Radical contextualists as Searle and Travis hold a general underdeterminacy claim according to which most non-indexical sentences in natural language are such that their tokens can have different truth-conditions. Moreover, these radical views avoid assuming that underdeterminacy is restricted to natural language (as opposed to mental representations). Because of that, they do not identify the content of an utterance with a structured proposition whose truth-conditions are independent of a context of use. Here I will put forward an alternative notion of utterance content that fits these views: Austinian propositions involving a lekton and an activity. I will argue that having Austinian propositions with different granularities allows us to have both contents that are closely tied to the context of use and contents that can be shared across contexts.
During the last decades, contextualist views have generalized the role of context in
the determination of the truth-conditions of our utterances. In this paper, I will focus
on radical contextualism1 in Searle [
Here is an oft-discussed Travis case (I slightly modify Travis’s [
This example motivates a principle of underdeterminacy (I will call it TypeUnderdeterminacy):
Type-Underdeterminacy: A non-indexical sentence S is typeunderdetermined if and only if there are tokens of S that have distinct truthconditions.
In the previous example, the meaning of the sentence uttered by Pia is adjusted so as to fit the context of use. In the first scenario, ‘green’ is understood as being satis fied by leaves that are superficially green. In the second, what matters is that the leaves are naturally green, or green by the effect of chlorophyll. What is said by the utterance depends on what is at stake at the context of use. In this sense, one can talk (as Searle and Travis suggest) of the sentence’s truth-value being relative to a background of implicit assumptions or to the character of the occasion.
Advocates of the radical view argue that what goes for ‘The leaves are green’ goes for most sentences in natural language. The justification for this claim is provided by the variety of examples that have been put forward, including more specific versions of ‘The leaves are green’ such as ‘The leaves are painted green’ (if the leaves have been painted by using yellow and blue dots, does that count as ‘green’?) or ‘The leaves are superficially green’ (How much of surface needs to be painted?). If Travis cases can be created for these sentences as well, then complexifying the sentence, i.e., adding more linguistic material so as to make everything explicit, does not seem to eliminate context-dependence.
Searle is especially clear about this: ‘our examples suggest that the assumptions
are not specifiable as part of the semantic content of the sentence, or as presupposi
tions of the applicability of that semantic content, for at least two reasons. First, they
are not fixed and definite in number and content; we would never know when to stop
in our specifications. And second, each specification of an assumption tends to bring
2 This claim has been disputed. I will not address this issue here.
in other assumptions, those that determine the applicability of the literal meaning of
the sentence used in the specification.’ (Searle, [
This possibility of iterating Travis cases, or of creating higher-order Travis cases, motivates Searle’s and Travis’s radical views, as opposed to more moderate contextu alist positions. According to more moderate approaches, the list of context-dependent expressions needs to be extended in order to include, for example, colour terms. By contrast, advocates of the radical views take this strategy to be insufficient. According to their views, we cannot articulate (encode in natural language) fully determinate contents, that is, contents whose truth-conditions cannot vary across contexts of use 3. If that is so, then Type-Underdeterminacy is a general feature of natural language.
One could argue that, whereas ‘is green’ does not express a constant property across contexts, there is nonetheless a context-independent property expressed in each context. Underdeterminacy would thus concern the relation between the meaning of the predicate ‘is green’ and the property expressed by this predicate at a context of use. Given the possibility of creating Travis cases for many sentence, the property should be conceived as non-encodable in natural language. If this option works, then Type-Underdeterminacy can be made compatible with a fairly standard approach to context-sensitivity.
As it is standardly understood following the model of indexicals, context-dependence motivates the distinction of two levels of content. On the one hand, there is lin guistic meaning, the kind of content that expressions have qua types, independently of the specifics of the context in which they are used. In the Kaplanian framework, the linguistic meaning is the character of the expression. On the other hand, we have the content that an expression expresses in an occasion of use. In the Kaplanian framework, the term ‘content’ is reserved for this level. The content, in this second sense, of a declarative sentence in an occasion of use is usually referred to as ‘the proposition expressed’, or ‘what is said’. Perhaps in Travis cases which proposition has been expressed is not determined by linguistic meaning, but the basic approach to content could be maintained.
However, it is not clear that this is the best way to model utterance content. One
reason is that there are doubts that the non-encodable properties one would need to
posit are playing any role in our cognitive lives. This is the case if
Type-Underdeterminacy is not restricted to natural language sentences. Searle [
If these authors are right, then there are no reasons to model mental representations as structured propositions with non-encodable properties that escape Type-Underdeterminacy, for these properties do not correspond to our concepts. And if that is so, and assuming that are no additional reasons for accepting the existence of proper ties that do not behave as GREEN, then there are no reasons to model the content of our speech acts as involving structured propositions with non-encodable properties. Instead, one can go for a relativist framework in which a shift in extension as the ones motivated by Travis cases is not necessarily equivalent to there being two different properties4. In these frameworks, the truth-value of the proposition expressed is relative to the circumstance of evaluation.
Let me pause on a different (and equally common) notion of proposition. So far I have considered structured propositions. There is another notion of proposition, namely a set of possible worlds. Advocates of the radical view need not reject this no tion of proposition. Although they reject the claim that truth-conditions are determined by linguistic meaning alone, they still take it that utterance content is truth-con ditional content, and this can be conceived as a set of possible worlds 5 or as a function from possible worlds to truth-values.
I think, however, that there can be cases in which the context of utterance only de termines a partial function from possible worlds to truth-values. When, in the first scenario depicted by Travis, Pia utters ‘The leaves are green’ it is plausible to model her utterance as being true of leaves that have been covered with green paint. How ever, it might be indeterminate whether a leaf that has been painted with blue and yel low dots (a leaf that only looks green at a certain distance) counts as ‘green’. By con trast, in other contexts this might be determinate (for example in context where there is no green paint and the interlocutors have decided that they will paint very small blue and yellow dots instead). This motivates what I will call Token-Underdeterminacy:
Token-Underdeterminacy: A token of a sentence S is tokenunderdetermined if and only if for some possible states of affairs its truth-value is indeterminate (i.e., if and only if it determines a partial function from possible worlds to truth-values).
In what follows I will assume that advocates of radical contextualism as Searle
and Travis are right in their general Type-Underdeterminacy claim and explore what
4 See MacFarlane [
Recanati [
The grounds for the distinction lekton-Austinian proposition is given by two principles:
[Duality] To get a truth-value, we need a circumstance of evaluation as well as a content to evaluate. (As Austin puts it, ‘It takes two to make a truth’.)
[Distribution] The determinants of truth-value distribute over the
two basic components truth-evaluation involves : content and circumstance.
That is, a determinant of truth-value, e.g. a time, is either given as an ingredi
ent of content or as an aspect of the circumstance of evaluation. ([
Duality is a well-accepted principle. It simply states the thesis that truth-value de
pends both on content and on how things are. Whether a given sentence is true
depends on what the sentence means together with the state of the world. Distribution
introduces a different idea: some of the elements that are needed to fix the truth-value
of an utterance need not be given by the articulated content of the sentence—they can
be given by the circumstance6.
6 Recanati follows here Perry [
The radical views provide another argument for the distinction lekton-Austinian proposition. They show that truth-conditions depend on something that cannot be articulated in the sentence. Why? Because there are reasons to think that, for most sen tences, a Travis case can be generated—for ‘The leaves are painted green’, ‘The leaves are naturally green’, and so on. The reason is not that the sentence would be too complex for us to articulate, but that all the terms involved can be understood in different ways. Thus, the possibility of iterating Travis cases forces a reading of the Distribution principle that extends the need for Austinian propositions, for they suggest that no sentence is fully articulated (truth-conditions are always relative to some implicit assumptions). As a result, the radical views motivate a stronger reading of the Distribution principle:
[Distribution*] The determinants of truth-value distribute over the two basic com ponents truth-evaluation involves : content and circumstance. Not all determinants can be fully articulated in the sentence. Some must be provided by the circumstance.
The distinction lekton-Austinian proposition can be easily adapted to Travis cases. Using Travis’s terminology, we could model Austianian propositions as <lekton, occasion> pairs. The question now is: what is an ‘occasion’? Or, better, what is the occasion adding to the lekton so that to make the satisfaction conditions of the predicate shift? How should we think of this second element?
Corazza and Dokic [
According to Corazza and Dokic, John and Jane utterance have different
truth-values because they concern different situation. On their account, the truth-conditions of
an utterance are given by relative T-sentences such as ‘An utterance u of “There’s
some beer in the fridge” is true iff there’s some beer in the fridge in the situation of u.’
[
Hence, occasion-sensitivity shows that, beyond including objects, places and times, situations must be conceived as involving linguistic and extralinguistic activi ties— perhaps more in line with Wittgenstein’s language games than with Barwise and Perry’s situations. Thus, we can take the salient variation in Pia’s example to be due to the activity in which the act of utterance is embedded. In the first context, Pia is painting leaves so as to match the colour they have in spring (or for decorative pur poses, let’s say), whereas the botanist is doing scientific research. The content of her utterances can be modeled as an Austinian proposition including a lekton (the mean ing of ‘The leaves are green’) and an activity (decoration, scientific research).
I am using the notion of activity here because it seems like a natural way to capture the idea that what counts as green on an occasion depends on what is going on in that occasion. Alternatively, we could talk about the purposes of the conversation. I take purposes and activities here to be roughly equivalent: if the conversation con cerns scientific research we can also say that its purpose is to do scientific research, and vice versa. In both descriptions, what makes it the case that painted leaves do not satisfy an utterance of ‘The leaves are green’ is that science is about natural proper ties.
Cappelen and Lepore [
According to contextualists, what is said in an utterance is the result of contextual adjustment. The point that Cappelen and Lepore find problematic is that this adjustment seems to be, according to contextualists, highly specific. The reason is that every small detail of a conversation might be relevant to how one understands words. Two people who have been trained in different painting traditions can understand differently an utterance of ‘The leaf is green’ when they are painting leaves—there can be a leaf that counts as ‘painted green’ for one of them but not the other. If content needs to be highly specific, then activities, as they figure in Austinian propositions, should be very fine-grained.
However, it should be possible to share content across contexts. Modeling activi ties too fine-grained makes it difficult. In particular, it makes most speech reports strictly speaking false. Imagine that I report Pia’s utterance by saying: ‘Pia said that the leaves are green. She is painting them’. In most circumstances, this report is enough for us to grasp what Pia said. We have a broad understanding of what painting consists in. Although there are different ways of painting things we know that, roughly speaking, painting consisting in applying paint to the surface so as to hide the original colour. However, if every small detail of the context matters, we should model Pia’s original utterance as involving a more fine-grained activity than the one conveyed by ‘She has painted them’.
As a consequence we need to find a middle course between contents that are as context-bound as Travis cases seem to call for but that cannot be shared across contexts, and contents that can be shared across contexts but that are too broad to respect the intuition that any detail might matter.
Corazza and Dokic’s situated minimalism seems follow that middle course. Ac
cording to them, propositional truth in cases similar to the ones I’ve been considering
should be relativized to the situation of use. This explains the variation in truth-value
and the intuition that any detail of the context might matter to what is said. But on the
other hand, Corazza and Dokic are minimalists when it comes to what an utterance
says, and claim that ‘Two people using the same alleged underdetermined sentence
can be characterized, pace contextualism, as having said the same thing even if they
are not co-situated.’ [
However, same-saying at least sometimes requires something else than sameness
of linguistic meaning. Some philosophers have argued that whether two utterances
share content is something that needs to be decided by taking into account not only
linguistic meaning (lekton) but also some features of the context of use. Wieland [
The problem with Corazza and Dokic’s account of same-saying is that a homophonic speech report, as ‘Jane said that there’s a beer in the fridge’ is false whenever the report context is relevantly different from the original context. And it is so because the sentence ‘There’s a beer in the fridge’ does not say the same in both occasions.
My solution to the tension will involve Austinian propositions with different granularities. Before that, let me note that conceiving Austinian propositions as including activity types, instead of particular situations, already allows us to explain some cases of same-saying without the need to identify same-saying with uttering the same sen tence. Imagine a teacher teaching geography to some children. In order to make it easier for them to identify countries in a map, he tells them: ‘Look, Italy is the one that looks like a boot. And France is hexagonal. You see?’. Her utterance concerns the activity of comparing countries with objects, let’s say. As it happens, this teacher uses the same example every year, as do many other teachers. She says the same year after year. If we model her utterance as including the activity type ‘comparing countries with objects’, we can account for that.
Instead of particular situations, we can think of Austinian propositions as includ ing situation-types. Thus, two utterances of the same sentence that refer to two differ ent particular situations can express the same Austinian propositions, because of the particular situations being of the same situation type. The content of an utterance, then, will be conceived as a pair <lekton, activity-type>. Two different activity-tokens can correspond to the same activity-type.
Despite this, I admit that a content that is closely tied to the context of use is difficult to share across contexts. If we model the content of Pia’s first utterance as including a very fine-grained activity, such as ‘painting the leaves in the way Pia likes’ or ‘painting the leaves according to such-and-such tradition’, then it is going to be diffi cult to have indirect reports that capture the content of the original utterance. But usually, less specific contents are good enough. For most purposes, it is enough to grasp that when Pia said ‘The leaves are green’ (first utterance) she only cared about the superficial aspect or, roughly, about the leaves being covered by paint (as opposed to them being naturally green).
In this sense, understanding can be said to come in degrees. When two interlocutors are co-situated and aware of each other’s intentions as well as the topic of the conversation, and so on, then their understanding of each other’s utterances can be very deep. By contrast, when they are not co-situated, or have only broadly grasped what the conversation is about, their understanding of the utterance will be more superficial and they will only roughly grasp what the speaker means—for instance, they might grasp that Pia’s utterance has to do with painting leaves, but not whether it is important that the whole surface is covered with paint.
My proposal goes as follows. We can classify activities more or less fine-grained. My suggestion is that we allow for different ways of classifying activities, ranging from very rough to very precise classifications in which any detail matters. If we look at the tension with this in mind, the threat will turn out not to be that serious. In the limiting case, the Austinian proposition can include the particular occasion—an activity so fine-grained that it will admittedly be difficult to share across contexts.
Given that we can class particular activities into activity-types with different fine ness of grain, we can represent the content of an utterance as being more or less finegrained, depending on the aim of the classification—ranging from very rough to very precise classifications in which any detail matters. Thus, we can have more or less fine-grained propositions. For example, we could have a range of propositions that could model the content of Pia and John’s utterance:
P1: <The leaves are green, Decoration> P2: <The leaves are green, Painting leaves>
P3: <The leaves are green, Painting leaves in a way that the original colour is no longer observable>
P4: <The leaves are green, Painting leaves in a way that the original colour is no longer observable and excluding pointillism>
And the limiting case: P5: <The leaves are green, Particular occasion>
We can class a given utterance as expressing one proposition or another, depending of the purposes of the classing. If we want to model the content of Pia’s utterance as being closely tied to the context of use, we can use P4 or P5. If we want to model the content that is shared across contexts (for example, in a speech report), we can use P1 or P2.
Propositions here can be understood, following Perry Korta and Perry [
[T]he reflexive-referential theory sees propositions as abstract
objects that are used to classify events of certain types (cognitive states and
utterances, paradigmatically) by conditions of truth (or other relevant forms of
success)—used explicitly by theorists such as ourselves, and implicitly in the
practice of those who have mastered the propositional attitudes and similar
constructions. We do not see propositions as denizens of a third realm to
which some quasi-causal relation relates us, but as devices by which we can
classify events along different dimensions of similarity and difference.
Different propositions can be used to classify the same act, relative to different
frameworks for associating success-conditions of various sorts. ([
Thus, the range of Austinian propositions I have described can be seen as a tool for classifying linguistic events. Sometimes, very fine-grained propositions are needed. These fine-grained propositions allow us to capture the insight that, as Travis cases suggest, any detail of the utterance context might matter to truth-conditions. On the other hand, less fine-grained propositions can be used in classifying linguistic events that are similar enough to count as same-saying.
To sum up, I have modelled utterance content as a <lekton, activity-type> pair and advocated a form of multipropositionalism according to which we can represent the content of an utterance with different granularities, depending on whether we want to capture a content that is shared across contexts or very detailed truth-conditions. This notion of content is compatible with radical contextualism is Searle-Travis style, for it respects the general Type-Underdeterminacy principle.
I will finish by noting that the approach presented here can be of some interest for the minimalist as well. According to the minimalist, non-indexical sentences are truthevaluable. Semantics determines classical propositions. However, minimalists typically agree that, when it comes to communication, these semantically determined propositions are insufficient7. A minimalist can distinguish a semantic notion of content (a minimal proposition whose truth-conditions are semantically determined) from a pragmatic notion of content, adjusted to the context of use. Thus, the notion of Austinian proposition I have put forward can be of interest in accounting for utterance content regardless of what one thinks of minimal propositions. A minimalist can have it that lekta are truth-evaluable and yet acknowledge that the truth of an utterance (not of the sentence) is relative to the relevant activity. The reason why a minimalist might find Austinian propositions useful is that of the ‘Claire has a good hand’ example: one thing is the truth-value of the sentence (in case it has one), another is the truth-value of the utterance. The latter arguably depends on what is at stake.