Structural Sentence Decomposition via Open Information Extraction

Authors

Abstract

The field of construction engineering is governed by an important volume of legal texts. Each of these texts provide a set of requirements by means of sentences written in natural language. These texts support conformity checking process of objects in construction engineering. Automate or at-least semi-automate this conformity checking process is the target of our project. This automation imposes to be able to make legal requirements processable. Thus, we envisage to rephrase n atural language sentences into a set of atomic requirements (i.e a triple: ). Further, we must identify between the facts found in a given sentence, the exist relations (e.g sentence = f act1 AND f act2 or IF f act1 THEN f act2, etc.), expressible using logical operators. For facts extraction, we have used Open Information Extraction-systems. Since they face a drop of precision mainly due to multi-word expressions, provide a method to handle them before OIE itself. Moreover, we also tackle the problem of enumeration items to improve OIE performance. Using OIE for legal sentences semantics handling, handling of multi-word expressions and contextual phrases and finally computation of relations between facts constitute the originalit y of this work. Such decomposition is the prerequisite to our target: automatic conformity checking.

Conference

18th International Conference Information Visualisation (IV2014), July 2014