Meaning in time: on temporal externalism and Kripkenstein’s skeptical challenge

Abstract

The main question of metasemantics, or foundational semantics, is why an expression token has the meaning (semantic value) that it in fact has. In his reading of Ludwig Wittgenstein’s later work, Saul Kripke presented a skeptical challenge that threatened to make the foundational question unanswerable. My first contention in this paper is that the skeptical challenge indeed poses an insoluble paradox, but only for a certain kind of metasemantic theory, against which the challenge effectively works as a reductio ad absurdum argument. My second contention is that as a result of rejecting the theory which entails a paradoxical outcome, we will see that the foundational question essentially involves a temporal dimension. After arguing that the skeptical challenge gives us a strong reason to adopt a historical view of meaning, I shall further argue against certain authors who claim that meanings not only have histories but futures as well, or that the meaning of a word may change retroactively in time as a consequence of counterfactual change in its future use. The major aim of the paper is thus to bring together the arguably interrelated debates about the skeptical challenge and temporal externalism in philosophy of language.

1 Introduction

The main question of metasemantics, or foundational semantics, is why an expression token has the meaning (semantic value) that it in fact has. In his (1982), Saul Kripke’s reading of Wittgenstein’s later work presented a skeptical challenge that threatened to make the foundational question unanswerable. My first contention in this paper is that the skeptical challenge indeed poses an insoluble paradox, but only for a certain kind of metasemantic theory, against which the challenge effectively works as a reductio ad absurdum argument. My second contention is that as a result of rejecting the theory which entails a paradoxical outcome, we will see that the foundational question essentially involves a temporal dimension.

The idea that meanings, facts grounding meaning facts, and ascriptions of meaning have an important temporal dimension is not new to philosophy of language. The more general purpose of this paper is to work towards linking the debate around temporal externalism with Kripkenstein’s skeptical challenge by gesturing towards certain parallels between the two. Specifically, while I think that the lessons of the challenge provide support for the claim that meanings have histories, at the end of the paper I shall argue against the idea that meanings would have futures as well, which has gained support in recent decades.

This paper is structured as follows. I shall start by examining in detail just what kind of a philosophical theory is the target of the skeptical challenge, in the sense that it effectively works as a successful reductio ad absurdum of the theory. Probably no one has done more work to read the challenge in this way than Kusch (2006). However, thorough as Kusch’s examination of the skeptical challenge is, I shall argue that he misses an important temporal aspect in the theory targeted by the challenge. In particular, I shall argue that the challenge functions as a successful reductio argument against an atemporal theory of meaning. The core of the atemporal view on which I will be focusing asserts that, for any term to be meaningful, it’s meaning must be absolutely determined. Since it will turn out that absolute determinacy is in fact impossible for finite speakers to achieve, in tandem with certain plausible assumptions about meaningful language (such that our actual language is meaningful and not merely illusorily so) it follows that our meanings, facts determining meaning facts, and meaning ascriptions cannot be absolutely determinate.

Second, I shall argue that there is in fact independent support for a positive temporal theory of meaning available in the literature that has sprouted under the heading of “temporal externalism”. Briefly, similarly to how natural kind or social externalism claims that the meaning of an expression as used by a speaker depends in part on factors not internal to her, temporal externalists claim that the expression’s meaning depends not just on the speaker’s (or her community’s) present use, but also on its past and future uses. To say that the present meaning is or can be affected by past use means that meanings have histories. To say that the present meaning is or can be affected by future use is to say that meanings have futures.

A central issue that has become contested in the wake of the skeptical challenge is whether meanings, facts determining meanings, or meaning ascriptions are “normative” in some interesting sense. As we shall see, the idea of normativity has spilled into the recent debate on temporal externalism as well. While I believe the skeptical challenge to convincingly show that meanings must have histories, the normativity of meaning controversy as it presents itself in the temporal externalism debate should be understood as the question whether meanings have futures as well. I will give some reasons to think that this is not the case.

In sum, this paper aims to contribute to two major, arguably interrelated debates in philosophy of language. First, I will argue that Kripkenstein’s skeptical challenge gives us a strong reason to reject an atemporal view of meaning, in particular regarding the way how meanings of expressions are determined. The challenge therefore has direct relevance for the ongoing debate around temporal externalism. Second, I shall criticize certain authors who believe that meanings should be understood not only as having histories, but as having futures as well. Furthermore, many of these authors think that normativity is a crucial factor in explaining how past use can depend on future use for its meaning. Relying on a powerful anti-normativist case made by Stephen Turner, I shall argue that the normativist authors face a problem of explaining what it would mean for the norms to succeed or fail in extending their force from future to past. A plausible story of how that is possible remains at large.

2 From a challenge to a paradox: two kinds of semantic indeterminacy

Natural language is ridden with what we might call “semantic indeterminacy”, thus denoting a certain family resemblance among various, ubiquitous linguistic phenomena. A standard classification should distinguish at least between vagueness, ambiguity, and polysemy, but also between indexicality, anaphora, context-dependence, translation, reference and many other related classes. What is perhaps most notably shared here is a kind of semantic polyvalence of words considered as sound-forms which can usually be detected by formulation of pseudo-contradictory statements:

Ambiguity. “The bank is around the corner, just below the bank.”

Vagueness. “Jane is tall (for a professor) and short (for a basketball player).”

Polysemy. “The suitcase won’t fit all my books, but it will fit my laptop, which has all my books and more besides.”

What is also shared between all these classes of semantic indeterminacy is their “ordinariness”, by which I simply mean their prevalence in the ordinary, non-theoretical discursive exchanges of everyday life. Not only the causes but also the cures of indeterminacy are ordinary, as the fluent ability of the natives both to produce and to reconcile the reefs testifies. How they actually do that has proved to be a highly complicated question to answer. Moreover, philosophical theories that deal with these matters are undoubtedly about a phenomenon that exists independently of such theories.¹

Now, contrast this realm of ordinary semantic indeterminacy with the indeterminacy we encounter in the paradox uncovered by Saul Kripke’s (1982) reading of Ludwig Wittgenstein’s later work. The story, familiar to many, starts with Jones, a beginner in elementary mathematics, who is tasked to learn addition. So far he has apparently (more or less) successfully calculated with numbers less than 57, and is currently posed to give an answer to the problem “58 + 67”. We naturally expect him to give the answer “125”. But then “a bizarre skeptic” enters the scene and demands to know what are the facts determining that Jones, in the context of his prior learning and intentions, should answer “125” and not, say, “5”. That is to ask, what is the fact determining that Jones has been following the addition function instead of an alternative “quaddition” (“ ⊕ ”) function, according to which

$$\begin{gathered} {\text{x }} \oplus {\text{ y }} = {\text{ x }} + {\text{ y}},{\text{ if x}},{\text{ y}} < \;{57} \hfill \\ = \;{5}\;{\text{otherwise}} \hfill \\ \end{gathered}$$

(Kripke, 1982, pp. 8–9).

The main issue I want to emphasize here is that, whatever else we may think about the skeptic’s intervention, it is clear that the kind of semantic indeterminacy she advocates is not to be found on the list of “ordinary semantic indeterminacy” alluded to above. The skeptical challenge is not encountered “in the wild”, with the notable exception of the philosophically attuned speakers. Of course, this is what Kripke himself observes early on, as the remark about the “bizarreness” of the intervention ought to imply. Even more explicitly, Kripke is right to say that Wittgenstein “invented a new form of skepticism” (Kripke, 1982, p. 60, my italics).²

Why is this in a sense obvious point worth raising at all? While it is uncontroversial that the folk do not as a matter of fact worry about indeterminacy in the sense of the skeptical challenge, it is far less clear whether the challenge is something they should worry about, at least supposing they have any philosophical inclinations. In other words, the clear implication of the “non-ordinariness” of skeptical semantic indeterminacy is that the challenge involved is a challenge for a theory about language and meaning. But what is the theory which the skeptical challenge strikes as a paradox? And is the theory “a folk theory” in the sense that ordinary discursive practitioners believe it, implicitly exhibit and rely on it in their linguistic interactions or something along those lines? Only in the case that the theory for which the skeptical challenge is a paradox is a folk theory and not “merely philosophical” is it a problem the folk should “worry about” (in as much as they should worry about philosophy of language at all).

In the following discussion I am guided by the assumption that the theory for which the skeptical challenge poses a paradoxical outcome (namely, the outcome that no word has any determinate meaning) is not a folk theory, whatever that precisely means.³ The theory is rather “merely philosophical” in nature, which is to say it is only believed by some of those folk with inclinations to do philosophy of language. In other words, while the skeptical challenge (i.e. the question raised by the hypothetical skeptic above) can be presented to any generic philosophical theory about language and meaning, it presents a paradox only to some of those theories. The first step in resolving the challenge, then, is to identify the theory which it strikes as a paradox.

3 Semantic temporality, absolute determinacy, and the mediating scheme

Let’s start by examining the kind of theory which Kusch (2006) claims to be the implicit target of the skeptical challenge as a whole. To be specific, the target counts as a “theory” only in a loose sense; in more exact terms it is a collection of interrelated, philosophically attuned ideas about the nature of meaning, facts determining meaning facts, and ascriptions of meaning.⁴ Kusch groups these ideas under seven distinct headings, of which two are of prominent interest here: “Classical Realism” and “Objectivity”. Briefly, Classical Realism states that declarative sentences, including meaning ascriptions such as “Jones means addition by ‘ + ’”, are meaningful in virtue of expressing propositions with truth conditions (Kusch, 2006, p. 10). Moreover, the key feature of semantic facts (i.e. truth conditions of meaning ascriptions) as understood by Classical Realism include their Objectivity. Briefly, Objectivity claims that meaning facts must determine for every logically possible application of an expression whether it is semantically correct or incorrect (Kusch, 2006, p. 9). This criterion for semantic facts is in my view better termed “absolute determinacy” since “objectivity” is a much more general term.

Kush’s tabulation of the theory which the skeptical challenge targets is very faithful to Kripke’s text, and although its details are debatable, I believe it is mostly accurate. (To what extent Kripke’s exposition is faithful to Wittgenstein is not my topic.) However, I think that there is a certain structure to the theory inherent in Kripke’s exposition of the challenge that is not explicitly brought out by Kusch’s commentary. In my proposed interpretation, this structure has three central elements:

• i. The mediating scheme

• ii. Atemporality of meanings and facts grounding meaning facts

• iii. Absolute determinacy of meanings

Perhaps the most important of the elements is the mediating scheme. Importantly, there are two ways to understand the metasemantic basic question raised by the skeptical challenge. On the one hand we can ask, in virtue of what fact does a given linguistic token (say, a predicate expression) denote a certain property and not some other? For example, why does my tokening of the utterance type “That is a table” mean table and not tabair, where “‘tabair’ is anything that is a table not found at the base of the Eiffel Tower, or a chair found there?” (Kripke, 1982, p. 19). On the other hand we can ask, in virtue of what fact does my tokening of “table” as I have used it before denote tables, and not chairs, even if I happen to encounter one at the foot of the Eiffel Tower?

The first way to ask the metasemantic basic question amounts to accepting “the mediating scheme” because it implicitly assumes that in addition to the linguistic item (e.g. a predicate) and a worldly item (e.g. the set of all tables) there must be a third, mediating item in virtue of which a certain tokening of a predicate denotes a certain set, kind or class of objects. The mediating item is required to solve the skeptical challenge of excluding the gerrymandered alternatives among the semantic value that is chosen as paradigmatic, for without it the meaning of the token will be absolutely indeterminate, i.e. it’s uniquely correct application is not determined in any logically possible case. Since this result ultimately generalizes to all expression tokenings, the challenge becomes a paradox.

Why does the other way of asking the metasemantic basic question not fall into the mediating scheme? To answer this we must take into account the other, closely related structural element which in my interpretation forms an important part of the theory targeted by the skeptical challenge. The first way of asking the question amounts to an atemporal understanding of the problem, for in order to answer it it is not in principle necessary to specify in what temporal context the tokening occurs. In other words, in order to explain why Jones’s tokening of “ + ” means addition and not quaddition it is irrelevant whether this is his first, second, or thousandth tokening of “ + ”. The reason why this is irrelevant is that, once it is determined why (in what truth conditions) any tokening of “ + ” in general denotes a unique mathematical function, it is then simply a question of looking at Jones’s actual use to see in which of those cases the truth conditions are fulfilled, in which they are not, to know when he means addition and when something else. (Of course, this only solves the metaphysical or constitutive and not the epistemic side of the challenge, but the former is usually, and rightly, taken as the more important side.)

We can now see the reason for why the second, temporal way of asking the metasemantic basic question avoids the mediating scheme. To begin with, this way amounts to a temporal understanding of the problem because answering it necessarily involves explaining why a given new object should be counted in the “continued” or “extended” extension of the predicate expression such as “is a table”. In short, in this view the particular question that essentially involves a certain temporal context comes before the general question that does not involve any actually existing context. As such, in order to answer the temporal way of asking the metasemantic basic question, it is not necessary to posit a third, mediating item (a fact) which in general relates a given linguistic tokening with a unique semantic value. The reason why no such item is required is that it is not yet determined whether the new object belongs to the continued extension or not. But once it becomes determined whether it belongs to the extension or not, then whatever did determine the matter is all that is needed to answer the metasemantic question in this particular instance. (Of course, any individual case of encountering new objects might fail to be determinately resolved by the existing predicates, so that a new one must be invented.)

Neither the temporal nor the mediating scheme readings are particularly new in the vast literature around the skeptical challenge, though they probably have not been as widely discussed as they ought to be. Already Colin McGinn (1984, p. 174) observed that the skeptical challenge is posed in transtemporal terms, which means that the central question is about the sameness of meaning between two consecutive tokenings of an expression.⁵ David Bloor (1997) also reads Wittgenstein as having understood the skeptical challenge primarily transtemporally. Moreover, Crisping Wright (1980) in his early works on the subject understood later Wittgenstein to be arguing against “Platonism”, which is very close to the mediating scheme as I describe it above. Finally, in an unpublished paper Jussi Haukioja has identified the mediating scheme in a way that has perhaps most clearly affected my understanding of the skeptical challenge and its rendering here.

The main reason for why the temporal understanding of the metasemantic basic problem avoids the mediating scheme is that there is no general answer, or a fact of its own kind, to why two temporally distinct applications of the same sound-form do or do not count as “going on the same way”. The question how and whether a term is to be applied in novel cases is, in this view, always left open logically speaking, although there are of course various psychological and physical limits to how creatures like us can apply expressions. This is the position which Bloor called “meaning finitism”:

According to meaning finitism, we create meaning as we move from case to case. We could take our concepts or rules anywhere, in any direction, and count anything as a new member of an old class, or of the same kind as some existing finite set of past cases. We are not prevented by ‘logic’ or by ‘meanings’ from doing this, if by these words we have in mind something other than the down-to-earth contingencies surrounding each particular act of concept application. (Some interpretive gloss can always be provided to render the step formally consistent.) The real sources of constraint preventing our going anywhere and everywhere, as we move from case to case, are the local circumstances impinging upon us: our instincts, our biological nature, our sense experience, our interactions with other people, our immediate purposes, our training, our anticipation of and response to sanctions, and so on through the gamut of causes, starting with the psychological and ending with the sociological. That is the message of Wittgenstein’s meaning finitism. (Bloor, 1997, p. 19).

At this juncture it might be objected that at least the semantic values of some expressions must be determined in the way of the mediating, atemporal scheme as opposed to the non-mediating, temporal one. That is, there are some restrictions on how words can be applied that are not merely “local”, whatever that precisely means. The ordinary meaning of “ + ” is first to spring to mind. This is where I think the last structural element above comes into consideration.

I already remarked that an important consequence of adopting the mediating scheme is an atemporal understanding of the metasemantic basic question. Another crucial consequence is what I have decided to call “absolute determinacy” of meanings, although as we shall see many other authors have identified something essentially similar by another name. As we already saw, Kusch defines the “Objectivity” of meaning so that the meaning-constituting fact for an expression must determine for any logically possible application of it whether the application would be semantically correct or incorrect, i.e. accord with the meaning or not (Kusch, 2006, p. 9). Elsewhere, Robert Brandom has characterized “Fregean determinateness” of senses as follows:

Fregean senses are required to determine classes of referents whose boundaries are sharp, fixed, and complete. To say that they are sharp is to say that it is impossible for any possible object to fall partially in the class determined by the sense (excluded middle), or both to fall in it and to fall outside it (non-contradiction). To say that the referents are fixed is to say that the boundaries of the class of referents determined by the sense do not change. (Which sense a given sign expresses may change, if the use of the sign changes, but the senses themselves do not change.) To say that the boundaries of the class of referents is complete is to say that the sense determines a partition of the possible candidates: every particular is classified by the sense either as falling under the concept it determines, or as not falling under it (excluded middle). This is Fregean determinateness, or determinateness in the Fregean sense. (Brandom, 2019a, p. 429)

Now, if we understand the key aspect of “meaning” which absolute determinacy relates to as its Fregean sense (i.e. mode of presentation of the referent), it is easy to see that absolute determinacy just is Fregean determinateness.⁶ Perhaps the most important difference between the two notions of determinacy concerns their respective scope. Fregean senses are usually understood as (only) descriptive in nature, whereas the kind of semantic determinacy which the skeptical challenge targets is much more general. A concise formulation of absolute determinacy in this general form comes from Alexander Miller:

In the case of a descriptive expression such as “+,” whatever fact that is proposed as making it the case that “+” means the addition function must be inconsistent with the hypothesis that “+” means some other function, such as quaddition. In the generalized version of the argument, which applies to both descriptive and non-descriptive language, this becomes: whatever fact that is proposed as making it the case that rule R_a is the rule governing Smith’s use of expression E must be inconsistent with the hypothesis that the rule governing his use of E is R_b, where R_a and R_b are such that for some possible use Δ of E, Δ is correct according to R_a but incorrect according to R_b. (Miller, 2010, p. 460)

There are differences in Kusch’s, Brandom’s and Miller’s formulations, but I think that they and others too have identified something essentially similar by a different name as playing a key role in the skeptical challenge.⁷ The essential idea behind absolute determinacy of meanings is that whatever semantic value one prefers for one’s semantic theory, it is a criterion of adequacy for one’s metasemantic theory that it in general assigns unique semantic values to expression tokens. Moreover, the “uniqueness” in question is logical in nature, which means a) that the possible alternative semantic values are only limited by logical possibility and that b) the semantic values must fulfill the laws of classical logic such as excluded middle. Absent such logically unique semantic value assignments, the skeptical challenge results inevitably in the paradox where no token expression has a determinate meaning, supposing that one presents the problem according to the mediating scheme—or so I shall argue below.

4 Refuting absolute determinacy

In this section I shall consider an argument to the effect that absolute determinacy is not true of us and our languages. To be more exact, I will offer my interpretation of Kripkenstein’s argument for why absolute determinacy must be rejected. The rejection of absolute determinacy in turn ultimately in my view entails also the rejection of atemporality of meaning and the mediating scheme, as I shall argue at the end of the section.

To begin with, it is good to get clear about the narrative structure of Kripke’s exposition of the skeptical challenge. Following Kush’s reading, the challenge is designed to result in a paradox for a certain theory, or picture, of interrelated philosophical ideas about meanings, facts grounding meaning facts, and meaning ascriptions. At the heart of the theory, I argued above, are the three structural elements of (i) mediating scheme, (ii) atemporality and (iii) absolute determinacy of meanings. In support of this theory, Kripkenstein considers altogether four different strategies. There is the dispositionalist response (1982, pp. 22–40), the qualia or picture response (1982, pp. 40–50), the primitivist response (1982, pp. 51–52), and finally the Platonist response (1982, pp. 53–54). According to Kusch, what these strategies share is a commitment to explain the truth conditions of ascriptions of meaning sentences (of the form “S means x by ‘y’”). These strategies all come with their individual merits and problems, and the arguments for and against them are many. Here, I shall consider what I believe to be Kripkenstein’s most general and powerful argument against all these strategies, and by that token against the tertiary core structure of the theory targeted by the skeptical challenge.

The most general argument I have in mind is closely related to the problem of finity. In fact, in a sense it just is the problem of finity. To put it briefly, the reason why the problem of finity cannot be solved by these strategies (or any other strategy that relies on the atemporal, mediating scheme) is that the problem is effectively an application of the regression of rules argument made famous by Lewis Carrol (1895). The solution to the problem of finity would effectively have to show that the regression of rules argument is invalid, which I think cannot be done, hence the finity problem remains insoluble for the atemporal, mediating scheme.

To recall, the problem of finity states that the fact that is to mediate (or “ground”) the relation between linguistic items as used by speakers and their semantic values cannot be determined by any actual fact that is true of speakers and their communities, for such facts must by necessity be finite in nature, whereas the mediating fact must by necessity be “infinite” since it must determine a logically unique rule while excluding countless others as governing the speaker (and her community). The only way how the unique rule can be determined is if every logically possible application of it is determined to be either correct or incorrect. While this uniqueness condition applies to all meaningful linguistic items and their meanings, I continue to use addition as the paradigmatic example.

Now, how is the problem of finity an “application” of the regress of rules argument? I think Kripke alludes to this very idea in a footnote where he considers the case of “super-Jones” (my term), i.e. the case where someone in fact could think of (or otherwise mentally “contain”) the complete addition table:

Suppose that I had explicitly thought of all cases of the addition table. How can this help me answer the question '68+ 57'? Well, looking back over my own mental records, I find that I gave myself explicit directions. "If you are ever asked about '68+57', reply '125'!" Can't the skeptic say that these directions, too, are to be interpreted in a non-standard way? (See Remarks on the Foundations of Mathematics, I, §3: “If I know it in advance, what use is this knowledge to me later on? I mean: how do I know what to do with this earlier knowledge when the step is actually taken?"). It would appear that, if finiteness is relevant, it comes more crucially in the fact that "justifications must come to an end somewhere" than in the fact that I think of only finitely many cases of the addition table, even though Wittgenstein stresses both facts. Either fact can be used to develop the skeptical paradox; both are important. (Kripke, 1982, p. 52, fn.)

Here Kripke seems to be saying that there are two ways to understand the problem of finity. The first way focuses on the point that no finite subject is actually capable of thinking of the complete addition table, or even to be in a state which would somehow “contain” the complete addition table. The second way focuses instead on “justifications” and the fact that the chain of justifications must end somewhere. It is the second way which Kripke thinks makes the skeptical challenge problematic even for super-Jones.

In my view these two ways correspond to two “directions” in which we can pose the problem of finity. The first direction is from the finite, particular subject to the infinite, universal addition function. The question here is how can the infinity be contained in any finite state such that it is fully determined at every given moment what mathematical function (if any) Jones is governed by. In contrast, the opposite direction to pose the question goes from the universal to the particular: why is it that the addition function as opposed to the quaddition function governs Jones’ behavior with “ + ”?

The common issue where these two directions cross is what mediates the relation of a unique universal and the particular subject and her behavior. Specifically, is the relation itself “universal” or “particular”, infinite or finite? On the one hand it must be particular and finite, because it is a relation that is partially about Jones (or his community), and Jones is a particular and a finite being. On the other hand the relation must be universal and infinite, for the addition function is both. But it is clear that the relation cannot be both particular and universal, finite and infinite. The point of Kripke’s second way of posing the problem of finity is that, even if we assume infinity on both sides of the mediating relation (super-Jones with the complete addition table and addition function in its Platonic heaven), “when the step is actually taken” and Jones produces a tokening of a solution to an addition problem, the key issue is simply shifted to a different location, namely between super-Jones’s pre-existing, explicit knowledge of the complete addition table and the particular calculation he performs. Does super-Jones also have knowledge about how his knowledge of the addition table is to be applied in this particular case? This step, too, can be gerrymandered with skeptical alternative interpretations. To wit, the skeptic proposes that whereas previously the addition table would give the answer 125 to 58 + 67, at this instance the answer should be 5 instead. Since the skeptic is not logically barred from asking “What is the justification for the correct interpretation of the addition table now as opposed to then?” (or vice versa), the original problem of “going on the same way as before” repeats itself.

I think the form of this argument is very close to the question of the Tortoise, that is, given the premises p and p → q, why should he accept the conclusion q? Well, because if one accepts p and p → q, then one must accept q. But if that claim is understood as a necessary premise for accepting q, then we only end up needing a yet further premise, and a further after that ad infinitum. Similarly, if every case of applying a rule needs to be justified by an interpretation that is itself understood as an application of a rule, one ends up needing an infinite chain of justifications: an interpretation is needed to apply an interpretation, and a premise to apply a set of premises to their conclusion.

In sum, the problem of finity comes with two prongs. The first prong challenges to explain how a finite state can determine an infinite number of correct-incorrect partitions of a term’s application. The second prong allows the logical possibility of super-Jones who meets the infinity requirement, but shifts the problem now between his state containing the complete addition table and the particular token calculation he performs. Here the question becomes: How can an explicit instruction given earlier determine a step that comes later? It seems that only a new explicit instruction can solve the problem, which then repeats itself at a yet later time. Insofar as there is always a later time, the regression of explicit instructions is inevitable.

The crucial thing to realize is that the problem of finity becomes an insoluble paradox only if we implicitly accept the tertiary core structure. The major element of the structure is the mediating scheme, which amounts to a commitment to find a general kind of facts that metaphysically glue together linguistic items and their semantic values. An important entailment of this commitment is the atemporal understanding of the challenge, where for any temporally contextual, actual tokening to be determinate, every logically possible application of it must be. This in turn amounts to commitment to what I have called absolute determinacy, which is where the weight of the challenge becomes a paradox, since no finite state that exists in time can be absolutely determinate without a regress ensuing.

5 Temporal determination of meaning

The major claim that I advanced in the previous sections was that the semantic values of our expressions are not, and cannot be, absolutely determined. This result in turn gives us a reason to believe that the way in which expressions in fact gain their semantic values has a temporal character; it is something that happens not only in but also over time. One important consequence which Bloor drew from his temporal understanding of the challenge is that “For a finitist there is no such thing as the ‘extension’ of a term or concept, or, if the word ‘extension’ is used, it radically changes its significance” (1997, p. 24). That is to say, if there are any so-called “semantic” constraints we finite speakers have on how to use our words and other expressions, those constraints cannot be logically strict in the way Fregean determinateness is. If that was the standard on our semantic constraints, we’d immediately be neck-deep in the paradoxical side of the skeptical challenge. However, although I accept this negative conclusion of the challenge, I am more inclined to side with authors such as Haukioja (2005) and Brandom (2019b) who work towards middle ground between unbounded finitism (which Brandom might call “semantic nihilism”) and absolute determinacy. That is to say, although neither intensions nor extensions are logically strict, philosophy of language can at least attempt to make systematic contributions to answer the metasemantic basic question of how words come to have the meanings they in fact do.

In this section my aim is to connect these thoughts gathered from Kripkenstein’s skeptical challenge to another debate that started around the same time and has developed mostly independently. Following Mark Wilson’s (1982), many authors have defended the idea of temporal externalism, which roughly states that the meaning of a token expression as used by a speaker (and her community) at t₁ may be metaphysically dependent on either an earlier or later use of that expression type in the community. (As we shall see, however, Wilson himself emphasized somewhat different aspects of the thought experiment.)

Wilson’s (1982) thought experiment is about a tribe of islanders called “the Druids”, whose syntax, semantics and vocabulary parallel closely that of pre-modern English. One peculiarity of their ecological and linguistic environment is that their term “bird” has as its extension all the flying things found on the island, which by happenstance contain only species of bird that can fly. One day, the Druids are visited by a modern airplane, the sighting of which in the sky naturally incites the Druids to extend their bird practice to cover it, e.g. by uttering “Lo, a great silver bird in the sky.”⁸

What makes this thought experiment interesting is the alternative account that we can imagine is true of the Druidic encounter with modern aviation technology. Call the situation above where the Druids become acquainted with an airplane by first sighting it in the sky and calling it a “bird” “scenario A”. In “scenario B”, the airplane and its crew crash on a remote location of the island to be found by the Druids some months after the event. In this scenario the Druids are naturally drawn to call the plane “a great silver house” instead of a “bird”. Only later, when their relations with the outsiders have developed to the point where acquaintance with flying airplanes becomes an established fact do the Druids come to realize that some of their “houses” can fly, and thus that this ability is not limited to “birds”.

The gist of the thought experiment is this. Up to the moment where scenarios A and B diverge, the Druidic word “bird” apparently has a determinate extensional meaning in both timelines, i.e. it means something like “flying thing, has feathers etc.”. The “bird” sentences to which the Druids in the pre-airplane situation (call it “the original Druids”) would unanimously assent to include: “Birds can fly”, “All birds have feathers”, “If it flies it has feathers” and so on. Now, in scenario A some of these sentences turn out to be false, for the extension of “bird” henceforth covers flying things without feathers, for example. (Similarly, some sentences about the Druidic term “house” will experience a change in truth-value in scenario B, but for simplicity I will only talk about the case of “bird”.) However, in scenario B the word “bird” retains its original extension, while the later developments of the Druidic “house” will probably have to accommodate a difference between immobile and flying houses. The question now arises: what extension for “bird” should we ascribe to the original Druids? Three possibilities are available:

• 1. Either scenario A or B gives the truth of “bird”. That is, “bird” for the original Druids really means (if scenario A is true) something like “a flying thing” but not e.g. “has feathers”, which property only accidentally coincides with their initial sample set of “birds” from which the true, proper extension is to be projected. This answer effectively claims that whatever is the correct way to extend the Druidic “bird” practice in response to the sighting of an airplane has one true answer available before the empirical fact of airplane encounter takes place, and is independent of how that encounter takes place.

• 2. “Bird” in original Druidic does not have “one true” extension in the above sense, but rather its semantics is to be interpreted in a looser fashion. Wilson names Hartry Field’s (1973) proposal for “partial denotation” as one answer, which he develops by a similar strategy of “implicit parameters”. (The core idea is to enrich extensions from sets to classes of sets that hang together according to some to-be-defined principles, but that is not the topic here.)

• 3. Whatever “bird” in original Druidic really means (what its proper extension assignments are) does not depend only on facts about the Druidic community’s use of the word preceding the encounter with airplanes, but rather may counterfactually change retroactively in time due to future developments in the Druidic community’s use. If the future for original Druids had been different, so would the past.

To anticipate the next section, in my view the right answer is either with (2) or (3), and there are some good reasons to criticize certain proposals rooting for (3). The reason for why (1) should be rejected is that it ultimately amounts to a commitment to absolute determinacy, which I have argued is false.⁹ Plausible as it may seem that the “one true extension” must be hiding somewhere in the original Druidic community, it is I think more plausible to conclude that “bird” has a history, not only as regards its use, but also as regards its meaning. But opting for (1) effectively entails denying that meanings have histories, since all the right answers must already always exist somewhere, be it in the Platonic heaven or the Druidic dispositions understood as rails onto infinity.

Now, there is clearly something similar between Wilson’s insight that many ordinary predicates have much more complicated extensions than often thought and the skeptical challenge that targets absolute determinacy. The common thread is in the notion that the way how extensions are in practice extended to cover new instances is not so logically strict as Frege for instance thought. However, there are also clear differences between the two authors. Wilson (1982) frames the discussion in terms of a radical interpreter who is tasked to produce an extensionally adequate translation manual for the Druids, whereas the skeptical challenge covers all possible semantic values, being a much more general argument for temporal determination of meaning.

Some authors have suggested, independently of the skeptical challenge, that meanings should be understood not only as having histories but as having futures as well. What this means is that the meaning of a term in the present and past depends for its determination not only on past uses but also on future ones. Although I don’t think the idea is completely without merits, the recent arguments in support of it do not seem to hold water, as I shall argue below. As a terminological clarification, the position I argue against can be termed “retroactive” temporal externalism, whereas temporal externalism in the “historical” sense is what I believe simply follows from the rejection of absolute determinacy. Roughly, while historical temporal externalism claims that semantic values are determined over time and are nowhere absolutely determined, retroactive temporal externalism claims that at least some token expressions can counterfactually change their semantic values because of future developments.

6 Against retroactive temporal externalism

In this final section I shall criticize certain positions that opt for the third option of resolving the Druidic example, and which thus accept temporal externalism not just in the sense that meanings have histories, but also in the sense that they have futures. To begin with, we have two rough camps on which I shall focus: normativists and non-normativists (or dispositionalists). Here, Jussi Haukioja is selected as representative of how to make retroactive temporal externalism plausible in a dispositionalist framework. The normativists further divide between two subcamps: Alessandra Tanesini represents what I shall call a “substantial” account of retroactive temporal externalism while Joseph Rouse and Jackman represent an “anaphoric” account of the same claim. I shall criticize all three positions while remaining neutral about what the right answer might be, although my sympathies are with Wilson and the second option.

Let’s start with the non-normative, dispositionalist solution. Haukioja claims that:

Temporal externalism would be true if, for some speaker S and word w, S is disposed to re-evaluate her use of w, in response to information about future use of w in his/her speech community. Roughly, this means that S is disposed to accept and go along with a range of different interpretations of w, and retract or not retract accordingly. Of course, in most cases speakers do not in fact receive the relevant kind of information about future use. What matters is the presence of such dispositions, not that they in fact be manifested. (Haukioja, 2020, p. 925)

The proposal builds on a broader metasemantic theory which Haukioja, together with Daniel Cohnitz, calls “dispositionalist metainternalism” (Cohnitz & Haukioja, 2013). Briefly, metainternalism is a second-order claim about what determines the first-order semantics of a given term, e.g. whether a term like “gold” has internalist or externalist semantics. While Haukioja and Cohnitz accept first-order externalism for many terms, especially natural kind terms and proper names, they argue that the reason why externalism is true for these terms is due to factors internal to the speaker using these terms. What makes their metainternalism dispositionalist is simply the added specification that the relevant facts about the speaker are to be identified (at least partially) with her dispositions to respond to information about e.g. the atomic number of gold as relevant for determining whether her term “gold” has externalist semantics.

I want to raise two problems for metainternalist dispositionalism as a hypothetical defense of retroactive temporal externalism. (Why it’s only hypothetical becomes clear below.) First, there is a principled epistemic problem regarding the counterfactual intervention by which information about future use of w is to be delivered to the speaker S. We may reasonably assume that how exactly S reacts to such information depends on the way it is delivered to her. For example, should John Locke be visited by a naked, time-traveling bodybuilder appearing to him in a sphere of visually impressive electromagnetic charge in the middle of the night, it is doubtful whether he would retain any capacity to react coherently to the visitor’s further attempts to depart information about the use of “gold” in the future to him. My point is, briefly, that the counterfactual intervention would have to be heavily idealized in some way if we were to draw any actual conclusions from it regarding the temporally external status of w for S.

However, the problem of epistemic idealization is not a principled objection to the metaphysical truth of temporal externalism as defended by meta-internalist dispositionalism. My second objection rather targets the heart of Haukioja’s proposal. The claim is that in order to work as a defense of temporal externalism, Haukioja needs to offer some method for distinguishing between two different scenarios that may transpire as a consequence of the counterfactual, idealized intervention. In scenario (1), the intervention merely reveals the already existing (second-order) disposition of S to change (or not to change) her use of w in response to information about its future use. In scenario (2), the intervention creates in S a (second-order) disposition to consider her use of w in response to information about its future use. Obviously, only in scenario (1) has temporal externalism for w as used by S been vindicated. Framed in slightly different terms, then, Haukioja’s challenge is to explain, in principle, how to ensure that the subject in the actual past and in the counterfactual intervention are dispositionally the same, i.e. no relevant change of dispositions is caused by the intervention itself. Without giving a methodological basis for making this distinction, not just any coherent reaction on Locke’s part regarding his use of “gold” in response to the counterfactual intervention would suffice to show the truth of retroactive semantic externalism, for it would remain unclear whether the intervention itself is the cause of Locke’s new disposition, and perhaps also his belief that these were his semantic dispositions all along.

How big of a problem is this challenge to Haukioja’s proposal? This is a difficult question in part because Haukioja does not invest great ambitions in his suggestion: all that it is meant to show is a principled explanation for how temporal externalism could be true for some terms, not a general defense of the idea as such (Haukioja, 2020, p. 929). Thus it is perfectly compatible with Haukioja’s proposal that in fact no w for any S is determinately temporally sensitive in the retroactive sense. Temporal externalism would be merely contingently false then. But it is also possible, perhaps even likely, that there is some w for some S out there for which it is true, or at least could be true. Indeed, it is precisely because epistemically the truth of temporal externalism is so vague that it may be practically impossible to evaluate Haukioja’s proposal any further. The more interesting question must, I believe, address the credibility of meta-internalist dispositionalism as such, though that is a task which I cannot undertake here.

Next, I shall critically examine Tanesini’s (2014) defense of temporal externalism. Her main idea is that the meaning of a word, like the Druidic “bird”, is not determined solely by dispositions to use the word, be it in question their past, present or future dispositions. Rather, what determines the meaning of “bird” for the original Druids are the norms governing the word’s (or concept’s) use, and these norms can be “retroactively instituted” so that future events may influence what meaning a word had in the past (Tanesini, 2014, p. 12). Ascriptions of meaning to past Druidic utterances of “bird” are not to be understood (only or primarily) as descriptions of their semantic dispositions, but rather (also) as retroactive enforcements of the norms common to past and future.

I have three kinds of objections to Tanesini. The first questions the notion of retroactive institution of meaning-constituting norms that plays an essential part in her proposal. She writes:

[T]emporal externalism is consistent with actual linguistic practice. Unless we have evidence to the contrary, we take the words of our ancestors to be governed by the same norms (express the same concepts) as our own, and for good reason. If we did not, any apparent disagreement would be explained as a case of talking across purposes. Since the norms governing the uses of our norms are instituted over time, they must have retroactive effect. (Tanesini, 2014, p. 15)

For one, it is one thing to claim that we often take the past community members to be governed by the same conceptual norms as we are and another to claim that the past community members really are governed by the norms as we enforce them now. Although Tanesini does not well articulate what exactly she means by “institution of norms”, there is at least one necessary criterion of adequacy any such account should meet, which is that the norms may be incorrectly enforced.¹⁰ So, she owes an account explaining what it would be for our retroactive institution or enforcement of conceptual norms to fail to hold the past members accountable to what we take to be the true, or correct, meaning of the word.

The problem here is that what would count as success or failure in a retroactive institution seems highly elusive, seeing that the past cannot actually respond to us. One intuitive notion of success would be where the past community members counterfactually accept our interpretation of the conceptual norm governing a word, with failure corresponding to a counterfactual rejection on their part. However, this interpretation cannot help Tanesini, for it effectively allows two outcomes that correspond to the two scenarios I discussed in the context of Haukioja’s criticism above. In the scenario where the past community has the counterfactual disposition to accept the future interpretation of the conceptual norm governing a word, no new content is really instituted, for it is only shown that the past community members had already accepted the norm as their own, perhaps without realizing it themselves. (Of course, to say that they accept the new meaning as a consequence of some kind of a dialogue with the future can only mean that a new community is created in the counterfactual present, not that one already existed in the actual past.) In the scenario where the past community members have the disposition to reject the new interpretation, the natural conclusion seems to be that the future and past communities have come asunder in regards to the meaning governing a certain word, and that they no longer count as “the same community” in this sense. In sum, unless Tanesini can offer some alternative explanation for how retroactive institution can fail and succeed, and not merely what it is to purport to retroactively institute and enforce conceptual norms, her proposal cannot work as a defense of retroactive temporal externalism.

This brings us to my second and third objections. I think the core problem of Tanesini’s position is that her view of norms as “fully determinate” (2006, p. 200) is actually incoherent with retroactive temporal externalism. To see this, notice first that “full determinateness” just is, in the relevant sense, “absolute determinacy”.¹¹ But it is clear that, if absolute determinacy is true (which it isn’t!), then whatever meaning w has for S or her community at t₁ cannot depend on its future use at t₂ by the consequent community members, simply because the meaning is determined atemporally. Absolutely determined meanings cannot be temporally determined, as I shall argue below.

Before going into my original objections to Tanesini, it is useful to show how I think Rouse’s (2014) criticism of her somewhat misses its mark. First, Rouse notes that there is apparent tension in Tanesini’s commitment to full determinacy of norms and her idea that normative practices are “open-textured”. Briefly, open-texture of practices does not only mean that it is always possible to missapply a fully determinate norm (which more or less immediately follows from full determinacy’s definition), but rather that which norms are in force and have authority for participants in the practice may be settled only by ex post facto institution. At first sight, then, open-textureness seems to be in conflict with the application-determinacy of norms. However, in her response to Rouse, Tanesini clarifies that “Open-texture is in my view not a feature of norms, it is a feature of our expressions that makes it possible for us to acknowledge the authority of future norms on our performances” (2014, p. 17). What I gather from this is that Tanesini distinguishes between the applicability of norms and their being in force, which allows her to hold onto both full determinacy of norms themselves and the open-texture of our attitudes and practices that are about those norms.

Rouse also has another worry about Tanesini’s defense of retroactive temporal externalism, which is that it leaves the door open to radical, foundational skepticism “concerning the legislative authority of our own meaning claims, since they always remain open to retroactive repeal in light of novel experience” (2014, p. 27). In response, Tanesini bites the bullet and admits that in her view, the possibility that “we never actually produced any meanings because our behavior never came to constitute a genuine linguistic practice” cannot be a priori denied (2014, p. 17). Leaving the door open is in her view harmless insofar as there is no further reason to believe that the skeptical scenario is true of us. However, as I shall argue below, Tanesini’s commitment to full (i.e. absolute) determinacy of norms does more than simply leave the door open to the abstract possibility of radical skepticism; it is an open invitation to the paradoxical side of Kripkenstein.

To continue my own objections to Tanesini, I think there is a certain vacillation in the meaning of “retroactive institution” in her use. If we understand institution epistemically, the claim is that the meaning of w at t₁, in order to be fully understood or known, depends on a future interpretation at t₂. In contrast, understood legislatively, the content-constituting norm that is in force for w at t₁ may change due to developments at t₂. I think the vacillation is present early on in Tanesini’s (2014) paper:

If one wants to understand [things with history], one must usually look backward onto them from the point of view of the future. This approach is forced upon one because historical entities can only be made fully intelligible by studying their histories. This study requires that one uncovers the significance of these entities in the context of the overarching projects, or cultures of which they are a part. As a result, the significance of present historical entities often partly depends on the significance of future developments. The relation between present and future (and vice-versa) in these instances is one of normative projection rather than causal prediction. The future casts its normative stamp on the past, as much as the past does on the future. (Tanesini, 2014, p. 2)

The sense in which historical entities like semantic norms are made “intelligible” by a future interpretation corresponds to what I call the epistemic reading of institution. Here Tanesini seems to be saying, however, that the epistemic reading entails the legislative reading, if “significance” corresponds to “meaning”. But this seems wrong: that the future understands a past meaning better than the past itself cannot entail that the past thereby changes, i.e. is dependent in the legislative sense, on the future! (Note that in question is normative, and not causal, counterfactual change.) One way to make this passage intelligible is to say that, in the view of future community members, if the past had known what the future knows, the past would have accepted the meaning which the future does. (For example, if the past community had known certain biological facts about whales, they would not have called them “fish”, at least in some ideal conditions.) But, again, this at most amounts to an elaboration of what it means for the future, together with the past, to take the past to be governed by the same meanings as the future, not that the past actually is so governed. And no amount of understanding or knowledge on the future’s part can as such make the past meaning dependent on future meanings in the legislative sense. Since this point recurs in my criticism of Rouse, I shall elaborate on it more later.

To clarify, I think Tanesini’s position suffers from three major problems:

• 1. She is committed to absolute determinacy of meaning and must thus show how it is possible for finite subjects.

• 2. She is committed to a retroactive institution of meaning via ex post facto legislation of norms and must show what it would take for the legislative purport to succeed or fail in general.

• 3. Even if these two problems are somehow solved, all they lead to is an incoherent pair of commitments. To show this, suppose that retroactive institution in the sense of legislation is possible. So, whatever (absolutely determinate) meaning w has for S and her community at t₁ depends in part on what that meaning is retroactively legislated to be at t₂. But, from the perspective of a yet more distant moment t₃, to which both t₁ and t₂ are past, the meaning of w at t₁ has not changed at any point in time. Since for t₃, as for any other moment in general, there can only be one actual past, it is impossible that the actual past could have changed at any point. Having “two pasts” in the relevant sense would mean identifying a moment t_n such that w is governed by two different absolutely determined meanings, which is a violation of the definition of absolute determinacy. From t₃’s perspective, w at no point in time was governed by any other meaning than what was legislated at t₂, and this was true even before t₂ actually happened, i.e. at t₁. But if there is no change in w’s meaning between t₁ and t₂, how can anything have been legislated in between?

What the argument (3) effectively shows is that the legislative sense of institution can only be apparent if meanings are absolutely determined. That is, there is no such thing as successful or failed retroactive legislation. There may be attempts (purport) to do such legislation, but both its success and failure must logically remain indeterminate, for otherwise the result would lead to an incoherence.¹²

Summed up, what the truth of retroactive temporal externalism necessarily demands is dependency of past meaning on future use. That is a temporal semantic relation: it entails that meanings themselves (and not merely our agreed upon understanding, interpretation, or knowledge of them) can counterfactually change backwards in time. In contrast, absolute determinacy necessarily demands an atemporal semantic relation such that the uniquely correct applicability of w (i.e. its meaning) is decided for every logically possible application of it irrespective of time. These two commitments cannot be combined without incoherence, supposing that a term’s applicability is partly dependent on what norm constitutes its meaning.

However, as I mentioned above in discussing Rouse’s criticism of Tanesini, she appears to make a distinction between the applicability of a norm (which is determinate) and its being in force (which is open-textured). Assuming that this reading is correct, Tanesini can avoid my incoherence objection. Yet she would still face the problem of finity as well as the problem of retroactive institution’s success and failure that builds on the Wittgensteinian criterion, and which I further develop below.

At this juncture we might observe an interesting similarity in Haukioja’s and Tanesini’s defense of retroactive temporal externalism. What retroactive temporal externalism requires is some kind of a notion of continuity between past and future linguistic practices which blocks option (2) of interpreting the Druids, i.e. the claim that the meanings are simply subtly different, or rather, their “sameness” is more fuzzy than often assumed, as Wilson argues at length. If we understand temporal externalism specifically as the claim that meanings not only have a past but a future as well, the general problem becomes how to explain how to “go on the same way” with the word w. What drives the solutions of Haukioja and Tanesini is their belief that sameness of meaning across time is to be explained by sameness of content. For Haukioja, this means dispositional content, i.e. w has temporally sensitive semantics if the past speaker S has the second-order disposition to change her use in response to information about future use to match the future speakers’ dispositions. For Tanesini, the sameness of meaning is explained by there being a single true content, absolutely determined, that is (legislated to be) in force for the speakers of the same community across time.

The final temporal externalist author I shall discuss here opts for a markedly different approach to cross-temporal continuity. The basic proposal by Rouse (2014, 2016) follows Robert Brandom (2002, 2014) in understanding the temporal continuity of linguistic community as governed by (something similar to) anaphoric relations, which explain how “the same content” can be relayed from speaker to speaker without specifications about what that content is.

To begin examining Rouse’s suggestion in detail, I think it’s illustrative to start by seeing how anaphora functions in ordinary linguistic interactions, which is I think best displayed by an example of how it malfunctions:

A wife asks her husband, “Could you please go shopping for me and buy one carton of milk, and if they have avocados, get 6.”

A short time later the husband comes back with 6 cartons of milk.

The wife asks him, “Why did you buy 6 cartons of milk?”

He replies, “They had avocados.”¹³

To use Rouse’s phrases, there is something “at issue” between the husband and the wife, and something to which they have a shared “stakes”. What is not shared between them is the content of the cardinal numeral expression “6”; instead, each has an “anaphoric commitment” to “6” referring to the same thing in both of their uses in the context. By analogy, Rouse thinks, something similar (if not identical) applies to virtually every aspect of our ongoing, temporally extended linguistic practices such as those exhibited by the Druids (Rouse, 2014, p. 31). The question of continuity between the original Druids and the later diverging scenarios is not whether something remains the same at the level of content, be it understood dispositionally or normatively, but whether the Druids are committed to their words being anaphorically cross-temporally linked so that future uses are authoritative in determining the meaning of past ones. While the force of authority in question is retroactive, it is not based on shared content, but rather:

A meaning-claim [ascription of meaning] gains its normative authority now from the difference it would make to subsequent practice to respect and enforce the norms it expresses. Of course, different speakers propose to use and understand words in different ways (often simply by so using them). What makes some correct and some incorrect is their retroactively applicable significance for subsequent discursive practice. (Rouse, 2014, p. 35)

Rouse’s theory of discursive, social practices, especially scientific practices, is sophisticated and in my view has many merits. My criticism aims first of all to show that one can accept a lot of this theory without having to commit oneself to retroactive temporal externalism. In particular, I think one can preserve a lot of the anaphoric normativist content of the theory without having to wrestle with the arguably unintuitive conclusion that past meanings depend on future use.

To begin with, let’s switch our paradigm example from the Druids to one from Wittgenstein’s Investigations:

Someone says to me: "Shew the children a game." I teach them gaming with dice, and the other says "I didn't mean that sort of game." Must the exclusion of the game with dice have come before his mind when he gave me the order? (Wittgenstein 1958, Sect. 70)

As with the Druids, we seem to have three basic answers available here. (1), what the subject issuing the order (call her “A”, the other “B”) meant by “game” at the time she gave the order was already determined in the sense that it excluded the game with dice. (2), what A meant at the time of the utterance was not so determined, yet once the issue arises with B, they find themselves with conflicting views about (i) what is at issue with “game” and possibly also (ii) what is at stake with the use of the word in the context. To elaborate, perhaps A might agree that a game with dice is within the extension of “game”, but think that what she rather meant at the time of the utterance was something like “game fit for children”. B might then retort that this way of explicating what is at issue is agreeable to him, but add that in his view, a game with dice is perfectly in accordance with the stake of shewing a game fit for children. This sort of dialogue can go on indefinitely and result either in agreement or disagreement in perpetuity. (3), finally, what A meant by “game” at the time of the utterance (what the stakes and issues were) depends on future developments between A and B in the counterfactual sense.

If I understand Rouse correctly, he would say that (3) is the right option, and he would defend it by saying that A’s utterance in the past must be understood as anaphorically accountable to the future interactions involving her and B. The reason why the temporal relation must be understood in this way is that what it is to be a word must be understood as an iterable, historical and anaphoric unit of semantic information (Rouse, 2016, p. 34). Moreover, Rouse also clearly endorses a retroactive version of temporal externalism, in which past meanings themselves are somehow supposed to depend on future use:

One source of resistance to temporal externalism may dissolve with this recognition that the retroactive determination of semantic content reflects the accommodation of the world's variety and recalcitrance within ongoing and continuous linguistic practices. We may bridle at the thought that subsequent choices by other speakers in response to new circumstances partially determine what I mean now. It is not their choices that are authoritative over my linguistic performances, however, but the stakes I share with them in linguistic articulation and communication. Subsequent developments and discoveries in linguistic and other practices may place those stakes in a new and previously unforeseeable light, but later responses to such developments are subject to comparable normative constraint. (2014, p. 37)

My main objection to Rouse is the same which I already posed to Tanesini, namely that there is a certain vacillation between (a) an epistemic thesis that the meaning of past utterances (or the stakes and issues of earlier controversies) may only become known in the future and (b) the metaphysical thesis that the determination of past meanings themselves is dependent on future use in the counterfactual sense that if future events had transpired differently, the past would have been different as well. It is important to realize that the epistemic thesis is perfectly agreeable to option (2), which amounts to historical temporal externalism as I have used the term here, and that it takes the stronger, non-epistemic, counterfactual notion of dependence to support retroactive temporal externalism. My claim is that Rouse does not actually give support to the stronger thesis, although he clearly means to defend it.

My reason for this claim comes in the shape of Wittgenstein’s criterion of adequacy for genuine norms on which we must now focus. Returning to the game example, it is one thing to say that A and B must necessarily understand their use of the word “game” as temporally extended and another to say that such understanding would amount to the past really being counterfactually dependent on the future for its meaning, stakes and issues. As Rouse says, what the stakes and issues really are is independent of their expression by A and B (2014, p. 29). So, in what sense do past meanings counterfactually depend on future usage if not through the actual stakes and issues? In the quote above, Rouse says that “Subsequent developments and discoveries in linguistic and other practices may place those stakes in a new and previously unforeseeable light,” which seems like an epistemic metaphor.

Rouse’s answer to what counterfactual, retroactive dependence consists in, I believe, has to do with his notion of anaphoric accountability, which serves the same theoretical role as ex post facto semantic legislation for Tanesini. Here is where the Wittgensteinian criterion comes into the picture. In order to distinguish the stronger metaphysical claim that supports retroactive temporal externalism in the sense of the past being counterfactually dependent on the future for its determination, Tanesini and Rouse offer alternative accounts for what it is for the future to normatively change the past. For Tanesini, this means ex post facto semantic legislation, which depends on full determinacy, whereas for Rouse:

Instead of identifying a practice by the accountability of all of its performances to a specifiable norm (which may not yet have been definitely settled or fully specified), we can identify the practice by the mutual normative accountability of its performances to one another. For example, the historical continuity of the Druid language provides the setting for understanding its performative instances as mutually accountable to one another. Whatever else Druid-speakers are doing when they utter 'ave' [“bird”], they understand one another as uttering a word that has been used before, and can be used again, in ways that are intelligible in relation to one another. (2014, p. 28)

Rouse and Tanesini differ, then, in whether normative continuity is understood to hold between the subjects and a fully determinate norm (Tanesini) or only between the subjects themselves (Rouse). In both cases the normative relation is supposed to cash in the metaphysical claim about the past being counterfactually dependent on the future for its determination. The crucial issue, however, is that if the future can really counterfactually change the past (in the normative, not causal, sense), then there must be some way for the future to fail in its attempt to change the past. What the future takes to be correct (perhaps with the past’s counterfactual consent) cannot determine what is correct if “correctness” is to amount to real normative change. The sense of “real” here is close to “mind-independent”, i.e. it means change that is independent from what potentially everyone (in the past or future) understands, interprets, agrees or otherwise takes to be a successful change.

To showcase, let’s return to the game example. For historical temporal externalism, we have two options here: either the exclusion of the meaning “game with dice” was determined at the time of A’s utterance or then it wasn’t, under some suitable definition of “determined” (which of course discounts absolute determinacy).¹⁴ Whatever A and B make of the issue later can at most epistemically clarify the truth of the matter, not change it as such, even if their actual memories of the event change as a result. In contrast, for retroactive temporal externalism the past meaning of A’s utterance itself is at stake, depending on what happens in the future. To support this latter alternative, one must show some standard against which the success of attempting to change the past is predicated (supposing that the change is due to A and B’s actions in a broad sense). For in the absence of such a standard that is independent of any and all subjects’ attitudes, it is difficult to see how the past could really be counterfactually dependent on the future, where “real” is understood in the sense of mind-independence that also covers such things as understanding-independence, agreement-independence, interpretation-independence etc.

I take it that we have sufficient reason to believe why any account of retroactive temporal externalism that somehow relies on normativity is tasked to answer the Wittgensteinian criterion of adequacy for genuine normative change. But what reason do we have to believe that meeting that criterion would be hard? To end this paper, I shall consider some reasons for why it is in fact hard that are, persuasively in my mind, made by Stephen Turner (2010, 2016). To specify, here I cannot mount an original defense of Turner’s arguments as such since the topic is vast. Instead, my modest aim is to show the relevance of Turner’s arguments to the problem of retroactive normative change described above, and how his “anti-normativist” position is able to avoid it.

To summarize my main point above, option (3) of interpreting the game example differs from (2) in that, once we have arrived to an adequate description and explanation of what happens in the self-reflective meaning negotiation between A and B, according to (2) we have said everything that needs to be said to understand the case. Option (3) essentially stands for the added question of whether anaphoric accountability or ex post facto meaning legislation was really successful in this instance, and supporting (3) also requires showing what would count as a standard of success here. This way of putting the issue effectively recapitulates on one of Turner’s main critical anti-normativist strategies. For what Turner argues is that, once we have a description of what happens in the self-reflective negotiation of meaning, where both A and B purport to identify how they should go on the same way (what the stakes and issues are), it adds nothing to the actual explanation of what is going on to say that what A and B achieved was a success or a failure in the normative sense. Or rather, what the normative notion of success or failure adds is a further level of justification for what it is that A and B have achieved during their exchange as they understand it themselves. The challenge by Turner is to offer an explanation for why the theorist describing the exchange would herself have to endorse the normative judgment as a theorist tasked to explain what is going on between A and B, i.e. apply normative vocabulary in her account.

Analogously to Turner, my objection to Rouse effectively asks what extra-value for explanation does it add to shift from option (2) to (3) in either the Druids’ or Wittgenstein’s example. After all, as far as (2) is concerned, what A really meant by “game” at the time of the utterance could well be indeterminate, not just at the level of semantic content, but also at the anaphoric level that she did not intend (implicitly or explicitly) to hold either herself or B accountable for the correct use of the word “game” in the context. It is only when she finds an unexpected, unfavorable outcome to unfold that A decides to interfere with an ex post facto meaning-claim as to what she “already always” meant in the past. A invents a miniature local history, as it were, where she always already had stakes in her command to B, for which reason B must now join forces with her to figure out what is really at issue in their exchange.

The question then becomes, do we need to understand the stakes and the issues as something transcendental to discursive practices as such, a condition of possibility without which there would not be discursive practices at all, or should we rather understand the quibbles about “how to go on the same way” as a prevalent yet ultimately contingent feature of our practices. Indeed, the latter option appears to be what Rouse himself actually has in mind:

The normative authority of scientific practices, concepts, and claims only emerges within an historically and biologically specific context, such that maintaining that authority requires also sustaining the way of life within which those practices, concepts, and claims could be authoritative for us. Recognizing the contingency of scientific practices and norms does not undercut their authority, I shall argue, but instead intensifies the significance of what is at stake in sustaining a scientific way of life. (2016, p. 40, fn.)

What Rouse claims here is that the (scientific, but the idea generalizes) authority of conceptual, semantic norms is historically contingent and precisely for that reason their force requires active uptake by the practitioners. The “transcendent conceptions of the normative authority” are helpful, he says, in maintaining the discursive, rational practices from within those very practices (Ibid.). So the claim to transcendental accountability itself could be understood in this sense as a rhetorical move within the self-reflective practices, and as such a proper object of sociological explanation – not a feature of its explanans! Really, the only thing that seems to separate Rouse from Turner here is that, for Turner, we do not need to add the transcendental rhetorics to our description of self-reflective, anaphoric temporally extended practices to make them better as explanations.¹⁵

Be that as it may, should Rouse wish to keep defending retroactive temporal externalism as a metaphysical thesis about meaning, he needs to give an explanation not only of retroactive anaphoric accountability’s purport, but also what it would mean to succeed or fail to uptake the purport in general. In this effort he is surely helped by the work of Brandom, which is more explicitly oriented to give such an explanation (2002; 2014; 2019a). Embracing Brandom’s three judges model of crosstemporal mutual recognition is not a free commitment for a self-avowed naturalist like Rouse, however, as Brandom arguably has a strong sui generis view of semantic normativity. Whether such a heavy-duty view of discursive normativity can or should be accommodated into the naturalist, empirical framework in philosophy of language remains to be seen.

Now, there is one counter-objection that I will consider against Turner’s criticism of normativism and my particular application of it to Rouse. Turner sometimes lets his polemic drive get the best of him (notoriously, he calls normativist theories “good bad theories”, likening them to animistic beliefs), and my dismissal of Rouse’s position as “transcendental rhetorics'' risks the same faux pas. For what Rouse rather has in mind is that the self-reflective perspective in which we identify ourselves as a part of an ongoing historical, conceptual tradition, as both givers and askers of reasons as Brandom says, is somehow necessary to the practice of giving and asking for reasons. In a recent paper, Henry Jackman (2020) has made the point perspicuously. His defense of retroactive temporal externalism is comparable to Rouse in that both accept a) a normativist understanding of use, b) the idea that use determines meanings, and c) that there is transtemporal continuity, not only from past to present, but from future to past as well. “Continuity” means that we both read our own meanings as already always having governed the past usage and also accept that future usage may change our own current meanings. Although meanings are not absolutely determinate in the “external”, theoretical or descriptive sense, Jackman thinks that from the “internal”, practical perspective, there is a sense in which absolute determinacy “must” be true of us. A common example here is the determinacy of content in the common law tradition, where new judgements are based on precedents deemed as relevant. While it is theoretically, descriptively implausible that the past judges would have had legislative intentions with contents so determinate that they could settle any future case, the practical commitment to find such an interpretation in the precedents is one important factor in making common law work at all.

Turner should have no problem admitting that this description of how e.g. the common law tradition actually operates is accurate, and that similar examples can be found in case of many developments of scientific and ordinary concepts. The question that remains, however, is that such descriptions of case examples only give us an idea of what it means for the practitioners to purport to extend their commitments (be their anaphoric or content-based) to past and future, not what it would actually mean to succeed or fail in such purports. To be sure, often enough the practice itself includes some standards of success. But insofar as those standards are internal to the practice, the question merely repeats itself, since arguably the subjects could be wrong about the standards too. Aside from polemical rhetoric, one of the core critical questions that Turner asks is why it is necessary for the theorist to accept the internal standard on herself in order to intelligibly explain how the practice operates. Of course, there is the practice of giving and asking for reasons in which the theorist herself partakes, and “outside” which she cannot step, if by that one means stopping to hold herself accountable to various anaphoric and semantic commitments. But that is hardly evidence that the norms “binding” (in force, true of etc.) the theorist are genuine in the sense of succeeding or failing to be binding independently of what anyone or everyone actually takes for the norms to be binding.

7 Conclusions

This paper aimed to contribute to two vast, ongoing and in my view strongly interrelated debates. The first debate concerned the temporality of meaning, namely whether determination of meaning should be understood as occurring not only in time but also over time. Kripkenstein’s skeptical challenge pushes us towards the view that meaning determination cannot be absolute for finite speakers, which more or less directly entails that it must be temporal in some sense. The sense in which it is temporal, I think, first of all means that meanings have histories.

The second debate picks up from where the first seems to terminate, namely whether meanings should be understood as having futures as well as histories. Especially the authors who believe that there is an important connection between meaning, use, and normativity argue that meaning determination occurs in both directions in time. I argued, first, that if meaning determination is temporal, it cannot be (contra Tanesini) absolute. Second, I argued that normativists who reject absolute determinacy must explain what it would be for the cross-temporal purport of meaning determination via anaphora to succeed or fail respectively, and following Turner, why should we take the normative purport to be an essential aspect of all our discursive practices to begin with.