Wednesday, October 18, 2006
Acquisition and Compositionality
I'm doing an independent study this semester with Dan Bonevac on semantic anti-realism, and this is an important issue which we spend a couple of weeks puzzling over. Let me introduce the debate as briefly as I know how before getting to the issue at hand.
A semantic realist holds the following two theses. Firstly, that understanding is knowledge of truth-conditions. Secondly, that the truth-conditions of some sentences can obtain or fail to obtain unknowably. But Michael Dummett suggested that realism so characterised gives a bad picture of what understanding a declarative sentence consists in, one that fails to take account of insights that can be distilled from the philosophy of the later Wittgenstein. Here's Alex Miller's summary of Dummett's complaints against the realist's claim that understanding could consist in knowledge of potentially verification-transcendent truth-conditions:
'...if our understanding of some sentences of D [D is some disputed aread of discourse, mathematics say] is constituted by grasp of potentially recognition-transcendent truth-conditions, how could we have acquired that understanding, given that our training in the use of sentences is a training to respond to situations which we are, necessarily, capable of recognising to obtain when they obtain? And if our understanding of some sentences of D is constituted by grasp of potentially recognition-transcendent truth conditions, how could we manifest that understanding in our use of those sentences, given that the situations to which we respond are, necessarily, situations which we are capable of recognising to obtain when they obtain?' (Miller 2002: 353)
These two challenges to realism have become known, for obvious reasons, as the acquisition challenge and the manfestation challenge. It's the acquisition challenge that I'll be concerned with here.
Last week Dan and I reread Dummett's classic 1959 paper 'Truth'. It's striking just how ahead of its time this paper is. For example, we were reflecting on the fact that the Investigations has only been in print for 6 years by the time this paper is published, and that Davidson's 'Truth and Meaning' is almost a decade away. It's also striking that there's pretty much no foreshadowing of the manifestation challenge in this paper, but there is a preliminary version of the acquisition challenge (see Truth and Other Enigmas: 17-8). The challenge gets developed throughout Dummett's extensive writings on these topics (which I'm still trying to acquaint myself with), but for me Dummett's own statements of the challenge have always been a little elusive. One of the clearer statements (or at least one of the statements I think I understand best) comes in Wright's introduction to Realism, Meaning & Truth:
'How are we supposed to be able to form any understanding of what it is for a particular statement to be true if the kind of state of affairs it would take to make it true is conceived, ex hypothesi, as something beyond our experience, something which we cannot confirm and which is insulated from any distinctive impact on our consciousness. Obviously such a conception cannot be bestowed ostensively. And the challenge is simply declined if the answer is offered, 'by description'. For it is of our ability to form an understanding of precisely such a description that an account is being demanded; there could be no better description of the relevant kind of state if affairs than the very statement in question.' (13)
Wright remarks: 'This is a pleasantly simple line of argument.'
Nonetheless, as Alex Millar has been keen to point out in a couple of his papers, both Wright and Hale have basically conceded that the acquisition challenge can be met, at least when it's presented in any strong form (Wright, for example, thinks there's still a challenge to the realist in the area, but that its force is parasitic on that of the manifestation challenge. See p16).
The passages where this concession is made are puzzling. Each makes a crucial concession to the realist in a single sentence, and without any reasons offered for why the anti-realist should concede so much. Here's Wright, from the intro again, and with the sentence in question italicized:
'...the most glaring example of our ability to work with concepts which transgress the limits of what we have actually experienced is provided by the understanding each of us has of no end of sentences in our language which we have never previously heard. The anti-realist can hardly deny that we have such an ability in general. Moreover his position will lose any credibility unless he can dissociate it from the old positivist dogma that untestable statements are meaningless. So he should grant that we are able to understand statements of the kind for which he finds the realist interpretation problematic; his quarrell is not with the supposition that we do have such understanding, but only with the realist account of it. But now the realist seems to have a very simply answer. Given that the understanding of statements in general is to be viewed as consisting in possession of a concept of their truth-conditions, acquiring a concept of an evidence-transcendent state of affairs is simply a matter of acquiring an understanding of a statement for which that state of affairs would constitute the truth-condition. And such an understanding is acquired, like the understanding of any previously unheard sentence in the language, by understanding the constituent words and the significance of their mode of combination.' (15-6)
So intially the challenge was to show how we could form a concept of what it would be for there to be sentences with verfication-transcendent truth-conditions, given that we learn a language through a process of training in use, and thus through learning how to respond to, to quote Miller again, 'situations which we are capable of recognising to obtain when they obtain'. The italicized phrase concedes that all it takes to meet this challenge is to show how we can, in a manner consistent with this view of language-acquisition, come to understand one of the sentences for which the anti-realist finds the realist account of understanding bankrupt. And now the answer is easy; compositionality does all the work.
But why on earth should we think that's all it takes to meet Dummett's challenge? Wright doesn't say.
Here's the relevant passage by Hale, again with my italics:
'...a very simple response is available: we come by a grasp of realist truth-conditions by coming to understand sentences having those truth-conditions, and we come to understand such sentences in just the way in which we come to understand the vast majority of sentences in our language, by understanding their words and semantically significant syntax.' ('Realism and its Oppositions' in the Blackwell Companion to the Philosophy of Language: 279)
Again, nothing at all is offered to suggest why the anti-realist is forced to concede the point in italics. Maybe it's obvious, but it certainly isn't obvious to me.
I need to finish reading Miller's 'What is the Acquisition Argument?', though the sections on this compositionality point I've read thus far sadly didn't shed much light on my particular point of confusion. I'm also guessing there might be some discussion in the literature on truth-value links concerning the past and other minds (for example, work by McDowell and Peacocke), where I think I similar move must be made to the one made in these italicized phrases. But I feel I need to do some serious further investigation.
A semantic realist holds the following two theses. Firstly, that understanding is knowledge of truth-conditions. Secondly, that the truth-conditions of some sentences can obtain or fail to obtain unknowably. But Michael Dummett suggested that realism so characterised gives a bad picture of what understanding a declarative sentence consists in, one that fails to take account of insights that can be distilled from the philosophy of the later Wittgenstein. Here's Alex Miller's summary of Dummett's complaints against the realist's claim that understanding could consist in knowledge of potentially verification-transcendent truth-conditions:
'...if our understanding of some sentences of D [D is some disputed aread of discourse, mathematics say] is constituted by grasp of potentially recognition-transcendent truth-conditions, how could we have acquired that understanding, given that our training in the use of sentences is a training to respond to situations which we are, necessarily, capable of recognising to obtain when they obtain? And if our understanding of some sentences of D is constituted by grasp of potentially recognition-transcendent truth conditions, how could we manifest that understanding in our use of those sentences, given that the situations to which we respond are, necessarily, situations which we are capable of recognising to obtain when they obtain?' (Miller 2002: 353)
These two challenges to realism have become known, for obvious reasons, as the acquisition challenge and the manfestation challenge. It's the acquisition challenge that I'll be concerned with here.
Last week Dan and I reread Dummett's classic 1959 paper 'Truth'. It's striking just how ahead of its time this paper is. For example, we were reflecting on the fact that the Investigations has only been in print for 6 years by the time this paper is published, and that Davidson's 'Truth and Meaning' is almost a decade away. It's also striking that there's pretty much no foreshadowing of the manifestation challenge in this paper, but there is a preliminary version of the acquisition challenge (see Truth and Other Enigmas: 17-8). The challenge gets developed throughout Dummett's extensive writings on these topics (which I'm still trying to acquaint myself with), but for me Dummett's own statements of the challenge have always been a little elusive. One of the clearer statements (or at least one of the statements I think I understand best) comes in Wright's introduction to Realism, Meaning & Truth:
'How are we supposed to be able to form any understanding of what it is for a particular statement to be true if the kind of state of affairs it would take to make it true is conceived, ex hypothesi, as something beyond our experience, something which we cannot confirm and which is insulated from any distinctive impact on our consciousness. Obviously such a conception cannot be bestowed ostensively. And the challenge is simply declined if the answer is offered, 'by description'. For it is of our ability to form an understanding of precisely such a description that an account is being demanded; there could be no better description of the relevant kind of state if affairs than the very statement in question.' (13)
Wright remarks: 'This is a pleasantly simple line of argument.'
Nonetheless, as Alex Millar has been keen to point out in a couple of his papers, both Wright and Hale have basically conceded that the acquisition challenge can be met, at least when it's presented in any strong form (Wright, for example, thinks there's still a challenge to the realist in the area, but that its force is parasitic on that of the manifestation challenge. See p16).
The passages where this concession is made are puzzling. Each makes a crucial concession to the realist in a single sentence, and without any reasons offered for why the anti-realist should concede so much. Here's Wright, from the intro again, and with the sentence in question italicized:
'...the most glaring example of our ability to work with concepts which transgress the limits of what we have actually experienced is provided by the understanding each of us has of no end of sentences in our language which we have never previously heard. The anti-realist can hardly deny that we have such an ability in general. Moreover his position will lose any credibility unless he can dissociate it from the old positivist dogma that untestable statements are meaningless. So he should grant that we are able to understand statements of the kind for which he finds the realist interpretation problematic; his quarrell is not with the supposition that we do have such understanding, but only with the realist account of it. But now the realist seems to have a very simply answer. Given that the understanding of statements in general is to be viewed as consisting in possession of a concept of their truth-conditions, acquiring a concept of an evidence-transcendent state of affairs is simply a matter of acquiring an understanding of a statement for which that state of affairs would constitute the truth-condition. And such an understanding is acquired, like the understanding of any previously unheard sentence in the language, by understanding the constituent words and the significance of their mode of combination.' (15-6)
So intially the challenge was to show how we could form a concept of what it would be for there to be sentences with verfication-transcendent truth-conditions, given that we learn a language through a process of training in use, and thus through learning how to respond to, to quote Miller again, 'situations which we are capable of recognising to obtain when they obtain'. The italicized phrase concedes that all it takes to meet this challenge is to show how we can, in a manner consistent with this view of language-acquisition, come to understand one of the sentences for which the anti-realist finds the realist account of understanding bankrupt. And now the answer is easy; compositionality does all the work.
But why on earth should we think that's all it takes to meet Dummett's challenge? Wright doesn't say.
Here's the relevant passage by Hale, again with my italics:
'...a very simple response is available: we come by a grasp of realist truth-conditions by coming to understand sentences having those truth-conditions, and we come to understand such sentences in just the way in which we come to understand the vast majority of sentences in our language, by understanding their words and semantically significant syntax.' ('Realism and its Oppositions' in the Blackwell Companion to the Philosophy of Language: 279)
Again, nothing at all is offered to suggest why the anti-realist is forced to concede the point in italics. Maybe it's obvious, but it certainly isn't obvious to me.
I need to finish reading Miller's 'What is the Acquisition Argument?', though the sections on this compositionality point I've read thus far sadly didn't shed much light on my particular point of confusion. I'm also guessing there might be some discussion in the literature on truth-value links concerning the past and other minds (for example, work by McDowell and Peacocke), where I think I similar move must be made to the one made in these italicized phrases. But I feel I need to do some serious further investigation.
Labels: Anti-Realism, Dummett, Epistemology, Philosophy of Language
# posted by Aidan @ 12:41 AM 
Comments:
<< Home
I'm not at all familiar with this literature, so let me make sure I've got the dialectic straight:
Dummett's challenge:
Let R = x understands a sentence S iff x knows S's truth-conditions. If R is true, then it is not possible to understand verification-transcendent sentences. But (realism allows that) it is possible to understand verification-transcendent sentences. So, R is false. Since R is part and parcel of realism, realism is false.
Wright & Hale's answer:
R does not entail that it is not possible to understand verification-transcendent sentences. For x understands a verification-transcendent sentence p iff x knows p's truth-conditions. And x knows p's truth-conditions iff x understands each of the words in p and their mode of combination. So, R -- and thus realism -- survives the challenge.
Good so far?
Assuming I've gotten the dialectic right, this is where you enter the picture. You find Wright & Hale's answer unsatisfactory. But I'm not sure I quite see your worry. I can think of (at least) 2 different potential sources of dissatisfaction:
1. You think that Wright & Hale fail to discharge a certain explanatory burden: namely, how x comes to know p's truth-conditions by merely understanding each of the words in p and their mode of combination. In the absence of this explanation, you find their answer incomplete.
2. You deny that x can understand p by understanding each word in p and their mode of combination. Thus, Wright & Hale have failed to answer the challenge
Which of these underlies your dissatisfaction? If it's (1), then I'd be interested to hear why you think (if you do) that understanding each word in p and their mode of combination isn't enough in the case of verification-transcendent sentences -- and why you think (if you do) it is in the case of non-verification-transcendent sentences. If it's (2), then I'd be interested to hear why you deny this. If it's neither, how would you put your worry?
Dummett's challenge:
Let R = x understands a sentence S iff x knows S's truth-conditions. If R is true, then it is not possible to understand verification-transcendent sentences. But (realism allows that) it is possible to understand verification-transcendent sentences. So, R is false. Since R is part and parcel of realism, realism is false.
Wright & Hale's answer:
R does not entail that it is not possible to understand verification-transcendent sentences. For x understands a verification-transcendent sentence p iff x knows p's truth-conditions. And x knows p's truth-conditions iff x understands each of the words in p and their mode of combination. So, R -- and thus realism -- survives the challenge.
Good so far?
Assuming I've gotten the dialectic right, this is where you enter the picture. You find Wright & Hale's answer unsatisfactory. But I'm not sure I quite see your worry. I can think of (at least) 2 different potential sources of dissatisfaction:
1. You think that Wright & Hale fail to discharge a certain explanatory burden: namely, how x comes to know p's truth-conditions by merely understanding each of the words in p and their mode of combination. In the absence of this explanation, you find their answer incomplete.
2. You deny that x can understand p by understanding each word in p and their mode of combination. Thus, Wright & Hale have failed to answer the challenge
Which of these underlies your dissatisfaction? If it's (1), then I'd be interested to hear why you think (if you do) that understanding each word in p and their mode of combination isn't enough in the case of verification-transcendent sentences -- and why you think (if you do) it is in the case of non-verification-transcendent sentences. If it's (2), then I'd be interested to hear why you deny this. If it's neither, how would you put your worry?
# posted by 
: 10/18/2006 11:36 PM
: 10/18/2006 11:36 PM
So I'm taking it the dialectic isn't quite like that. It's conceded by both sides that we understand the sentences under dispute. As Crispin says in the passage I quote, the anti-realist has to concede that to avoid verifiability being a criterion of meaning, as it was for the positivists (and the middle Wittgenstein in certain moods). That's not a distinctive committment of realism.
Here's another brief statement by Crispin of the acquisition challenge (this time from p86):
'the realist should explain how, when our training is necessarily restricted to confrontation with experienceable situations, we are supposed to be able to form a conception of what is is for an experience-transcendent situation to obtain'.
Now, I'm under no delusions that the challenge to the realist contained here is clear; there's an obvious need for a great deal of clarification and refinement. But that said, we can at least see that the challenge is to offer an account, consistent with these observations about the nature of language-acquisition, of how we could form a conception of what it would be for a verfication-transcendent truth-condition to obtain. Why does it suffice to meet *that* challenge to offer an account, consistent with these observations about the nature of language-acquisition, of how we could come to understand one of the statements in the disputed class? Why should the anti-realist concede that? But that's exactly what's conceded, without any reasons offered, in the italicized sentences from Hale and Wright.
So my worry was neither 1 nor 2. I don't dispute at all that the realist can offer an account of how we come to know the truth-conditions of some disputed sentence via our "training in use" by appeal to compositionality. What I don't see is how that suffices as an account of how we could come by a conception, through training in use, of what it is for such truth-conditions to obtain undetectably.
(I should note that Crispin in this book wants to adopt truth-conditional semantics, but render it invunerable to the challenges to realism by having a epistemically-constrained notion of truth. Since Crispin also agrees that we can understand the disputed sentences, presumably he's going to say with the realist we can grasp their truth-conditions. So again, this prompts the question; what's the connection between grasping the truth-conditions of one of these statements, and forming a conception of what it would be for such truth-conditions to obtain undetectably?)
Here's another brief statement by Crispin of the acquisition challenge (this time from p86):
'the realist should explain how, when our training is necessarily restricted to confrontation with experienceable situations, we are supposed to be able to form a conception of what is is for an experience-transcendent situation to obtain'.
Now, I'm under no delusions that the challenge to the realist contained here is clear; there's an obvious need for a great deal of clarification and refinement. But that said, we can at least see that the challenge is to offer an account, consistent with these observations about the nature of language-acquisition, of how we could form a conception of what it would be for a verfication-transcendent truth-condition to obtain. Why does it suffice to meet *that* challenge to offer an account, consistent with these observations about the nature of language-acquisition, of how we could come to understand one of the statements in the disputed class? Why should the anti-realist concede that? But that's exactly what's conceded, without any reasons offered, in the italicized sentences from Hale and Wright.
So my worry was neither 1 nor 2. I don't dispute at all that the realist can offer an account of how we come to know the truth-conditions of some disputed sentence via our "training in use" by appeal to compositionality. What I don't see is how that suffices as an account of how we could come by a conception, through training in use, of what it is for such truth-conditions to obtain undetectably.
(I should note that Crispin in this book wants to adopt truth-conditional semantics, but render it invunerable to the challenges to realism by having a epistemically-constrained notion of truth. Since Crispin also agrees that we can understand the disputed sentences, presumably he's going to say with the realist we can grasp their truth-conditions. So again, this prompts the question; what's the connection between grasping the truth-conditions of one of these statements, and forming a conception of what it would be for such truth-conditions to obtain undetectably?)
Hi Aidan,
I'm not quite sure I'm up to speed (always found this stuff really hard), but let me try something out, and see whether you agree.
Is the idea something like this. There's a sentence "a is F", say, about which the realist and the anti-realist differ. The realist thinks it has verification transcendent truth conditions, the anti-realist denies this.
Then there's the "training in use" thing. That's the idea that, whereever we end, up, we have to start with understandings that are acquired "use", e.g. the application of the sentence in circumstances where a certain state of affairs is recognized to obtain.
But then the thought is (no argument, just a challenge): how do we get verification-transcendent truth conditions from that basis?
Ok: compositionality response. This says that so long as we understand "a" and "F" (say, by recognizing a as the referent of "a" in circs where its apparent, and recognizing that "F" applies in circs where that's apparent) we'll be fine. And (says the realist), even though *some* of the circs where "F" applies may be recognizable, there's no need for all of them to be so. (So if we gained a grasp of "F" in application to b,c,d; and a grasp of "a" in circs where it had properties G, H, I, then there's no prima facie case that "a is F" must be something that we can recognize to obtain).
Lastly, a possible anti-realist rejoinder. Says she: this just pushes the question one stage back. How did we get a grasp on the predicate "F", such that its conditions of *application* to an entity can be VT in this way? Suppose the realist wants to say that "F" refers to property P, which sometimes unrecognizably obtains. But the way we come to understand "F" (by "training in use") at best is in circs where the objects *recognizably have the property P*. So how is that our understanding of "F" connects it to the (putative) VT property P, rather the non-VT property "recognizably having property P".
Does that sound right? There was some other stuff in what you said that I'm leaving out. Be interested to know if that's where you thought the distinctive interest was....
I'm not quite sure I'm up to speed (always found this stuff really hard), but let me try something out, and see whether you agree.
Is the idea something like this. There's a sentence "a is F", say, about which the realist and the anti-realist differ. The realist thinks it has verification transcendent truth conditions, the anti-realist denies this.
Then there's the "training in use" thing. That's the idea that, whereever we end, up, we have to start with understandings that are acquired "use", e.g. the application of the sentence in circumstances where a certain state of affairs is recognized to obtain.
But then the thought is (no argument, just a challenge): how do we get verification-transcendent truth conditions from that basis?
Ok: compositionality response. This says that so long as we understand "a" and "F" (say, by recognizing a as the referent of "a" in circs where its apparent, and recognizing that "F" applies in circs where that's apparent) we'll be fine. And (says the realist), even though *some* of the circs where "F" applies may be recognizable, there's no need for all of them to be so. (So if we gained a grasp of "F" in application to b,c,d; and a grasp of "a" in circs where it had properties G, H, I, then there's no prima facie case that "a is F" must be something that we can recognize to obtain).
Lastly, a possible anti-realist rejoinder. Says she: this just pushes the question one stage back. How did we get a grasp on the predicate "F", such that its conditions of *application* to an entity can be VT in this way? Suppose the realist wants to say that "F" refers to property P, which sometimes unrecognizably obtains. But the way we come to understand "F" (by "training in use") at best is in circs where the objects *recognizably have the property P*. So how is that our understanding of "F" connects it to the (putative) VT property P, rather the non-VT property "recognizably having property P".
Does that sound right? There was some other stuff in what you said that I'm leaving out. Be interested to know if that's where you thought the distinctive interest was....
Thanks for your comment Robbie. John persuaded me last night that I really hadn't been clear enough about the kind of point I was trying to make in my post, so I'll think about how to get clearer myself before I attempt to respond I think.
Post a Comment
<< Home
