Nancy Chih-Lin Chang

EECS Department, University of California, Berkeley

Technical Report No. UCB/EECS-2009-24

February 5, 2009

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2009/EECS-2009-24.pdf

Abstract

Constructing grammar: A computational model of the emergence of early constructions

by

Nancy Chih-lin Chang

Doctor of Philosophy in Computer Science University of California, Berkeley

Professor Jerome A. Feldman, Chair

In this thesis I explore and formalize the view that grammar learning is driven by meaningful language use in context. On this view, the goal of a first language learner is to become a better language user—in particular, by acquiring linguistic constructions (structured mappings between form and meaning) that facilitate successful communication. I present a computational model in which all aspects of the language learning problem are reformulated in line with these assumptions. The representational basis of the model is a construction grammar formalism that captures constituent structure and relational constraints, both within and across the domains of form and meaning. This formalism plays a central role in two processes: language understanding, which uses constuctions to interpret utterances in context; and language learning, which seeks to improve comprehension by making judicious changes to the current set of constructions. The resulting integrated model of language structure, use and acquisition provides a cognitively motivated and computationally precise account of how children acquire their earliest multiword constructions. I define a set of operations for proposing new constructions, either to capture contextually available mappings not predicted by the current grammar, or to reorganize existing constructions. Candidate constructions are evaluated using a minimum description length criterion that balances a structural bias toward simpler grammars against a data-driven bias toward more specific grammars. When trained with a corpus of child-directed utterances annotated with situation descriptions, the model gradually acquires the concrete word combinations and item-based constructions that constitute the first steps toward adult language.

Advisors: Jerome A. Feldman


BibTeX citation:

@phdthesis{Chang:EECS-2009-24,
    Author= {Chang, Nancy Chih-Lin},
    Title= {Constructing grammar: A computational model of the emergence of early constructions},
    School= {EECS Department, University of California, Berkeley},
    Year= {2009},
    Month= {Feb},
    Url= {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2009/EECS-2009-24.html},
    Number= {UCB/EECS-2009-24},
    Abstract= {Abstract

Constructing grammar: A computational model of the emergence of early constructions

by

Nancy Chih-lin Chang

Doctor of Philosophy in Computer Science
University of California, Berkeley

Professor Jerome A. Feldman, Chair

In this thesis I explore and formalize the view that grammar learning is driven by meaningful language
use in context. On this view, the goal of a first language learner is to become a better language
user—in particular, by acquiring linguistic constructions (structured mappings between form and
meaning) that facilitate successful communication. I present a computational model in which all
aspects of the language learning problem are reformulated in line with these assumptions. The
representational basis of the model is a construction grammar formalism that captures constituent
structure and relational constraints, both within and across the domains of form and meaning. This
formalism plays a central role in two processes: language understanding, which uses constuctions
to interpret utterances in context; and language learning, which seeks to improve comprehension
by making judicious changes to the current set of constructions.
The resulting integrated model of language structure, use and acquisition provides a cognitively
motivated and computationally precise account of how children acquire their earliest multiword
constructions. I define a set of operations for proposing new constructions, either to capture
contextually available mappings not predicted by the current grammar, or to reorganize existing
constructions. Candidate constructions are evaluated using a minimum description length criterion
that balances a structural bias toward simpler grammars against a data-driven bias toward more
specific grammars. When trained with a corpus of child-directed utterances annotated with situation
descriptions, the model gradually acquires the concrete word combinations and item-based
constructions that constitute the first steps toward adult language.},
}

EndNote citation:

%0 Thesis
%A Chang, Nancy Chih-Lin 
%T Constructing grammar: A computational model of the emergence of early constructions
%I EECS Department, University of California, Berkeley
%D 2009
%8 February 5
%@ UCB/EECS-2009-24
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2009/EECS-2009-24.html
%F Chang:EECS-2009-24