Non-native speaker perception of Intelligent Virtual Agents in two languages: the impact of amount and type of grammatical mistakes

Abstract Having a mixed-cultural membership becomes increasingly common in our modern society. It is thus beneficial several ways to create Intelligent Virtual Agents (IVAs) that reflect background as well, e.g., for educational settings. For research with such IVAs, it essential they are classified non-native by members of target culture. In this paper, we focus on variations IVAs’ speech the impression speakers identified two different mother tongues. particular, investigate grammatical mistakes and identify thresholds beyond which agents clearly categorised speaker. Therefore, conducted experiments: one native German, English. Results German study indicate 10% word order 25% infinitive German-speaking IVAs perceived speakers. English 50% omission English-speaking We believe these constitute helpful guidelines computational approaches speaker generation, simplifying