Date: Tuesday June 16, 2026 - 16:30–18:00 BST = 17:30–19:00 CEST

Convenors: Edward Ross (University of Reading), Chiara Zanchi (University of Pavia)

Youtube link: https://www.youtube.com/live/3wpP2iV1knI?si=JmMJkw_adqkNwSQ2

Slides: Slides session 8

Toward the Semi-Automated Population of the Ancient Greek WordNet: LLM-powered synonym generation and synset attribution

The first part of this session will explore the use of LLMs in the semi-automatic synonim generation for Ancient Greek with the aim of populating the Ancient Greek WordNet. Several approaches are investigated: zero-shot, few-shots, and fine-tuning. The results are compared against an English baseline. Zero-shot approach yields the highest accuracy, while fine-tuning leads to the highest number of potential synonyms. Our experiment also reveals that polysemy and PoS play a role in the model’s performance, as the highest scores are registered for polysemous words and for verbs and nouns.

In addition, I will present work in progress regarding the automated synset attribution to the Ancient Greek groups of synonyms. In this work, two complementary strategies for synset selection are investigated. The former leverages the hierarchical organization of WordNet by constructing bounded hypernym trees and ranking candidate roots according to the portion of the retrieved semantic space they organize. The latter relies on semantic comparison between candidate glosses and a derived metadefinition, using LLMs to perform pairwise preference judgments aggregated through an Elo ranking scheme.

Ethics and Efficacy: Teaching and Learning Ancient Greek and Latin in an AI World

The second part of this session introduces how artificial intelligence (AI) tools are implemented as part of Ancient Greek and Latin teaching and learning at the introductory level. We will discuss the importance and impact of introducing students to current AI ethical issues, including environmental impact, data corruption, worker exploitation, copyright infringement, and cultural misunderstanding. With this grounding, we will then present best practices for supporting Ancient Greek and Latin teaching and learning with AI tools, such as AI model personalization, using public domain datasets, and prompt engineering tips. Through this, we aim to support teachers and learners that approach AI tools and their outputs critically, ethically, and with an open mind.

• Beatrice Marchesi, Annachiara Clementelli, Andrea Maurizio Mammarella, Silvia Zampetta, Erica Biagetti, Luca Brigada Villa, Virginia Mastellari, Riccardo Ginevra, Claudia Roberta Combei, and Chiara Zanchi. 2025. Towards the Semi-Automated Population of the Ancient Greek WordNet. In Proceedings of the Eleventh Italian Conference on Computational Linguistics (CLiC-it 2025), pages 647–658, Cagliari, Italy. CEUR Workshop Proceedings.
• Daniela Santoro, Beatrice Marchesi, Silvia Zampetta, Marco Del Tredici, Erica Biagetti, Eleonora Litta, Claudia Roberta Combei, Stefano Rocchi, Tullio Facchinetti, Riccardo Ginevra, and Chiara Zanchi. 2025. Exploring Latin WordNet synset annotation with LLMs. In Proceedings of the 13th Global Wordnet Conference, pages 66–76, Pavia, Italy. Global Wordnet Association.
• Edward A. S. Ross, and Jackie Baines. 2025. “Navigating the Fog: The Effectiveness of Personalized Conversational GenAI Models for Supporting Ancient Language Learning.” AI & Antiquity 1, No. 1 (2025). pp. 35-52. https://doi.org/10.64946/aiantiquity.v1i1.002.
• Edward A. S. Ross, and Jackie Baines. 2024. “Treading Water: New Data on the Impact of AI Ethics Information Sessions in Classics and Ancient Language Pedagogy.” The Journal of Classics Teaching 25, No. 50 (2024). pp. 181-190. https://doi.org/10.1017/S2058631024000412.
• Jackie Baines, Edward A. S. Ross, Jacinta Hunter, Fleur McRitchie Pratt, and Nisha Patel. 2024. Digital Tools for Learning Ancient Greek and Latin and Guiding Phrases for Using Generative AI in Ancient Language Study. V3. March 12, 2024. Archived by figshare. https://doi.org/10.6084/m9.figshare.25391782.v3.

• Erica Biagetti, Martina Giuliani, Silvia Zampetta, Silvia Luraghi, and Chiara Zanchi. 2024. Combining Neo-Structuralist and Cognitive Approaches to Semantics to Build Wordnets for Ancient Languages: Challenges and Perspectives. In Proceedings of the Workshop on Cognitive Aspects of the Lexicon @ LREC-COLING 2024, pages 151–161, Torino, Italia. ELRA and ICCL.
• Erica Biagetti, Chiara Zanchi, and William Michael Short. 2021. Toward the creation of WordNets for ancient Indo-European languages. In Proceedings of the 11th Global Wordnet Conference, pages 258–266, University of South Africa (UNISA). Global Wordnet Association.
• Yuri Bizzoni, Federico Boschetti, Harry Diakoff, Riccardo Del Gratta, Monica Monachini and Gregory R. Crane. 2014. The Making of Ancient Greek WordNet. In: Nicoletta Calzolari et al. (eds.), Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC, vol. 2014), Reykjavik, Iceland, may 2014, 1140-1147. Accessed online at https://www.aclweb.org/anthology/L14-1054/.
• Christiane Fellbaum (ed.). 1998. WordNet: An electronic lexical database. MIT Press, Cambridge, MA.
• Stefano Minozzi. 2009. The Latin WordNet Project. In: Peter Anreiter and Manfred Kienpointner (eds.), Latin Linguistics Today. Akten des 15. Internationalem Kolloquiums zur Lateinischen Linguistik, Innsbrucker Beiträge zur Sprachwissenschaft 137, Innsbruck, AUT, 707-716.
• James O’Donnell, and Casey Crownhart. 2025. “Everything you need to know about estimating AI’s energy and emissions burden.” MIT Technology Review, 20 May. Available at: https://www.technologyreview.com/2025/05/20/1116331/ai-energy-demand-methodology.
• Ganna Pogrebna. 2024. “AI is a multi-billion dollar industry. It’s underpinned by an invisible and exploited workforce.” The Conversation, 8 October. Available at: https://theconversation.com/ai-is-a-multi-billion-dollar-industry-its-underpinned-by-an-invisible-and-exploited-workforce-240568.
• Cristian Randieri. 2025. “Bias and corruption in artificial intelligence: A threat to fairness.” Forbes, 14 March. Available at: https://www.forbes.com/councils/forbestechcouncil/2025/03/14/bias-and-corruption-in-artificial-intelligence-a-threat-to-fairness.
• Edward A. S. Ross. 2023. “A New Frontier: AI and Ancient Language Pedagogy.” The Journal of Classics Teaching 24, No. 48 (2023). pp. 143-161. https://doi.org/10.1017/S2058631023000430.
• Cheng Lim Saw, and Bryan Zhi Yang Tan. 2025. “Unpacking copyright infringement issues in the GenAI development lifecycle and a peek into the future.” Computer Law & Security Review 58, pp. 1–17. https://doi.org/10.1016/j.clsr.2025.106163.

1. GitHub repos related to the required readings:

• https://github.com/unipv-larl/llms-ag
• https://github.com/unipv-larl/mapping-for-agwn

2. Introductory Latin Personalized GenAI Tool Dataset. V3. June 6, 2025. Archived by figshare. https://doi.org/10.6084/m9.figshare.29261450.v3.

1. Select 8 groups of synonymous words from the GitHub repository at the following link: https://github.com/unipv-larl/llms-ag/blob/main/Data%20for%20fine%20tuning/train.jsonl.

• In the file, one Ancient Greek word is marked as "input" and the others as "synonyms." You can ignore this distinction and simply consider all the words listed on a single line as synonyms.
• Please note that these groups were compiled by conflating two back-translation dictionaries (refer to the slides).
• Avoid selecting groups that are too large.
• Try to balance your sample by Part of Speech (POS): select 2 noun groups, 2 verb groups, 2 adjective groups, and 2 adverb groups.
• If you encounter any difficulties while annoating, feel free to adjust your sample and focus exclusively on nouns and adjectives.

2. For each of these groups, choose the synset that you find most appropriate to describe the sense in which these words can be considered synonymous.

• Remember that WordNet uses a broad definition of synonymy. The words do not need to be synonymous in every possible context; rather, they should be interchangeable in at least some contexts.
• To find the synset, start with the Open English WordNet repository: https://en-word.net/lemma/free
• You can look them up by searching for the corresponding English lemma and then checking the synset's gloss (definition).

If you wish, it would be highly appreciated if you shared your completed annotations with Chiara Zanchi by sending your file to chiara.zanchi@unipv.it.