Predicting Verbal Presuppositions M´arta Abrus´an University of Oxford December 14, 2010
Abstract This paper offers a predictive mechanism to derive the presuppositions of verbs. The starting point is the intuition, dating back at least to Stalnaker (1974), that the information conveyed by a sentence that is in some sense independent from its main point is presupposed. The contribution of this paper is to spell out a mechanism for a diagnostic to decide what will become the main point of the sentence and how to calculate independence. It is proposed that this can be calculated by making reference to event times. As a very rough approximation, the main point of an utterance is what (in a sense to be defined) has to be about the event time of the matrix predicate and the information that the sentence conveys but is not (or does not have to be) about the event time of the matrix predicate is presupposed. The notion of aboutness used to calculate independence is based on that of Demolombe and Fari˜ nas del Cerro (2000).
1
Introduction
Why do verbs give rise to the presuppositions they do? One possibility, which has been the prevailing attitude in the presupposition literature, is that this question might not have an answer, and for a principled reason: the presuppositionality of verbs (and indeed any expression) is simply an arbitrary lexical property. This idea was implicitly captured by the term conventional implicature of Karttunen and Peters (1975, 1979), and has been Many thanks to Denis Bonnay, Nathan Klinedinst, Daniel Rothschild and Philippe Schlenker for many conversations on various versions, Emmanuel Chemla, Paul Egr´e, Giorgio Magri, David Nicholas and Tim Williamson for comments on a previous draft and Benjamin Spector, David Beaver, Mandy Simons, Craige Roberts, Sigrid Beck, Robert Demolombe, Nicholas Asher, Paolo Santorio, Kyle Rawlins, Jacopo Romoli, Fran¸cois Recanati, Chris Barker, Laurent Bartholdi, Danny Fox, Eytan Zweig, Ofra Magidor, Anna Szabolcsi, Bridget Copley, Matthew Towers and the audiences at LoLa10, The Aboutness Workshop in Toulouse, JSM10, University of York, University of T¨ ubingen, SALT20, the Riga Symposium on Semantics and IRIT for very helpful comments and questions at various stages of this research. All remaining errors are my own. This research was supported by the ESF (Euryi grant to P. Schlenker) and The Mellon Foundation.
1
inherited by much of the research on presuppositions. Thus it is usually assumed that while the lexicon specifies e.g. that know has to presuppose its complement, there is no similar lexical specification for believe. (1)
a. b.
John knows that it is raining John believes that it is is raining
There are reasons however to be dissatisfied with this answer. First, as it was pointed out by Levinson and Annamalai (1992) and Simons (2001), if presuppositions were simply conventional, they should manifest the property of detachability, i.e. it should be easy to find pairs of expressions which share their truth conditions, but differ in their presuppositional behavior. However it seems that words that express a similar meaning trigger a similar presupposition. Thus (2a) and its synonyms in (2b) all seem to imply that John indeed used to smoke. (2)
a. b.
Has John stopped smoking? Has John quit/finished/given up/ceased smoking?
Such examples show that there is a generalization to be captured about what type of meaning can give rise to what type of presuppositions. But they are entirely mysterious given a conventional view of presupposition triggering. Second, as it was argued by Levinson and Annamalai (1992), if presuppositions were purely conventional elements of the non-truth conditional meaning, one would expect there to be translation difficulties and conceptual mismatch when comparing the presuppositional items of different languages. However, this is generally not the case. Levinson and Annamalai (1992) offer a detailed comparison of English and Tamil and show that the presupposition triggers in these two unrelated languages are exactly parallel, and also manifest the same presupposition behavior in complex sentences. Such facts argue that presuppositions should follow somehow from the content of presupposition triggers. Yet how presuppositions could be predicted from the meaning of triggers has been an elusive and rarely addressed question. While the few attempts in the literature to explain presuppositions of at least certain items provided valuable insights (cf. Sperber and Wilson 1979, Simons 2001, Abusch 2002, 2010), they either did not make correct predictions or failed to be sufficiently explanatory. Stalnaker (1974) and Schlenker (2010) laid out a blueprint for a triggering mechanism, but did not provide a theory themselves.1 This paper offers a predictive mechanism to derive the presuppositions of verbs. The starting point is the intuition, dating back at least to Stalnaker (1974), that the information conveyed by a sentence that is in some sense independent from its main point is presupposed. The contribution of this paper is to spell out a mechanism for a diagnostic to decide what will become the main point of the sentence and how to calculate independence. It is proposed that this can be calculated by making reference to event times. As a very rough approximation, the main point of an utterance is what (in a sense to be defined) has 1
Recently, Beaver et al. (2010) proposed a mechanism for predicting “projective meaning”, a wider class of elements than just presupposition. This theory is still under development.
2
to be about the event time of the matrix predicate and the information that the sentence conveys but is not (or does not have to be) about the event time of the matrix predicate is presupposed. The notion of aboutness used to calculate independence will be that of Demolombe and Fari˜ nas del Cerro (2000). The paper proceeds as follows: Section 2 provides an overview of previous proposals and a short preview of the present one. Section 3 introduces the notion of aboutness that this paper will use, while Section 4 lays out the core of the proposal. In Section 5 it is argued that the system should also accommodate common knowledge. Section 6 discusses the issue of context sensitivity of presupposition triggering. Section 7 concludes the paper.
2
Previous Approaches
An intuition that has been around for quite some time now is that presuppositions are the part of the content of the sentence that is in some sense “not the main point” of an utterance. E.g. Karttunen and Peters (1979:1) define the presupposition of sentences as “propositions which the sentences are not primarily about but which have to be established prior to an utterance of the sentences in order for communication to go smoothly” (emphasis mine). Going one step further, Stalnaker (1974) suggested that some presuppositions are indeed generated precisely in order to make sure that an utterance does not make a heterogeneous contribution to the context, i.e. in order to allow speakers to know what the main point of the utterance was. It is clear that “x knows that P” entails that P. It is also clear that in most cases when anyone asserts or denies that x knows that P, he presupposes that P. Can this latter fact be explained without building it into the semantics of the word? I think it can. Suppose a speaker were to assert that x knows that P in a context where the truth of P is in doubt or dispute. He would be saying in one breath something that could be challenged in two different ways. He would be leaving unclear whether his main point was to make a claim about the truth of P, or to make a claim about the epistemic situation of x (the knower), and thus leaving unclear what direction he intended or expected the conversation to take. Thus, given what “x knows that P” means, and given that people normally want to communicate in an orderly way, and normally have some purpose in mind, it would be unreasonable to assert that x knows that P in such a context. (Italics mine)
So Stalnaker suggests that presuppositions are generated in order to avoid uncertainty as to what the main point of a speaker’s contribution to the context is.2 But Stalnaker’s remark does not make a prediction as to which part of the assertion should become the main point, and which should be presupposed.3 2 Cf. also Abbott (2000) for a related idea. Abbott claims further that “[t]ypically, the asserted proposition in an utterance will correspond to the main clause of the uttered sentence”, but does not offer further clarifications on this point. 3 In effect Stalnaker is making use of an (implicit) pragmatic principle that could be paraphrased as “Avoid heterogeneous assertion”. A similar idea can be found in Grice (1981). Schlenker’s (2008) principle
3
Another idea present in Stalnaker’s discussion (albeit implicitly) is that lexical entries have a structured meaning, namely that the content of “x knows that p” can be divided into the two conjuncts p ∧ K(x,p), where the latter is a predication attributing some suitable propositional attitude to x. The idea that the meaning of presupposition triggers is naturally split in this way raises some non-trivial issues. From a logical point of view, there are countless ways of picking two propositions p and q st. S= (p ∧ q), so we need a further mechanism to tell us why this particular structure should be salient for the hearer as opposed to some other (cf. discussion in Abusch 2002, 2010). Thus one could assume that the job of a triggering theory is to give a mechanism to split the meaning of a sentence into two parts and then decide which of these is presupposed. But this would be too hasty. While logically it is certainly possible to split the meaning of a sentence S into two conjuncts (e.g. S= p ∧ (p→S)), from a psychological/linguistic perspective the question arises whether we can really plausibly divide the meaning of “x knows that p” into two salient meaning constituents, and in particular the two conjuncts p and K(x,p). (cf. Williamson 2002, Yablo 2008). To demonstrate the problem with an easier example, one might wonder whether there is a salient paraphrase for p in the equation red = colored ∧ p, where p is supposed to be whatever proposition ‘remains’ after ‘deducting’ colored from the meaning of red ? Luckily, for a triggering theory to succeed it does not have to provide us with two conjuncts such that one of them will be the presupposed and the other the non-presupposed part of the meaning of the sentence. It is enough if it can provide a way of distinguishing the entailment(s) that is(are) presupposed among the set of all the entailments of the sentence, whether or not the ‘remainder’ part of the meaning is a definable or identifiable entity.
2.1
Wilson and Sperber (1979)
Wilson and Sperber’s (1979) paper is the first proposal that attempts to explain why certain entailments of sentences exhibit the special properties of presuppositions.4 Instead of assuming, as usual, that semantic entailments of a sentence are an unordered set of propositions, they argue that this set is ordered by certain syntactic and intonational factors. On the basis of this internal structure of entailments, they set out to distinguish entailments that are focalized from those which are peripheral, and within the first group, distinguish those which are in the foreground of attention from those which are in the background. These linguistically determined distinctions can then be used to predict the presuppositional behavior of utterances. Let’s illustrate this with an example. The entailment of S that we get by substituting “Be Articulate” is in the same spirit as well. Cf. Schlenker (2010) for more discussion. 4 Technically, they deny the existence of presuppositions as separate category. What this means is that they deny the existence of conventionally postulated presuppositions. If presuppositions are the type of entities that can be predicated based on the overall meaning of a sentence, the difference between saying– as they do–that certain entailments which show the typical projective behavior of presuppositions can be distinguished as opposed to saying that precisely these entailments should be called presuppositions becomes immaterial.
4
the focused expression in S by an existentially quantified variable is its first background entailment. First background entailments act as presuppositions. In the case of (3a), where by assumption the complement of the attitude verb is focused, the first background entailment is (3b): (3)
a. b. c.
Susan knows [that it is raining]F Susan knows something It is raining
An entailment that neither entails nor is entailed by the first background entailment is ‘not involved in normal interpretation’ (for our purposes: is presupposed). (3c) is such an entailment and therefore it will act as a presupposition, at least in contexts where the complement of the verb know is focused. The proposal seems to make correct predictions for the presuppositions (if that is indeed what they are5 ) of focus and clefts. Wilson and Sperber also attempt to capture the discourse sensitivity of presuppositions. Such effects have been recently discussed in Beaver (2004) and Kadmon (2001). Unfortunately, it does not capture these correctly and it makes further incorrect predictions for a wide range of facts. Firstly, one might wonder what happens if the focused constituent in (3) was the matrix subject? In this case the first background entailment would be that someone knows that it is raining. This entailment neither entails nor is entailed by the entailment that Susan believes that it is raining, which therefore should be presupposed, contrary to fact. Second, consider (4): (4)
a. b. c.
John killed [Bill]F John killed someone Bill is dead
The first background entailment of (4a) is (4b). This is independent from the entailment (4c), which should therefore be presupposed, contrary to fact. These problems are not unique to the particular examples mentioned above: a similar problem will arise with any attitude verb whose subject is focused, and any transitive verb whose object is focused. Thus it seems that Wilson and Sperber’s (1979) theory does not succeed in making correct predictions.
2.2
Simons (2001)
The idea that a conversational explanation can be given as to why certain items stand with a presupposition was revived recently by Simons (2001). Thus rather than a semantic mechanism, Simons (2001) attempts a purely pragmatic mechanism of presupposition triggering, treading along the path envisaged by Stalnaker (1974).6 Her idea in a nut5
Cf. Geurts and van der Sandt (2004) and replies to it in the same issue. The idea that at least some presuppositions should be conversationally triggered was also embraced by Chierchia and McConnell-Ginet (2000) and Kadmon (2001) and Schlenker (2010). Neither of these discussions offer a way to derive these presuppositions. 6
5
shell is as follows. By uttering a sentence S a speaker raises the question whether Q if S can be interpreted as addressing the question Q. Thus uttering an atomic sentence with content p counts as raising the question Q=Whether p?, and uttering a sentence with content ¬p counts as raising the question Q=Whether p? as well, and so does an utterance of a sentence with the content if p, then ..., and so on for other operators over which presuppositions project7 . Given this, she defines the triggering mechanism as follows: (5)
If a speaker A raises the question Whether p? by uttering S, and p asymmetrically entails8 some proposition q, then A indicates that she believes q to be true.
Let’s look at an example. A speaker who utters (6a) raises the question in (6b), and indicates that she believes the proposition in (6d) — which is asymmetrically entailed by (6c) — to be true: (6)
a. b. c. d.
S=John knows it is raining Q=Whether p? = Whether John knows it is raining p= John knows it is raining q=It is raining
Since the sentence John does not know that it is raining would raise the same question, namely (6)-b, it would be predicted to trigger the same presupposition, and so on for other contexts. As Simons (2001) herself points out, the proposal as it stands strongly overgenerates (cf. also discussion Abusch 2010). E.g. in connection with the example in (6), the proposal predicts that it should also presuppose that John believes that it is raining, and indeed any proposition that S asymmetrically entails. This is certainly incorrect. It also predicts (7) to presuppose (7d), contrary to fact. (7)
John killed Bill a. S=John killed Bill b. Q=Whether p? = Whether John killed Bill c. p= John killed Bill d. q=Bill is dead.
Further, from the fact that every asymetrically entailed proposition is presupposed, it follows that every sentence is predicted to presuppose itself.9 This is because any sentence S asymetrically entails both S ∨ Q and S∨ ¬Q, for any proposition Q. Therefore, S is predicted to presuppose both of these disjunctive propositions, and the intersection of the two is S itself. 7
The idea of presuppositions raising question alternatives was also used in Chemla (2007) to predict presupposition projection. 8 A proposition p asymmetrically entails q if p entails q, but q does not entail p. 9 This was pointed out to me by Emmanuel Chemla (pc.)
6
Because of these grave problems, Simons suggests that the triggering principle should be refined in such a way that only those entailments of S should end up being presupposed that in some sense count as a precondition for the truth of S. While the intuition is suggestive, the whole task of a triggering theory is to provide a definition for what it means for a proposition to be a precondition in this sense. Unfortunately Simons (2001) does not provide such a definition, and thus it is fair to say that the puzzle has not been solved by this paper.
2.3
Abusch (2002, 2010)
Abusch’s paper offers a prediction for a subset of presupposition triggers, which she identifies as soft triggers, including questions, focus and certain verbal triggers. These have a representation involving alternatives. It is standardly assumed that questions and focus introduce sets of alternative propositions which are derived by replacing the focused/questioned element by a contextually salient object of a suitable type. (Rooth 1992 wrt. focus, Hamblin 1973 and Karttunen 1977 wrt. questions.) (8)
a. b. c.
Who came? [Bill]F came The alternative set for both: ALT={that Bill came, that Mary came, ...}
Abusch proposes that sets of alternatives trigger the default presuppositional constraint according to which the proposition formed as the disjunction of the set is true. This amounts to the claim that some alternative in the set is true, which is what gives rise to the (defeasible) existential presupposition triggered by focus and questions. Abusch then goes on to extend the same idea to certain verbal presuppositional triggers. For these cases she stipulates that their lexical representation contains a set of alternatives. As before, the presupposition arises from the constraint that the disjunction of the alternatives is true. For example, the verbs know and stop trigger the following alternatives: (9)
a. b.
know triggers the alternatives ALT={know, be unaware} stop triggers the alternatives ALT={stop, continue}
If a sentence S that contains any of these triggers is uttered, the alternative propositions that we get by replacing the trigger by its lexical alternative are activated. Given the default presuppositional constraint, we then pragmatically presuppose that the disjunction of these alternative propositions is true: (10)
a. b.
John knows that it is raining or John is unaware that it is raining John stopped smoking or John continued to smoke
Since the disjunction still entails that it is raining (in the case of (10a)) or that John used to smoke (in the case of (10b)), the correct presupposition is predicted. Abusch’s account is very suggestive in the case of focus and questions, but somewhat more dubious when it comes to verbal triggers (cf. also discussion in Schlenker 2010). The 7
problem is that while for focus and questions the alternative sets were derived simply by compositional semantics and pragmatics and were independently motivated, in the case of verbal triggers she needs a lexical stipulation, namely it needs to be lexically defined for each trigger what their alternative is. Since alternatives uniquely define what the presupposition is, it is not less of a stipulation to postulate lexical alternatives for verbal triggers than to simply postulate what their presupposition should be. There is no reason for example why the lexical alternative of know could not be believe, and in this case Abusch’s system would predict that “x knows that p” should presuppose that “x believes that p”. There is also no principle from which it would follow that the verb kill should not trigger an alternative, and even less that this alternative could not be find dead, in which case the death of the object argument should be presupposed, incorrectly. Thus while Abusch makes correct predictions for the cases she discusses, what is missing from her proposal is a principled reason for where exactly the identity of the alternatives comes from, and thus her treatment of verbal triggers remains stipulative.
2.4
Preview of this proposal
The intuition behind this paper is that of Stalnaker’s: that entailments of a sentence that are in some sense independent from the main point of the sentence are presupposed. What makes a proposition independent from another? I will not attempt to answer this question in general. Instead, I will use a diagnostic tool that can be informally paraphrased as follows. The main point of a sentence is given by those entailments that are by nature about the event time of the matrix predicate. Propositions that describe events that are not (or do not have to be) about the event time of the matrix predicate of S are independent, hence presupposed. Unlike Stalnaker (1974) and Simons (2001), the present paper proposes a semantic (and not a pragmatic) mechanism to predict presuppositions of verbal predicates. This mechanism takes atomic sentences as its input. Presuppositions of complex sentences are derived from the presuppositions of atomic sentences they contain, via a separate projection mechanism, e.g. Heim (1983) or other. Following Partee (1973) and much subsequent work I assume that the event time of verbal predicates behaves analogously to a pronoun and refers to a particular time interval. Further, I will introduce the notion of a canonical temporal representation of a sentence. Canonical temporal representations of sentences (CT(S)) are sentences in which the independent tense argument positions of predicates are filled by choosing any constant of the appropriate type. Thus in canonical representations of sentences accidental co-temporal relations do not have to be preserved. (However, some lexically specified co-indexing relations have to be preserved, cf. Section 4.4). Some examples of CT(S) representations are shown below: (11)
a. b.
John sees Bill (at time t1 ) CT(S): sees (John,Bill,t) John believes (at time t1 ) that he is tired (at time t1 ) CT(S): believes(John, tired(John,t’),t) 8
c.
John stopped smoking (at time t1 ) CT(S): stopped (John, smoking, t)
I will call the temporal arguments replaced with new constants during the construction of CT(S) representations TS-arguments, and the constants that replace them CTS-arguments. Let’s now define the CT(S)-equivalent of an entailment p of S as follows: (12)
The CT(S)-equivalent p’ of an entailment p of S (abbreviated as p=CTS p’) is a. p itself, if (the linguistic form of) p does not contain TS-arguments b. if p contains TS-arguments, then p’ is the proposition that p can be turned into by replacing its TS-arguments with the corresponding CTS-arguments.
The triggering mechanism predicts an entailment p of S to be presupposed if it has a CT(S)-equivalent proposition p’ that is not about the event time of the matrix predicate of CT(S).10 (13)
The triggering mechanism for verbal presuppositions An entailment p of S is presupposed if S has a CT(S) representation such that the CT(S)-equivalent of p is not about the event time of the matrix predicate of CT(S).
Here is an illustration of the idea. For convenience, TS-arguments are represented with numerical subscripts, while CTS-arguments are represented by primes. Let’s take a CT(S) representation of (14) (given in (14a)) such that t1 , t and t’ all refer to non-overlapping intervals. Given the lexical semantics of know, its complement, which denotes the proposition that it is raining, is entailed. Since the CT(S) equivalent of this entailment is not about (in the sense to be defined in the next section) the matrix event time t of CT(S), it is predicted to be presupposed. (14)
S: John knows (at t1 ) [that it is raining (at time t1 .)] a. CT(S): knows (John, raining(t’), t) (by the method of constructing canonical temporal sentences) b. S|=raining (t1 ) c. raining (t1 )=CTS raining(t’) d. raining(t’) is not about t (given the definition of aboutness in Section 3) e. therefore, S presupposes that it is raining.
Compare this with the entailment of S that John believes that it is raining. This proposition is not predicted to be presupposed, because its corresponding CT(S)-entailment is necessarily about the matrix event time of its CT(S): (15)
S: John knows (at t1 ) [that it is raining (at time t1 .)]
10
It is an interesting question whether the present mechanism could also be defined using some version of the notion of generalized entailment of Schlenker (2008, 2010) that can apply to expressions whose type ‘ends in t’. An investigation of this question will have to be left for the future.
9
a. b. c. d. e.
CT(S): knows (John, raining(t’), t) S |=believes (John, t1 , that it is raining (at t1 )). believes (John, t1 , that it is raining (t1 )) =CTS believes (John, t, raining(t’)) believes (John, t, raining(t’)) is about the event time t therefore S does not presuppose that John believes that it is raining.
In the above case it was the canonical representation of the sentence that made the event time of the embedded clause different from that of the matrix event time, and thus presupposed. In some other cases, entailments are lexically specified to be true at some time other than the event time. This is what happens in (16), where the the lexical entailment that John used to smoke comes from the lexical semantics of the change of state verb. Thus, it is a lexical property of the verb stop that the entailment(16c) of the canonical sentence in (16a) is true at some time preceding the event time. As the CT(S) equivalent of this entailment is not about the event time, it is presupposed. (16)
S: John stopped smoking (at t1 ). a. CS: stopped (John, smoke, t) b. S |=smoke(j, t2 ), where t2 refers to some time interval preceding t1 c. smoke(j, t2 )=CTS smoke(j, t2 ) (because t2 is not a TS-argument) d. smoke(j, t2 ) is not about t (at least for some CT(S)) e. therefore, S presupposes that John used to smoke
The calculation of presuppositions thus crucially depends on the temporal argument of the sentence. Does this mean that all presuppositions depend on tense? No. Temporal arguments are simply used as diagnostic tools for telling independent events apart. I follow Stalnaker (1974) in assuming that presuppositions are also part of the entailed meaning. Presuppositions are simply entailments that are in some way distinguished. In this framework a presupposition triggering mechanism can be viewed as a function that takes as its input the bivalent meaning of a sentence S, and outputs one or more entailments of S, those which are also presupposed.11
3
Aboutness
As it was pointed out by Ryle (1933) the notion of ‘being about’ has many different uses in English. Accordingly, researchers addressing this topic have often talked about somewhat different issues. One notion of aboutness, which will be central to this paper, is the notion 11 Another option would be a view of presuppositions under which they are not part of the entailed meaning. In this case the triggering function would take as its input the ‘total’ meaning of S (call it TM(S)), i.e. the meaning we get by lumping together truth conditional and presupposed content, and output one or more entailment(s) of TM(S) as the presupposition. In this paper I use the Stalnakerian view and assume that presuppositions are also entailed, but it should be borne in mind that the present proposal is also compatible with a view of presuppositions where these are not entailed. Cf. Schlenker (2010) for a more detailed discussion of how these two views compare from a perspective of a triggering theory.
10
of a sentence being about an entity, in the sense of a sentence P(a) being about the entity denoted by a. This notion has been addressed from many angles, from a semantic (logic) point of view (e.g. Goodman 1961, Demolombe and Fari˜ nas del Cerro 2000) and from a pragmatic point of view (e.g. Strawson 1964, Reinhart 1981, Portner and Yabushita 1998). A second, related, notion concerns being about a class of things, in the sense of “All ravens are black” being about ravens, but not (or much less) about (non-)black things. This sense was addressed e.g. in Putnam (1958). A third notion of aboutness is that of being a subject matter, in the sense of the sentence “The number of planets is nine” saying something about the subject matter of astronomy. An influential analysis of this topic in the philosophy literature was given by Lewis (1988) and more recently by Humberstone (2000) and Yablo (2008).12 Lewis’s idea was that subject matters define equivalence relations on the set of all possible worlds. Simplifying somewhat, sentences can be said to address subject matters if they eliminate some cells from the partition defined by the equivalence relation. A somewhat similar idea has been around in the linguistic literature under the name “Question Under Discussion” (cf. Carlson 1983, Roberts 1996), which is also intended to capture the notion of subject matter, or what the “main point” of a sentence is. The similarity of this approach to Lewis’s idea becomes apparent once we recall Groenendijk and Stokhof’s (1984) proposal according to which questions define equivalence relations on the set of possible worlds.13 It would go beyond the scope of this paper to do justice to all the relevant approaches and provide an accurate survey. Instead, I will only introduce the particular treatment of the notion of aboutness that I use in this paper, the notion of being about an entity (object). This treatment is the straightforward extension of the notion of being about an argument worked out by Demolombe and Fari˜ nas del Cerro (2000) for first order logic. (cf. Appendix 1 for an exposition of Demolombe and Fari˜ nas del Cerro’s (2000) original ideas.)
3.1
Being About an Object
This section recasts the notion of being about an object of Demolombe and Fari˜ nas del Cerro (2000) in terms of possible worlds. The proposal has two parts: the definition of variants of a possible world with respect to an object and the definition of being about an object. 3.1.1
Variants
We define variants of possible worlds wrt to objects. Roughly speaking, two worlds w and w’ are c-variants if they only differ by the properties of c. In this case we need to allow 12
cf. also Demolombe and Jones 1996 and various works in relevance logic that address related issues. Although Roberts (1996) does not define her approach using Groenendijk and Stokhof’s (1984) theory of questions. It should also be mentioned that the different senses of aboutness are not independent from each other and that some of the analyses proposed for one or other types of aboutness might be extended to other types of aboutness as well. 13
11
that c-variants differ in the truth assignment to atomic sentences where the expression c appears as an argument of the matrix clause. Let M be a model hW, D, R,[[ ]]i, where • W is a set of possible worlds • D is a domain of things14 • R⊆ W× W is an accessibility relation on W • for a non-logical atomic predicate pn , [[pn ]]w ⊆ Dn For the purpose of calculating variants, a sentence such as a knows that p at t will be treated as if it had the more simple syntax K(a)(t), where K stands for knows that p. The language is assumed not to contain the identity predicate. We can then define for every possible world w∈ W, variants w’ of w as follows: • Dw’ =Dw 0
• [[x ]]w =[[x ]]w , for each constant or variable symbol x • if p is a predicate symbol of arity n – if t is an n+1-tuple the first n terms of which are from D of Lc and the final element is from W and which contains no occurrence of the constant c, then h[[t]], w’ i∈ [[p]] iff h[[t]], w i∈ [[p]]. – if an element hd1 , ..., dn i of Dn is such that for every j in [1,n], dj 6= [[c]], then 0 hd1 , ..., dn i ∈ [[p]]w iff hd1 , ..., dn i ∈ [[p]]w . 3.1.2
Aboutness
Given the above notion of variants, we might define aboutness as follows: (17)
Aboutness A sentence S is about the object denoted by the constant symbol c iff there are 0 two worlds w and w’ in W which are c-variants and [[S]]w =1 and [[S]]w =0
Conversely, we can also give a definition of what it means for a sentence to not be about an object c: (18)
The property of not-being about A sentence S is not about the object denoted by the constant symbol c iff for every 0 w, w’ in W st. w and w’ are c-variants [[S]]w = [[S]]w .
14
To keep the above discussion more simple, I assume that both individuals and time intervals belong to D. Things in D are typed however,(so D=De ∪ DI ) and predicates have a type restriction on the their arguments.
12
I assume that the domain of possible worlds is defined by the combinatorial possibilities of the elements in the language. For this reason, it is not possible to find two atomic propositions ϕ and ψ that are equivalent, but differ in the arguments they have. Had we restricted the set of possible worlds to allow such equivalences by rendering some worlds “impossible”, the notion of aboutness introduced here would fail to capture our intuitive notion of aboutness. Suppose it was the case that there were two sentences expressing the same proposition, but would differ in their arguments, e.g: [[Glup]]=[[Fido is tired]]. Now Fido is tired would not be predicted to be about Fido any more, because we would not be able to find two worlds w and w’ which are Fido variants differing only in the extension of the property tired. For the same reason, the domain of possible worlds W cannot be restricted to allow necessary connections among propositions. (Thanks to D. Rothschild and N. Klinedinst for discussion on this issue.)15 3.1.3
Examples
(19)
S=Fido is tired
The sentence (19) above is about Fido iff there are two Fido-variants w, w’, st. [[S]]w =1 0 and [[S]]w =0. Notice that the definition in (17) quantifies over all worlds, therefore the entailment ϕ of (19) (20)
ϕ=Some individual is tired
is also about Fido, because there are two worlds which differ only in the properties of Fido, 0 st. and [[ϕ]]w =1 and [[ϕ]]w =0, e.g. if Fido is the only tired individual in w. It is easy to see that in general sentences expressing existential quantification are about every individual. Similarly, a sentence expressing universal quantification is also about every individual in the domain. Suppose our sentence was (21)
ξ= Every individual is tired.
Now there are two worlds which differ solely in the properties of Fido, st. one of these makes the sentence true and the other false: Take world w in which every individual is in the denotation of the predicate tired, and another world w’ in which every individual except Fido is in this set. Finally, observe that the sentence ψ in (22) is not about Fido according to the definition above: (22)
ψ= Fido is tired or Fido is not tired
15
An anonymous reviewer for SALT20 has asked the following question: Suppose in w John has a unique brother, and Fred is his brother in w. Now take the sentence Fred has a unique brother. In every world w’ in which John has a unique brother is false, Fred’s properties will have to change as well. Wouldn’t this mean that John has a unique brother should fail to be about John? However, note that the proposition that Fred has a unique brother is simply an existential statement, and can be made true by individuals other than John. On the other hand the proposition that Fred’s unique brother is John is about John. Hence in fact this case does not pose a problem for the definition of aboutness.
13
This is because since ψ is a tautology, it is true in every two worlds w, w’ that are Fidovariants.
4
The Proposal: Some Core Examples
The present paper proposes a semantic mechanism which predicts presuppositions of verbal predicates. I follow Stalnaker (1974) in assuming that presuppositions are also part of the entailed meaning. Presuppositions are simply entailments that are in some way distinguished. In this framework a presupposition triggering mechanism can be viewed as a function that takes as its input the bivalent meaning of a sentence S, and outputs one or more entailments of S, those which are also presupposed. The main intuition behind this paper, similarly to that in Stalnaker (1974), is that entailments of a sentence that are in some sense independent from the main point of the sentence are presupposed. The main point of a sentence is given by those entailments that are by nature about the event time of the matrix predicate. Propositions that describe events that are not (or do not have to be) about the event time of the matrix predicate of S are independent, and hence presupposed. To calculate this, I will introduce the notion of a canonical temporal representation of a sentence.16 Canonical temporal representations of sentences (CT(S)) are sentences in which the independent tense argument positions of predicates are filled by choosing any constant of the appropriate type. (But note that some lexically specified co-indexing relations have to be preserved, cf. Section 4.4) Let’s call the temporal arguments replaced with new constants during the construction of CT(S) representations TS-arguments, and the constants that replace them CTS-arguments. We can now define the CT(S)-equivalent of an entailment p of S as follows: (23)
The CT(S)-equivalent p’ of an entailment p of S (abbreviated as p=CTS p’) is a. p itself, if (the linguistic form of) p does not contain TS-arguments b. if p contains TS-arguments, then p’ is the proposition that p can be turned into by replacing its TS-arguments with the corresponding CTS-arguments.
The triggering mechanism predicts an entailment p of S to be presupposed if it has a CT(S)-equivalent proposition p’ that is not about the event time of the matrix predicate of CT(S). (24)
The triggering mechanism for verbal presuppositions An entailment p of S is presupposed if S has a CT(S) representation such that the CT(S)-equivalent of p is not about the event time of the matrix predicate of CT(S).
16
I am grateful to Emmanuel Chemla for suggesting (pc) that the triggering mechanism could be calculated one a more abstract representation than the sentence itself, in order to avoid problems that plagued a previous version of this proposal.
14
4.1
Background Assumptions
Some clarifications are in order. I will assume that event times denote salient intervals whose value is assigned by the context. As such, they are rather like pronouns (cf. Partee 1973). For convenience I further assume that the tense argument is represented in the syntax as well.17 I use a type-theoretical system which in addition to the basic types e and t contains a type i whose domain is the set of time intervals. In this system, predicates have an extra argument slot for time, thus what are usually assumed to be one place predicates such as intransitive verbs are going to be two place predicates, taking an individual and a time argument: [[is tired]]w,g =λti .λxe . x is tired at t in w
(25)
Tense morphemes are time variables that saturate the time argument slot of predicates. The denotation of this variable is given by the assignment function g supplied by the context, which assigns it an element from the domain of time intervals. E.g. the sentence in (26) is true iff John is tired at the time assigned to t2 by g. That is, whatever the value of t2 , it denotes the time of John being tired.18 (26)
John is tired at t2
From the syntactic representation of sentences we can now create canonical temporal representations of sentences (CT(S)). These are sentences in which the tense argument positions of predicates are filled by choosing any constant of the appropriate type. The identity of the constants chosen for canonical representations is not relevant, except that the time they refer to should exist in the model. Thus in canonical representations of sentences accidental co-temporal relations do not have to be preserved.19 Some examples of CT(S) representations are shown below: (27)
a.
John sees Bill (at time t1 ) CT(S): sees (John,Bill,t)
17
Though this is not crucial. cf. Kusumoto (2005) for some recent arguments for this assumption, but also Keshet (2008) for arguments against Kusumoto’s position. 18 I will be agnostic however about the question whether it is the past/present/future feature (presupposition) on tense variable/morpheme itself that contributes the meaning of anteriority or simultaneity (as in Partee (1973), e.g.), or whether these derive from phonologically null elements that stand in some relation with tense morphemes, and give the ordering between event times (the time denoted by the tense morpheme, here: t2 ) and evaluation times (represented here as t*) as below: (cf Kusumoto (2005), e.g.) (i)
a. b.
[TP t* PAST λ2 t2 [VP John tired]] [[PAST]]w,g =λPhi, ti . λti . there is a time t’ st. t’
19
An exception is the case of (restrictive) relative clauses, where the CTS representation will have to use the same variable in the relative clause and the matrix clause. This is not unnatural considering that temporal dependencies in clausal complements and relative clauses are generally different. A well known example is the sequence of tense phenomenon, which shows wildly different properties in embedded relative clauses and clausal complements.
15
b. c. d.
John believes (at time t1 ) that he is tired (at time t1 ) CT(S): believes(John, tired(John,t’),t) John hopes (at time t1 ) to be promoted (at time t1 ) CT(S): hopes(John, promoted(John,t’),t) John stopped smoking (at time t1 ) CT(S): stopped (John, smoking, t)
Notice that sentences with infinitival complements (such as (27c)) are in principle treated the same way as sentences with tensed propositional complements. If, however, as in the case of implicative verbs such as manage, the verb lexically specifies that the tense of the embedded complement cannot be different from the tense of matrix verb, this co-indexing relation has to be preserved in the CTS as well, cf. Section 4.4. In what follows, I will call the temporal arguments replaced with new constants during the construction of CT(S) representations TS-arguments, and the constants that replace them CTS-arguments. For convenience, the former are represented using subscripts, and the latter by (zero or higher number of) primes.
4.2
Factive Presuppositions
This section spells out how factive presuppositions can be derived. The example I look at in detail is the verb know, but it will be shown that the same analysis carries over to the whole class of factive verbs. Some other members of this class in English include realize, discover, notice, recognize, find out, remember, forget, be aware that, admit, intuit and a subclass of sensory factives sense, see, smell, hear, detect, observe. A major subclass of factive verbs is the class of emotives, factive verbs used primarily to convey the subject’s emotional attitude towards information. This class, examined in subsection 4.2.2, includes predicates such as regret, be annoyed, be upset, be glad, be happy, be ecstatic. 4.2.1
Know
Suppose that our sentence S is the following example: (28)
a. b.
S: John knows (at time t1 ) that Mary is tired (at time t1 ) CT(S): know (John, t, tired (Mary, t’))
Let K be the set of all the propositions that (28a) entails.20 (29)
∩K=S
Assume that the proposition expressed by S in (28) corresponds to some set of worlds WS . Then for any set of worlds W1 st. WS ⊂ W1 , K contains the proposition that corresponds to W1 . 20
K is closed under logical entailment: If K logically entails ϕ, then ϕ∈ K
16
Intuitively, there are two types of entailments that are in K: logical entailments, and entailments that can be derived from the meaning of S. I will call the second type of entailment lexical entailment. Lexical entailments are not given in a formal way: they are only available to speakers by inspecting their intuitions about the lexical meaning of predicates and the meaning of S itself. Lexical entailments express restrictions on which possible worlds are in W, rendering some worlds ‘impossible’. Eg. the fact that ‘x believes p’ is a lexical entailment of ‘x knows that p’ is the restriction that there is no world in W in which ‘x knows that p’ is true, but ‘x believes that p’ is false. If we only looked at logical entailments, not taking the restrictions on possible worlds imposed by lexical entailments into account, then it can be shown that if S is about a matrix time t, and S is atomic, then its non-tautological logical entailments are also about t.21 Here is why. If the elements of W are assumed to be derived via the combinatorial possibilities of the elements in the language, and Q is not a tautology, there is a w that makes [[S]]w =0 and [[Q]]w =0. Since S is about t, and there are no necessary connections among predicates, there is a possible world corresponding to every possible combination of 0 predicates and arguments, and so there is a t-variant w’ of w st. and [[S]]w =1. Since S entails 0 Q, [[Q]]w =1. Thus no non-tautological logical entailment is predicted to be presupposed. Tautological entailments of S are predicted to be presupposed, as these are not about the matrix time, but this is harmless. Thus it is not necessary to look at purely logical entailments of S at all to find presuppositions. However, we need to look separately at lexical entailments, because given the restrictions that these impose on the set W, it is now possible that no t-variants w and w’ 0 can be found such that for a given lexical entailment Q, [[Q]]w =0 and [[Q]]w =1. The reason is that if a sentence P(a) lexically entails some other sentence P(b), there are no worlds in which P(a) is true but P(b) is false. Therefore the above reasoning will not go through any more. This means that we need to consider lexical entailments one by one, and check if they are about the matrix time t or not. (But notice that lexical entailments are not equivalences of the type discussed in section 3.1.2 above, so it is still possible to show that that S itself is about its matrix predicate.) Let’s return to example (28a). As argued above, the presupposition triggering mechanism only needs to look at the set of lexical entailments. Below is a list of some of the intuitively plausible lexical entailments of (28a). (30)
Some lexical entailments of John knows at t1 that Mary is tired at t1 a. ϕ=John knows at t1 that Mary is tired at t1 b. ψ=John believes at t1 that Mary is tired at t1 c. χ=Mary is tired at t1 d. ξ=John’s belief is justified at t1
NB: It is not claimed here that a sentence such as (28a) can be ‘factorized’ into its constituent lexical entailments, nor is it assumed that there is a solution to the equation John 21
Thanks to T. Williamson (pc.) for helping me to see that an older, more complicated reasoning about logical entailments was neither necessary nor correct.
17
knows that p= John believes that p ∧ p ∧ X.(cf. e.g. Williamson 2002, Yablo 2008 etc. on the dangers of such an assumption.) The only claim made is that speakers have intuitive access to plausible lexical entailments. The above list merely provides examples of such entailments and is not meant to be an exhaustive definition of the meaning of S. Which of the above lexical entailments, if any, are predicted to be presupposed? Here is the idea. Let’s take a CT(S) representation of (31) (given in (31a)) such that t1 , t and t’ all refer to non-overlapping intervals. Given the lexical semantics of know, its complement, which denotes the proposition that it is raining, is entailed. Since the CT(S) equivalent of this entailment is not about the matrix event time t of CT(S), it is predicted to be presupposed.22 (31)
S: John knows (at t1 ) [that Mary is tired (at time t1 .)] a. CT(S): knows (John, tired(Mary, t’), t) (by the method of constructing canonical temporal sentences) b. S|=tired (Mary, t1 ) c. tired(Mary, t1 )=CTS tired(Mary, t’) d. tired(Mary, t’) is not about t (given the definition of aboutness in Section 3) e. therefore, S presupposes that Mary is tired at t1 .
Compare this with the entailment of S that John believes that it is raining. This proposition is not predicted to be presupposed, because its corresponding CT(S)-entailment is necessarily about the matrix event time of its CT(S): (32)
S: John knows (at t1 ) [that Mary is tired (at time t1 .)] a. CT(S): knows (John, tired(Mary, t’), t) b. S |=believes (John, t1 , Mary is tired (at t1 )). c. believes (John, t1 , tired (Mary, t1 )) =CTS believes (John, t, tired(Mary,t’)) d. believes (John, t, tired(Mary, t’)) is about the event time t e. therefore S does not presuppose that John believes that Mary is tired at t1
It seems that no other entailment in (30) is such that its CTS-equivalent would not be about t.23 But if we found other lexical entailments whose CTS-equivalents would not be about the event time of some CT(S), these entailments would also be predicted to be presupposed.(Cf. Section 5) 22
Since infinitival complements are treated the same way as tensed ones, the same reasoning can apply to languages where factive verbs can stand with infinitival complements, e.g. in Italian. Thus (i)
Gianni sa di essere un genio Gianni knows to be a genius
is predicted to presuppose that Gianni is a genius the same way as the entailed tensed complement in (31) is predicted to be presupposed. 23 Even is John might have justification for his beliefs at other times than the matrix time itself, the only thing that is entailed by the sentence is that he has some justification for his belief at the time of believing it.
18
Notice finally that existential entailments that we get by replacing the matrix tense argument in the original sentence by an existentially bound tense variable are predicted to be about the matrix tense of the CT(S). This is because existentially quantified sentences are about every individual in the domain (cf. the discussion in connection with (19) in the previous section) and thus for any CT(S), the CTS-equivalent of the existential entailment will be about the matrix CTS-argument as well. Therefore, this entailment is not predicted to be presupposed.24 4.2.2
Emotive Factive Verbs
Emotive factive verbs such as regret deserve a special mention. Originally, these verbs were assumed to be just like cognitive factives in presupposing the truth of their complement (cf. Kiparsky and Kiparsky 1970, Karttunen 1971b). However their factivity has been called into question by various authors, and it has been been suggested that rather they presuppose that the subject believes the truth of the embedded proposition, or something analogous to this. (cf. e.g. Klein 1975, Egr´e 2008, Schlenker 2003).25 This section aims to defend the position that emotive factives are indeed factives, which is also what the present theory predicts. The statement that these predicates are factive is in accordance with the intuition that from (33a) one tends to infer that it is raining and that (33b) seems to be contradictory. (33)
a. I doubt that John regrets that it is raining b. #It is not raining but John regrets that it is raining.
However there are examples, first put forward by Klein (1975), which seem to challenge this picture: (34)
a. b.
Falsely believing that he had inflicted a fatal wound, Oedipus regretted killing the stranger on the road to Thebes (Klein 1975, quoted in Gazdar 1979:122) John wrongly believes that Mary got married, and he regrets that she is no longer single. (Egr´e 2008, based on Schlenker 2003)
In these examples even though the first conjunct entails the falsity of the complement of regret, the sentences are acceptable. Does this mean that a factive analysis is untenable? Gazdar (1979) gave a negative answer to this question, suggesting that Klein’s (1975) 24
In the case of sentences such as (i) it has to be assumed that the the CT(S) construction mechanism replaces the temporal adverb itself: (i)
John knows that sometimes he is tired.
Admittedly, this is a somewhat stipulative move. Also, as T. Williamson (pc) points out, it still needs to be explained why sentences such as John knows that some times are better than others presuppose their complement. I leave this problem for the future. 25 Another possibility, suggested by T. Williamson (pc.), is that emotive factives presuppose that the subject knows that the embedded proposition is true. I leave the detailed discussion of this option for the future.
19
example involves free indirect speech, and therefore does not give sufficient grounds for the claim that verbs such as regret do not entail their complements. Thus, according to Gazdar, (34a) is acceptable because it reports a situation in which Oedipus is saying to himself “I killed that stranger”. He points out that similar examples are possible with aware: (35)
Falsely believing that he had inflicted a fatal wound, Oedipus became aware that he was a murderer.
Egr´e (2008) suggests in turn that although it is conceivable that (34b) could also be analyzed as involving free indirect speech, it still seems that substituting regret with know in this example results in a contradiction, which at least suggests that there is a difference between regret and know : (36) #John wrongly believes that Mary got married, and he knows that she is no longer single. Yet, under the right conditions it is not hard in fact to construct cases analogous to (34b) with know. (37)
a. b. c.
John suffers from paranoia. He falsely believes that the police is spying on him and what is more he knows they are listening to his phone calls. The keys were not in the drawer but she knew that they were there, so she foolishly kept on searching. It’s not what he doesn’t know that bothers me, it’s what he does know for sure but just isn’t true. [Will Rogers about Ronald Reagan, from Yablo 2008]
All these cases become acceptable once they are understood as reporting a firm belief or feeling of “knowledge” on somebody’s part. In this sense they are analogous to the free indirect discourse cases discussed above. If this is on the right track then it is possible to maintain that regret and emotive factive verbs are indeed factive, and what requires explanation is under what condition free indirect discourse is more (or less) easily available.26 26
There might further be cross-linguistic differences in the behavior of emotive factives. In French e.g. these verbs require that their complement be in the subjunctive, which might be a contributing factor as to why the free indirect discourse interpretation could be more easily available. Indeed the discussion of the meaning of (i) Egr´e (2008) says: “[(i) ] does not necessarily imply that Pierre knows that Marie left without saying good bye. What the sentence presupposes is that it seems (at least to some participants in the conversation) that Marie left without saying good bye. The best way to approach the correct meaning would be to say that “Pierre is surprised that Mary might have left without saying good bye”. (i)
Pierre est surpris que Marie soit partie sans dire au revoir Pierre is surprised that Mary besubj left without saying good bye
However, according to my intuition, the meaning described by Egr´e is much too weak for the Hungarian equivalent of be surprised, which is normally formed with the indicative. However, it becomes more plausible once the subjunctive, the use of which is slightly marginal, is used:
20
4.3
Change of State Verbs
The section first looks at regular change of state verbs such as stop, after which I turn to examining cognitive change of state verbs such as discover. As in the previous discussion, the reasoning presented in connection with these predicates will carry over to the whole class of change of state predicates. As it was argued above, to predict which entailments of the sentence will be presupposed, we only need to consider the set of lexical entailments. Consider now (38), in which t1 denotes the event time of the predicate, in this case the time of the stopping. Let’s assume that the lexical entailments of (38) are ϕ, ψ and χ: (38)
John stopped smoking at t1 a. ϕ= John does not smoke at t1 b. ψ=John smoked at t2 (where t2 is some contextually given interval before t1 ) c. χ=John stopped smoking at t1
The event time of S is denoted by t1 . Its denotation is given by the contextually supplied assignment function g, which assigns it an element from the domain of time intervals. In this example, the event time denotes the interval that starts just before the onset of non-smoking, and goes on for a certain, potentially very short time. In some other cases, it might be reasonable to assume that the event time also includes a longer segment of the stage where the previous activity is still going on. This second option might be more intuitive with gradual transitions, e.g. stop the car.27 However, even in this second case the sentence also entails that the previous state held before the event time. Notice that this contrasts with the inference that the final state continues to hold, which is not an entailment. This is shown by the difference in the acceptability of the examples in (39) below. (39)
a. #John stopped smoking, but he has never smoked before. b. John stopped smoking, but then he started again.
In the case of factive verbs it was the canonical temporal representation of the sentence that allowed the event time of the embedded clause to be different from that of the matrix event time, and thus presupposed. In the case of change of state verbs some entailments are lexically specified to be true at some time other than the event time. This is what happens in (38), where the the lexical entailment that John used to smoke at some time (ii)
a. b.
Meglep˝ o, hogy Mari k¨ osz¨ on´es n´elk¨ ul t´avozott. It is surprising that Mari without saying good bye left ?Meglep˝ o, hogy Mari k¨ osz¨ on´es n´elk¨ ul t´avozzon. It is surprising that Mari without saying good bye leavesubj
In the case of (iib), but not (iia), what is implied is that someone has suggested that Mary left without saying good bye, and thus can be understood as reporting a belief on somebody’s part. 27 Further, it might not be possible to pin down the exact moment when the ‘change’ starts: as it is shown very nicely in Landman (1991) Chapter 4, such attempts inevitably run into the vagueness problem.
21
preceding the event time comes from the lexical semantics of the change of state verb. As the CT(S) equivalent of this entailment is itself, and thus not necessarily about the event time of CT(S), it is presupposed. (40)
a. b. c. d. e. f.
S: John stopped smoking (at t1 ). CT(S): stopped (John, smoke, t) S |=smoke(j, t2 ), where t2 refers to some time interval preceding t1 smoke(j, t2 )=CTS smoke(j, t2 ) (because t2 is not a TS-argument) smoke(j, t2 ) is not about t (at least for some CT(S)) therefore, S presupposes that John used to smoke
The rest of the lexical entailments in (38) are all about t1 , and thus their CTS-equivalent will be about t. Given the reasoning in Appendix 2 which shows that closure of lexical entailments under logical consequence does not introduce any new presuppositions, no other entailments of (38) are predicted to be presupposed. One might wonder about the following entailment:28 (41)
John smoked at some time t2 before t1
This proposition is technically about the arguments, according to the definition of aboutness. Don’t (41) and (38b) express equivalent propositions? Given that possible worlds are defined by the combinatorial possibilities in the language, they in fact do not express logically equivalent propositions. The two might be contextually equivalent, there is nothing however in the present system that would prohibit some proposition entailed by a sentence S to be a presupposition, while a contextually equivalent proposition is not.29 4.3.1
Cognitive Change of State Verbs
Cognitive change of state verbs work on the one hand as regular change of state verbs (presupposing the truth of a previous state), and on the other hand as factive verbs (having a factive presupposition). These presuppositions arise in the same way as with change of state verbs and with factives, respectively, and are therefore straightforwardly predicted by the present analysis. This predicts correctly that (42a) and (42b)will be the presuppositions of (42). (42)
Peter discovered at t1 that Mary is tired at t1 a. Peter did not know that Mary is tired at t2 (where t2 < t1 ) b. Mary is tired at t1
The entailment in (42a) does not contain a TS-argument (and the tense argument is not quantified over) and is therefore predicted to be presupposed. The entailment in (42b) does contain a TS-argument but its CTS-equivalent is not about the event time of the 28 29
This question was raised by P. Schlenker (pc.) and O. Percus (pc.). Thanks to Daniel Rothschild for a helpful comment on this issue.
22
CTS representation, and is predicted to be presupposed for this reason. The example further illustrates how the two types of principal arguments are treated separately by the system.
4.4
Implicative verbs
As Karttunen (1971a) has showed, implicative predicates such as manage, remember, see fit are veridical, but they do not presuppose their complement. This would be a problem for the present proposal if we assumed that the temporal argument of the embedded clause can be replaced by a different CTS argument from the one that replaces the matrix CTS argument, because in that case the proposal would predict that (42) should presuppose the truth of its embedded complement.30 (43)
a. b. c.
John managed to solve the problem. John saw fit to apologize. John remembered to lock the door.
However, as Karttunen shows, the tense of the embedded predicate is not independent of the matrix tense, in the sense that it cannot be modified by independent temporal adverbials. Cf. the following examples: (44)
a. #John managed to solve the problem next week. b. #John saw fit to arrive day after tomorrow. c. #John remembered to lock his door tomorrow.
The above facts contrast with other veridical predicates that combine with infinitival clauses, where such modification is available:31 (45)
John was happy to arrive tomorrow.
Intuitively it is clear why we observe this restriction: managing to do something and doing it are really the same event, and therefore the two cannot be modified by independent temporal (or spacial) adverbials. This means that in the case of implicative verbs the temporal argument of the embedded clause is keyed to the tense of the matrix verb, and therefore it cannot be replaced by a new constant when the CTS is computed. One way to implement this idea is to assume that the lexical semantics of implicatives specifies that the tense argument of the embedded clause is obligatorily co-indexed with the tense argument 30
Thanks to Jacopo Romoli and Kyle Rawlins (pc.) for bringing this issue to my attention. Abusch (2004) claims that predicates such as known are incompatible with future frame adverbials modifying the main predicate of the complement. However, as Abusch notes, with such predicates still at least a scheduling interpretation is available, which is not the case with implicatives: 31
(i)
John is known to be at Mary’s place tomorrow.
I assume therefore that the temporal structure of such sentences must be richer than (46).
23
of the matrix clause. This is similar to the treatment of the controlled subject of embedded infinitival clauses, which also has to be co-indexed with the matrix subject: (46)
Johni managed tj [PROi solve the problem tj ]
Given this lexical property of implicatives, the CTS construction mechanism cannot replace the embedded tense argument with a new constant: it has to be replaced with the same constant as the one that replaces the matrix tense. As a result, the embedded complement is not predicted to be presupposed by the present system.
4.5
Discussion
The algorithm offered in the present section for deriving presuppositions that depend on verbal predicates makes predictions about the types of entailments of sentences that will be presupposed. This prediction is not necessarily exhaustive, in that it does not exclude that other mechanisms could predict further entailments to be presupposed as well.32 But since the proposal itself is a function that associates presuppositions to sentence meanings, it follows that two sentences that have the same meaning should also trigger the same presuppositions (cf. also Schlenker (2010) for discussion). This explains the lack of crosslinguistic variation as well as the observed facts of non-detachability that were mentioned in the introduction of this paper. I examine some predictions of the present proposal in more detail below. 4.5.1
Predictions
In general the theory makes the prediction that verbs that entail the truth of their propositional complement will also presuppose the truth of this complement, unless it is lexically specified by the matrix verb that tense argument of the embedded complement has to be coindexed with the matrix tense. This is because, if the tenses are in principle independent, there will always be a CT(S) representation such that the CTS-equivalent corresponding to the proposition denoted by the complement is not about the event time of the CT(S). This property predicts not only the presuppositions of traditional factive verbs, but also the presuppositions of examples such as (47), where the truth of the embedded infinitival complement, namely that Mary run, is presupposed. 32 E.g., this paper has nothing to say about existential presuppositions of noun phrases. Note that some cases of alleged verbal presuppositions might belong to the case of existential presuppositions as well. One such case might be the verb accompany, which has been argued in Abusch (2010) to be presuppositional: In particular (i) is claimed to presuppose that Mary went to the airport.
(i)
John accompanied Mary to the airport presupposition: Mary went to the airport
But it seems that this presupposition arises from a syntactically more complex sentence, namely John accompanied Mary, who was going to the airport. Since in this case we have a relative clause, the inference we observe might also be an existential presupposition
24
(47)
John forced Mary to run
The difference then between a factive verb such as know and a non-factive one such as believe, or verbs such as force as opposed to want reduces to the fact that the latter do not entail that their complement is true, i.e. their veridicality. Further, the system makes the prediction that any entailment that does not contain a matrix TS-argument and whose tense argument is not quantified over is presupposed. Conversely, entailments of a sentence S whose temporal argument is the TS-argument corresponding to the matrix CTS argument of the CTS representation are not predicted to be presupposed.33 Therefore it is predicted that presuppositions of change of state verbs will be about some time other than the event time of a change of state verb. As far as I know this prediction is borne out.34 Further, it is also predicted that entailments of change of state verbs that are not about the event time cannot be not presupposed. Recall that the inference we might get from stop that the final state continues to hold is not itself an entailment, unlike the inference that the previous state held for some time before the event time, as it was shown in (39). Interestingly, true entailments of change of state verbs about events that occur after the event time are very hard to come by. But thus does not mean that propositions about events that are true in the future of the matrix tense can not be presupposed. Indeed, (47) was exactly such a case. Thus genuine entailments which are interpreted as being true in the future of the event time are predicted to be presupposed as well. The theory also makes the prediction that if two aspectual verbs were to differ only in the lengths of the interval denoted by the event time, their corresponding presuppositions should also differ. This prediction seems to be borne out as well. The following are examples from Hungarian: 33 34
(i)
At least not by the present mechanism. Mandy Simons (pc) has asked what happens in the case of examples such as (i): a. b.
John is winning the race (at time t1 ). John is participating in the race (at time t1 ).
The claim is that the sentence presupposes that John is participating in the race at the time of winning it, yet this cannot be predicted by the present system since (ib) is about the event time. I believe however that what is going on here is that the sentence in fact presupposes that John participated before the onset of the winning, (i.e. he entered the race) but this presupposition further gives rise to a pragmatic inference that he continued to participate while he was winning as well. This pragmatic inference arises from a contextually warranted inertia assumption, namely the belief that if nothing intervenes states and processes continue. I believe that this claim is backed by the observation that there is a difference in the status of the two inferences: (ii)
A: Look, John is winning the race now! B: Hey wait a minute! I did not know he entered the race. B’: ? Hey wait a minute! I did not know he was supposed to be participating now.
25
(48)
a. b.
J´anos elj¨ott ’John came’ J´anos megj¨ott ’John came/arrived’
In the above examples the overall meaning I think is the same, but there is a difference (resembling the difference btw. come and arrive in English, respectively) in that the event time seems to denote a different interval. (49)
a. b.
Tegnap J´anos elj¨ott ’Yesterday John came’ Tegnap J´anos megj¨ott ’Yesterday John came/arrived’
In the case of elj¨ott, modification with yesterday suggests that the whole event of John leaving the point of departure and arriving at target location happened yesterday and is not compatible with a scenario in which he departed weeks ago. In contrast, if yesterday modifies a sentence with megj¨ott it suggests that the ending phase of John’s trajectory happened yesterday, but is compatible with him departing weeks ago. Correspondingly, the presuppositions of the two examples are different too, as predicted: The negation of (49a) suggests that John stayed at home, while the negation of (49b) suggests that he left home. Looking beyond the scope of change of state verbs, it is generally predicted that entailments of atomic sentences that are not about the event time will be presupposed. This prediction seems to be borne out too. An example might be the sentence with the simple transitive predicate kill such as (50). Some plausible lexical entailments might be ϕ, ψ and ξ as shown below. (50)
John killed Bill a. ϕ=John killed Bill at t1 b. ψ=Bill is dead at t1 c. χ=Bill was alive at t2 (where t2 refers to some time before t1 )
Among the above, χ is not about t1 . This means that for any CT(S), χ’s CTS-equivalent will be itself and therefore it will be possible to find CT(S) representations such that χ is not about their event time. Notice again the contrast between the entailment of (50) that Bill was alive at t2 and the inference that Bill continued to be dead after the event time of the killing. While the first is indeed a lexical entailment, the second is only a pragmatic inference that follows from our world knowledge, as shown by the difference in the acceptability of the following pair. (51)
a. #John killed Bill, but he was never alive before. b. John killed Bill, but then he resurrected.
26
A second example might be the case of response-stance predicates. These predicates are usually taken to assert the truth of some proposition p and presuppose that someone (usually other than the subject, though this might not be necessary) has asserted or proposed that p is false. An example of such a verb is deny: (52)
John denied that it was raining a. ψ=John asserted at t1 that it was not raining b. ϕ=Someone(other than John) asserted at t2 that it was raining (where t2 is some interval before t1 )
Similarly as we have seen in the case of change of state verbs, ϕ is not about the event time t1 , and is therefore predicted to be presupposed, which seems to be correct. The present theory predicts presuppositions based on the meaning of atomic sentences and predicates. But notice that it is a separate question of why some predicates but not others happen to entail what they entail, for example why know but not believe happens to be veridical. Answering this second question would mean giving a general theory of concept formation. This daunting question however is beyond what a presupposition triggering theory could aspire to do. 4.5.2
Challenges? Fillmore (1971) and Abusch (2002,2010)
The prediction that atomic sentences that have the same meaning should trigger the same presuppositions seems largely correct. However, some cases have been offered in the literature that might seem to challenge this prediction. Fillmore (1971) has argued that there was a near-symmetry between the predicates accuse-criticize, in that ‘a accused b of p’ presupposed that a judged the action denoted by p bad, and asserted that b did p, ‘a criticized b for p’ presupposed that b did p and asserted that a indicated that p was bad. On closer observation though this near-symmetry turns out to be imprecise. First, it is not true that both predicates contribute the meaning that a judges/indicates that p is bad. While this might be approximately correct for criticize, accuse seems to contribute a stronger meaning suggesting that p is reproachable in general, as can be seen from the examples below. (53)
a. b.
John criticized Mary for cleaning the bathroom. John accused Mary of cleaning the bathroom.
Further, while criticize seems to entail and presuppose the truth of the embedded complement, accuse does not seem to entail the truth of the embedded complement, only that the subject asserted it. Interestingly, given these refinements the present proposal actually predicts the correct presupposition facts for the above pair: accuse triggers the sortal presupposition that p is reproachable, but does not in fact entail the truth of its complement, while criticize entails and presupposes the truth of its complement, but does not entail that p is morally reproachable, only that the subject finds it undesirable.
27
A more serious candidate for a pair where the same overall meaning might stand with different presuppositions was put forth by Abusch (2002, 2010). She has argued that the pairs be right-be aware are symmetric in the following way: (54)
a.
b.
John is right that dinner is ready asserts: Dinner is ready presupposes: John believes that dinner is ready John is aware that dinner is ready asserts: John believes that dinner is ready presupposes: Dinner is ready
However, as argued in Schlenker (2008, 2010) it seems that syntactically the two do not behave alike, and that (54)-a is syntactically more complex, akin to (55): (55)
John is right in claiming that dinner is ready
As Schlenker convincingly shows, the syntactic and semantic difference among the two predicates seems to be supported by various syntactic diagnostics such as weak crossover facts. This is a step towards distinguishing the two predicates, yet it does not fully grant the right predictions for the present proposal. While it is correctly predicted that be aware should presuppose ψ in (56), the sentence with is right is predicted to presuppose both ϕ and ψ, but it only presupposes ϕ. (56)
a. b.
ϕ=John claimed that dinner was ready (at some time t2 before the event time) ψ= Dinner is ready
One possibility for why the predicted presupposition ψ is not manifest is that it is canceled as in most embedded contexts it would be incompatible with the assertion or some implicature of the assertion. Thus (57a) asserts that John made an incorrect claim, while (57b,c) imply that the speaker does not know whether John’s claim is correct. Each of these cases is incompatible with the presupposition that dinner is ready. (57)
a. b. c.
John is not right that dinner is ready Is John right that dinner is ready? If John is right that dinner is ready, we should proceed to the dining hall.
Symmetric pairs, if real, would pose a challenge to any theory that attempts to derive presupposition triggering on the basis of the meaning of S alone. If the arguments in this paper and Schlenker (2010) are on the right track, to date no really convincing case has been found.
28
5
Adding Common Knowledge: Sortal Presuppositions
The triggering mechanism proposed in the previous sections was viewed as a function that takes as its input the bivalent meaning of a sentence S, and outputs one or more entailments of S. This section looks at cases that might neccesitate casting a wider net. It will be proposed that in certain cases at least, the input to the triggering mechanism is not only the set of entailments of the literal meaning of a sentence S, but the set of entailments of S given general common knowledge. The triggering mechanism will then operate on this contextually enriched set of entailments.35 Suppose we added the following lexical entailment, as seems reasonable, to the set of lexical entailments of (58): (58)
a. b.
John knows that Mary is tired at t1 John is sentient at t1
According to our rules, the CTS equivalent of (58b) is about the matrix event time of the CTS representation of (58a). Thus (58b) is not predicted to be a presupposition. However it was argued in the literature (cf. Simons 2001, e.g.) that the proposition that John is sentient is in fact a presupposition of sentences such as (58a). Indeed it seems that selectional restrictions in general should be treated as presuppositions (cf. McCawley 1968 and also Magidor 2007 and Asher 2010 for recent discussions). Further, it seems that most if not all sortal presuppositions have the property of being about the event time of the matrix predicate, thus it looks as if our system is unable to generate sortal presuppositions. We might ask however whether (58b) itself corresponds to the sortal presupposition of S. It seems more likely that the true sortal presupposition of S is not the episodic statement in (58b), but rather the characterizing statement that John is (in general) sentient. (59)
John is sentient (in general)
Is (59) itself a lexical entailment of (58a)? It is hard to tell, but we might safely assume that at least it is an entailment of (58b) given common knowledge (cf. Magri 2009). I.e. given world knowledge, speakers can be expected to assume that if John is sentient at a certain time t, then he is sentient in general. Given this it is also safe to say that (59) is an entailment of (58a) given common knowledge. Interestingly, the present proposal does predict (59), the contextual entailment of (58a) to be a presupposition. Here is why. Characterizing statements such as the above are typically assumed (cf. Chierchia 1995) to contain an instance of the generic operator Gen, and are thus analyzed as follows: (60)
Gen s [C(j,s)] [sentient (j,s)]
The variable C in the restriction of Gen expresses the property of being at an arbitrary 35
The previous, more lexical version of this section was rethought after skeptical comments of Nicholas Asher, for which I am grateful.
29
location. Thus (60) expresses that whenever John might be located in some situation s, he is a sentient being in s. For our present purposes, I will simplify the above by saying that the Davidsonian argument simply ranges over times t.36 (61)
Gen t [C(j,t)] [sentient (j,t)]
Now it is well known that generic statements, although in many ways similar to simple universal quantification over times, cannot be simply equated with universal quantification since unlike the latter, generic statements allow exceptions. Thus (62), but not (63) is contradicted by the existence of some after-dinner time at which John does not smoke. (62)
John smokes after dinner Gen t [after-dinner(t)∧ C(j,t)] [smokes (j,t)]
(63)
John always smokes after dinner ∀t [after-dinner(t)∧ C(j,t)] [smokes (j,t)]
In fact not only exceptions must be allowed, but not even a single instantiation in time is needed for the truth of a generic statement, as shown by the famous example (cf. Krifka et al. 1995), which can be true even if no mail from Antarctica has ever arrived. (64)
Mary handles the mail from Antarctica.
Similarly, (59) might be truthfully asserted even if John is lying in coma. Recall now that according to our definitions universal statements are about every individual: E.g. the sentence Everyone is tired is about Fido, if Fido is an element of the domain. This is because it is possible to choose two worlds w, w’ which are Fido-variants such that one of them make the assertion true and the other makes it false: suppose every individual including Fido is tired in w, and every individual except Fido is tired in w’. Then Everyone is tired is true in w, but false in w’, and thus the sentence is about Fido. Indeed by the same reasoning the universal statement in (63) above is about every individual in the domain. However, as it was argued above, generic statements are not equivalent to universal statements about times t in a given domain. According to our reasoning, the CTS equivalent of the contextually entailed generic statement is itself. However we cannot conclude that a generic statement such as (59) is about the event time of the CT(S). So while the CTS equivalent of the entailment (65a) is about the event time of the CT(S) of (58a), and is therefore not presupposed, the contextually entailed generic statement in (65b) cannot be proved to be about the matrix event time of the CT(S) of (58a), and is therefore presupposed. (65)
a. b.
John is sentient at t1 John is sentient (in general)
Sortal presuppositions of a sentence S can then be predicted by the present system as 36
cf. also Magri (2009) for a similar assumption.
30
presuppositions that arise from the set of entailments of S given by common knowledge.37 Looking further at the nature of sortal presuppositions, it is interesting to note that (65b) is very close in meaning to the (generic) modal statement below38 : (66)
John is able to have beliefs and feelings
Magidor (2007) has argued that when we try to formulate sortal presuppositions of predicates precisely, it often seems we have to settle for some modal statement. Such an argument was presented for the case of the sortal presupposition of pregnant in Magidor (2007:154): The presupposition of ‘x is pregnant’ turns out to particularly hard to formulate. In analogy with other cases above, I am inclined to resist the suggestion that the presupposition is that x is female (Consider the non-anomaly of an expert biologist uttering: ‘We now have a method which enables men to be pregnant as well. This man, for example, is pregnant’). Another suggestion might be that the presupposition is that x is an animal, but this hypothesis seems to too general as it fails to explain why in many standard contexts ‘This man is pregnant’ is anomalous. If English had a word such as ‘un-pregnant’ (the correlate of ‘unhappy’) we would have probably been able to express the presupposition as being that x is pregnant or unpregnant. But one only needs to reflect on how difficult it is to explain the exact truth-conditional meaning of ‘unhappy’ to see that this observation is not of much help. I think that roughly, the presupposition generated by ‘x is pregnant’ is that x can be pregnant. But the exact force of the modality of ‘can’ needs to be considered carefully. It cannot be so restrictive as to make (6-20) suffer from a presupposition failure: (6-20) In spite of trying for many years, my aunt isn’t pregnant. In fact, the doctors said that she cannot become pregnant. On the other hand, it cannot be so permissive as to include any metaphysical possibility. For it is at least conceivable that someone who is a man in the actual world is a pregnant woman in another (metaphysically) possible world. Even so, it still seems that in contexts where it assumed that men cannot be pregnant while remaining men (6-21) would suffer from a presupposition failure: 37
Restrictions on the grammatical category of verbal complements have been sometimes claimed to be presuppositions as well–e.g. whether a certain verb selects for an interrogative or declarative complement.(This was pointed out to me by L. McNally) I believe however that this is a misnomer, and is an example of the technical usage of the word ‘presupposition’ to designate some property for which we do not have an explanation. 38 It has been often noted in the literature (cf. Dahl 1975, Men´endez-Benito 2005) that modal statements expressing ability or possibility are very similar in meaning to dispositional/characterizing generic sentences, as shown by the following pair: (i)
a. b.
This car goes 200mph. This car can go 200mph.
31
(6-21) This man is pregnant.
With respect to the type of modality involved in sortal presuppositions we might assume that it is either a purely circumstantial existential modal or in some cases an ability modal. However, modal statements expressing ability, similarly to other types of modal statements, are also temporally sensitive in that the presence of an ability might vary at different times: (67)
John could cycle with the speed of 50 mph after he ate the breakfast of the champions.
What is the time then of the modal presupposition proposed by Magidor (2007)? Again, the literal lexical entailment of ‘x is pregnant at t1 ’ is presumably that x can be pregnant at t1 . However, x having the ability of being pregnant at t1 entails, given common knowledge, that x can be pregnant in general. The precise formulation of this latter statement should also involve a generic operator, just as in the case of characterizing sentences above: (68)
Gen t [ C(x,t)] [can be pregnant (x,t)]
Similarly to the reasoning presented above, it is this generic statement that will be predicted to be the presupposition of ‘x is pregnant’ by the present theory.39
6
Context Sensitivity?
This paper proposed a semantic mechanism to predict why certain entailments of sentences are presupposed. Alternatively, one might think of the present mechanism as one that predicts why certain entailments become lexicalized as presuppositions. The mechanism takes general common knowledge into account, as argued in the previous section, but otherwise it is not sensitive to the nature of context in which the sentence appears. Indeed it operates on atomic sentences and the presuppositions of complex sentences are assumed to be derived by applying a separate projection mechanism. Thus entirely different sets of rules determine how presuppositions are generated, and how they are transmitted. This is in contrast with Stalnaker (1974), Simons (2001) who assume that the triggering mechanism is a pragmatic mechanism operating on complex sentences, and also with Schlenker (2010) who goes even one step further and argues that the triggering mechanism should be sensitive to both the linguistic and extralinguistic context in which the expression triggering the presupposition appears. This section examines the various reasons that might support a pragmatic analysis. 39
Similar considerations might apply to the presuppositions of verbs such as manage, whose presupposition might be argued to express a dispositional statement as well: (i)
a. b.
John managed to solve the exercise presupposes The exercise is hard for John (in general)
32
As has been long observed, a pragmatic explanation of verbal presuppositions in particular might be supported by the fact that these presuppositions are less stable and fairly easily suspendable (cf. Stalnaker 1974Chierchia and McConnell-Ginet 2000, Kadmon 2001, Simons 2001, Abbott 2006, Abusch 2010). An example at hand might be the case of stop, of which the following example illustrates a non-presuppositional use. (69)
I notice you are chewing on your pencil. Have you recently stopped smoking? (Geurts 1994)
These cases might suggest that verbal presuppositions are a pragmatic phenomena, though it should be noted that other explanations that do not invoke pragmatic considerations are possible too (cf. Klinedinst 2009). A second, related, reason that might be an argument for the pragmatic analysis is the observation that cases of presupposition suspension and canceling often seem to depend on the presence of focus (cf. Beaver 2004). Thus (70b), but not (70a) suggests that the student has plagiarized his work, (as well as a sense of complicity between speaker and student): (70)
a. b.
If I discover that your work is [plagiarized]F , I will be [forced to notify the Dean]F . If the T.A. [discovers]F that your work is plagiarized, I will be [forced to notify the Dean]F .
Though it is not hard to imagine explanations where focus somehow suspends existing lexical presuppositions, it might at least be theoretically more elegant if focus interacted with the triggering mechanism itself. At the same time is it also well known that presuppositions are a somewhat fickle phenomenon, which makes it hard to draw far-reaching conclusions from the above observations. A more compelling reason for a pragmatic mechanism is provided by the cases of part time triggers noted in Schlenker (2010, 2008). Schlenker argues that there are expressions that trigger a presupposition when they appear in certain contexts but not in others. Such cases, he argues, point towards a triggering theory that predicts presuppositions not only based on the meaning of the expressions involved, but based on the literal meaning of the expressions together with the linguistic and extra-linguistic context in which these expressions appear. An example of such an expression, which Schlenker terms ‘part-time trigger’, is the verb announce. In some contexts, it does not entail the truth of its complement and in these contexts it does not presuppose the truth of its complement either. In other contexts, it entails and presupposes the truth of its complement. Let’s look at an example: (71)
Mary has announced that she is pregnant a. Scenario 1 : Mary is 30 years old and she is expected to be reliable. Therefore the context entails the truth of the embedded proposition. → (71) presupposes that Mary is pregnant b. Scenario 2 : Mary is 7 years old and is not expected to be reliable. Therefore 33
the context does not entail the truth of the embedded proposition. → (71) does not presuppose that Mary is pregnant As Schlenker points out, the verb announce contrasts minimally with the verb inform, which seems to lexically entail and presuppose the truth of its complement in the above context. It would not be impossible to extend the mechanism proposed in this paper to handle the above facts as they appear above. What would be needed is to allow contextual entailments to enter the pool of candidate entailments for presuppositions. Then if the embedded proposition is contextually entailed, it would also be predicted to be presupposed. Otherwise it would not. However, the situation is more complex. The complication arises from examining cases, discussed in Schlenker (2006), which show that inform itself is a part time trigger. In contexts in which the subject is not only assumed to be reliable, but extra-reliable, the factive inference of both announce and inform disappears. Look at the context below: (72)
George is the family butler. He is very reliable. If he says p, then we can infer that p is the case, and if he does not say p, we can infer that p is not the case. Has George announced/informed the guests that dinner is ready? → there is no implication that dinner is ready.
Plausibly, what happens in this case is that the verb in effect becomes a (quasi-)negraising verb when the context implies that the “middle” proposition, (not saying p and p) is excluded. (cf. Gajewski’s (2005) reasoning about the connection between the excluded middle and neg-raising in the case of “classic” neg-raising predicates such as want). While the triggering mechanism argued for in this paper could be made compatible with the facts in (71), it could not cover all the facts in (71) and (72). This is because even if we extended the mechanism to apply to contextually given entailments, it could not be sensitive to the presence of an additional contextual entailment to block the presupposition from arising. Notice that Schlenker’s (2010) formulation of the desiderata for a triggering theory is compatible with all the above facts. The present paper can thus be seen as part of the following picture. There is a lexical/semantic process of presupposition triggering, which is sensitive to world knowledge but otherwise insensitive to contextual effects. Further effects of presupposition-like phenomena such as the cases of part-time triggers described above must then result from a separate pragmatic mechanism that triggers presupposition-like inferences based on contextual factors. The workings of this second, yet to be discovered, mechanism are plausibly very different from the triggering mechanism offered here, for example because this mechanism will have to yield non-monotonic inferences, as shown above.
34
7
Conclusion
This paper proposed a mechanism for deriving the presuppositions of verbs. It was argued that an entailment p of an atomic sentence S is presupposed, if it is possible to find a CT(S) representation of S such that the CTS equivalent of p is not about the event time of CT(S). Further, it was proposed that in some cases the triggering mechanism can take into account not only the semantic entailments of a sentence, but also contextual entailments that arise from common knowledge. The proposed mechanism also made a number of predictions about the type and form of possible verbal presuppositions. These predictions seem to be correct.
35
APPENDIX: Demolombe and Fari˜ nas del Cerro (2000) The intuitive idea of Demolombe and Fari˜ nas del Cerro (2000)40 is that the truth value of a sentence that is not about an entity should not change if we change the truth value of the facts about that entity. To capture this intuition, they give a proposal that has two parts. The first is the definition of variants of an interpretation with respect to an object. Given this notion, the property of a sentence being about an object can be defined.
Variants of an interpretation with regard to an object Roughly speaking the notion of variants of an interpretation with regard to an object denoted by constant symbol c is the set of interpretations M c that only differ from M by the truth assignment of atomic sentences where c appears as an argument. Let Lc be a first order predicate calculus language that contains the constant c and does not contain the identity predicate.41 M’ is a c-variant of a model M iff it meets the constraints listed below: • DM’ =DM • iM’ =iM , for every variable symbol and constant symbol • iM’ is defined from iM for each predicate symbol as follows: if p is a predicate symbol of arity n – if t is an n-tuple of terms of language Lc that contain no occurrence of the constant symbol c, then iM’ (t)∈ iM’ (p) iff iM (t)∈ iM (p). – if an element hd1 , ..., dn i of Dn is such that for every j in [1,n], dj 6= iM (c), then hd1 , ..., dn i ∈ iM’ (p) iff hd1 , ..., dn i ∈ iM (p). M c will be used to denote the set of c-variant interpretations M’ defined from M. An example: Let Lc be a language with a unique unary predicate symbol p, and the constant symbols a, b, c. Let M be an interpretation of Lc defined by: D = {d1 , d2 , d3 , d4 }, iM (a) = d1 , iM (b) = d2 , iM (c) = d3 and iM (p) = {d1 , d3 , d4 }. For every variant M’ in M c , iM’ (p) contains d1 , because iM (p) contains d1 and d1 is the interpretation of the constant symbol a, which is different from constant symbol c. Therefore the sentence p(a) is true in every variant M’. At the other extreme, there are variants M’ of M such that d3 is not in iM’ (p), because d3 is the interpretation of c. In these variants p(c) is false, although it is true in M. 40
Cf. also Demolombe and Fari˜ nas del Cerro (2010) for a more recent exposition of the key ideas present in Demolombe and Fari˜ nas del Cerro (2000). 41 Though cf. Demolombe and Fari˜ nas del Cerro (2000) for some suggestions on how the identity predicate could be handled.
36
Aboutness Let F be a sentence of language Lc . F is not about an object named by the constant symbol c iff for every interpretation M, M |= F iff for every interpretation M’ in M c M 0 |= F : (73)
N A(F, c) holds iff ∀M (M |= F iff ∀M 0 ∈ M c M 0 |= F )
A formula F is about an object named by c if it is not the case that N A(F, c): (74)
A(F, c) holds iff ∃M (∃M 0 ∈ M c (M |= F and M 0 6|= F ))
Examples Now consider the following examples: (75)
S=Fido is tired
The sentence in (75) is about Fido, if there is an M’∈ MFido , such that M |= S and M’6|= S. Suppose that L is a language with a unique unary predicate symbol is tired, and the constant symbols Fido, John, Mary. Let M be an interpretation of L defined by: 1. D = {d1 , d2 , d3 , d4 }; 2. iM (Fido) =d1; iM (Mary) = d2; iM (John) = d3 3. iM (is tired) = {d1 , d3 , d4 } From this model, a Fido-variant M’ can be defined, where iM’ (is tired) = {d3 , d4 }. Thus M |= S and M’6|= S. Consider now: (76)
S=John is tired
The example in (76) is not about Fido, because for every M’∈ MFido , M |= S, and further for any model M, st. M|=S, for every M’∈ MFido , M’|=S. (77)
Fido is tired or Fido is not tired
The disjunction above is not about Fido, because it is a tautology hence is true in every model M. Interestingly, (78)
Some individual is tired
is about Fido, because there is an M’∈ MFido , such that M |=F and M’6|= F: suppose originally in our example we had iM (is tired) = {d1}, and iM’ (is tired) = ∅.
37
References Abbott, B. (2000). Presuppositions as nonassertions. Journal of Pragmatics 32 (10), 1419–1437. Abbott, B. (2006). Where have some of the presuppositions gone? In B. J. Birner and G. Ward (Eds.), Drawing the boundaries of meaning: Neo-Gricean studies in pragmatics and semantics in honor of Laurence R. Horn, pp. 1–20. Philadelphia: John Benjamins. Abusch, D. (2002). Lexical alternatives as a source of pragmatic presuppositions. In B. Jackson (Ed.), Proceedings of SALT XII. Ithaca, NY: CLC Publications. Abusch, D. (2004). On the temporal composition of infinitives. In J. Gu´eron and J. Lecarme (Eds.), The Syntax of Time. Cambridge, Mass.: The MIT Press. Abusch, D. (2010). Presupposition triggering from alternatives. Journal of Semantics 27 (1), 37–80. Asher, N. (2010). Lexical Meaning in Context: A Web of Words. Cambridge University Press. Beaver, D. (2004). Have you noticed that your belly button lint colour is related to the colour of your clothing? In R. Bauerle, U. Reyle, and T. E. Zimmerman (Eds.), Presupposition: Papers in Honor of Hans Kamp. Beaver, D., J. Tonhauser, M. Simons, and C. Roberts (2010). What projects and why. In N. Li and D. Lutz (Eds.), Proceedings of Semantics and Linguistic Theory 20. Cornell: Ithaca. Carlson, L. (1983). Dialogue games: An approach to discourse analysis. Kluwer Academic Publishers. Chemla, E. (2007). Similarity: Towards a unified account of scalar implicatures, free choice permission and presupposition projection. To appear in Semantics and Pragmatics. Chierchia, G. (1995). Individual level predicates as inherent generics. In M. Krifka, F. Pelletier, G. Carlson, A. ter Meulen, G. Chierchia, and G. Link (Eds.), The Generic Book. Chicago: The University of Chicago Press. Chierchia, G. and S. McConnell-Ginet (2000). Meaning and grammar: an introduction to semantics. MIT Press Cambridge, MA, USA. Dahl, O. (1975). On generics. In E. Keenan (Ed.), Formal Semantics of Natural Language, pp. 99–111. London and New York: Cambridge University Press. Demolombe, R. and L. Fari˜ nas del Cerro (2000). Towards a logical characterization of sentences of the kind “sentence p is about object c”. In S. Holdobler (Ed.), Intellectics and computational logic: papers in honor of Wolfgang Bibel. Kluwer Academic Press. Demolombe, R. and L. Fari˜ nas del Cerro (2010). Information About a Given Entity: From Semantics Towards Automated Deduction. Journal of Logic and Computation. Demolombe, R. and A. J. Jones (1996). On sentences of the kind “sentence ‘p’ is about topic t”: some steps toward a formal-logical analysis. In H.-J. Ohlbach and U. Reyle (Eds.), Logic, Language and Reasoning. Essays in Honor of Dov Gabbay, pp. 115–133. Kluwer Academic Press. Egr´e, P. (2008). Question-Embedding and Factivity. In F. Lihoreau (Ed.), Grazer Philosophische Studien 77, pp. 85–125. Fillmore, C. (1971). Types of lexical information. In D. Steinberg and L. Jakobovits (Eds.), Semantics. An interdisciplinary reader in philosophy, linguistics and psychology, pp. 370–392. Cambridge: Cambridge University Press. Gajewski, J. (2005). Neg-Raising: Polarity and Presupposition. Ph. D. thesis, Massachusetts Institute of Technology.
38
Gazdar, G. (1979). Pragmatics: Implicature, Presupposition and Logical Form. Academic Press, New York. Geurts, B. (1994). Presupposing. Ph. D. thesis, University of Osnabr¨ uck. Geurts, B. and R. van der Sandt (2004). Interpreting focus. Theoretical Linguistics 30 (1), 1–44. Goodman, N. (1961). About. Mind 70 (277), 1–25. Grice, H. (1981). Presupposition and Conversational Implicature. Radical Pragmatics, 183–198. Groenendijk, J. and M. Stokhof (1984). Studies in the Semantics of Questions and the Pragmatics of Answers. Ph. D. thesis, University of Amsterdam. Hamblin, C. (1973). Questions in Montague English. Foundations of Language, 41–53. Heim, I. (1983). On the projection problem for presuppositions. Proceedings of WCCFL 2, 114–125. Humberstone, L. (2000). Parts and partitions. Theoria 66 (1), 41–82. Kadmon, N. (2001). Formal Pragmatics: Semantics, Pragmatics, Presupposition, and Focus. Blackwell Publishers. Karttunen, L. (1971a). Implicative verbs. Language 47 (2), 340–358. Karttunen, L. (1971b). Some Observations on Factivity. Papers in Linguistics 5, 55–69. Karttunen, L. (1977). Syntax and semantics of questions. Linguistics and Philosophy 1 (1), 3–44. Karttunen, L. and S. Peters (1975). Conventional implicature in Montague Grammar. In BLS 1: Proceedings of the First Annual Meeting of the Berkeley Linguistic Society, pp. 266–278. Berkeley, CA. Karttunen, L. and S. Peters (1979). Conventional implicature. In C.-K. Oh and D. A. Dinneen (Eds.), Syntax and Semantics, Volume 11: Presupposition, pp. 1–56. New York: Academic Press. Keshet, E. (2008). Infinitival Complements. In A. Gronn (Ed.), Proceedings of Sinn und Bedeutung 12. Kiparsky, P. and C. Kiparsky (1970). Fact. In M. Bierwisch and K. Heidolph (Eds.), Progress in linguistics, pp. 143–173. The Hague:Mouton. Klein, E. (1975). Two sorts of factive predicate. Pragmatics Microfiche it 1.1. frames B5–C14. Klinedinst, N. (2009). Totally Hardcore Semantic Presuppositions. ms. UCL. Krifka, M., F. Pelletier, G. Carlson, A. ter Meulen, G. Chierchia, and G. Link (1995). The Generic Book. Chicago: The University of Chicago Press. Kusumoto, K. (2005). On the quantification over times in natural language. Natural Language Semantics 13 (4), 317–357. Landman, F. (1991). Structures for Semantics. Springer. Levinson, S. C. and E. Annamalai (1992). Why presuppositions aren’t conventional. In R. N. Srivastava (Ed.), Language and text: Studies in honour of Ashok R. Kelkar, pp. 227–242. Delhi: Kalinga Publications. Lewis, D. (1988). Relevant Implication. Theoria 54 (3), 161–174. Magidor, O. (2007). Category Mistakes. D. Phil dissertation, University of Oxford. Magri, G. (2009). A theory of individual-level predicates based on blind mandatory scalar implicatures. Natural Language Semantics 17 (3), 245–297. McCawley, J. D. (1968). Concerning the base component of a transformational grammar. Foundations of Language 4, 243–269. Men´endez-Benito, P. (2005). The Grammar of Choice. Ph. D. thesis, University of Massachusetts Amherst.
39
Partee, B. H. (1973). Some structural analogies between tenses and pronouns in English. The Journal of Philosophy 70 (18), 601–609. Portner, P. and K. Yabushita (1998). The semantics and pragmatics of topic phrases. Linguistics and Philosophy 21 (2), 117–157. Putnam, H. (1958). Formalization of the concept “About”. Philosophy of Science 25 (2), 125–130. Reinhart, T. (1981). Pragmatics and linguistics: An analysis of sentence topics. Philosophica 27, 53–94. Roberts, C. (1996). Information structure in discourse: Towards an integrated formal theory of pragmatics. Working Papers in Linguistics-Ohio State University Department of Linguistics, 91–136. Rooth, M. (1992). A theory of focus interpretation. Natural Language Semantics 1 (1), 75–116. Ryle, G. (1933). About. Analysis 1 (1), 10–11. Schlenker, P. (2003). The Lazy Frenchman’s approach to the Subjunctive. Proceedings of Romance Languages and Linguistic Theory, 269–309. Schlenker, P. (2006). Presupposed Entailments: The Triggering Problem Revisited. Talk presented at Sinn und Bedeutung 11. Schlenker, P. (2008). Be Articulate: A Pragmatic Theory of Presupposition Projection. Theoretical Linguistics 34 (3), 157–212. Schlenker, P. (2010). Local Contexts and Local Meanings. Philosophical Studies. Simons, M. (2001). On the conversational basis of some presuppositions. In J. B. Hastings, R. and S. Zvolenszky (Eds.), Semantics and Linguistic Theory, 11. Sperber, D. and D. Wilson (1979). Ordered entailments: An alternative to presuppositional theories. Syntax and Semantics XI: Presupposition, 299–323. Stalnaker, R. C. (1974). Pragmatic presuppositions. Semantics and Philosophy, 197–214. Strawson, P. (1964). Identifying reference and truth-values. Theoria 30, 96–118. Reprinted in D. Steinberg & L. Jacobovits, eds. Semantics, 86-99. Williamson, T. (2002). Knowledge and its Limits. Oxford University Press, USA. Yablo, S. (2008). Semantic Arithmetic. available at http://www.mit.edu/~yablo/semanticarithmetic.pdf.
40
