Multilingual Word Embeddings using Multigraphs
This work addresses the challenge of handling multilingual data in NLP, offering incremental improvements for tasks like machine translation.
The authors tackled the problem of learning multilingual word embeddings by proposing neural-network models that leverage both monolingual and multilingual text, resulting in higher accuracy on syntactic/semantic tasks and improved machine translation for out-of-vocabulary words.
We present a family of neural-network--inspired models for computing continuous word representations, specifically designed to exploit both monolingual and multilingual text. This framework allows us to perform unsupervised training of embeddings that exhibit higher accuracy on syntactic and semantic compositionality, as well as multilingual semantic similarity, compared to previous models trained in an unsupervised fashion. We also show that such multilingual embeddings, optimized for semantic similarity, can improve the performance of statistical machine translation with respect to how it handles words not present in the parallel data.