This repository is for the paper UAlberta at SemEval-2025 Task 2: Prompting and Ensembling for Entity-Aware Translation. In Proceedings of the 19th International Workshop on Semantic Evaluation (SemEval-2025), pages 1709–1717, Vienna, Austria. Association for Computational Linguistics.
🏆 1st Place — COMET Track
🔗 Task | 📊 Leaderboard 📄 Paper | 🖼️ Poster
gpt/– GPT-based Translation Moduletrans/– Other Translation Modules (e.g., Google Cloud)wiki/– Wiki Retrieval Modulewsd/– Word Sense Disambiguation Module
(Official submissions can be found in assets/submissions/)
- Ning Shi — mrshininnnnn@gmail.com
@inproceedings{shi-etal-2025-ualberta,
title = "{UA}lberta at {S}em{E}val-2025 Task 2: Prompting and Ensembling for Entity-Aware Translation",
author = "Shi, Ning and
Basil, David and
Hauer, Bradley and
Nawal, Noshin and
Riley, Jai and
Teodorescu, Daniela and
Zhang, John and
Kondrak, Grzegorz",
editor = "Rosenthal, Sara and
Ros{\'a}, Aiala and
Ghosh, Debanjan and
Zampieri, Marcos",
booktitle = "Proceedings of the 19th International Workshop on Semantic Evaluation (SemEval-2025)",
month = jul,
year = "2025",
address = "Vienna, Austria",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.semeval-1.224/",
pages = "1709--1717",
ISBN = "979-8-89176-273-2",
abstract = "We describe the methods used by our UAlberta team for the SemEval-2025 Task 2 on Entity-Aware Machine Translation (EA-MT). Our methods leverage large language models with prompt engineering strategies suited to this task, including retrieval augmented generation and in-context learning. Our best results overall are obtained with ensembles of multiple models, leveraging named entity knowledge in the dataset. Finally, we provide proof-of-concept experiments showing that lexico-semantic knowledge can be used to identify high-quality translations. We further demonstrate that our methods can function even without gold named entity translations, by using an alternative knowledge base such as BabelNet."
}