Attitude Reports, Events, and Partial Models
CUNY, Graduate Center
Clausal complements of different kinds of attitude verbs such as believe, doubt, be surprised, wonder, say, and whisper behave differently semantically in a number of respects. For example, they differ in the inference patterns they display. This paper develops a semantic account of clausal complements using partial logic which accounts for such semantic differences on the basis of a uniform meaning of clauses. It focuses on explaining the heterogeneous inference patterns associated with different kinds of attitude verbs, but it contributes also to explaining differences among clausal complements of attitude verbs regarding the possibility of de re reference, anaphora support, presupposition satisfaction, and the distribution of subjunctive in certain languages. Moreover, it gives a new account of factivity.
The point of departure of this paper is the general observation that the failure of inferences from attitude reports is relative in that it depends both on the general type of attitude and on the particular instance of the attitude described. Thus, from John is surprised that P and Q one cannot infer John is surprised that P and John is surprised that Q, though this is possible with believe. Conversely, one can infer from John believes that P and John believes that Q, to John believes that P and Q, but only as long as the same belief state of John is involved.
In order to capture this dependency of inferences from attitude reports on a particular mental state or act, I propose an account on which clausal complements of attitude verbs (as well as independent sentences) characterize the intentional state or act described by the attitude verb in question, rather than referring to independent propositions. The semantic account of attitude reports of this paper can hence be called an 'event-based account' of clauses.
Formally, the denotation of any sentence, both independent and embedded, is construed as a function mapping an intentional state or act to a function from situations (which form the content of the state or act) to truth values. The relation between such a function f and the mental state or act e consists in particular conditions that f(e) has to meet (that is, the assignment of truth values to the situations in the content of e). For example, in the case of an attitude report with a verb of acceptance such as John believes that Mary left, Mary left must be evaluated as true in every doxastic alternative (a possible situation) that belongs to the content of John's belief state. Other types of attitude verbs, for instance doubt or wonder, may involve different conditions on how the clause characterizes the content of the described intentional act or state. Such conditions may involve existential, rather than universal, quantification over doxastic alternatives; and negative, rather than positive, evaluations in such alternatives. A crucial feature of the semantic account of clausal complements in this paper is that attitude verbs, unlike other relational expressions, may impose rather complex conditions on the evaluation of their complement with respect to an intentional state or act. Moreover, such conditions may not just be part of the lexical meaning of an attitude verb, but may be manifest in sentence meaning as well (for example in the case of factive verbs).
The general semantic approach in this paper is governed by the assumption that certain general conditions on semantic theory are fundamental and take priority over decisions of how to analyze particular constructions or expressions in particular contexts and of how to construe the meanings or semantic values of expressions. Unlike many other semantic approaches to attitude reports, the account of attitude reports in this paper presupposes a sharp distinction between semantic theory, in particular the 'level' of semantic values for expressions of natural language, and the 'philosophy of semantics', which includes questions of how to possibly explain certain notions assumed in the semantic theory in terms of other notions. This means, in the present case, that the semantic theory does not attempt to explain the components of the contents of mental states or acts it assumes, but rather restricts itself to characterizing their semantically relevant properties.
There are two fundamental conditions on a semantic theory of clausal complements. The first can be called the 'Uniformity Condition'; the second is Compositionality:1
(1) (i) Uniformity Condition
A semantic theory should assign uniform meanings to expressions in different
contexts, in particular, independent and embedded sentence and clauses embedded
under different kinds of attitude verbs.
A semantic theory should assign meanings to sentences based on the meanings
of their constituents.
The event-based account of clausal complements meets the Uniformity Condition for at least the meaning of sentences. Also it is able to fulfill compositionality - though this will not be elaborated explicitly at the subsentential level.
The paper is organized as follows. In Section 1, the basic account of attitude reports is motivated and outlined for verbs of acceptance. In Section 2, a number of complex, nonfactive attitude verbs are analyzed in detail by extending the account given in Section 1. Section 3 gives a further elaboration of the event-based account for 'verbs of explicit saying'. Section 4 develops an analysis of factive verbs and presents a reassessment of the notion of factivity, which gives indirect support for the event-based account. Section 5 discusses other phenomena of natural language that arguably involve mental states or acts as semantic objects, namely epistemic modals, conditionals, and certain generic sentences. An appendix briefly discusses the relation of the theory to dynamic semantic phenomena in attitude reports such as anaphora and presuppositions and shows how the de re-de dicto distinction and de se reference can be construed within the event-based account.
2. Attitude reports with predicates of acceptance
2.1. Motivation of the event-based account
In this section, I will motivate and introduce the event-based account of clauses with the most basic kind of attitude verbs, namely what I call 'verbs of acceptance' (following the notion of acceptance of Stalnaker 1984). The prototypical verb of acceptance is believe. Others include think, assume, suppose, imagine, see, hear, and conclude. Verbs of acceptance form a semantic class in that, as we will see, they pattern the same with respect to relevant types of inference. They, of course, also differ among each other in one or the other respect. For example they may differ in the degree to which the agent commits himself to the content of the mental state or act. Some predicates such as be sure impose a strong degree of justification for accepting the propositional content of their complement; others imply a particular source for the acceptance of the propositional content (such as see, hear and conclude). Furthermore, verbs of acceptance differ in the extent to which the acceptance is restricted to a particular context (assume, suppose, imagine) or to which the propositional content has to cohere with other beliefs (belief vs. suppose or assume). Finally, verbs of acceptance differ in whether they describe a state, that is, a mental state or disposition to act (believe, know, assume), or an occurrent thought (imagine, think).2
For distinguishing different attitude verbs, there are basically six inference patterns that are most important.3 Four among them are Conjunction Introduction, Conjunction Distribution, Disjunction Introduction, and Disjunction Distribution:
(2) a. Conjunction Introduction
a V that S, a V that S' ==> a V that S & S'
b. Conjunction Distribution
a V that S & S' ==> a V that S & a V that S'
c. Disjunction Introduction
a V that S v a V that S' ==> a V that S v S'
d. Disjunction Distribution
a V that S v S' ==> a V that S v a V that S'
Two other relevant inference types are Downward Entailment (inference from weaker to stronger propositional content) and Upward Entailment (inference from stronger to weaker propositional content):
(3) a. Downward Entailment
a V that S, S' -> S ==> a V that S'
b. Upward Entailment
a V that S, S -> S' ==> a V that S'
These two inference types, however, should be understood only in an appropriately restricted way. In the context of propositional attitudes, Downward Entailment should be considered only to the extent that it involves 'relevant logical consequences'. For example, one would want to exclude inferences to propositions involving entities the relevant agent did not conceive of or to those involving predicates for which the agent did not consider the relevant entities to be applicable. Upward Entailment should be restricted in the same way; it should only involve propositions about relevant entities.
It is a matter of dispute whether monotonicity properties (Upward and Downward Entailment) should be part of the subject matter of semantics at all.4 Even with the restriction to relevant stronger or weaker propositions, usually there can still be counterexamples found in particular cases to whatever monotonicity property one might want to attribute to a verb. For example, in the case of believe, which appears to be upward entailing, it may be that John believes that P, but fails to believe that Q, even if Q is a relevant and rather obvious logical consequence of P (see Soames 1988 for detailed examples).
However, there is evidence that monotonicity properties are semantically relevant nonetheless (and thus, that such counterexamples may not be crucial). This evidence comes from attitude verbs like believe in the scope of negation and from 'negative verbs' like doubt and deny, which appear to be downward entailing, and as such would allow for the same sort of counterexamples as believe. Negated attitude verbs like believe and negative attitude verbs arguably are treated as downward monotone in natural language since they licence negative polarity items such as ever and anybody:
(4) a. John doubts / denies that he has ever been to France.
b. John does not believe that anybody won the race.
On the account of negative polarity licensing of Ladusaw (1979), negative polarity items are licenced precisely in downward-entailing environments.5 If believe in the scope of negation is downward entailing, then, clearly, believe itself must be upward entailing.
So even though monotonicity properties may not hold strictly in particular cases, speakers may treat attitude verbs as if they were upward or downward entailing. That is, they may classify attitude verbs semantically (for instance for the purpose of negative-polarity licensing) under idealized circumstances, disregarding any lack in logical skills on the part of the agent. So under such premises, monotonicity properties appear to be semantically relevant properties of attitude verbs.
Returning now to verbs of acceptance as a semantic class, we can observe the following behavior with respect to the inference patterns above, though a certain qualification has to be made concerning Conjunction Introduction:
Conjunction Introduction: OK6
(5) John believes that Mary left, and he believes that Bill arrived.
John believes that Mary left and Bill arrived.
Conjunction Distribution: OK
(6) John believes that Mary left and Bill arrived.
John believes that Mary left, and he believes that Bill arrived.
Disjunction Introduction: NO
(7) John believes that Mary left, or he believes that Bill arrived.
John believes that Mary left or Bill arrived.
Disjunction Distribution: NO
(8) John believes that Mary left or Bill arrived.
John believes that Mary left, or he believes that Bill arrived.
Upward Entailment (closure under 'relevant logical consequences'): OK
(9) John believes that a tall woman left.
John believes that a woman left.
Downward Entailment (closure under 'relevant stronger propositions'): NO
(10) John believes that somebody left.
John believes that Mary left.
Even though I have chosen the verb believe, the same observations hold for all verbs of acceptance.
Conjunction Introduction is not always valid, and the way in which it fails will lead us to one of the central ideas of the semantic account of attitude reports of this paper.7 For example, John may believe that P, and he may, at the same time, believe that Q, without realizing that P and Q are mutually contradictory and in particular without believing that P and Q (and consequently that P and ¬ P). What goes on in this case is that there are two distinct belief states of John involved: one in which John believes that P and one in which he believes that Q. Thus, Conjunction Introduction will hold only as long as the premises involve the same concrete belief state.8
The relativization of the attribution of an accepted proposition to a particular mental state or act also accounts for the possibility of change of belief or of the occurrence of different thoughts (at different times) with contradictory contents.
In cases of different beliefs or thoughts at different times, though, it would still be sufficient to relativize the attitude reports to a particular time, rather than particular mental states or acts. The inferences then would hold just in case the premises are about the same time interval. However, there are other cases where the relativization to concrete mental states or acts is necessary. These are cases where accepting a proposition depends crucially on the context, as pointed out by Stalnaker (1984) (See also Lewis (1986), Chap. 1). For example, John may, in front of his friends, accept that smoking is OK, but in front of his children that smoking is bad. Rather than attributing to John a belief in a contradictory proposition, it is more plausible in this case that John has two (possibly) simultaneous states of acceptance: one which manifests itself when John is among his friends and one which manifests itself in front of his children.
Thus, both the attribution of acceptance of a proposition to a person and the possibility of inferences of a certain kind should be relativized to a particular mental state or act. This mental state or act, in turn, can be considered an argument of the attitude verb, occupying an additional Davidsonian argument position (cf. Davidson 1967, Higginbotham 1985, Parsons 1990). I will henceforth assume that every (attitude) verb has such an additional event argument place. By convention, this will be the first argument place of the verb. Thus, believe now denotes a three-place relation between belief states, agents, and propositions.
A bit more has to be said about the notion of mental state or act that I assume. The relevant notion is not that of a mental state or act that is individuated independently of its external environment and thus has a 'narrow' content, based on its internal properties only. Rather it is the more abstract notion of mental state or act which has a 'wide content' and whose identity may be depended on the environment. It basically corresponds to the notion of mental state or act of Stalnaker (1984, 1990).
The dependency of inferences from attitude reports on the particular mental state or act that is described leads to one of the central ideas of the semantic account of attitude report of this paper: the function of the complement clause of an attitude verb is to characterize the mental state or act which is the Davidsonian event argument of the attitude verb. Different mental states or acts that act as arguments of the same attitude verb may have different 'global' contents, and a complement clause will be evaluated with respect to the content of the mental state or act. More formally, every mental state and act is associated with its own set of partial models, and the that-clause with be evaluated in a particular way in such a set of partial models. For this purpose, the denotation of a clause will be construed as a function mapping an event (a mental states or act) and to a function from situations to truth values, or equivalently, a function mapping an event and a situation to a truth value.
There is another, to some extent equivalent, way of capturing the dependency of the meaning of the clause on the event. On this account, the clause does not contribute an independent argument to the relation denoted by the attitude verb, but rather expresses a property that is attributed to the event. This would yield a seemingly rather different view of the semantic function of clauses. For then, the clause would act semantically as a predicate, predicated of the event argument, rather than as an independent argument. (Note that this view does not imply that the clause syntactically must act as an adjunct to the verb, like as event predicates such as today in John left today (as in Davidson's (1967) account). Also adverbs that have the status of arguments, rather than adjuncts, would have to be treated as event predicates in a Davidsonian account, for example, badly in John behaved badly.)
A construal of a clause denotation as an event property would correspond to the logical form of (11a) in (11b), where  is the denotation function. Here believe is, for the sake of simplicity, represented as a two-place relation between belief states and agents, though the proper denotation of believe should better be considered a three-place relation with the 'meaning postulate' in (11c) effecting the third argument being predicated of the first. [V]* is the relation between states or acts of acceptance of the sort specified by V and agents; thus [V]* is like [V] except that the third argument position is existentially quantified over.
(11) a. John believes that Mary left.
b. e([believe](e, [John]) & [that Mary left](e))
c. Lexical condition on verbs of acceptance (event-predicational format)
For a verb of acceptance V, an event e, an entity x, and an event property P,
[V](e, x, P) iff [V]*(e, x) and P(e).
If the view that clauses express properties of the mental state or act argument is not adopted, then we would instead get an analysis of the lexical meaning of attitude verbs as in (12), where CH is some condition relating the third argument of the belief-relation to the first (with the effect that the third argument appropriately characterizes the content of the first argument):
(12) Lexical condition on verbs of acceptance (propositional format)
For a verb of acceptance V, [V](e, x, F) iff [V]*(e, x) and CH(e, F).
I will call the view of the semantic structure of attitude reports given in (11c) the ‘event-predicational format’, and the one in (12) the ‘propositional format’. It is clear that the difference is rather notational. More precisely, the event-predicational format can be considered a special instance of the propositional format. The reason for me to adopt the propositional format is that with different kinds of attitude verbs, the clausal complement characterizes the content of the event argument in different ways. So different relations like CH have to be specified for different verbs, and this means that no uniform definition of a clause denotation as an event predicate can be given.
It is worthwhile, nonetheless, to give the event-predicational format some more general considerations. There are in principle a number of ways in which a clause could be considered an event predicate. Conceiving of clausal complements as predicates of the event argument of the embedding attitude verb allows for a compositional formulation of a number of proposals on which the evaluation of the clausal complement is, in some way, considered dependent on the intentional state or act described by the attitude verb.
For example, predicating a clause of an intentional state or act could mean that the clause, in some way or other, specifies the form of the intentional state or act. This corresponds to the view on attitude reports of, for example, Scheffler (1963) and perhaps Fodor (1986).9 On this view, the meaning of (11a) could roughly be formulated as in (13):
(13) (e)([believe](e, John) & [that Mary left](e)), whereby [that Mary left](e) only if e has the form
given by that Mary left.
Alternatively, it could mean that the constituents of the clause will be matched with 'constituents' of the mental state or act. This would correspond to the account of attitude reports of Crimmins/Perry (1988) and Crimmins (1990). On this view, the meaning of (11a) could be formulated as in (14), where n is what Crimmins and Perry call the 'cognitive notion' of Mary and i what they call the 'cognitive idea' of leaving (both of which are 'unarticulated constituents'):
(14) (e)([believe](e, John) & [that Mary left](e, n, i)), whereby [that Mary left](e,
n, i) iff is the content of e and n is a part of e and corresponds to
Mary in and i is a part of e and corresponds to the property of leaving
The event-predicational format, thus, provides a general way in which the evaluation of the clausal complement can be made dependent on the concrete intentional state or act that is described.10
It should also be mentioned that an idea similar to the event-predicational format for attitude reports has been proposed for perception (the 'adverbial theory of perception) (cf. Tye 1984 and references therein).
The event-based account of clauses raises a general question that should be addressed at the outset. It concerns the status of the attitude report itself. Clearly, given the Uniformity Condition, independent sentences should be given the same denotation as embedded clauses. The question then is, with respect to which event they will be evaluated. We will see that there is an answer to this question in Section 4. For the sake of simplicity, I will disregard the event component of independent sentences until then, focusing only on the way embedded clauses should be treated.
Another general remark is required concerning the event argument of attitude verbs and the status of clauses as characterizing the content of that event. A clausal complement of a verb does not necessarily specify the content of a mental state or act. It may also specify the ‘content’, that is, the intended goal, of an action, as in John tried to close the door. Both mental and nonmental acts or states may have a content, or, in other words, may be 'intentional' acts or states. Thus, more generally, the clausal complement specifies the content of an intentional act or state. In this paper, however, I will focus on verbs describing mental states and acts and disregard other kinds of intentional verbs.
Given the event-based account of clauses, what the inferences from attitude reports tell us is that if certain properties hold of a particular belief state, then also other properties always hold or fail to hold of that same belief state, depending on whether the inference is valid or invalid. So if the premises specify the content of the same belief state e, then the conclusion again involves e; and it has to be seen that e supports the relevant conditions imposed by the conclusion. So Conjunction Introduction, in particular, will be valid only if the premises and the conclusion are about the same belief state.11 The question that I will address in the next section is: how is the content of a belief state to be construed and how can the denotation of a clause be construed as a function from events?