Jan Niehues

🚀 Call for Participation: @iwslt Offline Speech Translation 2026

Break language barriers with new languages & real-world scenarios + a brand new source-language agnostic speech translation track 🌍

📅 Evaluation: Apr 1–15
👉 iwslt.org/2026/offline

#IWSLT2026 #SpeechAI

🚀 Call for Participation: @iwslt Model Compression 2026

Make large multilingual foundation models small ⚡ without losing power in EN→DE/ZH speech-to-text translation.

📅 Evaluation: Apr 1–15
iwslt.org/2026/compres...

#IWSLT2026 #SpeechAI #Qwen2 #EfficientAI

🚀 Call for Participation: @iwslt Subtitling 2026

Turn speech into ready-to-watch subtitles 🎬 across TV, News & YouTube!

📅 Evaluation: Apr 1–15
iwslt.org/2026/subtitl...

#IWSLT2026 #SpeechAI #MultimodalAI

📢 Fully-funded #PhD position in #NLProc, focusing on #SpeechLLMs & multimodal language processing. Reposts appreciated! 🙏

📅 Apply by March 15!
🔗 ai4lt.iar.kit.edu/english/team...

Quality estimation (automated metrics) are amazing. Truly. We would like to use them everywhere. That gets compute-expensive very quickly. We also don't know when they don't know.

In "Early-Exit and Instant Confidence Translation Quality Estimation" (at EACL26) we fix that.

Last week @ecekt.bsky.social gave a great talk on her recent work "Model Merging to Maintain Language-Only Performance in Developmentally Plausible Multimodal Models"!

Thank you so much for joining us, Ece! 🌟

聖誕老人究竟如何理解來自世界各地的眾多願望清單上的內容呢？

फादर क्रिसमस वास्तव में दुनिया भर की कई इच्छा-सूचियों पर लिखा हुआ कैसे समझ सकते हैं?

How can Santa actually understand what is written on the many wish lists from all over the world?

Find out at: ai4lt.iar.kit.edu

#kitinformatik #ai4lt #llm #machinelearning #adventskalender #advent

We’re collecting crowd-sourced translations in Piedmontese and Neapolitan.
🎯 Goal: see how well LLMs understand these languages.
👉 Participate here (in IT🇮🇹):
- Piedmontese: quest.ms.mff.cuni.cz/crowd-transl...
- Neapolitan: quest.ms.mff.cuni.cz/crowd-transl...
Anyone can join, no need to be fluent!

It was great seeing you all at EMNLP in Suzhou. 🤗🌆

I presented two papers, check them out if you are interested in
🎤 Speech (summarization) or
📊 MT evaluation with multiple candidates.

aclanthology.org/2025.emnlp-m...
aclanthology.org/2025.wmt-1.63/

I am heading to Suzhou 🌆 for EMNLP, presenting two papers with Fabian Retkowski, @zouharvi.bsky.social and Tu Anh Dinh.

Looking forward to see you at our posters about
🎤 Speech Summarisation (Fri. Nov 7, 14:00-15:30) and
📊 Machine Translation Evaluation (Sat. Nov 8, 11:00-12:00).

Excited to head over to #Suzhou to present 5 papers at #EMNLP2025 and affiliated venues!
Topics include quality estimation and evaluation 👩‍🔬, speech🗣️, and multilinguality🌐- see you soon! 🤩
@maikezufle.bsky.social @jan-niehues.bsky.social

Wir bei AI4LT arbeiten täglich mit 🤖 KI und 🎤 Sprache. Aber wie funktioniert Künstliche Intelligenz eigentlich – und wie kann sie Texte und gesprochene Sprache erzeugen?
@maikezufle.bsky.social aus unserer Gruppe erklärt es in einem Vortrag in der Stadtbücherei #Wörth 📚.
English version below👇

Grafik mit zwei Köpfen im Profil gegenüber: Rechts ein angedeutetes Gehirn, im linken Kopf ein Chip mit Leiterbahnen. Dazu der Text: Donnerstag, 30.10.25 18.30 - 20.00 Uhr Stadtbücherei Wörth am Rhein VORTRAG MIT MAIKE ZÜFLE VOM KARLSRUHER INSTITUT FÜR TECHNOLOGIE (KIT) KÜNSTLICHE INTELLIGENZ UND SPRACHE Spannende Einblicke in die Welt der Sprach-KI Eintritt frei Anmeldung erbeten: Tel.: 07271 / 131-225 E-Mail: vhs@woerth.de

Am 30.10.25 lädt die VHS Wörth zu einem spannenden Abend rund um KI und Sprache ein. Maike Züfle, Doktorandin in der Forschungsgruppe Artificial Intelligence for Language Technologies am @kit.edu erklärt, wie große Sprachmodelle Texte verstehen, schreiben und übersetzen u. zeigt praktische Beispiele

🎉 Starting the week with some great news: 3 of our papers have been accepted to #EMNLP Main! 🧵👇

Big congratulations to all co-authors! 📚✨ Looking forward to presenting in #Suzhou!

Our team is growing! Today, we’re excited to welcome our new colleague Yuka Ko, who joins us as a postdoc on the EU Project DVPS.

Looking forward to collaborating and achieving great things together. Welcome aboard! 🚀

#Interspeech2025 Excited to present our work on in-context language learning for endangered languages! 🤩

Come chat with Zhaolin Li today 14:30–16:30 (A09-P1-04)!

🔗 www.isca-archive.org/interspeech_...

Sara Papi, Maike Z\"ufle, Marco Gaido, Beatrice Savoldi, Danni Liu, Ioannis Douros, Luisa Bentivogli, Jan Niehues
MCIF: Multimodal Crosslingual Instruction-Following Benchmark from Scientific Talks
https://arxiv.org/abs/2507.19634

Findings of the IWSLT 2025 Evaluation Campaign Victor Agostinelli, Tanel Alumäe, Antonios Anastasopoulos, Luisa Bentivogli, Ondřej Bojar, Claudia Borg, Fethi Bougares, Roldano Cattoni, Mauro Cettolo, Lizhong Chen, William Chen, Raj Dabre, Yannick ...

🎤 Also involved in:

📍 Thu July 31 | 09:15–10:30 | IWSLT | Findings of the IWSLT 2025 Evaluation Campaign | 🔗 aclanthology.org/2025.iwslt-1...
📍 Thu July 31 | 16:30–17:30 | IWSLT | Demo Panel: Advances in the field of speech translation

NUTSHELL: A Dataset for Abstract Generation from Scientific Talks Maike Züfle, Sara Papi, Beatrice Savoldi, Marco Gaido, Luisa Bentivogli, Jan Niehues. Proceedings of the 22nd International Conference on Spoken Language Translation (IWSLT 2025). 2025.

📍 Fri Aug 1 | 14:00-14:30 | Oral Session D @ IWSLT | NUTSHELL: A Dataset for Abstract Generation from Scientific Talks | 🔗 aclanthology.org/2025.iwslt-1...

KIT’s Low-resource Speech Translation Systems for IWSLT2025: System Enhancement with Synthetic Data and Model Regularization Zhaolin Li, Yining Liu, Danni Liu, Tuan Nam Nguyen, Enes Yavuz Ugan, Tu Anh Dinh, Carlos Mullov, Alexander Waibel, Jan Niehues. Proceedings of the 22nd International Conference on Spoken Language Tran...

📍 Fri Aug 1 | 11:00-12:30 | Poster Session 2 @ IWSLT | KIT’s Low-resource Speech Translation Systems for IWSLT 2025: System Enhancement with Synthetic Data and Model Regularization | 🔗 aclanthology.org/2025.iwslt-1...

Quality-Aware Decoding: Unifying Quality Estimation and Decoding Sai Koneru, Matthias Huck, Miriam Exel, Jan Niehues. Proceedings of the 22nd International Conference on Spoken Language Translation (IWSLT 2025). 2025.

📍 Thu July 31 | 16:00–16:30 | Oral session B @ IWSLT | Quality-Aware Decoding: Unifying Quality Estimation and Decoding | 🔗 aclanthology.org/2025.iwslt-1...

KIT’s Offline Speech Translation and Instruction Following Submission for IWSLT 2025 Sai Koneru, Maike Züfle, Thai Binh Nguyen, Seymanur Akti, Jan Niehues, Alexander Waibel. Proceedings of the 22nd International Conference on Spoken Language Translation (IWSLT 2025). 2025.

📍 Thu July 31 | 14:00-15:30 | Poster Session 1 @ IWSLT | KIT’s Offline Speech Translation and Instruction Following Submission for IWSLT 2025 | 🔗 aclanthology.org/2025.iwslt-1...

Knockout LLM Assessment: Using Large Language Models for Evaluations through Iterative Pairwise Comparisons Large Language Models (LLMs) have shown to be effective evaluators across various domains such as machine translations or the scientific domain. Current LLM-as-a-Judge approaches rely mostly on indivi...

📍 Thu July 29 | 11:30-12:30 and 14:00-15:00 | Poster @ Gem Workshop | Knockout LLM Assessment: Using Large Language Models for Evaluations through Iterative Pairwise Comparisons | 🔗 www.arxiv.org/abs/2506.03785

Can a Large Language Model Keep My Secrets? A Study on LLM-Controlled Agents Niklas Hemken, Sai Koneru, Florian Jacob, Hannes Hartenstein, Jan Niehues. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop)...

📍 Thu July 29 | 10:30–12:00 | Poster @ SRW | Can a Large Language Model Keep My Secrets? A Study on LLM-Controlled Agents | 🔗 aclanthology.org/2025.acl-srw...

Middle-Layer Representation Alignment for Cross-Lingual Transfer in Fine-Tuned LLMs Danni Liu, Jan Niehues. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

📍 Tue July 29 | 10:30-12:00 | Poster Session 7 @ ACL| Middle-Layer Representation Alignment for Cross-Lingual Transfer in Fine-tuned LLms | 🔗 aclanthology.org/2025.acl-lon...

Contrastive Learning for Task-Independent SpeechLLM-Pretraining Maike Züfle, Jan Niehues. Findings of the Association for Computational Linguistics: ACL 2025. 2025.

📍 Mon July 28 | 18:00-19:30 | Poster Session 5 @ ACL Findings | Contrastive Learning for Task-Independent SpeechLLM-Pretraining | 🔗 aclanthology.org/2025.finding...

🚆 On my way to ACL 2025 in Vienna!

AI4LT (KIT) will present work at ACL, SRW, Gem Workshop, and IWSLT. If you're attending and would like to chat about our research, feel free to reach out or drop by one of the sessions below 🧵. Looking forward to see you there

Excited to share 8 papers at #ACL2025 and affiliated venues! 🎉

Our team explored topics including multimodality, cross-lingual transfer, evaluation, decoding, and low-resource languages.

Looking forward to presenting our work and connecting with the community! See you soon! 🤩

🇦🇹 I’ll be in Vienna for #ACL2025NLP!

Interested in training a speech-aware LLM without a lot of parameteres or data? Come to my poster:
🖼️ Mon, 18:00

Also into Speech Summarization? Join my IWSLT talk in collab with @fbk-mt.bsky.social
🎤 Fri, 14:00

Happy to chat - come say hi! 😎

Papers in 🧵

🚀 Today marks the official start of the #HorizonEurope DVPS project (dvps.translated.com). Happy that @kit.edu is part of this collaboration with 20 leading AI organizations!

📍 This week we're in Rome for the kickoff meeting. Looking forward to exciting collaborations!
#DVPS #KIT #MultimodalAI