Faeze Ghorbanpour

I am a PhD candidate in Computer Science at the Technical University of Munich, advised by Prof. Alexander Fraser and affiliated with the Munich Center for Machine Learning (MCML). My work focuses on the Efficient and Generalizable Transfer Learning, post-training of LLMs, multilingual NLP, and applied NLP. I started my PhD at LMU Munich in 2023 and moved to TUM with my group in 2026.

During my PhD my research has developed along three threads:

Alignment and post-training: adapting language models for tasks where human judgments genuinely disagree, where there is no single ground truth to fine-tune against.
Multilingual NLP: studying how models represent meaning across languages and how to steer their internal representations toward more consistent behavior.
Evaluation and safety: measuring how dependable model outputs are and how closely model judgments track human ones.

I am currently a guest researcher at the University of Edinburgh, working with Prof. Mirella Lapata. During my PhD I was also a guest researcher at the University of Copenhagen, working with Prof. Anders Søgaard, and earlier at the Max Planck Institute for Demographic Research and Aalto University. Before my doctoral studies I spent several years as an NLP Scientist, building and deploying production NLP systems.

news

Oct 17, 2026	📄 Two new preprints out! “Decomposing and Predicting Annotation Disagreement in Harmful Content Across Languages” and “Retrieval Disagreement as a Signal: Diagnosing and Steering Multilingual Dense Embedders.” ✨
Oct 15, 2026	🚀 Starting a research visit at the University of Edinburgh 🏴, collaborating with Prof. Mirella Lapata. Looking forward to the work ahead! 🤝
Sep 29, 2026	🎉 Our paper “On the Sensitivity of Instruction-tuned LLMs to Harmful Sentences in Long Inputs” will be presented at the Trustworthy AI for Good (AI4Good) Workshop @ NeurIPS 2026 in Paris! 🥳
Aug 25, 2026	🎉 Hat-trick! Three papers accepted to EMNLP 2026, one in the Main, two in Findings. 🥳
Mar 01, 2026	🚀 Excited to be visiting Prof. Anders Søgaard’s NLP group at the University of Copenhagen 🇩🇰. Looking forward to the collaboration and the project we will be working on together. 🤝

selected publications

EMNLP

Latent-Space Intervention for Cross-Lingual Factual Consistency: Consistency Improvements without Accuracy Drops

Faeze Ghorbanpour, Constanza Fierro, Alexander Fraser, and 1 more author

In EMNLP (Findings), 2026

To appear

arXiv Bib

@inproceedings{ghorbanpour2026latentspaceinterventioncrosslingualfactual,
  title = {Latent-Space Intervention for Cross-Lingual Factual Consistency: Consistency Improvements without Accuracy Drops},
  author = {Ghorbanpour, Faeze and Fierro, Constanza and Fraser, Alexander and S{\o}gaard, Anders},
  booktitle = {EMNLP (Findings)},
  year = {2026},
  publisher = {Association for Computational Linguistics},
  note = {To appear},
}

EMNLP

On the Sensitivity of Instruction-tuned LLMs to Harmful Sentences in Long Inputs

Faeze Ghorbanpour, and Alexander Fraser

In EMNLP (Main), 2026

To appear

arXiv Bib

@inproceedings{ghorbanpour2026sensitivityinstructiontunedllmsharmful,
  title = {On the Sensitivity of Instruction-tuned {LLM}s to Harmful Sentences in Long Inputs},
  author = {Ghorbanpour, Faeze and Fraser, Alexander},
  booktitle = {EMNLP (Main)},
  year = {2026},
  publisher = {Association for Computational Linguistics},
  note = {To appear},
}

EMNLP

Attribute-Based Diagnosis of LLM Alignment with Hate Speech Annotations

Mohammad Amine Jradi^*, Faeze Ghorbanpour^*, and Alexander Fraser

In EMNLP (Findings), 2026

To appear. ^* Equal contribution.

arXiv Bib

@inproceedings{jradi2026attributebaseddiagnosisllmalignment,
  title = {Attribute-Based Diagnosis of {LLM} Alignment with Hate Speech Annotations},
  author = {Jradi, Mohammad Amine and Ghorbanpour, Faeze and Fraser, Alexander},
  booktitle = {EMNLP (Findings)},
  year = {2026},
  publisher = {Association for Computational Linguistics},
  note = {To appear. <sup>*</sup> Equal contribution.},
}