DACE For Railway Acronym Disambiguation

2025-12-20

AI Summary

To address the challenge of automated acronym disambiguation in French railway technical documents—characterized by high lexical ambiguity and scarce annotated data—this paper proposes DACE, a novel framework featuring dynamic ambiguity-aware prompting. DACE integrates retrieval-augmented generation (RAG), context-sensitive candidate selection, and vote-based ensemble aggregation to enable adaptive fusion of heterogeneous domain knowledge. By explicitly modeling ambiguity during inference, it significantly mitigates large language model hallucination and enhances generalization and reliability in low-resource, domain-specific settings. Evaluated on the TextMine’26 international competition benchmark, DACE achieves an F1 score of 0.9069—the highest reported performance for acronym disambiguation in railway texts—securing first place. This result empirically validates the effectiveness of dynamic prompting coupled with ensemble-based disambiguation in highly ambiguous, vertical domains.

Technology Category

Natural Language Processing: Discourse, Pragmatics & Argument MiningKnowledge Representation and Reasoning: Knowledge AcquisitionData Mining & Knowledge Management: Knowledge Acquisition from the Web

Application Category

Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAGSemantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMsWeb Mining and Content Analysis: Robustness and generalizability of Web mining methods

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.

Problem

• Disambiguating acronyms in French railway documents
• Enhancing LLMs with dynamic prompting and knowledge injection
• Mitigating hallucinations in low-resource technical text processing

Innovation

• Dynamic prompting tailored to acronym ambiguity
• Retrieval augmented generation with external domain knowledge
• Ensemble aggregation of predictions to reduce hallucination