This paper describes the OpenU’s submission to the WMT26 Terminology Translation Task for Spanish into Basque (Track 1), covering the automotive and energy domains. We present AGENTEAK, a multi-agent system built on LangGraph in which three compact open-weight models (4-8B parameters) cooperate: a reasoning model that selects and disambiguates the terminology relevant to each segment, a domain fine-tuned translation model, and a verification model that checks and repairs terminology in the draft translation. To adapt the translator, we mined in-domain Spanish-Basque bitext from Wikipedia using multilingual sentence embeddings and complemented the data with in-domain synthetic parallel paragraphs generated by an instruction-tuned Basque LLM. Domain-tagged fine-tuning improves the base model by up to 10.4 BLEU and 3.9 COMET points on our in-domain test sets. On the shared task data, supplying the pipeline with correct terminology raises terminology success rate from 0.613 to 0.757 (automotive) and from 0.601 to 0.774 (energy).

Bioinformatics is almost invisible in Machine Translation (MT) research, despite being a hard case: dense specialised terminology, a high density of Multiword Expressions (MWEs), and an acute scarcity of parallel data. We address this gap for English-French. We compile a bioinformatics parallel corpus (14,467 sentence pairs) and expand it to 236k pairs through four augmentation strategies, each profiled for lexical diversity before use, then fine-tune five compact models (223M-8B parameters) under an incremental data ablation. We evaluate with automatic metrics and with a three-rater MQM evaluation involving professional life-sciences translators and using an error typology extended with a dedicated MWE category. Domain adaptation improves every system on every metric, with fine-tuned models surpassing much larger state-of-the-art commercial models on our domain test set. Expert judgements reveal a more structured picture with Accuracy errors falling by up to 68.8% and MWE errors by up to 51.1%, while also highlighting an accuracy-fluency trade-off that the automatic metrics conceal. Notably, the least lexically diverse augmentation sources contribute gains comparable to the most diverse ones, which we advance as one candidate account of the trade-off. We release all corpora, synthetic data, and adapted models.

Multiword Expressions (MWEs) are pervasive in scientific writing, and in specialized domains they include both multiword terminology (e.g., noun compounds) and recurrent academic phrasing. This study profiles MWEs in a large corpus of bioinformatics research articles segmented by IMRaD sections. Building on recent multi-method approaches to scientific MWE identification, we extract MWEs using complementary automated strategies (semantic matching, dependency parsing, controlled vocabularies, and academic formula lists) and compare the resulting inventories by size, form, and IMRaD section distribution. We further quantify cross-document dispersion using document frequency and Gries’ DP to distinguish widely reused expressions from items concentrated in a small subset of articles. Results show that bioinformatics MWEs are predominantly short and nominal, but that extraction methods differ in the extent to which they recover discourse and reporting phraseology. Dispersion is strongly long-tailed across sections with most MWEs being document-specific, while a smaller recurrent core aligns with section function and is enriched for conventional templates and standardized multiword terms. Overall, the findings argue for combining complementary identification methods with dispersion profiling to characterize domain “multiwordness” in a principled and section-sensitive way.