ChatGPT MT: Competitive for High- (but not Low-) Resource Languages

TL;DR

This study evaluates ChatGPT's machine translation performance across 204 languages, finding it excels in high-resource languages but underperforms in low-resource and African languages, highlighting resource level as a key factor.

Contribution

First comprehensive evaluation of ChatGPT's MT capabilities across a wide range of languages, revealing resource-dependent performance disparities.

Findings

GPT approaches or exceeds traditional MT in high-resource languages.

GPT underperforms in 84.1% of low-resource languages.

Resource level is the primary factor affecting translation quality.

Abstract

Large language models (LLMs) implicitly learn to perform a range of language tasks, including machine translation (MT). Previous studies explore aspects of LLMs' MT capabilities. However, there exist a wide variety of languages for which recent LLM MT performance has never before been evaluated. Without published experimental evidence on the matter, it is difficult for speakers of the world's diverse languages to know how and whether they can use LLMs for their languages. We present the first experimental evidence for an expansive set of 204 languages, along with MT cost analysis, using the FLORES-200 benchmark. Trends reveal that GPT models approach or exceed traditional MT model performance for some high-resource languages (HRLs) but consistently lag for low-resource languages (LRLs), under-performing traditional MT for 84.1% of languages we covered. Our analysis reveals that a language's resource level is the most important feature in determining ChatGPT's relative ability to translate it, and suggests that ChatGPT is especially disadvantaged for LRLs and African languages.