DACE For Railway Acronym Disambiguation

TL;DR

This paper introduces DACE, a novel framework that improves acronym disambiguation in railway documentation by leveraging adaptive prompting, external knowledge, and ensemble methods, achieving top performance in a competitive setting.

Contribution

The paper presents DACE, a new approach combining dynamic prompting, retrieval-augmented generation, and ensemble aggregation for improved acronym disambiguation in specialized domains.

Findings

Achieved top F1 score of 0.9069 in TextMine'26 competition.

Effectively handles low-resource scenarios and reduces hallucination.

Outperforms baseline models in railway acronym disambiguation.

Abstract

Acronym Disambiguation (AD) is a fundamental challenge in technical text processing, particularly in specialized sectors where high ambiguity complicates automated analysis. This paper addresses AD within the context of the TextMine'26 competition on French railway documentation. We present DACE (Dynamic Prompting, Retrieval Augmented Generation, Contextual Selection, and Ensemble Aggregation), a framework that enhances Large Language Models through adaptive in-context learning and external domain knowledge injection. By dynamically tailoring prompts to acronym ambiguity and aggregating ensemble predictions, DACE mitigates hallucination and effectively handles low-resource scenarios. Our approach secured the top rank in the competition with an F1 score of 0.9069.