Given a representation of each speech act, as above, there is no need, they suggest, for special ‘conversational postulates’ (Gordon and Lakoff 1975) to explain the interpretation of indirect speech acts. That is, it is not necessary to state that a question about the hearer’s ability to perform an act may be intended as a request to perform that act, since such connections can be inferred on a more general basis. We need a theory of plans anyway, for non-linguistic inferencing, and that will do the work for us. Furthermore, as Allen and Perrault (1980) argue, it will do more than this: given the ability to recognise an interlocutor’s plan, a speaker may offer additional information if he believes this will help the interlocutor in the execution of his plan. So, for example, if a traveller asks at the information booth in a station: ‘When does the Montreal train leave?’, the clerk may reply: ‘3.15 at gate 7’ (Allen and Perrault 1980:441); given what the traveller has revealed about his plans, the clerk has inferred that there may be other obstacles to the traveller achieving his goals, such as not knowing the gate number. The clerk then provides this information without being asked. Clearly there is a need to ensure that too much information is not given under such conditions, since this will then impose a tedious task of selection on the requester. This is a problem for the knowledge-base of the system: in particular, items must not be represented at too specific a level.