Presupposition, Assertion, and Lexical Items

transcript

Presupposition, Assertion, and Lexical ItemsAuthor(s): Deirdre WilsonSource: Linguistic Inquiry, Vol. 6, No. 1 (Winter, 1975), pp. 95-114Published by: The MIT PressStable URL: http://www.jstor.org/stable/4177858 .

Accessed: 14/06/2014 14:55

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact support@jstor.org.

The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to Linguistic Inquiry.

http://www.jstor.org

This content downloaded from 91.229.248.187 on Sat, 14 Jun 2014 14:55:37 PMAll use subject to JSTOR Terms and Conditions

Linguistic Inquiry Volume VI Number I (Winter, 1975), 95-114.

Deirdre Wilson Presupposition, Assertion, and Lexical Items*

In the considerable recent literature on presuppositions, two main approaches can be distinguished. On the first approach, presuppositional analysis lies within the scope of truth-conditional semantics; here presupposition is contrasted with entailment and defined in logical terms. On the second approach, presuppositional analysis lies within the scope of a use theory of meaning; here presupposition is contrasted with assertion and is defined in terms of appropriateness conditions for utterances. In what follows I shall be overtly concerned with the presupposition-assertion contrast. However, much of what I say has bearing on the presupposition-entailment contrast too, and I make some explicit remarks towards the end of the article about this.

The presupposition-assertion approach to semantic description rests on roughly the following assumptions. Every sentence S has two sets of associated conditions: the first expresses the semantic content of speech acts such as asserting, requesting information, and ordering, which S may be used to perform; the second expresses preconditions on the use, or the appropriate use, of S to perform a speech act at all. (I refer to the first type of conditions as assertions and to the second as appropriateness conditions or presuppositions.) Furthermore, every lexical item has two types of associated features: first, features that contribute to the assertions made by sentences in which it occurs; second, features that contribute to the appropriateness conditions of sentences in which it occurs. (I call the first type of features assertionfeatures and the second type presupposition features.) The job of a semantic description is to associate sentences with their assertions and appropriateness conditions on the basis of semantic rules sensitive, among other things, to the assertion features and presupposition features of lexical items.

Although there is much variation in detail among the proposals of individual writers using the presupposition-assertion distinction, I think that I have brought out above their common central claim: namely, that the distinction between assertions and appropriateness conditions is relevant for semantic description. For example, Kiparsky and Kiparsky (I97I, 349) say:

* This article is based on a chapter of my MIT Doctoral dissertation (Wilson I973) . I would like to thank Noam Chomsky, Sylvain Bromberger, and Paul Kiparsky for many valuable suggestions and criticisms.

96 DEIRDRE WILSON

In formulating the semantic structure of sentences, or ... the lexical entries for predicates, we must posit a special status for presuppositions, as opposed to what we are calling assertions.

Fillmore (I969, I 20) makes similar remarks:

I shall deal with a distinction between the presuppositional aspects of the semantic structure of a predicate on the one hand and the "meaning" proper of the predicate on the other hand. We may identify the presuppositions of a sentence as those conditions which must be satisfied before the sentence can be used in any of the functions just mentioned [ = "asking questions, giving commands, making assertions, expressing feelings, etc."].

Many other writers use a comparable distinction (cf. for example Langendoen and Savin (I97I, 55), Zwicky (I97I, 73)). This article is designed to show that analyses based on the presupposition-assertion distinction as at present conceived are grossly inadequate for purposes of semantic description.

The Framework for Presuppositional Analysis

It is generally held that there are considerable distributional differences between presuppositions and assertions. For example, presuppositions are held to remain constant under negation, while assertions are not. Indeed, in most treatments this difference is taken as criterial, as it is by Fillmore (I969, I23):

In the best-known meaning of BACHELOR ... only the property of "having never been married" is part of the meaning proper. Uses of this word (as predicate) presuppose that the entities being described are human, male and adult. We know that this is so because the sentence (i)

(i) That person is not a bachelor

is only used as a claim that the person is or has been married, never as a claim that the person is a female or a child. That is, it is simply not appropriate to use (i), or its nonnegative counterpart, when speaking of anyone who is not a human, male adult.

Fillmore's approach would yield analyses like the following, where the presupposition remains constant under negation and only the assertion is negated:

(I) My neighbor is a bachelor. Asserts: My neighbor is unmarried. Presupposes: My neighbor is human, adult, and male.

(2) My neighbor is not a bachelor. Asserts: My neighbor is not unmarried. Presupposes: My neighbor is human, adult, and male.

It is also generally assumed that the presuppositions of a given declarative sentence carry over intact to related nondeclaratives, yielding analyses like the following:

PRESUPPOSITION, ASSERTION, AND LEXICAL ITEMS

(3) Is Bill's teacher a bachelor? Asks: Is Bill's teacher unmarried? Presupposes: Bill's teacher is adult and male.

(4) Try to remain a bachelor. Orders: Try to remain unmarried. Presupposes: You are adult and male.

Finally, it is generally assumed that the presuppositions of a simple sentence will carry over intact to some or all complex sentences in which that sentence occurs embedded, conjoined or disjoined. On this assumption there is again a marked difference between the behavior of presuppositions and assertions. The words bachelor and spinster, for example, carry the assertion feature unmarried. This does not mean, however, that any declarative sentence in which these words occur may be used to assert that someone is unmarried. In (5)-(7) the words occur, but no such assertion would be made by use of these sentences:

(5) Either Bill's great-aunt Japonica is a spinster, or her marriage certificate was lost in the Great Fire of London.

(6) If your teacher was a bachelor, you were extremely lucky. (7) Sebastian thinks my husband is a bachelor.

In no circumstances could (5)-(7) be used to assert that someone was unmarried. In fact, on the analysis of bachelor and spinster given above, the assertions made by use of (5)-(7) will be those in (8)-(Io):

(8) Either Bill's great-aunt Japonica is unmarried, or her marriage certificate was lost in the Great Fire of London.

(g) If your teacher was unmarried, you were extremely lucky. (Io) Sebastian thinks my husband is unmarried.

In other words, assertion features remain within the scope of logical connectives and embedding predicates and do not function to make independent assertions.

Although this point about assertion features is obvious, it does raise the possibility that if an assertion feature like unmarried does not necessarily involve an assertion that someone is unmarried, then a presupposition feature like male need not necessarily involve a presupposition that anyone is male. As mentioned above, this approach to presuppositions is not the one generally adopted. It is more normal for those who use the presupposition-assertion distinction to treat presuppositions, unlike assertions, as preserved under some or all types of embedding-as falling outside the scope of some or all logical connectives and embedding predicates. It is, of course, on the basis of such differences in distribution that the distinction must ultimately be justified. I shall attempt to argue here that a number of theories based on the assumption that

98 DEIRDRE WILSON

these differences exist fail to do justice to the bahavior of presuppositions, both on simple and on complex sentences. In other words, I shall argue that there is no distributional basis for the presupposition-assertion distinction.

Summarizing the points I want to concentrate on, I think it can fairly be said that most people who use the presupposition-assertion distinction subscribe to one or more of the following hypotheses about the nature and behavior of presuppositions:

(i) The Detachability Hypothesis. According to this hypothesis, presuppositions never contribute to the assertive content of an utterance, but merely constrain its appropriateness. It follows that presuppositions and assertions must be separately represented. For example, Zwicky (I97I, 77) explicitly adopts this position, saying that "the presuppositions of the assertions are represented independently of the meanings of the assertions". It is also implicit in such typical analyses as that found in Kiparsky and Kiparsky (I97 I, 350), shown here as (i i) below:

(i i) Mary cleaned the room. Asserts: (a) Mary caused the room to become clean.

(b) The room became clean. Presupposes: The room was not clean.

Here the semantic content of the assertion is very clearly divorced from its presuppositional content.

(2) The Cumulative Hypothesis. According to this hypothesis, if a sentence S presupposes another sentence P, then any complex sentence in which S occurs as a subpart will also presuppose P. In other words, complex sentences presuppose what their constituent sentences presuppose. The Kiparskys (I97I) and Langendoen and Savin (I97I) subscribed to this hypothesis; (3) below presents a view that is currently more fashionable.

(3) The Filtering Hypothesis. This is a catchall term intended to cover various modifications of the cumulative hypothesis, so that certain presuppositions of embedded sentences need not emerge as presuppositions of the complex sentences in which they occur. A wide range of proposals have been made, including appeals to ambiguity, presupposition-cancelling or suspending devices, and plugging and filtering along the lines of Karttunen (I973). As far as I know, all those who use the presupposition- assertion distinction have adopted the detachability hypothesis and either the cumulative or the filtering hypothesis. I want to show that these hypotheses are inadequate to the facts and to draw some general conclusions from their inadequacy. I start with a consideration of complex sentences; the arguments will then be generalized to cover other cases.

Objections to the Presupposition-Assertion Approach

I. Arguments from Complex Sentences

Consider the following sentences:

(I 2) Your teacher will be either a bachelor or a spinster. (I 3) If your teacher was a bachelor you were lucky, but if your teacher was a

spinster you were unlucky.

I take it that (i!2) and (I3) are well-formed sentences and could be appropriately used to make assertions. If we take seriously the detachability and cumulative hypotheses, we analyze (I 2) and (I 3) as follows:

(i2) Asserts: Your teacher will be either unmarried or unmarried. Presupposes: (a) Your teacher will be adult and male.

(b) Your teacher will be adult and female.

Here the presuppositions are contradictory and the assertion is redundant. ( I3) presents worse problems, since on a similar analysis it will have both contradictory presuppositions and a contradictory assertion:

(I 3) Asserts: If your teacher was unmarried you were lucky, but if your teacher was unmarried you were unlucky. Presupposes: (a) Your teacher was adult and male.

(b) Your teacher was adult and female.

On the assumption that the presuppositions of a sentence must be true if it is to be capable of making an appropriate assertion, (I 2) and (I 3) will thus be predicted as making necessarily inappropriate assertions-a prediction that is clearly false.

It might be argued that the detachability hypothesis can be preserved and the problem of contradictory presuppositions eliminated if the cumulative hypothesis is dropped in favor of a filtering version. For example, one might postulate that disjunctive sentences have disjunctive presuppositions, so that (I 2) is reanalyzed as follows:

(I2') Your teacher will be either a bachelor or a spinster. Presupposes: Your teacher will be either adult and male or adult and female.

While this eliminates the contradictory presuppositions on (I 2), it does not eliminate the redundancy of the assertion. Nor will it yield a correct representation of sentences only one disjunct of which carries a presupposition about age or sex, as for example (I4):

(I 4) Either the victim was a bachelor or no one got killed at all. On the assumption that the second disjunct of this sentence carries no presuppositions

100 DEIRDRE WILSON

about age or sex, even on the disjunctive analysis (I4) will have (I5) as one of its presuppositions:

(I5) Presupposes: The victim was adult and male.

Since (I5) itself would standardly presuppose that there was a victim, it seems to be clearly wrong as an appropriateness condition on asserting (I4).

However, I think it is the analysis of (I3) that conclusively demonstrates the inadequacy of this particular approach. On my original analysis of (I3), both the assertion and the presuppositions were contradictory. Extending the view that disjunctive sentences have disjunctive presuppositions, we may assume that conditional sentences have conditional presuppositions and reanalyze the presuppositions of (13) as follows, thus eliminating one of the contradictions:

(I 3') If your teacher was a bachelor you were lucky, but if your teacher was a spinster you were unlucky. Presupposes: If your teacher was adult and male you were lucky, but if your teacher was adult and female you were unlucky.

Unfortunately, in addition to the fact that (I3) still has a contradictory assertion, the presupposition stated in (I3') totally misrepresents the semantic content of (I3).

(I 3) says that if your teacher was a bachelor you were lucky, but makes no claim at all about your position if your teacher was a married man. The presupposition in (I3') states that if your teacher was male, whether married or unmarried, you were lucky. This is clearly quite wrong as an appropriateness condition on asserting (I3).

The analysis of spinster is similar. The obvious way to resolve this dilemma about presuppositions, and at the same

time eliminate the contradictory assertion, is simply to represent (13) as follows:

(I3") If your teacher was a bachelor you were lucky, but if your teacher was a spinster you were unlucky. Asserts: If your teacher was adult, unmarried, and male you were lucky, but if your teacher was adult, unmarried, and female you were unlucky.

This representation, which seems to me the only adequate one, has a number of unfortunate consequences for the presuppositional approach. In the first place, it eliminates the presuppositions. In the second place, it treats the presupposition features on bachelor and spinster in exactly the same way as it treats their assertion features, leaving them in their embedded positions and thus involving a rejection of the cumulative hypothesis and also one version of the filtering hypothesis. In the third place, it treats the presupposition features and assertion features of bachelor and spinster as contributing equally to the assertion made by use of (I 3), thus involving a rejection of the detachability hypothesis, according to which presupposition features contribute

to the appropriateness conditions alone, while assertion features determine the whole semantic content of the assertion. At the very least, it seems that appropriateness conditions may on occasion contribute to speech-act content, and that any adequate filtering hypothesis must serve, not to cancel presuppositions entirely, but rather to incorporate them into the assertions made by use of complex sentences. It seems to me that if these conclusions can be generalized to other analyses of lexical items in terms of presupposition and assertion, then the whole approach will have to be radically rethought.

The presuppositional analysis of factives and aspectuals seems to me to run into the same difficulties as that of bachelor and spinster. Langendoen (I97I), who rejects the presuppositional approach to nouns, nonetheless concludes that it is still valid for verbs. I think that this conclusion is mistaken. On the presuppositional approach, (I6), for example, would be given the following analysis:

(I6) If Sartre knows that Chomsky is alive I'll be surprised but if he knows that Chomsky is dead I'll be amazed. Asserts: If Sartre is aware [nonfactive] that Chomsky is alive I'll be surprised, but if he is aware that Chomsky is dead I'll be amazed. Presupposes: (a) Chomsky is alive.

(b) Chomsky is dead.

This analysis would predict, falsely, that (I6) could never be appropriately used, since its presuppositions can never be true together. Similar examples can be constructed with regret and realize. Using conditional presuppositions would again misrepresent the content of (i 6), giving it (i6') as an appropriateness condition:

(I6') Presupposes: If Chomsky is alive I'll be surprised, but if Chomsky is dead I'll be amazed.

Clearly the surprise mentioned in (i 6) is not caused by Chomsky's being alive, but by Sartre's being aware of it. Hence (i 6') does not give a correct representation of (i6).

Similarly, (I7) would be given the following analysis:

(I 7) If Bill stopped smoking at midnight I'll be surprised, but if he started smoking at midnight I'll be amazed. Asserts: If Bill didn't smoke after midnight I'll be surprised, but if Bill smoked after midnight I'll be amazed. Presupposes: (a) Before midnight Bill was smoking.

(b) Before midnight Bill was not smoking.

Here too the presuppositions are contradictory. Again, the use of conditional presuppositions will misrepresent the content of (I 7):

I02 DEIRDRE WILSON

(I 7') Presupposes: If Bill smoked before midnight I'll be surprised, but if Bill didn't smoke before midnight I'll be amazed.

This is a particularly interesting example, since it seems to me that there is no way to capture a fairly standard interpretation of the sentence while holding the presuppositions distinct from the assertion. On this interpretation, my surprise would be caused by the fact that Bill had not smoked after midnight-part of the assertive content on the above analysis-while my amazement would be caused by finding out that after all Bill had never smoked-part of the presuppositional content on the above analysis. I cannot see any way of capturing this interpretation except by eliminating the presuppositions entirely and representing (I7) as (I7"):

(I 7") Asserts: If Bill smoked before midnight and didn't smoke after midnight I'll be surprised, but if Bill didn't smoke before midnight and smoked after midnight I'll be amazed.

It is well known that the sort of argument I have just presented for conditionals and disjunctions generalizes to other types of complex sentence containing "presupposition-carrying" items. The most obvious case is that of verbs of saying or re- porting speech. Thus (i 8) cannot presuppose (i 9), for (i 8) may be both true and appropriate while (i9) is necessarily false:

(i8) John said he regretted that Mary was dead, while Bill said he regretted that Mary was alive.

(I9) Mary was dead and Mary was alive.

On the basis of similar examples, Karttunen (I973) abandons the cumulative hypothesis, proposing instead a filtering analysis that will be discussed later. Similarly, there are many negative sentences that must have a nonpresuppositional interpretation; (20) and its putative presupposition (2i) are a case in point:

(20) John doesn't know that Nixon is dead, because Nixon isn't dead. (2 I) Nixon is dead.

Here the cumulative hypothesis would predict that (20) cannot be appropriately used unless Nixon is dead; but this is clearly a wrong prediction. Again, various filtering versions have been proposed to explain these facts, and I shall return to them later. In the meantime, though, I think it can safely be concluded that both the detachability hypothesis and the cumulative hypothesis are inadequate for the treatment of a wide range of complex sentences.

2. Generalization to Nondeclaratives and Simplex Sentences

In the last section I tried to show that in many cases, whatever the analysis of related simple sentences, complex sentences cannot be treated as carrying presuppositions.

I now want to generalize these conclusions in two directions: first, to the analysis of nondeclaratives, and second, to the analysis of simple sentences.

Arguments parallel to those used in the last section apply to the presuppositional treatment of the speech acts of questioning and ordering. On a presupposition-assertion analysis, (22) and (23) would be represented as follows:

(22) Point out to Jemima that Bill's teacher is a bachelor. Orders: Point out to Jemima that Bill's teacher is unmarried. Presupposes: Bill's teacher is adult and male.

(23) Is Bill's teacher a bachelor? Asks: Is Bill's teacher unmarried? Presupposes: Bill's teacher is adult and male.

Now clearly the content of the order expressed by (22) need not be merely thatJemima be given the information that Bill's teacher is unmarried. If it were, this would be a striking confirmation of the presuppositional analysis. Unfortunately the order expressed by (22) is that Jemima be made aware that Bill's teacher has the set of prop- erties unmarried, adult, and male. Hence, even if two of these features contribute to the appropriateness condition of the speech act of ordering that (22) is used to perform, they must also contribute to the semantic content of the speech act whose appropriateness they are supposed to constrain. Even allowing for such a double function, a presuppositional analysis of (22) will often make wrong predictions about the content of a given speech act. Imagine a situation in which all teachers are unmarried and all teachers are therefore known to be either bachelors or spinsters. The point of (22)

when used to make an order in such circumstances would clearly be that Jemima be told that Bill's teacher is male, which she may not know, rather than that he is unmarried, which would be common knowledge.

The analysis of (23) predicts that it is only possible or appropriate to ask whether someone is a bachelor when one knows or assumes that he is adult and male; or on some treatments, when he is in fact adult and male. In either case the prediction is false. In the situation imagined above, where all teachers are unmarried, the point of asking the question expressed by (23) would be to find out whether Bill's teacher was male. Again, it would be wrong to set up the analysis so that this question would be void or inappropriate unless either one knew or assumed that its answer was yes or its answer was in factyes. Moreover, on the above analysis the answeryes would merely confirm that Bill's teacher was unmarried rather than, as is needed in these imaginary circumstances, that he was male. Further problems are created by the interpretation of the answer no to (23). On the above analysis, this would mean unequivocally that Bill's teacher was adult, male, and married; in other words, it would preserve the presupposition of the question. However, (24) is a perfectly possible and appropriate response to (23):

(24) No, Bill's teacher is a spinster.

I04 DEIRDRE WILSON

If this response is to be allowed on a presuppositional analysis, this will involve postu- lating two possible interpretations of a negative response to a question. More seriously, if (23) can be appropriately asked when Bill's teacher is not male or is not known or assumed to be male, then it will be necessary to postulate two senses of questions: one presupposition-carrying and one non-presupposition-carrying. In the latter, at least, the alleged presuppositions will behave exactly like standard speech-act content conditions, and not according to either the detachability hypothesis or the cumulative hypothesis. Again, sentences like (25) confirm that the detachability and cumulative hypotheses lead to wrong results:

(25) Is Bill's teacher a bachelor or a spinster? Asks: Is Bill's teacher unmarried or unmarried? Presupposes: (a) Bill's teacher is adult and male.

(b) Bill's teacher is adult and female.

Hence these hypotheses seem to break down in the case of nondeclaratives in exactly the same ways as they break down in the case of declaratives.

As before, this conclusion can be generalized to other lexical items that would have been given presuppositional treatment. To take just one example, there is no presumption in (26) that Bill left, although on the presuppositional analysis it presupposes that Bill left:

(26) Does John regret that Bill left, or didn't Bill leave? Asks: Is John sorry that Bill left, or didn't Bill leave? Presupposes: Bill left.

In (27) one encounters the problem of contradictory presuppositions:

(27) Does John regret that he left, or that he didn't leave? Asks: Is John sorry that he left, or that he didn't leave? Presupposes: (a) John left.

(b) John didn't leave.

Again, one can imagine circumstances in which one could ask a question containing regret, in order to discover whether its complement was in fact true, rather than presupposing that it was true, as the above analysis dictates. Suppose that I know that for all P, if P is true, then John regrets that P. Then the point of asking a question like (28) might very well be to find out whether Bill left, rather than to find out anything about John's feelings.

(28) Does John regret that Bill left?

A speech act of this type is predicted as impossible by the proposed presuppositional analysis, which requires that the truth of the complement of regret be taken for granted if an appropriate speech act is to result.

Finally, these conclusions about complex declaratives and nondeclaratives may be generalized to simple sentences in the following way. (29) and (30) are valid arguments, one containing a disjunctive premise and one containing a conditional premise of the type already discussed.

(29) a. Your teacher will be either a bachelor or a spinster. b. Your teacher will not be a bachelor. c. Therefore your teacher will be a spinster.

(30) a. If your teacher was not a bachelor, you were unlucky. b. Your teacher was not a bachelor. c. Therefore you were unlucky.

According to Fillmore's analysis, (30) can be appropriately used only if the teacher in question was adult and male, since premise (b) presupposes this. This is simply false. (30) has no implications whatsoever about the sex of the teacher in question. According to Fillmore's analysis, (29) can never be appropriately used, since if the presupposition on (b) is satisfied, the presupposition on (c) will be automatically violated. Yet (29) is a paradigm case of rational argumentation; a seunantics that either marks it as necessarily inappropriate, or even deprives it of the ability to assert anything at all, is a bad semantics.

The conclusion drawn from (29) and (30)-that their simplex sentences cannot carry presuppositions-is confirmed by an argument adapted from Geach (i965). If (29) and (30) are valid arguments, then the words bachelor and spinster must have the same meaning in both the premises and the conclusions, if the fallacy of equivocation is to be avoided. An argument whose premises and conclusion punned on the alternative readings of bachelor as 'unmarried adult male' and 'young fur seal when without a mate during the breeding season' would clearly be invalid. Fillmore talks of distinguishing the presuppositions of a lexical item from its "meaning proper", where the meaning proper of bachelor is simply 'unmarried'. I have argued that the occurrence of bachelor in the first premises of arguments (29) and (30) cannot be analyzed in terms of presupposition and assertion; and that the "meaning proper" of the word in these premises must be 'unmarried, adult, and male'. But since the arguments are valid ones, the meaning proper of the word bachelor must remain constant throughout the premises and the conclusion. Hence, the occurrence of the word in the second premises and the conclusions must also be analyzed as having the meaning proper of 'unmarried, adult and male'. In this way the conclusion about the inadequacy of the presupposition-assertion distinction in the analysis of embedded sentences can be generalized to simplex sentences too.

Parallel arguments may be constructed for other lexical items. The following is an example:

(3I) a. IfJohn regrets that Nixon is dead, then Nixon is dead.

io6 DEIRDRE WILSON

b. Nixon is not dead. c. Therefore John does not regret that Nixon is dead.

Here, on a presuppositional analysis, the presuppositions of (3 I a) and (3 I c) will contra- dict (3 I b). Moreover, since it can be shown by arguments already used that (3 I a) does not carry a presupposition, it should follow that (3I c), a simple sentence, does not carry one either, or else that (3 I) commits the fallacy of equivocation.

More generally, one wants a semantics in which sentences such as (32)-(34) will be marked as necessary truths:

(32) If your neighbor is a bachelor, then your neighbor is male. (33) If John stopped smoking at midnight, then John had smoked before

midnight. (34) If Bill regrets that Fred left, then Fred left.

On the proposed presuppositional analysis, (32)-(34) will carry contingent presuppositions. But if a sentence carries a contingent presupposition, it cannot be a necessary truth. Hence the presuppositional approach must be wrong.

My arguments so far have been mainly designed to establish the following points. The cumulative hypothesis is invalid for a wide range of sentences, including conditionals and disjunctions, as well as sentences with such predicates as say and claim. In the cases in which the cumulative hypothesis is invalid, the putative presuppositions of the sentence in question do not constrain the appropriateness of the associated speech act but rather contribute directly to the assertion, thus refuting the detachability hypothesis. One version of the filtering hypothesis, whereby disjunctive sentences have disjunctive presuppositions and conditional sentences have conditional presuppositions, also seems to be inadequate. The conclusion for complex sentences of the type discussed must be that they either do not, or need not, carry presuppositions. Furthermore, this conclusion about lack of presuppositions on complex sentences can be generalized by simple arguments to apply to nondeclaratives and to simple sentences, showing that they too either do not, or need not, carry presuppositions. I now consider some proposals designed to avoid the conclusions I have drawn so far.

Some Modified Proposals

Fillmore (I969, I 22) offers an objection to some of the above arguments:

Certain apparent counterexamples to the claims I have been making about presuppositions can be interpreted as "semiquotations. " I believe some utterances are to be thought of as comments on the appropriate use of words. Uses of the verb chase presuppose that the entity defined as the direct object is moving fast. Uses of the verb escape presuppose that the entity identified by the subject noun-phrase was contained somewhere "by force" previous to the time of focus . . . It seems to me that sentences like (I) and (2)

are partly comments on the appropriacy of the words chase and escape for the situations

being described. These are sentences that would most naturally be used in contexts in which the word chase or escape had just been uttered:

(i) I didn't "chase" the thief; as it happened, he couldn't get the car started. (2) I didn't "escape" from the prison; they released me.

Kiparsky and Kiparsky (I97I, 35I) say very similar things about the presupposition- denying negation of the verb clean:

If you want to deny a presupposition, you must do it explicitly: Mary didn't clean the room: it wasn't dirty Abe didn't regret that he had forgotten: he had remembered.

The second clause casts the negative of the first into a different level; it's not the straightforward denial of an event or situation, but rather the denial of the appropriateness of the word in question-such negatives sound best with the inappropriate word stressed.

The common view expressed here is that sentences that fail to carry the predicted presuppositions are really denials of the appropriateness of the word that would standardly carry the presupposition. Fillmore sees such uses as semiquotations and as most naturally applied in rejecting a prior claim, while the Kiparskys see them as naturally accompanied by heavy stress on the inappropriate word and add that the sentences must contain an explicit rejection of the presupposition that would otherwise be conveyed.

This last claim is clearly false. I have given many examples of sentences that fail to carry the predicted presuppositions but that contain no explicit statement that the presuppositions are not true. (35) is a further example:

(35) I didn't clean the bathroom: I cleaned the kitchen.

(35) neither suggests that the bathroom was dirty nor explicitly states that it was not. One generally accepted solution to this problem is to treat negatives as ambiguous between presupposition-carrying and non-presupposition-carrying senses. (35) may then be seen as a non-presupposition-carrying negation and perfectly explicable as such. One might go on to claim that such negations should be seen as denials of appropriateness. Notice, however, that sentences rather like (35) can undergo conjunction reduction to yield (36) and (37):

(36) I cleaned not the kitchen but the bathroom: the kitchen wasn't dirty anyway.

(37) I cleaned the bathroom and not the kitchen.

It is a condition on conjunction reduction for these sentences that the deleted occurrence of the verb clean must be semantically identical to the one retained. Hence, if clean does not carry a presupposition in the negative clause-which conveys no sugges- tion that the kitchen was dirty-then it cannot do so in the positive clause. Similarly, if clean is not a semiquotation in the positive clause-which it clearly is not-then it

io8 DEIRDRE WILSON

cannot be one in the negative clause either. Moreover, even if some such account works for the problematic negative sentences, it does not follow that all cases in which a predicted presupposition fails to show up may be successfully treated as denials of appropriateness. It is hard to see how positive conditionals and disjunctions, of which I have given many relevant examples, can be subjected to such treatment. (38), which contains a modal, does not carry the predicted presuppositions either and can in no way be construed as a denial of appropriateness:

(38) Heath may at this very moment be chasing Wilson down Pall Mall-it's hard to know what politicians will do next.

As far as I can see, there is no presumption in (38) that Wilson is in fact running down Pall Mall at the moment, contrary to what Fillmore's analysis would suggest.

It would, of course, be possible to claim that non-presupposition-carrying negatives are denials of appropriateness, while non-presupposition-carrying conditionals are hypothesizations of appropriateness, modals are affirmations of the possibility of appropriateness, and questions are requests for answers about appropriateness. This naturally raises the question of whether there is any difference between a denial of appropriateness and a standard denial, a hypothesization of appropriateness and a standard hypothetical, and so on. Suppose that if an activity is truly to be called chasing, then both the person chasing and the object chased must be moving. If one of these conditions is not met, then the activity cannot truthfully (or appropriately) be called chasing. On the account offered by Fillmore and the Kiparskys, one of these conditions is given a favored treatment: if the person said to be chasing is not moving, then the statement is appropriate but false, while if the object said to be being chased is not moving, then the statement is said to be inappropriate or on another level. But this account will only be accepted if we have independent reasons for calling some truth conditions pragmatic presuppositions and denying this label to others. I have been arguing that there is in fact no justification for doing this. Hence the way out offered by Fillmore and the Kiparskys will only be accepted by one who already believes in pragmatic presuppositions; for someone who has been convinced by my arguments it will have no attractions at all.

Karttunen (I973) proposes a general treatment of complex sentences that allows for widespread loss of presuppositions in certain specified circumstances. I summarize the main points of his treatment here.' First, he argues that negative sentences are ambiguous between presupposition-carrying and non-presupposition-carrying senses. This would dispose of my negative counterexamples to the cumulative hypothesis. Second, he argues that certain predicates are plugs, which "block off all the presuppositions of the complement sentence". These predicates include the verbs say, mention, tell, ask, promise, and in general all verbs that introduce or describe reported

1 For detailed discussion of Karttunen's proposals, see Wilson (1973, Chapter II).

speech. This would dispose of my example (i 8). Third, he introduces a special filtering mechanism designed to deal only with the logical connectives if. . . then, and, and either .., or. It is this mechanism that would have to deal with my disjunctive and conditional counterexamples. According to Karttunen, filters are "predicates which, under certain conditions, cancel some of the presuppositions of the complement". The main body of his article is devoted to describing the conditions under which such cancellation occurs. His major claim is that there is an asymmetry between the behavior of presuppositions on the first and second clauses of conditionals, disjunctions, and conjunctions. Presuppositions on the first clauses can never be cancelled, but if the presupposition on a second clause is either entailed or presupposed by the first clause (in the case of conditionals and conjunctions) or by the negation of the first clause (in the case of disjunctions), then it is automatically cancelled.

If this asymmetry can be established, the presuppositional approach to semantics will be automatically justified. I have set out to show that there is no difference in the distributional behavior of presuppositions and assertions. Since there is no asymmetry condition for assertions, it should follow on my account that there is none for presuppositions. And in fact it can be shown that there is none. The conditional and disjunctive counterexamples I have used were constructed so as to fall outside the range of Karttunen's filtering mechanism. Thus even this modified treatment still makes the wrong predictions about them. The only way to set up the filter so that it makes the correct predictions is in fact to abandon the asymmetry condition, so that presuppositions on the first clauses of compound sentences may drop under just the same circumstances as those of the second clauses.2 But this move, while it would dispose of the remaining counterexamples, would also involve the admission that there is no distributional difference in the behavior of presuppositions and assertions and hence destroy the main justification for the presuppositional approach to semantics.

My argument against the filtering hypothesis can perhaps be put in another way. Within this modified theory, three different explanations are given for the fact that a presupposition associated with a given simple positive sentence fails to appear when that sentence is embedded in various ways. If the resulting sentence is negative, the presupposition need not appear, because negatives are ambiguous, and in one of their senses they do not carry presuppositions. If the resulting sentence has a main clause predicate that is a plug, then the presupposition never appears at all. There is no question of ambiguity; the presupposition simply vanishes. If the resulting sentence is a conjunction, conditional, or disjunction, then whether the presupposition appears or not will depend on the type of conditional, etc. Some such sentences always carry presuppositions, and others never do. Thus three different mechanisms are invoked to explain what seems to be clearly a unitary phenomenon: the failure of predicted presuppositions to appear.

2 In more recent work, Karttunen has modified his position to allow for symmetrical cancellation and also for contextual factors to play a much greater role in presupposition cancellation. See also Liberman (I 973).

110 DEIRDRE WILSON

In fact, for one who insists on retaining the presuppositional approach to semantics, the best solution would be to postulate a full-scale theory of ambiguity between presupposition-carrying and non-presupposition-carrying senses, along the lines of the standard treatment of negation. Both Karttunen's plugging mechanism and his filtering mechanism for logical connectives, whether symmetrical or asymmetrical, seem to me to have unfortunate consequences. For example, if (39) and (40) carry no presuppositions, as in most currently accepted analyses, then it is hard to see why they would quite often be interpreted as suggesting (4I) and (42), respectively:

(39) John says the Beatles regret that they were awarded an Oscar. (40) Marina says James realizes that his arguments are unsound. (41) The Beatles were awarded an Oscar. (42) James's arguments are unsound.

The fact that these sentences carry suggestions will now have to be given a quite different explanation from the normal account: that sentences suggest what they presuppose. On the other hand, if one simply treated (39) and (40) as ambiguous, as negations are treated as ambiguous, then the presuppositional approach to suggestions could be maintained. Similar accounts would have to be given of questions and orders.

I think an ambiguity analysis should also be preferred to Karttunen's account of logical connectives as filters. For example, if (43) carries no presupposition that the war is over, there are also many circumstances in which (44) would be treated as carrying no such presupposition; and by natural extension, there are also circumstances in which (45) could be interpreted in the same way:

(43) If Jowett knows that the war is over, then the war is over. (44) If Jowett knows that the war is over, I'll give you five dollars. (45) If Jowett knows that the war is over, Bill should leave at once.

And in general it seems to be the case that a given conditional or disjunction has two possible interpretations: one on which it carries a presupposition, and one on which it does not. Karttunen's account denies this; according to him, (43) will carry no presupposition, but (44) and (45) will. Again, the claim that conditionals and disjunctions are ambiguous would permit a unitary account of what seems to be a unitary phenomenon: that certain sentences have preferred interpretations on which they carry "presuppositions", but that they have other interpretations on which they do not. And of course it would be possible to go on to claim, as I would like to do, that the place for a theory of preferred interpretations is not in semantics at all, but in prag- matics. Karttunen's recent writings indicate that this is what he too believes.

Paul Kiparsky (personal communication) has proposed a very interesting alternative to the view that presuppositions are cancellable-a view that would accommodate the fact that "cancelled presuppositions" are not eliminated, but be-

come part of the content of the speech act performed by use of their associated sentence. Kiparsky's proposal is that instead of a cancellation mechanism, what is at work is a "perfectly general mechanism" whereby presuppositions may be promoted to become part of the assertion made. This mechanism would work in both positive and negative sentences, and it operates as follows. (46) has three slightly different possible interpretations, as indicated in (47)-(49):

(46) John stopped reading War and Peace. (47) John stopped reading War and Peace-as opposed to Bartholomewe Fair. (48) John stopped-as opposed to continued-reading War and Peace. (49) John stopped reading War and Peace-as opposed to taking up opium.

(47) and (48) might be treated as presupposing that John had been reading War and Peace and asserting that he is no longer reading it. In (49), however, one might feel that two assertions are made: one that John had been reading War and Peace, and the other that he is no longer reading it. Hence (46) might be treated as either presupposition-carrying or not presupposition-carrying, depending on the exact interpretation intended. Given this treatment, there is now an explanation for the alternative interpretations of negatives, questions, conditionals, modals, and disjunctions. These interpretations will themselves be parasitic on the two available interpretations of positives, and they will be expected to have both presupposition-carrying and non-presupposition-carrying uses on exactly the same pattern as their related positives.

Given that this proposal could be mrade explicit, it would yield the most adequate treatment of pragmatic presuppositions that I know of. Many of my arguments so far have been directed towards establishing what it assumes: namely, that a wide range of clause types that may be interpreted as presupposition-carrying have alternative, non-presupposition-carrying interpretations. Other of my arguments have been aimed at establishing that presuppositions may or must contribute to speech-act content and truth conditions; again Kiparsky's proposal accommodates this, since in his system certain conditions may function at one time as truth conditions and assertions and at another time as logical and pragmatic presuppositions. Given that the general mechanism for converting presuppositions into assertions can be made explicit, the question now arises of whether it is possible to choose between Kiparsky's presuppositional system and the approach I would propose, which would not use presuppositions at all and would treat negatives, etc. as vague or indeterminate as to their possible interpretations, rather than being systematically ambiguous.

I believe that there is one consideration that argues decisively against any system based on analyzing lexical items into pragmatic presuppositional elements and speech- act content elements, Kiparsky's system included. What is common to the approaches I have been considering here is the view that the syntactic categories of declarative, interrogative, and imperative are typically associated with certain speech acts: say

112 DEIRDRE WILSON

asserting, requesting information, and ordering. In giving the semantic analyis of a word, one examines the conditions in which it is "normally" used to perform the associated speech act. For example, one does not "normally" assert or deny that someone is a bachelor unless one is fairly sure that he is an adult male. One does not "normally" assert or deny that someone stopped doing something unless one is fairly sure that he has done it in the past. Hence, it must be mentioned in the semantic description of bachelor that it is appropriately used only of adult males, and in the semantic description of stop that it is appropriately used only of an action that has been done before. This conclusion, I think, is mistaken.

It is probably true that there are certain felicity conditions attached to the speech act of assertion. If I utter a declarative sentence without having evidence for its truth, for example, then what I say may not count as an assertion but rather as a guess or a fantasy. If I utter an imperative sentence without believing that what it enjoins can be carried out, then it may not count as an order, but rather as an ex- hortation, a wish, or part of an experiment. The declarative or imperative sentence is itself quite neutral as between the various speech acts which it may be used to perform, each of which will have its own idiosyncratic appropriateness conditions. What seems to me to be a fundamental mistake in the presupposition-assertion approach to semantics is to single out one of the many speech acts that can be performed by use of a syntactic sentence type and then build its appropriateness conditions into the semantic analysis of lexical items. The unfortunate consequences of this decision can best be seen by looking at the predictions made when the item in question figures in a less typical speech act. Take guessing, for example. Guessing is the opposite of asserting, to the extent that a guess is underevidenced, often nonevidenced. If I guess that the next person to come into the room will be a bachelor, I am obviously not taking for granted, but guessing also, that the next person to come into the room will be adult and male. If I guess that Martians criticize each other for looping the loop, then I am not taking for granted, but guessing too, that Martians occasionally loop the loop. Thus the construal of all declaratives as evidenced declaratives, and the incorporation of this fact about a particular speech act into the lexical entries for words, will often lead to radically wrong predictions about the appropriateness of other speech acts.

The fact that what may be an inappropriate assertion may be a perfectly appropriate denial, fantasy, or guess seems to me to argue decisively against pragmatic presuppositional theories as conceived here. Most of the counterexamples I have used have been the sort of sentences that would typically not be used to make evidenced assertions; they are necessary truths, or valid arguments, or are evidenced in one of their clauses but not in another. I do not think that one should conclude from this that these sentences should be given an entirely different semantic treatment from that given to evidenced assertions. One would not want to say that they differed in meaning from sentences with more standard uses. But if the semantic analyses of lexical items are to enable them to enter appropriately into the full range of sentences and

possible speech acts, then it does not seem possible to draw a single distinction in these analyses between appropriateness conditions on the one hand and speech-act content on the other. In other words, it seems to me fairly important not to assume, as Fillmore does, a "distinction between the presuppositional aspect of the semantic structure of a predicate on the one hand and the 'meaning' proper of the predicate on the other hand."

A Remark about Presupposition and Entailment

In this article I have restricted myself rather narrowly in two ways: first, I have talked only about the presupposition-assertion approach to semantics, and second, I have talked only about the presuppositional approach to lexical items. However, many of the examples and arguments I have used have a bearing on other presuppositional accounts. For example, the entailment-presupposition distinction is often set up in a way exactly parallel to the assertion-presupposition distinction, and what are treated as presuppositions in one framework will be treated as presuppositions in the other.3 In this case, my arguments carry over directly, and if they are good against the presupposition-assertion approach they will be equally good against the presupposition-entailment approach. Moreover, it is not only presuppositions carried by lexical items that create trouble for the presuppositional approach. Existential presuppositions carried by definite NPs in certain positions behave in exactly the same way as do those carried by lexical items; hence, for these cases too it is easy to construct counterexamples to the most immediately appealing presuppositional treatments. The problem I have tried to raise is thus perfectly general. I have also tried to suggest that the way around it is not to search for more and more refined filtering hypotheses; in doing this one loses sight of what seems to me to be a very simple underlying fact: that with a single category of entailments/assertions, and a single mechanism for predicting their possible distributions, one will in fact make exactly the correct claims about the possibilities of "presupposition cancellation". The correct claim seems to be this: "presuppositions" and entailments/assertions may drop in all places except in positive nonmodal main clauses. There is no distributional basis for setting up a separate category of presuppositions. There are cer- tainly many questions that remain to be answered about preferred interpretations, although I feel that these are not semantic questions. But as long as there is no clear class of cases in which a "presupposition" can be shown to require separate distributional treatment from an entailment/assertion, there seems to me to be no justification for incorporating presuppositions into semantic theory.

3For detailed discussion of the way in which the two approaches interconnect, see Wilson (1973, Chapter IV).

I 14 DEIRDRE WILSON

References Fillmore, C. (I969) "Types of Lexical Information," in F. Kiefer, ed. (I969). Fillmore, C. (I971) "Verbs of Judging: An Exercise in Semantic Description," in C. Fillmore

and D. T. Langendoen, eds. (I97I). Fillmore, C. and D. T. Langendoen, eds. (I97I) Studies in Linguistic Semantics, Holt, Rinehart

and Winston, New York. Geach, P. (1965) "Assertion," Philosophical Review LXXIV, no. 4, 449-465. Karttunen, L. (I973) "Presuppositions of Compound Sentences," Linguistic Inquiry Iv,

I69-I93. Kiefer, F., ed. (I969) Studies in Syntax and Semantics, Humanities Press, New York. Kiparsky, C. and P. Kiparsky (I97I) "Fact," in D. Steinberg and L.Jakobovits, eds. (I97I).

Langendoen, D. T. (I971) "Presupposition and Assertion in the Semantic Analysis of Nouns and Verbs in English," in D. Steinberg and L. Jakobovits, eds. (I97I).

Langendoen, D. T. and T. Savin (I97I) "The Projection Problem for Presuppositions," in C. Fillmore and D. T. Langendoen, eds. (I97I).

Liberman, M. (I973) "Alternatives," in C. Corum, T. Smith-Stark, and A. Weiser, eds., Papers from the Ninth Regional Alleeting of the Chicago Linguistic Society, University of Chicago, Chicago, Illinois.

Steinberg, D. and L. Jakobovits, eds. (97I Semantics: An Interdisciplinary Reader in Philos- ophy, Linguistics, and Psychologv, Cambridge University Press, Caitbridge.

Wilson, D. (1973) Presuppositions and Non-Truth-Conditional Semantics, Doctoral dissertation, MIT, Cambridge, Massachusetts (in press).

Zwicky, A. (97I) "On Reported Speech," in C. Fillmore and D. T. Langendoen, eds. (1I970I)

Department of General Linguistics University College London Gower Street London WCi, England

Presupposition, Assertion, and Lexical Items

Documents