The limits of speech recognisers make it difficult to assume very free conversations. Luckily, for task-oriented conversation, we may assume that two people having conversation with each other and having a common goal, will collaborate to reach that goal as smoothly as possible. Grice [1975] formulates this in one of the most influential theories of human conversation:

This principle is then expressed in more detail via a number of maxims divided into Kantian categories or aspects. The principle is very popular among spoken dialogue designers because it follows that the well-behaved user will cooperate with the system, and even find it natural to do so. It eases the burden on the system and its recogniser if one can assume that the human user acts cooperatively.

The principle and its maxims are too vague to be immediately constructive for human-computer dialogue design. However, one may use the theory as a basis for methodologies for the qualitative design and evaluation of spoken dialogue, probably most constructively expressed in our CoDial tool described on these pages. It also finds its way to other references, e.g. [Harris 2005, pp 495-501].

A different exploitation of the cooperative principle is to investigate what can be inferred from dialogue contributions if the maxims are assumed to be satisfied. In standard linguistic literature, including [Grice 1975], this exploitation of the maxims is called conversational implicatures. For instance, with an example from Grice:

  1. A: I am out of Petrol.
  2. B: There is a garage round the corner.
(Gloss: B would be infringing the maxim "be relevant" unless he thinks, or thinks it possible, that the garage is open, and has petrol to sell; so he implicates that the garage is, or at least may be open, etc.)
[Grice 1975]

Clearly, this use of Grice's theory is interesting if you want to make inferential analysis of the conversation, e.g. in a dialogue manager, and many formalisation attempts exist.

See also the "Grice?" page.