Issue #19 -- December 2007. Pp. 108-121
Reference, knowledge, and scepticism about meaning
Copyright © by Elisabetta Lalumera and Sorites
Reference, Knowledge, and Scepticism about Meaning
by Elisabetta Lalumera
Scepticism about meaning is the claim that there is no fact of the matter as to whether a term means something or anything else, as there is no privileged semantic relation between a term and what we would call its `meaning' in ordinary parlance. In its most radical version scepticism about meaning leads to Kripke's sceptic's paradoxical conclusion that that there is no right or wrong in the use of words. Moderate versions involve the abandonment of the reference relation as the fundamental semantic notion, in favour of truth of sentences or coherence of web of beliefs. Resistance to scepticism about meaning is therefore mandatory for any semantic project that assigns a central role to the reference relation. This paper explores the possibility of resisting meaning scepticism by appealing to the idea that the nature of reference is to maximize knowledge. If the reference relation is a knowledge-maximizing relation, then some candidate referents are privileged among the others -- i.e., those referents we are in a position to know about -- and a positive reason against meaning scepticism is thus individuated. A knowledge-maximizing principle on the nature of reference was proposed by Williamson in a recent paper (Williamson 2004). According to Williamson, such a principle would count as a defeasible reason for thinking that most of our beliefs tend to be true. My paper reverses Williamson's dialectic, and argues that (we get a defeasible reason for thinking that) reference is knowledge-maximizing from the premise that most of our beliefs tend to be true. I will therefore defend such premise on different grounds than Williamson's, and precisely by revisiting a Naturalist argument he rejected, centred on the role of true beliefs in successful action. In the conclusion, an opposition to meaning-scepticism comes out as motivated by the knowledge-maximizing nature of reference, and backed by the plausibility of the claim that beliefs tend to be true.
Let scepticism about meaning be the thesis that
(MS) sentences of the form `t means O' have no truth value, that is, there is no fact of the matter as to whether t means O.
Here, t is a predicate or a general term (like `is green' or `green'), and O belongs to the domain of non-linguistic counterparts of predicates (whatever they are, namely, properties for metaphysical realists, or nominalist alternatives to properties). The meaning relation is taken in its broadest sense to be the semantic relation holding between a term and its non-linguistic counterpart (whatever that is). It might be called `the reference relation'-- my use of `meaning' signals the fact that MS is not scepticism about reference and acceptance of senses, but rather, most often, rejection of bothFoot note 1. The above statement of MS is general enough to include mental terms, i.e. concepts, as well as linguistic terms, i.e. words. Arguably, either MS is about both words and concepts, or it poses no serious threatFoot note 2. It is not general enough, however, to include proper names and other singular expressions together with predicates. All the traditional arguments for MS typically involve predicates or general terms, while only some of them generalize to singular termsFoot note 3. Since singular expression are generally taken to be not analogous to predicates and general terms in the semantic sense, arguments for MS about names cannot be obtained by analogy from traditional arguments. I will therefore concentrate the discussion on predicate and general terms, and the question whether the conclusion might be generalized to names will be left open.
MS allows for a radical version and a moderate one. The radical version is championed by Kripke's character of the Sceptic in Wittgenstein on Rules and Private Language (Kripke 1982). The Sceptic argues for MS and concludes that `meaning just vanishes into thin air', and there is no right or wrong in using words. The position is so radical that it is unstable, or paradoxical: if the Sceptic's claim is true, then it has no meaning, therefore it cannot be trueFoot note 4. The Sceptic's case for MS is simple, at least prima facie. Given all the facts about an individual speaker available at a certain time to an omniscient point of view, there are infinite alternative assignments of meaning to an arbitrary term t she's been using before that time, that would make `t means O' true. For example, given the evidence allowed, her term `green' could mean green, but also grue, where grue is the property of being green or being blue after the time considered. If there are infinite alternative assignments of meaning, and if they are all equivalent given the evidence allowed, there is no unique fact that would make `t means O' true, and the idea of terms having a meaning (one meaning) turns out to be an illusion.
Moderate versions of MS agree that the idea of terms having one meaning turns out to be an illusion, but still concede that complete sentences have truth conditions and truth values. The case for moderate MS is usually made by Quinean or Davidsonian arguments, that is, by arguments starting from the premise that meanings are the outputs of a theory of translation or interpretation, and directed at the conclusion that having meaning is a holistic propertyFoot note 5. According to this kind of arguments for MS, to say that a term has a certain meaning is to say that a subject's behavior can be systematized by a certain kind of theory, but there is no guarantee that there is only one way of systematizing a subject's behavior. In particular, many different alternative assignments of meaning to individual terms are compatible with the same evaluation of a sentence. In the famous Quinean example, if `here's a rabbit' is evaluated as true, `rabbit' could mean rabbit but also rabbit-fly, provided that compensating adjustments are made to the interpretation of the surrounding context. In a similar fashion, Davidson specifies that
If some theory of truth (or translation or interpretation) is satisfactory in the light of all relevant evidence (actual or potential) then any theory that is generated from the first theory by permutation will also be satisfactory in the light of all relative evidence (Davidson 1984, 230).
With the logical device of permutation, endless assignments of meaning are generated from the first one, and they are all equivalent with respect to the evidence. There is thus no reason to pick out one of them and call it `meaning', no fact of the matter about what individual words refer to, and no substantial sense in which it can be said that terms have a meaning. Radical or paradoxical MS is eschewed here because the locus of meaning is allowed to be the theory (or better, the set of theories) as a whole.
Scepticism about meaning cannot be classified among the varieties of epistemological scepticism. MS doesn't claim that we can't get to know the meanings of terms, it claims that there are no facts about meanings, no truth-grounds for meaning attributions, and no possibility of a supervenience or reduction basis for meaningsFoot note 6. Kripke is explicit about the metaphysical character of MS (Kripke 1982, 27). Quine, on the other hand, is explicit about his dismissal of epistemological scepticism in general, given his conviction that we should `surrender the epistemological burden to psychology' (Quine 1969, 75). It can be objected that both Quinean and Davidsonian theories individuate meanings from a third-person perspective, and they both share the view that a theory of meaning is, basically, a theory of how someone can get to know someone else's meanings. Thus -- the objection goes -- their core insight is basically epistemological. Such theories, however, share the view that there is nothing more to the nature of meaning than what can be known about meaningsFoot note 7, so that epistemology actually collapses into metaphysics.
Moving backwards in the reconstruction of first-order and second-order reasons, one may ask: Why opposing MS at all? Here, at least three answers are available. Firstly, MS seems to clash with common sense. It is common sense that there is one thing to which «dog» refers on a normal occasion of use. People tend to assume (at least to a certain extent) the determination of meaning when they speak, when they interpret other's speech, and when they learn a new word. So, at least, one needs to be suspicious about a revisionary claim such as MS. Secondly, there are theoretical reasons for opposing MS. Among them, the belief that meaning, formation and attribution of individual concepts have some ontological and explanatory priority vis-à-vis the total assemblage of thoughts at which a person arrives. In other words, it can be argued that concepts are prior to thoughts in that they are acquired first both ontogenetically and phylogenetically, and in that all thoughts are conceptually structured, but not all concept-applications, or acts of categorization, involve structured thoughts. A second theoretical reason is the belief that the meaning relation is a somewhat causal relation between a mental item and a world item, which requires that both relata exist. Philosophical positions sharing such beliefs are direct reference theories and informational semantics for concepts, realism about intentionality and semantic naturalism. For these positions to be tenable, MS must be false.
Let's put aside second-order reasons, and go back to the main concern, that is, strategies for rejecting MS. Generally, they take the form of constraints, i.e., additional conditions that components in `t means O' must satisfy for `t means O' to have a truth value. Anti-MS constraints divide into two broad categories: constraints on the meaning relation, and constraints on the domain of possible non-linguistic counterparts of terms. Constraints on the second category provide further specifications of what it takes to be a referent, so to rule out the greatest part of arbitrary assignments proposed by the sceptics about meaning. Constraints of this kind are not very popular. They draw on metaphysical assumptions and argue, for example, that genuine non-linguistic counterparts of predicates have to be natural properties, and not gerrymandered classesFoot note 8.
Constraints of the first category, on the meaning relation, specify further conditions that a term-object relation must satisfy in order to qualify as a meaning relation. One could say, for example, that genuine meaning relations are also functional relations, that is, they connect the term to what it has been created or preserved for, by the organism or system it belongs to. Again, this move would have the effect of ruling out sceptical alternatives as spurious -- typically, meaning relations obtained via permutation devices are not likely to correspond to natural functions. Unlike the second category of anti-MS constraints, this one is crowded: there are many other proposals about constraints on the meaning relation, and a complete survey (let alone an assessment) would far exceed the limit of this paper. My aim here is rather to try and evaluate only one of them, the knowledge-maximization constraint on meaning (KM from now on). Employed by Williamson in a discussion of the veridicality of so-called intuition, I think KM can play the role of a positive reason against MSFoot note 9. Let's see how.
The core idea of KM can be stated simply as follows: terms mean what we can get knowledge about. Therefore, terms do not mean what we can't get knowledge about. For example, in Putnam's Twin-Earth scenario, Oscar's term `water' means water on Earth, whereas it means t-water on Twin Earth. The explanation is that on Earth Oscar can get knowledge about water, but he can't get knowledge about t-water, in fact there is no t-water; and viceversa for Twin Earth. Note that the explanation bears on the impossibility of getting to know (much) about the referent, given the alternative assignments of meaning, not on the fact that the alternative assignment of meaning is not present in the environment, as traditional externalism would have it. Non-e. Since, however, there are far more putative referents in the environment than we can know about, the emphasis on knowledge allows for a more fine-grained discriminationFoot note 10.
Also, it is important to distinguish between the idea that terms mean what we can get knowledge about, from the idea that terms mean what they mean through the mediation of our knowledge of their referentsFoot note 11. In the first case, the term-meaning relation can be direct, in the second it is not. The second idea can embodied, for example, by a descriptive theory. On a descriptive theory, `water' means whatever substance satisfies the conditions expressed by the theory of water we possess. The meaning is fixed by the theory. Instead, the first idea can be embodied by a direct-reference theory. On such a theory, `water' refers to that substance via causal chain. Only, the relevant causal chain (among many possible ones) is individuated as the knowledge-maximizing chain. On the quite neutral assumption that meaning relations (or reference relations) can be either causal or not causal, here's Williamson statement of the view:
Roughly: a causal connection to an object is a channel for reference to it if and only if it is a channel for the acquisition of knowledge about the object. Often, a causal connection is a channel for both. Equally, a non-causal connection to an object is a channel for reference to if and only if it is a channel for the acquisition of knowledge about the object. Sometimes, a non-causal connection is a channel for both. (Williamson 2005, 140-141).
For the present purposes, I can leave the notion of knowledge quite underspecified. Very generally, knowledge as a subject-object relation depends on facts about the subject as well as on facts about the object. Among the latter, the fact that the object exist. Among the former, facts about the subject's cognitive system, her inferential practices, her rational behaviour may be enlisted. The balance between the two classes of facts is obviously different for internalist accounts of knowledge and for externalist ones. Such a difference, however, has no bearing at this stage of the discussion. Whatever knowledge is, it may be used as a constraint on the meaning-relation.
More precisely, KM plays the role of a constraint on the nature of the meaning-relation. Its role is metaphysical. Though it employs the epistemic notion of knowledge, it is to be considered only secondarily as an epistemological principle about the proper methodology for the ascription of meaning, that is, only insofar as the nature of meaning determines how we should ascribe itFoot note 12. Thus, given our statement of MS as a metaphysical thesis, knowledge-maximization can qualify as an anti-MS constraint. Precisely, it is a constraint of the second category illustrated above, which specify what it takes to be a meaning-relation. According to knowledge-maximization, candidate meaning-relations will also have to be knowledge-relations.
But what is KM supposed to do in practice against a meaning sceptic? The sceptic would claim that all the possible assignments of meaning to t are equivalent, and therefore `t means O' is never true for a specific O. KM then allows one to reply that not all the possible assignments of meaning are equivalent, because some of them presents us with meanings we can know more about, and ideally one of them with meanings we know most about. This would be apparent in the case of multiple assignments generated by permutation devices. Take a theory of translation that individuates `Here comes a rabbit' as a true sentence, and assigns rabbit as the meaning of `rabbit', and then a second theory obtained by permutation from the same, which assigns the former Japanese prime minister Yasuhiro Nakasone as the meaning of `rabbit'. Given KM, we are in a position to say that `Here comes a rabbit' means that here comes a rabbit, and not that the former Japanese prime minister is coming along, and `rabbit' means rabbit, if at present the speaker is not in the position of getting knowledge about Yasuhiro Nakasone via perception. KM squares with the a posteriori hypothesis that, given a specific context of use, some aspects and objects in the world are in the cognitive background of speakers and hearers, while others are in the cognitive foreground, and this is a relevant fact about content individuation.
It may might be objected that, even granting that KM narrows down the wild explosion of sceptical meaning assignments, it will not suffice for narrowing it down to one. That is, KM may help us excluding Yasuhiro Nakasone but not rabbit-flies when «rabbit» is involved, given that by hypothesis rabbit-flies can be known about when rabbits are. Not all putative meaning-relations are knowledge-relations, but there can be more than one for each term. Since, however, knowledge admits gradation whereas exisence does not, KM as a constraint on meaning allows one to consider putative referents on a scale, that is, it allows to give weights to alternative choices. For example, presumably, rabbits are better referents than rabbit flies, given that they are more salient for a cognitive system shaped as oursFoot note 13. If putative referents are no more infinite in number, and if they can be assigned different weights, then MS is wrong. Thus, KM is a reason for rejecting MS, though a defeasible one, for the assignment of different weights would surely involve genuinely empirical facts and considerations, and not purely a priori ones.
If KM can count as a reason against MS, what we need now is a reason for accepting KM. Why KM should be taken as a plausible principle about the nature of meaning? How can the claim that meaning is knowledge-maximizing be supported? In Williamson (2004) KM is the conclusion of an inference to the best explanation. What has to be explained is the fact that beliefs tend to be true or, in other words, Williamson is looking for positive reason for believing that `beliefs tend to be true'. Let's call that thesis `Veridicality' for short. Williamson assesses some candidate explanation of Veridicality, and discharges them as not adequate. Among them, a naturalist argument, Davidson's principle of charity, and its spin-off the principle of humanity. Therefore, he concludes, `we need to make a new start', which is, the principle of knowledge-maximization (Williamson 2004, 131-139). Thus, in Williamson's paper, KM comes out in absence of other candidates, as the general explanatory principle from which Veridicality is supposed to follow.
My point here is that KM can have the role of a positive reason for thinking that terms have definite meanings -- here, KM comes out again as a general explanatory principle from which that thesis is supposed to follow. Nevertheless, I think that some independent reason for KM can be provided as well, beside establishing it as a conclusion of an inference to the best explanation. Precisely, I think that Veridicality supports KM, and provides a good (though defeasible) reason for accepting it. I am hereby reversing Williamson's dialectic -- or in other words, I'm exploiting the link he individuated between KM and Veridicality, but contrariwise. To repeat, according to Williamson KM is a reason for accepting Veridicality, and my point is that that Veridicality can constitute a positive reason for accepting KM. And KM, on its part, is going to have the role of an anti-sceptic constraint. As I'm not going to explain Veridicality via KM again (as Williamson does instead), my explanation will not be circular. I need, however, to argue for Veridicality on different grounds. Before turning to that, however, let's see what the truth of most of our beliefs has to do with the idea that the nature of meaning is to maximize knowledge.
To accept that Veridicality holds not by accident, but on a regular and systematic basis, is to accept the idea that there is something about our way of forming beliefs --some general principle about the functioning of our belief-forming mechanisms -- that makes it possible. It is also plausible that the general principles that make Veridicality possible are more than one, namely, a whole set including principles about our perceptual systems (optimality conditions for perception), our inferential processes (truth-preserving inferential rules), and about language and thought (knowledge-maximizing meaning assignments)Foot note 14. The intuitive idea is that if our representational system didn't work the way they do, our inferential rules were not the ones we take as valid, and the meaning relation was not the one that holds, Veridicality would non be in placeFoot note 15. Though none of the factors accounts for Veridicality alone, if one is missing it is likely that Veridicality fails. On the other hand, that Veridicality holds signals that it is likely that all the general rules of functioning in the set are operating properly. This is how Veridicality can act, indirectly, as a defeasible reason in favour of KM. In other words, on the hypothesis that KM and the other general rules make Veridicality possible, that Veridicality is actual may count as a reason to think that KM and the other general rules of functioning are in place. Schematically,
Veridicality ├ KM and other principlesFoot note 16
KM and other principles
Surely Veridicality comes out merely as a defeasible reason for KM alone, for KM is just one of the factors that jointly suffice for Veridicality. Nonetheless, if Veridicality can be accounted for on independent grounds, KM would gain in plausibility.
So how is it that Veridicality holds -- why is it that most of our beliefs tend to be true? Not my beliefs in particular, but our beliefs collectively -- if it was possible to collect them in a box and inspect them one by one, the true ones would outnumber the false one. Maybe not in any given instant of time -- there surely can be times of massive error. But with big numbers, just like the probability that a tossed coin lands head equals the probability that it lands tails, the frequency of true beliefs is higher than the frequency of false ones. You can either believe that it is so or just suppose that it is so -- in either case, why would that be?
This may seem to be a prohibitively big question. Note, however, that it is not as prohibitive as the task of providing a confutation to the traditional epistemological sceptic, who claims that beliefs, say, about the external world are false or unwarranted. The two tasks can be set quite apart. To individuate a positive reason for Veridicality would not be sufficient for traditional anti-sceptic aims, as the sceptic would probably run his sceptical argument on any claim you could present as a positive reason for Veridicality. Here my aim is more modest. What I would like to find is a positive reason for Veridicality that would support the second premise of the argument schema above. It need not be sceptic-proof.
As Williamson (2004) admits, it is very natural to support Veridicality with a `Naturalist' argument that appeals to the idea that true beliefs are more useful than false ones:
(N) True beliefs tend to cause one to get what one wants in a way in which false beliefs do not ... on the whole, truth is more conducive than falsity to survival (Williamson 2004, 131).
Taking (N) as a premise, and developing the evolutionary hint suggested by the word `survival', the Naturalist argument for Veridicality may go on as follows. According to (a simplified version of) the Evolutionary Theory, what is conducive to survival is preserved and enhanced. Therefore, Veridicality is preserved and enhanced. Here, reliable belief-forming mechanisms that account for Veridicality are identified with a phenotypic (heritable) trait, and the fact that our population came to have organisms with that phenotypic trait is explained by saying that our ancestors were selected for possessing that trait.
As it is well-known, evolutionary arguments invite objections when employed in connection with high-order cognitive capacities, such as conceptual thought, or logical reasoning. For example, it is often objected that evolutionary explanations could be plausible only for very simple beliefs about food, mating, etc., and generally only for limited subject matters. Such objections have nonetheless received convincing repliesFoot note 17. I do not think, however, that the Naturalist argument for Veridicality needs to be an evolutionary argument. The evolutionary reading is not crucial. In fact, the effect of true believing versus false believing can be seen within a lifespan perspective. If I am rational, I would preserve and prefer (as far as I can) those belief-forming practices that brought me to success in the past, given my personal experience. Thus, if evolution presupposes experience of the species, one may as well take experience of the individual as primary. This suggestion amounts to replacing `survival' with `success' in premise (N) above, thereby obtaining:
(N') True beliefs tend to cause one to get what one wants in a way in which false beliefs do not...on the whole, truth is more conducive than falsity to successFoot note 18.
The non-evolutionary version of the Naturalist argument would then have that given an agent's rationality, that agent is likely to preserve and enhance what is conducive to success; therefore, reliable belief-forming mechanisms that account for Veridicality are preserved and enhanced. This version of the Naturalist argument involves the concept of rationality and generalizes over rational agents. It does not mention the Evolutionary theory. It is weaker than the Evolutionary version, though, for individual agents are not in a position to control all their belief-forming mechanisms, but only some of them. We can learn from experience how much to trust testimony, intuition, divination, and other belief-forming practices, and consequently decide to enhance some of them and discard others. We are not, however, in the position of discarding our own perceptual belief-forming mechanism within a lifespan (at best, we can put on glasses). This version, on the other hand, is free from the traditional objections to evolutionary explanations of high-order cognitive capacities.
What I have said so far would suffice for suggesting that what is crucial to the Naturalist argument for Veridicality is not the particular theory (evolution or individual rationality) that fills in premise (2). What is crucial is the link between Veridicality and success or survival, that is, the link between true beliefs and what is good for one. It can be expressed in a general principle about rational action. Here is Williamson's version:
(RW) If an agent desires that p, and believes that if it does A then p, then cæteris paribus it acts so that it believes it does A.
Employing (RW), the role of true beliefs in successful action becomes apparent, as the following derivation shows:
S desires that p
S acts so that it believes it does A
The desired good, namely p, is arrived at only provided that S's beliefs (that if S does A then p, and that S does A) are true. This is how true beliefs explain successful action. But according to Williamson (2005), this is also how the Naturalist argument, both in the evolutionary and in the non-evolutionary version, falls prey of a fatal objection. In his words:
Unfortunately, such a derivation explains much less than it appears to. For one can show in the same way for infinitely many deviant properties true* and good* that the combination of true* beliefs and desires for what is good* for one yields (cæteris paribus) what is good (not just good*) for one (Williamson 2005, 133).
According to him, it is possible to prove that what is good for an agent (the desired p above) is also yielded by the combination of true* beliefs and desires for what is good*, where true* and good* are so defined:
(def1) that p is true* iff that ^p is true
(def2) that p is good* iff that ^p is good
where ^p is an arbitrary structure-preserving mapping on propositions (e.g. it maps `You are reading' with `Snow is white'). If Williamson is right, then the Naturalist argument for Veridicality fails, for the supposed link between true beliefs and success (or survival) comes out to be spurious. Let's see his proof in details.
Assumptions include the already mentioned (def1) and (def2), as well as his principle about rationality:
(RW) If an agent desires that p, and believes that if it does A then p, then cæteris paribus it acts so that it believes it does A.
The proof goes as follows.
|(1)||S desires that p, believes that if it does A then p, and acts so that it believes it does A||(RW)|
|(2)||that if it does A then p is true*||(assumption)|
|(3)||that it does A is true*||(assumption)|
|(4)||^(if it does A then p) is true||(2, def1)|
|(5)||if ^(it does A) then ^p||(4, commutation)|
|(6)||^ (it does A)||(3, def1)|
|(7)||that ^p is true||(5, 6, m. ponens)|
|(8)||that p is good* for S||(assumption)|
|(9)||that ^p is good for S||(8, def2)|
Conclusion 9 would show that what is good for an agent is yielded by a combination of arbitrary propositions such as good* and true* ones, thereby leaving no role for genuine true beliefs in the rational behaviour of agents. Permutation spoils the intuitive conceptual link between Veridicality and what is good. There is no reason to claim that success or survival have to do with the truth of our beliefs more than with any other arbitrary property of them. Williamson's proof would then amount to a fatal objection to the Naturalist argument. In order to employ the Naturalist argument to support Veridicality, therefore, it needs to be rejected.
How to block the undesired conclusion that true is no better than true* in goal-directed behaviour? I think the place to look at is Williamson's principle RW stated above. It contains a reflective specification -- the agent acts so that it believes it does A -- which plays an important role in the proof or the undesired conclusion, allowing for the insertion of the propositions obtained by permutation. Why this reflective specification? Why not a simpler principle, like the following?
(R) If an agent desires that p, and believes that if it does A then p, then cæteris paribus it does A.
This is Williamson's own explanation:
Williamson points to a difficult case for a simple principle like (R). The case is one where S's non-reflective belief (that if S does A then p) is true, whereas her reflective belief (that S does A) is false. What happens to principle (R) when, say, S goes north while believing she is going south, on the hypothesis that the desired good is in the northern direction? Williamson suggests that the hypothesis be dropped in favour of a new one, according to which the most desired good is southwards. We would then have that S goes north while believing that she is going south, and south is where she desires to go. In such a case, S would simply fail to accomplish action A -- the action that constitutes the means to reach the good p. And it would fail to reach p (which is, on the proposed reading, southwards). On this reading, Williamson argues, (R) still holds. But S's performance in the case considered is unsuccessful; we need thereby to isolate cases like that if we want to preserve the link between true beliefs and success. The only way, according to Williamson, is to assume within the cæteris paribus clause that reflective beliefs (beliefs about what one is doing) are mostly true. But then (R) would beg the question in an argument for Veridicality.
Williamson's reading of the proposed case, however, is not mandatory. One may follow this alternative line of reasoning. In a case where S goes north while believing she is going south, on the hypothesis that the desired good is in the northern direction, (R) is simply fulfilled. Given that the non-reflective belief is true, S reaches the desired good. The case would only show that reflective beliefs play no role in an explanation of successful action. We can just overlook them. No matter what we think we're doing, if we act on our desires and choose the appropriate means, we get through. In fact, this is what Freudian examples show -- they show us the irrelevance of (non-deep) reflective beliefs about one's actions (and desires, etc.) in the explanation of rational behaviour. Consider this other case. S believes she is posting an invitation for T. S is in fact putting the invitation for T in a dustbin. The explanation is that S desires T not to be present at her party, and S believes that if she does not send an invitation to T, T will not be present. S's reflective belief is false, but her non-reflective belief is true, and conducive to success.
This line of reasoning invites an objection. Even granted that there are many cases where we act without knowing what we do, most of the time we're reliable about ourselves. Maybe some desires remain hidden to introspection, but on many others the Ego has nothing to say, nor to censor. So, even the most enthusiastic Freudian would allow that lapsus linguae and missed acts are just episodes. To conclude that reflective beliefs about what one is doing are just irrelevant to action is somehow misleading, for it might suggest that agents are always passively driven by their desires, with no kind of cognitive control.
On the other hand, the Freudian case brings out an interesting point. The kind of cognitive control of agent on her actions need not be an explicit reflective belief of the form `I am doing A now', let alone of the form `S does A', where S is oneself. Maybe we are reliable about such beliefs when we entertain them, but it may be also true that often we act without being in the position of entertaining them, and not only for Freudian reasons. Sometimes actions are performed very quickly, on top of perceptual inputs -- for example, when screening up one's eyes with a hand as a light flashesFoot note 19. Some other time, an agent may lack the proper concepts that would compose the relevant belief. The more one conceives of concepts as word-like, the more one can find examples where an agent possesses the resources to behave rationally, but not the language to build up a belief about her actions (think of pre-linguistic children, of chimpanzees, etc.). For all these reasons, a conscious reflective belief seems to be too heavy a requirement for rational action.
It follows that Williamson's principle (RW), with its reflective specification, can be dropped. Dropping (RW), the proof of the conclusion that true* and good* yield what is good does not even start. The Naturalist argument for Veridicality is saved.
A problem, however, remains to be fixed. If (RW) is too demanding, Williamson's case of going north while believing one is going south shows that (R) is too loose. It leaves out the intuitions that rational agents do not act merely randomly or passively, but rather they have some kind of cognitive control over what they do. My proposal is that the concept of intention would suffice for expressing the required kind of cognitive control. A principle would then have this form:
(RI) If an agent desires that p, and believes that if it does A then p, then cæteris paribus it intends to do A, and cæteris paribus it does AFoot note 20.
(RI) establishes a link between success, true beliefs and fulfilled intentions. According to (RI), the agent's success depends on her having a true belief, the non-reflective belief about A being the appropriate means for the desired good. This preserves the link between true beliefs and success or survival. But the agent's success also depends on her forming the corresponding intention of doing A, and of that intention being fulfilled. The first cæteris paribus clause in (RI) specifies that rational agents, most of the time, intend to do what they believe is useful. To assume that in the cæteris paribus clause is not question-begging, as we are employing (RI) in an explanation of Veridicality, and Veridicality says nothing about the formation of intentions. Finally, according to (RI), the agent's success depends on her intention being fulfilled. The second cæteris paribus clause in (RI) is about the fulfilment of intentions. Intentions may not be fulfilled for a variety of reasons. Sometimes the world does not cooperate, and agents get frustrated. Again, in an explanation of Veridicality, in order to show that true beliefs are conducive to success, we may assume within the cæteris paribus clause that intentions of acting be fulfilled. This would not amount to begging the question, because Veridicality says nothing about the fulfilment of intentions. As for the case of going north while believing one is going south, it can be redescribed. If the agent goes intends to go north, then principle (RI) holds, and the falsity of the second-order belief is irrelevant. The agent may be deluded about the fulfilment of her own intention, like when one has the impression of getting lost when in fact one is going in the right direction. If, on the contrary, the agent intends to go south, and goes north, this does not constitute a counterexample to (RI), for (RI) assumes that most of the time rational agents do form the appropriate intention of acting -- (RI) absorbs the shock.
It might be objected that intentions, just like beliefs, are contentful states, so a permutation argument could be run on my new principle just like on Williamson's. An explanation of why an agent does what's good may be that the agent has true* beliefs and intends to do what's good*. My reply would be that the permutation argument in Williamson's version, relied on Veridicality, while the connection between intending to do something and doing something does not. Therefore, a new permutation argument would have to be thought up for the objection to be fatalFoot note 21.
So far, then, equipped with an intuitive concept of intention we may preserve the idea behind Williamson's objection, as well as avoid the reflective specification that brings to principle (RW).
The above defence of the Naturalist argument against Williamson's object makes a case for Veridicality, that is, the claim that beliefs tend to be true. That beliefs tend to be true, in its turn, signals that certain general principles about our though and language are in place; among them, the knowledge-maximizing principle about meaning. Veridicality is thus a defeasible reason in support of the knowledge-maximizing principle -- that is what I argued in section 4. Knowledge-maximization, in its turn, may give us a good though defeasible reason against meaning scepticism, the claim that there is no fact of the matter as to whether words and concepts mean. The required fact, according to my hypothesis, is that some putative meanings, and not others, are part of a channel of knowledge that ends up in our minds -- to put it simply, terms mean what we can get knowledge about.
A further question to engage in would concern the nature of the knowledge relation which constitutes the meaning relation. According to Williamson, KM makes essential use of an intentional vocabulary, and therefore cannot be employed within naturalist reductive accounts of meaning. Williamson's remark restricts the scope of KM as an anti-sceptic reason. KM would come out as available to non-naturalists only. I do not think, however, that such a conclusion is mandatory. If knowledge is naturalizable, then KM is compatible with a naturalist semantics, and its scope against MS is the widest. In particular, if the knowledge channel were a kind of causal channel, KM would be a generalization of Kripke's (1972) insight about names and general terms getting their reference via a causal channel. The truth of the antecedent of this conditional depends on the feasibility of reliabilist programmes in epistemology. As far as the present discussion is concerned, the question remains open -- once established that KM can have a role against meaning-scepticism, it remains to be settled how much it can do.
Università di Milano-Bicocca
Dipartimento di Psicologia
<elisabetta [dot] lalumera [at] unimib [dot] it>
[Foot Note 1]
It might be objected that this formulation of MS begs the question against Fregean accounts of meaning, as it does not mention senses at all. In fact, meaning sceptics tend to reject senses on independent grounds. This is the case for typical meaning-sceptics, i.e. Quinean, Davidsonian and Kripkensteinian-minded philosophers, and this is why I favour the reference-centred formulation here. Apart from this contingent fact, an interesting question would be whether a Fregean theory of meaning (with both reference and senses) is in principle immune to meaning- sceptical arguments, or not. I think it is not. As Kripke (1982) showed, if a putative meaning-relation is a satisfaction relation, i.e. it features a description that the putative referent ought to satisfy, the Sceptic can raise his doubts about the meaning of each of the terms contained in the description, thereby generating a regress. Given that the most plausible reconstruction of sense is via a definite description, such a move amounts to an extension of MS to senses. This, obviously, is not to say that there can be no Fregean reply to MS. Rather, the argument proposed in this paper should be taken as an alternative, not as an objection, to a possible Fregean stance on the issue.
[Foot Note 2]
Four views about the relationship between concepts and words are currently debated. According to thought-theorists, concepts are the primary bearers of meaning. For them, MS about linguistic meaning only would be trivial, and MS about concepts would automatically infect language. According to language-theorists, words of everyday languages are the primary bearers of meaning. For them, the above dilemma holds as well, mutatis mutandis. For identity-theorists and eliminativists (concepts are words, or there are no concepts but words), MS is again either trivially true, when about concepts only, or infectious, when about words only.
[Foot Note 3]
See Davidson 1984.
[Foot Note 4]
The position is unstable, but it is not as naïve as being overtly self-contradictory. The Kripkean sceptic raises his challenge about past meanings and past uses, on the assumption that meanings here and now are fixed and shared; then he goes on challenging the grounds of such assumption.
[Foot Note 5]
I'm not addressing here the exegetical question whether Quine and Davidson themselves could have accepted MS, couched in the formulation chosen here. They did surely agree on the dispensability of a strong notion of reference in a theory of meaning, as well as on the dispensability of senses -- such being the upshot of an holistic theory. Quine is generally taken to endorse the thesis of the dispensability of senses with his famous rejection of analyticity (Quine 1953.
[Foot Note 6]
Kripke is explicit about the metaphysical character of MS (Kripke 1982, 27). Quine, on the other hand, is explicit about his dismissal of epistemological scepticism in general, given his conviction that we should `surrender the epistemological burden to psychology' (Quine 1969, 75).
[Foot Note 7]
Given some (behavioural) constraints on the nature of evidence.
[Foot Note 8]
See Lewis 1997. More precisely, Lewis does not regard sceptical alternatives as absolutely ineligible for reference, but thinks that some properties (the natural ones) attract reference more than others (less natural ones). I have argued for a solution of this kind against Kripke's sceptic in Lalumera 2005.
[Foot Note 9]
Williamson uses «reference» where I use «meaning».
[Foot Note 10]
See Williamson's example where there are two putative assignments of meaning, and they both exist in the subject's environment, but only one is related to the subject through a reliable belief formation mechanism (perception) whereas the other is not (divination).
[Foot Note 11]
This latter stance is the Fregean stance, which comes out as an alternative to the view defended here.
[Foot Note 12]
Williamson 2005, 142.
[Foot Note 13]
See psychological studies on the notion of objecthood in relation to cognition, especially Spelke et al.(1993).
[Foot Note 14]
Why knowledge-maximizing meaning assignments rather than truth-maximizing meaning assignments? Basically, for the same reason expressed in section 3 above, that is, because meaning, formation and attribution of individual concepts have some ontological and explanatory priority vis-à-vis the total assemblage of thoughts at which a person arrives.
[Foot Note 15]
A similar strategy involving principles about cognition against scepticism is employed in Burge 2003.
[Foot Note 16]
Williamson's argument, as I understand it, includes a premise like `KM _ Veridicality'.
[Foot Note 17]
See Schechter (2005) for a defense of an evolutionary explanation of the validity of deductive rules.
[Foot Note 18]
Principle N', like its ancestor N, should not be confused with the basic idea of success semantics. Success semantics is the view that the content of a belief is individuated (roughly) by the actions it would bring to success (given some constraints). It is a thesis about the individuation of the content of beliefs Principles N' and N, on the other hand, simply state that true beliefs are more conducive to success than false ones, however their content (and their truth) comes to be individuated. One can hold (per absurdum) that true beliefs are read off from God One's mind, whereas false beliefs are read off from God Two's mind. N' would then say that God One's givings are more conduceive to success than God Two's ones. Thus, N and N' are compatible with virtually any theory of content whatsoever, and they are not vulnerable to objections typically raised against success semantics. For success semantics see e.g. Whyte (1990).
[Foot Note 19]
It might be objected that in such a case one would need a true belief about the position of one's head. A possible reply could be that such a content need not be a belief, but rather a non-conceptual piece of information.
[Foot Note 20]
The question about the nature of intentions and intending is not relevant for my purposes, as far as intending to do A is not equal to and doesn't entail believing one is doing A.
[Foot Note 21]
I would like to thank an anonymous reviewer of this journal for this point.