It's an image so evocative that we're still using it today. Mostly because we still haven't fixed the problem.

The word #gerrymandering is a combination of 1800's politician Elbridge Gerry and a salamander. The twisted electoral map signed into law by Gerry was said to resemble a mythic beast, an Elbridge horror.

#etymology #politics #linguistics #language

#boycott is named after a person. Not the person doing the boycotting but the one being boycotted. While most eponyms are in someone's honour, this is more of a name and shame.

Erne earned the ire of the Irish but, boy, Boycott bore the burning brunt.

#linguistics #etymology #language #words

The fact that most legalization legislation uses the term “cannabis” is proof it worked.

In honour of 4/20, the #etymology of #marijuana : from unknown origins to xenophobic propaganda. The use of “marijuana” (the word) was meant to make the use of marijuana (the drug) seem more dangerous.

#linguistics #words #420 #drugs #words #weed

On balance, male #privilege is pretty sweet but sometimes I just want to live my best #WinnieThePooh life. Society just isn't ready for it.

#standup #joke #DonaldDuck

Sometimes #family can be too close. I don't really believe they're #siblings but maybe step siblings...

#standup #crowdwork

Also, I keep saying "token" because that's the accurate term. They're a little shorter than a word (~4 characters) but if it helps your understanding, you can just think of them as words.

Despite their empirical performance, it's important to remember that at their core, an LLM's main goal is to generate fluent, plausible, and pleasing text. While truth and accuracy are one way to achieve that, it's not the only way.

#language #AI part 5: LLMs are stupidly brilliant. The tech isn't new but the scale is. Large Language Models are LARGE. 10k times more training data than earlier models. 50k times more neural network parameters. And context windows that are at least 400x larger.

#linguistics #computerscience #LLM

We're building our way towards modern natural language processing systems like LLMs but we aren't quite there yet.

#language #AI part 3: word embeddings encode word meaning AND relationships between #words . They solved a lot of the problems of bag-of-words and ngram models but still have issues with polysemy.

#linguistics #computerscience

Do you know how much time and tape it took to get that microphone grill to stick to my nose? And in the end, it just looks like I photoshopped it on. Please come to a show so I don't feel like my effort was all in vain.

TFIDF reweights the count (Term Frequency) based on how rare the word is across documents (Inverse Document Frequency). The idea being that if a word appears in few documents overall, it should carry more weight when it does appear.

There are variations on these count vector approaches that aren't as complicated as the advanced word representations we'll cover in part 3 but can reveal additional nuance. Binary bag-of-words uses not the count, just whether a word in is a document or not.

#language #AI part 2: bag-of-words and ngram language models are simple but powerful. More complex word representations exist but these are usually a good first attempt.

#linguistics #computerscience

It shouldn't be confused with Neuro Linguistic Programming. One is woo that's claimed to be life altering by people who use it without understanding how it's supposed to work. The other is... actually that describes both but Natural Language Processing has a stronger empirical evidence base

#Language #AI part 1: What's "natural" about Natural Language Processing?

#NLP is the AI subfield related to language.

#linguistics #computerscience

And I did get permission from her to post this

I'm twice her size and don't think I could get that kind of distance. She'd be great at shot put but only if she was half asleep.

Sometimes I ask myself which I hate more: being single or making the #bed

My #girlfriend doesn't hog the #blankets so much as banishes them to the shadow realm. I wake up shivering, she's burritoed in my duvet, top sheet's MIA, and my weighted blanket is thrown across the room

#sleep #relationship

If you're one of the 5 people in Vancouver who love both linguistics and wrestling and you happen to see me at a show, feel free to say hi. I'm awkward but I don't bite.

I know that #intelligence and liking #wrestling are completely unrelated but I'm not sure other people do. Sorry for having diverse #hobbies . Would you rather I watch hockey? That doesn't even have storylines!

Wrestling is theater and acrobatics and strength rolled into one. What's not to like?

If you thought my French pronunciation was bad, I assure you my Korean is worse. You're probably just less familiar with what it's supposed to sound like.

Why do the #korean speakers pronounce "coke" like it's a piece of male anatomy? I'm not entirely sure but it seems to be due to differences between Korean and English phonology and phonetic inventory.

#linguistics #pronunciation #language #ESL

I'm so fucking proud of this video but I fear my use of profanity will get it buried in the algorithm. Too bad there's not a word I can use to express my frustration.

The #grammar of #fuck is fascinating and includes some rules that are exclusive to #swearing . It's potentially the most versatile word in #English and yet few people notice its uniqueness.

#language

I think Quebecois sacres are so neat I'm willing to once again risk the ire of francophones by attempting to speak #French on the Internet.

I know my pronunciation is terrible but I tried my best. Can we please be chill about it?

#swearing #slang #Quebec #language

It's a process that can run as long as the stigma that drives it.

The Euphemism Treadmill is what makes #words #offensive . And like a real treadmill, falling off can leave you mocked and embarrassed.

It's a cycle where #problematic words are replaced with new, neutral ones only for them to become problematic themselves.

#linguistics #language