2

Am curious as to the number of homographs (sets of word meanings that share a common spelling) that occur in the English language. Also what the current state of the art would be for automatically differentiating senses.

The GloVe project makes available a word vector model with a 2.2 million vocabulary based on 840 billion tokens (taken from the Common Crawl dataset] - unfortunately from my understanding there model takes no account of homographs and I'm firstly trying to get a feel for how much of an effect that would have, and secondly thinking of how to go about recreating a similar model using word senses instead of words.

3
  • Is "hard" one or not? (goes to the question of counting meanings)
    – user6726
    Commented Dec 16, 2017 at 1:00
  • well 'hard' surfaces, and 'hard' problems shouldn't be using the same word vector in my view
    – norlesh
    Commented Dec 16, 2017 at 1:03
  • True homographs have different origins. If you really just mean common spellings then you should call them polysemous words instead.
    – amI
    Commented Dec 18, 2017 at 21:29

1 Answer 1

0

While this Wikipedia page on English homographs as it stands lists 164, as the following paper title suggests: University of Alberta norms of relative meaning frequency for 566 homographs there are substantially more, the same paper referencing other studies finding as much as 44% of randomly sampled words forms have multiple meaning (and 85% of the most common words in the language).

According to the University of Alberta study

many of the homographs in our list have more than two meanings; nearly half (230) have three or more meanings, and 75 have four or more meanings

as to using separated word senses to build word vector models - according to this answer on Quora something close to the state of the art (published 2015) can be found in the paper Breaking Sticks and Ambiguities with Adaptive Skip-gram which describes the AdaGram model that from my limited understanding estimates and differentiates word senses as it builds the model with similar computation requirements to word2vec.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.