2026
Franco, Arcangelo; Passaro, Lucia; Soldani, Jacopo
Intelligent Detection of Design Patterns in Existing Source Code Book Chapter
In: Spalazzese, Romina; Vaidhyanathan, Karthik; Capuano, Roberta; Cerny, Tomas; Reussner, Ralf; Russo, Barbara (Ed.): Software Architecture. ECSA 2026 Tracks and Workshops, pp. 152–168, Springer Nature Switzerland, Cham, 2026, ISBN: 978-3-032-39143-8.
@inbook{10.1007/978-3-032-39143-8_15,
title = {Intelligent Detection of Design Patterns in Existing Source Code},
author = {Arcangelo Franco and Lucia Passaro and Jacopo Soldani},
editor = {Romina Spalazzese and Karthik Vaidhyanathan and Roberta Capuano and Tomas Cerny and Ralf Reussner and Barbara Russo},
url = {https://link.springer.com/chapter/10.1007/978-3-032-39143-8_15#citeas},
doi = {10.1007/978-3-032-39143-8_15},
isbn = {978-3-032-39143-8},
year = {2026},
date = {2026-09-19},
urldate = {2026-09-19},
booktitle = {Software Architecture. ECSA 2026 Tracks and Workshops},
pages = {152–168},
publisher = {Springer Nature Switzerland},
address = {Cham},
abstract = {Detecting design patterns in source code is essential for understanding and maintaining software architectures, yet remains difficult due to variability in real-world implementations. This paper introduces a neuro-symbolic framework for design pattern detection that combines interpretable structural analysis with learned semantic representations. The approach integrates two complementary signals, i.e., topological conformance via abstract syntax tree analysis under a four-pillar heuristic, and semantic similarity to canonical pattern prototypes in a discriminative latent space. This latent space is obtained by fine-tuning CodeBERT with supervised contrastive learning, yielding cohesive and separable pattern clusters. To mitigate data scarcity and lexical bias, we generate and validate a synthetic dataset using large language models and apply identifier anonymization at the abstract syntax tree level. Evaluation on an unseen synthetic test set and on a manually validated Gold Standard benchmark demonstrates improved robustness over purely structural methods, with strong performance across standard metrics. Overall, combining structural constraints with optimised semantic representations offers a robust and scalable solution for design pattern detection in software architecture analysis.},
keywords = {},
pubstate = {published},
tppubtype = {inbook}
}
Detecting design patterns in source code is essential for understanding and maintaining software architectures, yet remains difficult due to variability in real-world implementations. This paper introduces a neuro-symbolic framework for design pattern detection that combines interpretable structural analysis with learned semantic representations. The approach integrates two complementary signals, i.e., topological conformance via abstract syntax tree analysis under a four-pillar heuristic, and semantic similarity to canonical pattern prototypes in a discriminative latent space. This latent space is obtained by fine-tuning CodeBERT with supervised contrastive learning, yielding cohesive and separable pattern clusters. To mitigate data scarcity and lexical bias, we generate and validate a synthetic dataset using large language models and apply identifier anonymization at the abstract syntax tree level. Evaluation on an unseen synthetic test set and on a manually validated Gold Standard benchmark demonstrates improved robustness over purely structural methods, with strong performance across standard metrics. Overall, combining structural constraints with optimised semantic representations offers a robust and scalable solution for design pattern detection in software architecture analysis.
Zisa, Bartolomeo; Passaro, Lucia; Soldani, Jacopo
Design Pattern-Based Code Refactoring with Llms Book Chapter
In: 2026 IEEE International Conference on Software Analysis, Evolution and Reengineering (SANER), pp. 222–227, 2026.
@inbook{11576692,
title = {Design Pattern-Based Code Refactoring with Llms},
author = {Bartolomeo Zisa and Lucia Passaro and Jacopo Soldani},
doi = {10.1109/SANER67736.2026.00033},
year = {2026},
date = {2026-01-01},
urldate = {2026-01-01},
booktitle = {2026 IEEE International Conference on Software Analysis, Evolution and Reengineering (SANER)},
pages = {222–227},
keywords = {},
pubstate = {published},
tppubtype = {inbook}
}
Lenci, Alessandro
GloVe Pre-trained Spaces Book Chapter
In: Nesi, Hilary; Milin, Petar (Ed.): International Encyclopedia of Language and Linguistics, Elsevier, 2026.
@inbook{Lenci2026,
title = {GloVe Pre-trained Spaces},
author = {Alessandro Lenci},
editor = {Hilary Nesi and Petar Milin},
url = {https://linkinghub.elsevier.com/retrieve/pii/B9780323955041011406},
doi = {10.1016/B978-0-323-95504-1.01140-6},
year = {2026},
date = {2026-01-01},
booktitle = {International Encyclopedia of Language and Linguistics},
publisher = {Elsevier},
keywords = {},
pubstate = {published},
tppubtype = {inbook}
}
Passaro, Lucia; Amadei, Davide; Bacciu, Davide
Emotion Recognition in Multimodal Social Data Conference
ESANN 2026 proceedings, European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning, 2026, ISBN: 9782875870964.
@conference{ 11568_1361188,
title = {Emotion Recognition in Multimodal Social Data},
author = {Lucia Passaro and Davide Amadei and Davide Bacciu},
url = {https://www.esann.org/sites/default/files/proceedings/2026/ES2026-287.pdf},
doi = {10.14428/esann/2026.es2026-287},
isbn = {9782875870964},
year = {2026},
date = {2026-01-01},
booktitle = {ESANN 2026 proceedings, European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning},
pages = {235–240},
abstract = {Emotion recognition on social media is often approached in unimodal or single-label settings, despite the multimodal nature of online communication. This paper presents a study of multilabel emotion recognition from paired text-image data. We evaluate vision–language encoders and compare them with strong unimodal baselines and a zero-shot multimodal LLM. A simple multimodal classifier built on CLIP achieves the most reliable performance. Data-centric additions such as emoji transcription, caption augmentation, and pseudo-labelling offer limited gains, whereas calibrated decision thresholds have a consistent effect. The results highlight the value of visual cues and show limitations of recent VLMs.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Emotion recognition on social media is often approached in unimodal or single-label settings, despite the multimodal nature of online communication. This paper presents a study of multilabel emotion recognition from paired text-image data. We evaluate vision–language encoders and compare them with strong unimodal baselines and a zero-shot multimodal LLM. A simple multimodal classifier built on CLIP achieves the most reliable performance. Data-centric additions such as emoji transcription, caption augmentation, and pseudo-labelling offer limited gains, whereas calibrated decision thresholds have a consistent effect. The results highlight the value of visual cues and show limitations of recent VLMs.
Oglio, Pietro Dell'; Bondielli, Alessandro; Marcelloni, Francesco; Passaro, Lucia C.
Enhancing Debunking Effectiveness Through LLM-Based Personality Adaptation Conference
Computational Intelligence 17th International Joint Conference, IJCCI 2025, Marbella, Spain, October 22–24, 2025, Proceedings, Part I, vol. 2827, 2026, ISBN: 9783032156327.
@conference{ 11568_1349898,
title = {Enhancing Debunking Effectiveness Through LLM-Based Personality Adaptation},
author = {Pietro Dell' Oglio and Alessandro Bondielli and Francesco Marcelloni and Lucia C. Passaro},
url = {https://link.springer.com/chapter/10.1007/978-3-032-15632-7_23},
doi = {10.1007/978-3-032-15632-7_23},
isbn = {9783032156327},
year = {2026},
date = {2026-01-01},
booktitle = {Computational Intelligence 17th International Joint Conference, IJCCI 2025, Marbella, Spain, October 22–24, 2025, Proceedings, Part I},
volume = {2827},
pages = {413–431},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Bushipaka, Praveen; Passaro, Lucia; Cucinotta, Tommaso
Standard vs. Modular Sampling: Best Practices for Reliable LLM Unlearning Book Chapter
In: Machine Learning and Principles and Practice of Knowledge Discovery in Databases. ECML PKDD 2025., vol. 2840, pp. 635–658, Springer Nature Switzerland, Cham, 2026, ISBN: 9783032190994.
@inbook{<LineBreak> 11568_1361187,
title = {Standard vs. Modular Sampling: Best Practices for Reliable LLM Unlearning},
author = {Praveen Bushipaka and Lucia Passaro and Tommaso Cucinotta},
url = {https://link.springer.com/chapter/10.1007/978-3-032-19099-4_43},
doi = {10.1007/978-3-032-19099-4_43},
isbn = {9783032190994},
year = {2026},
date = {2026-01-01},
urldate = {2026-01-01},
booktitle = {Machine Learning and Principles and Practice of Knowledge Discovery in Databases. ECML PKDD 2025.},
volume = {2840},
pages = {635–658},
publisher = {Springer Nature Switzerland},
address = {Cham},
keywords = {},
pubstate = {published},
tppubtype = {inbook}
}
Nissim, Malvina; Croce, Danilo; Patti, Viviana; Basile, Pierpaolo; Attanasio, Giuseppe; Musacchio, Elio; Rinaldi, Matteo; Borazio, Federico; Francis, Maria; Gili, Jacopo; Scalena, Daniel; Altuna, Begoña; Azurmendi, Ekhi; Basile, Valerio; Bentivogli, Luisa; Bisazza, Arianna; Bolognesi, Marianna; Brunato, Dominique; Caselli, Tommaso; Casola, Silvia; Cassese, Maria; Cettolo, Mauro; Collacciani, Claudia; Cosmo, Leonardo De; Buono, Maria Pia Di; Esuli, Andrea; Etxaniz, Julen; Ferrando, Chiara; Fidelangeli, Alessia; Frenda, Simona; Fusco, Achille; Gaido, Marco; Galassi, Andrea; Galli, Federico; Giordano, Luca; Goffetti, Mattia; Gonzalez-Dios, Itziar; Gregori, Lorenzo; Grundler, Giulia; Iannaccone, Sandro; Jiang, Chunyang; Quatra, Moreno La; Lagioia, Francesca; Lo, Soda Marem; Madeddu, Marco; Magnini, Bernardo; Manna, Raffaele; Mercorio, Fabio; Merlo, Paola; Muti, Arianna; Nastase, Vivi; Negri, Matteo; Onorati, Dario; Palmieri, Elena; Papi, Sara; Passaro, Lucia; Pensa, Giulia; Piergentili, Andrea; Potertì, Daniele; Puccetti, Giovanni; Ranaldi, Federico; Ranaldi, Leonardo; Ravelli, Andrea Amelio; Rosola, Martina; Ruzzetti, Elena Sofia; Samo, Giuseppe; Santilli, Andrea; Santin, Piera; Sarti, Gabriele; Sartor, Giovanni; Savoldi, Beatrice; Serino, Antonio; Seveso, Andrea; Siciliani, Lucia; Torroni, Paolo; Varvara, Rossella; Zaninello, Andrea; Zanollo, Asya; Zanzotto, Fabio Massimo; Zeinalipour, Kamyar; Zugarini, Andrea
Challenging the Abilities of Large Language Models in Italian: a Community Initiative Journal Article
In: IJCOL, vol. 12, no. 1, pp. 33–109, 2026.
@article{ 11568_1371727,
title = {Challenging the Abilities of Large Language Models in Italian: a Community Initiative},
author = {Malvina Nissim and Danilo Croce and Viviana Patti and Pierpaolo Basile and Giuseppe Attanasio and Elio Musacchio and Matteo Rinaldi and Federico Borazio and Maria Francis and Jacopo Gili and Daniel Scalena and Begoña Altuna and Ekhi Azurmendi and Valerio Basile and Luisa Bentivogli and Arianna Bisazza and Marianna Bolognesi and Dominique Brunato and Tommaso Caselli and Silvia Casola and Maria Cassese and Mauro Cettolo and Claudia Collacciani and Leonardo De Cosmo and Maria Pia Di Buono and Andrea Esuli and Julen Etxaniz and Chiara Ferrando and Alessia Fidelangeli and Simona Frenda and Achille Fusco and Marco Gaido and Andrea Galassi and Federico Galli and Luca Giordano and Mattia Goffetti and Itziar Gonzalez-Dios and Lorenzo Gregori and Giulia Grundler and Sandro Iannaccone and Chunyang Jiang and Moreno La Quatra and Francesca Lagioia and Soda Marem Lo and Marco Madeddu and Bernardo Magnini and Raffaele Manna and Fabio Mercorio and Paola Merlo and Arianna Muti and Vivi Nastase and Matteo Negri and Dario Onorati and Elena Palmieri and Sara Papi and Lucia Passaro and Giulia Pensa and Andrea Piergentili and Daniele Potertì and Giovanni Puccetti and Federico Ranaldi and Leonardo Ranaldi and Andrea Amelio Ravelli and Martina Rosola and Elena Sofia Ruzzetti and Giuseppe Samo and Andrea Santilli and Piera Santin and Gabriele Sarti and Giovanni Sartor and Beatrice Savoldi and Antonio Serino and Andrea Seveso and Lucia Siciliani and Paolo Torroni and Rossella Varvara and Andrea Zaninello and Asya Zanollo and Fabio Massimo Zanzotto and Kamyar Zeinalipour and Andrea Zugarini},
doi = {10.17454/ijcol121.02},
year = {2026},
date = {2026-01-01},
journal = {IJCOL},
volume = {12},
number = {1},
pages = {33–109},
abstract = {The rapid progress of Large Language Models (LLMs) has transformed natural language processing and broadened its impact across research and society. Yet, systematic evaluation of these models, especially for languages beyond English, remains limited. "Challenging the Abilities of LAnguage Models in ITAlian" (CALAMITA) is a large-scale collaborative benchmarking initiative for Italian, coordinated under the Italian Association for Computational Linguistics. Unlike existing efforts that focus on leaderboards, CALAMITA foregrounds methodology: it federates more than 80 contributors from academia, industry, and the public sector to design, document, and evaluate a diverse collection of tasks, covering linguistic competence, commonsense reasoning, factual consistency, fairness, summarization, translation, and code generation. Through this process, we not only assembled a benchmark of over 20 tasks and almost 100 subtasks, but also established a centralized evaluation pipeline that supports heterogeneous datasets and metrics. We report results for four open-weight LLMs, highlighting systematic strengths and weaknesses across abilities, as well as challenges in task-specific evaluation. Beyond quantitative results, CALAMITA exposes methodological lessons: the necessity of fine-grained, task-representative metrics, the importance of harmonized pipelines, and the benefits and limitations of broad community engagement. CALAMITA is conceived as a rolling benchmark, enabling continuous integration of new tasks and models. This makes it both a resource – the most comprehensive and diverse benchmark for Italian to date – and a framework for sustainable, community-driven evaluation. We argue that this combination offers a blueprint for other languages and communities seeking inclusive and rigorous LLM evaluation practices.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The rapid progress of Large Language Models (LLMs) has transformed natural language processing and broadened its impact across research and society. Yet, systematic evaluation of these models, especially for languages beyond English, remains limited. "Challenging the Abilities of LAnguage Models in ITAlian" (CALAMITA) is a large-scale collaborative benchmarking initiative for Italian, coordinated under the Italian Association for Computational Linguistics. Unlike existing efforts that focus on leaderboards, CALAMITA foregrounds methodology: it federates more than 80 contributors from academia, industry, and the public sector to design, document, and evaluate a diverse collection of tasks, covering linguistic competence, commonsense reasoning, factual consistency, fairness, summarization, translation, and code generation. Through this process, we not only assembled a benchmark of over 20 tasks and almost 100 subtasks, but also established a centralized evaluation pipeline that supports heterogeneous datasets and metrics. We report results for four open-weight LLMs, highlighting systematic strengths and weaknesses across abilities, as well as challenges in task-specific evaluation. Beyond quantitative results, CALAMITA exposes methodological lessons: the necessity of fine-grained, task-representative metrics, the importance of harmonized pipelines, and the benefits and limitations of broad community engagement. CALAMITA is conceived as a rolling benchmark, enabling continuous integration of new tasks and models. This makes it both a resource – the most comprehensive and diverse benchmark for Italian to date – and a framework for sustainable, community-driven evaluation. We argue that this combination offers a blueprint for other languages and communities seeking inclusive and rigorous LLM evaluation practices.
Mannocci, Lorenzo; Naretto, Francesca; Passaro, Lucia; Monreale, Anna
RAG-Enhanced LLMs for Interactive Explainability in Clinical Decision Support Systems Conference
Communications in Computer and Information Science, vol. 2830, Springer Science and Business Media Deutschland GmbH, GEWERBESTRASSE 11, CHAM, CH-6330, SWITZERLAND, 2026, ISBN: 9783032167088.
@conference{ 11568_1366368,
title = {RAG-Enhanced LLMs for Interactive Explainability in Clinical Decision Support Systems},
author = {Lorenzo Mannocci and Francesca Naretto and Lucia Passaro and Anna Monreale},
doi = {10.1007/978-3-032-16708-8_4},
isbn = {9783032167088},
year = {2026},
date = {2026-01-01},
booktitle = {Communications in Computer and Information Science},
volume = {2830},
pages = {50–63},
publisher = {Springer Science and Business Media Deutschland GmbH},
address = {GEWERBESTRASSE 11, CHAM, CH-6330, SWITZERLAND},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Fidone, Giacomo; Passaro, Lucia; Guidotti, Riccardo
Evaluating Online Moderation Via LLM-Powered Counterfactual Simulations Conference
Proceedings of the AAAI Conference on Artificial Intelligence, vol. 40, no. 45, AAAI Press, Washington, DC, 2026, ISBN: 978-1-57735-906-7.
@conference{ 11568_1333475b,
title = {Evaluating Online Moderation Via LLM-Powered Counterfactual Simulations},
author = {Giacomo Fidone and Lucia Passaro and Riccardo Guidotti},
url = {https://ojs.aaai.org/index.php/AAAI/article/view/41186},
doi = {10.1609/aaai.v40i45.41186},
isbn = {978-1-57735-906-7},
year = {2026},
date = {2026-01-01},
booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence},
volume = {40},
number = {45},
pages = {38451–38459},
publisher = {AAAI Press},
address = {Washington, DC},
abstract = {Online Social Networks (OSNs) widely adopt content moderation to mitigate the spread of abusive and toxic discourse. Nonetheless, the real effectiveness of moderation interventions remains unclear due to the high cost of data collection and limited experimental control. The latest developments in Natural Language Processing pave the way for a new evaluation approach. Large Language Models (LLMs) can be successfully leveraged to enhance Agent-Based Modeling and simulate human-like social behavior with unprecedented degree of believability. Yet, existing tools do not support simulation-based evaluation of moderation strategies. We fill this gap by designing a LLM-powered simulator of OSN conversations enabling a parallel, counterfactual simulation where toxic behavior is influenced by moderation interventions, keeping all else equal. We conduct extensive experiments, unveiling the psychological realism of OSN agents, the emergence of social contagion phenomena and the superior effectiveness of personalized moderation strategies.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Online Social Networks (OSNs) widely adopt content moderation to mitigate the spread of abusive and toxic discourse. Nonetheless, the real effectiveness of moderation interventions remains unclear due to the high cost of data collection and limited experimental control. The latest developments in Natural Language Processing pave the way for a new evaluation approach. Large Language Models (LLMs) can be successfully leveraged to enhance Agent-Based Modeling and simulate human-like social behavior with unprecedented degree of believability. Yet, existing tools do not support simulation-based evaluation of moderation strategies. We fill this gap by designing a LLM-powered simulator of OSN conversations enabling a parallel, counterfactual simulation where toxic behavior is influenced by moderation interventions, keeping all else equal. We conduct extensive experiments, unveiling the psychological realism of OSN agents, the emergence of social contagion phenomena and the superior effectiveness of personalized moderation strategies.
Dell'Oglio, P.; Bondielli, A.; Marcelloni, F.; Passaro, L. C.
An experimental comparison of the most popular approaches to fake news detection Journal Article
In: INFORMATION SCIENCES, vol. 745, 2026.
@article{ 11568_1356949,
title = {An experimental comparison of the most popular approaches to fake news detection},
author = {P. Dell'Oglio and A. Bondielli and F. Marcelloni and L. C. Passaro},
url = {https://www.sciencedirect.com/science/article/abs/pii/S0020025526003385},
doi = {10.1016/j.ins.2026.123407},
year = {2026},
date = {2026-01-01},
journal = {INFORMATION SCIENCES},
volume = {745},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Carta, Antonio; Marinelli, Alberto Roberto; Passaro, Lucia C.
Adapting language models with continual learning for temporal drifts Journal Article
In: EVOLVING SYSTEMS, vol. 17, no. 2, 2026.
@article{ 11568_1356948,
title = {Adapting language models with continual learning for temporal drifts},
author = {Antonio Carta and Alberto Roberto Marinelli and Lucia C. Passaro},
url = {https://link.springer.com/article/10.1007/s12530-026-09817-x},
doi = {10.1007/s12530-026-09817-x},
year = {2026},
date = {2026-01-01},
journal = {EVOLVING SYSTEMS},
volume = {17},
number = {2},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
2025
Capone, Luca; Bondielli, Alessandro; Lenci, Alessandro
CLASS-IT: Conversational and Lecture-Aligned Small-Scale Instruction Tuning for BabyLMs Conference
Proceedings of the First BabyLM Workshop, Association for Computational Linguistics, Suzhou, China, 2025, ISBN: TODO.
@conference{capone-etal-2025-class,
title = {CLASS-IT: Conversational and Lecture-Aligned Small-Scale Instruction Tuning for BabyLMs},
author = {Luca Capone and Alessandro Bondielli and Alessandro Lenci},
editor = {Lucas Charpentier and Leshem Choshen and Ryan Cotterell and Mustafa Omer Gul and Michael Y. Hu and Jing Liu and Jaap Jumelet and Tal Linzen and Aaron Mueller and Candace Ross and Raj Sanjay Shah and Alex Warstadt and Ethan Gotlieb Wilcox and Adina Williams},
url = {https://aclanthology.org/2025.babylm-main.30/},
doi = {10.18653/v1/2025.babylm-main.30},
isbn = {TODO},
year = {2025},
date = {2025-11-01},
urldate = {2025-11-01},
booktitle = {Proceedings of the First BabyLM Workshop},
pages = {436–444},
publisher = {Association for Computational Linguistics},
address = {Suzhou, China},
abstract = {This work investigates whether small-scale LMs can benefit from instruction tuning (IT). We compare conversational and question–answering IT datasets, applied either in a merged or sequential curriculum, using decoder-only models with 100M and 140M parameters. Evaluation spans both fine-tuning (SuperGLUE) and zero-shot (BLiMP, EWoK, WUGs, entity tracking, and psycholinguistic correlation) settings. Results show that IT yields small but consistent gains in fine-tuning scenarios, with sequential curricula outperforming merged data; however, improvements do not consistently transfer to zero-shot tasks, suggesting a trade-off between interaction-focused adaptation and broad linguistic generalization. These results highlight both the potential and the constraints of adapting human-inspired learning strategies to low-resource LMs, and point toward hybrid, curriculum-based approaches for enhancing generalization under ecological training limits.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
This work investigates whether small-scale LMs can benefit from instruction tuning (IT). We compare conversational and question–answering IT datasets, applied either in a merged or sequential curriculum, using decoder-only models with 100M and 140M parameters. Evaluation spans both fine-tuning (SuperGLUE) and zero-shot (BLiMP, EWoK, WUGs, entity tracking, and psycholinguistic correlation) settings. Results show that IT yields small but consistent gains in fine-tuning scenarios, with sequential curricula outperforming merged data; however, improvements do not consistently transfer to zero-shot tasks, suggesting a trade-off between interaction-focused adaptation and broad linguistic generalization. These results highlight both the potential and the constraints of adapting human-inspired learning strategies to low-resource LMs, and point toward hybrid, curriculum-based approaches for enhancing generalization under ecological training limits.
Bondielli, Alessandro; Miliani, Martina; Paglione, Luca; Auriemma, Serena; Passaro, Lucia C.; Lenci, Alessandro
LLMs Struggle on Explicit Causality in Italian Conference
Proceedings of the Eleventh Italian Conference on Computational Linguistics (CLiC-it 2025), CEUR Workshop Proceedings, Cagliari, Italy, 2025, ISBN: 979-12-243-0587-3.
@conference{bondielli-etal-2025-llms,
title = {LLMs Struggle on Explicit Causality in Italian},
author = {Alessandro Bondielli and Martina Miliani and Luca Paglione and Serena Auriemma and Lucia C. Passaro and Alessandro Lenci},
editor = {Cristina Bosco and Elisabetta Jezek and Marco Polignano and Manuela Sanguinetti},
url = {https://aclanthology.org/2025.clicit-1.10/},
isbn = {979-12-243-0587-3},
year = {2025},
date = {2025-09-01},
urldate = {2025-09-01},
booktitle = {Proceedings of the Eleventh Italian Conference on Computational Linguistics (CLiC-it 2025)},
pages = {83–94},
publisher = {CEUR Workshop Proceedings},
address = {Cagliari, Italy},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Proietti, Mattia; Passaro, Lucia C.; Lenci, Alessandro
Proceedings of the Eleventh Italian Conference on Computational Linguistics (CLiC-it 2025), CEUR Workshop Proceedings, Cagliari, Italy, 2025, ISBN: 979-12-243-0587-3.
@conference{proietti-etal-2025-leveraging,
title = {Leveraging LLMs to Build a Semi-synthetic Dataset for Legal Information Retrieval: A Case Study on the Italian Civil Code and GPT4-O},
author = {Mattia Proietti and Lucia C. Passaro and Alessandro Lenci},
editor = {Cristina Bosco and Elisabetta Jezek and Marco Polignano and Manuela Sanguinetti},
url = {https://aclanthology.org/2025.clicit-1.87/},
isbn = {979-12-243-0587-3},
year = {2025},
date = {2025-09-01},
urldate = {2025-09-01},
booktitle = {Proceedings of the Eleventh Italian Conference on Computational Linguistics (CLiC-it 2025)},
pages = {933–941},
publisher = {CEUR Workshop Proceedings},
address = {Cagliari, Italy},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Lombardi, Agnese; Lenci, Alessandro
Doing Things with Words: Rethinking Theory of Mind Simulation in Large Language Models Conference
Proceedings of the Eleventh Italian Conference on Computational Linguistics (CLiC-it 2025), CEUR Workshop Proceedings, Cagliari, Italy, 2025, ISBN: 979-12-243-0587-3.
@conference{lombardi-lenci-2025-things,
title = {Doing Things with Words: Rethinking Theory of Mind Simulation in Large Language Models},
author = {Agnese Lombardi and Alessandro Lenci},
editor = {Cristina Bosco and Elisabetta Jezek and Marco Polignano and Manuela Sanguinetti},
url = {https://aclanthology.org/2025.clicit-1.59/},
isbn = {979-12-243-0587-3},
year = {2025},
date = {2025-09-01},
urldate = {2025-09-01},
booktitle = {Proceedings of the Eleventh Italian Conference on Computational Linguistics (CLiC-it 2025)},
pages = {613–624},
publisher = {CEUR Workshop Proceedings},
address = {Cagliari, Italy},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Radaelli, Matteo; Chersoni, Emmanuele; Lenci, Alessandro; Baggio, Giosuè
Proceedings of the 16th International Conference on Computational Semantics, Association for Computational Linguistics, Düsseldorf, Germany, 2025, ISBN: 979-8-89176-316-6.
@conference{radaelli-etal-2025-context,
title = {Context Effects on the Interpretation of Complement Coercion: A Comparative Study with Language Models in Norwegian},
author = {Matteo Radaelli and Emmanuele Chersoni and Alessandro Lenci and Giosuè Baggio},
editor = {Kilian Evang and Laura Kallmeyer and Sylvain Pogodalla},
url = {https://aclanthology.org/2025.iwcs-main.7/},
isbn = {979-8-89176-316-6},
year = {2025},
date = {2025-09-01},
urldate = {2025-09-01},
booktitle = {Proceedings of the 16th International Conference on Computational Semantics},
pages = {78–88},
publisher = {Association for Computational Linguistics},
address = {Düsseldorf, Germany},
abstract = {In complement coercion sentences, like *John began the book*, a covert event (e.g., reading) may be recovered based on lexical meanings, world knowledge, and context. We investigate how context influences coercion interpretation performance for 17 language models (LMs) in Norwegian, a low-resource language. Our new dataset contained isolated coercion sentences (context-neutral), plus the same sentences with a subject NP that suggests a particular covert event and sentences that have a similar effect but that precede or follow the coercion sentence. LMs generally benefit from contextual enrichment, but performance varies depending on the model. Models that struggled in context-neutral sentences showed greater improvements from contextual enrichment. Subject NPs and pre-coercion sentences had the largest effect in facilitating coercion interpretation.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
In complement coercion sentences, like *John began the book*, a covert event (e.g., reading) may be recovered based on lexical meanings, world knowledge, and context. We investigate how context influences coercion interpretation performance for 17 language models (LMs) in Norwegian, a low-resource language. Our new dataset contained isolated coercion sentences (context-neutral), plus the same sentences with a subject NP that suggests a particular covert event and sentences that have a similar effect but that precede or follow the coercion sentence. LMs generally benefit from contextual enrichment, but performance varies depending on the model. Models that struggled in context-neutral sentences showed greater improvements from contextual enrichment. Subject NPs and pre-coercion sentences had the largest effect in facilitating coercion interpretation.
Miliani, Martina; Auriemma, Serena; Bondielli, Alessandro; Chersoni, Emmanuele; Passaro, Lucia; Sucameli, Irene; Lenci, Alessandro
ExpliCa: Evaluating Explicit Causal Reasoning in Large Language Models Conference
Findings of the Association for Computational Linguistics: ACL 2025, Association for Computational Linguistics, Vienna, Austria, 2025, ISBN: 979-8-89176-256-5.
@conference{miliani-etal-2025-explica,
title = {ExpliCa: Evaluating Explicit Causal Reasoning in Large Language Models},
author = {Martina Miliani and Serena Auriemma and Alessandro Bondielli and Emmanuele Chersoni and Lucia Passaro and Irene Sucameli and Alessandro Lenci},
editor = {Wanxiang Che and Joyce Nabende and Ekaterina Shutova and Mohammad Taher Pilehvar},
url = {https://aclanthology.org/2025.findings-acl.891/},
doi = {10.18653/v1/2025.findings-acl.891},
isbn = {979-8-89176-256-5},
year = {2025},
date = {2025-07-01},
urldate = {2025-07-01},
booktitle = {Findings of the Association for Computational Linguistics: ACL 2025},
pages = {17335–17355},
publisher = {Association for Computational Linguistics},
address = {Vienna, Austria},
abstract = {Large Language Models (LLMs) are increasingly used in tasks requiring interpretive and inferential accuracy. In this paper, we introduce ExpliCa, a new dataset for evaluating LLMs in explicit causal reasoning. ExpliCa uniquely integrates both causal and temporal relations presented in different linguistic orders and explicitly expressed by linguistic connectives. The dataset is enriched with crowdsourced human acceptability ratings. We tested LLMs on ExpliCa through prompting and perplexity-based metrics. We assessed seven commercial and open-source LLMs, revealing that even top models struggle to reach 0.80 accuracy. Interestingly, models tend to confound temporal relations with causal ones, and their performance is also strongly influenced by the linguistic order of the events. Finally, perplexity-based scores and prompting performance are differently affected by model size.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Large Language Models (LLMs) are increasingly used in tasks requiring interpretive and inferential accuracy. In this paper, we introduce ExpliCa, a new dataset for evaluating LLMs in explicit causal reasoning. ExpliCa uniquely integrates both causal and temporal relations presented in different linguistic orders and explicitly expressed by linguistic connectives. The dataset is enriched with crowdsourced human acceptability ratings. We tested LLMs on ExpliCa through prompting and perplexity-based metrics. We assessed seven commercial and open-source LLMs, revealing that even top models struggle to reach 0.80 accuracy. Interestingly, models tend to confound temporal relations with causal ones, and their performance is also strongly influenced by the linguistic order of the events. Finally, perplexity-based scores and prompting performance are differently affected by model size.
Radaelli, Matteo; Chersoni, Emmanuele; Lenci, Alessandro; Baggio, Giosuè
Proceedings of the 29th Conference on Computational Natural Language Learning, Association for Computational Linguistics, Vienna, Austria, 2025, ISBN: 979-8-89176-271-8.
@conference{radaelli-etal-2025-compositionality,
title = {Compositionality and Event Retrieval in Complement Coercion: A Study of Language Models in a Low-resource Setting},
author = {Matteo Radaelli and Emmanuele Chersoni and Alessandro Lenci and Giosuè Baggio},
editor = {Gemma Boleda and Michael Roth},
url = {https://aclanthology.org/2025.conll-1.31/},
doi = {10.18653/v1/2025.conll-1.31},
isbn = {979-8-89176-271-8},
year = {2025},
date = {2025-07-01},
urldate = {2025-07-01},
booktitle = {Proceedings of the 29th Conference on Computational Natural Language Learning},
pages = {469–480},
publisher = {Association for Computational Linguistics},
address = {Vienna, Austria},
abstract = {In sentences such as John began the book, the complement noun, lexically denoting an entity, is interpreted as an event. This phenomenon is known in linguistics as complement coercion: the event associated with the verb is not overtly expressed but can be recovered from the meanings of other constituents, context and world knowledge. We investigate whether language models (LMs) can exploit sentence structure and compositional meaning to recover plausible events in complement coercion. For the first time, we tested different LMs in Norwegian, a low-resource language with high syntactic variation in coercion constructions across aspectual verbs. Results reveal that LMs struggle with retrieving plausible events and with ranking them above less plausible ones. Moreover, we found that LMs do not exploit the compositional properties of coercion sentences in their predictions.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
In sentences such as John began the book, the complement noun, lexically denoting an entity, is interpreted as an event. This phenomenon is known in linguistics as complement coercion: the event associated with the verb is not overtly expressed but can be recovered from the meanings of other constituents, context and world knowledge. We investigate whether language models (LMs) can exploit sentence structure and compositional meaning to recover plausible events in complement coercion. For the first time, we tested different LMs in Norwegian, a low-resource language with high syntactic variation in coercion constructions across aspectual verbs. Results reveal that LMs struggle with retrieving plausible events and with ranking them above less plausible ones. Moreover, we found that LMs do not exploit the compositional properties of coercion sentences in their predictions.
Tosi, Filippo; Arzilli, Guglielmo; Baglivo, Francesco; Fondelli, Serena; Renda, Alessandro; Bondielli, Alessandro; Dell’Oglio, Pietro; Acampora, Vittoria; Marcelloni, Francesco; Rizzo, Caterina
Atti 58° Congresso Nazionale SItI, 2025.
@conference{ 11568_1329807,
title = {Un approccio basato su NLP per la sorveglianza delle infezioni del sito chirurgico utilizzando le lettere di dimissione ospedaliera},
author = {Filippo Tosi and Guglielmo Arzilli and Francesco Baglivo and Serena Fondelli and Alessandro Renda and Alessandro Bondielli and Pietro Dell’Oglio and Vittoria Acampora and Francesco Marcelloni and Caterina Rizzo},
url = {https://siti2025.it/Download/EasyCms/SITI_ott_2025_provvisorio_26307.pdf},
year = {2025},
date = {2025-01-01},
booktitle = {Atti 58° Congresso Nazionale SItI},
pages = {438–439},
abstract = {Background
Le infezioni del sito chirurgico (ISC) rappresentano una per- centuale significativa delle infezioni associate all’assistenza sanitaria e costituiscono un onere importante in ambito ospe- daliero. La sorveglianza tradizionale delle ISC, basata sulla re- visione manuale delle cartelle cliniche a testo libero, richiede molto tempo ed è scarsamente modulabile. Questo studio si propone di sviluppare e valutare un sistema di sorveglianza semi-automatizzato basato sull’intelligenza artificiale (AI) per il rilevamento delle ISC utilizzando le cartelle cliniche in lin- gua italiana di un grande ospedale universitario.
Metodi
È stata condotta un’analisi retrospettiva sulle lettere di dimissio- ne ospedaliera non strutturate e anonime dell’Ospedale di Pisa (Italia). Il testo clinico è stato elaborato e vettorizzato utilizzan- do TF-IDF, Word2Vec e BERT (incluso il pre-training e il fine tuning). I classificatori comprendevano regressione logistica, Decision Trees, random forest e XGBoost. Le prestazioni finali sono state valutate utilizzando la convalida incrociata stratificata a 10 volte, il punteggio F1 e l’AUC. Dato lo sbilanciamento 1:100 tra record ISC e non ISC, sono state testate tecniche di oversampling and undersampling. Inoltre, sono stati valutati due modelli linguistici di grandi dimensioni (LLM) - OpenBioLLM e Minerva - utilizzando prompting a zero e a pochi colpi.
Risultati Le migliori prestazioni sono state ottenute da un modello BERT ulteriormente pre-addestrato sul corpus specifico del dominio e messo a punto per la classificazione (BERT-FT), con XGBoost come classificatore: F1-score=0,79},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Background
Le infezioni del sito chirurgico (ISC) rappresentano una per- centuale significativa delle infezioni associate all’assistenza sanitaria e costituiscono un onere importante in ambito ospe- daliero. La sorveglianza tradizionale delle ISC, basata sulla re- visione manuale delle cartelle cliniche a testo libero, richiede molto tempo ed è scarsamente modulabile. Questo studio si propone di sviluppare e valutare un sistema di sorveglianza semi-automatizzato basato sull’intelligenza artificiale (AI) per il rilevamento delle ISC utilizzando le cartelle cliniche in lin- gua italiana di un grande ospedale universitario.
Metodi
È stata condotta un’analisi retrospettiva sulle lettere di dimissio- ne ospedaliera non strutturate e anonime dell’Ospedale di Pisa (Italia). Il testo clinico è stato elaborato e vettorizzato utilizzan- do TF-IDF, Word2Vec e BERT (incluso il pre-training e il fine tuning). I classificatori comprendevano regressione logistica, Decision Trees, random forest e XGBoost. Le prestazioni finali sono state valutate utilizzando la convalida incrociata stratificata a 10 volte, il punteggio F1 e l’AUC. Dato lo sbilanciamento 1:100 tra record ISC e non ISC, sono state testate tecniche di oversampling and undersampling. Inoltre, sono stati valutati due modelli linguistici di grandi dimensioni (LLM) - OpenBioLLM e Minerva - utilizzando prompting a zero e a pochi colpi.
Risultati Le migliori prestazioni sono state ottenute da un modello BERT ulteriormente pre-addestrato sul corpus specifico del dominio e messo a punto per la classificazione (BERT-FT), con XGBoost come classificatore: F1-score=0,79
Le infezioni del sito chirurgico (ISC) rappresentano una per- centuale significativa delle infezioni associate all’assistenza sanitaria e costituiscono un onere importante in ambito ospe- daliero. La sorveglianza tradizionale delle ISC, basata sulla re- visione manuale delle cartelle cliniche a testo libero, richiede molto tempo ed è scarsamente modulabile. Questo studio si propone di sviluppare e valutare un sistema di sorveglianza semi-automatizzato basato sull’intelligenza artificiale (AI) per il rilevamento delle ISC utilizzando le cartelle cliniche in lin- gua italiana di un grande ospedale universitario.
Metodi
È stata condotta un’analisi retrospettiva sulle lettere di dimissio- ne ospedaliera non strutturate e anonime dell’Ospedale di Pisa (Italia). Il testo clinico è stato elaborato e vettorizzato utilizzan- do TF-IDF, Word2Vec e BERT (incluso il pre-training e il fine tuning). I classificatori comprendevano regressione logistica, Decision Trees, random forest e XGBoost. Le prestazioni finali sono state valutate utilizzando la convalida incrociata stratificata a 10 volte, il punteggio F1 e l’AUC. Dato lo sbilanciamento 1:100 tra record ISC e non ISC, sono state testate tecniche di oversampling and undersampling. Inoltre, sono stati valutati due modelli linguistici di grandi dimensioni (LLM) - OpenBioLLM e Minerva - utilizzando prompting a zero e a pochi colpi.
Risultati Le migliori prestazioni sono state ottenute da un modello BERT ulteriormente pre-addestrato sul corpus specifico del dominio e messo a punto per la classificazione (BERT-FT), con XGBoost come classificatore: F1-score=0,79
Mala, Chandana Sree; Maio, Christian Di; Proietti, Mattia; Gezici, Gizem; Giannotti, Fosca; Melacci, Stefano; Lenci, Alessandro; Gori, Marco
Towards Building a Trustworthy RAG-Based Chatbot for the Italian Public Administration Journal Article
In: Frontiers in Artificial Intelligence and Applications, vol. 408, pp. 196 – 204, 2025.
@article{Mala2025196,
title = {Towards Building a Trustworthy RAG-Based Chatbot for the Italian Public Administration},
author = {Chandana Sree Mala and Christian Di Maio and Mattia Proietti and Gizem Gezici and Fosca Giannotti and Stefano Melacci and Alessandro Lenci and Marco Gori},
editor = {Pedreschi D. and Milano M. and Tiddi I. and Russell S. and Boldrini C. and Pappalardo L. and Passerini A. and Wang S.},
url = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-105020935873&doi=10.3233%2fFAIA250637&partnerID=40&md5=590a91bfd34f9af45a6c2e4b8990b204},
doi = {10.3233/FAIA250637},
year = {2025},
date = {2025-01-01},
journal = {Frontiers in Artificial Intelligence and Applications},
volume = {408},
pages = {196 – 204},
publisher = {IOS Press BV},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Auriemma, Serena; Miliani, Martina; Madeddu, Mauro; Bondielli, Alessandro; Passaro, Lucia; Lenci, Alessandro
Prompting encoder models for zero-shot classification: a cross-domain study in Italian Journal Article
In: LANGUAGE RESOURCES AND EVALUATION, 2025.
@article{ 11568_1321427,
title = {Prompting encoder models for zero-shot classification: a cross-domain study in Italian},
author = {Serena Auriemma and Martina Miliani and Mauro Madeddu and Alessandro Bondielli and Lucia Passaro and Alessandro Lenci},
url = {https://link.springer.com/article/10.1007/s10579-025-09853-0},
doi = {10.1007/s10579-025-09853-0},
year = {2025},
date = {2025-01-01},
journal = {LANGUAGE RESOURCES AND EVALUATION},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Muscato, Benedetta; Passaro, Lucia; Gezici, Gizem; Giannotti, Fosca
Perspectives in Play: A Multi-Perspective Approach for More Inclusive NLP Systems Conference
Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, IJCAI-25, International Joint Conferences on Artificial Intelligence, 2025, ISBN: 978-1-956792-06-5.
@conference{ 11568_1322168,
title = {Perspectives in Play: A Multi-Perspective Approach for More Inclusive NLP Systems},
author = {Benedetta Muscato and Lucia Passaro and Gizem Gezici and Fosca Giannotti},
url = {https://doi.org/10.24963/ijcai.2025/1092},
doi = {10.24963/ijcai.2025/1092},
isbn = {978-1-956792-06-5},
year = {2025},
date = {2025-01-01},
booktitle = {Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, IJCAI-25},
pages = {9827–9835},
publisher = {International Joint Conferences on Artificial Intelligence},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Lenci, Alessandro; Auriemma, Serena; Miliani, Martina
Linguistica computazionale. Natural Language Processing e Intelligenza Artificiale Book
Hoepli, Milano, 2025, ISBN: 978-88-360-1830-7, (Hoepli Academy).
@book{lenci_auriemma_miliani_2025,
title = {Linguistica computazionale. Natural Language Processing e Intelligenza Artificiale},
author = {Alessandro Lenci and Serena Auriemma and Martina Miliani},
url = {https://www.hoeplieditore.it/universita/articolo/linguistica-computazionale-alessandro-lenci/9788836018307/3144},
isbn = {978-88-360-1830-7},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
publisher = {Hoepli},
address = {Milano},
abstract = {Il volume presenta un quadro generale dei principi teorici e delle metodologie pratiche della linguistica computazionale, e fornisce gli strumenti per la corretta comprensione e implementazione dei modelli di machine learning e di intelligenza artificiale per l’analisi del linguaggio naturale. Il testo si articola in tre parti: la prima è una panoramica delle nozioni fondamentali, la seconda si focalizza sui metodi di machine learning per il natural language processing (NLP), sia sui modelli più tradizionali, sia su quelli basati su reti neurali; la terza parte è dedicata ai modelli fondazionali, che rappresentano l’approccio metodologico più recente nell’ambito del NLP. La trattazione teorica è integrata da approfondimenti su temi applicativi delle tecnologie in contesti reali e da risorse online come tutorial per l’apprendimento del linguaggio di programmazione Python applicato al NLP. Al termine di ciascun capitolo sono previste domande sui principali argomenti trattati per l’autovalutazione delle conoscenze acquisite. La sequenza degli argomenti segue un percorso graduale di sviluppo dai concetti di base fino ai temi più avanzati; tuttavia, ogni capitolo è concepito come un’unità autonoma, offrendo così maggiore flessibilità nell’organizzazione dello studio.},
note = {Hoepli Academy},
keywords = {},
pubstate = {published},
tppubtype = {book}
}
Il volume presenta un quadro generale dei principi teorici e delle metodologie pratiche della linguistica computazionale, e fornisce gli strumenti per la corretta comprensione e implementazione dei modelli di machine learning e di intelligenza artificiale per l’analisi del linguaggio naturale. Il testo si articola in tre parti: la prima è una panoramica delle nozioni fondamentali, la seconda si focalizza sui metodi di machine learning per il natural language processing (NLP), sia sui modelli più tradizionali, sia su quelli basati su reti neurali; la terza parte è dedicata ai modelli fondazionali, che rappresentano l’approccio metodologico più recente nell’ambito del NLP. La trattazione teorica è integrata da approfondimenti su temi applicativi delle tecnologie in contesti reali e da risorse online come tutorial per l’apprendimento del linguaggio di programmazione Python applicato al NLP. Al termine di ciascun capitolo sono previste domande sui principali argomenti trattati per l’autovalutazione delle conoscenze acquisite. La sequenza degli argomenti segue un percorso graduale di sviluppo dai concetti di base fino ai temi più avanzati; tuttavia, ogni capitolo è concepito come un’unità autonoma, offrendo così maggiore flessibilità nell’organizzazione dello studio.
Baccheschi, Corrado; Bondielli, Alessandro; Lenci, Alessandro; Micheli, Alessio; Passaro, Lucia; Podda, Marco; Tortorella, Domenico
Investigating Time-Scales in Deep Echo State Networks for Natural Language Processing Book Chapter
In: Artificial Neural Networks and Machine Learning. ICANN 2025 International Workshops and Special Sessions, vol. 16072, pp. 188–200, Springer, 2025, ISBN: 9783032045522.
@inbook{<LineBreak> 11568_1324255,
title = {Investigating Time-Scales in Deep Echo State Networks for Natural Language Processing},
author = {Corrado Baccheschi and Alessandro Bondielli and Alessandro Lenci and Alessio Micheli and Lucia Passaro and Marco Podda and Domenico Tortorella},
url = {https://link.springer.com/chapter/10.1007/978-3-032-04552-2_18},
doi = {10.1007/978-3-032-04552-2_18},
isbn = {9783032045522},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
booktitle = {Artificial Neural Networks and Machine Learning. ICANN 2025 International Workshops and Special Sessions},
volume = {16072},
pages = {188–200},
publisher = {Springer},
keywords = {},
pubstate = {published},
tppubtype = {inbook}
}
Muscato, Benedetta; Bushipaka, Praveen; Gezici, Gizem; Passaro, Lucia; Giannotti, Fosca; Cucinotta, Tommaso
Embracing Diversity: A Multi-Perspective Approach with Soft Labels Conference
Proceedings of HHAI 2025: The 4th International Conference Series on Hybrid Human-Artificial Intelligence, IOS Press, Amsterdam, 2025, ISBN: 978-1-64368-611-0.
@conference{ 11568_1322167,
title = {Embracing Diversity: A Multi-Perspective Approach with Soft Labels},
author = {Benedetta Muscato and Praveen Bushipaka and Gizem Gezici and Lucia Passaro and Fosca Giannotti and Tommaso Cucinotta},
url = {https://ebooks.iospress.nl/doi/10.3233/FAIA250654},
doi = {10.3233/FAIA250654},
isbn = {978-1-64368-611-0},
year = {2025},
date = {2025-01-01},
booktitle = {Proceedings of HHAI 2025: The 4th International Conference Series on Hybrid Human-Artificial Intelligence},
pages = {370–384},
publisher = {IOS Press},
address = {Amsterdam},
abstract = {In subjective tasks like stance detection, diverse human perspectives are often simplified into a single ground truth through label aggregation i.e. majority voting, potentially marginalizing minority viewpoints. This paper presents a Multi-Perspective framework for stance detection that explicitly incorporates annotation diversity by using soft labels derived from both human and large language model (LLM) annotations. Building on a stance detection dataset focused on controversial topics, we augment it with document summaries and new LLM-generated labels.
We then compare two approaches: a baseline using aggregated hard labels, and a multi-perspective model trained on disaggregated soft labels that capture annotation distributions. Our findings show that multi-perspective models consistently outperform traditional baselines (higher F1-scores), with lower model confidence, reflecting task subjectivity. This work highlights the importance of modeling disagreement and promotes a shift toward more inclusive, perspective-aware NLP systems.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
In subjective tasks like stance detection, diverse human perspectives are often simplified into a single ground truth through label aggregation i.e. majority voting, potentially marginalizing minority viewpoints. This paper presents a Multi-Perspective framework for stance detection that explicitly incorporates annotation diversity by using soft labels derived from both human and large language model (LLM) annotations. Building on a stance detection dataset focused on controversial topics, we augment it with document summaries and new LLM-generated labels.
We then compare two approaches: a baseline using aggregated hard labels, and a multi-perspective model trained on disaggregated soft labels that capture annotation distributions. Our findings show that multi-perspective models consistently outperform traditional baselines (higher F1-scores), with lower model confidence, reflecting task subjectivity. This work highlights the importance of modeling disagreement and promotes a shift toward more inclusive, perspective-aware NLP systems.
We then compare two approaches: a baseline using aggregated hard labels, and a multi-perspective model trained on disaggregated soft labels that capture annotation distributions. Our findings show that multi-perspective models consistently outperform traditional baselines (higher F1-scores), with lower model confidence, reflecting task subjectivity. This work highlights the importance of modeling disagreement and promotes a shift toward more inclusive, perspective-aware NLP systems.
Taddei, Andrea; Lenci, Alessandro; D'Angelo, Caterina
Detecting Semantic Reuse in Ancient Greek Literature: AComputational Approach. Conference
Proceedings of the Eleventh Italian Conference on Computational Linguistics (CLiC-it 2025), 2025.
@conference{<LineBreak> 11568_1331500,
title = {Detecting Semantic Reuse in Ancient Greek Literature: AComputational Approach.},
author = {Andrea Taddei and Alessandro Lenci and Caterina D'Angelo},
editor = {Cristina Bosco and Elisabetta Jezek and Marco Polignano and Manuela Sanguinetti},
url = {https://clic2025.unica.it/wp-content/uploads/2025/09/33_main_long.pdf},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
booktitle = {Proceedings of the Eleventh Italian Conference on Computational Linguistics (CLiC-it 2025)},
pages = {1–10},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Rebillard, C.; Hurtado, J.; Krutsylo, A.; Passaro, L.; Lomonaco, V.
Continually learn to map visual concepts to language models in resource-constrained environments Journal Article
In: Neurocomputing, vol. 652, 2025.
@article{Rebillard2025,
title = {Continually learn to map visual concepts to language models in resource-constrained environments},
author = {C. Rebillard and J. Hurtado and A. Krutsylo and L. Passaro and V. Lomonaco},
url = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-105011582431&doi=10.1016%2fj.neucom.2025.131013&partnerID=40&md5=d9d8b55dad71120e084ef332e6948f8f},
doi = {10.1016/j.neucom.2025.131013},
year = {2025},
date = {2025-01-01},
journal = {Neurocomputing},
volume = {652},
abstract = {Continually learning from non-independent and identically distributed (non-i.i.d.) data poses a significant challenge in deep learning, particularly in resource-constrained environments. Visual models trained via supervised learning often suffer from overfitting, catastrophic forgetting, and biased representations when faced with sequential tasks. In contrast, pre-trained language models demonstrate greater robustness in managing task sequences due to their generalized knowledge representations, albeit at the cost of high computational resources. Leveraging this advantage, we propose a novel learning strategy, Continual Visual Mapping (CVM), which continuously maps visual representations into a fixed knowledge space derived from a language model. By anchoring learning to this fixed space, CVM enables training small, efficient visual models, making it particularly suited for scenarios where adapting large pre-trained visual models is computationally or data-prohibitive. Empirical evaluations across five benchmarks demonstrate that CVM consistently outperforms state-of-the-art continual learning methods, showcasing its potential to enhance generalization and mitigate challenges in resource-constrained continual learning settings. © 2025 The Authors},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Continually learning from non-independent and identically distributed (non-i.i.d.) data poses a significant challenge in deep learning, particularly in resource-constrained environments. Visual models trained via supervised learning often suffer from overfitting, catastrophic forgetting, and biased representations when faced with sequential tasks. In contrast, pre-trained language models demonstrate greater robustness in managing task sequences due to their generalized knowledge representations, albeit at the cost of high computational resources. Leveraging this advantage, we propose a novel learning strategy, Continual Visual Mapping (CVM), which continuously maps visual representations into a fixed knowledge space derived from a language model. By anchoring learning to this fixed space, CVM enables training small, efficient visual models, making it particularly suited for scenarios where adapting large pre-trained visual models is computationally or data-prohibitive. Empirical evaluations across five benchmarks demonstrate that CVM consistently outperforms state-of-the-art continual learning methods, showcasing its potential to enhance generalization and mitigate challenges in resource-constrained continual learning settings. © 2025 The Authors
Bouchardet, Cecília Carvalho; Gonçalves, Kênia Cristina; Passaro, Lucia; Almeida, Jussara Maria; Marques-Neto, Humberto Torres
Call2Go: A Cross-Platform Strategy Of Kids Online Social Media Influencers Conference
Proceedings of the 31st Brazilian Symposium on Multimedia and the Web, Sociedade Brasileira de Computação (SBC), 2025.
@conference{ 11568_1331522,
title = {Call2Go: A Cross-Platform Strategy Of Kids Online Social Media Influencers},
author = {Cecília Carvalho Bouchardet and Kênia Cristina Gonçalves and Lucia Passaro and Jussara Maria Almeida and Humberto Torres Marques-Neto},
url = {https://sol.sbc.org.br/index.php/webmedia/article/view/37953},
doi = {10.5753/webmedia.2025.16108},
year = {2025},
date = {2025-01-01},
booktitle = {Proceedings of the 31st Brazilian Symposium on Multimedia and the Web},
pages = {112–120},
publisher = {Sociedade Brasileira de Computação (SBC)},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Suozzi, A.; Capone, L.; Lebani, G. E.; Lenci, A.
BAMBI: DEVELOPING BABY LANGUAGE MODELS FOR ITALIAN Journal Article
In: LINGUE E LINGUAGGIO, vol. 24, no. 1, pp. 83–102, 2025.
@article{ 11568_1327954,
title = {BAMBI: DEVELOPING BABY LANGUAGE MODELS FOR ITALIAN},
author = {A. Suozzi and L. Capone and G. E. Lebani and A. Lenci},
url = {https://rivisteweb.it/doi/10.1418/117444},
doi = {10.1418/117444},
year = {2025},
date = {2025-01-01},
journal = {LINGUE E LINGUAGGIO},
volume = {24},
number = {1},
pages = {83–102},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Capone, Luca; Suozzi, Alice; Lebani, Gianluca E; Lenci, Alessandro
BAMBI Goes to School: Evaluating Italian BabyLMs with Invalsi-ITA Conference
Proceedings of the Eleventh Italian Conference on Computational Linguistics (CLiC-it 2025), Università di Cagliari, 2025.
@conference{ 11568_1327955.2,
title = {BAMBI Goes to School: Evaluating Italian BabyLMs with Invalsi-ITA},
author = {Luca Capone and Alice Suozzi and Gianluca E Lebani and Alessandro Lenci},
url = {https://clic2025.unica.it/wp-content/uploads/2025/09/15_main_long.pdf},
year = {2025},
date = {2025-01-01},
booktitle = {Proceedings of the Eleventh Italian Conference on Computational Linguistics (CLiC-it 2025)},
pages = {1–12},
publisher = {Università di Cagliari},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Arzilli, G; Baglivo, F; Angelis, L De; Casigliani, V; Renda, A; Bondielli, A; Dell'Oglio, P; Acampora, V; Marcelloni, F; Rizzo, C
An NLP-Based Approach for Surgical Site Infection Surveillance Using Hospital Discharge Letters Journal Article
In: EUROPEAN JOURNAL OF PUBLIC HEALTH, vol. 35, no. Supplement_4, 2025.
@article{ 11568_1334567,
title = {An NLP-Based Approach for Surgical Site Infection Surveillance Using Hospital Discharge Letters},
author = {G Arzilli and F Baglivo and L De Angelis and V Casigliani and A Renda and A Bondielli and P Dell'Oglio and V Acampora and F Marcelloni and C Rizzo},
url = {https://academic.oup.com/eurpub/article/35/Supplement_4/ckaf161.1037/8303229},
doi = {10.1093/eurpub/ckaf161.1037},
year = {2025},
date = {2025-01-01},
journal = {EUROPEAN JOURNAL OF PUBLIC HEALTH},
volume = {35},
number = {Supplement_4},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Testa, Davide; Bonetta, Giovanni; Bernardi, Raffaella; Bondielli, Alessandro; Lenci, Alessandro; Miaschi, Alessio; Passaro, Lucia; Magnini, Bernardo
All-in-one: Understanding and Generation in Multimodal Reasoning with the MAIA Benchmark Conference
Findings of the Association for Computational Linguistics: EMNLP 2025, Association for Computational Linguistics, 2025, ISBN: 979-8-89176-335-7.
@conference{ 11568_1331520,
title = {All-in-one: Understanding and Generation in Multimodal Reasoning with the MAIA Benchmark},
author = {Davide Testa and Giovanni Bonetta and Raffaella Bernardi and Alessandro Bondielli and Alessandro Lenci and Alessio Miaschi and Lucia Passaro and Bernardo Magnini},
url = {https://aclanthology.org/2025.findings-emnlp.1091/},
doi = {10.18653/v1/2025.findings-emnlp.1091},
isbn = {979-8-89176-335-7},
year = {2025},
date = {2025-01-01},
booktitle = {Findings of the Association for Computational Linguistics: EMNLP 2025},
pages = {20030–20050},
publisher = {Association for Computational Linguistics},
abstract = {We introduce MAIA (Multimodal AI Assessment), a native-Italian benchmark designed for fine-grained investigation of the reasoning abilities of visual language models on videos. MAIA differs from other available video benchmarks for its design, its reasoning categories, the metric it uses, and the language and culture of the videos. MAIA evaluates Vision Language Models (VLMs) on two aligned tasks: a visual statement verification task, and an open-ended visual question-answering task, both on the same set of video-related questions. It considers twelve reasoning categories that aim to disentangle language and vision relations by highlighting the role of the visual input. Thanks to its carefully taught design, it evaluates VLMs' consistency and visually grounded natural language comprehension and generation simultaneously through an aggregated metric revealing low results that highlight models' fragility. Last but not least, the video collection has been carefully selected to reflect the Italian culture, and the language data are produced by native-speakers.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
We introduce MAIA (Multimodal AI Assessment), a native-Italian benchmark designed for fine-grained investigation of the reasoning abilities of visual language models on videos. MAIA differs from other available video benchmarks for its design, its reasoning categories, the metric it uses, and the language and culture of the videos. MAIA evaluates Vision Language Models (VLMs) on two aligned tasks: a visual statement verification task, and an open-ended visual question-answering task, both on the same set of video-related questions. It considers twelve reasoning categories that aim to disentangle language and vision relations by highlighting the role of the visual input. Thanks to its carefully taught design, it evaluates VLMs' consistency and visually grounded natural language comprehension and generation simultaneously through an aggregated metric revealing low results that highlight models' fragility. Last but not least, the video collection has been carefully selected to reflect the Italian culture, and the language data are produced by native-speakers.
2024
Capone, Luca; Auriemma, Serena; Miliani, Martina; Bondielli, Alessandro; Lenci, Alessandro
Lost in Disambiguation: How Instruction-Tuned LLMs Master Lexical Ambiguity Conference
Proceedings of the Tenth Italian Conference on Computational Linguistics (CLiC-it 2024), CEUR Workshop Proceedings, Pisa, Italy, 2024, ISBN: 979-12-210-7060-6.
@conference{capone-etal-2024-lost,
title = {Lost in Disambiguation: How Instruction-Tuned LLMs Master Lexical Ambiguity},
author = {Luca Capone and Serena Auriemma and Martina Miliani and Alessandro Bondielli and Alessandro Lenci},
editor = {Felice Dell'Orletta and Alessandro Lenci and Simonetta Montemagni and Rachele Sprugnoli},
url = {https://aclanthology.org/2024.clicit-1.19/},
isbn = {979-12-210-7060-6},
year = {2024},
date = {2024-12-01},
booktitle = {Proceedings of the Tenth Italian Conference on Computational Linguistics (CLiC-it 2024)},
pages = {148–156},
publisher = {CEUR Workshop Proceedings},
address = {Pisa, Italy},
abstract = {This paper investigates how decoder-only instruction-tuned LLMs handle lexical ambiguity. Two distinct methodologies are employed: Eliciting rating scores from the model via prompting and analysing the cosine similarity between pairs of polysemous words in context. Ratings and embeddings are obtained by providing pairs of sentences from Haber and Poesio (2021) to the model. These ratings and cosine similarity scores are compared with each other and with the human similarity judgments in the dataset.Surprisingly, the model scores show only a moderate correlation with the subjects' similarity judgments and no correlation with the target word embedding similarities. A vector space anisotropy inspection has also been performed, as a potential source of the experimental results. The analysis reveals that the embedding spaces of two out of the three analyzed models exhibit poor anisotropy, while the third model shows relatively moderate anisotropy compared to previous findings for models with similar architecture (Ethayarajh 2019). These findings offer new insights into the relationship between generation quality and vector representations in decoder-only LLMs.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
This paper investigates how decoder-only instruction-tuned LLMs handle lexical ambiguity. Two distinct methodologies are employed: Eliciting rating scores from the model via prompting and analysing the cosine similarity between pairs of polysemous words in context. Ratings and embeddings are obtained by providing pairs of sentences from Haber and Poesio (2021) to the model. These ratings and cosine similarity scores are compared with each other and with the human similarity judgments in the dataset.Surprisingly, the model scores show only a moderate correlation with the subjects' similarity judgments and no correlation with the target word embedding similarities. A vector space anisotropy inspection has also been performed, as a potential source of the experimental results. The analysis reveals that the embedding spaces of two out of the three analyzed models exhibit poor anisotropy, while the third model shows relatively moderate anisotropy compared to previous findings for models with similar architecture (Ethayarajh 2019). These findings offer new insights into the relationship between generation quality and vector representations in decoder-only LLMs.
Cascione, A.; Cerulli, A.; Manerba, M. M.; Passaro, L. C.
Women's Professions and Targeted Misogyny Online Conference
vol. 3878, 2024.
@conference{Cascione2024,
title = {Women's Professions and Targeted Misogyny Online},
author = {A. Cascione and A. Cerulli and M. M. Manerba and L. C. Passaro},
url = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-85214429769&partnerID=40&md5=f2d3707476fb121b4200010236a6b194},
year = {2024},
date = {2024-01-01},
journal = {CEUR Workshop Proceedings},
volume = {3878},
abstract = {With the increasing popularity of social media platforms, the dissemination of misogynistic content has become more prevalent and challenging to address. In this paper, we investigate the phenomenon of online misogyny on Twitter through the lens of hurtfulness, qualifying its different manifestation in English tweets considering the profession of the targets of misogynistic attacks. By leveraging manual annotation and a BERTweet model trained for fine-grained misogyny identification, we find that specific types of misogynistic speech are more intensely directed towards particular professions. For example, derailing discourse predominantly targets authors and cultural figures, while dominance-oriented speech and sexual harassment are mainly directed at politicians and athletes. Additionally, we use the HurtLex lexicon and ItEM to assign hurtfulness scores to tweets based on different hate speech categories. Our analysis reveals that these scores align with the profession-based distribution of misogynistic speech, highlighting the targeted nature of such attacks. © 2024 CEUR-WS. All rights reserved.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
With the increasing popularity of social media platforms, the dissemination of misogynistic content has become more prevalent and challenging to address. In this paper, we investigate the phenomenon of online misogyny on Twitter through the lens of hurtfulness, qualifying its different manifestation in English tweets considering the profession of the targets of misogynistic attacks. By leveraging manual annotation and a BERTweet model trained for fine-grained misogyny identification, we find that specific types of misogynistic speech are more intensely directed towards particular professions. For example, derailing discourse predominantly targets authors and cultural figures, while dominance-oriented speech and sexual harassment are mainly directed at politicians and athletes. Additionally, we use the HurtLex lexicon and ItEM to assign hurtfulness scores to tweets based on different hate speech categories. Our analysis reveals that these scores align with the profession-based distribution of misogynistic speech, highlighting the targeted nature of such attacks. © 2024 CEUR-WS. All rights reserved.
Gili, J.; Patti, V.; Passaro, L.; Caselli, T.
VeryfIT - Benchmark of Fact-Checked Claims for Italian: A CALAMITA Challenge Conference
vol. 3878, 2024.
@conference{Gili2024,
title = {VeryfIT - Benchmark of Fact-Checked Claims for Italian: A CALAMITA Challenge},
author = {J. Gili and V. Patti and L. Passaro and T. Caselli},
url = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-85214372761&partnerID=40&md5=089c1b0a0891e71e54e849d4a06e77c5},
year = {2024},
date = {2024-01-01},
journal = {CEUR Workshop Proceedings},
volume = {3878},
abstract = {Achieving factual accuracy is a known pending issue for language models. Their design centered around the interactive component of user interaction and the extensive use of “spontaneous” training data, has made them highly adept at conversational tasks but not fully reliable in terms of factual correctness. VeryfIT addresses this issue by evaluating the in-memory factual knowledge of language models on data written by professional fact-checkers, posing it as a true or false question. Topics of the statements vary but most are in specific domains related to the Italian government, policies, and social issues. The task presents several challenges: extracting statements from segments of speeches, determining appropriate contextual relevance both temporally and factually, and ultimately verifying the accuracy of the statements. © 2024 CEUR-WS. All rights reserved.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Achieving factual accuracy is a known pending issue for language models. Their design centered around the interactive component of user interaction and the extensive use of “spontaneous” training data, has made them highly adept at conversational tasks but not fully reliable in terms of factual correctness. VeryfIT addresses this issue by evaluating the in-memory factual knowledge of language models on data written by professional fact-checkers, posing it as a true or false question. Topics of the statements vary but most are in specific domains related to the Italian government, policies, and social issues. The task presents several challenges: extracting statements from segments of speeches, determining appropriate contextual relevance both temporally and factually, and ultimately verifying the accuracy of the statements. © 2024 CEUR-WS. All rights reserved.
Alberto, Roberto Marinelli; Carta, Antonio; Passaro, Lucia
Updating knowledge in Large Language Models: an Empirical Evaluation Conference
Conference Proceedings: 2024 IEEE International Conference on Evolving and Adaptive Intelligent Systems (EAIS), 2024.
@conference{ 11568_1254427.2,
title = {Updating knowledge in Large Language Models: an Empirical Evaluation},
author = {Roberto Marinelli Alberto and Antonio Carta and Lucia Passaro},
url = {https://ieeexplore.ieee.org/document/10570019},
doi = {10.1109/eais58494.2024.10570019},
year = {2024},
date = {2024-01-01},
booktitle = {Conference Proceedings: 2024 IEEE International Conference on Evolving and Adaptive Intelligent Systems (EAIS)},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Marinelli, A. R.; Carta, A.; Passaro, L. C.
Updating knowledge in Large Language Models: An Empirical Evaluation Conference
2024.
@conference{Marinelli2024,
title = {Updating knowledge in Large Language Models: An Empirical Evaluation},
author = {A. R. Marinelli and A. Carta and L. C. Passaro},
url = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-85199261839&doi=10.1109%2fEAIS58494.2024.10570019&partnerID=40&md5=2751592a1b7dff48c02cfea01ae84475},
doi = {10.1109/EAIS58494.2024.10570019},
year = {2024},
date = {2024-01-01},
journal = {IEEE Conference on Evolving and Adaptive Intelligent Systems},
abstract = {Natural Language Processing (NLP) has witnessed a paradigm shift with Large Language Models (LLMs), yet the static knowledge from pre-Training can lead to knowledge obsolescence. This study focuses on the dynamic relationship between LLMs and evolving knowledge, using GPT-2 as a case study. Leveraging an existing framework, we update models with monthly Wikipedia dumps and Wikidata probes, addressing the stability-plasticity trade-off. We introduce a novel synthetic data generation method for experimental control and present SMARTREVIEW, a state-of-The-Art continual learning method. This work advances understanding and methodologies in tackling knowledge obsolescence in evolving language models. © 2024 IEEE.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Natural Language Processing (NLP) has witnessed a paradigm shift with Large Language Models (LLMs), yet the static knowledge from pre-Training can lead to knowledge obsolescence. This study focuses on the dynamic relationship between LLMs and evolving knowledge, using GPT-2 as a case study. Leveraging an existing framework, we update models with monthly Wikipedia dumps and Wikidata probes, addressing the stability-plasticity trade-off. We introduce a novel synthetic data generation method for experimental control and present SMARTREVIEW, a state-of-The-Art continual learning method. This work advances understanding and methodologies in tackling knowledge obsolescence in evolving language models. © 2024 IEEE.
Aru, Giacomo; Emmolo, Nicola; Marzeddu, Simone; Piras, Andrea; Raffi, Jacopo; Passaro, Lucia C.
RoBEXedda: Sexism Detection in Tweets Conference
CEUR Workshop Proceedings, vol. 3740, CEUR-WS, 2024.
@conference{ 11568_1272686,
title = {RoBEXedda: Sexism Detection in Tweets},
author = {Giacomo Aru and Nicola Emmolo and Simone Marzeddu and Andrea Piras and Jacopo Raffi and Lucia C. Passaro},
url = {https://ceur-ws.org/Vol-3740/paper-88.pdf},
year = {2024},
date = {2024-01-01},
booktitle = {CEUR Workshop Proceedings},
volume = {3740},
pages = {942–957},
publisher = {CEUR-WS},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Cerini, L.; Bondielli, A.; Lenci, A.
Representing Abstract Concepts with Images: An Investigation with Large Language Models Conference
Workshop on Cognitive Aspects of the Lexicon, CogALex 2024 at LREC-COLING 2024 - Workshop Proceedings, European Language Resources Association (ELRA), 2024, ISBN: 978-2-493814-45-6.
@conference{ 11568_1256367,
title = {Representing Abstract Concepts with Images: An Investigation with Large Language Models},
author = {L. Cerini and A. Bondielli and A. Lenci},
url = {https://aclanthology.org/2024.cogalex-1.12},
isbn = {978-2-493814-45-6},
year = {2024},
date = {2024-01-01},
booktitle = {Workshop on Cognitive Aspects of the Lexicon, CogALex 2024 at LREC-COLING 2024 - Workshop Proceedings},
pages = {107–113},
publisher = {European Language Resources Association (ELRA)},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Proietti, Mattia; Lebani, Gianluca E.; Lenci, Alessandro
On the proto-role properties inferred by transformer language models Journal Article
In: LINGUE E LINGUAGGIO, vol. 1, no. 1, pp. 111–140, 2024.
@article{<LineBreak> 11568_1287831,
title = {On the proto-role properties inferred by transformer language models},
author = {Mattia Proietti and Gianluca E. Lebani and Alessandro Lenci},
url = {https://colinglab.fileli.unipi.it/proietti_lebani_lenci_24_final/},
doi = {10.1418/113930},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
journal = {LINGUE E LINGUAGGIO},
volume = {1},
number = {1},
pages = {111–140},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Rambelli, Giulia; Chersoni, Emmanuele; Testa, Davide; Blache, Philippe; Lenci, Alessandro
Neural Generative Models and the Parallel Architecture of Language: A Critical Review and Outlook Journal Article
In: TOPICS IN COGNITIVE SCIENCE, no. 4, 2024.
@article{ 11568_1287828,
title = {Neural Generative Models and the Parallel Architecture of Language: A Critical Review and Outlook},
author = {Giulia Rambelli and Emmanuele Chersoni and Davide Testa and Philippe Blache and Alessandro Lenci},
doi = {10.1111/tops.12733},
year = {2024},
date = {2024-01-01},
journal = {TOPICS IN COGNITIVE SCIENCE},
number = {4},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Muscato, Benedetta; Bushipaka, Praveen; Gezici, Gizem; Passaro, Lucia; Giannotti, Fosca
Multi-Perspective Stance Detection Conference
CEUR Workshop Proceedings, vol. 3825, CEUR-WS, 2024.
@conference{ 11568_1280367,
title = {Multi-Perspective Stance Detection},
author = {Benedetta Muscato and Praveen Bushipaka and Gizem Gezici and Lucia Passaro and Fosca Giannotti},
year = {2024},
date = {2024-01-01},
booktitle = {CEUR Workshop Proceedings},
volume = {3825},
pages = {208–214},
publisher = {CEUR-WS},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Kauf, Carina; Chersoni, Emmanuele; Lenci, Alessandro; Fedorenko, Evelina; Ivanova, Anna A.
Proceedings of the 7th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP, 2024, ISBN: 979-8-89176-170-4.
@conference{ 11568_1287830,
title = {Log Probabilities Are a Reliable Estimate of Semantic Plausibility in Base and Instruction-Tuned Language Models},
author = {Carina Kauf and Emmanuele Chersoni and Alessandro Lenci and Evelina Fedorenko and Anna A. Ivanova},
url = {https://aclanthology.org/2024.blackboxnlp-1.18},
doi = {10.18653/v1/2024.blackboxnlp-1.18},
isbn = {979-8-89176-170-4},
year = {2024},
date = {2024-01-01},
booktitle = {Proceedings of the 7th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP},
pages = {263–277},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Cascione, A.; Cerulli, A.; Manerba, M. M.; Passaro, L. C.
Investigating the Hurtfulness of Misogynistic Tweets Across Professions Conference
Proceedings of the Discovery Science Late Breaking Contributions 2024 (DS-LB 2024) co-located with 27th International Conference Discovery Science 2024 (DS 2024), vol. 3928, 2024.
@conference{ 11568_1327848,
title = {Investigating the Hurtfulness of Misogynistic Tweets Across Professions},
author = {A. Cascione and A. Cerulli and M. M. Manerba and L. C. Passaro},
url = {https://ceur-ws.org/Vol-3928/paper_174.pdf},
year = {2024},
date = {2024-01-01},
booktitle = {Proceedings of the Discovery Science Late Breaking Contributions 2024 (DS-LB 2024) co-located with 27th International Conference Discovery Science 2024 (DS 2024)},
volume = {3928},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Tamponi, Lucia; Bondielli, Alessandro
Exploring Latin epigraphy with Distributional Semantic Models: A pilot study Journal Article
In: STUDI E SAGGI LINGUISTICI, vol. 62, no. 2, pp. 31–56, 2024.
@article{ 11568_1287408,
title = {Exploring Latin epigraphy with Distributional Semantic Models: A pilot study},
author = {Lucia Tamponi and Alessandro Bondielli},
year = {2024},
date = {2024-01-01},
journal = {STUDI E SAGGI LINGUISTICI},
volume = {62},
number = {2},
pages = {31–56},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Cassese, Maria; Bondielli, Alessandro; Lenci, Alessandro
Evaluation of event plausibility recognition in Large (Vision)-Language Models Journal Article
In: IJCOL, vol. 10, no. 2, 2024.
@article{ 11568_1322153,
title = {Evaluation of event plausibility recognition in Large (Vision)-Language Models},
author = {Maria Cassese and Alessandro Bondielli and Alessandro Lenci},
url = {https://journals.openedition.org/ijcol/1422},
doi = {10.17454/IJCOL102.02},
year = {2024},
date = {2024-01-01},
journal = {IJCOL},
volume = {10},
number = {2},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Ferri, P.; Lomonaco, V.; Passaro, L. C.; Castro, A. Félix-De; Sánchez-Cuesta, P.; Sáez, C.; García-Gómez, J. M.
Deep continual learning for medical call incidents text classification under the presence of dataset shifts Journal Article
In: Computers in Biology and Medicine, vol. 175, 2024.
@article{Ferri2024,
title = {Deep continual learning for medical call incidents text classification under the presence of dataset shifts},
author = {P. Ferri and V. Lomonaco and L. C. Passaro and A. Félix-De Castro and P. Sánchez-Cuesta and C. Sáez and J. M. García-Gómez},
url = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-85192148870&doi=10.1016%2fj.compbiomed.2024.108548&partnerID=40&md5=dfc4ec4fb6df9d4011c89a8643c881ea},
doi = {10.1016/j.compbiomed.2024.108548},
year = {2024},
date = {2024-01-01},
journal = {Computers in Biology and Medicine},
volume = {175},
abstract = {The aim of this work is to develop and evaluate a deep classifier that can effectively prioritize Emergency Medical Call Incidents (EMCI) according to their life-threatening level under the presence of dataset shifts. We utilized a dataset consisting of 1 982 746 independent EMCI instances obtained from the Health Services Department of the Region of Valencia (Spain), with a time span from 2009 to 2019 (excluding 2013). The dataset includes free text dispatcher observations recorded during the call, as well as a binary variable indicating whether the event was life-threatening. To evaluate the presence of dataset shifts, we examined prior probability shifts, covariate shifts, and concept shifts. Subsequently, we designed and implemented four deep Continual Learning (CL) strategies—cumulative learning, continual fine-tuning, experience replay, and synaptic intelligence—alongside three deep CL baselines—joint training, static approach, and single fine-tuning—based on DistilBERT models. Our results demonstrated evidence of prior probability shifts, covariate shifts, and concept shifts in the data. Applying CL techniques had a statistically significant (α=0.05) positive impact on both backward and forward knowledge transfer, as measured by the F1-score, compared to non-continual approaches. We can argue that the utilization of CL techniques in the context of EMCI is effective in adapting deep learning classifiers to changes in data distributions, thereby maintaining the stability of model performance over time. To our knowledge, this study represents the first exploration of a CL approach using real EMCI data. © 2024 The Author(s)},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The aim of this work is to develop and evaluate a deep classifier that can effectively prioritize Emergency Medical Call Incidents (EMCI) according to their life-threatening level under the presence of dataset shifts. We utilized a dataset consisting of 1 982 746 independent EMCI instances obtained from the Health Services Department of the Region of Valencia (Spain), with a time span from 2009 to 2019 (excluding 2013). The dataset includes free text dispatcher observations recorded during the call, as well as a binary variable indicating whether the event was life-threatening. To evaluate the presence of dataset shifts, we examined prior probability shifts, covariate shifts, and concept shifts. Subsequently, we designed and implemented four deep Continual Learning (CL) strategies—cumulative learning, continual fine-tuning, experience replay, and synaptic intelligence—alongside three deep CL baselines—joint training, static approach, and single fine-tuning—based on DistilBERT models. Our results demonstrated evidence of prior probability shifts, covariate shifts, and concept shifts in the data. Applying CL techniques had a statistically significant (α=0.05) positive impact on both backward and forward knowledge transfer, as measured by the F1-score, compared to non-continual approaches. We can argue that the utilization of CL techniques in the context of EMCI is effective in adapting deep learning classifiers to changes in data distributions, thereby maintaining the stability of model performance over time. To our knowledge, this study represents the first exploration of a CL approach using real EMCI data. © 2024 The Author(s)
CAO, Mattia DE; Cao, Nicola De; Colonna, Angelo; Lenci, Alessandro
Deep Learning Meets Egyptology: a Hieroglyphic Transformer for Translating Ancient Egyptian Conference
Proceedings of the 1st Workshop on Machine Learning for Ancient Languages (ML4AL 2024), Association for Computational Linguistics, 2024, ISBN: 979-8-89176-144-5.
@conference{ 11568_1276282,
title = {Deep Learning Meets Egyptology: a Hieroglyphic Transformer for Translating Ancient Egyptian},
author = {Mattia DE CAO and Nicola De Cao and Angelo Colonna and Alessandro Lenci},
url = {https://aclanthology.org/2024.ml4al-1.9},
doi = {10.18653/v1/2024.ml4al-1.9},
isbn = {979-8-89176-144-5},
year = {2024},
date = {2024-01-01},
booktitle = {Proceedings of the 1st Workshop on Machine Learning for Ancient Languages (ML4AL 2024)},
pages = {71–86},
publisher = {Association for Computational Linguistics},
abstract = {This work explores the potential of Transformer models focusing on the translation of ancient Egyptian hieroglyphs. We present a novel Hieroglyphic Transformer model, built upon the powerful M2M-100 multilingual translation framework and trained on a dataset we customised from the Thesaurus Linguae Aegyptiae database. Our experiments demonstrate promising results, with the model achieving significant accuracy in translating hieroglyphics into both German and English. This work holds significant implications for Egyptology, potentially accelerating the translation process and unlocking new research approaches.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
This work explores the potential of Transformer models focusing on the translation of ancient Egyptian hieroglyphs. We present a novel Hieroglyphic Transformer model, built upon the powerful M2M-100 multilingual translation framework and trained on a dataset we customised from the Thesaurus Linguae Aegyptiae database. Our experiments demonstrate promising results, with the model achieving significant accuracy in translating hieroglyphics into both German and English. This work holds significant implications for Egyptology, potentially accelerating the translation process and unlocking new research approaches.
Bondielli, A.; Dell'Oglio, P.; Lenci, A.; Marcelloni, F.; Passaro, L.
Dataset for multimodal fake news detection and verification tasks Journal Article
In: DATA IN BRIEF, vol. 54, 2024.
@article{ 11568_1241070.2,
title = {Dataset for multimodal fake news detection and verification tasks},
author = {A. Bondielli and P. Dell'Oglio and A. Lenci and F. Marcelloni and L. Passaro},
url = {https://www.sciencedirect.com/science/article/pii/S2352340924004098},
doi = {10.1016/j.dib.2024.110440},
year = {2024},
date = {2024-01-01},
journal = {DATA IN BRIEF},
volume = {54},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Cossu, A.; Carta, A.; Passaro, L.; Lomonaco, V.; Tuytelaars, T.; Bacciu, D.
Continual pre-training mitigates forgetting in language and vision Journal Article
In: Neural Networks, vol. 179, 2024.
@article{Cossu2024,
title = {Continual pre-training mitigates forgetting in language and vision},
author = {A. Cossu and A. Carta and L. Passaro and V. Lomonaco and T. Tuytelaars and D. Bacciu},
url = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-85197809738&doi=10.1016%2fj.neunet.2024.106492&partnerID=40&md5=9da883314248cc3696a82b3c6c43bd64},
doi = {10.1016/j.neunet.2024.106492},
year = {2024},
date = {2024-01-01},
journal = {Neural Networks},
volume = {179},
abstract = {Pre-trained models are commonly used in Continual Learning to initialize the model before training on the stream of non-stationary data. However, pre-training is rarely applied during Continual Learning. We investigate the characteristics of the Continual Pre-Training scenario, where a model is continually pre-trained on a stream of incoming data and only later fine-tuned to different downstream tasks. We introduce an evaluation protocol for Continual Pre-Training which monitors forgetting against a Forgetting Control dataset not present in the continual stream. We disentangle the impact on forgetting of 3 main factors: the input modality (NLP, Vision), the architecture type (Transformer, ResNet) and the pre-training protocol (supervised, self-supervised). Moreover, we propose a Sample-Efficient Pre-training method (SEP) that speeds up the pre-training phase. We show that the pre-training protocol is the most important factor accounting for forgetting. Surprisingly, we discovered that self-supervised continual pre-training in both NLP and Vision is sufficient to mitigate forgetting without the use of any Continual Learning strategy. Other factors, like model depth, input modality and architecture type are not as crucial. © 2024 The Author(s)},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Pre-trained models are commonly used in Continual Learning to initialize the model before training on the stream of non-stationary data. However, pre-training is rarely applied during Continual Learning. We investigate the characteristics of the Continual Pre-Training scenario, where a model is continually pre-trained on a stream of incoming data and only later fine-tuned to different downstream tasks. We introduce an evaluation protocol for Continual Pre-Training which monitors forgetting against a Forgetting Control dataset not present in the continual stream. We disentangle the impact on forgetting of 3 main factors: the input modality (NLP, Vision), the architecture type (Transformer, ResNet) and the pre-training protocol (supervised, self-supervised). Moreover, we propose a Sample-Efficient Pre-training method (SEP) that speeds up the pre-training phase. We show that the pre-training protocol is the most important factor accounting for forgetting. Surprisingly, we discovered that self-supervised continual pre-training in both NLP and Vision is sufficient to mitigate forgetting without the use of any Continual Learning strategy. Other factors, like model depth, input modality and architecture type are not as crucial. © 2024 The Author(s)