Nils Feldhus - NLP researcher


Welcome 👋

Welcome to my homepage! I am Nils Feldhus and am currently working on my PhD thesis in explainable natural language processing at the German Research Center for Artificial Intelligence and the Technische UniversitÀt Berlin under the supervision of Sebastian Möller.

Research interests 👀

My main research interest is making (neural) language models more interpretable by building applications that democratize access to explanations. Topics of interest are rationale generation, data-centric interpretability, information-seeking dialogue, and evaluation measures for generated text.

News đŸ€©

2024-03-23 : Excited to give a talk about Explanation Dialogues and the Role of Didactics in Explainability at the inaugural BIFOLD Tutorial Day on April 30.
2024-02-28 : Two new papers in submission: A follow-up to InterroLang on the conversational examination of self-explaining LLMs (LLMCheckup) and a resource and evaluation paper on instructional explanations in teacher-student dialogues (ReWIRED).
2023-11-13 : InterroLang will be presented as an in-person poster at BlackboxNLP (Thu, Dec 7, 11:00 AM) and Findings (Sat, Dec 9, 09:00 AM).
2023-11-03 : Invited talk at Human-Centric AI group of NEC Labs Europe, Heidelberg
2023-10-08 : “InterroLang: Exploring NLP Models and Datasets through Dialogue-based Explanations” accepted to EMNLP 2023 Findings! 🩁 This is my fourth first-author paper and an implementation of the Mediators precursor paper.
2023-05-30 : Saliency Map Verbalization (first-author paper) accepted to ACL 2023! See you in Toronto! 🍁
2023-05-19 : Inseq accepted to ACL 2023! MultiTACRED which I reviewed for my colleagues has been accepted as well.
2023-02-27 : Inseq pre-print published on arXiv.
2022-12-26 : Three papers accepted to ESSV and HUCAPP. New interpretability library Inseq (project led by Gabriele Sarti) now available on GitHub.
2022-11-21 : Journal paper about “Interactive Explainable AI” accepted to KI.
2022-10-14 : First-author paper on “Saliency Map Verbalization” is now on arXiv.
2022-10-13 : One paper I reviewed for my colleagues was accepted to the EMNLP 2022 main track.
2022-07-04 : Companion paper to “Personalized Conversational Agents” accepted to SIGDIAL 2022.
2022-06-15 : Paper on “Personalized Conversational Agents” accepted to INTERSPEECH 2022.
2022-06-14 : “Mediators” is now available on arXiv.
2022-06-04 : First-author paper “Mediators: Conversational Agents Explaining Language Model Behavior” accepted to the IJCAI-ECAI 2022 Workshop on XAI.
2022-04-04 : Paper on “Textual Explanations for Clinical Decision Support” accepted to LREC as a poster.
2022-03-31 : Two papers I reviewed for my colleagues were accepted to the NLP-Power! and Repl4NLP workshops at ACL 2022
2022-03-22 : Project report on “XAI and meaningful information in automated decision-making” now available in full.

Publications 📚

2024

ReWIRED: Instructional Explanations in Teacher-Student Dialogues

Nils Feldhus, Aliki Anagnostopoulou, João Lucas Mendes de Lemos Lins, Qianli Wang, Milad Alshomary, Henning Wachsmuth, Daniel Sonntag, and Sebastian Möller
In submission
OpenReview

LLMCheckup: Conversational Examination of Large Language Models via Interpretability Tools

Qianli Wang, Tatiana Anikina, Nils Feldhus, Josef van Genabith, Leonhard Hennig, and Sebastian Möller
In submission
arXiv | GitHub

2023

InterroLang: Exploring NLP Models and Datasets through Dialogue-based Explanations

Nils Feldhus, Qianli Wang, Tatiana Anikina, Sahil Chopra, Cennet Oguz, and Sebastian Möller
EMNLP 2023 Findings & BlackboxNLP Workshop
ACL Anthology | arXiv | GitHub

Saliency Map Verbalization: Comparing Feature Importance Representations from Model-free and Instruction-based Methods

Nils Feldhus, Leonhard Hennig, Maximilian Dustin Nasert, Christopher Ebert, Robert Schwarzenberg, and Sebastian Möller
ACL 2023 Workshop on Natural Language Reasoning and Structured Explanations (NLRSE)
ACL Anthology | arXiv | GitHub

Inseq: An Interpretability Toolkit for Sequence Generation Models

Gabriele Sarti, Nils Feldhus, Ludwig Sickert, Oskar van der Wal, Malvina Nissim, and Arianna Bisazza
ACL 2023 System Demonstrations
ACL Anthology | arXiv | GitHub | Project page

Pre-trained Language Models for the Automatic Evaluation of Customer Chatbot Dialogs

Mika Rebensburg, Stefan Hillmann, and Nils Feldhus
ESSV 2023
Proceedings

Adapters for Resource-Efficient Deployment of NLU Models

Jan Nehring, Akhyar Ahmed, and Nils Feldhus
ESSV 2023
Proceedings | GitHub

Fighting Disinformation - Overview of Recent AI-based Collaborative Human-Computer Interaction for Intelligent Decision Support Systems

Tim Polzehl, Vera Schmitt, Nils Feldhus, Joachim Meyer, and Sebastian Möller
HUCAPP 2023
SciTePress

2022

XAINES: Explaining AI with Narratives

Mareike Hartmann, Han Du, Nils Feldhus, Ivana Kruijff-KorbayovĂĄ, and Daniel Sonntag
KI - KĂŒnstliche Intelligenz
Journal article on Springer

Mediators: Conversational Agents Explaining NLP Model Behavior

Nils Feldhus, Ajay Madhavan Ravichandran, and Sebastian Möller
IJCAI-ECAI 2022 Workshop on XAI
arXiv | Slides

Towards Personality-aware Chatbots

Daniel Fernau, Stefan Hillmann, Nils Feldhus, Tim Polzehl, and Sebastian Möller
SIGDIAL 2022
ACL Anthology | Video (Live presentation)

Towards Automated Dialog Personalization using MBTI Personality Indicators

Daniel Fernau, Stefan Hillmann, Nils Feldhus, and Tim Polzehl
INTERSPEECH 2022
ISCA Proceedings

A Comparison of Feature Extraction Models for Medical Image Captioning

Sebastian Germer, Hristina Uzunova, Jan Ehrhardt, Nils Feldhus, Philippe Thomas, and Heinz Handels
GMDS-TMF 2022
PDF

An Annotated Corpus of Textual Explanations for Clinical Decision Support

Roland Roller, Aljoscha Burchardt, Nils Feldhus, Laura Seiffe, Klemens Budde, Simon Ronicke, and Bilgin Osmanodja
LREC 2022
ACL Anthology

What to explain when explaining is difficult? An interdisciplinary primer on XAI and meaningful information in automated decision-making

Hadi Asghari, Nadine Birner, Aljoscha Burchardt, Daniela Dicks, Judith Fassbinder, Nils Feldhus, Freya Hewett, Vincent Hofmann, Matthias C. Kettemann, Wolfgang Schulz, Judith Simon, Jakob Stolberg-Larsen, and Theresa ZĂŒger
Project report (published 2022-03-22)
Full report

2021

Thermostat: A Large Collection of NLP Model Explanations and Analysis Tools

Nils Feldhus, Robert Schwarzenberg, and Sebastian Möller
2021 Conference on Empirical Methods in Natural Language Processing (EMNLP): System Demonstrations
ACL Anthology | arXiv | GitHub | Video

Efficient Explanations from Empirical Explainers

Robert Schwarzenberg, Nils Feldhus, and Sebastian Möller
4th workshop on analyzing and interpreting neural networks for NLP (collocated with EMNLP 2021)
BlackboxNLP 2021 proceedings | arXiv | GitHub

Combining Open Domain Question Answering with a Task-Oriented Dialog System

Jan Nehring, Nils Feldhus, Harleen Kaur, and Akhyar Ahmed
1st Workshop on Document-grounded Dialogue and Conversational Question Answering (DialDoc 2021)
ACL Anthology

European Language Grid: A Joint Platform for the European Language Technology Community

Georg Rehm et al.
16th Conference of the European Chapter of the Association for Computational Linguistics (EACL): System Demonstrations
EACL 2021 Proceedings

2020

Evaluating German Transformer Language Models with Syntactic Agreement Tests

Karolina Zaczynska, Nils Feldhus*, Robert Schwarzenberg, Aleksandra Gabryszak, and Sebastian Möller
5th Swiss Text Analytics Conference (SwissText) & 16th Conference on Natural Language Processing (KONVENS)
SwissText/KONVENS 2020 Proceedings | arXiv | GitHub
* joint first authorship

Towards an Interoperable Ecosystem of AI and LT Platforms: A Roadmap for the Implementation of Different Levels of Interoperability

Georg Rehm et al.
1st International Workshop on Language Technology Platforms (c/w LREC 2020)
IWLTP 2020 Proceedings

Education 👹‍🎓

2021-2025 – Computer Science, PhD, Technische UniversitĂ€t Berlin. Supervised by Sebastian Möller: “Approaches for Generating and Evaluating Natural Language Explanations of Language Models”

2016-2020 – Cognitive Systems: Language, Learning and Reasoning, MSc, University of Potsdam. Supervised by Manfred Stede: “Utilizing machine translation for bootstrapping abstractive text summarization”.

2013-2016 – Computational Linguistics, BA, Heidelberg University. Supervised by Katja Markert: “An in-depth investigation on timeline summarization evaluation”.

Jobs đŸ‘šâ€đŸ’Œ

2021-ongoing – Researcher/Software Engineer @ German Research Institute for Artificial Intelligence (DFKI), Berlin, Speech and Language Technology group.
XAINES - Explaining AI with Narratives project, supervised by Sebastian Möller.

2020-2021 – Researcher/Software Engineer @ German Research Institute for Artificial Intelligence (DFKI), Berlin, Speech and Language Technology group.
European Language Grid project, supervised by Georg Rehm.

2014 – Student assistant at the Institute for Computational Linguistics @ Heidelberg University

Supervision đŸ‘šâ€â€đŸ«

Supervised theses, open topics for prospective students and taught courses

Invited Talks

2024-04-30 : BIFOLD Tutorial Day on Foundation Models at Max DelbrĂŒck Center, Berlin – Explanation Dialogues for Understanding Foundation Model Behavior and Teaching Concepts

2023-11-03 : NEC Labs Europe, Heidelberg – Generating and Evaluating Human-Centric Explanations of Language Model Behavior

Reviews ⭐

ACL 2024 Area Chair for Interpretability and Analysis Models for NLP track
NAACL 2024
EACL 2024
EMNLP 2023 (Main conference & BlackboxNLP workshop)
ACL 2023 (Reality Check theme track & Interpretability track)
EACL 2023
EMNLP 2022 (Interpretability, Interactivity and Analysis of Models track; BlackboxNLP 2022 workshop)
ACL Rolling Review (2021 November – ongoing)
ACL 2022 (+ Emergency Reviews)

As secondary reviewer:
NLDB 2022, BlackboxNLP 2021, EMNLP 2021, ACL 2021, NAACL 2021, WebConf 2021, IWLTP 2020, EMNLP 2020, ACL 2020

Recommended

Video tutorials and publications I recommend

Mail 📹

nils (dot) feldhus (at) dfki (dot) de
feldhusnlp (at) gmail (dot) com

Links 🌐

DFKI Profile
GitHub
GitLab
Semantic Scholar
Google Scholar
OpenReview
Twitter
Mastodon (sigmoid.social)

Tools I love to work with 🧰

PyTorch, Hugging Face datasets + transformers and Captum : My “Explainable NLP toolbox”
PyCharm + Atom : Preferred editors for writing code
Obsidian.md (+ dataview), Zotero and Semantic Scholar (API) : Paper management

Leisure activities and other interests đŸŽ”

Hosting two web radio shows where I mix ambient (Neptunian @ FRISKY Radio, since 2014) and atmospheric electronic music (Idolatry @ Proton Radio, since 2015). I’ve been producing mixes for Mixcloud since 2011. All of these are available on helioscope.net.
Cycling and hiking
Nature photography