Cihan Ünlü

researcher #t9n , #1nt / AI in translation and interpreting.
PhD student @UniBogazici , research assistant @YeniYuzyilEduTR

[The Interpreter and Translator Trainer (ITT) @ITT_journal]

We invite proposals for special, guest-edited issues of The Interpreter and Translator Trainer (ITT) to be published in 2029.

See ITT's News section: https://t.co/mrSTY3OcQ2

[Cihan Ünlü @cihanunlu_]

@AlicanKiraz0 @VNGRS MT için kaç bin pair'li bir test seti kullandınız ve hangi domain'deydi?

[translation attached @tra_attached]

Proud moment 🥹 Our very first book, Stories of Exile, makes it onto the same syllabus for the second time, and now has a review by translation scholar @cihanunlu_ Links (or scan the image 👀):
https://t.co/mOOvD6tvEi (academia edu)
https://t.co/B6UXhZbXOv (journal)

[EST @estrans]

New journal special issue Linguistica Antverpiensia, Machine and Computer-assisted Interpreting. Edited by: Xinchao Lu & Claudio Fantinuoli. Volume 24 (2025). https://t.co/E2bQdOteWS

[Cihan Ünlü @cihanunlu_]

Ebru Diriker, delivering her keynote speech on interpreting and technology at the 2nd National Congress on Translation and Interpreting Technologies @mcbu1992

[Cihan Ünlü @cihanunlu_]

Sight-Terp is on air!

[Sight-Terp @SightTerp]

Introducing Sight‑Terp Toolbox — a suite of AI‑powered tools for interpreters

What’s inside
#SightTerpAssist: Your digital booth‑mate.
#SightTerpEvaluate: Turn practice into progress
#SightTerpPractice: Practice deliberately

Explore features: https://t.co/FYFetbYQCf

[Cihan Ünlü @cihanunlu_]

With eternal respect and longing, I commemorate Mustafa Kemal Atatürk — the Commander-in-Chief of our War of Independence, the Founder of Republic of Türkiye, whose ideas still light our way today.

[Ilya Sutskever @ilyasut]

if you value intelligence above all other human qualities, you’re gonna have a bad time

[Cihan Ünlü @cihanunlu_]

@DrFantinuoli Great. Thanks for the insight.

[Cihan Ünlü @cihanunlu_]

@DrFantinuoli Were you able to detect any quality increase in the output with the context injection through frames?

[Cihan Ünlü @cihanunlu_]

I got a limited access. Soon will compare it with what Apple has for us.

[Slator @slatornews]

.@Microsoft unveils 🚀 Live Interpreter API for real-time #speech #translation across 76 languages and 143 locales with automatic language detection 🌏🔍 and voice 🗣️ preservation.
#AI #xl8 #t9n @Azure
https://t.co/y85gO5iwTv

[Cihan Ünlü @cihanunlu_]

We've been there many times. Not a new phenomenon. But the fact that Apple now pursues and invests in real-time translation in wearable tech shows the new level of maturity of the advancements in speech translation research.

[Morning Brew ☕️ @MorningBrew]

Apple's new Live Translation works flawlessly

[Cihan Ünlü @cihanunlu_]

Slowly to the dead internet theory

[Ahmet @ia_met]

Video Overviews in NotebookLM coming soon! :)

@testingcatalog I think you already talked about this one.

[Cihan Ünlü @cihanunlu_]

@DrFantinuoli 2025 has become the year of open source and proprietary TTS models (apart from agents). Let’s see when the year of S2ST comes.

[Cihan Ünlü @cihanunlu_]

Another hype-loaded machine interpreting reveal. This time it is directly presented for the end user. We have seen the attempt of Meta's seamlessMT with their new "waiting policy" (AI's decalage), having terrible results for non-Eng. I wonder what Google has done for that,if beta

[Exec Sum @exec_sum]

Google just introduced real time translation for Google Meet video calls

[Cihan Ünlü @cihanunlu_]

English-dominant yet.

But text-to-speech is slowly reaching the point of being a ‘solved problem'.

[Vaibhav (VB) Srivastav @reach_vb]

HOLY SHITT, Sesame Labs just dropped CSM (Conversational Speech Model) - Apache 2.0 licensed! 💥

> Trained on 1 MILLION hours of data 🤯
> Contextually aware, emotionally intelligent speech
> Voice cloning & watermarking
> Ultra fast, real-time synthesis
> Based on llama architecture & Mimi like decoder
> Apache 2.0 licensed
> Weights on the Hub

So cool to see such a strong Speech backbone out in the wild! Kudos @sesame team! 🤗

[Cihan Ünlü @cihanunlu_]

@mervenoyann Thanks for the notebook. I would personally recommend extracting different frames based on a similarity threshold and creating 4x4 or 6x6 grids from these frames. Then sending to the model. (To be effective in longer videos).

[Cihan Ünlü @cihanunlu_]

CAT tool providers stand to benefit greatly from this.

[Sophia Yang, Ph.D. @sophiamyang]

Announcing @MistralAI OCR -  the world’s best document understanding API.

🔍 State-of-the-art understanding of complex documents
🌍 Natively multilingual and multimodal
⚡ Fastest in its category
📄 Doc-as-prompt, structured output
🔒 Available for on-prem deployment

[Cihan Ünlü @cihanunlu_]

@Chao1013324 Great work. Congrats.

[Cihan Ünlü @cihanunlu_]

I think the reliance of CS researchers on Mean Opinion Score (MOS) for SI quality eval. shows a fundamental gap. For a truly benchmark of MI against human expertise, surface-level assessments should be avoided. An interdisciplinary evaluation framework is clearly needed.

[Cihan Ünlü @cihanunlu_]

A French AI research company Kyutai released a new speech-to-speech model, outperforming the "Seamless" of Meta. Paper and code available. Could not test it yet. It supports EN-FR only now. Supposedly, real-time inference on GPU and even on a smartphone is possible.

[kyutai @kyutai_labs]

Meet Hibiki, our simultaneous speech-to-speech translation model, currently supporting 🇫🇷➡️🇬🇧.
Hibiki produces spoken and text translations of the input speech in real-time, while preserving the speaker’s voice and optimally adapting its pace based on the semantic content of the source speech.
Based on objective and human evaluations, Hibiki outperforms previous systems for quality, naturalness and speaker similarity and approaches human interpreters. 🧵

[Cihan Ünlü @cihanunlu_]

Interestingly, in the evaluation section, authors bring in the human interpretation data for comparison, too. The demo page compares some outputs.

https://t.co/Yd5Rlykvo9