DyLex: Incorporating Dynamic Lexicons into BERT for Sequence Labeling

TL;DR

DyLex introduces a novel approach to incorporate dynamic lexicons into BERT for sequence labeling, effectively handling large-scale lexicons and reducing noise, leading to state-of-the-art results across multiple datasets.

Contribution

The paper presents DyLex, a plug-in framework that uses word-agnostic tag embeddings and a denoising method to incorporate dynamic lexicons into BERT without retraining.

Findings

Achieves new state-of-the-art performance on ten datasets

Effectively handles large-scale lexicons with reduced noise

Demonstrates robustness across three sequence labeling tasks

Abstract

Incorporating lexical knowledge into deep learning models has been proved to be very effective for sequence labeling tasks. However, previous works commonly have difficulty dealing with large-scale dynamic lexicons which often cause excessive matching noise and problems of frequent updates. In this paper, we propose DyLex, a plug-in lexicon incorporation approach for BERT based sequence labeling tasks. Instead of leveraging embeddings of words in the lexicon as in conventional methods, we adopt word-agnostic tag embeddings to avoid re-training the representation while updating the lexicon. Moreover, we employ an effective supervised lexical knowledge denoising method to smooth out matching noise. Finally, we introduce a col-wise attention based knowledge fusion mechanism to guarantee the pluggability of the proposed framework. Experiments on ten datasets of three tasks show that the proposed framework achieves new SOTA, even with very large scale lexicons.