CSC2540S Machine Learning and Universal Grammar

 

Department of Computer Science, University of Toronto

Tuesday 2-4 PM, SF3207, Winter Semester, 2009

 

Shalom Lappin

Department of Philosophy

King's College London

shalom.lappin@kcl.ac.uk

shalom@cs.toronto.edu

 

 

 

Please note that the class has moved from MS3268 to SF3207

 

Recent advances in the application of machine learning to grammar induction and related

NLP tasks have shown that it is possible to acquire a significant amount of linguistic

knowledge through domain general learning algorithms constrained by relatively weak

biased language models. Computational learning theory has also provided important

results for understanding the limits and possibilities of different sorts of learning

algorithms for acquiring the grammars of formal and natural languages. The seminar

considers the implications of this work for human language acquisition and the theory

of universal grammar, the language model that defines the space of possible hypotheses

for natural language grammars. While the seminar focuses on research in machine

learning and computational learning theory, it will also look at some current work in

theoretical linguistics, psycholinguistics, cognitive science, and genetics in exploring

these issues.

 

                                                                    References

 

 

 

Week 1: January 6, 2009, Structure of the Course

    Overview of the problem,

    The argument from the poverty of stimulus (APS) and Universal Grammar

 

Week 2: January 13, Cognitive Architecture and Language Evolution

    Nativism and cognitive architecture,

    The evolution of language

 

Week 3: January 20, Clarifying the APS

     A critical examination of some linguistic arguments for the APS

 

Week 4: January 27, The Nature of Primary Linguistic Evidence,

    The role of negative evidence in language acquisition

 

Week 5: February 3, Identification in the Limit

    The Gold model of grammar induction

 

Week 6: February 10, PAC Learning

    An alternative computational learning paradigm

 

Week 7: February 17, Reading Week

 

Week 8: February 24, Modifying PAC Learning with Distributions

    Using probability distributions on learning samples to facilitate grammar induction

 

Week 9: March 3, Machine Learning and Grammar Induction

    Unsupervised grammar induction,

    Recent work on acquiring grammars through the application of machine learning

      methods to large natural language corpora

 

Week 10: March 10, Parameters in Linguistic Theory and in Probabilistic Language Models

     Parametric explanations of language universals and variation in linguistic theory,

     A bootstrapping approach to parameter estimation for language models

 

Week 11: March 17, Genetic and Psycholinguistic Evidence

    The FOXP2 gene and specific language impairment,

    Psychological evidence for Bayesian grammar induction

 

Weeks 12-14: March 24, March 31, and April 7, Project Presentations