An Empirical Study of Discriminative Sequence Labeling Models for Vietnamese Text Processing

TL;DR

This study empirically compares CRFs and LSTMs for Vietnamese text tasks, showing simple features and word embeddings yield high accuracy, with LSTMs not always outperforming CRFs on moderate datasets.

Contribution

It provides a comprehensive empirical comparison of CRFs and LSTMs for Vietnamese text processing, highlighting the effectiveness of simple features and word embeddings.

Findings

Simple word-based features achieve over 90% accuracy.

Word embeddings significantly improve model performance.

LSTMs do not always outperform CRFs on moderate datasets.

Abstract

This paper presents an empirical study of two widely-used sequence prediction models, Conditional Random Fields (CRFs) and Long Short-Term Memory Networks (LSTMs), on two fundamental tasks for Vietnamese text processing, including part-of-speech tagging and named entity recognition. We show that a strong lower bound for labeling accuracy can be obtained by relying only on simple word-based features with minimal hand-crafted feature engineering, of 90.65\% and 86.03\% performance scores on the standard test sets for the two tasks respectively. In particular, we demonstrate empirically the surprising efficiency of word embeddings in both of the two tasks, with both of the two models. We point out that the state-of-the-art LSTMs model does not always outperform significantly the traditional CRFs model, especially on moderate-sized data sets. Finally, we give some suggestions and discussions for efficient use of sequence labeling models in practical applications.