Patent claim decomposition for improved information extraction

Abstract

In several application domains research in natural language processing and information extraction has spawned valuable tools that support humans in structuring, aggregating and managing large amounts of information available as text. Patent claims, although subject to a number of rigid constraints and therefore forced into foreseeable structures, are written in a language even good parsing algorithms tend to fail miserably at. This is primarily caused by long and complex sentences that are a concatenation of a multitude of descriptive elements. We present an approach to split patent claims into several parts in order to improve parsing performance for further automatic processing.