High-quality Data-to-Text Generation for Severely Under-Resourced Languages with Out-of-the-box Large Language Models

TL;DR

This study demonstrates that pretrained large language models can significantly improve data-to-text generation in severely under-resourced languages, achieving near-human performance and surpassing previous methods.

Contribution

The paper shows that out-of-the-box LLMs can effectively generate high-quality text in under-resourced languages, highlighting their potential to bridge NLP performance gaps.

Findings

LLMs outperform previous methods for under-resourced languages

Human evaluation shows near-human performance for top systems

BLEU scores are unreliable for non-task-specific evaluation

Abstract

The performance of NLP methods for severely under-resourced languages cannot currently hope to match the state of the art in NLP methods for well resourced languages. We explore the extent to which pretrained large language models (LLMs) can bridge this gap, via the example of data-to-text generation for Irish, Welsh, Breton and Maltese. We test LLMs on these under-resourced languages and English, in a range of scenarios. We find that LLMs easily set the state of the art for the under-resourced languages by substantial margins, as measured by both automatic and human evaluations. For all our languages, human evaluation shows on-a-par performance with humans for our best systems, but BLEU scores collapse compared to English, casting doubt on the metric's suitability for evaluating non-task-specific systems. Overall, our results demonstrate the great potential of LLMs to bridge the performance gap for under-resourced languages.