Dirk Hovy
Associate Professor

Biografia
Dirk Hovy è Associate Professor presso l'Università Bocconi, dove si occupa di Natural Language Processing and Computational Social Science.
In precedenza, è stato docente e ricercatore post-dottorato a Copenaghen. È anche direttore scientifico dell'unità di ricerca Data and Marketing Insights (DMI) di BIDSA e responsabile del laboratorio MilaNLP. Ha organizzato una conferenza (EMNLP 2017) e vari workshop su linguaggio offensivo, etica nell'elaborazione del linguaggio naturale e scienze sociali computazionali. Il suo lavoro si concentra sull'elaborazione del linguaggio naturale e sulle scienze sociali computazionali, con un interesse particolare nel modellare ciò che il linguaggio può dirci sulla società e ciò che i computer possono dirci sul linguaggio. Lavora anche su etica ed equità nell'elaborazione del linguaggio naturale e nell'intelligenza artificiale. Dal punto di vista tecnico, è interessato al trasferimento di apprendimento, all'apprendimento multi-task, all'apprendimento per rinforzo e all'apprendimento avversario. Nel 2020 ha ricevuto una sovvenzione ERC Starting Grant per esplorare l'effetto delle variabili sociodemografiche sui modelli di elaborazione del linguaggio naturale e per integrare i due aspetti.
Ha conseguito un Dottorato in Natural Language Processing presso l'USC e un Master in Sociolinguistics in Germania.
Pubblicazioni recenti
- 2024
IA & società
HOVY, D., "IA & società", Eco, July 2024 - 2024
Beyond the Stats: Realities, Perception, and Social Media Discourse on Poverty
BOSE, P., L. LUPO, M. HABIBI, D. HOVY, C. , R. SCHWARZ, "Beyond the Stats: Realities, Perception, and Social Media Discourse on Poverty", American Economic Association Papers and Proceedings, 2024, vol. 114, pp. 690-694 - 2024
Wisdom of Instruction-Tuned Language Model Crowds. Exploring Model Label Variation
PLAZA DEL ARCO, F. M., D. NOZZA, D. HOVY, "Wisdom of Instruction-Tuned Language Model Crowds. Exploring Model Label Variation" in Proceedings of the 3rd Workshop on Perspectivist Approaches to NLP (NLPerspectives) @ LREC-COLING 2024, May 20-25, 2024, Torino, Italy - 2024
“My Answer is C”: First-Token Probabilities Do Not Match Text Answers in Instruction-Tuned Language Models
WANG, X., B. MA, C. HU, L. WEBER-GENZEL, P. RÖTTGER, F. KREUTER, D. HOVY, B. PLANK, "“My Answer is C”: First-Token Probabilities Do Not Match Text Answers in Instruction-Tuned Language Models" in Findings of the Association for Computational Linguistics ACL 2024, 2024, Bangkok, Thailand - 2024
Twists, Humps, and Pebbles: Multilingual Speech Recognition Models Exhibit Gender Performance Gaps
ATTANASIO, G., B. SAVOLDI, D. FUCCI, D. HOVY, "Twists, Humps, and Pebbles: Multilingual Speech Recognition Models Exhibit Gender Performance Gaps" in Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, 2024, Miami, FL, United States of America - 2024
Divine LLaMAs: Bias, Stereotypes, Stigmatization, and Emotion Representation of Religion in Large Language Models
PLAZA DEL ARCO, F. M., A. C. CURRY, S. PAOLI, A. CERCAS CURRY, D. HOVY, "Divine LLaMAs: Bias, Stereotypes, Stigmatization, and Emotion Representation of Religion in Large Language Models" in Findings of the Association for Computational Linguistics: EMNLP 2024, 2024, Miami, FL, United States of America
Grants & Premi
Bocconi Teaching Award per l'anno accademico 2023-2024 - “Innovation in Teaching”, Università Commerciale Luigi Bocconi, 2025
Teaching Award - PhD School, Università Commerciale Luigi Bocconi, 2022
ERC Starting Grant INTEGRATOR, PI (1.5M EUR), 2020

