Rules, Acts and Interpretation: A Wittgensteinian View of Linguistic Communication

I've written another paper for another graduate "course," entitled "Theories of Linguistic Communication." It's not as strong or thorough as I'd like it to be, and it's a bit disorganized, but I don't have more time to work on it. I'm not sure how committed I am to the views I'm expressing here, either. How's that for a disclaimer? In a nutshell, my arguments and views are still in development, but hopefully this short essay will be of some interest.

Rules, Acts and Interpretation:

A Wittgensteinian View of Linguistic Communication

I

According to the standard view of linguistic communication, to know a language is to know a set of rules which allows one to deduce the truth-conditions of any well-formed sentence in that language (Recanati 2002, p. 105). These rules are semantic, which means they relate sentences to the propositions they express. Such rules may be context-sensitive: some linguistic entities may even “wear their context-dependent nature on their sleeve,” as Jason Stanley puts it (Stanley 2002, p. 150). Kent Bach thus distinguishes between wide and narrow context: there is “a short list of variables, such as the identity of the speaker and the hearer and the time and place of an utterance” which combines “with linguistic information to determine content (in the sense of fixing it)" (Bach 1997, quoted in Recanati 2002, pp. 110-111). This list comprises the narrow pragmatic context, the identification of which is recognized as a part of semantic interpretation, because the goal is not the evaluation of speaker intentions or beliefs, but the recovery of the truth-conditional content expressed by a sentence. When the goal of interpretation is the speaker’s intentions and beliefs, a different process of interpretation must occur, one which takes into account the wide context—in other words, one which is sensitive to any and every possible fact. There is no set of linguistic rules which limits the number, type or range of entities which can influence our interpretation of speaker intentions and beliefs. Speaker meaning is thus distinguished from sentence meaning: the former requires pragmatic processes of interpretation while the latter requires semantic processes. Semantic processes are distinguished both by their reliance on linguistic rules and their goal of identifying the values of lexical items. A composed list of such values is called an explicated proposition, or what is said by an utterance. Pragmatic processes may result in the construction of another proposition, one which is implicated by an utterance. Pragmatic processes do not tell us what a sentence means, but only what a speaker means when they use a sentence. Speaker meaning and sentence meaning may often be identical, but in many cases diverge.

Since speaker meaning and sentence meaning often diverge, it would be misleading to suppose that one knows a language if one does not know the conventional uses of the language which account for this divergence. To fully speak and understand a given language, one must be privy to the idioms which allow language users to easily recognize when words are being used to mean something other than what they say. Furthermore, this divergence cannot be explained by semantic rules and is not identified by means of semantic interpretation—indeed, that is what distinguishes implicature as such. If it could be determined by semantic interpretation, it would not be an implicature. Thus, knowing a language cannot be simply a matter of knowing the semantic rules of deduction and the syntactic rules of combination. Rather, knowing a language must also include knowing the conventional rules which govern implicatures. If these rules are sensitive to any and every possible fact, however, it is not clear how determinate they can be. Unlike semantic interpretation, pragmatic interpretation is in some fundamental sense indeterminate.

II

The question I have so far been discussing concerns the nature and extent of the rules governing linguistic communication. At this point, I want to raise and address a few related questions which have divided contemporary theorists.

1. What is the relationship between semantic and pragmatic interpretation? According to the standard view, they are distinct processes and semantic interpretation is primary in normal linguistic communication. Pragmatic interpretation is not always necessary; it is only vital when speaker meaning diverges from sentence meaning. According to Recanati (2002), however, semantic interpretation is always dependent upon pragmatic interpretation: the fixing of the narrow context can only be done by appealing to the wide context. Semantic rules are not sufficient to determine what is said. Then there are relevance theorists, who claim that semantics and pragmatics are combined in a single process of interpretation, that the identification of speaker meaning occurs through the same process as the interpretation of sentence meaning, whether or not these meanings are identical or divergent (Carston 2002). Recanati agrees with the relevance theoretic view that no propositional content is recovered prior to pragmatic interpretation, though he still distinguishes between semantic and pragmatic processes of interpretation. In his view, the distinction rests on the conscious availability of premises about speaker intentions. Recanati claims there are pragmatic processes (secondary pragmatic processes, to be exact) which take beliefs about speaker intentions and beliefs as arguments in inferential processes concerning the intended meanings of utterances. Thus, when Recanati claims that pragmatic interpretation is always involved in linguistic communication, he means that primary pragmatic interpretation is always involved; he does not suppose that normal communication requires inferences about speaker meaning and intention.

2. Is semantic interpretation constrained by syntax? According to Stanley (2002), it is: “all the constituents of the propositions hearers would intuitively believe to be expressed by utterances are the result of assigning values to the elements of the sentence uttered, and combining them in accord with its structure" (Stanley 2002, pp. 150-151). According to relevance theorists, however, there are elements of what is said which cannot be determined by syntax and semantics alone.

3. Does semantic interpretation rely on inferences about speaker intentions and beliefs? According to inferentialists, semantic interpretation is an inferential process which relies on premises about the intentions and beliefs of the speaker. According to anti-inferentialists, no such inferences normally occur in linguistic communication. What is said, as opposed to what is meant, can be recovered without inferences from speaker intentions and beliefs. Linguistic communication is, in normal cases, as direct as perception.

Relevance theorists do not claim that linguistic interpretation rests on inferences about speaker beliefs and intentions. Rather, they claim that interpretation produces judgments about such beliefs and intentions as output. They deny that implicatures require prior outputs about sentence meaning as arguments in an inferential process of interpretation (what Recanati calls “secondary pragmatic processes.”) Recanati also rejects inferentialism, claiming that normal linguistic communication requires only primary pragmatic processes, and that no inferences of any kind need occur at all.

Carston (2002), a relevance theorist, claims that, because of the “highly context-sensitive nature of sense selection and reference assignment . . . they are matters of speaker meaning, not determinable by any linguistic rule or procedure for mapping a linguistic element to a contextual value, and so just as dependent on pragmatic principles as the process of implicature derivation” (Carston 2002, p. 134). Recanati observes this same context-sensitivity, which he calls “semantic underdetermination.” According to this thesis, the values of lexical items can only be determined by appealing to the wide context of utterance. The notion of wide context is distinguished by the fact that it is indeterminate: any fact at all might enter the wide context and influence the interpretation of an utterance. Thus, the correct interpretation of an utterance requires pragmatic interpretation: it cannot rely on semantic rules alone.

As noted above, Recanati distinguishes between two types of pragmatic interpretation: primary and secondary. Primary pragmatic processes are not determinable by linguistic rules and procedures for mapping lexical items to their values. Primary pragmatic processes do identify what a speaker means, and not simply what a sentence says by itself. Sentences do not, on Recanati’s view, say anything by themselves. Yet, Recanati observes, primary pragmatic processes do not rest on inferences about what a speaker means. They need not involve any inferences at all.

Carston goes further. Instead of relying on the relatively uncontroversial claim that semantic interpretation is deeply sensitive to context, she claims that semantic interpretation is not normally constrained by syntax:

There is a wide range of cases where it seems that pragmatics contributes a component to the explicitly communicated content of an utterance although there is no linguistic element indicating that such an element is required. That is, there is no overt indexical, nor is there any compelling reason to suppose there is a covert element in the logical form of the utterance, and yet a contextually supplied constituent appears in the explicature (Carston 2002, p. 135).

These constituents are what Stanley calls “unarticulated elements.” Stanley’s method of refuting the existence of such elements is to identify, on a case by case basis, hidden (but articulated) lexical items and so account for the explicature (aka “what is said”) in terms of the logical form (syntactical structure) of the uttered sentence.

Stanley’s methodological suggestion is that “an unpronounced element exists in the structure of a sentence just in case there is a behavior that would be easily explicable on the assumption that it is there, and difficult to explain otherwise” (Stanley 2002, p. 152). As an illustration of the application of this principle, Stanley presents an argument for the existence of unpronounced by-phrases in passive constructions, such as “The ship was sunk to collect the insurance.” The by-phrase, he says, must be present as a local controller of the unpronounced pronominal element, ‘PRO,’ which is supposedly the subject of infinitival clauses. How does Stanley know that there is an unpronounced pronominal element which must be controlled? Because, he says, sentences like “The ship sank to collect the insurance” are ungrammatical. Ships are not the sorts of things that can collect insurance. For Stanley, to say that a ship is not the sort of thing that can collect insurance is to say that “the ship” cannot control the pronominal element in “The ship sank [PRO to collect the insurance].”

Pace Stanley, it is not obvious that “the ship sank to collect the insurance” is ungrammatical. It is unusual, perhaps, and this is because we usually do not attribute to ships the ability to collect insurance. The only obvious fact is that collecting insurance is an intentional activity, and ships are not intentional agents. So we reject the content of the sentence, but not its grammatical form. If the sentence is grammatical, however, there is no need to postulate an unpronounced pronominal element. It does not seem easier to stipulate an unpronounced pronominal element, and so Stanley’s criterion is not satisfied.

I take the preceding argument as grounds for rejecting Stanley’s example, though it is worth exploring Stanley’s full application of his proposed criterion. Stanley’s argument is that the alleged unpronounced pronominal element in passive constructions (such as “The ship was sunk to collect the insurance”) is controlled by unpronounced by-phrases. Stanley concludes that all passive constructions must include such unpronounced by-phrases. For example, I cannot say “The student was well-informed” without saying who or what informed the student. I doubt Stanley would claim that I necessarily know who or what informed the student. Rather, for the sentence to be true, somebody or something must have informed the student well, even if I don’t know who or what it is, and that somebody or something is what enters into the proposition.

My objection is that it need not be the case that a determinate somebody or something is responsible for a student being well-informed. Arguably, we sometimes use the passive voice in such cases where no determinate element exists. So it is implausible to suppose that a by-phrase must be saturated for such sentences to have meaning. Stanley might be better off saying that there is an unpronounced by-phrase only in case it is required to control the alleged unpronounced pronominal element. Yet, we lack sufficient grounds for supposing any such element exists, and it would be circular to argue for their existence by appealing to them. True, it does seem that “The ship was sunk to collect the insurance” is true only in case there is some value for the entity which intended to collect the insurance by sinking the ship, even if I do not know who or what that entity is. Yet, we can claim that the by-phrase is implicated by the sentence and not a part of its syntactical form. The existence of an agent is implied—it is a logically necessary consequence of the sentence meaning—without being an articulated element of the sentence.

It is not obvious that there are ever cases in which the only, or even the easiest, way to explain linguistic behavior is to postulate unpronounced lexical items. Stanley’s project appears to have an insurmountable methodological difficulty. In any case, this does not necessarily support Carston’s position that the logical form of an explicature underdetermines its meaning. That point ultimately depends on how we understand logical form.

III

I propose a distinction between the syntax of a sentence and its logical form. If the meaning of a sentence has logical entailments, then we might suppose those entailments are a part of its logical form. Indeed, there does not seem to be any principled way of distinguishing between what a sentence means and what is logically entailed by that sentence. This was evident in the case of “The ship was sunk to collect the insurance.” In so far as that sentence logically entails that somebody intended to collect insurance by sinking the ship, then we might suppose that agent is part of the proposition expressed by the sentence. Certainly, if no such agent exists, the sentence is false. Is it false because of the content expressed by its logical form or is it false because of a proposition which is logically entailed by its logical form? I do not suppose there could be any principled way of deciding this question. There is no methodological basis for distinguishing between meaning and logical entailment. If we take the proposition expressed by a sentence to have a logical form (as Carston and Stanley agree must be the case), then we have no reason to reject any logical entailments as being distinct from the proposition so expressed.

Stanley and Carston both equate the logical form of an explicature with the syntactical construction of its linguistic elements. Thus, while Stanley claims that logical form constrains the proposition expressed by a sentence, Carston claims that it does not. Yet, I propose that we instead take the logical form of an explicature to be the logical form of the proposition it expresses. Logical form is not determined by the sentence alone, but rather by the speaker’s intentions and beliefs as they are manifested in the use of a sentence.

Stanley claims that the logical form (taken as identical to the syntactical structure of a sentence) expressed by a sentence is determined by the speaker’s linguistic intentions (Stanley 2002, p. 150). However, according to Stanley, speaker intentions are constrained by the rules of the language. Speaker intentions determine logical form in so far as a speaker intends to utilize such-and-such properties of a language. It is because a speaker intends to assert X that the sentence uttered has the logical form of X. Another speaker might utter the same words without the intent to make an assertion, and so would not express a proposition at all. What distinguishes an assertion from the mere production of sounds or markings is an intention to make an assertion. I suppose that this, for Stanley, is why the logical form is determined by speaker intentions: the speaker intends to follow the rules of the language.

The crucial question, then, is whether or not the intended sentence contains elements corresponding to all of the elements of the proposition it expresses. Does syntax determine logical form? If we take a sentence to be more than what is physically produced—more than an assortment of sounds or markings—but as a combination of elements which can be used to express a proposition, then the sentence may be said to have a logical form—to be a token of a logical type. What makes it a token is its logical form, and this seems to be what we mean when we talk about syntax. And if the logical form is identical to its logical entailments, then the sentence contains everything that is logically entailed by the utterance.

Yet, there must be some constraint on what can be considered a logical entailment. For what counts as logical entailment in ordinary discourse depends on assumptions about relational properties. What a person logically entails is a matter of what that person believes about the world. These beliefs cannot be deduced from logic or language alone. So it appears that logical entailment is really just a matter of intended entailment. The proposition expressed by an utterance cannot be deduced by the rules of logic and language alone, but must follow from the speaker’s beliefs and intentions. This is not to say that the process must take speaker beliefs and intentions as argument. It is only to say that some accordance with speaker intentions is required, and this accordance cannot be produced by strict adherence to the rules of logic or language. In so far as logical form is a matter of logical entailment, it cannot be a matter of syntax alone.

IV

We now have a picture of logical form which is not limited by the syntactical properties of a sentence, but by the truth-evaluable properties of what a speaker intends a sentence to say. Sentence meaning is always intended sentence meaning, which is still distinguishable from speaker meaning—not because the latter alone is uniquely intended by the speaker, but because only the former incorporates syntactical features of the language. There are no principled, rule-based criteria for assigning logical forms to sentences. Syntax must play a part in sentence meaning, but it does not completely determine its logical form.

If the question is “What constitutes sentence meaning?”, then we seem to end up with a dilemma: we cannot identify any proper constituents of sentence meaning, because we cannot identify the essential properties of a sentence apart from its syntactical components, and these are not sufficient to determine logical form. Perhaps the question should not be What constitutes sentence meaning?, but rather, How are notions of sentence meaning constructed? This question is a question about how interpreters of speech acts go about creating notions such as what is said and what is meant. If we approach the topic this way, we may regard the notion of sentence meaning as the result of interpretation. Of course, we still want to be able to say things such as, “That is not what I meant.” We want to be able to test interpretations against intended meanings. Yet, we cannot do so by appeal to facts about particular sentences. We can only do so by fixing sentence meanings as interpreters of our own discourse. The notion of sentence meaning, then, is not determined by speaker intentions; rather, speaker intentions and sentence meaning are co-determined by the interpretation of utterances.

Sentences have meaning in so far as there are interpreters who interpret them. And it is permitted to postulate a sentence meaning in so far as our interpretation of certain sorts of behavior requires it. So, if there are linguistic rules which determine the logical form of sentences, these rules must exist as elements of interpretation. There are no facts about what a sentence means—or even about whether or not a sentence has been produced—prior to the interpretation of a speech act as such.

If we are going to investigate linguistic communication, then, we should focus on the processes of interpretation. This is not to say that speech production is irrelevant. Rather, it is to say that speech production cannot be identified unless we understand what it means to interpret speech. We might even define speech production as the intentional triggering of processes of linguistic interpretation. Whatever intentionally triggers processes of linguistic interpretation is, by definition, a speech act. Thus, there can be no study of speech acts apart from the study of interpretation.

The view I am promoting owes some gratitude to Wittgenstein and his rule-following paradox. In the controversial but influential formulation proposed by Kripke (1982), the dilemma takes the following shape: there can be no facts about the intended meaning of an expression which determine any future uses of that expression. According to Kripke, Wittgenstein held that there are no facts about intended meaning at all. I am not convinced this is correct. I believe we can read Wittgenstein as denying that there are any facts about intended meaning which logically necessitate future uses of expressions. However, I do not think Wittgenstein thereby supposes that there are no facts about intended meaning at all. Wittgenstein is only rejecting the ability of any such facts to constitute a logical foundation for our metalinguistic discourse.

If we attempt to determine what rule we are following by our use of a particular expression, we end up with an infinite regress: we can only appeal to other rules or other expressions of the same rule, and so never arrive at a final interpretation of the rule we are intending to follow. Wittgenstein concludes: “there is a way of grasping a rule which is not an interpretation, but which is exhibited in what we call ‘obeying the rule’ and ‘going against it’ in actual cases” (Wittgenstein 1953/2001, section 201, p. 69) In other words, processes of semantic and pragmatic interpretation cannot themselves be wholly circumscribed by rules. The indeterminacy of pragmatic interpretation was noted at the outset, when it was observed that sensitivity to every possible fact indicates a fundamental lack of linguistic control. The Wittgensteinian view of logical form should not be surprising, then, once it has been acknowledged that semantic interpretation relies on the wide context.

V

According to inferentialists, you cannot understand an utterance without inferring from beliefs about the speaker’s own beliefs and desires. To correctly interpret an expression, you must first interpret the speaker’s beliefs and then deduce the correct interpretation of their utterance. Anti-inferentialists, on the other hand, claim that the correct interpretation of an utterance does not depend on any inferences from speaker beliefs, but can be determined solely by knowledge of the language and determinate features of the context of utterance. My view is perhaps more sympathetic to anti-inferentialism. We cannot identify that any of a speaker’s beliefs and desires are entailed by her utterance without first interpreting that utterance. We cannot infer from the beliefs and desires to sentence meaning when that very meaning is what leads us to stipulate beliefs and desires. It appears that both the beliefs/desires we attribute to a speaker and the meaning we attribute to her utterance are interpreted together; we can speak of what a speaker says only because we can speak of her relevant beliefs and desires, and we can speak of the latter only because we can speak of the former. Our ability to interpret speech must depend on our ability to interpret personal qualities. The interpretation of speech is an interpretation of personal qualities. While the personal qualities might not enter the process of interpretation as premises, they are represented in judgments about that speech. Still, this process need not be inferential.

I am also sympathetic to Carston’s relevance theoretic approach. I see no reason to reject her position that semantic and pragmatic interpretation (even Recanati’s “secondary” pragmatic interpretation) cannot occur in a single process of interpretation. There does not seem to be a principled method for distinguishing between explicature and implicature. No set of conventions could decide ahead of time whether or not a particular speech act was conventional or unconventional, and whether or not an entailment was explicated or implicated. This is not to deny that some meanings are not more implicit than others, or that some are not less conventional than others, nor is it to deny that there is any such thing as “literal meaning.” It is rather to make the Wittgensteinian point that these notions are not wholly circumscribed by rules, and that there is no criteria to satisfy prior to acts of interpretation.

Recanati’s distinction between primary and secondary pragmatic processes relies on the notion of conscious availability. Supposedly, when an utterance defies normal interpretation, secondary pragmatic processes kick in and, using the sentence meaning as argument, infer possible speaker meanings. But the notion of conscious availability seems too weak to do much work here. We do not always need to take sentence meanings as argument when we identify implicatures. The identification of an implicature can be as automatic as any act of interpretation.

There is a temptation to say that some uses of a linguistic construction are simply incorrect or unconventional. However, we necessarily lack the means of determining what distinguishes the conventional from the unconventional, or the correct from the incorrect, in advance of particular cases. In other words, whether or not we should regard any given use of a linguistic construction as “correct” or “conventional” (or, conversely, as “incorrect” or “unconventional”) is theoretically undecidable.

This Wittgensteinian view is not opposed to analyzing content in terms of logical form. However, it is opposed to the notion that we could define logical form in advance of our analysis of content. Thus, when we interpret a speech act, we do not first identify the complete logical form and then apply rules to identify the content. We rather identify the logical form by virtue of our understanding of the content.

VI

Theorists like Jason Stanley will always be able to revert to hypotheses about hidden lexical elements in attempts to defend the view that “every feature” of the communicated content “must be the semantic value of something” in the syntactical form “or introduced via a context-independent construction rule” (Stanley 2000, quoted in Stanley 2007, p. 36). Yet, since there is no way to determine the correct analysis of logical form in advance of our treatment of particular cases, Stanley’s strategy is suspect. It will always be possible to stipulate some hidden elements to account for the particular use of a linguistic construction. This does not mean that logical form somehow preceded that particular use. For Stanley’s method to work, we need an independent indication of logical form apart from actual speech acts—a purely semantic and syntactic theory—against which we can test our empirical observations. Yet, Stanley seems open to the possibility that any such theory will be open to possible revision when new pragmatic evidence is revealed.

What is the point of supposing that logical form precedes speech acts? Presumably, those who favor such a view suppose that correct linguistic interpretation requires adherence to rules for correctness, and such rules require some kind of linguistic discipline: Communication must be “under linguistic control,” a Stanley puts it (Stanley 2007, p. 10). From the Wittgensteinian view, this is not even wishful thinking: it is simply untenable. Linguistic control does play a role in communication, but it is not foundational. Linguistic control is the result of acts of linguistic communication, and so cannot be a condition of such acts.

The view I am promoting is not that there is no role for rules in linguistic communication. Rather, it is that this role is not foundational—or, rather, that the foundational aspects of rules do not precede our formulation of them. There are unquestionable cases in which disagreements or uncertainties about meaning are resolved by appeal to linguistic rules. Yet, it is a mistake to suppose that those rules somehow preceded our use of them in resolving our disputes.

Works Cited

Bach, Kent. 1997. The semantics-pragmatics distinction: what it is and why it matters. Linguistische Berichte, 8, 33-50.

Carston, Robyn. 2002. Linguistic Meaning, Communicated Meaning and Cognitive Pragmatics. Mind & Language 17, Nos 1 and 2 (February/April): 127-148.

Kripke, Saul. 1982. Wittgenstein on Rules and Private Language. Harvard University Press.

Recanati, Francois. 2002. Does Linguistic Communication Rest on Inference? Mind & Language 17, Nos 1 and 2 (February/April): 105-126.

Stanley, Jason. 2000. Context and Logical Form. Linguistics and Philosophy 23: 391-434, reprinted in Stanley 2007.

Stanley, Jason. 2002. Making It Articulated. Mind & Language 17, Nos 1 and 2 (February/April): 149-168.

Stanley, Jason. 2007. Language in Context. Oxford University Press.

Wittgenstein, Ludwig. 1953/2001. Philosophical Investigations. Blackwell Publishing.