https://hdl.handle.net/10807/60 2026-05-13T21:00:14Z https://hdl.handle.net/10807/335518 Titolo: Evaluating Hierarchical Aggregation and LLM-Based Matching for Synset Selection in Ancient Greek Autori: Brigada Villa Luca; Passarotti Marco; Zanchi Chiara; Ginevra Riccardo; Fratellini Erica; Litta Eleonora Abstract: This paper presents a structured framework for WordNet synset selection applied to Ancient Greek lexical material. Starting from synonym definitions extracted from the Liddell–Scott–Jones (LSJ) lexicon, we compare two strategies: hierarchy-driven aggregation via bounded hypernym trees and LLM-based definitional matching with pairwise ranking. Graded human evaluation shows that structure-aware methods provide a robust baseline, particularly for nouns and verbs, while LLM-based reranking does not consistently improve performance, especially for highly ploysemous groups of synonyms. Beyond supporting the development of an Ancient Greek WordNet, the study highlights the methodological portability of the framework to other languages and lexical resources. 2026-01-01T00:00:00Z https://hdl.handle.net/10807/335516 Titolo: The UD_Latin-PROIEL as Linked Open Data: Integrating a Latin Treebank into the LiLa Knowledge Base Autori: Dezotti Lucas; Passarotti Marco; Iurescia Federica; Moretti Giovanni Abstract: This paper presents the steps taken to integrate data from the UD_Latin-PROIEL treebank into the LiLa Knowledge Base of interoperable linguistic resources for Latin. It describes how the lexical, morphological, syntactic, and citation information from the source was modeled using the Linked Open Data principles as adopted by the LiLa Knowledge Base. The process of linking tokens to the LiLa collection of Latin lemmas is detailed, addressing challenges such as ambiguities, new lemmas, and errors encountered in the source. The outcome is a syntactically annotated textual resource that is interoperable with the (meta)data of other Latin linguistic resources linked within the LiLa Knowledge Base. This integration enables new ways of analyzing linguistic information and using the content as a starting point to explore connections with other interlinked resources. A use case demonstrates this interoperability. 2026-01-01T00:00:00Z https://hdl.handle.net/10807/335483 Titolo: Overview of the Dependency Parsing Task at EvaLatin 2026 Autori: Iurescia Federica; Passarotti Marco; Sprugnoli Rachele Abstract: This paper presents the organization, methodology, and outcomes of the Dependency Parsing shared task held within the fourth edition of EvaLatin, a campaign dedicated to the evaluation of Natural Language Processing tools for Latin. EvaLatin aims to promote and advance research in language technologies for Latin, fostering the development of robust and linguistically informed computational approaches. The paper provides a detailed description of the data released for the shared task. It also outlines the evaluation framework and metrics adopted for assessing system performance. The results achieved by participating teams are reported and comparatively analyzed, highlighting strengths, limitations, and emerging trends in current approaches to Latin dependency parsing. Finally, the paper discusses the main challenges posed by the task and suggests directions for future research in the field. 2026-01-01T00:00:00Z https://hdl.handle.net/10807/335482 Titolo: From Lemmas to Links: A Lemma Bank for Ancient Greek Autori: Swaelens Colin; Mambrini Francesco; Passarotti Marco Abstract: This paper presents the initial construction of a lemma bank for Ancient Greek, developed according to the Linked Data principles. The need for interoperable linguistic infrastructures capable of supporting interoperability among historical variation, divergent annotation practices, and resource-specific lemmatisation conventions was highlighted by the increasing availability of digital linguistic resources. This lemma bank is developed as the core component of the Linking Greek knowledge base and inspired by the architecture of the LiLa project for Latin. The proposed lemma bank adopts a descriptive, lemma-centric approach that preserves alternative canonical forms and dialectal variation while enabling consistent linking across lexical and semantic resources. Its population combines data extracted from the Ancient Greek WordNet and the Liddell–Scott–Jones lexicon, integrating semantic structure, part-of-speech information, and lexicographically encoded gender assignment. Additional normalisation steps, including the rule-based correction of closed-class part-of-speech categories and harmonisation to the Universal Dependencies tagset, were applied to improve consistency and computational usability. The resulting dataset provides a foundation for interlinking corpora, lexica, and NLP tools for Ancient Greek within a Linked Open Data framework. 2026-01-01T00:00:00Z