We describe PTT-MULTITAN, our submission to the WMT26 MT Evaluation shared task, Task 3 (error-free segment detection). The task requires a binary decision — does a machine translation contain any error? — without ac- cess to reference translations, and is scored with the Matthews Correlation Coefficient (MCC). Our system combines a supervised classifier fitted on WMT25 human error-span annotations with three families of reference- free indicators: structural and phraseological cues, corpus-linguistic association measures computed over an LLM-as-judge model, and format/locale conformity checks. We have im- proved our system submissions, whose MCC rises from 0.159 to 0.168. We submitted six language pairs from the official WMT26 test set. The pairs were from English into Belaru- sian, German, Russian, Ukrainian, Simplified Chinese, and Traditional Chinese.

This paper describes our participation in the WMT25 Terminology Shared Task, specifically Track 1 (Spanish to English) focused on translation within the Information Technology (IT) domain. The shared task challenges participants to improve machine translation systems by effectively incorporating terminology constraints to ensure accurate and consistent translation of specialised technical terms. We experimented with several approaches to tackle terminology and lexical constraints with both NMT systems and LLMs with a small amount of training data and a glossary. Experimental results demonstrate that systems behave differently with and without glossary. The NMT system seems rather limited in adapting to special lexicon and resizing embeddings, which is the opposite of the case with LLMs preferring structured instructions. Through this participation, our objective is to improve terminology accuracy and overall translation quality, highlight the potential of specialised terminology-aware translation models for technical domains, and explore possibilities of fine-tuning of LLMs and NMT models with domain and lexical constraints.

This paper discusses the WMT 2021 terminology shared task from a “meta” perspective. We present the results of our experiments using the terminology dataset and the OpenNMT (Klein et al., 2017) and JoeyNMT (Kreutzer et al., 2019) toolkits for the language direction English to French. Our experiment 1 compares the predictions of the two toolkits. Experiment 2 uses OpenNMT to fine-tune the model. We report our results for the task with the evaluation script but mostly discuss the linguistic properties of the terminology dataset provided for the task. We provide evidence of the importance of text genres across scores, having replicated the evaluation scripts.