OMCS Concept Relations

A project by Thomas Lin.


The OpenMind Common Sense project collected a large database of English sentences from volunteer contributors. Using the Brill part of speech tagger, I parsed 190,000 of the sentences into concepts (noun phrases) and relations (the shell of the sentence that's left after you turn all the noun phrases into variable slots). I then represented the data as a graph where the nodes are the concepts and the edges are the relations. One feature of this graph is that paths between nodes represent ways in which real life noun phrases are related. I put the graph into MySQL and wrote a Bidirectional Breadth-First search (in Perl) that runs over it to return the shortest path between any two nodes.


The program will tell you how two nouns (e.g. "dog," "cat") of your choice are related. If you enter two completely unrelated concepts, it's amusing to see what obscure and indirect relation is returned. If you enter two related concepts, it should find something similar to the expected relation.


(Currently offline)