Kilian Evang (@texttheater)

Berlin, Germany

The below is an off-site archive of all tweets posted by @texttheater ever

August 2nd, 2019

spacy_io Say hello to spacy-pytorch-transformers!

🛸 BERT, XLNet & GPT-2 in your spaCy pipeline
🤗 Based on @huggingface's pytorch-transformers
🎚️ Fine-tune pretrained models on your task
📦 Model packages for English & German
🚀 Token alignment, similarity & more
explosion.ai/blog/spacy-pyt…

via Twitter Web App (retweeted on 11:11 PM, Aug 2nd, 2019 via Twitter for Android)

wayfu Parkett - der beste Bodenbelag der wellt

via Twitter for Android (retweeted on 11:07 PM, Aug 2nd, 2019 via Twitter for Android)

drguidoknapp 1969: Die dt Brauer werben ua in der BRAVO, um Bier ein jugendlicheres Image zu verpassen. Vielen gilt es als Altherrenbrause. Slogan: "Bier ist nicht lapschig, klebrig-süß. Hat nichts Künstliches. Bier ist ehrlich. Echt und gut. Bier hat Wumm."

via Twitter Web App (retweeted on 10:56 PM, Aug 2nd, 2019 via Twitter for Android)

Matrix Reloaded. The Phantom Menace. Tron Legacy. All above 50%? Well, Equilibrium then. twitter.com/xkcdComic/stat…

via Twitter for Android

"In Deutschland haben wir keine Minikuchenmentalität."

via Twitter for Android

nsaphra QA: why scores and not probability distributions? Noah: Softmax is the source of a lot of uninterpretability!

via Twitter for iPhone (retweeted on 5:53 PM, Aug 2nd, 2019 via Twitter for Android)

nsaphra On WSJ PTB: "I'm tired of parsing the writing of Reaganist newsroom white men in the 80s."

via Twitter for iPhone (retweeted on 5:53 PM, Aug 2nd, 2019 via Twitter for Android)

nsaphra Noah just said he only likes methods he understand mathematically. A bold methodological diet!

via Twitter for iPhone (retweeted on 5:53 PM, Aug 2nd, 2019 via Twitter for Android)

nsaphra "We should care about interpretability because it's very hard to improve a model you don't understand." pic.twitter.com/vxDNsQGTiQ

via Twitter for iPhone (retweeted on 5:53 PM, Aug 2nd, 2019 via Twitter for Android)

nsaphra "Sometimes reduced capacity can be beneficial!" Oooo I'm feeling a return to old school learning theory here

via Twitter for iPhone (retweeted on 5:52 PM, Aug 2nd, 2019 via Twitter for Android)

nsaphra "Rational Recurrence" is a category of RNNs with a mapping to WFSAs -- unlikely to contain LSTMs, but does include a lot of recent architectures.

via Twitter for iPhone (retweeted on 5:52 PM, Aug 2nd, 2019 via Twitter for Android)

nsaphra But RR accuracy is really good for the very small parameter size!

via Twitter for iPhone (retweeted on 5:52 PM, Aug 2nd, 2019 via Twitter for Android)

nsaphra RRs are weighted automata, but with the weights as functions of vector representations, giving "soft patterns"

via Twitter for iPhone (retweeted on 5:52 PM, Aug 2nd, 2019 via Twitter for Android)

phonologist Alright, folks, share widely.

I'm tired as a disabled computational linguist to see the needs of disabled folx at *ACL events ignored, so I'm going to formally complain.

Please fill out this form if you've experienced accessibility issues
forms.gle/Dr5dH9mqMrx4od…

via Twitter for Android (retweeted on 5:50 PM, Aug 2nd, 2019 via Twitter for Android)

tinawirnsberger In Grönland sind gestern binnen 24 Stunden 12 Milliarden Tonnen Eis geschmolzen, in Sibirien wüten Waldbrände von der Größe Belgiens.

Bloß keine S c h n e l l s c h ü s s e in der Klimapolitik jetzt bitte!

via Twitter for Android (retweeted on 5:48 PM, Aug 2nd, 2019 via Twitter for Android)

nsaphra QA now. Willem Zudema wants to ban the Chomsky hierarchy. Bob agrees that we should be more granular about abstract properties!

via Twitter for iPhone (retweeted on 5:45 PM, Aug 2nd, 2019 via Twitter for Android)

nsaphra Much Ado about result that LSTMs can't generalize Dyck (CFG-complete) languages to new recursive depths at inference. But are RNNs even closed under homomorphism? If not, then even if they can learn Dyck, it means nothing for general capacity to learn CFGs!

via Twitter for iPhone (retweeted on 5:45 PM, Aug 2nd, 2019 via Twitter for Android)

ArneKoehn So at I just gave several poster presentations (~15min) for a poster where I don't even know the authors -- they seemed to be on lunch break at the start of the session.

Author is now here,will see whether I did good. Yes, I told everyone that I am not an author etc twitter.com/texttheater/st…

via Twitter for Android (retweeted on 2:52 PM, Aug 2nd, 2019 via Twitter for Android)

ML_NLP State-of-the-art result for all Machine Learning problems
github.com/RedditSota/sta…

via Twitter Web App (retweeted on 2:36 PM, Aug 2nd, 2019 via Twitter for Android)

nimirea_ @ACL_NLP @ACL2019_Italy but I'm not a wheelchair user. I'm gonna talk abt another problem that directly affects me: food labeling

photo 1: these sandwiches were on same plate with same (veg) label, even though one has meat and tuna cream. photo 2: no label on table at all pic.twitter.com/FhSuQmi3L8

via Twitter Web App (retweeted on 2:33 PM, Aug 2nd, 2019 via Twitter for Android)

nimirea_ hey @ACL_NLP @ACL2019_Italy let's talk accessibility

this venue was NOT wheelchair accessible. will it be different at other conferences like EMNLP, COLING? unclear. info abt locations focused more on proximity of beach/Disneyland than access (mentioned for ACL 2020 at least)

via Twitter Web App (retweeted on 2:33 PM, Aug 2nd, 2019 via Twitter for Android)

OscarKocken Ik liep net door Hoog Catherijne en zag daar héél véél mensen die gedwongen waren om vernederende kleren te dragen. Tenminste, ík kan me niet voorstellen dat mensen die kleren vrijwillig dragen en wat je je niet kunt voorstellen is de waarheid toch?

via Twitter Web App (retweeted on 2:30 PM, Aug 2nd, 2019 via Twitter for Android)

Die Säle sind hier fast so intuitiv nummeriert wie die Sitzplätze der Deutschen Bahn.

via Twitter for Android

@sjmielke Deep Learning & Formal Languages: Building Bridges keynote 4 by John Kelleher

via Twitter for Android in reply to sjmielke

yoavgo The "Deep Learning and Formal Languages: Building Bridges" ACL workshop today is amazing, exactly what a workshop should be: focused, well defined, non-popular topic, a real meeting of interested experts working on a subject from different angles.

via Twitter for Android (retweeted on 2:24 PM, Aug 2nd, 2019 via Twitter for Android)

tfw you mistake someone for the author of a poster and they not only walk you through it but also offer a critique of the work /cc @ArneKoehn

via Twitter for Android

Iwillleavenow The 3 stages of horrific new tech developments:

1) "All of you freaking out about this are paranoid - it's great!"
2) "WHO COULD HAVE PREDICTED THIS WOULD GO HORRIBLY AND BREAK EVERYTHING???"
3) "We are creating a revolutionary new* field called 'studying the thing we broke.'" twitter.com/hypervisible/s…

via Twitter Web Client (retweeted on 2:05 PM, Aug 2nd, 2019 via Twitter for Android)

bomelino in einem kleinen dorf lebte einst ein mädchen namens dörte die für ihren rhabarberkuchen bekannt war
deshalb nannte man sie rhabarberdörte
sie eröffnete eine kneipe
die rhabarberdörtekneipe
dort hatte sie schnell drei schlingel als stammkunden
die rhabarbererdörtekneipenschlingel

via Twitter for Android (retweeted on 11:47 AM, Aug 2nd, 2019 via Twitter for Android)

sleepinyourhat I just bought credible carbon offsets to offset emissions from lab travel, computing, and events. It was _trivially_ cheap. Consider doing the same.
wp.nyu.edu/ml2/carbon-neu… pic.twitter.com/VuXp6TpySm

via Twitter Web Client (retweeted on 11:45 AM, Aug 2nd, 2019 via Twitter for Android)

lutzmache An update on Android for search providers in Europe @Google blog.google/around-the-glo…

via Twitter Web Client (retweeted on 11:45 AM, Aug 2nd, 2019 via Twitter for Android)

hzulla Zusammen mit Flugtaxi ist Hyperloop ein Paradebeispiel dafür, wie der naive Glaube an eine zukünftige "innovative" Technologie davon ablenken soll, dass man mit ÖPNV, Bahn & ICE längst funktionierende Lösungen des Problems hat, welche man aber kaputt gespart hat. twitter.com/ZDFheute/statu…

via Twitter Web App (retweeted on 11:42 AM, Aug 2nd, 2019 via Twitter for Android)

durst Typisch ich: Musste heute zusammen mit dem Pöbel zur Fabrik weil der Zeppelin ein Loch hatte und kann nun den Zigarrenabschneider nicht finden weil ich das Monokel im Bus vergessen habe.

via Twitter Web App (retweeted on 8:55 AM, Aug 2nd, 2019 via Twitter for Android)

GalaxyKate "This AI can generate text"
!=
"This AI can generate semantically-coherent text"
!=
"This AI can generate causally-coherent text"
!=
"This AI can generate narratively-coherent text"
!=
"This AI can generate interesting text"

via Twitter for Android (retweeted on 8:54 AM, Aug 2nd, 2019 via Twitter for Android)

kle_blatt Könnten wir in Erste Hilfe Kursen bitte mal zwei Sätze über psychische Notfälle wechseln?

via Twitter for iPhone (retweeted on 8:54 AM, Aug 2nd, 2019 via Twitter for Android)

texttheater Das ist schon eher meine Krakenweide!

via Twitter for Android (retweeted on 6:50 AM, Aug 2nd, 2019 via Twitter for Android)

illadvisedno I want "small and accessible computing environment" to mean that I can train your model in my two hour time slot on the free computer at the public library

via Twitter for Android (retweeted on 6:47 AM, Aug 2nd, 2019 via Twitter for Android)

derwahremawa – Spielst du mit mir und Hugo Ball
– Nein
– Warum
– Weil ich schon mit Norbert Bolz

via Twitter for iPhone (retweeted on 5:03 AM, Aug 2nd, 2019 via Twitter for Android)