4 ms·
Word Embeddings as Metric Recovery in Semantic Spaces (2016) [pdf]
- trashtoss 10y agoVery late to see this but am curious how these approaches handle words with numerous meanings (like run, which arguably has dozens if not hundreds of idiomatic meanings in English). Distinct vectors for each meaning? One vector and the metric still works better than you'd think? It's not clear from the paper at a skim.