A probabilistic search for the best solution among partially completed candidates

Authors:
Filip Ginter;Aleksandr Mylläri;Tapio Salakoski
Affiliations:
University of Turku, Turku, Finland;University of Turku, Turku, Finland;University of Turku, Turku, Finland
Venue:
CHSLP '06 Proceedings of the Workshop on Computationally Hard Problems and Joint Inference in Speech and Language Processing
Year:
2006

Citing 5
Cited 1

Finite-state parsing and disambiguation

COLING '90 Proceedings of the 13th conference on Computational linguistics - Volume 2
Active Feature-Value Acquisition for Classifier Induction

ICDM '04 Proceedings of the Fourth IEEE International Conference on Data Mining
Combining lexical, syntactic, and semantic features with maximum entropy models for extracting relations

ACLdemo '04 Proceedings of the ACL 2004 on Interactive poster and demonstration sessions
Using semantic relations to refine coreference decisions

HLT '05 Proceedings of the conference on Human Language Technology and Empirical Methods in Natural Language Processing
Regularized least-squares for parse ranking

IDA'05 Proceedings of the 6th international conference on Advances in Intelligent Data Analysis

Regular approximation of link grammar

FinTAL'06 Proceedings of the 5th international conference on Advances in Natural Language Processing

Quantified Score

Hi-index	0.00

Visualization

Abstract

We consider the problem of identifying among many candidates a single best solution which jointly maximizes several domain-specific target functions. Assuming that the candidate solutions can be generated incrementally, we model the error in prediction due to the incompleteness of partial solutions as a normally distributed random variable. Using this model, we derive a probabilistic search algorithm that aims at finding the best solution without the necessity to complete and rank all candidate solutions. We do not assume a Viterbi-type decoding, allowing a wider range of target functions. We evaluate the proposed algorithm on the problem of best parse identification, combining simple heuristic with more complex machine-learning based target functions. We show that the search algorithm is capable of identifying candidates with a very high score without completing a significant proportion of the candidate solutions.