When counterfactuality rescues positive polarity items: A note on Szabolcsi (2004) Bernhard Schwarz McGill University Baker (1970a) observed that positive polarity items antilicensed by negation can be rescued by the larger linguistic context. Szabolcsi (2004) proposes that the relevant contexts are just those licensing weak negative polarity items. This proposal ignores the contributions in Baker (1970b) and Karttunen (1971b), according to which antilicensed positive polarity items are in certain cases rescued by counterfactual implications. Reviewing and extending the relevant observations, this note argues that in clauses with fake past tense morphology in the sense of Iatridou (2000), rescuing of positive polarity items indeed correlates with counterfactuality and is independent of the licensing of weak negative polarity items. 1. Introduction So-called positive polarity items (PPIs) usually cannot be interpreted in the immediate scope of a clause mate negation. As (1) illustrates for already and sometimes, PPIs whose logical scope is fixed by word order typically cannot be in the immediate scope of negation at the surface. Noun phrases headed by the PPI some can more freely appear in the immediate surface scope of negation but they must usually be interpreted with inverse scope. Thus (2) means that there were some typos they did not find, not that they found no typos. (1) a. * They haven’t already talked to me. b. * They haven’t sometimes talked to me. (2) They didn’t find some typos. Baker (1970a) observed that in certain linguistic contexts, PPIs can be rescued from negation, that is, can exceptionally be interpreted in the immediate scope of negation after all. These contexts include the restrictor of no and the scope of surprised. For example, each of the sentences in (3) to (6) below is acceptable in a surface scope interpretation. (3) a. There is no one here who hasn’t already talked to me. b. There is no one here who doesn’t sometimes talk to me. c. (4) There is no one here who didn’t find some typos. (5) a. I am surprised that they haven’t already talked to me. b. I am surprised that they don’t sometimes talk to me. c. (6) We are surprised that they didn’t find some typos.

-1-

The question is how the class of contexts where PPIs are rescued from their antilicensers can be characterized. Szabolcsi (2004) suggests a correlation between PPI rescuing and licensing of weak negative polarity items (NPIs) like ever, proposing the generalization in (7). (7) An antilicensed PPI is rescued if and only if it appears in a context where weak NPIs are licensed. This generalization is consistent with the data in (3) to (6), as weak NPIs are known to be licensed in the restrictor of no and in the scope of surprised. Szabolcsi (pp. 417-419) moreover demonstrates that PPIs can be rescued from potential antilicensers in a variety of other familiar contexts for weak NPIs, including the scope of the adversative predicate regret, the scope of various downward entailing noun phrases, the restrictor of every, antecedents of conditionals, and the scope of sentential negation in a higher clause. In addition, Szabolcsi (p. 426) reports that rescuing is subject to the kind of intervention constraint that Linebarger (1987) detected in negative polarity licensing. For example, Szabolcsi’s (8)a illustrates Linebarger’s observation that because-clauses cannot scopally intervene between a negative polarity item and a potential licenser: the parenthesized continuation forces an interpretation where the first because-clause is interpreted in the immediate scope of the higher negation, and licensing of the NPI anything is impossible in this reading. Analogously, Szabolcsi reports that the parenthesized continuation in (8)b prevents the matrix negation from rescuing something, which suggests that because-clauses also cannot scopally intervene between an antilicensing negation and a potential rescuer. (8) a. b.

I didn’t expect that John would say anything because this was a public event (*but because I know how he is). I didn’t expect that John wouldn’t say something because this was a public event (*but because I know how he is).

The generalization in (7), then, appears to have ample empirical support. It leads Szabolcsi to the conclusion that negation composes with a PPI in its immediate logical scope into a derived weak NPI and hence that rescuing is a case of NPI licensing. However, (7) does not take into account discussion in Baker (1970b) and Karttunen (1971b), which suggests that in a certain class of cases antilicensing correlates with counterfactuality. Below I will review and extend the data pattern described by Baker and Karttunen, concluding that in clauses hosting fake past tense morphology in the sense of Iatridou (2000), rescuing is independent of licensing of weak NPIs and instead correlates with counterfactuality. 2. Rescuing by counterfactuality Baker (1970b) does not explicitly state the generalization in (7) but nevertheless discusses an example showing that it undergenerates. Baker observes that PPIs can be rescued in the clausal complement of wish even though NPIs are not licensed there. Despite the ungrammaticality of (9)b, for example, sentence (9)a is acceptable.

-2-

(9) a. I wish he didn’t already have a cat. b. * I wish they had ever helped us. Baker proposes that rescuing under wish is made possible by the counterfactuality of wishsentences, that is, by the implication that the complement of wish is false. Thus according to Baker, (9)a is acceptable by virtue of having the counterfactual implication in (10). (10)

He already has a cat.

In combination with the data presented by Szabolcsi, Baker’s observation suggests that there are two kinds of rescuing, rescuing by NPI licensing and rescuing by counterfactual implication, hence that the generalization in (7) is to be amended as shown in (11). (11)

An antilicensed PPI is rescued if and only if (i) or (ii): i. The PPI appears in a context where weak NPIs are licensed. ii. The containing sentence implies that the clause hosting the PPI is false.

Generalization (11) does not specify whether the relevant implications are truth conditional entailments, presuppositions, or conversational implicatures. Baker (1970b) is actually more specific and considers them semantic presuppositions. The reasons for the vaguer formulation in (11) may become apparent shortly and are discussed explicitly in section 6 below.1 Additional support for (11) comes from data not presented in Baker (1970b). To begin, (12)a demonstrates that the complement of suppose in imperative sentences supports PPI rescuing, and (12)b makes the same point for clauses introduced by as though. As shown in (13), these contexts do not license NPIs and therefore present additional counterexamples to (7). Moreover, since the sentences in (12) also imply (10), they are covered by (11). (12) a. b.

Suppose he didn’t already have a cat. You talk as though he didn’t already have a cat.

(13) a. * Suppose they had ever talked to me. b. * You talk as though you had ever been here.

1

Generalization (11) also remains vague on the notion of falsity, which cannot always be explicated as falsity in the actual world. This is because Baker’s characterization of the counterfactuality of wish-sentences is not quite accurate. As Iatridou (2000) observes, a wishsentence need not imply the falsity of its complement, but merely that the referent of the subject believes it to be false. Thus (i)a is consistent with me not knowing as many jokes as John, as long as he wrongly believes that I actually do. The same is true for (i)b, indicating that rescuing under wish does not require the complement to be false in the actual world, either. (i) a. b.

John wishes I didn’t know as many jokes yet as he does. John wishes I didn’t already know as many jokes as he does. -3-

Consequents of would-conditionals present yet another counterexample to (7).2 As (14) illustrates, this context also supports PPI rescuing despite not licensing NPIs. Generalization (11) again applies correctly, given that (14)a implies (15). (14) a. If there were no oil in the tank, the furnace wouldn’t already be running. b. * If there were oil in the tank, the furnace would ever start. (15)

The furnace is already running.

Rescuing in consequents of would-conditionals actually makes a particularly clear case for rescuing by counterfactual implication. This is because would-conditionals do not invariantly imply that their consequents are false. Would-conditionals whose consequent contains the word still illustrate this. As Karttunen (1971a) notes, the consequents of such conditionals are often understood to be true, rather than false. For example, sentence (16)a implies that we are not in Stuttgart yet. According to (11), therefore, PPIs should not be rescuable in the consequents of such would-conditionals with still. As sentence (16)b illustrates, this prediction is correct. (16) a. If we had left a minute earlier, we still wouldn’t be in Stuttgart yet. b. # If we had left a minute earlier, we still wouldn’t be in Stuttgart already. Karttunen (1971a) also notes that an anaphoric connection between the antecedent and the consequent can prevent a would-conditional from implying the falsity of its consequent. The conditional in (17)a illustrates this. Since the sentence is taken to imply that I do not actually have a car, the consequent cannot be classified as either true or false in the actual world. According to (11), therefore, PPIs should not be rescuable in such cases. Example (17)b illustrates that this prediction is again correct. (17) a. If I had a car, I would never wash it. b. # If I had a car, I wouldn’t sometimes wash it. In summary, there is a class of contexts for PPI rescuing that do not license weak NPIs. These contexts show that Szabolcsi’s generalization (7) undergenerates. In accordance with the amended generalization in (11), rescuing in each of these cases comes with a counterfactual implication, that is, the implication that the clause hosting the PPI is false. Particularly compelling support for (11) comes from the observation that would-conditionals support rescuing in their consequents only if their interpretation allows for the requisite counterfactual implication.

2

The term “would-conditional” is due to Arregui (2005) and of course refers to conditionals headed by the modal verb would. Would-conditionals are more commonly known as “counterfactual conditionals” or “subjunctive conditionals”. However, Arregui (2005) and Iatridou (2000), among others, note that both terms are misleading: English would-conditionals do not actually feature subjunctive morphology nor, as we will see, are they always interpreted counterfactually. -4-

3. Rescuing by counterfactuality and fake past tense While motivated by the examples in section 2, clause (ii) of generalization (11) is too permissive as it stands. This is shown by the failure of the verb pretend to rescue PPIs in its complement. As Lakoff (1972) observes, complements of pretend are interpreted counterfactually. Sentence (18)a, for example, implies that they have a cat already. However, as illustrated by (18)b, pretend does not rescue PPIs. (18) a. They pretend that they don’t have a cat yet. b. * They pretend that they don’t have a cat already. Faced with this counterexample, it is time to draw attention to a property that all the contexts for rescuing by counterfactuality described above have in common. Complements of wish, complements of suppose in imperatives, as though-clauses, and consequents of wouldconditionals feature fake past tense marking in the sense of Iatridou (2000), that is, past tense marking that does not actually express anteriority. Thus despite the past tense form of did, (9)a conveys that I would prefer for him not to have a cat now, (12)a instructs you to imagine that he does not have a cat now, and (12)b conveys that what you said implies that he does not have a cat now. Also, (14)a implies that a lack of oil in the tank would prevent the furnace form running now. Assuming with Iatridou (2000) and many others that would is the past tense counterpart of present tense will, this suggests that consequents of would-conditionals feature fake past tense as well. This common feature of the relevant examples would be unremarkable if, as Iatridou (2000) suggests, fake past tense were a necessary ingredient of counterfactuality in English. However, the case of pretend suggests that this is not in fact so. Examples (18)a shows that pretend can embed present tense complements, and past tense marking in the complement of pretend expresses anteriority, as shown by the incoherence of They pretend that now you were sick. The observations presented here suggest that rescuing by counterfactual implication only applies in cases where the relevant PPI appears in a clause featuring fake past tense. This invites the question whether fake past tense marking also has a role to play in rescuing by NPI licensing. The observations presented in the next section show that this is indeed the case. 4. Rescuing by NPI licensing and fake past tense In illustrating PPI rescuing by counterfactual implication, Baker (1970b) focuses on a type of context not considered in section 2, namely antecedents of would-conditionals. In Baker’s view, for example, the PPI in (19) is acceptable in the immediate scope of negation by virtue of the counterfactual implication (20). This analysis is consistent with the proposal in section 3, for antecedents of would-conditionals feature fake past tense morphology as well: the counterfactual condition that (19)a introduces is that I do not have a bicycle now. 3

3

Ippolito (2003) and Arregui (2005) actually argue that fake past tense in conditional antecedents does express anteriority after all. Still, these authors agree that past tense in antecedents of would-conditionals fails to constrain the temporal location of the eventuality described by the antecedent clause. -5-

(19) (20)

If I didn’t already have a bicycle, I would go and buy one today. I already have a bicycle.

Example (19) by itself is less informative than the rescuing cases in section 2 in that it is consistent with Szabolcsi’s generalization (7). After all, conditional antecedents in general are contexts for NPIs and as (21) illustrates, antecedents of would-conditionals are no exception. (21)

If they had added any oil, the furnace would have started.

However, in antecedents of would-conditionals, too, rescuing depends on counterfactuality. It is possible to show this because, just like consequents of would-conditionals, antecedents of would-conditionals are not in all cases implied to be false. I will examine three different types of would-conditionals that do not have counterfactual antecedents, showing that in each case the antecedent does not support rescuing. The first type of case, discussed in Anderson (1951) and much later work, is illustrated in (22)a. This text is an argument to the effect that the tank was empty and it is accordingly inconsistent with the implication that there was oil in the tank. In this case, then, the wouldconditional cannot be read as implying that its antecedent is false. That the antecedent does not support rescuing in this case is illustrated by the unacceptability of (22)b.4 (22) a.

If there had been no oil in the tank, the furnace would have made just exactly the kind of noise that it in fact did. So it’s likely the tank was indeed empty. b. # If there hadn’t been some oil in the tank, the furnace would have made just exactly the kind of noise that it in fact did. So it’s likely the tank was indeed empty.

A second type of would-conditional that does not imply the falsity of its antecedent is illustrated in (23)a, which is based on an example in von Fintel (1999). The first sentence of this text implies that there was no oil in the tank, hence the lower would-conditional cannot consistently be read as implying that its antecedent is false. The unacceptability of (23)b shows that the lower antecedent clause in this case does not allow for PPI rescuing. (23) a.

The fact that there was no oil in the tank could have been a problem. If they had come to fix the furnace, then if there had been no oil in the tank, they couldn’t have done anything. b. # The fact that there was no oil in the tank could have been a problem. If they had come to fix the furnace, then if there hadn’t been some oil in the tank, they couldn’t have done anything.

A third type of would-conditional that does not imply that its antecedent is false, discussed in Karttunen (1971b), is exemplified in (24)a below. The nested would-conditional is understood to be counterfactual with respect to the higher antecedent, implying that I do not have a sister. The anaphoric link between the two antecedents prevents the sentence from carrying any 4

The judgment on (22)b also indicates that, for reasons that need not interest us here, some oil cannot be interpreted as taking inverse scope over negation. The same comment applies to example (23)b below. -6-

implication as to the truth of the lower antecedent. Given the implication that I do not have a sister in the first place, the sentence cannot be read as implying either that she visits me or that she does not. Hence the lower antecedent cannot be interpreted counterfactually.5 As Karttunen (1971b) observes, PPIs contained in such an antecedent again cannot be rescued. The unacceptability of example (24)b illustrates this. (24) a. If I had a sister, then if she never came to visit me, I would probably get depressed. b. # If I had a sister, then if she didn’t sometimes come to visit me, I would probably get depressed. The conclusion is that rescuing in antecedents of would-conditions is dependent on counterfactuality. Of course, this observation would be consistent with Szabolcsi’s original generalization (7) if it turned out that licensing of weak NPIs is subject to the same condition. This, however, seems not to be the case. In the sentences in (25), weak NPIs replace the PPI and its antilicenser in the (b) examples in (22) to (24). In each case, the weak NPI is judged to be acceptable despite the absence of a counterfactual implication.6 (25) a. b. c.

If there had been any dirt in the tank, the furnace would have made just exactly the kind of noise that it in fact did. So it’s likely there was indeed some dirt in there. The fact that there was dirt in the tank could have been a problem. If they had come to fix the furnace, then if there had been any dirt in the tank, they couldn’t have done anything. If I had a sister, then if she ever said anything about my hairstyle, I would get upset.

Thus not all contexts for weak NPIs support PPI rescuing. Generalization (7) not only undergenerates, but also overgenerates. The more specific conclusion is that, just like clause (ii) of generalization (11), clause (i) must be restricted to PPIs occurring in clauses of a certain type: while rescuing by counterfactual implication appears restricted to PPIs in clauses featuring fake past tense morphology, rescuing by NPI licensing only seems to apply to PPIs in clauses that do not contain such morphology. A suitably revised version of generalization in (11) is provided in (26).

5

Moreover, as Karttunen (1971b) notes for an analogous example, the lower antecedent in (24)a is also not counterfactual relative to the possible state of affairs the higher antecedent invites one to consider. That is, the sentence does not imply that in the relevant counterfactual situations where I have a sister, my sister will sometimes visit me. 6 A reviewer reports that some speakers, while finding examples like (25)a initially acceptable, perceive incoherence upon closer reflection, finding that the conditional antecedent needs to be interpreted counterfactually after all. I speculate that this relates to the observation, discussed in section 5 below, that so-called minimizers in antecedent of would-conditionals indeed force counterfactuality in this type of conditional. According to Krifka (1995), stressed any has much the same semantics as minimizers, suggesting that variations in judgments may be due to varying intonation assigned to the relevant sentences. -7-

(26) i. ii.

A PPI that is antilicensed in a clause whose finite verb is not a fake past tense is rescued if and only if it appears in a context where weak NPIs are licensed. An PPI that is antilicensed in a clause whose finite verb is a fake past tense is rescued if and only if the containing sentence implies that this clause is false.

5. A new type of polarity sensitivity As mentioned in section 1, Szabolcsi (2004) concludes from the proposed generalization (7) that negation composes with a PPI that it antilicenses into a derived weak NPI. While we have seen that the distribution of PPI rescuing is actually more complex, it is still possible to maintain, of course, that negation always composes with the antilicensed PPI into some type of polarity item. According to generalization (26), polarity items of this type would behave like weak NPIs in clauses that do not feature fake past tense, but would be licensed by counterfactual implications in clauses that do. An observation reported in Giannakidou (1999, p. 408) may in fact invite the speculation that polarity items composed from negation and a PPI are distributed like a well-known type of NPI, namely like minimizers such as say a word, lift a finger, or give a damn. Giannakidou presents the example in (27), which is analogous to (25)a above except that the conditional antecedent features a minimizer instead of a weak NPI. (27) # If he gave a damn about his job, he’d be working exactly as hard as he is now. Giannakidou concludes from the unacceptability of (27) that licensing of minimizers in the antecedents of would-conditionals is dependent on the presence of a counterfactual implication. If this is correct, (27) and (22)b are unacceptable for the same reason, which raises the prospect that the contexts for PPI rescuing can be described by the minimal revision of generalization (7) stated in (28). (28)

A PPI can be rescued if and only if it appears in a context where minimizers are licensed.

Generalization (28) may improve on (7) in predicting that PPI rescuing in the antecedent of a would-conditional requires a counterfactual interpretation of this antecedent. Thus (28) may avoid the problem of overgeneration for generalization (7) identified in section 4. Unfortunately, however, (28) undergenerates as much as (7) does. We saw in section 2 that counterfactual implications can rescue antilicensed PPIs in certain contexts where weak NPIs are not licensed, contexts such as the complement of wish and counterfactual consequents of would-conditionals. As illustrated in (29), minimizers are no more acceptable in these contexts than weak NPIs are. Thus there remains a significant difference in distribution between PPI rescuing and minimizers. (29) a. * I wish he had lifted a finger to help us. b. * If he liked his job, he would give a damn about this project. Moreover, it appears that even in antecedents of would-conditionals, the licensing of minimizers does not in fact correlate with counterfactuality in the same way PPI rescuing does. This is suggested by the examples in (30), which are analogous to those in (25)b and (25)c

-8-

except that the relevant antecedent clause hosts a minimizer instead of weak NPI. The fact that these examples are acceptable indicates that counterfactuality is not in general required for the licensing of minimizers in the antecedents of would-conditionals. (30) a. b.

Given that John spoke at the meeting, I’m glad that Mary wasn’t there. If she had been there, then if John had said a word, they would have got into an argument. If you had a bike, then if you gave a damn about its condition, you would add fenders.

If PPIs and their antilicensers indeed compose into polarity items, therefore, the distribution of these polarity items cannot be derived by analyzing them as minimizers. These polarity items would instead seem to be of a novel type whose distribution does not track the distribution of any type of polarity item discussed in previous literature. 6. New questions for the theory of rescuing Naturally, the proper theory of rescuing will need to explain how the complex generalization in (26) can be derived from more basic principles of grammar. While I will not attempt such a derivation, I will briefly raise four questions concerning (26) that a complete analysis of rescuing would have to address. The first question is whether a general theory of rescuing should actually make reference to fake past tense marking directly, as the descriptive generalization in (26) does for English, or whether it should instead refer to a semantic property that correlates with fake past tense marking.7 This question could be addressed by looking for PPI rescuing in languages where counterfactuality does not correlate with past tense morphology. If such a language featured rescuing by counterfactuality, then the theory of rescuing should presumably make reference to the semantic contribution of fake past tense rather than fake past tense morphology itself. Future work will need to examine the relevant data.8 The second question is how the semantic contribution of fake past tense could be characterized. The discussion in Baker (1970b) suggests that fake past tense marking triggers the semantic presupposition that the containing clause is false. However, while semantic presuppositions are not expected to be cancelable, we saw in sections 2 and 4 that neither consequents nor antecedents of would-conditionals are always counterfactual. This suggests that counterfactual implications in would-conditionals are conversational implicatures rather than semantic presuppositions conventionally associated with a particular linguistic form. Therefore, while fake past tense may contribute to the emergence of counterfactuality in conditionals, it cannot be held solely responsible. To be more concrete on this point, I will sketch two existing proposals to this effect, namely the analysis of counterfactual antecedents in Stalnaker (1975) and the analysis of counterfactual consequents in Karttunen (1971a). In an informal rendering of Stalnaker’s (1975) analysis, fake past tense in the antecedent of a would-conditional suspends the default assumption that the possible situation the antecedent invites the listener to consider is consistent with the speaker’s beliefs. The most obvious possible 7

I thank a reviewer for raising this question. Nevins (2002) reports that counterfactuality is indeed not universally associated with fake past tense, listing Chinese, Tagalog, Slovenian, Hebrew, and Turkish as counterexamples. Naturally, though, Nevins does not discuss whether these languages allow for rescuing by counterfactuality.

8

-9-

reason for the speaker to suspend this assumption would be that the antecedent is false. This explains why would-conditionals in isolation are typically interpreted counterfactually with respect to their antecedents. However, the speaker might have other reasons to suspend the relevant default assumption, and non-counterfactual interpretations of antecedents are predicted to be possible in contexts that provide such alternative reasons.9 Turning to consequents, Karttunen (1971a) identifies two ingredients involved in their counterfactual interpretation. The first ingredient is the counterfactual interpretation of the antecedent. The second is the phenomenon of conditional strengthening described in Geis and Zwicky (1971), that is, the general tendency for conditional antecedents to be interpreted as specifying necessary conditions for the truth of their consequents, rather than merely sufficient conditions. Thus in Karttunen’s analysis, the sentence If John had been here, Mary would have been happy implies that Mary was unhappy because it implies that John was not here and also that John’s presence was a necessary condition for Mary’s happiness, from which it follows that Mary is not in fact happy. This analysis allows for would-conditionals with non-counterfactual consequents because would-conditionals do not always have counterfactual antecedents and also do not always undergo conditional strengthening.10 Note that according to these analyses, fake past tense makes different contributions to counterfactuality in conditional antecedents and consequents; in fact, fake past tense in consequents does not contribute to their counterfactuality at all. Also, the role of fake past tense in other constructions considered in section 2 remains to be investigated. Hence it is presently unclear whether a general semantic contribution of fake past tense can be identified. The third question concerns the nature of the counterfactual implications attested in the rescuing cases described by clause (ii) of generalization (26). Two possible views can be distinguished. First, the relevant implications may have the same sources as in corresponding examples without rescuing, that is, they may be conversational implicatures associated with the use of fake past tense morphology. The reason the relevant implications in rescuing cases cannot be cancelled would then have to be that PPI rescuing depends on their presence. In this view, then, conversational implicatures can rescue PPIs in much the way they can license NPIs in the analysis of Linebarger (1987). Second, it is possible to assume with Baker (1970)b that the relevant implications are semantic presuppositions. Of course, for reasons that will have become clear, the presuppositions in question could not be triggered by fake past tense marking. It is conceivable, however, that they are triggered by the very negation from which the PPI is rescued. In this view, the main task would be to explain why presuppositional negation can only occur in clauses hosting fake past tense. Whether such an explanation can be found and how the two views outlined here compare is a topic for future work. The fourth question is whether the two clauses of generalization (26) can or should be derived from a common underlying principle. This is perhaps the hardest question and it cannot be fruitfully addressed before the three previous questions have been given satisfying answers. I will therefore refrain from speculating on the issue. However, this is the time to note that the

9

All three of the non-counterfactual cases presented in section 4 can be analyzed in this way. The reader is referred to von Fintel (1999) for relevant discussion. 10 Both of the non-counterfactual cases discussed in section 2 arguably fail to undergo conditional strengthening. - 10 -

generalization in (26) actually cannot be the final description of the class of context where rescuing is found. This is because of examples like those in (31).11 (31) a. b.

We need to make sure that they haven’t already left. The law requires that one not be already in charge of this task when applying for the job.

(32) a. * We need to make sure that they ever arrived. b. * The law requires that we ever file a report. As shown in (32), the PPIs in these cases do not appear in NPI contexts. Moreover, the clauses containing them do not feature past tense morphology and are not interpreted counterfactually. These cases are accordingly counterexamples to generalization (26). While this much is clear, it is less clear how the generalization is to be amended. The fact that the PPIs in (31) occur under verbs expressing necessity seems relevant, but a detailed investigation of such cases must wait for another occasion. 7. Conclusion Szabolcsi’s (2004) analysis of PPI rescuing as the licensing of weak NPIs ignores cases of rescuing by counterfactual implications of the sort noted in Baker (1970b) and Karttunen (1971b). I have proposed that for PPIs antilicensed in clauses that feature fake past tense in the sense of Iatridou (2000), rescuing by counterfactuality replaces rescuing by NPI licensing. The proposed generalization raises many new questions that a complete theory of PPI rescuing will need to answer, in particular questions concerning the role of fake past tense marking and the nature of counterfactual implications. References Anderson, Alan Ross: 1951, ‘A note on subjunctive and counterfactual conditionals’, Analysis 11, 35-38. Arregui, Ana: 2005, On the accessibility of possible worlds: The role of tense and aspect, PhD dissertation, University of Massachusetts at Amherst. Baker, C. Lee: 1970a, ‘Double Negatives’, Linguistic Inquiry 1, 169-186. Baker, C. Lee: 1970b, ‘Problems of Polarity in Counterfactuals’, in J. Sadock and A. Vanek (eds.) Studies presented to Robert B. Lees by his Students, Linguistics Research Inc., Edmonton. Von Fintel, Kai: 1999, ‘The presupposition of subjunctive conditionals’, in U. Sauerland and O. Percus (eds.), The Interpretive Tract, MIT Working Papers in Linguistics 25, MITWPL, Cambridge, pp. 29-44. Geis, Michael L. and Arnold M. Zwicky: 1971, ‘On invited inferences’, Lingustic Inquiry 2, 561-566. Giannakidou, Anastasia: 1999, ‘Affective dependencies’, Linguistics and Philosophy 22, 367421. 11

I owe example (31)b to a reviewer. - 11 -

Iatridou, Sabine: 2000, ‘The grammatical ingredients of counterfactuality’, Linguistic Inquiry 31, 231-270. Ippolito, Michela: 2003, ‘Presuppositions and implicatures in counterfactuals’, Natural Language Semantics 11, 145-186. Karttunen, Lauri: 1971a, ‘Counterfactual conditionals’, Lingustic Inquiry 2, 566-569. Karttunen, Lauri: 1971b, ‘Subjunctive conditionals and polarity reversal’, Papers in Linguistics 4, 279-298. Krifka, Manfred: 1995, ‘The semantics and pragmatics of polarity items’, Linguistic Analysis 25, 209-257. Lakoff, George: 1972, ‘Linguistics and natural logic’, in D. Davidson and G. Harman (eds.), Semantics of Natural Language, Reidel, Dordrecht, pp. 545-665. Linebarger, Marcia: 1987, ‘Negative polarity and grammatical representation’, Linguistics and Philosophy 10, 387-437. Nevins, Andrew Ira: 2002, ‘Counterfactuality without past tense’, in Masako Hirotani (ed.), Proceedings of NELS 32, GLSA, University of Massachusetts, Amherst, pp. 441 – 450. Stalnaker, Robert: 1975, ‘Indicative conditionals’, Philosophia 5, 269-286. Szabolcsi, Anna: 2004, “Negative Polarity – Positive Polarity”, Natural Language and Linguistic Theory 22, 409-452.

- 12 -

counterfactual rescuing

If I had a sister, then if she ever said anything about my hairstyle, I would get upset. Thus not all contexts for weak NPIs support PPI rescuing. Generalization (7) ...

121KB Sizes 2 Downloads 215 Views

Recommend Documents

On Counterfactual Computation
cast the definition of counterfactual protocol in the quantum program- ... fact that the quantum computer implementing that computation might have run.

rescuing flaming projects!
Mar 1, 2004 - architecture are used to create a detailed design and a project plan, using the project's .... ABC Company noted a weakness in other web-based ... schedules, distributed across the Houston and India development centers.

Counterfactual Thinking and Posttraumatic Stress ... - Semantic Scholar
Traumatic Stress Service, St. George's Hospital, London and University of Surrey. Preoccupation with ... dictive of recovery in individuals with either depression or PTSD, whereas .... These tasks were used to generate data on the availability of ...

Counterfactual Thinking and Ascriptions of Cause and ...
belief should be that C was a necessary cause of E. To use a well- .... come information. Because only upward counterfactuals (i.e., thoughts that improve on reality; Markman et al., 1993) can mentally restore negative outcomes to positive or neutral

A Comment on Diagnostic Tools for Counterfactual ...
Feb 12, 2008 - London School of Economics, Department of Economics,. Houghton Street ... The further from the data we take a counterfactual, the ..... for n 5 100, high correlation (rho 5 0.8) does not have a big effect on the results of the.