Semantic Role Labeling Approach for Evaluation of Text Coherence

Detection of semantic roles associated with linguistic elements is important to the textual classification of communicative context into specific identities. In this paper, a new model for semantically identifying sentences is presented through contextual patterns. The proposed contextual pattern originated its structure from a labeling process of the semantic roles provided by constituents of a sentence within a semantic frame. Semantic roles of the pattern elements are properly identified through word sense disambiguation and accordingly the entire patterns sense is evaluated. Such semantic identification of text sentences is a generic semantic role labeling approach that could support many computational linguistic applications. A utilization of the proposed semantic labeling approach is introduced in the paper through a novel algorithm for text coherence evaluation. Coherence evaluation is provided by a matching task to individual semantic patterns and their relations to each other as well as patterns organization within the text segments. Results proved good capability of the modelling of contextual pattern, addressing semantic roles, to accurately evaluate text coherence. It has been shown that both contextual patterns labeling and coherence evaluation algorithm proposed here are generic, topic free and semantically arbitrated by the global concept within context.