The story of Pinocchio is the story of a man who creates a puppet with the ability to speak using human language, and to converse naturally – even to lie. Since the early days of the industrial revolution, people have imagined creating agents that can emulate human skills, including language production and conversation (Mayer, 1878; Wood, 2003). A modern example of this is Turing's test for artificial intelligence – a machine that is indistinguishable from a human being when evaluated through disembodied conversation (Turing, 1950).

This book concerns the intersection of two areas of computational research involved in the production of conversational agents – natural language generation and interactive systems.

Natural language generation

Natural language generation (NLG) systems are systems that produce human language artifacts (including speech, text, and language-rich multimedia presentations). NLG systems differ in the inputs they take, the types of output they produce and the degree to which they support interactivity. However, there are some common challenges across all types of NLG system. All end-to-end NLG systems perform the following three tasks: content selection, or determination of “what to say”; surface realization, or determination of “how to say it” (including assignment of content to media, selection of words, and arrangement of the content); and production, or the presentation/performance of the generated material (which in the case of an embodied agent may include production of body postures, gestures, and sign language).