Researchers have analyzed an online thesaurus and found that relationships between words appear similar to human social networks. Like humans, words are connected by a small world network, each word being only a few semantic steps away from any other.
I have to wonder at how much this research is really measuring the properties of the thesaurus itself, and not the language. This isn’t drawing from a corpus - it’s looking at a book compiled by human beings.
If a word has very few actual synonyms, an editor might take considerable license in choosing words to put under the entry. Ultimately you end up studying human perceptions of language, rather than actual empirical data. To make a claim like this, you would really need a more rigorous definition of ‘relationships between words’.
It’s a little disheartening to see initial results for one book in English being hyped up as a universal property of language. Even though I suspect the property holds true ( an exciting result ) this kind of methodology just makes us empirical types look bad.
This is the personal weblog of Cameron Marlow, research scientist at Facebook. You can find a comprehensive list of content in the archives, or you can simply search using the box above.
