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

May 27th, 2019

Vilinthril Bei aller Kritik an beiden Projekten find ich's trotzdem symbolisch schön, dass DieM25 und Volt mit je einem MdEP vertreten sind.

via TweetDeck (retweeted on 9:20 PM, May 27th, 2019 via Twitter for Android)

davidschlangen Conference over. Very nice and instructive 3 days. One puzzle I’m left with is what the logo of Gothenburg University is meant to show. A person standing on a boat carrying a bicycle in front of the setting sun? pic.twitter.com/aj7vm2E5lK

via Tweetbot for iΟS (retweeted on 8:24 PM, May 27th, 2019 via Twitter for Android)

"Sweden has provinces that don't have any political meaning. They're just there to give titles to stray royalty."

via Twitter for Android

_roryturnbull This is a schwa
ə
Now there are two of them
ə ə
There are two ___

via Twitter Web App (retweeted on 2:21 PM, May 27th, 2019 via Twitter for Android)

MeaningBank Embedding spaces have not been treated fairly! Our colleagues at CLCG Groningen looked critically at the much cited analogy "party tricks". Check out their paper "Fair is Better than Sensational: Man is to Doctor as Woman is to Doctor": arxiv.org/abs/1905.09866

via Twitter Web Client (retweeted on 2:18 PM, May 27th, 2019 via Twitter for Android)

evanmiltenburg Nice work by Matthijs Westera and Gemma Boleda: “Don’t blame distributional semantics it can’t do entailment” (IWCS 2019): mwestera.humanities.uva.nl/downloads/West…

The authors argue that DS provides a model of *expression meaning*, which differs from *speaker meaning* (where entailment resides).

via TweetDeck (retweeted on 2:17 PM, May 27th, 2019 via Twitter for Android)

perseveresther The main issue vastly outmattered all the side issues to us

via Twitter for Android (retweeted on 11:54 AM, May 27th, 2019 via Twitter for Android)

labenzticker Schmarzmasser, der: Nicht zum Nieser führender Niesreiz. labenz.neutsch.org/Schmarzmasser

via Labenzticker (retweeted on 11:49 AM, May 27th, 2019 via Twitter for Android)

ojahnn tfw on the last conf day finally a woman asks something in the Q&A after a talk but it's yourself so you're not sure if it counts

via Twitter for Android (retweeted on 10:59 AM, May 27th, 2019 via Twitter Web App)

rikvannoord 6/7 This is not beneficial for dealing with the problem of actual biases in word embeddings. Moreover, these are usually not captured by the analogy task anyway (we agree that they are "party tricks" as stated by @yoavgo and @hila_gonen in this fine work: arxiv.org/abs/1903.03862)

via Twitter Web Client (retweeted on 10:59 AM, May 27th, 2019 via Twitter Web App)

rikvannoord 5/7 (ii) then, the analogy task doesn't work that well (no "queen" for "man is to king as woman is to X", but "king") and (iii) analogy-based biases have often been overemphasized in the literature.

via Twitter Web Client (retweeted on 10:59 AM, May 27th, 2019 via Twitter Web App)

rikvannoord 1/7 Do word embeddings really say that man is to doctor as woman is to nurse? Apparently not. Check out this thread for a description of a short paper I co-wrote with Malvina Nissim and Rob van der Goot, available here: arxiv.org/abs/1905.09866

via Twitter Web Client (retweeted on 10:58 AM, May 27th, 2019 via Twitter Web App)

ojahnn R-grams perform only slightly worse than standard tokenization on word embedding benchmarks. Word bias is a partial explanation for the lack in performance. Future work: Can we build embeddings directly on the grammar? Is it limited to text? Does it improve SoA models?

via Twitter for Android (retweeted on 10:57 AM, May 27th, 2019 via Twitter Web App)

ojahnn Putting "word" in scare quotes because treating everything between two spaces as one word breaks down quickly when you look at Mandarin or Finnish. R-grams are supposed to be more useful across languages with different morpheme-per-word densities.

via Twitter for Android (retweeted on 10:57 AM, May 27th, 2019 via Twitter Web App)

ojahnn The best student paper at is being presented by Amaru Cuba Gyllensten. It sounds really cool! It's about creating R-grams, which are kind of between the "word" and character level, with the aim of getting more semantically meaningful reps. Paper: aclweb.org/anthology/W19-…

via Twitter for Android (retweeted on 10:57 AM, May 27th, 2019 via Twitter Web App)

asayeed Gyllensten: evaluating word/character compromise embeddings like R-grams is hard with standard benchmarks because of word-basedness of the benchmarks.⬅️ Biased against non-word embeddings which may actually be better.

via Twitter for Android (retweeted on 10:57 AM, May 27th, 2019 via Twitter Web App)

asayeed The R-gram concept is an unsupervised way of finding an embeddings space that is more agnostic about the notion of "word" @iwcs2019

via Twitter for Android (retweeted on 10:57 AM, May 27th, 2019 via Twitter Web App)

StigAbell Elton John wins the award for the best diary entry ever (quoted in today’s Observer).

“Woke up, watched Grandstand. Wrote Candle in the Wind. Went to London, bought Rolls-Royce. Ringo Starr came for dinner”.

via Twitter for iPhone (retweeted on 10:56 AM, May 27th, 2019 via Twitter Web App)

niggi Für alle, die meinen, das Wahlergebnis in UK sei ein klares Votum für : twitter.com/EuropeElects/s…

via Tweetbot for Mac (retweeted on 9:20 AM, May 27th, 2019 via Twitter for Android)

JeremyCliffe Le Pen: down on 2014
FPÖ: down on 2014
AfD: little higher than pre-surge 2014 result
Wilders: massive fall
Baudet: underperforms

Predictably, proving more complicated than "march of the populists" narrative.

via Twitter Web Client (retweeted on 8:41 AM, May 27th, 2019 via Twitter for Android)

FridayForFuture Wenn die Parteien in Deutschland nach der heutigen jetzt endlich handeln wollen - hier eine Hilfe:

Für 2019:
Ende fossiler Subventionen
1/4 Kohlekraft abschalten
CO2-Abgabe, die Folgekosten entspricht

Nettonull 2035
Kohleausstieg bis 2030
100% Erneuerbare bis 2035

via Twitter for Android (retweeted on 8:40 AM, May 27th, 2019 via Twitter for Android)

"Like the heraldic shields of ancient knights, these morphemes were painted onto the names of scientific technologies to proclaim one’s history and achievements to friends and enemies alike." twitter.com/aeonmag/status…

via Twitter for Android

Whack-a-Mole, aber als Kosmetikum

via Twitter Web App