Our pick of the week by Dennis Fucci: 'Homophone Disambiguation Reveals Patterns of Context Mixing in Speech Transformers' by Hosein Mohebbi, Grzegorz Chrupała 🇪🇺🇺🇦, afra alishahi, et al., EMNLP 2023.

#NLP roc #NLP #EMNLP2023 #speech #transformer #contextmixing #homopohone #disambiguation

The moment for “Transformer-specific Interpretability” tutorial is very approaching!
Check out our detailed program & an extensive suggested reading list here: projects.illc.uva.nl/indeep/tutoria…

🗓️ Thursday, 21 March
⏱️ 14:00-17:30 pm
📍 Radisson room
#EACL2024 #NLProc #XAI

If you're at EACL and still unsure how to spend your Thursday afternoon, consider attending the Transformer-specific Interpretability workshop at 14:00!
Recommended for people with and without experience in interpretability, it will start with a gentle intro in interpretability!

نوروز مبارک.

Sharing some great news today. My paper titled 'Encoding of lexical tone in self-supervised models of spoken language' has been accepted to the main NAACL 2024 conference!
Many thanks to my co-authors Michaela Watkins afra alishahi Arianna Bisazza Grzegorz Chrupała 🇪🇺🇺🇦

A brief 🧵below:

Value Zeroing, a faithful approach for analyzing context mixing in Transformers, is now available on Inseq main branch for all Hugging Face text generation models! 🔀

🔍Paper introducing VZ: aclanthology.org/2023.eacl-main…
🐛VZ in Inseq: tinyurl.com/inseq-vz

Hosein will be on the road; don't miss it if you around these places.

Hi NS online I'm afraid what you put on my discount card does not look like my real surname.
Your Name Is Invalid!

I have 2 openings for my VIDI project on modeling low-res morphologically rich languages, taking inspiration from child language acquisition insights!
GroNLP University of Groningen
📢 a Postdoc (apply by 15 Feb)
rug.nl/about-ug/work-…
📢 a PhD (apply by 11 Mar)
rug.nl/about-ug/work-…

Our upcoming #EACL2024 tutorial “Transformer-specific Interpretability” will focus on the trending type of interpretability that makes use of specific features of transformers for understanding LLMs, and discuss their pros & cons!
More info: projects.illc.uva.nl/indeep/tutoria…
#NLProc #XAI

Happy to announce that PECoRe, our framework to verify the trustworthiness of LM generations from model internals, was accepted to #ICLR2024 ! 🎉 Details here ⬇️

This week I taught a course on neural models of spoken language at the Netherlands National Graduate School of Linguistics, also know as the LOT Winter School 2024. My lecture slides & notes as well as other course materials are available at github.com/gchrupala/neur…

The temptation to use half a page to display a single scalar can be hard to resist.

Join us at the BlackboxNLP poster session today at 11:00 (in-person at #EMNLP2023 ) for our PECoRe presentation and to grab some Inseq stickers! 🐛 #nlproc

Very proud of this release and all the amazing new contributors working towards democratizing interpretability research tools for the NLP community! 🚀

I’m hiring a postdoc for my VIDI project on improving modeling of low-res morphologically rich languages taking inspiration from child language acquisition insights. Are you the one? Come talk to me at #EMNLP2023 ! (more details here: cs.rug.nl/~bisazza/)