André F. T. Martins · page 2
Papers
- Findings of the WMT 2021 Shared Task on Quality Estimation (2021 · 80 citations)
Quality Estimation (QE) predicts how good a machine translation is without access to a reference translation. The 2021 edition of the WMT QE shared task ran three subtasks on output from neural MT systems: - MLQE-PE: A Multilingual Quality Estimation and Post-Editing Dataset (2022 · 76 citations)
MLQE-PE is a public dataset for training and testing machine translation Quality Estimation (QE), which predicts how good a translation is without a reference. - Turbo Parsers: Dependency Parsing by Approximate Variational Inference (2010 · 69 citations)
Rich non-projective dependency parsers that score sibling and grandparent arcs make exact inference intractable, so they need approximate decoding. - MaLA-500: Massive Language Adaptation of Large Language Models (2024 · 58 citations)
Most large language models are trained mainly on English or a few dozen languages, so they perform poorly on low-resource languages. - Transferring Coreference Resolvers with Posterior Regularization (2015 · 35 citations)
Coreference resolvers usually depend on annotated corpora, which most languages lack. This paper builds resolvers for low-resource languages by transferring knowledge from an English system, using English→Spanish and English→Brazilian Portuguese as test cases. First, the Berkeley Coreference Resolution system labels… - EuroBERT: Scaling Multilingual Encoders for European Languages (2025 · 34 citations)
Bidirectional encoders still power most retrieval, classification and regression systems, but recent modeling progress has mostly gone into decoder-only LLMs. - Findings of the WMT 2020 Shared Task on Chat Translation (2020 · 29 citations)
This report presents the inaugural WMT competition devoted to translating chat. Its focus is bilingual customer-support conversations in which a German-speaking customer and an English-speaking agent each write in their own language. Such chat text is short, informal, noisy and heavy in pronouns, and public data for… - Assessing the Role of Context in Chat Translation Evaluation: Is Context Helpful and Under What Conditions? (2024 · 8 citations)
Most automatic MT metrics were built and validated on well-formed text such as news. It is unclear how well they judge machine-translated chats, which are short, informal and depend heavily on earlier turns. The authors first compare MQM error annotations for chat and news in WMT22 English–German. Errors in chat turn… - Fine-Grained Reward Optimization for Machine Translation using Error Severity Mappings (2026 · 7 citations)
Most RL training for neural MT scores a translation with one number per sentence. This sparse signal does not say where errors occur, so learning is slow and unstable, especially for long inputs. The authors instead derive a reward for each token from XCOMET, a quality estimation model that marks error spans and… - A Context-aware Framework for Translation-mediated Conversations (2026 · 4 citations)
Machine translation can connect people, or people and English-centric assistants, who lack a shared language.
This page in Markdown → https://md.research.elamur.ai/free/en/authors/andre-f-t-martins/page/2/