First, we incorporate the semantic role system defined in PropBank into the Predicate-argument structure Thesaurus. This extension allows us to capture the semantic relations between predicates and their arguments even in different constructions, such as the causative, passive and adversative passive forms, by describing both numbered arguments and named arguments (e.g., Agent and Theme). The dictionary entries have also been maintained and published on the website. Second, for construction of automatic annotation system of predicate-argument structures, we have applied various kinds of models such as deep learning models and models based on Bayes' theorems and revealed the approaches to improve the performance. Finally, we have developed a concordancer using an automatic annotation system of predicate-argument structures. In the concordancer, the users can combine block-based patterns that allow us to extract matched texts .