., & Carlson, G. N. (1990). Comprehension of deep and surface verb phrase anaphors. Language and Cognitive Processes, 5(4), 257–280. [Uses sensicality judgments to investigate the role of syntactic parallelism in the comprehension of deep and surface anaphors. Surface anaphors make sense more often in syntactically parallel contexts than in non-parallel contexts. Parallelism does not affect judgments of deep anaphors]

Bibliography - EllipsisMasaya Yoshida, Michael Walsh Dickey & Patrick Sturt, (to appear, 2010). Predictive Processing of Syntactic Structure: Sluicing and Ellipsis in Real-Time Sentence Processing. Language and Cognitive Processes. [Uses self-paced reading measures to investigate the prediction of syntactic structure in potentially sluiced constructions. Argues that parser chooses sluicing over other possible structures when possible. Argues for syntactic structure at the ellipsis site]

Masaya Yoshida, Jiyeon Lee, Isaac Rottman and Michael Dickey. (to appear). Islands under the predicted structure. In J. Sprouse & N. Hornstein (eds.), edited volume on syntax and psycholinguistics of islands. [Self-paced reading study investigating the processing of potential sluicing/sprouting structures. Argues that the parser posits the structure of ellipsis when an embedded wh-phrase is processed, based on presence/absence of a reading time cost associated with sprouting.]

Other works citedDalrymple, M., Stuart, M., Shieber, S. & Pereira, F. (1991). Ellipsis and higher-order unification.

Linguistics and Philosophy: 14: 399–452.Gibson, E. & Hickok, G. (1993). Sentence processing with empty categories. Language and

Cognitive Processes: 8(2): 147-161.Kempen, G. (2011): Nontransformational reinterpretation of the purported psycholinguistic

evidence for grammatical movement operations and movement traces. Presentation at the 85th Annual Meeting of the Linguistic Society of America. Jan 6-9. Pittsburgh, PA.

Hardt, D. (1993). Verb Phrase ellipsis: Form, meaning, and processing. Ph.D. Dissertation, University of Pennsylvania.

Hofmeister, P. (2007). Representational Complexity and Memory Retrieval in Language Comprehension. Ph.D. Dissertation, Stanford University.

Lago, S., Chow, W. Y., & Phillips, C. (in prep.). Word frequency affects pronouns and antecedents identically: Distributional evidence. U of Maryland.

McKoon, G., Ratcliff, R., & Ward, G. (1994). Testing theories of language processing: An empirical investigation of the on-line lexical decision task. Journal of Experimental Psychology: Learning, Memory, and Cognition: 20(5): 1219-1228.

Nicol, J. & Swinney, D.A. (1989). The role of structure in coreference assignment during sentence comprehension. Journal of Psycholinguistic Research, Special Issue: Sentence Processing: 18: 5-19.

Nicol, J., Fodor, J.D., & Swinney, D. (1994). Using Cross-Modal lexical decision tasks to investigate sentence processing. Journal of Experimental Psychology: Learninq, Memory, and Cognition: 20(5): 1229-1238.

Other works citedOmaki, A., Trock, A., Wagers, M., Lidz, J., Phillips, C. (2009). Active gap search in the visual world

with lexical competitors. CUNY Sentence Processing Conference, 22, Davis, CA.Phillips, C., & Wagers, M. (2007). Relating Structure and Time in Linguistics and Psycholinguistics.

In G. Gaskell, ed. Oxford Handbook of Psycholinguistics. Oxford University Press.Phillips, C. & Lewis, S. (2010). Derivational Order in Syntax: Evidence and Architectural

Consequences. In C. Chesi, ed. Directions in Derivations. Elsevier.Schmitt, B. Meyer, A., & Levelt, W.J.M., (1999). Lexical access in the production of pronouns.

Cognition: 69: 313-335.Swinney, D., Ford, M., Frauenfelder, U., & Bresnah, J. (1988). On the temporal course of gap-filling

and antecedent assignment during sentence comprehension. In B. Grosz, R. Kaplan. M. Macken. & 1. Sag, eds. Language structure and processing. Stanford, CA: CSLI.

Tanenhaus, M.K., Carlson, G. & Seidenberg, M.S. (1985). Do listeners compute linguistic representations? In D. Dowty, L. Kartunnen & A. Zwicky, eds., Natural Language Parsing: Psychological, Computational, and Theoretical Perspectives. Cambridge University Press.

Townsend, D. J., and Bever, T. G. (2001). Sentence Comprehension: The Integration of Habits and Rules. MIT Press.

Van Gompel, R.P.G., & Majid, A. (2004). Antecedent frequency effects during the processing of pronouns. Cognition: 90: 255-264.

Documents

The Psycholinguistics of Ellipsis Colin Phillips & Dan Parker Department of Linguistics Neuroscience & Cognitive Science Program University of Maryland