Overfitting avoidance for stochastic modeling of attribute-value grammars

Authors:
Tony Mullen;Miles Osborne
Affiliations:
University of Groningen;University of Edinburgh
Venue:
ConLL '00 Proceedings of the 2nd workshop on Learning language in logic and the 4th conference on Computational natural language learning - Volume 7
Year:
2000

Citing 12
Cited 1

Wrappers for feature subset selection

Artificial Intelligence - Special issue on relevance
Inducing Probabilistic Grammars by Bayesian Model Merging

ICGI '94 Proceedings of the Second International Colloquium on Grammatical Inference and Applications
Parsing with Context-Free Grammars and Word Statistics

Parsing with Context-Free Grammars and Word Statistics
Inducing Features of Random Fields

Inducing Features of Random Fields
Maximum entropy models for natural language ambiguity resolution

Maximum entropy models for natural language ambiguity resolution
Building a large annotated corpus of English: the penn treebank

Computational Linguistics - Special issue on using large corpora: II
Stochastic attribute-value grammars

Computational Linguistics
Automatic extraction of subcategorization from corpora

ANLC '97 Proceedings of the fifth conference on Applied natural language processing
A probabilistic corpus-driven model for lexical-functional analysis

COLING '98 Proceedings of the 17th international conference on Computational linguistics - Volume 1
Statistical decision-tree models for parsing

ACL '95 Proceedings of the 33rd annual meeting on Association for Computational Linguistics
A new statistical parser based on bigram lexical dependencies

ACL '96 Proceedings of the 34th annual meeting on Association for Computational Linguistics
Estimation of stochastic attribute-value grammars using an informative sample

COLING '00 Proceedings of the 18th conference on Computational linguistics - Volume 1

Improved iterative scaling can yield multiple globally optimal models with radically differing performance levels

COLING '02 Proceedings of the 19th international conference on Computational linguistics - Volume 1

Quantified Score

Hi-index	0.00

Visualization

Abstract

We present a novel approach to the problem of overfitting in the training of stochastic models for selecting parses generated by attribute-valued grammars. In this approach, statistical features are merged according to the frequency of linguistic elements within the features. The resulting models are more general than the original models, and contain fewer parameters. Empirical results from the task of parse selection suggest that the improvement in performance over repeated iterations of iterative scaling is more reliable with such generalized models than with ungeneralized models.