On the consequences of event quantification in counterfactual conditionals
Introduction. Lewis began one of the most influential books on counterfactual conditionals with the
sentence If kangaroos had no tails, they would topple over (Lewis 1973). Yet, if his antecedent clause had
been eventive, the choice of auxiliaries would have been different. With eventive antecedents, a past
perfect is needed to formulate a counterfactual hypothesis (compare (1)/(2)). Lewis concluded that a
unified analysis of would-conditionals was not possible. I am more hopeful. In this paper I show that a
unified analysis is possible once we understand the role of perfect morphology as a quantifier over events,
and contrast perfect with perfective morphology, which makes reference to events. The differences in
(1)/(2) will be shown to follow from the differences between quantificational vs. referential aspect.
The proposal. Following Bennett & Partee (1978), I take default aspect in English to be perfective.
Taking seriously the definite nature of perfective aspect (a.o. Katz 1995), I argue that it introduces an
event variable that refers to an event in the actual world, and to its corresponding events in other worlds:
(3)[[perfective-ei]]w0=Ptw[P([[ei]]w0)(w) & part-of ((([[ei]]w0)(t)] (adapted from Kratzer 1998)
The perfect is given a 'resultant state' analysis (see a.o. Parsons 1990 and Kratzer 1998 for discussion):
(4) [[perfect]] = Ptwe[P(e)(w) & (e)<t] (Kratzer 1998)
Given this view of aspectual heads, the proposition corresponding to the perfective antecedent in (1) will
be true in worlds in which there is an event corresponding to the actual world event of your plants dying
(via counterparts or Lewis-events), whereas the proposition corresponding to the perfect antecedent in (2)
will simply be true in worlds in which there is an event of your plants dying. The difference is that, in the
case of quantificational aspect, no correspondance with actual world events is required. Under the
standard assumption that statives are not perfective, stative antecedents are expected to pattern with the
perfects,. The data bear this out: plain statives can give rise to counterfactual hypothesis (5).
Consequences. Given the distinctions at the level of the interpretation of aspect, a Stalnaker-Lewis style
analysis of would-conditionals is able to derive the difference between (1) and (2). In a Stalnaker-Lewis
style analysis, the modal quantifies over the most similar worlds in which the antecedent is true (adopting
the Limit Assumption and allowing for ties in similarity). If the antecedent has perfective aspect, the
modal will quantify over the most similar worlds that include events corresponding to the actual world
event of your plants dying. If the actual world does not itself include such an event, the proposition will
not be true of any world. Arguably, with the domain of quantification of the modal empty, the sentence
will be unacceptable. If the antecedent has perfect aspect, the modal will quantify over the most similar
worlds in which there is an event of your plants dying. The restriction will arguably be weaker than in the
perfect case. Conclusion: the quantificational domains of modals vary depending on aspect, so the
assertions made by the conditionals also vary depending on aspect.
Supporting evidence. Evidence further supporting the view that perfective aspect is tied to the possibility
that there be an actual world event satisfying the event predicate comes from examples like (6), with
perfective/perfect in the relative clause: if there is no actual world event fitting the description, perfective
aspect in the relative clause leads to infelicity, even if the main clause is past perfect.
Recent debate. The difference between examples like (1) and (2) has been discussed recently by Ippolito
(2003), Ogihara (2002), and to some extent, Iatridou (2000). All of these accounts have characterized the
have auxiliary we observe in (2) as an instance of a past tense interpreted outside the antecedent clause. In
the paper I show that perfect aspect is not an outside past tense in disguise: we see that past tense triggers
sequence of tense phenomena (7), but perfect doesn't (8); past perfect antecedent clauses can host since,
but since can only combine with a perfect, not with a simple past tense (eg. Kamp & Reyle 1990),
indicating that perfect is not a past interpreted outside the antecedent (see contrast in 9a, b, c). Ippolito
(2003) has argued that in past perfect would-conditionals, perfect (past) shifts the evaluation time of
presuppositions, and allows the conditional to be evaluated with respect to a conversation background that
preceded the utterance (past). However, presupposition projection facts (10) do not support this view.
Conclusion. Once we distinguish the semantic contributions made by different aspectual heads, we can
give a unified analysis for would-conditionals. The crucial difference is that between referential and
quantificational aspect. In interaction with modality, this difference has far reaching consequences.
Context: You are about to go on holidays, and ask me to look after your plants while you are away. Some
days before you are supposed to leave, however, you cancel your request. Your plants have died. I am
very sorry, but also somewhat relieved.
(1) #If your plants died while you were on holidays, I would have been very upset.
(2) If your plants had died while you were on holidays, I would have been very upset.
(5) If your plants had enough light, they would be fine.
(6) a. #Unfortunately, Professor Smith died before finding a cure for insomnia.
But if the cure the professor discovered had been very expensive, the
insurance companies would not have made it available to the general
public anyway.
b. Unfortunately, Professor Smith died before finding a cure for insomnia.
But if the cure the professor had discovered had been very expensive, the
insurance companies would not have made it available to the general
public anyway.
(7) She said that he lived in California.
(8) She has said that he lived in California.
(9) a. Mary has lived in Amsterdam since 1975.
Mary had lived in Amsterdam since 1975,
Mary will have lived in Amsterdam since 1975. (Kamp and Reyle 1995: 628)
b. *Mary lives in Amsterdam since 1975.
*Mary lived in Amsterdam since 1975.
*Mary will live in Amsterdam since 1975.
c. If you had lived in this house since 1963, you would have qualified for a rent
subsidy.
(10) a. There was a robbery last night. If the witness had managed to see the
thief's face, he would have been apprehended already.
b. #There was a robbery last night. Nobody saw it happen. If the witness had
managed to see the thief's face, he would have been apprehended already.
References:
A r r e g u i , Ana, 2004, On the accessibility of possible worlds, PhD dissertation, University of
Massachusetts, Amherst
Condoravdi, Cleo: 2001, 'Temporal interpretation of modals', in David Beaver, Stefan
Iatridou, Sabine: 2000, `The Grammatical Ingredients of Counterfactuality', Linguistic Inquiry 31,
231270.
Ippolito, Michela: 2003, 'Presuppositions and Implicatures in Counterfactuals', Natural Language
Semantics 11, 145-186.
Kamp, Hans and Uwe Reyle: 1995, From discourse to logic, Kluwer Academic Publishers, Dordrecht.
Katz, Graham: 1995, Stativity, Genericity, and Temporal Reference, unpublished PhD dissertation,
University of Rochester, New York.
Kratzer, A.: 1998, `More Structural Analogies Between Pronouns and Tenses', in Devon Strolovitch and
Aaron Lawson (eds.), Proceedings of Salt VIII. CLC Publications,
Cornell University, Ithaca, N.Y.
Lewis, D.: 1973, Counterfactuals. Harvard University Press, Cambridge, Mass.
Ogihara, Toshiyuki: 2000, `Counterfactuals, Temporal Adverbs, and Association with Focus', in
Proceedings of SALT 10, CLS Publications, Cornell University, Ithaca, N.Y. 115131.
Parsons, Terence: 1990, Events in the semantics of English, MIT Press, Cambridge.