Introducing... the Subspace Chronicles (ft. @robvanderg, @barbara_plank, @iatitov)—a riveting tale of linguistic information emerging, shifting and disentangling across LM training! Coming to an #EMNLP2023 near you! 📜⏳
📑https://t.co/GJIXHEr5uR
📽https://t.co/p0gJuGDHpv
🙏🥳 Super grateful and honoured to receive this @VILLUMFONDEN grant which will finance my three year postdoc at @ITUkbh including a two year visit at @MilaNLProc working with @dirk_hovy
Looking forward to starting this project 🙌
Tremendous thanks to @iatitov@EdinburghNLP for being extremely welcoming hosts this spring.
And, special thanks to @robvanderg for helping with the video shenanigans: https://t.co/p0gJuGDHpv
Introducing... the Subspace Chronicles (ft. @robvanderg, @barbara_plank, @iatitov)—a riveting tale of linguistic information emerging, shifting and disentangling across LM training! Coming to an #EMNLP2023 near you! 📜⏳
📑https://t.co/GJIXHEr5uR
📽https://t.co/p0gJuGDHpv
By treating probes as characterisations of task subspaces in an LM, we can further quantify linguistic relatedness via how much information is shared across tasks, over time.
📢 Calling all #NLProc enthusiasts!
Join us at the NLP Workshop on Linguistic Variation at @ITUkbh 🇩🇰 on Oct. 5th & 6th.
🎙️ Exciting speakers and an open to all!
This event is made possible by Prof. Dr. @barbara_plank's Multi⭐ grants.
Register + Info: https://t.co/1JViaPphCV
Super happy to announce the first publication result of my research visit at the @TurkuNLP lab in 2022!
Many thanks to Filip Ginter and Sampo Pyysalo for hosting me and to the group for the nice time spent there 🇫🇮💬
#NLProc
1/3
Hi! I'm Max (@mxmeij), a 3rd-year PhD looking into representation learning for quantifying linguistic variation, together with @barbara_plank and @robvanderg. #IntroFriday#NLProc 🧵 [1/5]
https://t.co/RluFDIqjXS
Great job @jsbaan! Very clear appeal for why you should take human disagreement into account when calibrating your models 🎲 #EMNLP2022
📑 https://t.co/ymwHuzRbax
What a poster session: Thank you all for stopping by!
If you missed it and want to know the secret on how to choose the best LM for your task without fine-tuning, find us (@mxmeij@mjjzha@barbara_plank) around the conference in the next days 😁👋
#EMNLP2022
For #EMNLP2022, we (w/ @robvanderg, @barbara_plank) look through differentiale, rainbow-colored glasses to find linguistic timescale profiles for 7 #NLProc tasks across 6 languages 🌈
📑 https://t.co/jYvR67UhuX
📽️ https://t.co/4tloTLSU5x
💬 10th Dec 9:00 at Poster Session 7 & 8
The resulting spectral profiles are visually and linguistically intuitive across 7 diverse tasks, e.g. PoS -> short, Syntax -> medium, Sentiment -> long waves🌊
Ever wanted to compare performance of several encoders *before* fine-tuning on your task? 🔮
“Evidence > Intuition: Transferability Estimation for Encoder Selection” has been accepted at #EMNLP2022! This was a cool collab with @EliBassignana@mxmeij@barbara_plank 🚀
🧵 (1/6)
If you are at #NAACL2022 I hope you get a chance to meet & check out the works by Max @mxmeij, Mike @mjjzha and Rob @robvanderg on choosing a LM encoder for parsing via probing, IE for computational job markets and MTL for SemEval tasks @NLPnorth@CisLmu#NLProc - times & loc 👇
Why choose LMs in a Chinese room when you can probe to rank?
📄https://t.co/3E71cIHWF5
Learn more at our 14:30 #NAACL2022 poster and check out our reenactment of the current SOTA:
🎬https://t.co/ptXgw0TE3o
(ft. @mjjzha, @robvanderg, @barbara_plank)