Conversational Frames: Improving Conversation Context in Smart Personal Assistants

摘要

This paper extends the previous work of Almousa and Migdady by proposing and implementing three models to improve the correctness and naturality of the given answers of smart personal assistants (SPAs). The motivation behind the three proposed models is the failure of some well-known existing SPAs (Siri and Salma) to extract contextual information from a conversation. The authors call this kind of information: the conversational frame. For evaluating the suggested models, the authors also implement an abstraction of the existing SPAs that they call non-framed model. To evaluate the four implementations, 36 respondents answered an on-line questionnaire after watching four videos that the authors recorded per implementation. The authors statistically prove that their three suggested models precede the non-framed model. The researchers attribute this to the fact that our models overcome the shortcoming present in non-framed model. Moreover, their model that implements the conversational frame precedes all other models.