Heading to @emnlpmeeting in Budapest! 🇭🇺✈️
I’m thrilled to present our latest research on speech translation! Here is where and when you can find our work:
👇👇👇👇👇👇👇👇
📌 How to Evaluate Speech Translation with Source-Aware Neural MT Metrics (Computational Linguistics)
• 🗓️ Date & Time: October 26, 2026 | 11:00 – 12:30 CEST
• 📍 Location: Poster Session D, Hall H
📖 Paper: doi.org/10.1162/COLI.a.627
A big thanks to all my incredible co-authors: @luisabentivogli, @mgaido91, @negri_teo, @javigg_gg, @zouharvi, @geiongallego, @jorirsan, @dominikmachc, @PSchmidtova, @MaikeZufle, and others.
If you're attending EMNLP, stop by our posters and chat about speech translation! 💬✨
Sara Papi retweeted
We are still accepting submissions to Last Translation Benchmark (10 accepted in any language for coauthorship 🤩).
Also add your model & beat LTB. We added all the latest models: North Small Translate (@cohere), Mistral Large 4 (@MistralAI), Qwen 3.8 (@Alibaba_Qwen) congrats!
Sara Papi retweeted
Machine translation is not solved and it will take a while for it to be done arxiv.org/abs/2609.04173
Bittersweet milestone: today is my last day at @FBK_research, wrapping up an incredible 6+ year chapter from my PhD to now! 😢🚀
To cap it off, our paper on streaming SpeechLLMs was accepted to @emnlpmeeting Main! 🎉
📄 Paper: arxiv.org/abs/2605.31432
DOA policy enables simultaneous translation directly on off-the-shelf models:
• 🧠 Training-free (zero fine-tuning)
• ♾️ Long-form speech processing
• ⚡ Low latency (<1s)
• 🔌 Works on Phi4-Multimodal, Qwen3-Omni, etc.
DOA is integrated into Simulstream for easy benchmarking: 👉 github.com/hlt-mt/simulstrea…
A huge thank you to @luisabentivogli and @fbk_mt! 🌟
Sara Papi retweeted
There are many things machine translation still can't do. Help us steer the next direction by contributing hard-to-translate inputs (and be on a cool paper).
Curious to explore the capabilities of the new audio-text NVIDIA Audex model supporting speech translation from 7 source languages other than English. It comes in two sizes, a 30B MoE architecture with 3B active parameters, and a 2B-parameter-only model
arxiv.org/pdf/2607.05196
Sara Papi retweeted
🚨 DEADLINE EXTENDED! 🚨
More time to shrink your LLMs! The submission deadline for the Model Compression task at #WMT2026 (#EMNLP2026) is now July 15th. 📅
Test data is out. Show us how small you can go without losing translation quality! 🧠🔧
👉shorturl.at/5zXsp
Sara Papi retweeted
We are at @EAMTee with a large-scale stealth project called "[redacted] Translation [redacted]". Looking for contributors across all languages to challenge MTs. Talk to @PSchmidtova, @sethjsa, @rajaee_sara and @gsarti_.
Sara Papi retweeted
We're excited to share that three papers from our lab have been accepted at #Interspeech2026 !! 🍾
#SpeechTranslation #SpeechAI #NLProc #FBK #Interspeech
🇧🇷 Heading to 𝗜𝗖𝗟𝗥 2026 to present our paper 𝗠𝗖𝗜𝗙 🚀 The first human-annotated crosslingual multimodal instruction-following benchmark.
🗣️ Drop by the poster if you are interested in Speech AI!
Check it out:
iclr.cc/virtual/2026/poster/…
Sara Papi retweeted
🚀 New Shared Task: Model Compression for Machine Translation at #WMT2026 (co-located with #EMNLP2026)!
📅 Test data out on June 18th, submissions by July 2nd!
Can you shrink an LLM and keep translation quality high? 🧠🔧
👉 www2.statmt.org/wmt26/model-… #NLP #ML #LLM #ModelCompression
Sara Papi retweeted
Amirbek Djanibekov, Luisa Bentivogli, Matteo Negri, Sara Papi, "SimulU: Training-free Policy for Long-form Simultaneous Speech-to-Speech Translation," arxiv.org/abs/2603.16924
