Grammar Engineering Meets LLMs: Development of Cantonese and Irish ParGram Treebanks

2026-08-07

AI Summary

This study investigates the efficient development of formal grammatical resources for Cantonese and Irish while preserving cross-linguistic consistency, and evaluates the potential of multilingual large language models (LLMs) to assist in grammar engineering for low-resource languages. Building upon the ParGram framework, we present the first systematic application of multilingual LLMs—such as gpt-oss-120b—to parallel treebank construction, leveraging model-assisted translation and syntactic structure generation to maintain alignment at the level of abstract functional representations. Experimental results indicate that model-generated translations show limited efficacy and are unaffected by the choice of prompt language; although syntactic generation captures predicate–argument relations to some extent, it underperforms on cross-linguistically abstract tasks, necessitating expert intervention. This work establishes a novel paradigm and empirical foundation for formal syntactic modeling of low-resource languages.

Abstract

Grammar engineering requires expertise in linguistic formalism and computational implementation, especially in parallel grammar projects that balance cross-linguistic consistency with language-specific properties. This paper presents the development of Cantonese and Irish treebanks within the Parallel Grammar (ParGram) Project, where linguistic parallelism is maintained at an abstract functional level. We also investigate the methodological potential and limitations of using multilingual LLMs to support grammar engineering, focusing on Cantonese-Irish translation and the generation of formal syntactic structures using OpenAI's gpt-oss-120b model. The results show that translation performance was generally unsatisfactory and unaffected by prompt language. For syntactic structure generation, the model produced some structurally meaningful outputs, but performed poorly on tasks requiring cross-linguistic abstraction. Nonetheless, LLM-generated outputs may still offer some reference value by suggesting alternative analyses and (partially) capturing predicate-argument relations. Overall, our findings highlight both the potential and limitations of using LLMs in collaborative grammar engineering, while underscoring the continued importance of expert-driven analysis and verification.

Problem

• grammar engineering
• parallel grammar
• Cantonese
• Irish
• LLMs

Innovation

• grammar engineering
• multilingual LLMs
• ParGram treebanks
• cross-linguistic abstraction
• syntactic structure generation