Miscommunicating with Computers.
Matthew Purver, QMUL
Indicators of miscommunication and repair are important in human-human conversation – so those of us who work on computer dialogue systems should probably pay attention to them too. In human-computer dialogue interfaces, a lot of work has gone into detecting possible misunderstandings by the computer, and into strategies for avoiding or repairing these. Less has gone into detecting possible misunderstandings by humans – but as systems get more advanced, and as we apply computers to problems of analysing & understanding human-human conversation, this problem starts to surface. This talk will examine computational approaches to detecting and analysing repair phenomena, including clarification and self-repair, and explain why this problem is slightly different from standard natural language processing problems.