Arturo Oncevay

Senior AI Researcher · NLP & Evaluation · London

About

I'm an NLP researcher working on multilingual data, low-resource languages and evaluation. In industry (financial services), I focus on evaluating LLM and agentic systems for high-stakes applications, where reliability matters as much as raw capability. I hold a PhD in Informatics from the University of Edinburgh, where I studied how linguistic typology shapes neural machine translation.

Outside my day job, I'm committed to open research: through AmericasNLP, WMT shared tasks and my collaboration with PUCP, I help build public benchmarks and openly released data, so that better models are possible for every language, not only the best-resourced ones.

Interests

• Multilingual & low-resource NLP
• Data curation & quality
• Evaluation & benchmarking
• Agentic systems evaluation
• LLM-as-judge reliability
• Machine translation
• Linguistic typology
• Open models & open data

Community & open research

• AmericasNLP — co-founder & organiser, 2021–present Workshop series and open shared tasks on NLP for Indigenous languages of the Americas, with public benchmarks for languages that had none.
• WMT · Translation with Terminologies — co-organiser, 2025–2026 Shared task on terminology-constrained translation: benchmark design, terminology data curation and system evaluation.
• PUCP — external research advisor, 2018–present Supervising student research on long-tail and low-resource NLP at Pontificia Universidad Católica del Perú.
• ACL — area chair, 2025–present Reviewer for ACL, EMNLP, NAACL, WMT, SIGTYP and LoResMT since 2019.

Selected publications

• Simple Beats Spend: Cost–Performance Trade-Offs for Multi-Agent Machine Translation of Specialized Long Documents Oncevay et al. · Findings of EMNLP 2026 (to appear)
• On the Robustness of Morphosyntactic Transformation with Large Language Models: The Case of Quechua Collao Pocco, Oncevay · AmericasNLP 2026Peru
• Translating Domain-Specific Terminology in Typologically-Diverse Languages: A Study in Tax and Financial Education Oncevay et al. · EMNLP 2025
• Findings of the WMT25 Terminology Translation Task: Terminology is Useful Especially for Good MTs Semenov, Huang, Zouhar, Berger, Zhu, Oncevay, Chen · WMT 2025
• AfroCS-xs: Creating a Compact, High-Quality, Human-Validated Code-Switched Dataset for African Languages Olaleye, Oncevay, Sibue et al. · ACL 2025
• The Impact of Domain-Specific Terminology on Machine Translation for Finance in European Languages Oncevay et al. · NAACL 2025
• AmericasNLI: Evaluating Zero-shot Natural Language Understanding of Pretrained Multilingual Models in Truly Low-resource Languages Ebrahimi, Mager, Oncevay et al. · ACL 2022
• Quantifying Synthesis and Fusion and their Impact on Machine Translation Oncevay, Ataman, van Berkel, Haddow, Birch, Bjerva · NAACL 2022
• SchAman: Spell-Checking Resources and Benchmark for Endangered Languages from Amazonia Oncevay et al. · AACL-IJCNLP 2022Peru
• Findings of the AmericasNLP 2021 Shared Task on Open Machine Translation for Indigenous Languages of the Americas Mager, Oncevay et al. · AmericasNLP 2021
• No Data to Crawl? Monolingual Corpus Creation from PDF Files of Truly Low-Resource Languages in Peru Bustamante, Oncevay, Zariquiey · LREC 2020Peru
• Bridging Linguistic Typology and Multilingual Machine Translation with Multi-View Language Representations Oncevay, Haddow, Birch · EMNLP 2020

Full list on Google Scholar and the ACL Anthology.