Discovering semantic relations from the web and organizing them with PATTY

Abstract

PATTY is a system for automatically distilling relational patterns from the Web, for example, the pattern "X covered Y" between a singer and someone else's song. We have extracted a large collection of such patterns and organized them in a taxonomic manner, similar in style to the WordNet thesaurus but capturing relations (binary predicates) instead of concepts and classes (unary predicates). The patterns are organized by semantic types and synonyms, and they form a hierarchy based on subsumptions. For example, "X covered Y" is subsumed by "X sang Y", which in turn is subsumed by "X performed Y" (where X can be any musician, not just a singer). In this paper we give an overview of the PATTY system and the resulting collections of relational patterns. We discuss the four main components of PATTY's architecture and a variety of use cases, including the paraphrasing of relations, and semantic search over subjectpredicate- object triples. This kind of search can handle entities, relations, semantic types, noun phrases, and relational phrases.