1 Introduction phenomenon Introduction phenomenon Methods & - PDF document

Introduction phenomenon Methods & results background Conclusion & afterthoughts hypothesis Sense-contingent lexical bias & Structure of the talk QITL-2 its role for initial parsing University of Osnabrück  Introduction Institute of Cognitive Science decisions June 2006  Phenomenon: Local NP/S-ambiguity  Assumptions and hypothesis  Sense-contingent lexical guidance  Methods and Results  Corpus-based estimation of lexical bias  form-based vs. sense-contingent preferences Daniel Wiechmann  Comparison with experimental data (Hare et al. 2003) Friedrich-Schiller-Universität Jena  Conclusion and afterthoughts FSU Jena Anglistische Sprachwissenschaft: 1 FSU Jena Anglistische Sprachwissenschaft: 2 Sprache und Kognition Sprache und Kognition Introduction phenomenon Introduction phenomenon Methods & results background Methods & results background Conclusion & afterthoughts hypothesis Conclusion & afterthoughts hypothesis local syntactic ambiguity local NP/S-ambiguity  The student knew the  The student knew the solution solution to the problem. was incorrect. Since Jay always jogs a mile … doesn’t seem that long. (clause boundary) (construction DO NP ) (construction DO S )   The horse raced past the barn … fell. (MV/ RR )     The student knew the solution … was wrong. (NP/ S )    t syntactic structure is ambiguous at t FSU Jena Anglistische Sprachwissenschaft: 3 FSU Jena Anglistische Sprachwissenschaft: 4 Sprache und Kognition Sprache und Kognition Introduction phenomenon Introduction phenomenon Methods & results background Methods & results background Conclusion & afterthoughts hypothesis Conclusion & afterthoughts hypothesis local NP/S-ambiguity Some accounts of sentence processing  Structural models  Kimball’s ‘7 Principles’  ‘Garden Path Theory’´ (Frazier, Fodor) Some variants are usually preferred over  Grammar-based models  Theta-Attachment (Pritchett, Abney) possible alternatives.  Argument Attachment (Crocker)  Dependency (Pickering) The question is: Why?  Memory-/Capacity-based (Gibson, Just und Carpenter)  Experience-based  Lexical Guidance (Ford)  Tuning (Mitchell)  Referential Theory (Altman, Steedman)  Completely probabilistic (Jurafsky, Crocker und Brants)  Attractor-based (dynamical systems) (Elman, Tanenhaus, … )  Embodied sentence comprehension (Zwaan, Madden, Bergen and Chang; … ) FSU Jena Anglistische Sprachwissenschaft: 5 FSU Jena Anglistische Sprachwissenschaft: 6 Sprache und Kognition Sprache und Kognition 1

Introduction phenomenon Introduction phenomenon Methods & results background Methods & results background Conclusion & afterthoughts hypothesis Conclusion & afterthoughts hypothesis Some accounts of sentence processing constraints on constructing an interpretation  Phrase-formation constraints serial  Contextual constraints informationally restricted  Plausibility  Referential contexts parallel  Computational resource constraints parsing  Locality (memory cost constraints)  Phrase-level contingent (and other) frequency constraints  Prosodic constraints serial ? informationall  Lexically-based constraints y unrestricted  Grammatical category of perceived element  Subcategorization preferences parallel  ... FSU Jena Anglistische Sprachwissenschaft: 7 FSU Jena Anglistische Sprachwissenschaft: 8 Sprache und Kognition Sprache und Kognition Introduction phenomenon Introduction phenomenon Methods & results background Methods & results background Conclusion & afterthoughts hypothesis Conclusion & afterthoughts hypothesis lexically specific information: constraints on constructing an interpretation SUBCAT preferences  Phrase-formation constraints  Contextual constraints transitive verbs  Plausibility  Referential contexts  Computational resource constraints  Locality (memory cost constraints)  Phrase-level contingent (and other) frequency constraints  Prosodic constraints only sentential only nominal  Lexically-based constraints think,... like,... feel, find, claim, bet, ...  Grammatical category of perceived element  Subcategorization preferences  ... FSU Jena Anglistische Sprachwissenschaft: 9 FSU Jena Anglistische Sprachwissenschaft: 10 Sprache und Kognition Sprache und Kognition Introduction phenomenon Introduction preliminaries Methods & results background Methods & results data Conclusion & afterthoughts hypothesis Conclusion & afterthoughts method lexical guidance hypothesis estimating lexical preferences Means to estimate lexical preferences Parsing preferences are guided by structural expectations resulting from sense-contingent lexically specific preferences . derived from corpora experimentally derived measures of association sentence completion sentence production string frequencies, ... FSU Jena Anglistische Sprachwissenschaft: 11 FSU Jena Anglistische Sprachwissenschaft: 12 Sprache und Kognition Sprache und Kognition 2

Introduction preliminaries Introduction preliminaries Methods & results data Methods & results data Conclusion & afterthoughts methods Conclusion & afterthoughts methods lexical preferences: form vs. sense procedure Estimations of lexical bias on the level of verb-form are problematic  Verbs are (conventionally) used to express a number of semantic relations    Corpus-based estimation of LB both on the level of and each of these have their own preferences verb form and sense Peter VP [ V admitted 1 NP [ his ex-girlfriend ] PP [to the club]].   2. LBs are compared with reading-time latencies in Peter VP [ V admitted 2 S [ NP [ his ex-girlfriend ] was nicer than his current    self-paced moving window experiment one]].  Peter VP [ V admitted 2 NP [ his error ]]    (Hare et al. 2003)  LB Verb(form) ≠ LB Verb(sense)  (cf. Roland et al. 2000: experimental vs. corpus-based norms) FSU Jena Anglistische Sprachwissenschaft: 13 FSU Jena Anglistische Sprachwissenschaft: 14 Sprache und Kognition Sprache und Kognition Introduction data Introduction data Methods & results method Methods & results method Conclusion & afterthoughts results Conclusion & afterthoughts results data data corpus : 17 mio word sample (BNC) Step 2 → All instances of the relevant construction were coded with respect to [compiled as to mirror structural properties of ICE-GB corpus] the grammatical role played by the postverbal NP  the verb sense instantiated by the verb in the matrix clause using a lexical Step 1 : Extraction of all [V NP]-constructions for 20 verbs  database ( WordNet 2.0) [pronominal NPs have been ignored (case)] Step 3 → Lexical bias was calculated both on the level of form and sense Number of tokens per type: N > 3000 : random sample: 10% using Distinctive Collexeme Analysis (DCA) 300 < N < 3000 : random sample: n=300 N < 300 : all tokens ___________ token total : 4960 FSU Jena Anglistische Sprachwissenschaft: 15 FSU Jena Anglistische Sprachwissenschaft: 16 Sprache und Kognition Sprache und Kognition Introduction data Introduction data Methods & results method Methods & results method Conclusion & afterthoughts results Conclusion & afterthoughts results distinctive collexeme analysis (DCA) distinctive collexeme analysis (DCA) contingency contingency table table DCA DCA find find other other verbs verbs row totals row totals  DCA is a variant of ‘ Collostructional Analysis ’ (Stefanowitsch and Gries 2003) NP-complementation NP-complementation observed observed (expected) (expected) observed observed (expected) (expected) R 1 R 1  Assumes a construction grammar (CxG) framework S-complementation observed S-complementation observed (expected) (expected) observed observed (expected) (expected) R 2 R 2  CxG assumes that linguistic knowledge at all levels can be characterized as constructions, column column totals totals C1 C1 C2 C2 N i.e. pairings of form and meaning  Assesses the degree of association between two constructions of arbitrary degrees of 1. Collection of data above specificity 2. Statistical analysis of the observed distribution; → Application of Fisher‘s exact test (FET) as a measure of association  DCA measures the relationship of a lexical construction towards more abstract  Fisher 1922, Pedersen 1996, Gries und Stefanowitsch 2004 for a discussion constructions it can occur in  Evert 2004 for a comprehensive discussion of a wide array of association measures  Lexical preferences are expressed in terms of association scores  → outputs a gradual measure FSU Jena Anglistische Sprachwissenschaft: 17 FSU Jena Anglistische Sprachwissenschaft: 18 Sprache und Kognition Sprache und Kognition 3