Character-level Representations Improve DRS-based Semantic Parsing Even in the Age of BERT

EMNLP 2020  ·  Rik van Noord, Antonio Toral, Johan Bos ·

We combine character-level and contextual language model representations to improve performance on Discourse Representation Structure parsing. Character representations can easily be added in a sequence-to-sequence model in either one encoder or as a fully separate encoder, with improvements that are robust to different language models, languages and data sets. For English, these improvements are larger than adding individual sources of linguistic information or adding non-contextual embeddings. A new method of analysis based on semantic tags demonstrates that the character-level representations improve performance across a subset of selected semantic phenomena.

PDF Abstract EMNLP 2020 PDF EMNLP 2020 Abstract

Datasets


  Add Datasets introduced or used in this paper

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
DRS Parsing PMB-2.2.0 Bi-LSTM seq2seq: BERT + characters in 1 encoder F1 88.3 # 1
DRS Parsing PMB-3.0.0 Bi-LSTM seq2seq: BERT + characters in 1 encoder F1 89.3 # 1

Methods


No methods listed for this paper. Add relevant methods here