Can Lexical Semantics Improve Translation In Low-resource And Endangered Languages?

Abstract

Low-resource, especially endangered languages are often overlooked in a majority of NLP tasks. For instance, in neural machine translation (NMT), the aim is to map text from one language into another. Although many advances have been made in developing NMT systems for natural language, little research has been done on understanding how the word ordering and lexical similarity between the source and target language affect translation performance. Specifically, in the case of low-resource and endangered languages, translation performance can often be enhanced through model architectures that leverage this understanding. Here, we investigate these relationships on a variety of low-resource, endangered Eurasiatic language pairs from the OpenSubtitles2016 database, where the source language is English, and find that the more similar the target language is to English, the greater the translation performance. In addition, we study the impact of providing NMT models with part of speech (POS) of words in the English sequence and find that, for Transformer-based models, the more dissimilar the target language is from English, the greater the benefit provided by POS. Our findings could further enhance interdisciplinary collaboration to develop NLP models to benefit speakers of many endangered languages