Our goal in this paper is to investigate the effectiveness of relation extraction techniques for the slot-filling task. We discuss two relation extraction systems. YRES follows the traditional paradigm in relation extraction, where a system takes advantage of available examples for each relation to be extracted. On the other hand, SONEX follows the open relation extraction paradigm, where the relations to be extracted are assumed to be unknown a priori. In particular, SONEX applies clustering techniques to identify relations in an unsupervised way. The results of our submissions show that the performances of the two systems are fairly similar.
展开▼
