Comparing Command Construction in Native and Non-Native Speaker IPA Interaction through Conversation Analysis

Intelligent Personal Assistants (IPAs) are limited in the languages they support, meaning many people are left to interact using a non-native language. Yet, we know little about how people interact with IPAs in this way. Through a conversation analysis (CA) perspective, we examine native (L1) and non-native (L2) English speaker interactions with Google Assistant, comparing how both user groups produce IPA commands. Our work shows that L1 and L2 speakers similarly used pauses, partial or complete repetition, and hyper-articulation when constructing commands. However, L2 speakers tended to experience issues in lexical access, syntactic construction and pronunciation, resulting in the use of code-mixing, increased pause lengths and off-task rehearsal to help generate commands. We consider reasons for such effects, whilst exploring ways to design IPA interaction to ensure it is sensitive to L2 challenges in command production.