Factual Consistency of Multilingual Pretrained Language Models
This work addresses the reliability of factual knowledge extraction for multilingual NLP applications, but it is incremental as it extends prior monolingual analysis to a multilingual setting.
The paper tackles the problem of factual consistency in multilingual pretrained language models by extending analysis to 45 languages, finding that models like mBERT and XLM-R are highly inconsistent across languages, with inconsistency often worse than in English.
Pretrained language models can be queried for factual knowledge, with potential applications in knowledge base acquisition and tasks that require inference. However, for that, we need to know how reliable this knowledge is, and recent work has shown that monolingual English language models lack consistency when predicting factual knowledge, that is, they fill-in-the-blank differently for paraphrases describing the same fact. In this paper, we extend the analysis of consistency to a multilingual setting. We introduce a resource, mParaRel, and investigate (i) whether multilingual language models such as mBERT and XLM-R are more consistent than their monolingual counterparts; and (ii) if such models are equally consistent across languages. We find that mBERT is as inconsistent as English BERT in English paraphrases, but that both mBERT and XLM-R exhibit a high degree of inconsistency in English and even more so for all the other 45 languages.