Part-of-Speech Tagging with Minimal Lexicalization

TL;DR

This paper introduces LegoTag, a PoS tagger using a Dynamic Bayesian Network with minimal lexical features, achieving state-of-the-art accuracy and demonstrating that a small, linguistically motivated feature set suffices for effective tagging.

Contribution

The paper presents a compact, flexible PoS tagger that reduces feature complexity while maintaining high accuracy, emphasizing the importance of minimal, linguistically motivated features.

Findings

Small suffix sets improve cross-corpora generalization

Minimal lexicon with function words suffices for good performance

State-of-the-art error rate of 3.6% on benchmark corpus

Abstract

We use a Dynamic Bayesian Network to represent compactly a variety of sublexical and contextual features relevant to Part-of-Speech (PoS) tagging. The outcome is a flexible tagger (LegoTag) with state-of-the-art performance (3.6% error on a benchmark corpus). We explore the effect of eliminating redundancy and radically reducing the size of feature vocabularies. We find that a small but linguistically motivated set of suffixes results in improved cross-corpora generalization. We also show that a minimal lexicon limited to function words is sufficient to ensure reasonable performance.