Measuring Closure Properties of Patent Sublanguages

Proceedings of Recent Advances in Natural Language Processing, pages 659–666, Hissar, Bulgaria, 7-13 September 2013. Measuring Closure Properties of Patent Sublanguages Irina P. Temnikova Institute of ICT Bulgarian Academy of Sciences irina.temnikova@gmail.com Negacy D. Hailu Computational Bioscience Program U. Colorado School of Medicine negacy.hailu@ucdenver.edu Galia Angelova Institute of ICT Bulgarian Academy of Sciences galia@lml.bas.bg K. Bretonnel Cohen Computational Bioscience Program U. Colorado School of Medicine kevin.cohen@gmail.com Abstract Patent search is an important information retrieval problem in scientific and busi- ness research. Semantic search would be a large improvement to current technolo- gies, but requires some insight into the lan- guage of patents. In this article we test the fit of the language of patents to the sublan- guage model, focussing on closure prop- erties. The research presented here is rel- evant to the topic of sublanguage identi- fication for different domains, and to the study of the language of patents. We in- vestigate the hypothesis that fit to the sub- language model increases as one moves down the International Patent Classifica- tion hierarchy. The analysis employs a general English corpus and patent docu- ments from the MAREC corpus. It is shown that patents generally fit the sub- language model, with some variability be- tween categories in the extent of the fit. 1 Introduction The study presented in this article aims to con- tribute to two important Natural Language Pro- cessing (NLP) applications: patent search and sub- language identification. 1.1 Patents and Patent Search We define patents as “legal documents issued by a government that grant a set of rights of exclu- sivity and protection to the owner of an invention” (Alberts et al., 2011). Patent search is an impor- tant Information Retrieval (IR) problem due to the financial risks involved in accidentally breaking previously registered patent rights, and due to the complexity of the phenomenon. Patent search is carried out by a variety of users, including patent specialists, managers, researchers, attorneys, and inventors. There are multiple scenarios requir- ing patent search (Alberts et al., 2011), as well as multiple types of patent search tasks—state-of- the-art, novelty, patentability, infringement, free- dom to operate, and due diligence (Hunt et al., 2007; Joho et al., 2010). Different user types are prompted to adopt dif- ferent and often complex search techniques, re- flecting their different search aims and search tasks (Hunt et al., 2007). Search techniques in- clude classification code search, keyword search, full-text search, forward and backward citation of related documents, inventor or author search, patent assignee search, patent family search, legal status, and cross-language search (Alberts et al., 2011). Among these, full-text search is considered to have relatively more advantages than the other types of search techniques, as it allows the user to access the full semantic contents of the patent doc- ument (Adams, 2010a). However, in its present state, full-text patent search still exhibits several shortcomings, such as poor precision and lack of disambiguation (Adams, 2010a; Adams, 2010b). Besides the increased IR field attention towards patent search (see the CLEF-IP1, TREC-CHEM2, …

Content truncated to 500 words