“Do you mean I was wrong?” A Preliminary Approach on a Graph-based Framework for Suggesting Alternate Interpretations on Japanese Conversations

When a person cannot predict how his or her speech will be interpreted by others, communication problems will happen in person-to-person communications. In the case of communication at workplaces, junior staff may receive his or her supervisor’s words as verbal violence even if the supervisor spoke no offense words. This research aims to achieve the method that shows the candidates of other person’s interpretations in advance. If the interpretations were shown in advance, we can avoid speaking the words eliciting misunderstanding. As a concrete application, this research focuses on the conversation on text chat software. The text chat software shows the candidates of text interpretation which the other person will feel. An opinion mining research has reported that building a semantic tree is effective for text meaning recognition. The research of misinformation detection also has reported the effectiveness of graph data use. In this study, we construct a semantic tree to recognize Japanese text conversations. We also implement the function that transforms the text based on the grammar to show malicious meaning the receiver may perceive. The evaluation showed that the proposed method can transform texts into other texts that clearly express malicious meanings. A translation process was done in practical time, which was 0.32 seconds on average.