Publications by Daniela Occhipinti on persona-based dialogue generation, conversational AI evaluation, and NLP — including papers at ACL 2025, ACL 2024 Findings, and NAACL 2024 Findings.
Start Here
These papers give the clearest entry point into my current research agenda: how dialogue systems represent people, how data quality shapes conversational models, and how interlocutors influence generation.
This paper is the strongest expression of my current research direction: dialogue models should adapt not only to a target persona, but also to the interlocutor and the relationship between speakers.
HED-IT asks how human post-editing affects the quality of dialogue training data, connecting model behavior to the often-hidden question of what counts as good conversational supervision.
PRODIGy introduces a dataset for studying richer speaker representations in dialogue generation and lays the groundwork for much of my later work on persona-aware systems.
No publications match this filter. Try a title, venue, year, or author keyword.
In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Jul 2025
Explores how dialogue generation changes when models know not just who is speaking, but who they are speaking to — and what that relationship means for consistency.
Endowing dialogue agents with persona information has proven to significantly improve the consistency and diversity of their generations. While much focus has been placed on aligning dialogues with provided personas, the adaptation to the interlocutor’s profile remains largely underexplored. In this work, we investigate three key aspects: (1) a model’s ability to align responses with both the provided persona and the interlocutor’s; (2) its robustness when dealing with familiar versus unfamiliar interlocutors and topics, and (3) the impact of additional fine-tuning on specific persona-based dialogues. We evaluate dialogues generated with diverse speaker pairings and topics, framing the evaluation as an author identification task and employing both LLM-as-a-judge and human evaluations. By systematically masking or disclosing information about interlocutor, we assess its impact on dialogue generation. Results show that access to the interlocutor’s persona improves the recognition of the target speaker, while masking it does the opposite. Although models generalise well across topics, they struggle with unfamiliar interlocutors. Finally, we found that in zero-shot settings, LLMs often copy biographical details, facilitating identification but trivialising the task.
@inproceedings{occhipinti-etal-2025-superman,title={When Harry Meets Superman: The Role of The Interlocutor in Persona-Based Dialogue Generation},author={Occhipinti, Daniela and Guerini, Marco and Nissim, Malvina},editor={Che, Wanxiang and Nabende, Joyce and Shutova, Ekaterina and Pilehvar, Mohammad Taher},booktitle={Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},month=jul,year={2025},address={Vienna, Austria},publisher={Association for Computational Linguistics},url={https://aclanthology.org/2025.acl-long.879/},doi={10.18653/v1/2025.acl-long.879},pages={17964--17985},isbn={979-8-89176-251-0},note={Explores how dialogue generation changes when models know not just who is speaking, but who they are speaking to — and what that relationship means for consistency.},}
Automatic methods for generating and gathering linguistic data have proven effective for fine-tuning Language Models (LMs) in languages less resourced than English. Still, while there has been emphasis on data quantity, less attention has been given to its quality. In this work, we investigate the impact of human intervention on machine-generated data when fine-tuning dialogical models. In particular, we study (1) whether post-edited dialogues exhibit higher perceived quality compared to the originals that were automatically generated; (2) whether fine-tuning with post-edited dialogues results in noticeable differences in the generated outputs; and (3) whether post-edited dialogues influence the outcomes when considering the parameter size of the LMs. To this end we created HED-IT, a large-scale dataset where machine-generated dialogues are paired with the version post-edited by humans. Using both the edited and unedited portions of HED-IT, we fine-tuned three different sizes of an LM. Results from both human and automatic evaluation show that the different quality of training data is clearly perceived and it has an impact also on the models trained on such data. Additionally, our findings indicate that larger models are less sensitive to data quality, whereas this has a crucial impact on smaller models. These results enhance our comprehension of the impact of human intervention on training data in the development of high-quality LMs.
@inproceedings{occhipinti-etal-2024-fine,title={Fine-tuning with {HED}-{IT}: The impact of human post-editing for dialogical language models},author={Occhipinti, Daniela and Marchi, Michele and Mondella, Irene and Lai, Huiyuan and Dell{'}Orletta, Felice and Nissim, Malvina and Guerini, Marco},editor={Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek},booktitle={Findings of the Association for Computational Linguistics: ACL 2024},month=aug,year={2024},address={Bangkok, Thailand},publisher={Association for Computational Linguistics},url={https://aclanthology.org/2024.findings-acl.707},doi={10.18653/v1/2024.findings-acl.707},pages={11892--11907},note={Investigates how human post-editing of machine-generated dialogues affects model training, finding that data quality matters most for smaller models.}}
In Findings of the Association for Computational Linguistics: NAACL 2024, Jun 2024
Introduces PRODIGy, a dataset pairing movie-script dialogues with rich speaker profiles — personality, biography, and communication style — to support persona-aware dialogue generation.
Providing dialogue agents with a profile representation can improve their consistency and coherence, leading to better conversations. However, current profile-based dialogue datasets for training such agents contain either explicit profile representations that are simple and dialogue-specific, or implicit representations that are difficult to collect. In this work, we introduce the PRODIGy (PROfile-based DIalogue Generation) dataset, which brings diverse representations together, providing a more comprehensive profile dimension set for each speaker. This resource comprises more than 20k dialogues, sourced from movie scripts, aligned with speaker representations such as communication style, biography, personality and gender. Initial experiments with diverse baselines show that providing generative language models with these aspects of a profile, both separately and jointly, enhances models’ performance. This improvement holds true in both in-domain and cross-domain settings, for both fine-tuned and instruction-based LLMs.
@inproceedings{occhipinti:etal-2024-prodigy,title={{PRODIG}y: a {PRO}file-based {DI}alogue Generation dataset},author={Occhipinti, Daniela and Tekiro{\u{g}}lu, Serra Sinem and Guerini, Marco},editor={Duh, Kevin and Gomez, Helena and Bethard, Steven},booktitle={Findings of the Association for Computational Linguistics: NAACL 2024},month=jun,year={2024},address={Mexico City, Mexico},publisher={Association for Computational Linguistics},url={https://aclanthology.org/2024.findings-naacl.222},doi={10.18653/v1/2024.findings-naacl.222},pages={3500--3514},note={Introduces PRODIGy, a dataset pairing movie-script dialogues with rich speaker profiles — personality, biography, and communication style — to support persona-aware dialogue generation.}}
In this paper we describe the systems we used to participate in the TAG-it task of EVALITA 2020. The first system uses a linear Support Vector Machine as the learning algorithm, while the other two are based on the pretrained Italian language model UmBERTo, following multi-task and single-task learning settings. The systems ranked first in all TAG-it subtasks on the official test sets, supporting the effectiveness of the proposed approaches.
@inproceedings{occhipinti:etal:2020italianlp,title={ItaliaNLP @ TAG-IT: UmBERTo for Author Profiling at TAG-it 2020},author={Occhipinti, Daniela and Tesei, Andrea and Iacono, Maria and Aliprandi, Carlo and De Mattei, Lorenzo},booktitle={International Workshop on Evaluation of Natural Language and Speech Tools for Italian},pages={263},year={2020},url={https://ceur-ws.org/Vol-2765/paper143.pdf},note={Author profiling in Italian — an early study on inferring speaker attributes from text, and the system that won every subtask.}}
