wow this was hard but fun. apparently i don't understand the structure of word2vec at all
semantle.novalis.org
Conversation
i was thrown off by a bunch of nearby words that had several very distinct meanings, i think some of the similarities were for different meanings and that took me a long time to sort out
1
7
the word2vec manifold must be folded in on itself in weird ways by words with multiple meanings, i was i guess naively expecting it to be a lot "flatter"
3
1
10
word2vec distance is not a measure of semantic similarity!!!
1
3
no kidding 😅 but what *is* it a measure of then?
1
1
some kind of loose notion of "substitutability / arises in same context" in the training corpus
1
3
e.g. words and their opposites are often kinda close together
1
2
Replying to
yeah that makes sense and is roughly consistent with my vague sense of how it must've been trained. hmmmm