TOKEN STATES
Guy Longworth[1]
Warwick University, February 2009
1. Introduction. According to James Higginbotham, ‘There is no doubt that reference to events and states is a pervasive feature of human thought and language.’[2] But Higginbotham’s claim might appear to be falsified by the doubt Helen Steward expresses in the following passage:
There are certainly facts about pain which relate to particular people and which we could refer to by means of gerundive nominals—e.g., my being in pain—and there are individual pains, like headaches and toothaches, and there is the phenomenon of pain in general. But why should we add to all these a further mysterious class of entities: token pain states? No natural-language category refers to them, no philosophical purpose is served by introducing them. My claim is that, in view of these facts, we should say that there simply are no such things.[3]
In fact, Higginbotham’s claim is compatible with the existence of Steward’s doubt. Steward makes clear, in her ubiquitous talk of states, that she does not deny that—in a suitably broad sense of ‘reference’—reference to states is a (pervasive) feature of human thought and language. What she denies is that reference is made to what she calls token states. And here she is in disagreement with another of Higginbotham’s claims, one for which he, and others, have presented argument: the claim that states play a role in natural language semantics analogous to that played by token events.[4] And Steward is joined in doubting that token states play such a role by, amongst other philosophers, Jennifer Hornsby and, perhaps, Timothy Williamson.[5]
At an abstract level, we can characterise the locus of disagreement as follows. Consider sentences involving stative predication, for example (1), and nominalizations of such predications, for example (2):[6]
(1) Kim is happy.
(2) Kim’s happiness…
For each sentence like (1), we might propose either of the following semantic treatments where ‘EQ’ is one or another type of existential quantifier, restricted or sorted so as to quantify over states:
(3) happy (Kim).
(4) EQV [subject(V, Kim) & happy(V)].
Nominalizations like (2) will then be open analogues of (4), with a definite quantifier—i.e., a quantifier requiring exhaustion of the nominal—in place of EQ, as in (5):
(5) DEF/EQV[subject(V, Kim) & happy(V) & …
Now ‘EQ’ (and ‘DEF/EQ’) might be taken either to quantify properties or to quantify what only has properties (so isn’t ‘had’)—e.g., particulars.[7] The former option threatens to make (3) and (4) equivalent. The latter leads to further options. For if we take ‘EQ’ to quantify what only has properties, then it might quantify plurally or non-plurally, and it might quantify count-wise or mass-wise.[8] A sufficient condition for a semantic analysis to invoke token states will be that it takes (4) as a model for the treatment of stative predication and takes ‘EQ’ to quantify tokens, so definitely excluding properties and definitely including, amongst what only has properties, plural and non-plural countables.
Necessary conditions are more delicate to specify. Steward appears to argue that ‘EQ’ cannot quantify tokens since it quantifies mass-wise rather than count-wise.[9] But she also appears to hold that the existence of an appropriately related type suffices for tokenhood.[10] And types of what only has properties—including masses—are cheap. They are derivable via a seemingly productive formal operation—perhaps intensional abstraction over an open position, as in (6):
(6a) ÙX gold(X, V)
(6b) ÙV happy(x, V).
I propose to take the locus of dispute over token states to turn on the question whether ‘EQ’ quantifies over what only has properties, for the most part ignoring distinctions amongst those. I shall return briefly to this issue below. So for purposes of this paper, the dispute concerns the question whether a generalisation of (4), with ‘EQ’ an existential quantifier over what only has properties, supplies the proper treatment of (at least some) stative predications, with Higginbotham et. al. affirming, and Steward et. al. denying, that it does. I shall refer to the affirmation as the State Hypothesis.
Before expounding the case in favour of the State Hypothesis, I note a complication. Compare the nominalization in (2), repeated here, with the gerundive in (7):
(2) Kim’s happiness…
(7) Kim’s being happy…
As has often been pointed out, there are numerous distinctions between these two constructions.[11] For instance, one can admire Kim’s happiness, but not her being happy. Crudely, (2) appears to make reference to some sort of token, or particular, while (7) appears to make reference to something more like a fact, or state-of-affairs.[12] Without wishing to prejudge their semantical differences, or the basis for those differences, let us label (2) a stative nominalization and (7) a state-of-affairs nominalization. The complication is that, while I’ve characterised the dispute over token states as concerning stative predications and their nominalizations as in (2), Steward often characterizes her target in terms of state-of-affairs nominalizations.[13] Moreover, she appears to allow that constructions like (2) can make reference to a sort of token, namely a trope or property instance.[14] It might seem, then, that my characterization of the dispute means that Steward is not really a participant.
In fact, however, Steward’s major target is the claim that stative constructions, like (1), are to be given the same treatment as eventive constructions, like (8).
(8) Kim smoked a cigarette.
And she argues that the latter involve quantification over tokens in the sense specified above and not over tropes. Dispute can be reinstated, then, at two points. First, Steward is committed to the claim that the analogue for the eventive nominalization in (9) is the state-of-affairs nominalization in (10), rather than the stative nominalization in (11).
(9) Kim’s smoking a cigarette…
(10) Kim’s being happy…
(11) Kim’s happiness…
The defender of the State Hypothesis is committed to denying Steward’s first claim. As we shall see, the denial is underwritten by provision, and defence, of a unified treatment of (9) and (11). Steward’s first claim appears to be driven by her believing, first, that (11) makes reference to a trope and, second, that (9) makes reference, not to a trope—something that is a property instance—but rather to something that—in my terms—only has properties.[15] Given those beliefs, (11) couldn’t be the required analogue of (9), forcing consideration of the next most plausible candidate, (10). A second source of dispute, then, concerns the fundamental claim made by the defender of the State Hypothesis, namely that (9) and (11) should be given analogous logico-semantical treatment. The defender of the Hypothesis will therefore disagree with Steward in taking arguments against a tropist treatment of (9) as, inter alia, arguments against a tropist treatment of (11), and vice versa.[16]
The remainder constitutes a defence of the State Hypothesis. I shall begin by expounding some elementary observations that have been offered in favour of generalising the token-quantificational treatment of predication from eventive to stative predication. Although this will involve a brief rehearsal of parts of the case in favour of token events, I shall be assuming that case to have been made.[17] I shall also be allowing that the proper treatment of token events is whatever it is, compatibly with its invoking tokens. Hence, I will be neutral over the question whether token events are tropes.[18] The question I wish to address here concerns the generalisation, not the precise nature of its eventive basis. As should be expected, given the differences between eventive and stative predication, not every observation carries over from one case to the other. But in light of its potential for underwriting a fully unified treatment of predication, the case in favour of token states is compelling.[19] However, at least one failure of generalisation is ostensibly problematic for the unified treatment of predication. On the basis solely of its quantificational form, the unified treatment appears to be encumbered with a prediction that fails to square with reflective judgement. An important aim in what follows is to defend the State Hypothesis in the face of the apparent failure.
2. Token Events and States. In this section, I shall begin by marking some basic aspectual distinctions amongst types of predication. Then I shall detail some of the evidence that has been marshalled in favour of appeal to token events in the semantics of a certain range of those predications, the eventives. Then I shall indicate how related arguments have been used in an attempt to motivate the State Hypothesis generalisation to non-eventive predications. The purpose is to motivate and explain the generalisation and its commitment to token states.
2.1. States, Events, and Processes. Numerous aspectual distinctions between types of predicate have been proposed.[20] For present purposes, we require only a tripartite classification of predicates into eventives, processives, and statives.[21] One major distinction is between those predicates that accept temporal modification by ‘in a time’ and resist modification by ‘for a time’—eventive predicates—and those that exhibit the converse pattern—stative and processive predicates. Thus, the VPs in (12a) and (12b) fall within the range of eventive predicates, while those in (12c) and (12d) fall without.
(12a) Kim finished the cigarette in a few minutes/*for a few minutes.
(12b) Kim smoked the cigarette in a few minutes/*for a few minutes.[22]
(12c) Kim smoked for a few minutes/*in a few minutes.
(12d) Kim was a smoker for a few months/*in a few months.[23]
(12c) involves an example of a processive predication. (12d) involves an example of stative predication.
The first distinction tracks a second. Eventive predications are telic, in that each has a semantically specified telos, a required upshot, outcome, or eventuality. Thus, the lexically specified upshot of ‘smoked a cigarette’ is that a cigarette be, or become, smoked. Processive and stative predications are, by contrast, atelic. With a little simplification, we can see the temporal modification, ‘in a time’ as measuring an event from instigation to completion and, hence, as requiring an outcome to have been specified.[24] Manfred Krifka has proposed the following as a sufficient condition for telicity:
If V satisfies ‘VP’, then all parts of V that satisfy ‘VP’ must have the same starting and stopping point as V.[25]
The point of the condition is that a telic VP specifies a particular outcome, so that no event that ends sooner, and so lacks the specified outcome, can satisfy the VP. For that reason, ‘starting and’ should be excised from Krifka’s condition. As Susan Rothstein points out, this amendment allows that we can classify ‘run to Paris’ as an eventive (telic) VP, in line with its pattern of acceptable temporal modification. Parts of a run to Paris can satisfy ‘run to Paris’ even if they start later, as long as they end at the same point.[26] Even with that amendment, Krifka’s condition does not specify what it is in the nature of telic predicates that underwrites their meeting it. On the assumptions, first, that each distinct application of a telic predicate requires exactly one achievement of its telos and, second, that each such achievement occupies the same temporal stretch, the initial characterisation can supply the required account. The characterisation of telic predicates in terms of lexically specified upshot requirements is therefore more basic. Stative and processive predications are atelic.
A third distinction, based upon the first two, can be seen by considering what Alexander Mourelatos calls nominalization equivalents of predications.[27] Thus, (13a) seems to be equivalent to (13b), with the event reference apparently made explicit in (13c):
(13a) Kim smoked a cigarette.
(13b) There was a smoking of a cigarette by Kim.
(13c) There was an event that was a smoking of a cigarette by Kim.
The apparent equivalence between (13a) and (13c) is a major reason why these predications are classified as eventives, and are taken to refer to—or to involve quantification over—events. Compare here the putative analogues of (13) for processives—(14a)–(14c)—and statives—(15a)–(15c).
(14a) Kim smoked.
(14b) ??There was a smoking by Kim.
(14c) ??There was an event that was a smoking by Kim.[28]
(15a) Kim was a smoker.
(15b) ??There was a being a smoker by Kim.
(15c) ??There was an event that was a being a smoker by Kim.[29]
Although the relevant judgements are not perfectly transparent, many find (14b, c) and (15b, c) unacceptable. On this basis, Mourelatos claims that eventive predications are distinctive in allowing count quantification in their nominalization equivalents. Compare here the (16a)/(16b) alternation with those in (16c)–(16f), with ‘sm’ the unstressed mass existential:
(16a) There was at least one smoking of a cigarette by Kim.
(16b) ?There was sm smoking of a cigarette by Kim.
(16c) ?There was at least one smoking by Kim.
(16d) There was sm smoking by Kim.
(16e) ??There was at least one being a smoker by Kim.
(16f) ?There was sm being a smoker by Kim.[30]
On its more acceptable reading, (16b) appears to involve a processive predication, hence to be compatible with Kim’s failing to complete her smoking of the cigarette. Not just any mere quantity of a process aimed towards the smoking of a cigarette satisfies the eventive VP ‘smoke a cigarette’; to satisfy the latter VP, such a process would have to have a cigarette’s becoming smoked as its outcome.[31] Whether or not the appropriate forms of quantification are reflected at the linguistic surface, we can see this third feature as a natural upshot of telicity/atelicity alternation. Thus, atelics are akin to mass nominals in being homogeneous, where a predicate is homogeneous if and only if it is both divisive and cumulative: