Contribution of Automatically Recognized User Attributes to Dialogue Strategy Acquisition

Determine the extent to which automatically recognized user attributes contribute to dialogue strategy acquisition in real-world social-robot tasks.

Background

The paper discusses the use of vision-LLMs to recognize attributes such as age, gender, emotions, facial expressions, personality, knowledge level, urgency, and engagement for adaptive dialogue. Although automatically recognized attributes may provide information unavailable from dialogue history alone, the experiments found no substantial improvement in task-success prediction when attributes were incorporated, particularly for the relatively simple route-guidance task.

The unresolved issue is whether automatically recognized attributes meaningfully improve the acquisition of dialogue strategies in more realistic social-robot tasks. This question is important because identity-related attributes can introduce risks of misclassification, stereotyping, and discriminatory behavior, whereas transient interactional cues such as engagement and interest may provide safer and more useful bases for adaptation.