You are currently browsing the category archive for the 'Contextualism' category.
It’s a cliché that you always think of the right thing to say after the discussion is over. I think it’s got some truth to it though. Here is what I ought to have said recently when I was talking about differences between epistemic contextualism and contextualism in the semantics of personal taste predicates.
The semantic contextualist has a real problem with keeping content fixed. It seems reasonable to say that two speakers are disagreeing iff their utterances have the same content, plus or minus a negation operator. The details may get messy but the central idea seems to be sound. It may be that some versions of contextualism try and work with only suitably similar content, but most try and claim it is identical. This tends to lead to problems for them. Not least, how is their view then contextualist rather than relativist?
On the contrary the whole point of epistemic contextualism, at least as I see it, is to deny that ‘knows’ has the same content in different contexts. Again the details are difficult but the line of approach is clear.
From this it emerges that the links between semantic and epistemic contextualism are actually pretty tenuous, at least as far as a unified approach to both topics is concerned. What helps one project in a theory is likely to harm the other.
I’ve been re-reading John MacFarlane’s paper ‘Relativism and Disagreement’ this afternoon. It has a nice account of the key problem for contextualists when it comes to the semantics of predicates of personal taste: preserving what he calls the ‘phenomenology of disagreement’. I’m also a fan of the discussion at the end of how and why such terms could ever come to be part of our linguistic practices when they’re so strange. I think that this sort of issue is often ignored when people write about these things.
I did come across something that I found puzzling in his proposed solution to the problem. At one point he writes:
To be a relativist, then, is not to relativize propositional truth to “nonstandard parameters” like standards of taste, but to adopt a certain view about how the accuracy of certain acts or states is to be assessed.
It looks like there is a distinction between accuracy and truth playing a role here. Earlier in the paper such a distinction has indeed been introduced, but it is used to mark the distinction between utterances and propositions. The thought is that an utterance(-token) is an act and an act is not the kind of thing that can be true or false, merely accurate or inaccurate. I think that that’s a reasonable claim to make but I remember wondering when I first read it whether it was going to do any real work. It seems that it does, but, read like that, the work it’s doing looks rather dubious. If the distinction is in play only because utterances are acts then it seems obscure how that can be the bone of contention between relativists and contextualists.
I also wanted to to mention something that has been exercising me about the whole field recently. Looking at MacFarlane’s view I find it hard to discern much difference between it and the position of Peter Lasersohn. I have similar difficulty in finding any big picture differences between Lasersohn and the views of Jonathan Schaffer, at least as they were presented at his recent talk to Arché’s Contextualism and Relativism seminar. But MacFarlane is a relativist and Schaffer is a contextualist. If they disagree more about detail than they do about grand philosophy, then where is the importance of the distinction between these two terms? That’s assuming that I’m not lumping together views that should be kept apart due to a poor understanding of them. I hope I’m not.
I’m still puzzled by linguistic tests and related issues. I think that the following will do as a paradigm example. We want to test e to see if it plays the role of a gradable adjective in English. Suppose we run the test with two candidates, ‘tall’ and ‘knows’. We might then have:
‘Jane is taller than Jill.’
# ‘Jill knows more than Jane that snow is white.’
The ‘#’ is, I believe, commonly used to indicate the supposed unacceptability of an expression. One trouble i have is in being precise about just what defect is being marked. The problem is that I do see that the expression is defective, I suppose I want to say that it is in some sense not well-formed. Of course I can’t say that it’s because the sentence is false. Things in displayed contexts like the above aren’t even assertions in the usual way. (This is an important point in the later sections of the Cappelen and Lepore book that I mentioned in a earlier post.)
I suppose my main worry is in what sense such things are to count as tests, as a way to sort expressions once we have decided what categories are available for them to belong to but before we have decided to which categories they do belong. For example, take the second test described in Cappelen and Lepore’s book. This test relies on collecting expressions together in a single context. So, if we want to test ‘yesterday’ we construct certain sentences and observe that even though we can have a context in which:
‘Jane left yesterday’
and one in which:
‘Jill left yesterday.’
there need not be one in which:
‘Jane and Jill left yesterday’
Context-sensitive expressions block such collective descriptions, not in the sense that such a collective description is always false, but in that just because there are true utterances of the components there need not be any context in which there could be a true utterance of the collected expression.
Of course this is true, and plausibly all the basic set block collective descriptions and certain intuitively insensitive expressions (e.g. ‘Penguin’) do not. What bothers me about the whole practice of testing is that if I thought ‘Penguin’ were context-sensitive, wouldn’t I report that it blocked collective descriptions? And if I had no idea whether it was or not, wouldn’t I avoid committing myself? It looks like the test marks a sharp distinction between the sensitive and the insensitive but does it really count as a way of discriminating between them?
Maybe the story to be told is that the withholding or applying the ‘#’ on the basis of intuition is the real result of the test. It seems to me that everyone who uses tests, i.e. many philosophers of language, owe a story of what this consists in. I’ve not yet come across it.
I’ve been looking at some of the literature on linguistic tests for context-sensitivity. I’d like to write something here about some thoughts I’ve had about the notion in general, but I’m going to leave that for another time. Now I’m just going to mention something that struck me about one of the tests described in Herman Cappelen and Ernie Lepore’s Insensitive Semantics. It’s test 3 described on pp. 104-106. The thought is that we can test the context-sensitivity of expression e by thinking up a sentence S that includes e. Then we put S in the following schema and see what happens:
There can be false utterances of ‘S’ even though S.
An example of something to test would be ’she’, as in:
There can be false utterances of ‘She is French’ even though she is French.
And this is indeed so, people are frequently mistaken about who is French even if I am right that the woman I’m pointing at is. So far so good for the test as ’she’ is a common ground context-sensitive expression. What about things that we think shouldn’t be, like ‘penguin’? The problem I have is that I think, purely intuitively, that the following is true as well:
There can be false utterances of ‘Penguins are happy’ even though penguins are happy.
Penguins might not have been happy and even though they’re happy now they may well have been unhappy in the past and they may have dark times ahead. It seems to me that you could rule out enough things that these possibilities don’t seem relevant. In this case we might say that nobody in @ and now can utter it falsely and that seems right. But if you fix who you’re pointing at then we can’t get the result in the ‘She is French’ case. This is not the intuition I’m supposed to have. I would also like to report that I’m not inclined to say that ‘penguin’ ought to behave like ’she’. Of course it’s possible that I’m just missing the point, but it makes me wonder about the test a little.
If my chosen project is to give a contextualist alternative to relativism it looks like I’m going to need to be clear that they’re different things. Here is one way I might try and explain the distinction.
It’s clear that in order for it to be an interesting case the relevant tokenings of ‘The haggis is tasty’ and ‘The haggis is not tasty’, by John and Mary respectively, must be both in contexts that have one and the same salient haggis. Otherwise they’re simply not disagreeing. And they may well have to have the same contextually determined comparison class for tastiness. It seems that the contextualist holds that each tokening is able to come out as true in each one’s context. But they do not hold that there is some single context (of assessment) such that both come out as true. Whereas the relativist does allow for such a case.
I don’t know if that quite reports the positions taken in the literature by people like Michael Glanzberg but it seems to make a distinction clear. It also makes the relativist look rather like a dialethist, which I believe is a respectable but rather rare thing to be in the philosophy of logic. That might be a connection worth following up.
I’ve been thinking about the following puzzle. Suppose Jane wakes up in the cockpit of a crashed plane in a jungle clearing in Borneo. A blow to the head has given her complete amnesia. It’s 18.55 on 1/2/08. She says ‘I am here now’. A story I think is plausible would tell us that she has said that Jane is in a particular jungle clearing at the relevant time. But she doesn’t believe that she is. That might seem fair enough, but it looks like it clashes with the equally plausible claim that, whenever S says that P, she believes that P.
I have to admit that I’m not sure how to respond.
I think things went ok on Thursday. As always there were questions I couldn’t answer and objections I couldn’t meet. Here is a selection of them, and what I think might be a reasonable response.
Several people thought that it would be better to go with one of the other sorts of response to cases such as ‘Jones’ Note’. I think Keith and Ed were the main advocates of this line. The popular option seemed to be a kind of ambiguity for ‘I’ (and other indexicals). For example, perhaps ‘now’ can pick out either the time of utterance or the time at which the utterance is ‘consumed’. I have two connected worries about this. The first is that indexical expressions are now ambiguous, potentially between lots of senses. I think that this is more troubling than the thought that they can be used in lots of ways. Secondly, on this picture there is no obvious reason to think that the use of indexical expressions is governed by determinate rules at all. As was pointed out, I reject the most obvious rule for their use, but not the idea that there is a rule at all.
Various people asked for an example of a case where ‘I am here now’ turns out false. Here is one: Suppose I were to say ‘I am Aristotle. I am here now.’ The first sentence creates a content where the index for assessing the second will have Aristotle as its speaker parameter. If I said it in St Andrews on the 1st of March we would get those in the index as well. What would then be said by ‘I am here now’ would turn out to be the false claim that Aristotle is in St Andrews on the 1st of March. (Assume a tenseless ‘is’ if it helps.) That is false.
Ruth objected that I was assuming that the meaning of certain indexical containing expressions was equivalent to some expression not containing them. It’s certainly true that I would need to say a lot more about meaning to make my view work. Nonetheless I think that I’m actually quite happy to say that what the particular utterance of an indexical expression says is equivalent to an indexical-free expression.
Elia had two criticisms. Firstly he suggested that if I reject that ‘I am here now’ is an analytic truth then I must reject that other candidates such as ‘P iff P‘ are as well. I think I tried to say something complicated at the time. Looking back, I think it would have been better to point out that ‘P iff P‘ doesn’t have an indexical in, so for all I’ve said it is safe as an expression of an analytic truth. Admittedly it’s a schema, and if we put in a sentence with indexicals we may have a problem. The solution is to have a restriction such that if we have two occurrences of a sentence in such a schema they must get the same index. Then everything works out ok.
The second worry was about the use I wanted to put this in the contextualism vs. relativism debate. The thought was that if we had a language L for which there was no context-sensitive terms, but there was still pressure towards relativism, then the project would collapse. So far my only response would be to say that what I’m interested in is English and I don’t care about L. But I can see that that’s not going to be terribly satisfying.
I think that’s enough writing for a while. Thank-you for all the feedback so far and for any comments you’d like to leave now…
I’m giving a little talk at St Andrews’ M Litt seminar on Thursday. Here is what I’m going to say. I’m planning to write a more detailed post after the dust has settled to record all the interesting comments and criticisms I’ll be getting.
This final chapter aims to build on the points made earlier to provide a defence of neo-traditionalism against a common kind of counterexample. The aim is to explain how an extremely simple system along these lines can account for belief reports. Predelli does not argue for the assumptions that underlie his particular system, simply against the charge that such a system cannot handle belief reports.
The alleged problem is with the substitution of co-referring terms in belief reports (and so it’s almost as old as this kind of philosophy). The problem is as follows:
(1) Tom believes that Bush is the President.
(2) Tom believes that Dubya is the President.
We seem to be able to represent one and the same belief by both (1) and (2), but not in all cases. But to count these as having different truth-conditions suggests that we should be able to detect a significant semantic difference, some ‘unarticulated constituent’ perhaps.
Predelli appeals to his account of the real workings of the neo-traditionalist paradigm and argues that (1) and (2) will indeed be associated with identical T-distributions, but not what are intuitively thought of as truth-conditions.
This and ch. 4 suggest lots of very interesting questions which I’m now thinking about how to answer. All in all it was a worthwhile read.
This chapter introduces a lot of new ideas, at least they’re new to me. It’s also pretty long. I’m just going to try and summarise what I take to be the main point: a discussion of how T-distributions work and how that understanding can be used to rebut a contextualist argument.
The problem is to deal with the following kind of case. Pia has painted the (brown) leaves on the tree in her garden green. We can imagine the sentence:
(1) The leaves are green
being uttered in two distinct circumstances, to a photographer and to a botanist. It seems that in the former case what is uttered is true and in the latter it is false. But it isn’t any obviously indexical feature of the language that does it. So we have different truth-conditions associated with the same sentence and so a counterexample to the traditional theory’s claim to empirical adequacy. (I think that the definition of ‘empirical adequacy’ and ‘meeting intuitions’ is worth a serious discussion, but here isn’t the place or the time.)
Predelli’s strategy seems to be the following. The worry comes from taking the points at which the sentence is to be evaluated as ‘possible worlds’ construed as a kind of snapshot of a possible state of a world. In fact the theory will not give any such thing but will tell us whether (1) is true or false at every index. Indices can include features that fix what is to count as ‘green’ in various sorts of case. Which of these conventions is in play determines which indices are relevant. Only those in which it is fixed a certain way are relevant to an utterance in a context that so fixes the extension of ‘green’. Relevance is contextually determined, but this does not amount to conceding everything to the contextualist; there has been no alteration in the traditionalist claim that it is the sentence (or rather clause) that is associated with a T-distribution.
I’m not entirely sure what I think of this proposal yet, or even if I’ve got it right. At least there is now a pleasant symmetry established between the substitution of clause-index pairs for sentence-context pairs as inputs to the system and T-distributions for truth-conditions as outputs.
This chapter deals with an alternative to the type-orientated semantics of neo-traditionalism: token-reflexive semantics. On this view we have rules for determining the referent of each token of an indexical.
Predelli discusses a first possible objection to this: that it undergenerates logical truths i.e. that there are logical truths that it will deny to be logical truths. His response is that when the view is suitable augmented with the considerations about the difference between contexts and indices discussed in the previous chapter this worry disappears. It is reasonable to restrict the assessment of premises and conclusion of an argument to the same index. This will allow both types of view under discussion to get the right results.
The real problem with the token-reflexive view is that it overgenerates logical truths. This is because it restricts suitable indices to those in which there is a tokening. Kaplan argued that there should be a distinction between “the verities of meaning and the vagaries of action”; the token-reflexive view does not respect this.
As part of the discussion summarised above there are useful comments made about several puzzles, for example that ‘addressing puzzle’. The general solution is an appeal to the possibility of features of the conversation in raising and lowering appropriate standards of assessment. This is reminiscent of Lewis’ ‘accommodation’, as Predelli points out.
I have to say that some of this goes by rather quickly in this chapter. A lot of pragmatics seems now to be involved in the assessment, and it also seems that features not part of an austere theory of ’semantic profiles’ are part of the assessment of some sentences. In particular in the solution to the addressing puzzle; for example a feature of the sentence that is nothing to do with fixing the reference of indexicals can trigger the adoption of a different index. I need to think about this some more…
In this chapter we get a fuller account of the ‘index’ component of clause-index pairs, and just how this is different to a straightforward context. First Predelli outlines what he calls the ’simple-minded view’. On this view an index is an ordered n-tuple consisting of, for example, an agent, a location, a time and a world. These will be the speaker, her location, the time of her utterance and the world she is in. This is equivalent to what Kaplan calls a ‘proper index’, which becomes important later.
We then have a discussion of ‘answering machine’ type situations where it becomes clear that the simple-minded view will not work. We need to evaluate some utterances at indexes other than those it prescribes. The solution is to drop the simple-minded view and allow improper indexes. There are then sketches of how this makes Kaplan’s remarks about the logical truth of ‘I am here now’ and ‘∃x Exists x’ false. There is also a discussion of how this picture can account for truth in fiction without appeal to a fiction operator but, to a suitable fictional world filling the ‘world’ parameter of the index.
The point I’m keen to take home is the difference between indexes and contexts. It’s worth emphasising that in a proper index we have <agent, location, time, world> constrained by the requirement that the agent is the speaker and the utterance occurred at the location, time and world. This corresponds to what would intuitively be thought of as the context in which something is uttered. Allowing improper indexes means that a perfectly good index can be nonetheless not obviously related to anything that could reasonably be called a possible context for an utterance. I think that this is the point of the context / index distinction.
I thought I might start posting the notes I take. I’ve started reading Stefano Predelli’s book Contexts. Here is what I thought about chapter 1 on a first reading.
The key point seems to be a kind of terminological refinement. We do not talk of sentence-context pairs being assigned truth-conditions (which are those sentences meanings). Instead we talk about clause-index pairs being assigned T-distributions. These will be distributions of truth values at points, points being where things are assessed for truth.
Two things strike me as immediately interesting:
1) I don’t yet understand what a T-distribution is, why it’s an improvement on truth-conditions etc. My naïve worry would be that unless we want to specify the truth-value of φ point by point we are going to need to say in general which sorts of points will assign φ ‘True’. That looks to me like truth-conditions, unless I’m missing something.
2) The move from assessing sentences to assessing clauses is very interesting. It seems that this is intended to side-step debates about interpreting what is said by a particular sentence, or even sentence token. What we need is to be able to say for each interpretation (which is a clause) how it interacts with the theory. There is an immediate methodological payoff here: we can avoid a lot of the tedious debates about how a particular sentence is to be ‘read’. Probably not every such debate, though. It also seems that it has been argued against traditionalism that it cannot provide the disambiguation needed (Predelli provides some textual evidence from critics.) So it looks like it’s an important point for the defence of traditionalism as well. (Traditionalism is basically the view that Kaplan style semantics is the way to go; Predelli is a ‘neo-traditionalist’.)
