English

Dict-BERT: Enhancing Language Model Pre-training with Dictionary

Computation and Language 2022-03-22 v2 Artificial Intelligence Machine Learning

Abstract

Pre-trained language models (PLMs) aim to learn universal language representations by conducting self-supervised training tasks on large-scale corpora. Since PLMs capture word semantics in different contexts, the quality of word representations highly depends on word frequency, which usually follows a heavy-tailed distributions in the pre-training corpus. Therefore, the embeddings of rare words on the tail are usually poorly optimized. In this work, we focus on enhancing language model pre-training by leveraging definitions of the rare words in dictionaries (e.g., Wiktionary). To incorporate a rare word definition as a part of input, we fetch its definition from the dictionary and append it to the end of the input text sequence. In addition to training with the masked language modeling objective, we propose two novel self-supervised pre-training tasks on word and sentence-level alignment between input text sequence and rare word definitions to enhance language modeling representation with dictionary. We evaluate the proposed Dict-BERT model on the language understanding benchmark GLUE and eight specialized domain benchmark datasets. Extensive experiments demonstrate that Dict-BERT can significantly improve the understanding of rare words and boost model performance on various NLP downstream tasks.

Keywords

Cite

@article{arxiv.2110.06490,
  title  = {Dict-BERT: Enhancing Language Model Pre-training with Dictionary},
  author = {Wenhao Yu and Chenguang Zhu and Yuwei Fang and Donghan Yu and Shuohang Wang and Yichong Xu and Michael Zeng and Meng Jiang},
  journal= {arXiv preprint arXiv:2110.06490},
  year   = {2022}
}

Comments

ACL 2022 (Findings)

R2 v1 2026-06-24T06:50:57.522Z