Trust and Language Choice in Conversational Agents for Non-Native Speakers

Contact: Shatha Degachi (C.Degachi@tudelft.nl)

In this project, we investigate how the choice of language, English, or mother tongue, in conversational agents changes the perception of the agent and the user trust in the system. We focus on non-native speakers of English, who are also fluent in a low online resource language. Low online resource languages are those for which there is little examples available online, few experts, few new speakers. Thus, conversational agents trained on internet data often perform worse in these languages. Little work has been to investigate how this affects user perception, especially now that large language models have emerged which claim to show stronger performance on many tasks in comparison to previous models and which may perform differently on this class of languages. Trust is an especially complex and influential aspect of the user-system relationship, which impacts intent to use and meaningfulness of interactions. When designing conversational agents for less prioritized user groups, such as non-native English speakers, it is thus imperative to understand how language choice shapes their relationship to the system.
 
*highly preferable: Technical skill (can prototype a conversational agent)