This talk is on simple and naked infinitive perception reports, such as (1) and (2).

(1)     Jack saw Chris
(2)     Jack saw Chris Walk

These perception reports are used to show that models developed within cognitive psychology, as in Marr's Vision (1982), can be fruitfully applied to the semantics and pragmatics of natural language.

Barwise 1981 and Barwise and Perry 1983 develop a semantics for direct perception reports based on the assumption that normally the reports are veridical; if (2) is true, so is (3).

(3)     Chris walked.

However, they also noted that NI reports have non-veridical uses for which (2) need not imply (3).

In this talk, I will argue that the veridicality of NI reports belongs to pragmatics rather than semantics. To do so, I present a non-veridical semantics of the reports, which is not only partial but also retractable (e.g., we may correct reports in case of misperception). That the retractability of the reports often goes unnoticed is a pragmatic phenomenon: we normally expect that what we perceive will remain the case under growth of information. This pragmatic expectation can be given as a non-monotonic rule, which makes the perception reports veridical.

I will indicate that the techniques presented have a wider application within linguistics. Among other things, they can also be used to explain the retract of information in such dialogues as:

A:      'Did you see that hawk there?'
B:      'I saw something, but it was not a hawk.
        No, you're right, it is a hawk after all.'

[ The hand-out of this talk will be available on my home page: http://turing.fwi.uva.nl/~jvddoes/ .]