Alexander Koller
iPhone : Saptarshi Sengupta Rachel Thomas Emily M. Bender Oh wow, thank you! I’m glad you found the paper valuable to you. Happy new year!
Bio Professor (Computational Linguistics, Saarland University, @LstSaar). Musician (barbershop a cappella, singer, arranger, coach). Twitter newbie.
Tweets 89
Followers 681
Following 204
Account created 08-03-2009 19:00:29
ID 23340278
iPhone : Saptarshi Sengupta Rachel Thomas Emily M. Bender Oh wow, thank you! I’m glad you found the paper valuable to you. Happy new year!
iPhone : (Prof. Dr.) Noah A. Smith (but call me Noah) Haha. :) Have you seen this blog post in GPT-3? The problems seem similar. lacker.io/ai/2020/07/06/…
iPhone : Ernie Chang Oh wow, congratulations!
iPhone : It’s so fun to see how current NLP models can be fooled by sentences with false presuppositions. twitter.com/xkcd/status/13…
Twitter Web App : Jason Baldridge Who is the woman?
Twitter Web App : Join us as tenure-track faculty at our department! International applicants very welcome. We are looking for an excellent research record in any area of computational linguistics / #NLProc, after postdoc or potentially right after PhD. Contact me with any questions. twitter.com/LstSaar/status…
TweetDeck : Very excited to share our recent work with Alan Alan Ramponi: a survey in NLP on unsupervised adaptation - including an awesome repository
Paper: arxiv.org/pdf/2006.00632…
Repository: github.com/bplank/awesome… ✍️🥳
Thx to Nils Rethmeier Sebastian Ruder for feedback on the draft 😊
TweetDeck : Tatjana Scheffler I guess proper names are just hard. But their model does seem biased towards US names. "Martinez" is actually a German "Matthias". (I'll admit that "Martinez" is the cooler first name.)
TweetDeck : Slidelive's automatic subtitling strikes again for #emnlp2020: "My name is Martinez Lindemann, and today I'm going to talk about semantic parsing with, well, tightness guarantees. This is joint work with UNESCO, Auschwitz, and Alexander Column." 🤣
iPhone : Since jsLDA seems to have some traction I want to say a bit about its design goals. These are based on experience teaching ML to a range of experience levels in classes and workshops. twitter.com/Ted_Underwood/…
TweetDeck : Come get your MSc in computational linguistics at Saarland University! Join a very international crowd (2/3 non-German from ~20 countries) of brilliant students. Classes taught in English by internationally renowned faculty. Scholarships available. twitter.com/LstSaar/status…
Twitter Web App : (((ل()(ل() yoav)))) (3) Others will stay in research, and our job is primarily to help them solve the problems we can't. Then it's okay that we don't have finished answers for anything. Perhaps you need to teach two classes?
Twitter Web App : (((ل()(ل() yoav)))) (2) I can relate to your frustration, but I suppose it is worth asking why we teach. Some students will go out into the world and do real-life engineering, and they need NLP tools that can get _something_ done, plus perhaps a healthy does of humility towards language.
Twitter Web App : (((ل()(ل() yoav)))) These are excellent questions. Some thoughts: (1) If you can say with any kind of precision why language is hard and what the unsolved problems are, you should teach that not just to your students, but to the community. Everyone needs to hear that.
Twitter Web App : Christopher Potts ilyaeck Emily M. Bender Right, we don't argue that an AI system couldn't in principle achieve understanding. We also don't argue that this system has to be symbolic (neural is fine too) - just that it needs richer training data than just text.
TweetDeck : Six scientific contributions of our department were accepted for EMNLP. Congratulations to all authors!
Authors and titles in a thread >> 1/7
TweetDeck : Check out our new paper on fast (1-2k tokens/sec on CPU, 10k/sec on GPU) and accurate semantic parsing into graphs! Includes nontrivial, but provably dead-end-free transition system. #emnlp2020
Paper: arxiv.org/abs/2009.07365
Online demo: amparser.coli.uni-saarland.de:8080
iPhone : Here’s a less facetious take on the paper I linked to last night. As usual, Emily M. Bender makes some good and uncomfortable points. twitter.com/emilymbender/s…
iPhone : Turns out that GPT-3 _is_ evil after all. middlebury.edu/institute/site…
iPhone : Amy Isard It really isn’t. It is also a neat rhetorical trick to let GPT-3 talk about its intentions, obscuring the fact that it doesn’t have any.