As promised a while back, a draft of “CIA Leaks” is now available. It is a smallish paper I wrote with Thony Gillies laying out a number of problems for some recent relativistic approaches to the semantics of epistemic modals.
We have two larger papers in progress: (i) a tutorial/survey on work on epistemic modals, destined for the second volume of Oxford Studies in Epistemology, and (ii) a paper laying out and defending our own approach to epistemic modals, destined for a volume on epistemic modals edited by Andy Egan and Brian Weatherson. Drafts of those two papers should surface sometime this summer.
[Update: There is a post at Certain Doubts about one of the points we raise in our paper.]
“CIA theories instead say that the index is pretty much invisible to epistemic modals.” I wonder how charitable this attribution is. It clearly makes pretty bad predictions for embedded contexts. (Especially for ep modals under attitudes.) Wouldn’t it be more plausible to read these theorists as (1) enriching the index so as to include some parameter connectable to a body of information (e.g., an agent parameter), and then (2) letting context of assessment sometimes initialize index values?
May 20th, 2006, at 11:23 am #I’m not sure I see excatly what the proposal is here. But the basic point we are making at that point in the paper is that CIA theories take the modal base that the modals quantify over to be determined by what is known at the point of assessment. The time associated with a point of assessment can differ from the time associated with the index (that’s a good thing, from the CIA’s point of view, since that’s what opens the door to offering an explanation of diachronic disagreement). So whatever shifting to t_i that gets done by tense operators higher up doesn’t constrain the modal base. And that leads to bad predictions.
As we note in the paper, there are some more conservative strains of the CIA (Stephenson’s, and, we assume, Lasersohn’s preferred way of going for modals) that don’t have this feature since they opt for less robust points of assessment.
May 24th, 2006, at 1:27 pm #I think Seth’s suggestion is that relativists want the epistemic modal to somehow indicate that the index parameters (wi, ti, xi) (where xi is the agent parameter of the index) are initially set to the point-of-assessment values, rather than the utterance-context values as they are in the standard “CI” theory. This is how I always understood relativism (having learned relativism at Seth’s knee) prior to seeing Kai and Thony’s paper. According to this formulation, the modal base is given the following Kratzer-style entry: the set of worlds compatible with what xi knows at ti in wi. When MODAL P is not embedded under an index-shifting operator, this defaults to the set of worlds compatible with what xa knows at ta in wa, since nothing shifts the index parameters from the point-of-assessment values to which they are initially set. I think these two formulations of relativism will disagree only on cases in which the modal is embedded under an index-shifting operator (much the way, a semantics for “now” which assigns it the utterance time tc will differ from one which assigns it the evaluation time ti only on cases in which “now” is embedded under a tense operator).
Two things about the formulation Seth suggests: (i) It isn’t clear how epistemic modals manage to get the index parameters to be initially set to the point-of-assessment values. This is presumably a lexical property of epistemic modals; what would the lexical entries for the modals need to look like in order to accomplish this task? (ii) Concerning Kai and Thony’s example (21a): On this formulation of relativism, PAST (MIGHT (B)(P)) will be true at a c,i,a just in case there is a recentish t’ , a. When Kai and Thony write concerning Sophie’s utterance of (21a), “It is possible for Sophie to have said something true…”, I take it that they mean true at the point of assessment a = c = the utterance context (since on p. 11 they assume Sophie is the judge). But then this formulation of relativism will, I think, yield the right prediction for this example, since B will be the set of worlds compatible with what Sophie knows in wa (= wc) at t’ (t’ , a. When Kai and Thony write concerning Sophie’s utterance of (21a), “It is possible for Sophie to have said something true…”, I take it that they mean true at the point of assessment a = c = the utterance context (since on p. 11 they assume Sophie is the judge). But then this formulation of relativism will, I think, yield the right prediction for this example, since B will be the set of worlds compatible with what Sophie knows in wa (= wc) at t’ (t’ < tc).
May 26th, 2006, at 1:18 pm #I think that Dilip has put his finger exactly on where it hurts with his point (i):
There’s the rub. The obvious way to go (following Dilip’s analogy with now) is to make the point of assessment a separate parameter of evaluation and to write the lexical entry for the epistemic modals so as to access that parameter. That’s exactly what one does for indexicals like now. So, now one does this for “assessicals”. Hence, three parameters of evaluation, hence CIA. And now, like in the case of now, you can embed until you’re blue in the face but won’t be able to budge the point of evaluation for the item. That is the problem we point out.
So, if one is going to tell a different story about assessment-sensitivity, one should do so. We didn’t mean to say that there can be no assessment-based story that works. All we are saying is that what we have seen so far runs into trouble and since we have a different story to tell, we aren’t going to be the ones to come up with a fix.
May 26th, 2006, at 1:51 pm #Given that Kai von Fintel in one of his works correctly points to the fact that human languages are multimodal, the idea that some lexical items are equivalent to modal operators must be abandoned.
May 28th, 2006, at 9:33 am #A multimodal Semantics has certain advantages and one of them is its capacity to attach a role to the epistemic agent, which is in any case a modal parameter. In addition to the parameters and to the type of modal operator, i.e., (boulomaic,) universal or existential, the formation operation needs also needs to be defined.
The lexical items of a sentence and their particular features provide this information and together make up what in Propositional Multi-Modal Logic is called a (multi-)modal operator.
The other alternative is to think in terms of a Predicate meta-language, rather than a Propositional one (((multi-)modal) first order, second order. etc), where the epistemic agent may be written as an individual variable associated with a Predicate. But in such case you are just providing different ways to represent the same things, wherefore it is just a matter of choice.
PS: Sorry, I did not want to make all the characters bold-faced.
May 28th, 2006, at 9:38 am #The abstract story about assessment-sensitivity I sketched above is discussed (e.g.) by MacFarlane in his ‘Making Sense of Relative Truth’ (2005). As I understand it, the idea is supposed to be that context of assessment initializes certain index parameters just by stipulation, in the definition of truth for sentences. (Maybe you stipulate that all parameters are initialized by Ca; maybe you stipulate that just the “nonstandard” parameters are.)
This is more or less the way I interpret Egan (2005), and various sections of the EHW paper.
May 31st, 2006, at 1:01 am #Campbell Brown and I are wondering why the relativist shouldn’t say that the Boss should reason as follows. Suppose (this picks up a bit on Matt’s post no. 10 suggestion on the Certain Doubts page) the Boss, prior to Jack’s or Zack’s assertions, knows that each is reliable and a good evaluator of spy-related evidence. So prior to either’s assertion, the Boss knows that if Jack or Zack were to pronounce on spy-related matters in a way that is inconsistent with what he knows, he should throw out the inconsistent worlds.
So after Jack says “it must be either P or Q”, the Boss knows two things (1) that Jack has asserted that it must be either P or Q that is the turncoat, and (2) that Jack is very reliable. The Boss should then look at the worlds that are compatible with what he believes in which Jack makes that assertion. No world in which (1) and (2) are both true is also a world in which R is the turncoat. Hence, on the CIA view, Jack’s assertion is true when evaluated from the Boss’s new point of assessment.
So the CIA theorist should say that the Boss should throw the R-worlds out. The Boss should reason likewise as a result of Zack’s utterance, loosen his tie, sucks back a double martini, and order the hit on Q.
I think (perhaps only temporarily!) this also allows the relativist to take up Thony’s challenge in post no. 3, also on Certain Doubts. In, e.g. MacFarlane’s George/Sally case, Sally says “oh, I guess I was wrong” because she reasons in the same way that the Boss does above, throws out the worlds in which Joe is in Boston, and as a result of her new point of assessment, says of her earlier assertion that it was wrong.
Dowell
June 10th, 2006, at 4:12 pm #Thinking distributively:
(a) (P∨Q)&(Q∨R)
⇒
(P&R)∨(Q&Q)
Wherefrom:
(b) (P&R)∨(Q)
Why should the boss in question conclude that the copperhead is Q? Why shouldn’t he guess that there are two ratters, namely P and R? If he has no previous guesses or knowledge of who the traitor might be/is, he cannot discard P and R, nor can he limit the number of traitors to one.
June 12th, 2006, at 9:45 am #But again, why at the very first place would the three of them be picked as hypotheses?