Wietse de Vries (contact me by email) (@Wietsedv) profile picture

PhD student at the @GroNlp group of @univgroningen. I work on multilingual and cross-lingual NLP for Dutch and underresourced languages.

Joined April 2018
Photo by Wietse de Vries (contact me by email) (@Wietsedv)Photo by Wietse de Vries (contact me by email) (@Wietsedv)Photo by Wietse de Vries (contact me by email) (@Wietsedv)Photo by Wietse de Vries (contact me by email) (@Wietsedv)Photo by Wietse de Vries (contact me by email) (@Wietsedv)Photo by Wietse de Vries (contact me by email) (@Wietsedv)
Our new @GroNLP paper w/@martijnwieling and @MalvinaNissim, about cross-lingual transfer for >100 languages is accepted by @aclmeeting! We show that English is NOT the best training language and we discuss what factors do influence effective transfer. (Paper+models+demo+code) 1/n
10
46
6
198
Wietse de Vries (contact me by email) retweeted
After a year of restless development, I'm finally happy to announce Inseq, a new tool to democratize post-hoc interpretability of sequence generation models 🐛 github.com/inseq-team/inseq #nlproc #xai Some highlights 👇 1/
6
37
8
167
Wietse de Vries (contact me by email) retweeted
🥳We're back! Fresh start of the season @gronlp reading group, with plenty of ideas and enthusiasm for the year ahead. Always happy to host external #NLProc speakers. If you're interested to talk to us about your research, get in touch!
6
1
34
Wietse de Vries (contact me by email) retweeted
📢🥳 Our group is opening two 2y post-doc positions in NLP in the wider context of Humane AI! Come work with us! All details: rug.nl/let/sectorplan #NLProc Main gist of the two specific projects below 👇 Interested? Get in touch with @MalvinaNissim
1
19
2
28
Wietse de Vries (contact me by email) retweeted
Want to know more about our @univgroningen @FacultyofArtsUG @GroNlp #NAACL2022 paper? Check it out in our 🌟new🌟 video! 📽️ by the great @musicbyblanks youtu.be/c5nowiEItA4
Can we distinguish closely-related Dutch regional languages using only 6⃣ seconds of speech? With multilingual pre-training and fine-tuning on Dutch we can! Even more accurately compared to using a (new) monolingual Dutch model. Read it in our new @naaclmeeting paper! 🌟
7
19
Wietse de Vries (contact me by email) retweeted
That's a wrap for the @GroNLP team at #ACL2022! Now, on to the workshops! #acl2022nlp
5
41
Wietse de Vries (contact me by email) retweeted
Conference season is here!🎓 @Wietsedv just presented his paper (with @MalvinaNissim & @martijnwieling, @GroNlp) on cross-lingual transfer for >100 languages at #ACL2022 (@aclmeeting). 🎉 If you missed it, you can still catch him at his poster tomorrow!
4
15
Wietse de Vries (contact me by email) retweeted
🌍 "Make the Best of Cross-lingual Transfer: Evidence from POS Tagging with over 100 Languages" by @Wietsedv et al. will be presented in session OS4-6 at 12:00 CEST nitter.cf/Wietsedv/status/149678…
Our new @GroNLP paper w/@martijnwieling and @MalvinaNissim, about cross-lingual transfer for >100 languages is accepted by @aclmeeting! We show that English is NOT the best training language and we discuss what factors do influence effective transfer. (Paper+models+demo+code) 1/n
1
1
4
I am a bit confused by the #ACL2022 program. The list of accepted papers shows 699 long+short accepted papers (2022.aclweb.org/papers), but the program only shows 289 main ACL papers (mixed oral+poster) how did 410 papers vanish?
1
And why are there additionally 129 poster presentations for Findings papers mixed between all other poster presentations? That means that these papers should have been accepted and recognized as main conference papers, right? Am I missing something?
2
Wietse de Vries (contact me by email) retweeted
Can we distinguish closely-related Dutch regional languages using only 6⃣ seconds of speech? With multilingual pre-training and fine-tuning on Dutch we can! Even more accurately compared to using a (new) monolingual Dutch model. Read it in our new @naaclmeeting paper! 🌟
3
20
4
81
Wietse de Vries (contact me by email) retweeted
Our work w/ @MalvinaNissim on the Italian T5 is finally out! IT5 is the first encoder-decoder NLM pre-trained on Italian, advancing the state-of-the-art in many 🇮🇹 NLG & NLU tasks. Paper: arxiv.org/abs/2203.03759 Code: github.com/gsarti/it5 Demo: hf.co/spaces/it5/it5-demo 🧵👇
6
17
2
98
I think annotation efforts should focus on (higher-resource) languages from different language families and other clusters that can be identified by other discussed factors. Zero-shot transfer can be a solution for low-resource if there is training data for related languages.
1
1
9
Models and a demo can be found on @huggingface Spaces (second link). Evaluation results for each model can be found on the linked model pages. 📝 Paper: wietsedv.nl/files/devries_ac… 🤗 Models + Demo: hf.co/spaces/wietsedv/xpos 💻 Code: github.com/wietsedv/xpos
1
1
15
Our paper in which we ultimately recommend using Wav2Vec2-based acoustic representations to model subjective non-native pronunciation differences, has been accepted by the Journal of Phonetics! Transformers everywhere!
This Post is from an account that no longer exists. Learn more
1
5
Wietse de Vries (contact me by email) retweeted
And just in time for the holidays, exciting news 🥳: we're hiring! Assistant professorship in CL/NLP, permanent after 18mo if criteria are met. Vibrant atmosphere, great group! See full ad for details and contact https://nitter.cf/t.co/rTeh9F1r6F #NLProc @univgroningen @FacultyofArtsUG