Analogy Training Multilingual Encoders

Language encoders encode words and phrases in ways that capture their local semantic relatedness, but are known to be globally inconsistent. Global inconsistencycan seemingly be corrected for, in part, by leveraging signals from knowledge bases, but previous results are partial and limited to monolingual English encoders.We extract a large-scale multilingual, multi-word analogy dataset from Wikidata for diagnosing and correcting for global inconsistencies and implement a four-way Siamese BERT architecture for grounding multilingualBERT (mBERT) in Wikidata through analogy training. We show that analogy training not only improves theglobal consistency of mBERT, as well as the isomorphism of language-specific subspaces, but also leads to significant gains on downstream tasks such as bilingual dictionary induction and sentence retrieval.