1 / 18

Tanmoy Bhattacharya Nguyen Chi Duy Khuong tanmoy1@gmail khuongvn2000@yahoo

MODELLING OF SHALLOW PARSING OF INDIAN LANGUAGES IIT MUMBAI April 2-4, 2006 Identification of Relative Clause as a Nominal Dependency Relation. Tanmoy Bhattacharya Nguyen Chi Duy Khuong tanmoy1@gmail.com khuongvn2000@yahoo.com Department of Linguistics University of Delhi. The Problem.

shakti
Download Presentation

Tanmoy Bhattacharya Nguyen Chi Duy Khuong tanmoy1@gmail khuongvn2000@yahoo

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. MODELLING OF SHALLOW PARSING OF INDIAN LANGUAGESIIT MUMBAIApril 2-4, 2006Identification of Relative Clause as a Nominal Dependency Relation Tanmoy Bhattacharya Nguyen Chi Duy Khuong tanmoy1@gmail.comkhuongvn2000@yahoo.com Department of Linguistics University of Delhi MSPIL IIT Mumbai

  2. The Problem • Implementable formalisms (e.g. HPSG) have two problems with adjuncts: • Identifying the adjunct (RECOGNITION) • Determining the place the adjunct belongs (ADDRESSING TECHNIQUE) • Solution using a “selectional” RC theory within Principles & Parameters framework MSPIL IIT Mumbai

  3. The Solution in Brief Both the RECOGNITION and the ADDRESSING problems can be bypassed if there is no adjunct to start with: 3 The [flower that John bought] selectional relation MSPIL IIT Mumbai

  4. The Problem in GeneralFunctor versus Argument The ARGUMENT view: Head XP COMPLEMENT The FUNCTOR view: Head XP ADJUNCT Head-Complement idea is conducive to HPSG as it is easier to see complements as semantic arguments of their heads MSPIL IIT Mumbai

  5. The Argument View • Can explain diagnostics: • (i) Semantic Constancy: a. Sharma sleeps/ snores/ laughs in the seminar b. Sharma depends/ relies on a mouse • (ii) Iterability: Sharma opened the drawer with a key, with a hammer • (iii) Order: The police blamed the riot on the residentswithout checking the facts MSPIL IIT Mumbai

  6. The Functor View • Cannot be imported to HPSG easily • SUBCAT cannot handle it since: • Different adjuncts attach to different heads • Not a Head-Filler semantics • Adjunct relation is syntactically different since: • Broader range of categories modified • Number of adjuncts is not pre-fixed How do we handle it then? MSPIL IIT Mumbai

  7. Classic HPSG: The duality of representation • Pollard and Sag (1987): adjunct-main clause dependency determined by rules of grammar; specification neither on N or the RC • Assumes Type hierarchy • Rule of RC: HEAD MAJ N HEAD-DTR|SYN|LOC NFORM NORM DTRS LEX __ ADJ-DTR|SYN|RELCLAUSE MSPIL IIT Mumbai

  8. HPSG Structure for an RC PHON solution that I like HEAD 1 SYN | LOC SUBCAT 2 LEX _ PHON solution MAJ N HEAD-DTR HEAD 1 NFORM NORM DTRS SYN|LOC SUBCAT 2 DET LEX _ PHON that I like ADJ-DTR SYN RELCLAUSE MSPIL IIT Mumbai

  9. Problems with the Classic HPSG • A large number of such rules required • The solution rests on hierarchy of types/ subtypes (head-structure and head-adjunct-structure respectively) Phrase qo CLAUSALITY HEADEDNESS qo qo Clause non-clause hd-ph non-hd-ph • tu .. … rel-cl hd-adj-ph MSPIL IIT Mumbai

  10. Unification • Desirable: Unification within the rule • BUT, who is selecting whom? (Functor/ Argument problem) • Functor alternative is attractive but not formalizable • So, modified heads “selecting” adjuncts (Kayne) • Every common N bearing head feature ADJUNCT • a head-adjunct rule: HEAD-DTR|SYN|LOC HEAD|ADJUNCT{..[1]..} DTRS LEX __ ADJ-DTR|SYN [1] MSPIL IIT Mumbai

  11. HPSG schemata for solution that I like PHON solution that I like HEAD 1 SYN | LOC SUBCAT 2 LEX _ PHON solution MAJ N HEAD-DTR HEAD 1 NFORM NORM ADJUNCTS <..3..> DTRS SYN|LOC SUBCAT 2 DET LEX _ PHON that I like ADJ-DTR SYN 3 RELCLAUSE MSPIL IIT Mumbai

  12. Revised HPSG Highest V rather than a null relativizer heads the clause: S [MOD N’] qo NP VP [MOD N’] wi NP V [MOD N’] John-I chayk-ul neh-un -nom book-acc put-rel ‘The book that John put’ Sag (1997) MSPIL IIT Mumbai

  13. Highest V in RC need not show RC morphology Watashi ga inu ga taberu ring-o miru I NOM dog NOM eats apple-ACC see ‘I see the apple which the dog eats.’ IP 3 watashi ga I’ 3 VP I 3 NP V 3 miru CP NP 6 inuga taberu ringo • Similarly with Vietnamese MSPIL IIT Mumbai

  14. Sag (1997) Solution who (relative) CAT NP CONT [INDEX 3] REL { 3 } QUE { } • The head words have a REL feature • Inheritance of REL is governed by WHIP MSPIL IIT Mumbai

  15. Drawbacks of Revised HPSG • Hierarchy of types as before • Wh-relatives are wh-rel-cl and subject to a separate constraint • Subject Wh-relatives belong to yet another type and subject to yet another constraint • Non-subject relatives belong to another subtype and subject to another constraint • And so on … MSPIL IIT Mumbai

  16. Semantics of the Gap Hunter (2004) I eat the apples which the men bought. MSPIL IIT Mumbai

  17. Linking the two clauses MSPIL IIT Mumbai

  18. Back to the Duality of Representation • “CP generated from the semantics of the RC is added as an adjunct to the NP.” • Doesn’t address either the RECOGNITION or the ADDRESSING problem • One feature for the RC (relativewh) and one for the MC (subclause0) Can we bypass this problem? MSPIL IIT Mumbai

More Related