Towards cloud-based speech interfaces for open-domain CogInfoCom systems

Abstract

A web browser accessing Wikipedia is a classic example of an open-domain CogInfoCom system, providing instant high-quality information about any topic, and thereby extending human cognitive capabilities. The WikiTalk system provides an alternative spoken dialogue interface to Wikipedia. By implementing WikiTalk on humanoid robots, the robots become open-domain CogInfoCom systems, capable of instantly talking about any topic. This capability for “open-domain talking” can be useful for different applications, such as serving as conversational companions. The WikiTalk system does not currently implement a parallel capability for “open-domain listening”. However, in view of recent advances in cloud-based speech processing, the paper proposes steps towards open-domain listening that combine anticipated progress in open vocabulary speech recognition with recent developments in named entity recognition, where Wikipedia is now used as a dynamically updated knowledge source instead of fixed gazetteer lists. The paper proposes that Wikipedia-based open-domain talking and open-domain listening will be combined in a new generation of cloud-based open-domain CogInfoCom systems.