One way of accounting forthe well known problems that belief reports raise for compositionality consists in drawing a distinction between two different, albeit intertwined, speech acts they contain: the speech act of the ascriber, directed to an addressee and oriented towards some goal involving both the ascriber and the addressee, and the virtual speech act of the believer encoded in the that-clause. That is, in understanding belief reports we have to take into consideration not merely what is said in them (something supposedly derived from the truth conditional meanings of their component linguistic expressions), but also what the semiotic tradition (in particular, Jakobson) has called "the scene of enunciation". At least two scenes of enunciation are appealed to in a belief report, and the meanings of the words used within either of them cannot be composed with one another in a straightforward way.
Since "enunciation" involves an agent who appropriates language and is attributed responsibility for the speech act, the scene of enunciation is clearly connected, if not coincident, with the contextual instantiation of the deictic coordinates regarding speaker and addressee. Thus the perspective on belief reports here proposed assigns a determining role to context. Other perspectives too, such as the hidden indexical theory, have highlighted the role of context, but with different aims, presuppositions and consequences. According to the hidden indexical theory, contexts permit us to individuate the modes of presentation of the object of belief, which are supposed to be referred to by the belief report, and whose specification seems to be necessary in order to assign the belief report its truth conditions. On the perspective proposed here, contexts determine the interpretation and evaluation of the two speech acts involved in the belief report, specifying what assertion or kind of an assertion it is appropriate to make as if the believer were making, and in what words it is appropriate to encode it.
The discussion of examples will show how both the virtual speech act of the believer and the actual speech act of the ascriber are rooted in their respective contexts. The context of the former speech act is virtual and therefore is exhausted by the relevant assumptions of the ascriber. The context of the latter speech act is delimited by the goals of the conversational exchange in which it occurs and contains facts likely to be influential on the achievement of those goals. The usual distinction between de dicto and de re belief reports can be made to correspond to a range of differences in the communicative function of the report with respect to its addressee, to be analyzed in terms of the scenes of enunciation involved and their participant structure. The lack of linguistic markers delimiting the influence of each scene of enunciation and associated context raises some problems for the approach (as well as for other approaches invoking hidden quotation or semi-quotation). These problems can be tackled by a broader consideration of the ways in which interlocutors grasp contexts.