Interacting with Intelligent Personal Assistants : Blending Voice and Chat Interaction to Improve Learnability

Abstract: The largest software companies in the world are racing to develop their Intelligent Personal Assistants. Each with more and more capabilities and ways for users to access them. The different assistants have different capabilities and one of the ways the companies are differentiating is in the interaction techniques their assistants that are controlled with, most uses voice, and some uses text. This thesis walks through these interaction techniques, give a brief overview of intelligent personal assistants today and explores how learnability differentiate when using these interaction techniques. The thesis also investigates when users have access to multiple interaction techniques to use in combination when interacting with such an assistant. To evaluate the field, a literature study was conducted covering Intelligent Personal Assistants, interaction techniques, conversational user interfaces, modalities, multimodal interaction, voice and chat interfaces. Interviews with early adopters of Intelligent Personal Assistants was performed to better understand the current state of the technology and how users were using it. To evaluate the learnability differences in different interaction techniques, a series of user tests in a wizard-of-oz setup was conducted. In the user tests the different groups was using different interaction techniques to interact with their assistants. The main findings of the user tests was that there are indications to that the type of language used differs depending on the interaction techniques. Users who interact through voice is more prone to use a more natural language, while users using a chat interface quickly started to treat the interactions with the Intelligent Personal Assistant as commands. The result observed was that users who interacted with the assistant using chat and then switched to interacting through voice used a shorter, more command-like language when they interacted through voice as well.