Debjit Paul

Hi there! I’m a Research Scientist at Noah’s Ark Lab, London.

Previously, I was a Postdoctoral Researcher at EPFL, working with Antoine Bosselut, Boi Faltings, and Robert West. Before EPFL, I completed my Ph.D. under the supervision of Anette Frank in the Department of Computational Linguistics at Heidelberg University, where I was also part of the Research Training Group AIPHES.

My overarching research objective is to enhance the reasoning capabilities of artificial agents by advancing their ability to understand the complex interplay between language, knowledge representation, and cognitive reasoning processes. To accomplish this goal, I have focused my investigations on three areas:

Neuro-symbolic Reasoning - AI systems must integrate structured, symbolic knowledge with neural representations to capture knowledge dynamics and generalise robustly to unseen situations. My research has focused on developing computational models that perform reasoning by combining the expressiveness of symbolic structures with the adaptability of neural architectures.
Learning to generate explanations for reasoning - When a model explains its decision-making steps or logical steps it took to reach that conclusion, it enhances transparency and helps build trust in the system. My research has focused on designing computational models that generate explanations before making the final decisions.
Aligning Models with Human or AI Feedback - Text generation models are observed to display undesired and inconsistent behaviours, such as hallucination and unfaithful reasoning. My research has focused on designing methods to rectify the undesired behaviours of text generation models through interaction.

news

May 04, 2025	Our paper FIPO: Fallacy-Informed Preference Optimisation received the Outstanding Paper Award 🏆 at NAACL 2025.
Jan 23, 2025	One paper is accepted at ICLR 2025 (INCLUDE)
Jan 23, 2025	One paper is accepted at NAACL 2025 (FIPO: Fallacy-Informed Preference Optimization)
Nov 07, 2024	Talk at Microsoft Research (MSR), Bangalore, India.
Sep 22, 2024	Two papers are accepted at EMNLP 2024

latest posts

Sep 30, 2025	What Children Know That AI Doesn't: Learning Through Experience, Not Text
Sep 28, 2025	Understanding FLOPs, MFU, and Computational Efficiency in LLM Training: From Dense Transformers to MoE Architectures
Jan 15, 2025	FIPO: Fallacy-Informed Preference Optimization - Steering LLMs Toward Logically Sound Arguments

selected publications

A Logical Fallacy-Informed Framework for Argument Generation

Luca Mouchel, Debjit Paul, Shaobo Cui, and 3 more authors

In 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), Apr 2025

DOI Bib PDF Code

@inproceedings{mouchel-etal-2025-logical,
  title = {A Logical Fallacy-Informed Framework for Argument Generation},
  author = {Mouchel, Luca and Paul, Debjit and Cui, Shaobo and West, Robert and Bosselut, Antoine and Faltings, Boi},
  editor = {Chiruzzo, Luis and Ritter, Alan and Wang, Lu},
  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)},
  month = apr,
  year = {2025},
  address = {Albuquerque, New Mexico},
  publisher = {Association for Computational Linguistics},
  url = {https://aclanthology.org/2025.naacl-long.374/},
  doi = {10.18653/v1/2025.naacl-long.374},
  pages = {7296--7314},
  isbn = {979-8-89176-189-6},
}

Making Reasoning Matter: Measuring and Improving Faithfulness of Chain-of-Thought Reasoning

Debjit Paul, Robert West, Antoine Bosselut, and 1 more author

In Findings of EMNLP: 2024, Nov 2024

Bib PDF Code

@inproceedings{paul2024reasoning,
  title = {Making Reasoning Matter: Measuring and Improving Faithfulness of Chain-of-Thought Reasoning},
  author = {Paul, Debjit and West, Robert and Bosselut, Antoine and Faltings, Boi},
  month = nov,
  year = {2024},
  publisher = {Association for Computational Linguistics},
  booktitle = {Findings of EMNLP: 2024},
  url = {https://debjitpaul.github.io/reasoningmatter},
  journal = {arXiv preprint arXiv:2402.13950},
}

REFINER: Reasoning Feedback on Intermediate Representations

Debjit Paul, Mete Ismayilzada, Maxime Peyrard, and 4 more authors

In Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers), Mar 2024

Bib PDF Code

@inproceedings{Paul2023REFINERRF,
  title = {REFINER: Reasoning Feedback on Intermediate Representations},
  author = {Paul, Debjit and Ismayilzada, Mete and Peyrard, Maxime and Borges, Beatriz and Bosselut, Antoine and West, Robert and Faltings, Boi},
  booktitle = {Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)},
  month = mar,
  year = {2024},
  url = {https://aclanthology.org/2024.eacl-long.67},
}

CRoW: Benchmarking Commonsense Reasoning in Real-World Tasks

Mete Ismayilzada, Debjit Paul, Syrielle Montariol, and 2 more authors

In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP), Dec 2023

DOI Bib PDF

@inproceedings{ismayilzada2023crow,
  title = {CRoW: Benchmarking Commonsense Reasoning in Real-World Tasks},
  author = {Ismayilzada, Mete and Paul, Debjit and Montariol, Syrielle and Geva, Mor and Bosselut, Antoine},
  booktitle = {Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP)},
  month = dec,
  year = {2023},
  url = {https://aclanthology.org/2023.emnlp-main.607},
  doi = {10.18653/v1/2023.emnlp-main.607},
  pages = {9785--9821},
}