Building Machine Translation Systems for the Next Thousand Languages

Abstract

In this paper we share findings from our effort to build practical machine translation (MT) systems capable of translating across over one thousand languages. We describe results in three research domains: (i) Building clean, web-mined datasets for 1500+ languages by leveraging semi-supervised pre-training for language identification and developing data-driven filtering techniques; (ii) Developing practical MT models for under-served languages by leveraging massively multilingual models trained with supervised parallel data for over 100 high-resource languages and monolingual datasets for an additional 1000+ languages; and (iii) Studying the limitations of evaluation metrics for these languages and conducting qualitative analysis of the outputs from our MT models, highlighting several frequent error modes of these types of models. Using this approach, we add 24 new languages to Google Translate, the product's largest increase in language coverage to-date. We hope that our work provides useful insights to practitioners working towards building MT systems for currently understudied languages, and highlights research directions that can complement the weaknesses of massively multilingual models in data-sparse settings.

BigSSL: Exploring the Frontier of Large-Scale Semi-Supervised Learning for Automatic Speech Recognition

“…Web-NTL: For pre-training with unlabeled text, we use a web-crawled corpus of monolingual text containing over 28B sentences [76]. The dataset spans 1140 languages, 205 of which have over 1M sentences and 199 of which have between 100k and 1M sentences.…”

“…We up-sample lower resource languages using temperature-based sampling [77] with T = 3.0. More details about the dataset and the mining approach have been described in Section 2 of [76].…”

Transforming Assessment: The Impacts and Implications of Large Language Models and Generative AI

“…Machine translation has rapidly advanced in the past few years, expanding its scope to most languages. Meta's No Language Left Behind translates 200 different languages with high-quality results (Meta, 2022), and Google Translate, as of 2022, supports 133 languages, including 24 low-resource languages (Bapna, 2022). OpenAI's GPT models also emerge as excellent translators by generating context-relevant translation (Hendy et al, 2023).…”

Tencent AI Lab - Shanghai Jiao Tong University Low-Resource Translation System for the WMT22 Translation Task

“…A common approach is to expand the vocabulary and the word embedding matrix to contain the extra tokens. However, the incoming embeddings must be randomly initialized (Garcia et al, 2021;Bapna et al, 2022), which leads to inconsistency with the original embeddings and increases training difficulty. Fortunately, Rikters et al (2022) has released a translation model for En-Liv called Liv4ever-MT 4 .…”