Human conversants in dialog adjust their behavior to their conversational partner in many ways. In terms of language use, they adapt to their partners both lexically and syntactically, by using the same referring expressions or sentence structure. In this paper, we describe a natural language generator PERSONAGE-PRIMED, which can produce utterances entrained to a range of utterance features used in prior utterances by a human user, and represented in the discourse context. PERSONAGE-PRIMED can entrain to the user's referring expressions, tense-modality selection, verb and noun lexical selection, hedge and cue word choice, and syntactic template selection, or any combination of these. To our knowledge, there is no other NLG engine that can dynamically generate all these types of entrainment in any utterance. We report an experiment testing all possible combinations of entrainment in a particular discourse context in order to test whether some types of entrainment are preferred, either because they make the utterance more natural, or because humans perceive the system as more friendly. Our experimental results suggest that human judgements of naturalness are distinct from friendliness: entraining on a user's hedges increase perceptions of friendliness while reducing naturalness, while entraining on user's referring expressions, syntactic template selection and tense/modal choices increase perceptions of both naturalness and friendliness.
CITATION STYLE
Hu, Z., Halberg, G., Jimenez, C. R., & Walker, M. A. (2016). Entrainment in Pedestrian Direction Giving: How Many Kinds of Entrainment? (pp. 151–164). https://doi.org/10.1007/978-3-319-21834-2_14
Mendeley helps you to discover research relevant for your work.