Wilfred Sellars (1974) argues that we should not construe claims about meanings as expressing a semantic relation between a verbal entity (a word, sentence, etc.) and a language-independent entity (abstract or concrete) but as claims about the functional roles of linguistic tokens. Thus we should construe
“chat” (in French) means cat
*chat*’s (in French) are •cat•’s
“La neige est blanche” (in French) means Snow is white
*La neige est blanche*’s (in French) are •Snow is white•’s
Where the expression “*chat*’s” is a metalinguistic distributive term that refers to all non-semantically individuated tokens with a certain shape or sound and the dot quotation expression “•cat•’s” uses the English token “cat” to exemplify its functional role in English. This expression says, in effect, that characters and letters of a certain shape in French have the same functional role as “cat” in English.
This device allows Sellars to construct a conception of meaning which is not committed to extra-linguistic abstract entities such as propositions. The meaning of s is not constituted by its relation to some abstract entity p but by its functional role in a given linguistic community (its role within its economy of language-entry, transition and exit rules). This is obviously an attractive notional device for nominalists who wish to rein in metaphysical commitments to non-linguistic abstracta. It reframes metaphysical issues about the existence of propositions or attributes as questions about the status of functional roles. Of course, functional roles are not metaphysically innocent or unproblematic. We can ask of the Sellarsian whether normative facts supervene on non-normative ones and what the consequences of this relationship are. If we can do no better than supervenience to describe their relationship, this will be a problematic outcome for many naturalists. A second question – not unrelated to the first – is that of how functional-inferential roles are individuated. Presumably, they cannot be individuated semantically if Sellars’ account of meaning is to be non-question-begging.
In this post I want to consider a puzzle that is related to the second problem. I have discussed an analogous issue with regard to Davidson’s interpretation-based semantics in “Radical Quotation and Real Repetition” (Roden 2004). I’m not confident about the metaphysical solution I proposed in that paper, but if something like it can begin to address the issue for Sellars account of functional classification this might help us think through the ontological underpinnings of interpretation.
The problem anatomized in “Radical Quotation” arose with regard to Davidsonian truth theories.
As Olaf Gjelsvik (1994) points out, the formal model used by Davidson presupposes that we can pick out bits of the language we want to interpret syntactically. Davidson’s account requires that radical interpreters have a stock of primitive terms referring to constituent expressions of the object language and that these can be assembled into ‘structural descriptions’ reflecting the syntactic composition of its sentences (Davidson 1984, p. 133). For example, an axiom in a truth theory for a language might say of a certain concatenation of three symbols that it is satisfied by a sequence of objects if the first member of the sequence is larger than the second member (i.e. giving it the extension of the predicate “….larger than….”.)
Why might this be a problem for Davidson? Well, it is a problem if we recall that Davidson’s use of model theory is designed to explicate an informal semantic notion: meaning. He proposes to do this by way of a notion he takes to be better understood: truth. Sellars’ approach (as I understand it) is procedural rather than model-theoretic. But it one might expect that it needs to meet analogous constraints (even if not the same ones).
So here’s where Gjelsvik thinks that Davidson’s account hits a bump.
If languages are individuated by the syntactic types composing their expressions – roughly, by the physical shape and structure of grammatical strings – the semantic properties of their sentences must be non-essential. It is thus possible for a sentence to have different semantic properties in different speech communities. But then a truth theory for one community can be made false if another uses tokens of these types differently. For example, on Twin Earth a language, Twinglish, might be spoken in which English-shaped predicates have contrary ‘meanings’.
The existence of Twinglish would be enough to falsify the T sentence:
‘Snow is white’ is True(E) if and only if Snow is white
Since it is the syntactic string referred to by ‘Snow in white’ which relativises a truth predicate, not the abbreviations “E” and “Tw”, there is nothing to distinguish it from a statement about a sentence of Twinglish:
‘Snow if white’ is True(Tw) if and only if Snow is white
If ‘. . . is white’ in Twinglish were a contrary of its English counterpart (meaning is green, say) the ‘only if’ would make it false.
According to Gjelsvik, the only alternative is to specify English sentences semantically. A formal theory of the Tarksian kind achieves this by defining a predicate that holds of all and only the true sentences of a language. But its theorems flow by stipulation and logical necessity. Davidsonian theories are supposed to express contingent, empirical claims about semantically uncharacterised sentences. Thus, Gjelsvik argues, a competent radical interpreter must assume that the world’s distribution of semantic properties is not of the Twinglish/English sort (Gjelsvik 1994, p. 34). The problem, here, is that this assumption utilizes pretheoretic concepts of subsentential meaning (using semantic concepts like “satisfaction” in the formalism of semantic theories is OK, according to Davidson, because they are part of the logical machinery of the theories. They are not explicatory as such)
It seems that a similar problem afflicts the metalinguistic statements that occur in Sellarsian functional role ascriptions.
*chat*’s (in French) are •cat•’s
Would be false if Twin French speakers used *chat*’s differently to •cat•’s . Indeed it would be false if anyone, anywhere used *chat*’s in a way that ended up giving it a contrary functional role . Thus there must be other assumptions built into the ascription of metalinguistic types that are not evident in this formalism.
Well, it might seem that the Sellarsian is in a more favorable position than the Davidsonian here. For Gjelsvik, Davidson cannot constrain the scope of truth based theories without introducing meaning by the back door. But the Sellarsian only has to to claim that the distribution of functional roles is not of the silly type that would have *chat*’s acquiring contrary functional roles all over the place.
The problem with this fix is that there is absolutely nothing silly about contrary functional roles. As Robert Brandom’s inferentialist account implies, a term can acquire different functional roles where people have contrary beliefs. We would expect dancing inferential roles to be par for the course within any speech community. In any event, Sellars cannot preclude rampant homonymy without making their functional roles essential to interpretants in metalinguistic sortal sentences. But this would also render them trivial.
In consequence many metalinguistic sortal claims are falsified by inferential nuances within and between language communities, while it would be perfectly conceivable that there are no true ML sortals at all (allowing for sufficient homonymic variability across speech communities).
It is not clear to me where this would leave Sellars’ metaphysics of meaning. For example, can we build in a tacit reference to a given speech community which can be expected to exhibit the uniformities described by metalinguistic sortal claims? Maybe, but as well as being questionable for the Davidsonian/Brandomian reasons mentioned above, it also seems to require an explicit notion of reference. If we cannot plausibly restrict the scope of ML sortals in such a way, however, it would seem to follow that most or many ML sortal claims are false (thus there are no ML functional types, or very few) Thus the claim that the meaning of a term is its functional role would have to be judged false as well.
My solution to the problem that Davidson faces is to treat metalinguistic statements in a constructionist spirit. Syntactical types – accordingly – are not contingent owners of functional roles. They are individuated by functional role. So English “white” and Twinglish “white” are distinct characters and not the same character used in different ways. The problem, then, is to account for the empirical, contingent character of claims like
*chat*’s are •cat•’s
For on this account *chat* is not part of a “language” (like French) in a conventional sense but of a local idiom constructed purely for purposes of interpretation. For reasons similar to those discussed by Davidson in “A Nice Derangement of Epitaphs” we are no longer supposing that the notion of a language is the basic one here. The issue, then, is what is the function of an interpretant such as *chat* in this sortal statement?
My solution circa 2004 was to say that its function is to repeat the utterances or parts of utterances used by native speakers of languages under interpretation (we return to the primal scene of radical interpretation, as it were). *La neige est blanche* is designed to quote expressions in one idiom in another idiom (that of the interpreting discourse). So
*La neige est blanche*’s (in French) are •Snow is white•’s
refers to a set of historically instantiated utterance events by repeating them. Thus there must be a historical-causal relation of some kind between the interpreter and users of the interpreted idiom which can explain its purchase on these (rather, say, than on users of an orthographically identical language on Twin Earth).
The ontological basis of this quotation is not exemplification of a common semantic type. It is an ontologically primitive relation of repetition or “iteration” (to use Derrideanese) which operates transversely between languages and language communities (non-language-relative repetition). Some events, it must be assumed, just repeat other events without having to fall under a common description. The worry, now, is that the interpreted terms in ML sortings are being used as instances of the items they repeat rather being merely structural descriptions or examples of sign-designs. They are being used, so to speak, to refer to themselves. But if this is right, then the very act of interpreting them constitutes a variation in functional role. It is also a function that cannot obviously be expressed in inferential terms.
Finally, if the interpretants are essentially repeatabilia, then it is part of their job description (so to speak) that they can always accrue functional roles which differ from the ones they have had (otherwise interpretation would have no text). But then it cannot be inappropriate to use them in these “deviant” ways. Thus there no longer seems to be room for the normative facts which (supposedly) undergird the functionalist account.
Davidson, Donald (1984). Inquiries into Truth and Interpretation (Oxford: Clarendon Press).
____1986. ‘A Nice Derangement of Epitaphs’, in Ernest LePore (ed.) Truth and
Interpretation: Perspectives on the Philosophy of Donald Davidson (Oxford: Blackwell).
Derrida, Jaqcues. 1988. Limited Inc. Samuel Weber and Jeffrey Mehlman (trans.) (Evanston Ill.: Northwestern University Press).
Gjelsvik, Olav. 1994. ‘Davidson’s Use of Truth in Accounting for Meaning’, in Language, Mind and Epistemology: on Donald Davidson’s Philosophy, Gerhard Preyer, Frank Siebelt and Alexander Ul?g (eds.) (Dordrecht: Kluwer), pp. 21–43.
Lewis, Kevin. 2013. ”Carnap, Quine and Sellars on Abstract Entities”, https://www.academia.edu/2364977/Carnap_Quine_and_Sellars_on_Abstract_Entities (Accessed 12-7-14)
Roden, D., 2004. Radical quotation and real repetition. Ratio, 17(2), pp.191-206.
Sellars, W. (1974). Meaning as Functional Classification (A Perspective on the Relation of Syntax to Semantics). Synthese, (3/4). 417.