It's that time of the year! 🎁

The Apple Machine Learning Research (MLR) team in Paris is hiring a few interns, to do cool research for ±6 months 🚀🚀 & work towards publications/OSS.

Check requirements and apply: ➡️ jobs.apple.com/en-us/detail...

More❓→ ✉️ mlr_paris_internships@group.apple.com

Leveraging Audio-Visual Data to Reduce the Multilingual Gap in Self-Supervised Speech Models Self-supervised learning (SSL) has made significant advances in speech representation learning. Models like wav2vec 2.0 and HuBERT have achieved state-of-the-art results in tasks such as speech recogn...

New preprint!


In this work led by María Andrea Cruz Blandón during her internship at Apple, and with the help of Jie Chi and Zakaria Aldeneh, we show that visual grounding helps reduce the multilingual gap present in self-supervised speech models.

arxiv.org/abs/2509.17523

🎙️Slides from our @interspeech.bsky.social tutorial “Speech Technology Meets Early Language Acquisition” (with Emmanuel Dupoux & Okko Räsänen) are now online: zenodo.org/records/1701...

We hope they’re useful for anyone interested in connecting speech tech and early language learning!

In Vienna for ACL 2025! 🇦🇹 Come chat about speech, multilinguality, and cognitive modelling : I’ll be at the Apple booth Tuesday 1:30–3:30.
#ACL2025NLP

⚠️ Correction: The tutorial is on August 17th (not September 21)

We’ll show how bridging speech processing and psycholinguistics benefits both fields, and why speech researchers (you!) should care.

Collab. between ENS/Meta (Emmanuel Dupoux), Tampere University (Okko Räsänen) and Apple (me).

Hope to see many of you on September 21!

(2/2)

Now that @interspeech.bsky.social registration is open, time for some shameless promo!

Sign-up and join our Interspeech tutorial: Speech Technology Meets Early Language Acquisition: How Interdisciplinary Efforts Benefit Both Fields. 🗣️👶

www.interspeech2025.org/tutorials

⬇️ (1/2)

New preprint out! 👇

We adapt the ABX task, commonly used in speech models, to investigate how multilingual text models represent form (language) vs content (meaning).

📄 arxiv.org/pdf/2505.17747

🙌 With Jie Chi, Skyler Seto, @maartjeterhoeve.bsky.social, Masha Fedzechkina & Natalie Schluter

In this work, we investigate whether learning statistical regularities in speech supports word learning, and find that while people can detect these regularities, they rarely remember the words; suggesting a dissociation between pattern learning and memory. 🧠

Check it out!

This work was primarily carried out by
Ansgar Endress, who also gave me my first taste of research, of cognition, and even introduced me to programming. Couldn't have asked for a better introduction!

New paper out in Cognition! 🎉
This one actually includes some of my (very) early work: my first-ever research project as an undergrad psychology student, now a decade ago. Nice to see it finally in print!

👉 authors.elsevier.com/sd/article/S...

👀

NeurIPS participation in Europe We seek to understand if there is interest in being able to attend NeurIPS in Europe, i.e. without travelling to San Diego, US. In the following, assume that it is possible to present accepted papers ...

Would you present your next NeurIPS paper in Europe instead of traveling to San Diego (US) if this was an option? Søren Hauberg (DTU) and I would love to hear the answer through this poll: (1/6)

🥳 A huge thank you to all of our co-authors @hadware.bsky.social, @acristia.bsky.social, @hbredin.bsky.social, Guillaume Wisnewski & Emmanuel Dupoux, and of course to a larger extent to everyone in the CoML team!

🧵7/7

💡 Finally, we aim to illustrate with this research how advanced computational models can simulate broad and realistic language learning, and as such help advance psycholinguistic research.

For a theoretical take, check our other paper ⬇

📄 www.cambridge.org/core/journal...

🧵6/7

📚 We also analyse STELA's representations and find they don't match traditional linguistic categories like phonemes or words.

📈This indicates that early phonetic and word learning can occur without these categories, which may develop later in life.

🧵5/7

🔬After training, we evaluate STELA on a phone discrimination and a word recognition task.

📊 We find the model replicates the *gradual* and *simultaneous* learning patters found in infants, suggesting statistical learning can alone bootstrap early language patterns.

🧵4/7

🤖 Here, we isolate statistical learning by developing STELA, a computational model that learns by predicting the near future of speech utterances.

🗣️ STELA is trained on raw speech only, with gradually increasing quantities, equivalent to those an infant would hear.

🧵3/7

👶 We know that infants learn multiple aspects of language - like distinguishing native sounds and recognising words - *gradually* and *simultaneously*.

Yet, it is still difficult to pinpoint which mechanisms are responsible for which aspect of language development.

🧵2/7

Simulating Early Phonetic and Word Learning Without Linguistic Categories Before they even talk, infants become sensitive to the speech sounds of their native language and recognize the auditory form of an increasing number of words. Traditionally, these early perceptual c...

🎉 Super happy to announce that a significant part of Marvin Lavechin's and my PhD research on modelling early language acquisition has finally been published in Developmental Science!

📄 doi.org/10.1111/desc...
(OS version : onlinelibrary.wiley.com/share/author...)

🧵1/7

Amazing opportunity for researchers with cognitive/psychology/linguistics (or related) backgrounds!

There is an open position to work as an AIML resident with Apple MLR in Copenhagen, with the great @maartjeterhoeve.bsky.social.

Details (topic + deadline) in the 🐦 thread: x.com/maartjeterho...

👀

I've started putting together a starter pack with people working on Speech Technology and Speech Science: go.bsky.app/BQ7mbkA

(Self-)nominations welcome!