ClimateBen Remember:
0.5°C safety
1°C irreversible feedbacks (now)
1.5°C a dangerous world
2°C an extremely dangerous world
2.5°C moving towards Hell
3°C horror
3.5°C death of organised society
4°C human survival in doubt
5°C likely extinction
6°C our current trajectory
Climate 'change'.
publictorsten Hätte dieser Falschparker vier Räder und einen Benzintank, würde er nicht weiter auffallen. Ohne wirkt er kurios. pic.twitter.com/OxacAmWg7d
annargrs Here's a summary post on problems with huge models that dominate #NLProc these days. I put together several different discussion threads with/by @yoavgo, @jaseweston, @sleepinyourhat, @bkbrd, @alex_conneau, @SeeTedTalk.
hackingsemantics.xyz/2019/leaderboa…
timoconnorbl Idly reflecting: there's been a psychological sea-change in regard to Brexit. I don't think the EU27 care any more. The feeling now is that a country willing to give power to someone like Johnson is not a serious country, and is just an external problem to be managed.
BuckRivera Du (ein Farbtourist): Ich fliege in die Türkei, um die Farbe Türkis zu erleben.
Greta (kommt und streicht deinen Vorgarten türkis): O no, you don’t. twitter.com/Turkey_Home/st…
laetusaeneas Komisches Wetter in Mannheim. Bei Thomas Mann würde jetzt die Cholera ausbrechen.
living_marble Imagine being the richest man on Earth when scientists tell us we have a decade to get our carbon emissions under control and instead of spending your billions on THAT you're spending your billions on rockets while having the gall to claim you're doing it for future generations. twitter.com/mims/status/11…
joshraclaw It’s almost like we got an entire book explaining the actual functions of these kinds of words over thirty years ago! twitter.com/The_GrammarGee… pic.twitter.com/MWq6nwa7VX
nlpmattg In general, I think this a strong argument for making smaller datasets, and testing the *same* model on many different datasets. These models are incredibly good feature extractors, if you give them enough data. Don't give them enough data to let them exploit quirks. twitter.com/LeonDerczynski…
Bleeptrack There is Videotext in this car. I'm done with technology for today. pic.twitter.com/T3kdqc1jDI
GilesPalaeoLab This is a BAD TAKE.
Being a scientist (and/or academia) is a JOB. It should not have to be an obsession. You can be passionate about it as a JOB. You can also do science as a hobby. But it does not have to rule your life, and you should not have to live-eat-breath science. twitter.com/blotreport/sta…
zirkuspony gut: meine chefin hat nicht bemerkt, wie lange ich mittagspause gemacht habe
schlecht: meine chefin bin ich
derwahremawa Gleithörnchen, aber als Fahrwerksbauteil
mdlhx We find that our parser does not learn similar information for these two, unless we use recursive composition as in this work: aclweb.org/anthology/N19-…, this is good news, maybe we have not yet found the best way to use this recursive layer! 3/4
mdlhx <TLDR> We do a linguistic investigation of a neural parser, looking at agreement and transitivity information in auxiliary verb constructions (AVCs) in comparison to finite verbs (FMVs). 1/4
c_borstell OK, so some points on posters in #academia:
Posters shouldn't be seen as a 2nd grade stage presentation. They are often seen this way, unfortunately, but some topics work better in a poster format. But conference organizers need to ensure posters are dealt with adequately. 1/6 pic.twitter.com/Bq5pIoP6fx
Chaosfux Das beste an Community Management ist dass man da pluralis majestatis benutzen kann ohne dass es seltsam wirkt
benbenhh BERT is very good at being right for the wrong reasons.
Great analysis of BERT's ability to learn exploiting annotation artifacts better than other models: arxiv.org/abs/1907.07355
Performance drops from 77% to random chance level when these artifacts are removed.
gdm3000 Amazing!! Deep Learning-based NLP techniques are going to revolutionize the way we write software. Here's Deep TabNine, a GPT-2 model trained on around 2 million files from GitHub. Details at tabnine.com/blog/deep #nlproc pic.twitter.com/VWB4XsyD4T
EnglishGibson Up with which I will not put pic.twitter.com/rPPhE58Qin
FSHornschild Erneuerbare-Energien-Gesetz (EEG) hat Deutschland zum 🌍 Champion in Photovoltaik-Branche gemacht. Dann hat Industrie-Lobby @insm 2012 Kampagne gegen #ErneuerbareEnergien gestartet. Kurz darauf haben @CDU und @fdp das EEG und damit die PV-Branche inklusive 80.000 Jobs zerstört. pic.twitter.com/m4kVoSl1Lr
InkyFirefly if you pronounce gif as "jiff", then i'm pronouncing jpg as "gaypeg", it's only fair
separating the grain of truth from the chaff of exaggeration
GretchenAMcC Anyone know of research on non-Latin writing systems and keysmash?
(Or, if you're a regular user of a different writing system, feel free to keysmash in the replies below! Let us know whether you're on a phone vs computer) twitter.com/C_Shoup/status…
LeonDerczynski This reproducibility exercise is a great step for NLP, with a really broad range of complex tasks: wordpress.let.vupr.nl/lrec-reproduct… -- I wonder if anyone will manage to produce any of these often quite important results? #nlproc