Large Scale Relation Detection

We present a technique for reading sentences and producing sets of hypothetical relations that the sentence may be expressing. The technique uses large amounts of instance-level background knowledge about the relations in order to gather statistics on the various ways the relation may be expressed in language, and was inspired by the observation that half of the linguistic forms used to express relations occur very infrequently and are simply not considered by systems that use too few seed examples. Some very early experiments are presented that show promising results.