1. Generate bilingual synthetic data based on the pre-approved terms, using an LLM, namely ChatGPT.
2. Fine-tune a generic model, OPUS (Tiedemann and Thottingal, 2020), on a mix of the terminology-based synthetic data generated in (i) and a randomly sampled portion of the original generic training data.
3. Generate translations of the dev, test, and blind datasets provided by the organisers with the fine-tuned model from (ii).
4. Apply terminology-constrained automatic post-editing using ChatGPT to incorporate missing terms into translations that do not yet include the required terminology.