Three different ways of introducing a relevant topic have been illustrated for now. It is of course essential to get the complete gamut of admissible topic initiations. The main prediction of the model is that, for the kind of conversations considered here, at least one of the three effects evoked above : improbability, paradox, (un)desirability, is required.
These constraints on topic introduction in spontaneous conversation is a predictive, and therefore falsifiable, model. The breakdowns observed in [ex_train p.6], [ex_lunch p.6], [ex_mercedes p.9] and [ex_antenna p.9] are predicted by our model : in each case the addressee had no means to perceive the event as problematic[3]. As we noticed, these breakdowns are predicted by no other pragmatic theory[4]. It is easy to get such a breakdown, just by mentioning any event which is not problematic, even if this event is in some way related to the current situation (as it was the case in [ex_train]).
The above principle is claimed to be also valid when the topic is introduced with a question. Notice that new topics are quite often introduced in an assertive way, unlike what some models seem to take as granted[5]. Nonetheless questions are frequent. We observed several of them, in [ex_Goffman p.6], [ex_mercedes p.9] or [ex_lunch p.6]. In such case the first speaker has a precise idea of at least one possible answer or of some of its characteristics, and this expected answer must lead to something problematic (either paradoxical, or improbable, or (un)desirable). Furthermore, the listeners most of the time need to be aware of this anticipation [cf. Fox 1987:375]. For sure we may ask questions for which we have no precise idea of the answer: "What is the capital of Burundi?". But one has again to imagine such a question in situation, as a way to introduce a new topic. What would be said after a reaction like "Why do you want to know ?" ? According to the model, the first speaker would then reveal a problematic situation. It may go from an undesirable lack of instantiation (e.g. if the speaker is doing crosswords) to a more structured problem (e.g. some inconsistency when reading a novel about Africa). But even the lack of instantiation has to be specifically problematic. In other words, the model excludes unmotivated questions.
We may represent the logical context that characterizes the problematic issue by the following relation :
[ p1 & p2 & . . . & pn ] => MOD
where MOD represents one of the modalities F, IMPR, DES and UND, and where all the pi are believed to be true. We will refer to this relation by the term problematic clause.
We have now to investigate further by studying the constraints that limit the range of relevant replies that may be uttered after a relevant (i.e. problematic) topic has been introduced.
[3] This term does not take the desirable mode into account, but it is a convenient way to summarize how a new topic should logically appear.
[4] Actually some models based on plan recognition [Grosz, Pollack & Sidner 1989], [Scholtens 1991], [Young, Moore & Pollack 1994] may be able to provoke a breakdown when no plan is recognized. Such models, used this way, would be much too restrictive, rejecting many relevant topics introduced in the improbable or the paradoxical modes, for these modes do not involve any plan.
[5] For instance in the IRF model of conversation structure, which is often mentioned ([Coulthard 1977:135]; [Stubbs 1983:136]; [McTear 1985:35]), the role of initiation is most often played by a question.