Annotation System for Dialogue Datasets of Older Adult’s with Photos and Storytelling

We have developed a state-of-the-art “Dialogue-Based System with Photo and Storytelling” (DBSPS). The system runs automatically when the experimental schedule is on time to reduce cumbersome screen-based operation (e.g., tap) as much as possible for older adults to be ready for home-based use. In this paper, we present an original annotation system that is designed to create a corpus by collecting data with our system. Along with our experimental conditions, it is essential to display the images and stories used as experimental conditions and the user’s questions, answers, and responses. Using the system, we conducted three annotations tasks (i.e., question form annotation, naturalness annotation, and similarity annotation). First, question form annotation is checking whether the user’s utterance is a question form or not. The second is naturalness, which aims to identify how our system could respond naturally. Third, the similarity annotation is designed to check the coverage of our system’s prepared datasets. Five annotators could use our system smoothly, and all tasks were completed without system errors. As a result, we identified areas to improve our dialogue system.