AGAINST USAGE-BASED APPROACHES TO...

AGAINST USAGE-BASED APPROACHES TO RECURSION: THE GRAMMAR-PERFORMANCE

DISTINCTION IN A BIOLINGUISTIC PERSPECTIVE �

Andreas Trotzke & Markus Bader�

GLOW 36�Lund University, April 2, 2013�

1

THE GRAMMAR-PERFORMANCE DISTINCTION �

… in a biolinguistic perspective !“To study actual linguistic performance, we must consider the interaction of a variety of factors, of which the underlying competence of the speaker-hearer is only one. […] 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). […] in the technical sense, linguistic theory is mentalistic, since it is concerned with discovering a mental reality underlying actual behavior.” (Chomsky 1965: 4) !

… in a usage-based perspective !“[…] grammar is the cognitive organization of experience with language. [...] Grammar cannot be thought of as pure abstract structure that underlies language use [...]. Grammar is built up from specific instances of use that marry lexical items with constructions; it is routinized and entrenched by repetition and schematized by the categorization of exemplars.” (Bybee 2006: 730)�

2


A prominent argument in favor of the grammar-performance distinction: �

Multiple center-embedding cannot successfully be interpreted despite being generable by the grammar.�

(1) The rat the cat the dog chased killed ate the malt. "(Chomsky & Miller 1963: 286) �

Claim: Such sentences are quite incomprehensible, but this has no bearing on the possibility of generating them on the level of grammar.�

Chomsky’s (1963: 327) analogy: “the inability of a person to multiply 18,674 times 26,521 in his head is no indication that he has failed to grasp the rules of multiplication.” �

3


However, usage-based approaches do not accept this line of reasoning.�

“[…] from the point of view of modeling psychological processes, we need not take the purported unbounded recursive structure of natural language as axiomatic. Nor need we take for granted the suggestion that a speaker/hearer’s knowledge of language captures such infinite recursive structure. Rather, the view that ‘unspeakable’ sentences which accord with recursive rules form a part of the knowledge of language is an assumption of the standard view of language developed by Chomsky.” (Christiansen & Chater 1999: 158)�

We took up this issue by focusing on certain properties of the processing of multiple center-embeddings, thereby investigating some of the relevant ‘psychological processes’ Christiansen & Chater (1999) allude to. �

4

PROCESSING GRAMMATICAL AND UNGRAMMATICAL CENTER-EMBEDDINGS " "!

(Frazier 1985: 178)�(2) "a. "The patient the nurse the clinic had hired sent to the doctor met "

" "Jack.�

"b. * "The patient the nurse the clinic had hired __ met Jack.�

Multiple center-embeddings missing the second VP are often accepted as grammatical.�

This missing-VP effect is experimentally confirmed for English (Gibson & Thomas 1999; Vasishth et al. 2010) and for French (Gimenes et al. 2009). For English, Christiansen & MacDonald (2009) even found that cases like (2b) are rated better than cases such as (2a).�

5


However, for German, Vasishth et al. (2010) claim that the missing-VP effect �does not exist.�

Vasishth et al. (2010: 550)�(3) "a. "Der "Anwalt, "den der Zeuge, den der Spion betrachtete, schnitt, überzeugte �

" "the "lawyer "who " the witness who the spy looked-at " cut " convinced�" "den Richter.�" "the "judge �"b. * Der Anwalt, den der Zeuge, den der Spion betrachtete, __ überzeugte �" "the "lawyer who the witness who the spy looked-at convinced�" "den Richter.�" "the "judge �

6


Results from Vasishth et al.‘s (2010) self-paced reading study:�

English " " " " " " " " "German�

7

Reading time

The mean reading times and 95% confidence intervals are summarised inFigure 2 and the statistical analyses in Table 1. The VP-forgetting hypothesismakes predictions about the V1 and post-V1 region. We therefore presentanalyses for these regions in Table 1. Here and in all subsequent analyses,the grammaticality factor was coded as 1 for grammatical sentences, and!1 for ungrammatical sentences. This coding aids in the interpretation of theestimated coefficients; a positive sign indicates slower reading time forthe grammatical condition, whereas a negative sign indicates faster readingtime.

As predicted by Gibson and Thomas’ VP-forgetting hypothesis andthe DLT, in the post-V1 region the ungrammatical condition was significantlyeasier to process than the grammatical; in the V1 region the differencebetween the conditions had the predicted pattern but did not reachsignificance.

Figure 2. Mean reading times and 95% confidence intervals for the three-verb regions and thepost-V1 word in the English self-paced reading study (Experiment 1). The figure shows readingtimes with respect to the grammaticality manipulation.

TABLE 1Coefficients, standard errors and t-scores for the grammaticality effect in the Englishself-paced reading Experiment 1. The grammatical condition was coded as 1 and the

ungrammatical condition as "1

Region Coefficient SE t-score HPD.lower HPD.upper

V1 0.007273 0.017104 0.43 !0.02693120 0.04134948Post-V1 0.141575 0.014559 9.72* 0.11161532 0.16940903

*Statistically significant at a#0.05.

SHORT-TERM FORGETTING 543

Dow

nloa

ded

by [U

nive

rsity

of K

onst

anz]

at 0

2:01

22

Febr

uary

201

3

Results

Question!response accuracy

Question!response accuracy (for all items, including distractors) was79.51%. The mean accuracy for the grammatical condition was 65%, whichwas significantly lower than in the ungrammatical condition, 71.5%(coefficient "0.19423, SE 0.09953, z#"1.952, p#0.05).

Reading time

The mean reading times and 95% confidence intervals at the verb regionsare summarised in Figure 5 and the results of the mixed-effects modelanalysis are shown in Table 4.

The reading-time analyses show that, contrary to the predictions of DLTand the VP-forgetting account, at the matrix verb V1 as well as the post-V1word, the ungrammatical condition was read slower than the grammaticalone. This is the opposite of the result found for English in Experiments 1, 1aand 2.

Figure 5. Mean reading times and 95% confidence intervals for the critical regions in theGerman self-paced reading study (Experiment 3). The figure shows reading times with respect tothe grammaticality manipulation.

TABLE 4Coefficients, standard errors and t-scores for the grammaticality effect in the German

self-paced reading Experiment 3

Region Coefficient SE t-score HPD.lower HPD.upper

V1 "0.067315 0.015935 "4.22 !0.097603787 "0.03427237Post-V1 "0.070175 0.016191 "4.33 !0.10162431 "0.03635828

*Statistically significant at a#0.05.

SHORT-TERM FORGETTING 551D

ownl

oade

d by

[Uni

vers

ity o

f Kon

stan

z] a

t 02:

01 2

2 Fe

brua

ry 2

013


A prominent explanation for the missing-VP effect �

The working-memory overload account:!Forgetting the prediction of the second VP reduces working-memory load.�The comprehender’s acceptance of (or even preference for) the ungrammatical structure is due to memory overload induced by distance (Gibson & Thomas 1999; Gibson 2000).�

However, language-independent working-memory constraints seem to offer no explanation for the purported absence of the missing-VP effect in German.�

8


A usage-based explanation for the missing-VP effect �

The experience-based account:�“[…] constraints on recursive regularities do not follow from extrinsic limitations on memory or processing; rather they arise from interactions between linguistic experience and architectural constraints on learning and processing […], intrinsic to the system in which the knowledge of grammatical regularities is embedded.” (Christiansen & MacDonald 2009: 127)�

Claim: Due to the head-finality of German relative clauses, German readers have a stronger expectation of a VP. The missing-VP effect is thus due to experience with language-specific structures.�

9

THE MISSING-VP EFFECT IN GERMAN �

The results of Vasishth et al. (2010) and the claim by Christiansen &�MacDonald (2009) contrast with our own experimental results (Bader,�Bayer & Häussler 2003).�

However, Bader, Bayer & Häussler (2003) used different sentence structures�(4) and a different experimental procedure (end-of-sentence speeded grammaticality judgments; cf. Warner & Glass 1987). �

(4) "Heute morgen ist das Programm, das den Programmierer, der die �"this " morning is the program which the programer " who the �"Dokumentation völlig" "ohne irgendeine Hilfe erstellen musste, abgestürzt. "documentation completely "without any " " help write " had-to " crashed�

We therefore run a new experiment with the same procedure as�Christiansen & MacDonald (2009) and the same sentence structures as�Vasishth et al. (2010). �

10


Experiment�

Procedure:! Word-by-word self-paced reading with mid-screen presentation � Word-by-word grammaticality judgments (go – no-go)� End-of sentence rating task on a scale from 1 (very good) to 7 (very bad)�

Materials: !10 sentences in two versions each: complete or missing-VP !

Sentences with doubly center-embedded relative clauses in SpecCP�(5) a. Der Chef, den der Angestellte, den die Sekretärin liebt, betrügt, betritt den Raum.!

" the boss who the criminal who the secretary likes betrays enters the room!

b. Der Chef, den der Angestellte, den die Sekretärin innig liebt, __ betritt den Raum.! "the boss who the criminal who the secretary devotedly likes " enters the room!

11


Results:!

12


Results:!

Mean end-of-sentence ratings�(1=very good, 7=very bad) �

13

COMPARISON OF ENGLISH AND GERMAN �

English " " " " " " " "German�

14

Results:" "The rejection rates are substantially higher for English than for German." "Missing-VP cases in German were not rejected significantly more than

complete center-embeddings."

(Christiansen & MacDonald 2009: 148)


Grammatical prediction error (GPE; cf. Christiansen & Chater 1999): � measure of processing difficulty used by simple recurrent networks� ranges from 0 (very easy) to 1 (very difficult)�

" " " " " " " " " "V1 " " " " " " " "" "

"High predictability means low GPEs.�"For German, V1 is predictable. Thus,�"the missing-VP effect, according to�"these connectionist models, should�"not exist.�

"In contrast, we found that the missing-�"VP effect in German does exist �"(although processing center-embeddings�"in general seems to be easier in German than in English). !

15


Interim Conclusions:!

"The missing-VP effect appears in both English and German.�"Thus, it is not a reaction to the particular configuration created by a "language-specific word order.�

#Language-independent working-memory constraints seem to offer an "adequate explanation for the missing-VP effect; usage-based models "do not provide us with a full picture of processing center-embeddings.�

#Since constraints on multiple center-embedding can thus be explained "by extrinsic limitations on processing (cf. Miller & Chomsky 1963), "there is no reason to question the grammar-performance distinction. �

16

THE PERFORMANCE INTERFACE IN LANGUAGE DESIGN �

According to Chomsky (2005: 6), three factors have to be explored when one aims at an explanation of language design: �

The genetic endowment (= UG) � Linguistic experience � Principles not specific to the language faculty �

Third factor: �“[...] (a) principles of data analysis that might be used in language acquisition and other domains; (b) principles of structural architecture and developmental constraints [...] including principles of efficient computation.” Chomsky (2005: 6)� “[…] properties of the human brain that determine what cognitive systems can exist.” (Chomsky 2007: 3, fn. 4)

We argue in Trotzke, Bader & Frazier (2013) that systematic properties of performance systems provide third-factor explanations for crucial design features of human language. 17


The systematicity of the language processing system�

The Missing-VP effect also occurs in language production.�

Corpus study: deWaC corpus (cf. Baroni et al. 2009) ! Corpus of written German built by web crawling ! 1.600.000.000 tokens of text in roughly 92.000.000 sentences� Size enhances the chance of obtaining quantitative information even for� low-frequency constructions like multiple center-embeddings. !

Result: !In the case of doubly center-embedded relative clauses contained within an embedded clause, VP2 was missing in about 15% of all cases.�

18


The systematicity of the language processing system�

Missing-VP example !(6) "Dieser Typ entsteht, wenn lin-3 oder ein Gen, das für die Induktion, die von der Ankerzelle �

"this " type emerges when lin-3 or a gene that for the induction that from the anchor-cell�"ausgeht, __ mutiert ist.�

originates mutated is !"‘This type emerges when lin-3 or a gene that for the induction that originates�"from the anchor cell ___ has mutated.’�"(Dewac-1/95201, http://www.zum.de)�

"The VP of the lower relative clause was almost never omitted (0,2%).� "The VP of the superordinate clause was missing only in a small number of cases

"(2,2%).�

"In accordance with Frazier (1985: 178-179), the missing-VP effect only arises�"when the VP of the highest relative clause is missing. �

19

CONCLUSION �

"The regularities involved in the missing-VP effect are language-"independent, since they are not a reaction to language-specific word "order; there are no dramatically different processing systems in "different languages.�

"Since we presented converging data from language comprehension "and language production, the extrinsic constraints on recursive center-"embedding "are deep and systematic.�

#The performance systems, beside the grammar, constitute a distinct "cognitive component of biolinguistic inquiry, since the constraints they "impose on the grammar are not random but characteristic. Therefore, "systematic properties of performance systems belong to the domain of "third-factor effects.

20

THANK YOU! �

21

REFERENCES�

Bader, Markus, Josef Bayer & Jana Häussler. 2003. Explorations of center-embedding and "missing VPs. Poster presented at the 16th CUNY Conference on Sentence Processing, MIT, "Cambridge, MA. [27–29 March 2003] �

Baroni, Marco, Silvia Bernardini, Adriano Ferraresi & Eros Zanchetta. 2009. TheWaCkyWide "Web: A collection of very large linguistically processed web-crawled corpora. Language "Resources and Evaluation Journal 23, 209–226.

Bybee, Joan. 2006. From usage to grammar: The mind’s response to repetition. Language 82, "711–733.�

Chomsky, Noam. 1963. Formal properties of grammars. In R. Duncan Luce, Robert R. Bush "& "Eugene Galanter (eds.), Handbook of Mathematical Psychology: Vol. 2, 323–418. New York: "Wiley. �

Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT Press.�Chomsky, Noam. 2005. Three factors in language design. Linguistic Inquiry 36, 1–22. �Chomsky, Noam. 2007. Approaching UG from below. In Uli Sauerland & Hans-Martin

"Gärtner (eds.), Interfaces + Recursion = Language? Chomsky’s Minimalism and the View "from Syntax–Semantics, 1–29. Berlin: Mouton de Gruyter. �

Chomsky, Noam & George A. Miller. 1963. Introduction to the formal analysis of natural "languages. In R. Duncan Luce, Robert R. Bush & Eugene Galanter (eds.), Handbook of "Mathematical Psychology: Vol. 2, 269–321. New York: Wiley. �

22

REFERENCES�

Christiansen, Morton H. & Nick Chater. 1999. Toward a connectionist model of recursion in "human linguistic performance. Cognitive Science 23, 157–205.�

Christiansen, Morten H. & Maryellen C. MacDonald. 2009. A usage-based approach to "recursion in sentence processing. Language Learning 59, 126–161.�

Engelmann, Felix & Shravan Vasishth. 2009. Processing grammatical and ungrammatical "center embeddings in English and German: A computational model. Proceedings of 9th "International Conference on Cognitive Modeling, Manchester, UK. !

Frazier, Lyn. 1985. Syntactic complexity. In David Dowty, Lauri Karttunen & Arnold Zwicky "(eds.), Natural Language Processing: Psychological, Computational, and Theoretical "Perspectives, 129–189. Cambridge: Cambridge University Press.�

Gibson, Edward. 2000. The dependency locality theory: A distance-based theory of linguistic "complexity. In Alec Marantz, Yasushi Miyashita & Wayne O’Neil (eds.), Image, Language, "Brain. Papers from the First Mind Articulation Project Symposium, 95–126. Cambridge, MA: "MIT Press. �

Gibson, Edward & James Thomas. 1999. Memory limitations and structural forgetting: The "perception of complex ungrammatical sentences as grammatical. Language and Cognitive "Processes 14, 225–248. �

Gimenes, Manuel, François Rigalleau & Daniel Gaonac’h. 2009. When a missing verb makes a "French sentence more acceptable. Language and Cognitive Processes 24, 440–449. �

23

REFERENCES�

Miller, George A. & Noam Chomsky. 1963. Finitary models of language users. In R. Duncan "Luce, Robert R. Bush & Eugene Galanter (eds.), Handbook of Mathematical Psychology: Vol. "2, 419–491. New York: Wiley. �

Trotzke, Andreas, Markus Bader & Lyn Frazier. 2013. Third factors and the performance "interface in language design. Biolinguistics 7, 1–34.�

Vasishth, Shravan, Katja Suckow, Richard L. Lewis & Sabine Kern. 2010. Short-term forgetting "in sentence comprehension: Crosslinguistic evidence from verb-final structures. "Language and Cognitive Processes 25, 533–567.�

Warner, John & Arnold L. Glass. 1987. Context and distance-to-disambiguation effects in "ambiguity resolution: Evidence from grammaticality judgements of garden path "sentences. Journal of Memory and Language 26, 714–738.�

24

AGAINST USAGE-BASED APPROACHES TO...

Documents

Transcript of AGAINST USAGE-BASED APPROACHES TO...