A neural interlingua for multilingual machine translation

TL;DR

This paper introduces a neural interlingua in multilingual NMT that enables zero-shot translation and cross-lingual classification, achieving comparable performance with fewer parameters.

Contribution

It presents a neural interlingua that learns language-independent representations, facilitating zero-shot translation and cross-lingual tasks within a unified model.

Findings

Enables zero-shot translation without pivot languages.

Achieves comparable BLEU scores with fewer parameters.

Supports cross-lingual classification across multiple languages.

Abstract

We incorporate an explicit neural interlingua into a multilingual encoder-decoder neural machine translation (NMT) architecture. We demonstrate that our model learns a language-independent representation by performing direct zero-shot translation (without using pivot translation), and by using the source sentence embeddings to create an English Yelp review classifier that, through the mediation of the neural interlingua, can also classify French and German reviews. Furthermore, we show that, despite using a smaller number of parameters than a pairwise collection of bilingual NMT models, our approach produces comparable BLEU scores for each language pair in WMT15.