Domain Terminology Integration into Machine Translation: Leveraging Large Language Models

TL;DR

The paper tackles domain-specific terminology translation in MT within the WMT 2023 Terminology Shared Task by proposing a four-step LLM-assisted pipeline: generate bilingual synthetic data from pre-approved terms, fine-tune an OPUS MT model on a mix of synthetic and generic data, translate with the fine-tuned model, and apply terminology-constrained post-editing using an LLM to insert missing terms. It demonstrates that this approach substantially increases term usage in translations (from about 36.7% to 72.9%) while maintaining or improving overall translation quality across German–English, English–Czech, and Chinese–English. The study also discusses data generation quality controls, model fine-tuning details, and the benefits of post-editing over pure inference using LLMs for domain adaptation. The work suggests practical implications for terminology adherence in production MT systems and outlines avenues for future work with larger LLMs and alternative MT architectures to broaden language coverage and efficiency.

Abstract

This paper discusses the methods that we used for our submissions to the WMT 2023 Terminology Shared Task for German-to-English (DE-EN), English-to-Czech (EN-CS), and Chinese-to-English (ZH-EN) language pairs. The task aims to advance machine translation (MT) by challenging participants to develop systems that accurately translate technical terms, ultimately enhancing communication and understanding in specialised domains. To this end, we conduct experiments that utilise large language models (LLMs) for two purposes: generating synthetic bilingual terminology-based data, and post-editing translations generated by an MT model through incorporating pre-approved terms. Our system employs a four-step process: (i) using an LLM to generate bilingual synthetic data based on the provided terminology, (ii) fine-tuning a generic encoder-decoder MT model, with a mix of the terminology-based synthetic data generated in the first step and a randomly sampled portion of the original generic training data, (iii) generating translations with the fine-tuned MT model, and (iv) finally, leveraging an LLM for terminology-constrained automatic post-editing of the translations that do not include the required terms. The results demonstrate the effectiveness of our proposed approach in improving the integration of pre-approved terms into translations. The number of terms incorporated into the translations of the blind dataset increases from an average of 36.67% with the generic model to an average of 72.88% by the end of the process. In other words, successful utilisation of terms nearly doubles across the three language pairs.