ESTUDIOS SOBRE HISPANIA

Following this intuitive presentation, let me now show in detail how Krifka (1998)’s account wrongly rules out sentence (21) even though it is more sophisticated than a naïve implementation of the subregion-based approach. I will argue that the presupposition predicted by Krifka’s entry in (18) is violated by the predicate denoted by (21). As a result, Krifka’s account wrongly predicts that (21) should be unacceptable due to presupposition failure. His prediction is that (21) presupposes that any eventein its denotation satisfies the following condition:

(23) ∃e0_[e0 _<

τ e] ∧ ∀e00[e00≤τ e→e00 ∈JJohn push carts to the storeK]

As a reminder, I call the first conjunct of this presupposition the existence clause and the second conjunct the divisiveness clause. To argue that this presupposition is violated, I will proceed in four steps. First, I will argue that sentence (21) applies to a proper sum event, that is, an event that has subevents. This means that the existence clause of the presupposition is satisfied, which means that the problem must lie in the divisiveness clause and ultimately in Krifka’s reliance on divisive reference. Second, I will argue that some of these shorter subevents are not in the denotation of the predicate John push carts to the store. Third, I will argue that some of these subevents stand in the relation≤τ and therefore≤to the sum event (they are parts of the sum event and their runtime is shorter than that of the sum event). Finally, I will argue that these subevents cannot be excluded on pragmatic grounds from the range of the universal quantifier in (23). These four steps together have the consequence that the presupposition in (23) does not hold.

I start by arguing that (21) applies to a sum event that has subevents. I have assumed that a verbal predicate that has been modified byto the storeonly applies to events whose spatial extent ends at the store (see Section 2.9). This assumption is intuitively plausible and corresponds to the account of spatial prepositional phrases in Krifka (1998). Suppose that (21) is uttered in a scenario in which John repeatedly pushed some carts from location A to the store. (In a pragmatically plausible scenario, he pushes a different set of carts on each of his back-and-forth trips, but nothing depends on this.) Call B the point halfway between location A and the store. Standard assumptions of mereological event semantics commit us to the (intuitively plausible) claim that (21) entails the existence of a sum evente which can be divided at least into two complex subevents: an evente1in which

John pushed carts (iteratively) from A to B and an evente2 in which he pushed

carts (iteratively) from B to the store. I will leavee2 aside and concentrate on the

evente1. This event does not itself qualify as an event of John pushing carts to the

store, because its spatial extent does not include the store. To see thateentails the existence ofe1, note that (21) entails (24), and moreover (24) entails (25):

(24) John pushed carts to the store.

(25) John pushed carts halfway towards the store.

These sentences are of course less informative than (21) because they describe parts of the scenario evoked by it, so it is not immediately obvious that the entailment relations that I have mentioned hold. For example, one might take (25) to be false in such a scenario because it conveys that the carts in question did not reach the store. However, that the carts did not reach the store is not a part of the literal meaning of but an implicature. This is clear in contexts such as questions, where implicatures are usually not computed. For example, when we turn (25) into a question, it is possible to answer it affirmatively with (24), but the converse is not possible:

(26) Did John push carts halfway towards the store?

(27) Yes/*No – in fact, he pushed carts to the store. (28) Did John push carts to the store?

(29) No/*Yes, he pushed carts halfway towards the store.

I conclude that the entailment relations do indeed hold. Each of these sentences entails the existence of an event, and we can plausibly model the fact that (21) entails (25) by assuming that any event denoted by (21) has an event denoted by (25) as one of its parts. This means that (21) indeed denotes a sum event, as claimed above.

The second step is to argue that the predicate denoted by (24), which applies to the sum event of (21), call ite, does not apply to the sum event of (25), call ite1.

This is easy to see, given that we know that (24) entails (25) but not vice versa. If John pushed carts to the storeapplied to the event in (25), then (25) should entail (24). The intuition behind this reasoning is that an event of pushing carts halfway towards the store neither is nor entails an event of pushing carts to the store.

Third, I argue thate1 ≤τ e, that is,e1 is a part ofeand there is another part

ofewhose runtime does not overlap with the runtime ofe1. This is how Krifka

expresses that e1 has a shorter runtime than e. As mentioned, that e1 ≤ e is

intuitively plausible. Any assumption to the contrary would make it at best very difficult to explain why (21) entails (25). Moreover, observe thate1 is a proper part

ofe, that is, it is not identical toe. This is so because (25) does not entail (24). By the Unique Separation axiom (see Section 2.3.1), there is an evente2 that is a proper

part ofeand does not overlap withe1, so that the sum ofe1 ande2 ise. Intuitively,

this models the following fact: eis a sum event in which John pushed carts from A to the store.e1 is a part ofein which John pushed carts from A halfway towards

the store, that is, from A to B; ande2 is an event in which John pushed carts from

B to the store, that is, the result of subtractinge1 frome. Now, given the fact that

John cannot be in two different locations at the same time, the runtimes ofe1 and

e2 do not overlap. This entailse1 ≤τ e, which I set out to show.29

Summing up, we have the following situation. A for-adverbial can modify a

sentence, namely (21), whose sum eventehas a proper parte1 that is not in the

denotation of the sentence. This is in contradiction to divisive reference. Moreover, e1 has a shorter runtime than the sum evente, therefore not only e1 < e holds

but alsoe1 <τ e. This is in contradiction to Krifka (1998)’s divisiveness clause, by

which he implements the subregion-based approach as seen in (18). I will refer to e1 as the “offending event” since it violates the divisiveness clause. We will

see later that the strata-based approach fares better becausee1 does not lead to a

violation of stratified reference.

Finally, let us consider what additional assumptions would be necessary to rescue the subregion-based approach. Although this is not always explicitly stated, universal quantifiers in semantic representations are generally assumed to be restricted to “relevant” entities. The following classical example, taken from Kratzer (1989), illustrates the point. We have an orchard whose trees are all laden with wonderful apples. A man who wants to buy the orchard asks us whether all its trees are apple trees. We answer: “Yes, and every tree is laden with wonderful 29_{The runtimes of each of these events will be discontinuous in scenarios where John goes back}

and forth between A and the store and pushes the carts to the store one by one or at least little by little. This does not affect the main point.

apples.” It is clear from context that the universal quantifier supplied byevery tree is implicitly restricted to the trees in the orchard. Otherwise, the sentence would be false because there are many trees in the world which are not apple trees and cannot have any apples. The relevance issue is a bit of a wildcard, because no commonly accepted semantic theory provides clear criteria for deciding whether a given entity is relevant. However, a common intuition is that all irrelevant entities must be “outside the situation” in which the sentence is understood. In the orchard example, most if not all of the trees inside the orchard are relevant from an intuitive point of view since the truth of the sentence depends on any one of them having apples, while the trees outside the orchard are irrelevant. This intuition is formally implemented in situation semantics (Kratzer 1989).

In our event-based framework, the situation in which a sentence is understood can be thought of as the sum event over which a sentence existentially quantifies (Dekker 1997; Kratzer 2010). According to this view, the relevant events in the example discussed above are the subevents of the sum event e. Event e1, the

“offending event” whose location does not contain the location of the store, is indeed a subevent ofeand is therefore relevant. As shown in Figure 6.2b, there are many other such offending subevents. It would be difficult to explain why none of these subevents should be seen as relevant.

However, let us grant that an argument to the contrary can be made, and that the only relevant events turn out to be those subevents ofewhose location contains the store. In Figure 6.2b, these subevents are the ones whose right-hand boundary is also the right-hand boundary of the sum event. These subevents all have checkmarks. This indicates that they all qualify aspush carts to the store, which is the case because their location contains the store. It would then follow that all relevant subevents ofewould qualify as push carts to the store, and the presupposition of (21) would indeed be satisfied. This would appear to rescue the subregion approach.

I see three problems with this line of thinking. First, there is no reason to assume that these subevents are indeed irrelevant for the felicity and truth condi- tions of the sentence, and that so many of them should be. If they were, we should expect most trees inside our orchard to be irrelevant for the truth conditions of Every tree is laden with applesas well. Second, there is no explanation why (21) requires an iterative interpretation. Suppose John starts pushing a set of carts towards the store all at once, and stops doing anything once he has arrived there. Although the locations of most subevents of such an event do not contain the store, those events whose location does contain the store all qualify as events in which John pushes carts to the store. If these are the only relevant subevents, the sentence should be modifiable by any temporalfor-adverbial. Finally, there is no explanation why spatialfor-adverbials cannot modify (21), since an analogous reasoning shows

that they should be acceptable just like temporalfor-adverbials. I conclude that an appeal to relevance is unlikely to save the subregion-based approach.