| tags | |
|---|---|
|
I was randomly reading a paper from a journal whose name sounded legitimate but actually is one of those predatory journals (and supposedly has an acceptance rate of 90+%, hence why I'm not linking to them). The authors raised something provocative which I thought was worth debunking:
One of core insights in (statistical) NLP is the idea of using embeddings (vectors in Euclidean space) to represent words (concepts). However, this seems to raise some philosophical quandaries; namely, now that we're in Euclidean space, which is equipped with a natural metric, and allows us to
Aside: distributional hypothesis vs distributed representation? I'm pretty sure I use these terms interchangeably.