In this paper we provide a formal model for occasion-sensitive semantics motivated by so called `Travis cases' (Travis 1978, 2000, 2008, 2009). We suggest that understanding of an utterance of ' (knowing its truth conditions) can be modelled as a twofold partitioning of worlds in logical space, where the initial partition is induced by the contextinvariant meaning of the sentence uttered and the latter on the basis of context-dependent goals. Our model uses only a single parameter to capture occasion-sensitivity of sentences: practical goals.1 Context-sensitivity is a phenomenon that a ects natural language expressions. Frege already pointed out that some sentences express thoughts only when certain features of context are known, such as the time of an utterance or the speaker. The case in point are sentences containing context-sensitive terms such as indexicals or demonstratives (see Frege 1956: 296). For Frege, as for many others, context-sensitivity seemed a fairly isolated phenomenon, a ecting only a small class of expressions. More recently, however, radical contextualists (e.g. Travis 2008, Recanati 2004, Carston 2002) have suggested that contextsensitivity is pervasive, that it a ects all open class expressions (i.e. not just pure indexicals), and that it does so in a way that classical truth-conditional semantic theories may nd di cult to track.
1.1
Radical contextualism is motivated by examples such as following:2
State of a airs: Pia's Japanese maple is full of russet leaves. Pia paints them green.
Zoe: Do you have green leaves? Pia: These leaves are green.
Botanist: Do you have green leaves? Pia: # These leaves are green.4
State of a airs: Sid, an impoverished student, uses a door over two stacks of milk crates as a desk to write on.
Pia: Does Sid have a desk? Max: Sid has a desk.
Pia: Does Sid have a desk? Max: # Sid has a desk
State of a airs : Pia is looking for her shoes. Sid sees them, heels protruding from beneath the bed.
2 In the literature these are known as Travis cases (sometimes referred to as
contextshifting arguments, see
Pia: Are the shoes under the bed? Sid: # The shoes are under the bed. 1.2
Travis cases are taken as evidence for a number of claims. First, they demonstrate what we deem to be an uncontroversial fact that the interpretation and evaluation of utterances are somehow sensitive to contextual factors. However, the reason why Travis cases are particularly interesting is because the relevant sentences contain no pure indexicals (such as I, you, here, today ) or other standard context-sensitive expressions (e.g. gradable adjectives, predicates of personal taste, modals, quanti er phrases), and yet their interpretation and evaluation varies across contexts of utterance. Second, in each example, the sentence type used in both contexts is the same, and it is assumed that its conventional meaning remains invariant. One important consequence of the fact that the semantics of sentence types is invariant is that it would be impossible to avoid the denial of the law of non-contradiction, since the same sentence appears to be evaluated as true and false of the same state of the world, if, that is, the sentences used in these examples are taken at their face value (namely, as not being context-dependent in some sense). Third, the fact that the interpretation of utterances varies in this way is typically taken as evidence for the variability of intuitive truth-conditions or truth-conditional content of utterances.7
A number of analyses of Travis cases have been proposed, many whose
primary concern is to preserve a compositional account of truth-conditions.8 With
the exception of semantic minimalists who reject that the intuitions about
truthvalue of utterances are good evidence for truth-conditions of sentences, almost all
other solutions assume that at least some of the constituents in these sentences
have context-dependent interpretations. Thus the appeal to context-dependence
seems by far the most common way to avoid the attribution of contradictory
7 See
Lepore (2008),
In this paper we aim to take a di erent approach because we believe that
occasion-sensitivity is di erent from indexicality, and also that it is not a threat
to compositional semantics. Thus, although we will propose what is, broadly
speaking, a contextualist treatment of occasion-sensitivity, unlike
The approach defended in this paper is formally and conceptually closest (albeit not identical) to the position known as non-indexical contextualism developed in MacFarlane (2007, 2009). We agree with MacFarlane that contextsensitivity is a broader phenomenon than indexicality and it shouldn't be conated with the latter. We also think that occasion-sensitivity is a good example of non-indexical context-sensitivity, where (i) semantic content of a predicate is invariant, (ii) the world of evaluation is invariant, and yet (iii) the extension of the predicate (and the truth-value of an utterance) varies. This is possible, MacFarlane argues, because the circumstance of evaluation includes more parameters than just a possible world (where these parameters are initiated by the context of use rather than the context of assessment).
With regards to Travis cases, MacFarlane's suggestion is that predicates such
as green or weighs 80 kilos invariantly express a property of being green or
weighing 80 kilos yet such a property has an intension which is a function from
the circumstance of evaluation { consisting of a pair of parameters (counts-as
parameter and world parameter) rather than one { to extension. A non-standard,
all encompassing count-as parameter \settles what things have to be like to have
various properties: e.g. the property of weighing 160 pounds, or of being tall"
We believe that this picture, generally speaking, adequately captures the
phenomenon illustrated in Travis cases. However, we also contend that because
of its generality it potentially su ers from similar problems as
Our proposal departs from MacFarlane's on several points (although the rst two may be only terminological departures): rst, we take it that properties that feed into the counts-as function can also be zero-place properties such as the property of these leaves being green or Sid's having a desk ; second, we don't consider a zero-place property to be the same as the proposition expressed by an utterance on occasion (i.e. intuitive truth-conditions); we consider the proposition expressed by an utterance on occasion (i.e. the Austinian proposition)9 to be an output of, rather than an input to, the counts-as function. Third, whereas MacFarlane leaves it open which feature of a context drives modi cations of properties and why such modi cations would ever take place, we suggest that a speci c contextual feature mandates shifts in when things count as having certain properties.10 By considering pragmatic processes as not linguistically mandated (rightly so), existing non-indexicalist approaches fail to put enough emphasis on why, in the course of communication, we nevertheless do systematically modify standing semantic contents of our expressions. We suggest that Travis cases point towards one speci c contextual factor as playing part in xing when things would count as having a certain property: relevance (of the information) for the goals that speakers pursue within particular practical projects. Even though objective predicates have their context-invariant semantic contents, these contents are uninformative unless they are restricted so that they become valuable for achieving contextually salient goals. 2
The role of standing meaning
We will assume that each declarative11 sentence ' has a context-invariant,
compositional meaning in virtue of syntactic and lexical structure and linguistic
9 See
To capture this notion of semantic constraint on what is said, we suggest that the conventional meaning of an (atomic) declarative sentence can be viewed as the set of worlds which are compatible with its meaning, that is, those worlds where the sentence is true on some of its uses but false on others. For any atomic (NP+VP) sentence (used to state a piece of information) there will be: (i) some possible worlds or situations which are incompatible with what it means, and which can thus be eliminated purely with respect to its conventional meaning, and (ii) others which are compatible with the meaning of the sentence but in which an utterance of the sentence is not always true.
In standard truth-conditional frameworks, a world is considered to be compatible (incompatible) with the meaning of ' only when ' is true (or false) at that world. Accordingly, the set of worlds in W' would thus correspond to the proposition expressed by ', modelled as the set of worlds in which ' is true. However, as evidenced in Travis cases, this common identi cation of meaning with truth conditions is problematic insofar as the truth valuation of an utterance of ' may shift even when the world of evaluation and conventional meaning remain unchanged. So we believe it is crucial to resist the common assumption that conventional meaning is su cient to determine when a sentence uttered in context would be true. Hence, on our view, the meaning function does not partition the worlds into those at which ' is true and those at which ' is false. Rather ' is considered to be truth-evaluable and worlds are considered to be truth-makers only after taking into account a further factor: an occasion on which ' is used.
Take The shoes are under the bed Travis case. The sentence The shoes are under the bed may be true if the shoes are well hidden under the bed, or if their heels are poking out. This sentence could also be true if the shoes are three oors down aligned with the bed. Or if they are buried in cement under the bed etcetera. However, an utterance of the sentence on an occasion won't necessarily be true at all these worlds: this is the lesson we learned from Travis cases. Even though all these very di erent conditions of the world are compatible with what the sentence means not all of them will make an utterance of the sentence on some occasion true since an occasion places some further constraints on success (i.e. truth) of the utterance.
Let us call this set generated by the meaning function the compatibility set of ', or W'. If a world is semantically incompatible with the meaning of ', then there will be no possible understanding on which ' is true at that world (where an understanding of ' is a re nement of W'). If a world is semantically compatible with ' then there will be some understanding on which an utterance of ' is true at that world, but also some understandings on which it is false at that world.
11 01 (a) 10 00 For simplicity's sake, let us assume there are only two possible ways of being green: painted green and naturally green. Figure 1 represents the set of all alternatives, where (11) is the world where the salient leaves are both painted and naturally green, (10) is the world where they are painted green but not naturally green, (01) is the world where they are only naturally green, and (00) is the world where they are neither painted nor naturally green. Since in our simpli ed model painted and naturally green exhaust the ways of being green, in the (00) world the leaves are not green. The set of worlds which are semantically compatible with the sentence The leaves are green is thus Wgreen : f11; 10; 01g, whilst the set of semantically incompatible worlds is Wgreen : f00g.
Travis cases show that not all the worlds in Wgreen will make an utterance of The leaves are green on some occasion true: for instance, in the botanist context, the world (10) is ruled out as a truth-maker because it matters in this context that the leaves are naturally green (see Figure 2). By contrast in the decorator context, it does not matter if the leaves are painted or natural, so the only world ruled out is f00g.
Hence it depends on which understanding of green is relevant, which worlds will make an utterance of ' true. The resolution of the issue which understanding of ' is relevant for truth evaluation is here independent from its meaning. To x which understanding of a predicate is correct we need to ask which information (about the world) would be relevant to an agent on some occasion. In the next section we suggest that the worlds in W' are kept or eliminated on the basis of how conducive they are to the goal that the agent aims to achieve on an occasion.12 12 In our diagram the world (01) is thus not t for the botanist's goal. So far we used the term occasion and occasion-sensitivity without properly de ning them, or showing what the relation between an occasion and a world is supposed to be. In this section we will try to give a more detailed account of these notions by identifying a feature that we consider necessary for determining which proposition is expressed by an utterance of a sentence on occasion. We shall suggest that a key feature of some occasion of using ' that gives rise to the phenomenon of occasion-sensitivity is a practical goal which is salient of that occasion and which agents strive to achieve. Recognising which goal is salient is necessary to determine which proposition someone expressed by an utterance of '.
What does a practical goal have to do with determining when a sentence is true? In this section we try to motivate the connection between goal directed actions and correct understanding of utterances.
To get the idea that practical goals constrain interpretation o the ground,
we take the following remark made by Robert Stalnaker (who cites Grice) as our
guiding principle:
\[W]e should see speech as action to be explained, like any other kind
of action, in terms of the beliefs and purposes of the agent"
To Stalnaker's principle we add the following observation about the successconditions of actions: namely, that an action (including a speech action) is successful when it successfully achieves its goal. But what constitutes the goal of a speech act?
Ever since
Our answer to this question is negative because we believe that goals (and
hence successes) that agents aim to achieve in the course of communication
are typically much broader (and much more practical) than the characteristic
goals of assertions, questions, or commands. That is, by virtue of performing an
assertive speech act, an agent does aim to say something true and, if true, add the
proposition to the common ground
If communication, in general, is embedded in other goal-directed, practical
activities, then we can also assume that what someone communicates on a
particular occasion is directed at ful lling some such goal. To take a Travis case
again as our example, if I know that my friend is a botanist and is searching for
green leaves for her experiment (assuming I am not totally ignorant about this
topic) then I also should know that her request for green leaves bears on that
particular goal and that so does my response. If I sincerely say to my friend,
\Search no more, the leaves on my tree are green", then what I say is that the
world is indeed such that her project can be successfully carried out. In other
words, by asserting this, I suggested that the world is conducive to my friend's
goal. If it turns out, however, that I have the leaves that are painted green but are
naturally red, then her reasonable expectations as to the state of the world won't
be satis ed and her project won't take o . That is, this state of the world is not
conducive to the salient goal and so what I said to my friend was false (although
it could have been true were her goal decorating rather than chemistry).
13 This idea of communication is based on
It is in this way that we take practical goals to be an important factor in interpreting and evaluating what someone has said. 3.2
Agents communicate to exchange or gain the information about the world which helps them achieve their goals, which, it was argued, outstrip narrow goals that de ne illocutionary forces of particular speech acts. We also suggested that the state of the world can be such that it is more or less conducive to ful lling some practical goal, and so, when we provide some information to someone about the state of the world, it goes without saying that this will have certain practical consequences, i.e., some positive or negative e ect on salient plans and goals. For instance, the world where a room is such that the shadows of objects can be seen in it is not conducive to the goal of opening a camera and developing a lm since such a project won't be successfully achieved in this state of the world. Consequently, if someone says that the room is dark, where this particular goal is salient in the conversation, then he said something false. If, however, he says this when the salient goal is to put a child to sleep, then he might well be saying something true.
Our main suggestion is that the worlds which are compatible with the meaning of a sentence (the members of W') may include those with very di erent potentials in terms of enabling a successful action (some very low, some high). For a particular goal g and world w we can take it that w will be more or less conducive to g (i.e. achieving g in w is more or less di cult). We will give a formal de nition of this notion below.
We propose that, among the worlds that are semantically compatible with the sentence uttered (those belonging to the compatibility set), only those worlds that are conducive to the salient goal on an occasion constitute the proposition expressed by an utterance of the sentence on that occasion. 4
A model In order to determine which proposition is expressed by an utterance of the sentence on some occasion we propose to rst evaluate all possible worlds in terms of how goal-conducive they are, and to assign each a numerical value, thus producing a ranking over worlds. The set of worlds with the highest rank that is also semantically compatible with the sentence will correspond to the proposition expressed by the utterance of that sentence on that occasion. 4.1
Because we take simple NP+VP sentences as our atom of analysis, we will use a language L for propositional logic, with its standard set of propositional letters and connectives. We assume the set of atomic sentences A and de ne formulas as follows: ' := p; ' _ '; :'; ' ^ '; ' ! ' where p 2 A, the set of atomic sentences. A model M for L is a tuple M = hW; G; V; C; i in which W is a non-empty set of possible worlds, G is a non-empty set of goals, V is a valuation function which assigns to a sentence a numerical value 1 for semantically compatible or 0 for semantically incompatible, C is a goal-conduciveness function which maps a pair of worlds and goals to a value in the interval [0; 1] ( the c-value of the world w given the goal g ).
C : G
W ! [0; 1] where g 2 G is a goal that agents are aiming to reach on occasion.
8w; w0 2 W; w0 g w i C(w0; g)
C(w; g) We can produce a ranking of all worlds in logical space with respect to how conducive they are to some goal. For any pair of possible worlds fw; w0g the world with the lower conduciveness value will be ranked lower than the world with the higher conduciveness value. If they have the same value they will also have the same rank.
w 2 MAXg i 8w0 2 W; w0
g w The set of worlds with the highest ranking, MAXg, will form the top tier of the worlds wrt to their goal-conduciveness. For any world w with a given ranking one can conceive of another world w0 that has the same ranking despite di ering in some way from w. So, in principle, there will also be more than one world with the MAXg value.
We were able to order the worlds in terms of their conduciveness to a given goal quite independently of language. This is because a successful action that brings about some goal does not necessarily presuppose a use of language (i.e. a performance of a speech act). However, since our primary interest is in the interpretation and evaluation of sentences used in communication (i.e. speech acts), our model should capture how pragmatic constraints on interpretation we de ned by means of contextual goals interact with semantic constraints we de ned by means of the initial partition induced by conventional meaning.
V : A
W ! f0; 1g where A is the set of atomic sentences, and values 1 and 0 designate semantically compatible and semantically incompatible, respectively (NOT: true and false).
M; w j=g ' i w 2 MAXg ^ V ('; w) = 1 M; w j=g :' i M; w 2g ' M; w j=g (' ^ ) i M; w j=g ' and M; w j=g The proposition expressed by an utterance of an atomic sentence p is de ned in model M as the intersection of W', the set of worlds that are semantically compatible with the sentence, and the set of worlds that have the MAXg value. 5
Examples We can now analyse Travis cases using our present semantics in the form of the following conditional T-sentence: (T) The utterance u of the sentence ' is true at world w relative to goal g i w is conducive to g and w is semantically compatible with '.
The leaves are green: In this example we have a shift of goals between the two occasions: on one occasion the salient goal is to decorate a table centrepiece, in which case the natural property of the leaves is not relevant; on another occasion the salient goal is to examine the green leaf chemistry, in which case the natural property is relevant. The stipulated world of evaluation is such that the leaves are only painted green (but are naturally red). This world is semantically compatible with the uttered sentence on both occasions, but is only conducive to the former goal and not to the latter. Therefore, the utterance is true at w only wrt to the goal of decorating and is false wrt the goal of doing chemistry. (O1) ugreen is true at w wrt gdecorate (O2) ugreen is false at w wrt gchemistry The shoes are under the bed: Again we have a shift of goals on the two occasions. On one occasion the goal is to retrieve the shoes and the extent to which they are visible is not relevant to a successful achievement of this goal. On another occasion the goal is to make sure the shoes are well hidden, so not visible at all. The world of evaluation is such that the heels of the shoes are visible from under the bed. Saying that the shoes are under the bed is conducive only to the former goal, whilst saying this on the second occasion would prevent the achievement of the latter goal. Therefore, the information conveyed by the sentence is true with respect to the former goal and false wrt to the latter. (O1) ushoes is true at w wrt gretreive (O2) ushoes is false at w wrt ghide Sid has a desk: In this scenario, each occasion of utterance is paired with a di erent goal such that on one occasion the goal is to have something to sit and write on, and on another the goal is to nd a piece of expensive o ce furniture. The world of evaluation is such that Sid has a door over two stacks of milk crates. Saying that Sid has a desk in the rst case facilitates the ful lment of the goal, whilst in the latter case it probably doesn't. Hence, the information conveyed by the utterance is true on the former occasion and false on the latter. (O1) udesk is true at w wrt gwrite (O2) udesk is false at w wrt goffice 6
Conclusion
This completes our model for occasion-sensitive semantics. It is important to
bear in mind that we do not claim that all meaning is occasion-sensitive
insofar as we assumed that the conventional meaning of a sentence creates an
initial partition which is not sensitive to any particular occasion. The outcome
of the meaning partition corresponds to what is in the literature known as
minimal proposition.14 Unlike proponents of the notion of minimal proposition, we
take intuitions behind Travis cases as evidence for the variability of intuitive
truth-conditions across occasions. Although we decided to sidestep the
terminological side of this debate by stipulating that occasion-sensitive utterances of
' are successful (an not merely true), the substantial point of Travis cases and
occasion-sensitivity nonetheless remains: `minimal proposition' is not only
something with no `psychological reality'