This paper describes our participation in Task 11 of SemEval-2026, which evaluates the ability of models to determine logical validity of syllogisms independent of real-world content. We develop and compare three approaches for Subtask 1: (1) an encoder-based classification baseline using both classical ML methods and fine-tuned BERT with debiasing strategies; (2) an autoformalization pipeline combining DPO-aligned models with first order logic translation and formal inference via Prover9; and (3) a hybrid neuro-symbolic approach using GPT to generate OWL 2 ontologies evaluated with the HermiT reasoner. Our best result was achieved by the encoder-based classifier, obtaining a 72.25% accuracy and a combined score of 20.37, placing 40th out of 45 participating teams. Analysis shows that classification methods exhibit lower content bias, autoformalization approaches suffer from translation inconsistencies and syntax incompatibilities, and ontology-based reasoning is hindered by prompt design limitations and verbose serialization formats. All our code can be found in the paper’s repository.

Metaphor identification remains challenging in specialized and low-resource domains, where large annotated datasets are unavailable and general-domain models often fail to transfer effectively. In this paper, we evaluate FLAVORS-AECC, a Spanish dataset of oncological discourse that provides transparent, instance-level annotations of basic meaning (BM) and contextual meaning (CM) following the Metaphor Identification Procedure (MIP). We test the state-of-the-art Contrast-WSD model under two splits: a random split and a lemma-based split to control for lexical memorization. We compare three configurations: (i) a control model with no meaning information, (ii) manually curated basic meanings, and (iii) first dictionary entry as an approximation of basic meaning. Results show that explicitly modeling meaning contrast substantially improves performance in low-resource settings (from below 0.30 to above 0.50 F1). However, contrary to expectations, manually annotated BM does not consistently outperform first dictionary entries, suggesting that definition length rather than theoretical fidelity may introduce noise. We also find that models perform best on cases with high annotator agreement and that verbs remain the most challenging part of speech. Overall, our findings highlight the importance of linguistically grounded modeling for metaphor detection in specialized domains.

Metaphors play a central role in cancer narratives, helping patients and practitioners articulate complex experiences and technical concepts. While cancer metaphors in English have been extensively studied, Spanish remains underexplored in this regard, despite its global importance and rich cultural variation. This paper presents a new dataset of Spanish cancer metaphors designed to address these gaps. The resource comprises over 80K annotated words drawn from diverse forum posts, with detailed documentation of lexical units, contextual versus basic meanings, and inter-annotator agreements. To construct the dataset, we adapted the Metaphor Identification Procedure (MIP) for Spanish medical discourse, proposing methodological refinements to challenges such as defining lexical units or domain-specific Basic Meaning labels.

In this paper we introduce MUSCLE, a dataset for MUltilingual lexico-Semantic Classification of Links between Entities. The MUSCLE dataset was designed to train and evaluate Lexical Relation Classification (LRC) systems with 27K pairs of universal concepts selected from Wikidata, a large and highly multilingual factual Knowledge Graph (KG). Each pair of concepts includes its lexical forms in 25 languages and is labeled with up to five possible lexico-semantic relations between the concepts: hypernymy, hyponymy, meronymy, holonymy, and antonymy. Inspired by Semantic Map theory, the dataset bridges lexical and conceptual semantics, is more challenging and robust than previous datasets for LRC, avoids lexical memorization, is domain-balanced across entities, and enables enrichment and hierarchical information retrieval.

Understanding the relation between the meanings of words is an important part of comprehending natural language. Prior work has either focused on analysing lexical semantic relations in word embeddings or probing pretrained language models (PLMs), with some exceptions. Given the rarity of highly multilingual benchmarks, it is unclear to what extent PLMs capture relational knowledge and are able to transfer it across languages. To start addressing this question, we propose MultiLexBATS, a multilingual parallel dataset of lexical semantic relations adapted from BATS in 15 languages including low-resource languages, such as Bambara, Lithuanian, and Albanian. As experiment on cross-lingual transfer of relational knowledge, we test the PLMs’ ability to (1) capture analogies across languages, and (2) predict translation targets. We find considerable differences across relation types and languages with a clear preference for hypernymy and antonymy as well as romance languages.

Metaphors shape the way we think by enabling the expression of one concept in terms of another one. For instance, cancer can be understood as a place from which one can go in and out, as a journey that one can traverse, or as a battle. Giving patients awareness of the way they refer to cancer and different narratives in which they can reframe it has been proven to be a key aspect when experiencing the disease. In this work, we propose a preliminary identification and representation of Spanish cancer metaphors using MIP (Metaphor Identification Procedure) and MetaNet. The created resource is the first openly available dataset for medical metaphors in Spanish. Thus, in the future, we expect to use it as the gold standard in automatic metaphor processing tasks, which will also serve to further populate the resource and understand how cancer is experienced and narrated.

The accurate prediction of lexical relations between words is a challenging task in Natural Language Processing (NLP). The most recent advances in this direction come with the use of pre-trained language models (PTLMs). A PTLM typically needs “well-formed” verbalized text to interact with it, either to fine-tune it or to exploit it. However, there are indications that commonly used PTLMs already encode enough linguistic knowledge to allow the use of minimal (or none) textual context for some linguistically motivated tasks, thus notably reducing human effort, the need for data pre-processing, and favoring techniques that are language neutral since do not rely on syntactic structures. In this work, we explore this idea for the tasks of lexical relation classification (LRC) and graded Lexical Entailment (LE). After fine-tuning PTLMs for LRC with different verbalizations, our evaluation results show that very simple prompts are competitive for LRC and significantly outperform graded LE SoTA. In order to gain a better insight into this phenomenon, we perform a number of quantitative statistical analyses on the results, as well as a qualitative visual exploration based on embedding projections.