conventional implicature

[harder at COCO.IHI.KU.DK]

I think Marta's question is one that has not received a terribly satisfying
answer in the standard literature because of the traditional identification
between 'meaning' and 'descriptive content'. The best approximation to
descriptive content that one can find in 'but', i.e.
'contrast-to-expectation' is in some sense defeasible, if examples like 'he
didn't succeed, but no-one expected him to' show anything.

I like to think that the issue can only be understood if functionalists
take their own position seriously enough to believe that coded meaning is
functional rather than descriptive. As I see it, 'but' has the function
(roughly) of alerting the addressee that what comes next may cancel some
'natural' inferences of the preceding utterance (something like 'hold your
horses for a second!'). In this, it contrasts with 'and', which (roughly)
signals that what follows collaborates with the preceding context. These
coded, conventional functions (=meanings) then collaborate with the context
in giving rise to variable, defeasible inferences.

In order for those functions to make sense there have to be ways in which
the two utterances are contrasting and collaborative, respectively. But
this is not claimed, but rather presupposed in using these signals
(otherwide they would make no sense), cf. 'he is Republican but honest' etc.
                                                Peter Harder, Copenhagen