2025
Mousumi Akter, Erion Çano, Erik Weber, Dennis Dobler, Ivan Habernal
A Comprehensive Survey on Legal Summarization: Challenges and Future Directions Artikel
In: ACM Comput. Surv., Bd. 58, Nr. 7, S. 1-32, 2025, ISSN: 0360-0300.
@article{10.1145/3776586,
title = {A Comprehensive Survey on Legal Summarization: Challenges and Future Directions},
author = {Mousumi Akter and Erion Çano and Erik Weber and Dennis Dobler and Ivan Habernal},
url = {https://doi.org/10.1145/3776586},
doi = {10.1145/3776586},
issn = {0360-0300},
year = {2025},
date = {2025-12-01},
urldate = {2025-12-01},
journal = {ACM Comput. Surv.},
volume = {58},
number = {7},
pages = {1-32},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
abstract = {This article provides a systematic up-to-date survey of automatic summarization techniques, datasets, models, and evaluation methods in the legal domain. Through specific source selection criteria, we thoroughly review over 120 papers spanning the modern ‘transformer’ era of natural language processing (NLP), thus filling a gap in existing systematic surveys on the matter. We present existing research along several axes and discuss trends, challenges, and opportunities for future research.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
This article provides a systematic up-to-date survey of automatic summarization techniques, datasets, models, and evaluation methods in the legal domain. Through specific source selection criteria, we thoroughly review over 120 papers spanning the modern ‘transformer’ era of natural language processing (NLP), thus filling a gap in existing systematic surveys on the matter. We present existing research along several axes and discuss trends, challenges, and opportunities for future research.
Subhabrata Dutta, Timo Kaufmann, Goran Glavaš, Ivan Habernal, Kristian Kersting, Frauke Kreuter, Mira Mezini, Iryna Gurevych, Eyke Hüllermeier, Hinrich Schütze
Problem Solving Through Human–AI Preference-Based Cooperation Artikel
In: Computational Linguistics, S. 1–36, 2025, ISSN: 0891-2017.
@article{colt-2025-problem,
title = {Problem Solving Through Human–AI Preference-Based Cooperation},
author = {Subhabrata Dutta and Timo Kaufmann and Goran Glavaš and Ivan Habernal and Kristian Kersting and Frauke Kreuter and Mira Mezini and Iryna Gurevych and Eyke Hüllermeier and Hinrich Schütze},
url = {https://doi.org/10.1162/COLI.a.19},
doi = {10.1162/COLI.a.19},
issn = {0891-2017},
year = {2025},
date = {2025-11-09},
urldate = {2025-11-09},
journal = {Computational Linguistics},
pages = {1–36},
abstract = {While there is a widespread belief that artificial general intelligence—or even superhuman AI—is imminent, complex problems in expert domains are far from being solved. We argue that such problems require human–AI cooperation and that the current state of the art in generative AI is unable to play the role of a reliable partner due to a multitude of shortcomings, including difficulty in keeping track of a complex solution artifact (e.g., a software program), limited support for versatile human preference expression, and lack of adapting to human preference in an interactive setting. To address these challenges, we propose HAI-Co2, a novel human–AI co-construction framework. We take first steps towards a formalization of HAI-Co2 and discuss the difficult open research problems that it faces.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
While there is a widespread belief that artificial general intelligence—or even superhuman AI—is imminent, complex problems in expert domains are far from being solved. We argue that such problems require human–AI cooperation and that the current state of the art in generative AI is unable to play the role of a reliable partner due to a multitude of shortcomings, including difficulty in keeping track of a complex solution artifact (e.g., a software program), limited support for versatile human preference expression, and lack of adapting to human preference in an interactive setting. To address these challenges, we propose HAI-Co2, a novel human–AI co-construction framework. We take first steps towards a formalization of HAI-Co2 and discuss the difficult open research problems that it faces.
Ivan Habernal, Peter Schulam, J"org Tiedemann
Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing: System Demonstrations Konferenzberichte
Association for Computational Linguistics, Suzhou, China, 2025, ISBN: 979-8-89176-334-0.
@proceedings{emnlp-2025-demos,
title = {Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing: System Demonstrations},
author = {Ivan Habernal and Peter Schulam and J"org Tiedemann},
editor = {Ivan Habernal and Peter Schulam and J"org Tiedemann},
url = {https://aclanthology.org/2025.emnlp-demos.0/},
doi = {10.18653/v1/2025.emnlp-demos.0},
isbn = {979-8-89176-334-0},
year = {2025},
date = {2025-11-01},
urldate = {2025-01-01},
publisher = {Association for Computational Linguistics},
address = {Suzhou, China},
abstract = {Welcome to the proceedings of the system demonstrations session of the 2025 Conference on Empirical
Methods in Natural Language Processing, held in Suzhou, China on November 4–9, 2025.
The system demonstrations session includes papers describing systems ranging from early research prototypes to mature production-ready software. We put particular emphasis on publicly available open-source
or open-access systems. Out of 211 submissions in total, four were desk-rejected, and another five had
been withdrawn by the authors. The remaining papers were reviewed by a team of over 150 reviewers
(out of 343 originally invited) and the final recommendations were proposed by a team of 31 area chairs.
We finally accepted 77 papers to be included in the proceedings, resulting in a 38% acceptance rate.
We would like to deeply thank all the authors, area chairs, and reviewers.},
keywords = {},
pubstate = {published},
tppubtype = {proceedings}
}
Welcome to the proceedings of the system demonstrations session of the 2025 Conference on Empirical
Methods in Natural Language Processing, held in Suzhou, China on November 4–9, 2025.
The system demonstrations session includes papers describing systems ranging from early research prototypes to mature production-ready software. We put particular emphasis on publicly available open-source
or open-access systems. Out of 211 submissions in total, four were desk-rejected, and another five had
been withdrawn by the authors. The remaining papers were reviewed by a team of over 150 reviewers
(out of 343 originally invited) and the final recommendations were proposed by a team of 31 area chairs.
We finally accepted 77 papers to be included in the proceedings, resulting in a 38% acceptance rate.
We would like to deeply thank all the authors, area chairs, and reviewers.
Methods in Natural Language Processing, held in Suzhou, China on November 4–9, 2025.
The system demonstrations session includes papers describing systems ranging from early research prototypes to mature production-ready software. We put particular emphasis on publicly available open-source
or open-access systems. Out of 211 submissions in total, four were desk-rejected, and another five had
been withdrawn by the authors. The remaining papers were reviewed by a team of over 150 reviewers
(out of 343 originally invited) and the final recommendations were proposed by a team of 31 area chairs.
We finally accepted 77 papers to be included in the proceedings, resulting in a 38% acceptance rate.
We would like to deeply thank all the authors, area chairs, and reviewers.
Lena Held, Ivan Habernal
Contemporary LLMs struggle with extracting formal legal arguments Proceedings Article
In: Aletras, Nikolaos; Chalkidis, Ilias; Barrett, Leslie; Goanță, Cătălina; Preoțiuc-Pietro, Daniel; Spanakis, Gerasimos (Hrsg.): Proceedings of the Natural Legal Language Processing Workshop 2025, S. 292–303, Association for Computational Linguistics, Suzhou, China, 2025, ISBN: 979-8-89176-338-8.
@inproceedings{held-habernal-2025-contemporary,
title = {Contemporary LLMs struggle with extracting formal legal arguments},
author = {Lena Held and Ivan Habernal },
editor = {Nikolaos Aletras and Ilias Chalkidis and Leslie Barrett and Cătălina Goanță and Daniel Preoțiuc-Pietro and Gerasimos Spanakis},
url = {https://aclanthology.org/2025.nllp-1.20/},
isbn = {979-8-89176-338-8},
year = {2025},
date = {2025-11-01},
urldate = {2025-11-01},
booktitle = {Proceedings of the Natural Legal Language Processing Workshop 2025},
pages = {292–303},
publisher = {Association for Computational Linguistics},
address = {Suzhou, China},
abstract = {Legal Argument Mining (LAM) is a complex challenge for humans and language models alike. This paper explores the application of Large Language Models (LLMs) in LAM, focusing on the identification of fine-grained argument types within judgment texts. We compare the performance of Flan-T5 and Llama 3 models against a baseline RoBERTa model to study if the advantages of magnitude-bigger LLMs can be leveraged for this task. Our study investigates the effectiveness of fine-tuning and prompting strategies in enhancing the models’ ability to discern nuanced argument types. Despite employing state-of-the-art techniques, our findings indicate that neither fine-tuning nor prompting could surpass the performance of a domain-pre-trained encoder-only model. This highlights the challenges and limitations in adapting general-purpose large language models to the specialized domain of legal argumentation. The insights gained from this research contribute to the ongoing discourse on optimizing NLP models for complex, domain-specific tasks. Our code and data for reproducibility are available at https://github.com/trusthlt/legal-argument-spans.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Legal Argument Mining (LAM) is a complex challenge for humans and language models alike. This paper explores the application of Large Language Models (LLMs) in LAM, focusing on the identification of fine-grained argument types within judgment texts. We compare the performance of Flan-T5 and Llama 3 models against a baseline RoBERTa model to study if the advantages of magnitude-bigger LLMs can be leveraged for this task. Our study investigates the effectiveness of fine-tuning and prompting strategies in enhancing the models’ ability to discern nuanced argument types. Despite employing state-of-the-art techniques, our findings indicate that neither fine-tuning nor prompting could surpass the performance of a domain-pre-trained encoder-only model. This highlights the challenges and limitations in adapting general-purpose large language models to the specialized domain of legal argumentation. The insights gained from this research contribute to the ongoing discourse on optimizing NLP models for complex, domain-specific tasks. Our code and data for reproducibility are available at https://github.com/trusthlt/legal-argument-spans.
Ivan Habernal, Sepideh Ghanavati, Vijayanta Jain, Timour Igamberdiev, Shomir Wilson (Hrsg.)
Proceedings of the Sixth Workshop on Privacy in Natural Language Processing Konferenzberichte
Association for Computational Linguistics, Albuquerque, New Mexico, 2025, ISBN: 979-8-89176-246-6.
@proceedings{privatenlp-ws-2025-main,
title = {Proceedings of the Sixth Workshop on Privacy in Natural Language Processing},
editor = {Ivan Habernal and Sepideh Ghanavati and Vijayanta Jain and Timour Igamberdiev and Shomir Wilson},
url = {https://aclanthology.org/2025.privatenlp-main.0/},
isbn = {979-8-89176-246-6},
year = {2025},
date = {2025-04-01},
publisher = {Association for Computational Linguistics},
address = {Albuquerque, New Mexico},
keywords = {},
pubstate = {published},
tppubtype = {proceedings}
}
Sebastian Ochs, Ivan Habernal
Private Synthetic Text Generation with Diffusion Models Proceedings Article
In: Chiruzzo, Luis; Ritter, Alan; Wang, Lu (Hrsg.): Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), S. 10612–10626, Association for Computational Linguistics, Albuquerque, New Mexico, 2025, ISBN: 979-8-89176-189-6.
@inproceedings{ochs-habernal-2025-private,
title = {Private Synthetic Text Generation with Diffusion Models},
author = {Sebastian Ochs and Ivan Habernal},
editor = {Luis Chiruzzo and Alan Ritter and Lu Wang},
url = {https://aclanthology.org/2025.naacl-long.532/},
isbn = {979-8-89176-189-6},
year = {2025},
date = {2025-04-01},
urldate = {2025-04-01},
booktitle = {Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)},
pages = {10612–10626},
publisher = {Association for Computational Linguistics},
address = {Albuquerque, New Mexico},
abstract = {How capable are diffusion models of generating synthetics texts? Recent research shows their strengths, with performance reaching that of auto-regressive LLMs. But are they also good in generating synthetic data if the training was under differential privacy? Here the evidence is missing, yet the promises from private image generation look strong. In this paper we address this open question by extensive experiments. At the same time, we critically assess (and reimplement) previous works on synthetic private text generation with LLMs and reveal some unmet assumptions that might have led to violating the differential privacy guarantees. Our results partly contradict previous non-private findings and show that fully open-source LLMs outperform diffusion models in the privacy regime. Our complete source codes, datasets, and experimental setup is publicly available to foster future research.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
How capable are diffusion models of generating synthetics texts? Recent research shows their strengths, with performance reaching that of auto-regressive LLMs. But are they also good in generating synthetic data if the training was under differential privacy? Here the evidence is missing, yet the promises from private image generation look strong. In this paper we address this open question by extensive experiments. At the same time, we critically assess (and reimplement) previous works on synthetic private text generation with LLMs and reveal some unmet assumptions that might have led to violating the differential privacy guarantees. Our results partly contradict previous non-private findings and show that fully open-source LLMs outperform diffusion models in the privacy regime. Our complete source codes, datasets, and experimental setup is publicly available to foster future research.
Mousumi Akter, Erion Çano, Erik Weber, Dennis Dobler, Ivan Habernal
A Comprehensive Survey on Legal Summarization: Challenges and Future Directions Artikel
In: CoRR, Bd. abs/2501.17830, 2025.
@article{DBLP:journals/corr/abs-2501-17830,
title = {A Comprehensive Survey on Legal Summarization: Challenges and Future
Directions},
author = {Mousumi Akter and Erion Çano and Erik Weber and Dennis Dobler and Ivan Habernal},
url = {https://doi.org/10.48550/arXiv.2501.17830},
doi = {10.48550/ARXIV.2501.17830},
year = {2025},
date = {2025-01-01},
journal = {CoRR},
volume = {abs/2501.17830},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Anna Leschanowsky, Zahra Kolagar, Erion Çano, Ivan Habernal, Dara Hallinan, Emanuël A. P. Habets, Birgit Popp
Transparent NLP: Using RAG and LLM Alignment for Privacy Q&A Artikel
In: CoRR, Bd. abs/2502.06652, 2025.
@article{DBLP:journals/corr/abs-2502-06652,
title = {Transparent NLP: Using RAG and LLM Alignment for Privacy Q&A},
author = {Anna Leschanowsky and Zahra Kolagar and Erion Çano and Ivan Habernal and Dara Hallinan and Emanuël A. P. Habets and Birgit Popp},
url = {https://doi.org/10.48550/arXiv.2502.06652},
doi = {10.48550/ARXIV.2502.06652},
year = {2025},
date = {2025-01-01},
journal = {CoRR},
volume = {abs/2502.06652},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Elisabeth Kirsten, Ivan Habernal, Vedant Nanda, Muhammad Bilal Zafar
The Impact of Inference Acceleration on Bias of LLMs Proceedings Article
In: Chiruzzo, Luis; Ritter, Alan; Wang, Lu (Hrsg.): Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), S. 1834–1853, Association for Computational Linguistics, Albuquerque, New Mexico, 2025, ISBN: 979-8-89176-189-6.
@inproceedings{kirsten-etal-2025-impact,
title = {The Impact of Inference Acceleration on Bias of LLMs},
author = {Elisabeth Kirsten and Ivan Habernal and Vedant Nanda and Muhammad Bilal Zafar},
editor = {Luis Chiruzzo and Alan Ritter and Lu Wang},
url = {https://aclanthology.org/2025.naacl-long.91/},
doi = {10.18653/v1/2025.naacl-long.91},
isbn = {979-8-89176-189-6},
year = {2025},
date = {2025-01-01},
booktitle = {Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)},
pages = {1834–1853},
publisher = {Association for Computational Linguistics},
address = {Albuquerque, New Mexico},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
2024
Ivan Habernal, Daniel Faber, Nicola Recchia, Sebastian Bretthauer, Iryna Gurevych, Indra Spiecker Döhmann, Christoph Burchard
Mining legal arguments in court decisions Artikel
In: Artif. Intell. Law, Bd. 32, Nr. 3, S. 1–38, 2024.
@article{DBLP:journals/ail/HabernalFRBGDB24,
title = {Mining legal arguments in court decisions},
author = {Ivan Habernal and Daniel Faber and Nicola Recchia and Sebastian Bretthauer and Iryna Gurevych and Indra Spiecker Döhmann and Christoph Burchard},
url = {https://doi.org/10.1007/s10506-023-09361-y},
doi = {10.1007/S10506-023-09361-Y},
year = {2024},
date = {2024-01-01},
journal = {Artif. Intell. Law},
volume = {32},
number = {3},
pages = {1–38},
keywords = {},
pubstate = {published},
tppubtype = {article}
}