Identifying Phrase Structure in Sequential Data

Title: Identifying Phrase Structure in Sequential Data

Abstract: This research investigates the problem of combining the outcomes of several different classifiers to provide a coherent infer-ence that satisfies some constraints. The study focuses on two general approaches: a Markovian approach that extends standard Hidden Markov Models (HMMs) to allow the use of a rich observation structure and general classifiers to model state-observation dependencies, and an extension of constraint satisfaction formalisms. The research develops efficient combinationalgorithms under both models and studies them experimentally in the context of shallow parsing. The results show that the Constraint Satisfaction with Classifiers (CSCL) approach performs better than the Projection-based Markov Models (PMM) on both tasks, with a more significant improvement on the harder, Subject-Verb (SV) task. These findings suggest that CSCL's ability to cope better with the length of the phrase and long-term dependencies contributes to its better performance. The research also demonstrates the improvements of the standard HMM when allowing states to depend on a richer structure of the observation via the use of classifiers.

Link to Article: https://arxiv.org/abs/0111003v1 Authors: arXiv ID: 0111003v1