A pilot study on automatic inference rule discovery from Turkish text

In this paper we present a framework for extraction of inference rules from Turkish documents, such as "A solved B ~ A found a solution to B". Many natural language processing tasks, such as question answering, information retrieval and machine translation can benefit tremendously from inference rules. Our framework consists of three layers: construction of dependency trees; determining predicates and their complements and finally discovery of inference rules via clustering. We use Silhouette Index (SI) of the clusters as an automatic evaluation metric.