Login

Publications  •  Project Statistics

Glossary  •  Schools  •  Disciplines
People Search: 
   
Title/Abstract Search: 

Dissertation Information for Nina Wacholder

NAME:
- Nina Wacholder

DEGREE:
- Ph.D.

DISCIPLINE:
- Linguistics

SCHOOL:
- City University of New York (USA) (1995)

ADVISORS:
- None

COMMITTEE MEMBERS:
- None

MPACT Status: Incomplete - Not_Inspected

Title: Acquiring syntactic generalizations from positive evidence: An HPSG model

Abstract: "This dissertation investigates how children can acquire, from positive evidence only, a correct grammar which does not miss significant generalizations or license ungrammatical constructions. A learning model based on the GPSG prototype of Fodor (1992) is adapted to HPSG. Fodor showed that correct, compact, language-particular syntactic generalizations can be acquired from input tree structures, if the learning mechanism strips out of syntactic trees all universally determined feature values, including, crucially, default values. However, HPSG differs from GPSG in that it assumes that all language-particular information is lexical rather than syntactic, and it renounces syntactic defaults because of their computational intractability. Two questions are therefore addressed: (i) Can apparently syntactic phenomena be acquired by lexical learning? and (ii) How can the work of syntactic defaults be done in HPSG?

These issues are illustrated using complementizers as the initial working example. A complementizer has a lexical feature in which selectional information is stored, so that properties of the clauses types with which it co-occurs can be acquired by lexical learning. A lexical entry is acquired when the feature specifications on a lexical node in an input tree are copied into the lexicon for storage. However, unless lexical entries are compacted by eliminating feature values, they would be costly to store, redundant, and linguistically unsatisfactory. It is argued here that eliminating only fully predictable features is insufficient. Default values must also be factored out of lexical entries. What is stored is then maximally compact feature descriptions consisting only of marked feature values. Default specifications are thus needed in UG, but since in this model they are restricted to the lexicon, they do not create the computational problems that unrestricted defaults can engender.

Evidence is presented that lexical interrogatives (e.g., who) and phrasal interrogatives (e.g., to which panda) also select properties of their sister node. It is shown that even in the phrasal case, these selectional properties can be stored in the lexicon and acquired by lexical learning. The question is then raised whether other types of apparently syntactic selection (as between adjunct and matrix clauses) can similarly be re-cast as lexical."

MPACT Scores for Nina Wacholder

A = 1
C = 6
A+C = 7
T = 1
G = 1
W = 1
TD = 1
TA = 0
calculated 2012-07-31 16:41:34

Advisors and Advisees Graph

generating graph, please reload

Students under Nina Wacholder

ADVISEES:
- Liu Ying-Hsang - Rutgers University (2009)

COMMITTEESHIPS:
- Yang-Woo Kim - Rutgers University (2004)
- Ying Sun - Rutgers University (2005)
- Robert John Rittman - Rutgers University (2007)
- Catherine L. Smith - Rutgers University (2009)
- Heather Lea Moulaison - Rutgers University (2010)
- Catherine L Smith - Rutgers University (2010)