Semantic Graphs

Construct a graph of words step by step

Graph a particular word

Find a cluster which a word belongs to

The Infomap Project is developing graph models representing words and their relationships. This focus has so far been on finding classes of nouns -- musical instruments, academic subjects, names of diseases -- directly from free text.

This work is important for developing automatic dictionaries. For example, in medical research, many many expert hours are devoted to reading documents and marking the occurence of new terms for diseases, syndromes and drugs. Due to the sheer volume of new information available every day, automatic methods are needed to classify new words, especially in languages other than English.

The following diagram shows the sort of results we obtain, directly from free-text and without human intervention.

Senses of the word <I>arms</I>.

Note the different clusters or components. We can infer from this that arms can refer to both body-parts and weapons.

This sort of technology can be used to identify the meanings of ambigous words and automatically work out which meaning is being used in a particular situation. Pretty neat, huh?

Try it for yourself!

Graph a particular word Construct a graph of words step by step Find a cluster which a word belongs to

You can find out more in the following paper:

Dominic Widdows and Beate Dorow. A Graph Model for Unsupervised Lexical Acquisition. 19th International Conference on Computational Linguistics, Taipei, August 2002, pages 1093-1099. (.ps)


Last modified: Wed Feb 16 19:45:28 PST 2005