Learning through failure: speech accommodation in questions to Siri
Anna Konstantinova
(Universität Bonn)
Friday (April 21st), 13:15-13:40
Lipsius 147

Can we learn more about human communication via observing people talking to a machine, (e.g. a person checking the weather with speech interaction and recognition interface (Siri)? The answer is certainly “Yes”.

Despite numerous attempts to understand the mechanistic nature of dialogue (e.g. Pickering, M. J., & Garrod, S. 2004), it is still uncertain, how we manage to communicate. According to Common Knowledge theory, people share some symbolic conventions which make the communication possible. The “emergence-through-use” approach, on the contrary, suggests that we develop those symbolic conventions trough failing and learning during communication. Moreover, Barr (2004:943) ran simulations where the agents could converge even being “egocentric”, in the sense “indifferent to the identity of their conversational partners”.

In this study, we test how inexperienced in voice interaction with machines L1 and L2 speakers of English would cope with the task of communicating with “egocentric” Siri. The participants will be instructed to ask 20 questions to software. If Siri misunderstands the question, the participants should feel free to make changes in the phasing that we originally provided.

We expect that the participants will try to fine-tune their interaction with Siri by failing and trying to accommodate on the phonological, lexical and syntactic level. This would support the “emergence-through-use” approach confirming that common knowledge is a result of communication and not a prerequisite for it.

Barr, D. J. (2004). Establishing conventional communication systems: Is common knowledge necessary?. Cognitive Science, 28(6), 937-962.
Pickering, M. J., & Garrod, S. (2004). Toward a mechanistic psychology of dialogue. Behavioral and brain sciences, 27(02), 169-190.