From the ALWS archives: A selection of papers from the International Wittgenstein Symposia in Kirchberg am Wechsel

1. The Bounds of Linguistic behavior

My objective in this section is to show that both Quine and Davidson, as representatives of what may be described as an interaction oriented approach to language, presuppose in their theorizing a distinction between linguistic and non-linguistic behavior. However, such a distinction needs to be accounted for rather than be presupposed. I show what kind of delineation of the bounds of linguistic behavior can be given within their (respective) frameworks, but argue that the details of such an answer are in need of explicit elaboration.

One distinctive aspect of Quine's account of language, as first presented in his Word and Object (Quine 1960), is the way it combines a general empiricist outlook with an interaction oriented view of language. Quine views the interpersonal set-up of radical translation as exhausting whatever there is to be said about the function of language and the way it is associated with sensory stimulation. This is a major point where he breaks away from logical empiricists such as Carnap, in response to whose views he forms and defines his own.

Thus the hypothetical situation of radical translation is the starting point of Quine's (partly destructive) analysis of linguistic meaning. However, throughout the discussion of this situation Quine never asks which actions of the foreigner (whose speech needs to be translated) are linguistic utterances, and which are not. Quine's whole treatment of radical translation, from Gavagai onwards, presupposes that the translator can distinguish between language and non-language in the other's behavior. Quine never mentions a step (or aspect) of the translation process where the linguistic behavior of the other is delineated—he seems just to presuppose that the relevant input for translation is auditory, and that the translator can pick out of the stranger's auditory behavior those actions that are utterances. But what justifies this supposition? How is the delineation of language to be accounted for in this context? I claim that these are legitimate questions, that Quine seems not to be aware of them, and that therefore he does not provide an answer to them.

Donald Davidson follows Quine in this respect. He describes his thought experiment of radical interpretation (a descendant of Quine's radical translation) as providing the grounds for an answer to the question (Davidson 1984, p. xiii) ".. what is it for words to mean what they do?", but never addresses the (seemingly) simpler question which acts of the person being interpreted are utterances and which are not. The bounds of linguistic behavior—i.e., behavior that is amenable to radical interpretation—are taken for granted.

In a sense, Davidson's need to face this question is even more urgent than Quine's, for the following reason. As opposed to Quine, Davidson makes room in his theory for talk about the mental, and views interpretation as involving not only the assignment of meaning to a person's utterances but rather also the assignment of propositional content to her beliefs and desires (as well as other propositional mental states). This multidimensional task is to be carried out not only on the basis of the linguistic behavior of the person he is engaged with, but rather also taking into account her non-linguistic behavior, which can bear witness to her propositional attitudes (and, indirectly, to the meaning of her utterances). Thus according to Davidson all of the agent's behavior is subject to interpretation in a general sense (i.e. as being related to her mental states), but only some of this behavior is subject to interpretation in the more restricted sense of being assigned T-sentences (of a Tarskian truth theory). So, the question arises, what are the grounds for deciding which actions are subject to which type of interpretation?

Before we consider how Davidson and Quine can delineate the bounds of linguistic behavior, let us first note two kinds of grounds for such delineation that are quite often appealed to for this purpose, but that are not available to them. The first is the demarcation of linguistic behavior on the basis of its relation to internal psychological processes. According to Chomsky (2002, 2006), for example, language is essentially an internal, computational mechanism. Whatever behavior that is directly related to this internal mechanism (through phonological (or similar) encoding) is linguistic, and whichever behavior that is not so related to it is not language. Surely this is a widely accepted way to delineate the bounds of language; however, it is not available to Davidson and Quine. The reason is this. Both reject the idea that meaningfulness arises from the connection of language to internal states and processes. It follows that the domain of meaningful behavior (i.e. its extension) cannot be demarcated on the basis of an appeal to internal states and processes as well. Hence if language is characterized in the way suggested by Chomsky, an unacceptable gap is opened between what is (potentially) meaningful and what is linguistic: In principle, there could be actions that are translatable/interpretable without being linguistic, and vice versa. I take this result to be untenable.

A different avenue for grounding the bounds of language that is clearly unavailable to Davidson (and most probably cannot be pursued by Quine as well) is conventional. It is a widely accepted view (even if not by Chomskian linguists) that language is constituted by a set of social conventions. Among other things (or, rather, prior to them), these conventions determine which actions are linguistic—that is, which are legitimate moves in the conventionally constituted language game, and which are not. Now as Davidson famously rejects the view that language depends in an essential way on convention (Davidson 1984b), this way of delineating language is obviously not available to him. By the same token, it seems that this conclusion applies also to Quine (although he does not say so explicitly himself): Nothing in his radical translation depends on the translator's being able to place the foreigner within a web of social conventions, and therefore whatever is necessary for such translation—including the delineation of the bounds of language—cannot have an essential conventional basis.

Is all this to say that Quine and Davidson are left without means to distinguish between language and non-language? I argue that this is not the case. The alternative to the above mentioned ways of delimiting language is that the very same process through which language is assigned meaning (be it radical translation or interpretation) will be also the context where language is distinguished from non-language. This is the only way that seems to be open to Davidson and Quine: The processes that they describe as constituting meaning must also be burdened with the extra task of giving rise to the extension of the term 'language'. In other words, whatever is interpretable/translatable is linguistic.

This conclusion, though, is not enough. It still needs to be articulated how language is distinguished from non-language in the context of translation/interpretation. The simplistic response that this is a matter of trial and error will not do: Some indication must be given which actions are good candidates for being utterances, or else too much freedom is allowed. Similarly, the identification of linguistic behavior on the basis of its overt syntactical complexity is not enough: Such complexity is arguably discerned in (or ascribed to) some types of behavior on the basis of their characterization as linguistic rather than prior to such characterization. Without an articulation of the way language is delineated through interaction a key aspect of the projects in question (and of the whole interaction oriented perspective on language) is left standing on shaky grounds. In the next section of this paper I show (albeit in outline) how such an articulation may be provided.

2. Triangulation and the Identification of Language

In a series of papers from the early nineties onwards Donald Davidson presents an application of the notion of triangulation to the analysis of the fundamentals of language (Davidson 1991). The objective of this section is to present a brief outline of Davidson's use of this notion, and then to show how it bears upon the question raised in the previous section.

One of the major concerns of Davidson's later work is the analysis of the basic inter-subjective setup that allows linguistic communication to emerge, in the form of talk about mutually accessible objects in a shared environment. (In his earlier work Davidson paid less attention to situations of this kind.) In order to pursue such analysis Davidson considers a communicative scenario that does not exhaust all there is to language (as he emphasizes), but that nevertheless can help us better understand some of the essentials of language and how it is related to more basic forms of communication. The scenario includes two creatures in a shared environment. Each creature's behavior is in partial correlation with the environment, in the sense that it forms patterns of responses to some changes in the environment. Furthermore, each creature perceives both the environment and the other creature's behavior, observing some correlations between the two, and reacting to such observed correlations. Thus each creature's behavior can be affected not only by, e.g. the appearance of predators, or by the other creature's hand movements, but also by an observed correlation between the two. Such a correlation is a necessary condition for the said movements to be signs of predators, i.e. proto-linguistic.

Davidson invokes this setup for various purposes. One is that it helps him support his claim that it is the so called distal stimulus that is the content of linguistic expressions: Triangulation helps pick out of the causal chain leading to my utterance the link that is its content by intersecting this chain with another, leading to my interlocutor. Another use of this scenario is to point out how social externalism with respect to language (that Davidson, following Wittgenstein, endorses) comes into play: One creature's behavior can be described as regularly correlated with the environment only from another creature's perspective.

I propose to use this conceptual construction in order to answer the question raised in the first section (which, as argued above, Davidson fails to consider). The characteristics of acts vis-à-vis this setup are those that must be appealed to in order to delineate language, rather than connections of acts to some internal psychological reality, or their conformity to social convention of some kind. Thus we get the result that those actions of a creature that may be correlated with the environment by another creature are those that are potentially linguistic. There are several ways in which this characterization may be developed and elaborated. Let me conclude by indicating two of them.

First, note that correlation is a relatively weak connection. Potential utterances need not be efficacious in bringing about the changes in the environment they are correlated with, nor need they be acts that are necessary responses to such changes (in the sense of ensuring survival). There are things that we do in order to change the environment (such as building fences) or to face changes in it (e.g., run away from danger), but utterances need not be this way—their effect on the environment is to arise from their observed correlation with the environment rather than to underlie it. This fact allows for the use of voice in order to make utterances: Our world is such that the production of sounds is typically not directly operative in, e.g., building fences or moving us away from danger, but it can be modulated so as to be correlated with fences and dangers. Indeed, actions like auditory expression—that typically do not have a direct impact on the environment nor seem to be an observable result of a direct impact from it—are arguably good candidates for being utterances. They are acts the effects of which on the environment are produces in virtue of their being correlated with it be other creatures. (This is not claimed here to be a necessary or sufficient condition, though.)

The second point is that correlation may be viewed as underlying the notion of correspondence (or infusing new content into this notion). An expression may be said to correspond to reality (i.e., the environment) not because its internal syntactic structure somehow represents a part or aspect of reality (e.g., a fact), but rather because its production (in a given context) is in accord with an established correlation pattern. Davidson (1984c) is correct in saying (with respect to full fledged language) that this way we get correspondence with the world (a.k.a. the Big Fact) rather than parts or aspects of it, but wrong in maintaining that this renders the notion useless. Indeed, the foregoing discussion indicates that such a weak notion of correspondence may be operative in answering the question this paper started with. That is, as opposed to fence constructions and flights from danger, utterances may be characterized as actions that have the potential to correspond to the environment in the sense suggested above. A translator/interpreter may say to himself: "I have seen behavior of this kind several times before, in similar contexts. It does not seem to directly change the environment, nor to be direct result from changes in it. So maybe it just corresponds to the way the environment is. Maybe it is linguistic—I'll try to figure out what it means."