Recap

In the previous post, I discussed how we could have context-sensitive thoughts. Or more precisely, within the framework of a Language of Thought hypothesis, whether a mind/brain could deal with context-sensitive LoT sentences.

I (rightly) said that on the face of it, you expect a mind/brain with such a LoT to get confused when the relevant context changes. This is the confusion problem. That's ok when we do in fact observe that speakers get confused (e.g. when thinking about time in relativistic physics!). But what when they don't?

I (wrongly) argued there that there were only three ways a mind/brain could think with context-sensitive LoT sentences without getting confused:

  1. Eternalisation. Every context-sensitive LoT sentence is backed up with a context-insenstive LoT sentence in which indexical are replaced by proper-names-like things. (And real proper names: one per person, not like "Peter" or "Mary" who apply to thousands of people.)
  2. Inferential Isolation. The subject is kept frow drawing inferences from two LoT sentences issued in different contexts. This can be done by:
    1. Wiping out. Context-sensitive LoT sentences are simply wiped out when the subject changes context.
    2. Situation Boxes. Or maybe I should call that Inferential Islands. Context-sensitive LoT sentences issued from a single context are kept within a "situation box" where it can only inferentially interact with other sentences of that box or eternal sentences.

Now last week, at the excellent Epistemology, Context, Formalism conference, John Hawthorne has presented a third way in which a mind/brain could have context-sensitive thoughts without getting confused that I hadn't thought of: Change of Mind. It's a (simple and) great idea; here it is.

(NB, this is arguably a version of Wiping Out, so you could say that I wasn't wrong after all. But still the picture is different from what I said about Wiping Out.)

Change-of-mind

The confusion problem again

Suppose you introduce in the LoT the context-sensitive predicate IS-TALL, whose semantic value is given by the rule:

  • "X is TALL" tokened by a thinker S when C is the most salient (to S) type of thing X belongs to is true iff X is tall for a C.

Let Al be a 1.80m tall guy. Suppose S is at a dinner in which Al is the tallest guest, and assume that to S the most salient type of thing Al belongs to is the class of guests. Suppose S tokens:

Al is TALL

A true thought, in that context.

A few minutes later, the conversation switches to basketball - Al is a player at the local team. The most salient (to S) type of thing Al belongs to is now the class of local basketball players. Somebody remarks that Al is the smallest player of the team. Suppose S tokens:

Al is not TALL

A true thought, in the new context. But now our thinker has the two following sentences in her belief box:

Al is TALL. Al is not TALL.

And she is confused.

Change-of-mind

The Change-of-Mind idea is just this: when the thinkers gets in a new context, she revises her previous sentences containing the relevant context-sensitive word. When she gets to the second context, she tokens:

Al is not TALL.

And she revises her previous sentences that %%Al is not TALL%%, replacing it with %%Al is TALL%%. She basically thinks: "oh, I was wrong, Al is not tall after all". If she gets back to a context like the first one, she changes her mind again: "oh, I was wrong in fact, Al is tall indeed". And so on.

Such an agent is partly confused. She treats what is in effect a change of context as a difference in the world. This means in particular that she wrongly regards as contradicting thoughts that are not contradictory (because they belong to different contexts). But she does get around pretty well; in particular, (all things being equal) in no context she thinks false first-order thoughts (e.g. about Al being tall). The only false thoughts she is lead to have are thoughts about other thoughts.

For instance, in the second context she has the two following sentences that express truths given the context:

Al is not TALL. It is false that Al is TALL.

But she may have the following sentences that expresses something false:

"Al is TALL" tokened 30 mins ago was false.

Predictions and applications

The model predicts:

  • Retraction: when put in a new context the thinker assesses utterances made in different contexts as false. (Provided the relevant changes in contextual parameters and that the thinker knows the facts)
  • Relapse: the thinker has a seemingly inconsistent behaviour, constantly "relapsing" into saying things that she previously declared false.

As Hawthorne pointed out, this much closer to what we "observe" (proper data is merely coming up!) in the case of knowledge attributions. (I think Hawthorne mentioned vagueness as well; at least, that would seem to me a good story to apply there two.)

In the previous post I said that Récanati-style moderate relativism would be a nice way to represent what's going on in such cases, but that I would draw the line between "contextualist" and "relativist" content at whether the thinker deals with the relevant context-sensitive parameter, while Récanati would (I take it) ascribe relativist content to LoT sentences in the Situation Box case. Now where does the Change of Mind picture fits? At first sight I would say that it should be deal with in the relativist way. In no way does the thinker represent the relevant context-sensitivity, it just changes its mind as if the thoughts in question where context-insensitive.