Evaluating Contextualized Embeddings on 54 Languages in POS Tagging, Lemmatization and Dependency Parsing

TL;DR

This paper extensively evaluates three recent contextualized embedding methods across 54 languages and three NLP tasks, demonstrating state-of-the-art results in POS tagging, lemmatization, and dependency parsing.

Contribution

It provides a comprehensive comparison of BERT, Flair, and ELMo embeddings within a strong baseline, highlighting their effectiveness over previous pretrained embeddings.

Findings

State-of-the-art results achieved in all tasks

BERT, Flair, and ELMo outperform previous embeddings

Significant improvements over UD 2.2 results

Abstract

We present an extensive evaluation of three recently proposed methods for contextualized embeddings on 89 corpora in 54 languages of the Universal Dependencies 2.3 in three tasks: POS tagging, lemmatization, and dependency parsing. Employing the BERT, Flair and ELMo as pretrained embedding inputs in a strong baseline of UDPipe 2.0, one of the best-performing systems of the CoNLL 2018 Shared Task and an overall winner of the EPE 2018, we present a one-to-one comparison of the three contextualized word embedding methods, as well as a comparison with word2vec-like pretrained embeddings and with end-to-end character-level word embeddings. We report state-of-the-art results in all three tasks as compared to results on UD 2.2 in the CoNLL 2018 Shared Task.