The 6th International Workshop on Designing Meaning Representations
NounAtlas, VerbAtlas, BMR, MOSAICo and other marvels: Towards a Unified Multilingual Semantic Framework
Abstract: In the era of Large Language Models (LLMs), the pursuit of a unified, language-independent representation of meaning remains both essential and complex. In this talk I will revisit the rationale for advancing semantic understanding beyond the capabilities of LLMs and focuses on the challenge of multilingual Semantic Role Labeling (SRL) enabled by the unification of VerbAtlas and NounAtlas, two resources developed at Sapienza NLP that provide language-independent frame inventories connected to BabelNet. I will then survey the BabelNet Meaning Representation (BMR) formalism, which enables language-independent semantic representations of sentences, and MOSAICo, a new large-scale dataset annotated with several semantic layers, namely Word Sense Disambiguation, SRL, Semantic Parsing, and Relation Extraction. Central to our work is the creation of wide-coverage multilingual inter-task resources as well as the design of innovative methods that bridge word- and sentence-level meanings across languages. Taken together, these efforts lay the groundwork for practical, robust multilingual semantic tools and their integration into today’s Large Language Models.
Bio: Roberto Navigli is Professor of Natural Language Processing at the Sapienza University of Rome, where he leads the Sapienza NLP Group. He has received two ERC grants on multilingual semantics, highlighted among the 15 projects through which the ERC has transformed science. He has received several prizes, including two Artificial Intelligence Journal prominent paper awards and several outstanding/best paper awards from ACL. He leads the Italian Minerva LLM Project — the first LLM pre-trained in Italian — and is the co-founder and scientific director of Babelscape, a successful deep-tech company developing next-generation multilingual NLU and NLG. He is a Fellow of ACL, AAAI, ELLIS, and EurAI, and serves as General Chair of ACL 2025.
Quantum machine learning for natural language processing
Abstract: The observation that natural language obeys rules similar to rationals, gave rise to a series of algebraic theories of syntax. These are known as type-logical or categorial grammars. Their two prominent examples are Combinatory Categorial Grammar and the Lambek Calculus. Due to their algebraic nature, these theories are easily mapped to vector spaces and linear maps and given a statistical semantics. In this talk, we show how they can also be mapped to qubits and quantum gates, and develop a quantum circuit semantics for them. Our model is tested on a range of tasks, from question answering to classification, to anaphora resolution, and recently also to content-based recommendation. I will go through the setting, present a summary of the results and show how the quantum methods are able to firstly, solve the tasks with many less parameters and secondly, learn some rare linguistic correlations.
Bio: I studied computer software engineering and then logic at Sharif University of Technology in Iran. My PhD was from University of Quebec at Montreal, while being an academic visitor at Oxford, where I later stayed for 10 years, funded by a Wolfson College Junior Research Fellowship, an EPSRC PDRA and then an EPSRC Career Acceleration Fellowship. Since 2022, I am a Professor of Computer Science in UCL. At the moment, I hold a Royal Academy of Engineering Research Chair and lead a lab on mathematical and quantum methods for natural language.