This is an exciting new development in computational linguistics which is likely to have a large impact on the future development of search engines and conversational software.

http://www.physorg.com/news/2011-09-mathematical-enable-web.html

“The model provides a theory about how to represent words and phrases using vectors, or sequences of numbers… Our theory tells you what the vector for a phrase should look like in terms of the vectors for the individual words that make up the phrase.”

Great find, Andrew! I ran into similar issues while developing my little EMO module. It works by assuming words have some associated emotional vector (in a 7D “emotion space”). The program assumes that the emotional content of phrases could be derived by summing up the emotional content of the words that comprise the phrase (that is, summing the vectors associated with each word). The resulting vector would then be matched against the closest known emotion or combination of emotions. One can immediately imagine situations in which such an ansatz might fail.

The context here is different, but the principle is the same. How do you identify words and phrases that have equivalent meaning? Is it possible to combine the “meaning” vectors of individual words in a phrase using some scheme that results in a vector close to those of other synonyms?

I’ll have to read the paper. Found it on arXiv: http://arxiv.org/abs/1101.4479

C R Hunt - Sep 27, 2011:

I’ll have to read the paper. Found it on arXiv: http://arxiv.org/abs/1101.4479

Thanks for the link, I naively thought that I would have to wait for it to be published in “Computational Linguistics” before I could read it.

You know, it’s interesting that you should be bringing this subject up today, Andrew. I was also reading something about Computational Linguistics.

WARNING! VERY “tongue in cheek”!, also NFO* and NSFW**!

*Not Family Oriented
**Not Safe for Work

Ha ha ha—nice, Dave.

Andrew: Gotta love the arXiv. Probably the published version will be edited/updated, so it might still be worth checking out.

I wasn’t going to post that, but then you mentioned your work with your EMO module, and it tied in with the comic so nicely that I felt I had no choice but to mention it.

CR, You’re approach to affect analysis is very naive (consider the sentence “You do not have cancer”)  Check this out: http://linguistlist.org/pubs/papers/browse-papers-action.cfm?PaperID=33467

Rob: Yes, cases like your example are exactly why I said “One can immediately imagine situations in which such an ansatz might fail.” (Although come to think of it, my naive little program would probably handle this particular example well considering it contains the word “not”. But in general, I completely agree with you.) The module was just a little playing around on my part. Eventually I plan to expand on it, but it’s low on the priority list at the moment. Thanks for the link, I’ll check out the paper.

Edit: I’m looking through the paper, and it seems as though the authors handle noun phrases in the same manner I did. It’s interesting reading—they really delve into some unexpected nuances of phrasing with their sentence level analysis.