Relationship extraction from Thai children's tales for generating illustration

Abstract

Telling tales is a great way to boost brain power and imagination of childrens. When kids listen to tales, they imagine in their mind and create images of the characters and scenes. These kinds of intelligence exist in humans, but it is a challenge for machines. Imitating human creativity is one of the challenges in artificial intelligence field. This paper proposes the extraction of characters, scenes and relationship between one character and another from Thai children's tales. We construct a corpus for Thai children's tales called Nithan Thai and represent the semantic of the tales using a conceptual graph. The extracted information are evaluated by experts with the three questions, (i) which characters that you think their images should be appeared in the scene, (ii) which location that you think it should be presented in the scene, and (iii) what is the most noticeable relationship in the scene. The experiment results show that the correctness of the proposed method in terms of F-measure is 80.74%.