DFKI-LT - A Multi-Dimensional Classification Approach towards Recognizing Textual Semantic Relations

Rui Wang, Yi Zhang
A Multi-Dimensional Classification Approach towards Recognizing Textual Semantic Relations
2 Proceedings of the 12th International Conference on Intelligent Text Processing and Computational Linguistics (CICLing), Tokyo, Japan, Springer, 2/2011
 
Recognizing textual entailment has been known as a challenging task, with many proposed approaches focusing on solving it independently. From a broader perspective, there are other semantic relations between pairs of texts, e.g., paraphrase, contradiction, overlapping, independence, etc. In this paper, we propose three basic measurements: relatedness, inconsistency, and inequality, to characterize these closely related Textual Semantic Relations. We show empirically the effectiveness of these measurements for the recognition tasks (e.g. an improvement of 3.1% of accuracy for entailment recognition) with features extracted from dependency paths of the joint syntactic and semantic graph. With the semantic relation space based on these three dimensions, we show this is a way to achieve a better understanding of general semantic relations between texts.
 
Files: BibTeX, CICLing2010.pdf