System, Method, and Apparatus for Information Extraction of Textual Documents

Abstract

A method and system for extraction of text from a set of text document(s). A plurality of document segments are identified within a given text document. For each document segment, at least one structured annotation is embedded within the document and associated with the given segment. The structured annotation specifies the start and end of the given document segment and a rhetorical relation associated with the given segment. The structured annotations are processed generate a plurality of variables that represent document segments and associated rhetorical relations as specified by the structured annotations, and such variables are stored in a data repository. A user interacts with a computer to define query input that specifies at least one rhetorical relation of interest. The query input specified by the user is processed to query the variables stored in the data repository to identify zero or more document segments that are associated with a rhetorical relation that matches the at least one rhetorical relation of interest specified by the query input. Information corresponding to the zero or more matching document segments is returned to the user. In the preferred embodiment, the rhetorical relations represented by the user supplied query input as well as the variables stored in the data repository include a set of RST relations whose meaning is dictated by nuclearity of the associated text. Such RST relations can include a plurality of mononuclear RST relations each having a nucleus and a satellite and a plurality of multinuclear RST relations each having a plurality of nucleus. The rhetorical relations represented by the user supplied query input as well as the variables stored in the data repository can also include a set of Speech Act relations whose meaning extends beyond the situational semantics of the associated text.

Dependencies are inferred from the claim text.