No title

that
(1991) on entailments with }active predicates

SentailsS' iff bothSandS' have truth value and whenSis true in a context,S' is as well.8

Let us assume then that the problems about the conclusion adding new information can be circumvented by moving to Strawson Entailment as the criterion for NPI licensing. Adversatives are more complicated though. Even when we presuppose that Robin bought a Honda Civic (and that Sandy is aware of that), it does not seem obvious at all that from Sandy regrets that Robin bought a car we can safely infer that Sandy regrets that Robin bought a Honda Civic. After all , the following seems perfectly coherent: (3 I) Sandy regrets that Robin bought a car, but Sandy does not regret that Robin bought a Honda Civic. The story that goes with it may be that Sandy really didn't want Robin to have a car but among the cars Robin could have gotten, Sandy by far prefers the Honda Civic. . Now, if one wants to maintain that regret is DE, one has to explain the prima facie coherence of (3 I). At the same time, one has to ensure that the story does not in tum make predicates like gl4d DE in the same sense. The latter is a suspicion voiced by Linebarger: as soon as we apply all the tricks in the book to make regret and its cousins DE, the same fate befalls glad and its cousins. 3.1

Kadmon & Landman's move

Kadmon & Landman (1993), who develop a variant of the Fauconnier­ Ladusaw analysis, respond to the problem raised by (3 1 ) as follows: What 8

Of course, this is cleat.

Entailment

needs

to be fixed up a little but the intent of defining something like Strawson

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Katz

o

.Kai von Fintel I I 3

makes (31) coherent is a change of'perspective' between the two conjuncts. On a consistent perspective, the DE inference will be valid. Here's what they have to say:

(33) Given my high ·opinion on his moral character, I was surprised that he stole the watch. The perspective enters into the semantics of surprised and affects the truth conditions of sentences containing it. That is, for example, why we (normally) don't take B in (34) to express a contradiction. (34) A: Were you surprised that he stole the watch? B: I was and I wasn't. B is saying that relative to one perspective she was surprised and relative to another perspective she wasn't. Thus, while surprised may superficially look like a two-place relation, it can be semantically regarded as a THREE-PLACE relation between a subject (experiencer), a contextual perspective, and a proposition. Given that, to .see whether surprised is DE, we have to check whether the following pattern holds. (A. B propositions, x subject, p. q perspectives.) If A � B then Surprised(x, p. B) :::? Surprised(x, p. A) That is, we have to check whether surprise is DE 'on a constant perspective'. The apparent DE failure case of 'surprised by the purchase of a car but not by the choice of car' is not relevant, since it involves a different pattern, viz., Surprised(x, p. B) � Surprised(x, q, A). •

.

.

There is nothing prima facie objectionable about such an analysis. It is quite standard to assume that validity of inferences is to be checked against the background of a constant context. Violations of the constancy of context are branded as follacies of equivocation. There are two reasons why I would like to explore this approach into much further detail than Kadmon & Landman do: (i) I want to know what the 'fine-grained semantics of these factive attitude predicates looks like and what this contextual parameter 'perspective' does. (ii) I am skeptical that'constant context' is a notion that can be uncritically assumed: after all, we have plenty of reason to suspect that context continually changes. This is the essence of the myriad of dynamic approaches to interpretation. What are we to make of a notion of DE entailment that relies on a constant context? I suppose the best perspective on this would be to make a distinction between two kinds of discourse: evolving deliberative discourse and static statements of lo�cal

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

To be surprised that A is always relative to a certain perspective on A. a perspective that determines what it is about A that is surprising and in virtue of what it is surprising. The perspective is a contextually determined parameter in the interpreta­ tion of surprised, very much in the same way that a 'modal base' (Kratzer 1981} is a contextually determined parameter in the interpretation of modals. It can be at least partly specified by explicit linguistic text, as in (32) and (33), or be left entirely implicit. (32) I was surprised that he stole the watch, in as far as that was a daring thing to do.

I 14

NPI Licensing. Strawson Entailment, and ConteXt Dependence

.

infinitum:

(3 s} Sandy is sorry Robin bought a car. But Sandy is not sorry that Robin bought a Honda Civic. However, Sandy is sorry that Robin bought a Honda Civic EX. Then again, Sandy is not sorry that Robin bought a Honda Civic EX with :i sunroo£ 9 An explicit attempt to apply Kratzer's notions _to the semantics of rtgrd and to thereby- salvage DE can be found in Katz (1991� Instead of reporting directly what he has to say, I will unfold the analysis in a slightly different way.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

inferences. Someone who puts forward an inference from any number of premises (including tacidy assumed ones) to a target conclusion arguably guarantees implicidy that the inference does not rely on- any context changes midway through the argument. But such large issues should be put on the backburner while we tackle the nitty-gritty. What is this contextual parameter 'perspective' that Kadmon & Landman appeal to and how does it enter into the semantics of the attitudes? Kadmon & Landman are not very specific in their discussion. Perhaps it would be best to expand on their reference to Kratzer's analysis of the context-dependency of modals.9 In a Kratzer­ style semaritics, we would make attitude predicates sensitive to two contextually supplied parameters of interpretation. One parameter will be a . set of worlds, another will be an ordering among these worlds. For attitudes like want, wish, glad, regret, sorry the ordering will be one of 'preference'. For attitudes like expect, amazed, surprised the ordering will be one of 'expectation/likelihood'. Kadmon & Landman do not say whether we should think of . their 'perspective' parameter as corresponding to something like the set of worlds among which an ordering is established or as corresponding to whatever supplies the ordering itsel£ Looking at the examples that they discuss, (32) and (3 3), one would have to say that it is the ordering that is being grounded: what makes his stealing the watch more surprising than the possibility that he didn't steal the watch are the facts that stealing is a daring thing to do or that I have a high opinion of his moral character. But when one looks at the crucial Honda Civic scenario, it seems that what changes is not the ordering. Instead, there appears to be a narrowing of the set of worlds among which the ordering is specified. What makes (3 I} coherent appears to be that when one takes into consideration worlds in which Robin does not buy a car, Sandy prefers those over worlds in which Robin does buy a car But when we narrow the view to worlds thai: differ only in which car Robin buys, Sandy prefers those in which Robin buys a Honda Civic. This game can of course be continued at

Kai

. von Fintd I I S

So, perhaps we should look for a semantics for the attitudes that is specifically sensitive to a shifting domain of ordered worlds. 3.2

Some attitude semantics

Kadmon & Landman argue that we should understand the semantics of and sorry as being intimately linked to the semantics of want. Very roughly-we shall have reason to become more sophisticated-the idea is that glad that p is almost the same as want that p, and that sorry that p is almost the same as want that not p. Now, if want is UE, then glad will be too and sorry will be DE, which is what we want Therefore, let's start with investigating the semantics of want, about which we can find much detailed discussion in Heim (I 992).

glad

A

want

rather simple-minded semantics for

want may look like this:

(36) [wantst•S(p)

(a) (w) = True iff 'v'w' E lilaXg(a,w) (f(a , w)) : w' E p 'Among the worlds f(a, w), the ones that maximally correspond to a's preferences in w are all p-worlds.'

This semantics is built on the two ingredients already mentioned: a set of worlds among which the subject a is said to have preferences and a way of ordering these worlds according to a's preferences in the evaluation world:

(37)

(i) The 'modal base function' f is a function from pairs of an individual and a world to a set of worlds. (ii) The 'ordering source function' is a function from pairs of an individual and a world to a set of propositions (in the case of want to a set of propositions forming the subject's preferences).

In the semantics of the attitude predicate, the ordering source (a set of propositions) is used to induce an ordering on the worlds in the modal base. The attitude predicate then makes a claim about those worlds in the modal base that maximally satisfy the preferences given by the ordering source. Here is how a set of propositions induces an ordering on a set of worlds: (38)

For any set of propositions P, we define a strict partial order < p: 'v'w', w": (w'

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

A Kratzer-style semantics for

116 NPI Licensing. Strawson Entailment, and Context Dependence

Then we pick the worlds in the modal base that are best by the ordering
(39) For a given strict partial order

(4o) [wants;] f, g (p) (a) (w) = True iff.V w ' E maxg ; (a,w) (fi ( a , w ) ) : w ' E p is still not complete. The predicate want will carry conditions on which kinds of modal base and ordering source need to be supplied to it. We'll tum to the question of what modal base it wants in a moment. But for now, note that by squinting at (4o) one can see that it will make want an upward entailing operator (which is what we need to carry out the Kadmon & Landman project). If all of the maximally preferred worlds in f (a, w) are p-worlds and every p-world is a q-world, then all maximally preferred worlds in f (a , w) will be q-worlds. If p entails q, then John wants p should entail John wants q.

As it stands, this

What is the right m.odal base for want? is that f (a, w) for want is the set of worlds compatible with a's beliefs in w, the set of doxastically accessible worlds for => in w.

A simple idea

(41 ) [wants;] f,g(p) (a) (w) is defined only if f i (a, w} = DOX(a, w), where DOX(a, w) is the set of worlds compatible with a's beliefs in w. If defined, [wants;] f,g(p) (a) (w) = True iff 'Vw ' E max g ;(a,w) (fi(a, yv ) ) : w ' E p This idea has some nice immediate consequences: it predicts that worlds which run counter to the subject's beliefs will be irrelevant for evaluating want-sentences. Consider in this light an example from Heim:

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The semantics of attitude predicates will be sensitive to these two parameters, and in tum the lexical semantics of an attitude can specify which kind of parameters it needs to work with. Since there are plenty of different kinds of attitudes, there must be plenty of different kinds of modal bases and ordering sources. It is therefore not quite right to specify the semantics as in (36) with just a particular f and g. Perhaps it would be best to give attitude predicates an index and think of the parameters f and g as being functions from indices to modal base functions and ordering source functions. We revise (36) to this:

Kai von Fintel 1 1 7

(42) I want to teach Tuesdays and Thursdays next semester. ·

Suppose I am sick. I want to get well. But getting well entails having been sick, and I do not want to have been sick. Suppose there was a murder. I want to know who committed the murder. But my knowing who committed the murder entails that the murder was committed, and I never wanted the murder to have been committed (Stalnaker 1984: 89).

Heim discusses a possible solution to this problem. Suppose that a wants that p presupposes that a neither believes that p nor believes that not p. We may try something like this: (43) [wants;] f,g(p) (a) (w) is defined only if (i) fi(a, w) = DOX(a, w) (ii) fi (a, w) n p "# 0 (iii) fi(a, w) - p "# 0 If defined, [wants;] f·S(p) (a) (w) = True iff 'tlw ' E nl.a.Xg;(o, w) (qa, w)) : w' E p You can only want something of whose fact you are not (yet} convinced. This would not make it false of the speaker of (42) that she wants to teach next semester, but it would make the claim suffer from presupposition failure. Heim suggests we can live with that result. But Heim argues that we need to refme the analysis yet some more. It seems that you can actually want something which you firmly believe will happen, as long as you think it's up to you whether it will happen. Heim's example is this:

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Heim writes about this: 'Suppose this sentence is intuitively true as spoken by me today. Is it therefore the case . . . that I teach Tuesdays and Thursdays next semester in all the worlds that are compatible with everything I desire? No. In worlds that are compatible with everything I desire I actually don't teach at all' (Heim 1992: 195). If a want-predication only makes a claim about the relative desirability of the worlds compatible with the subject's beliefs, we can explain (42). If the speaker assUlll.es that she will teach on some days next semester, all (42) claims is that among those worlds (in all of which she does teach) the most desirable ones are ones where she teaches Tuesdays and Thursdays. Unfortunately, this analysis makes not so nice predictions as well. If a believes that p, then we would predict directly that a wants that p. If all of a's belief worlds are p-worlds then the most desirable ones among those will also be p-worlds. This is not good. For one, it would predict that since the speaker of (42) believes she will teach next semester, it would be true to say of her that she wants to teach next semester. That does not seem right. Similar problems were noted by Stalnaker:

I I 8 NPI Licensing, Strawson Entailment, and Context Dependence

(44) Uohn hired a babysitter because) he wants to go to the movies tonight.

(45) [wantr;] f, g(p) (a) (w) is defined only if (i) fi(a, w) = DOX* (a, w) (ii) fi(a, w) n p # 0 (iii) fi(a, w) p -:/: 0 If defined, [wants;] f,g(p) (a) (w) = True iff 'Vw ' E lllaXg ,(o:,w) (fi (a , w)) : w' E p -

This semantics describes how the meaning of want (through its defined­ ness conditions) requires a certain kind of domain of quantification. It leaves open the nature of the ordering source, which is a gap that will have to be filled. For our purposes it suffices to say that want will take as its ordering source a set of propositions specifying the preferences of a m w. The apparent non-monotonicity of want

There is a kind of case, introduced by Asher (1987), that may suggest that we are wrong in analyzing want as an UE operator. Asher says that the following inference is invalid: ·

(46) Nicholas wants to get a free trip on the Concorde. So, Nicholas wants to get a trip on the Concorde. He writes: 'If I want to ride on the Concorde and not pay for it, it doesn't necessarily follow that I also want to ride on the Concorde simpliciter. It may mean bankruptcy!' (Asher i987: 171). One reaction to such examples, which is the one found in Heim {1992), is to see them as reasons to give up the monotonic analysis. What may be going on here is this: Nicholas thinks that among the worlds compatible

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

According to Heim., (44) 'certainly does not suggest in any way that John has the slightest doubt about where he will be tonight'. So, Heim suggest that instead of setting f(a, w) to the set of doxastically accessible worlds it should be taken to be the set of worlds compatible with 'everything that a in w believes to be the case no matter how he or she chooses to act'. This will always be a superset of the set of worlds compatible with everything a in w believes simpliciter. Let's call this new set DOX*(a, w). So far so good, then. It seems that with the proper understanding of what modal base want-claims quantify over, we have a satisfactory analysis of want that makes it an upward entailing operator. Here for future reference is the analysis we have arrived at:

·

Kai von Fintel 1 1 9

with his beliefs some are more likely than others. It is not likely that he will get a free trip, more likely he will have to pay $3000. What Nicholas wants a free trip means is that the most likely worlds in which he gets a free trip are better than the most likely ones in which he doesn't. This can be true, while at the same time it is false that Nicholas wants a trip, because the most likely worlds in which he takes a trip are expensive worlds and thus worse than the most likely worlds in which he doesn't take a trip at all. Heim presents such a non-monotonic analysis (involving both. a desirability ordering and a likelihood or similarity ordering). Here's a version:10

[wants;] f•B(p) (o:} Sim w1 ( • p) where Sim (p) w'

world in p}

(w) = True =

{w

'

iff 'V w ' E qo: , w ) :

Simw' ( p )


' E p : w resembles w no less than any other

This is, of course, much more complex than what we had before. The claim is not anymore that within the modal base (roughly given by the subject's beliefs) the worlds that maximally conform to the subject's preferences are ' p-worlds. Now, we would say this: for any world w in the modal base (still ' roughly given by the subject's beliefs in w) the p-worlds most similar to w are better (according to the subject's preferences in w) than ·the non p­ 1 11 worlds most similar to w • If this analysis is right, want is not UE. This does not leave much hope for the claim that regret is DE. Can it be that Asher and Heim are wrong? Well, I am not convinced that the Asher inference in (46} is really invalid. What is clearly invalid is the following: Nicholas wants a free trip on the

Concorde. Therefore, no matter how much it costs Nicholas_ wants a trip on the 10

This

formuhrion is essenrial.ly found in (F ) on p. l9J of Heim's paper. There is another possibility which doesn't use an additional ordering but makes want-claims much stronger and thereby also destroys UE inferences: (i) [wants;t·B(p) (a) (w) = True iff f; ( a, w) n p
This

than

This

than

take

any of the scenarios where Nicholas either doesn't go on the ttip or scenarios are more desirable pays through the nose for it. One cannot infer that all of the scenarios where Nicholas a ttip, whether for free or for a lot of money, are better any of the scenarios where he doesn't go on a ttip.

than

think

takes

that the semantics in (i) is far too srrong. It would make I want a Hontla Civic However, I equivalent ro something like I want any Honda Civic that I can think of. To me, for I want a Hontla Civic to be true it is enough if there is a small number of Honda Civics (perhaps even just one) with which I would be maximally happy. So, (i) cannot be right.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(47} If defined,

I 20 NPI Licensing, Sttawson Entailment, and Context Dependence

Concorde. But, why should we read the conclusion Nicholas wants a trip on the Concorde as being equivalent to No matter how much it costs Nicholas wants a trip on the Concorde? It may well be that Asher reads the conclusion of (46) as

having such a strong meaning. At least, I am not sure what else he might mean by his paraphrase I want to ride on the Concorde simpliciter. Heim (class notes) has a variant on Asher's example, which makes matters clearer: ·

What are we entided to conclude? At the time of John's first answer, it would seem that one could truthfully say:John does not want to buy this couch. At the time ofJohn's second answer, one could truthfully say: john wants to buy this couch at a 25% discount. But it is not clear that we have to conclude that at the very same time in the very same context the following statements are true:

(49) (i) John doesn't want to buy this couch but he wants to buy this couch at a 25% discount. {ii) John wants to buy this couch at a 25% discount but he doesn't want to buy this couch. In fact, these seem hopelessly contradictory. It seems that John's first statement in (48) has to be understood against the background of a set of worlds in which the couch has exactly the price stated on the price tag. Among those, the most desirable worlds are not ones where John buys the couch. By the time ofJohn's second stat�ment, more worlds are made accessible: apparendy the couch can be bought at a 25% discount. Worlds where John does buy the couch at that discount are highly desirable, so John wants to buy the couch at that price. Is the first statement still true in the new situation? I don't think so: it is now false that John doesn't want to buy the couch. In the new context, we would have to say that John doesn't want to buy the couch at its original price. But that is not the same {anymore) as saying that John doesn't want to buy the couch. Note that with the proposal in (45), we can say more precisely what .is going on. At first, John's set of worlds DOX*Uohn. w) will only contain worlds where the couch costs as much as is stated on the price tag that he s�es. Then the salesman asks the question whether John wants to buy the

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(48) John is in a furniture store, looking. at a couch that has a very scary price-tag. The salesman comes up to him and the following conversation takes place: Salesman: Would you like to buy this couch? John: No. Salesman: Would you like to buy it at a 25% discount? John: Yes.

Kai von Fintel 12. I

3·3

(Glad' and �orry'

As I mentioned, Kadmon & Landman argue that there is a very special connection between glad and sorry (and their cousins) on the one hand and want on the other. They write It is deeply rooted in the meaning of sorry that if you are sorry ;bout a fact A. given a certain perspective, that is because you want the opposite, i.e., -.A, relative to the same perspective . . . Vice versa, if the facts are opposite (the negation) of a certain wish you have, then you must be sorry about them, at least given the same perspective relative to which you have your Similarly, there is a special connection between being glad and one's wishes. If you are glad that A. that is because you want A, relative to the same perspective. Vice versa; if A is a fact, and you want A, then you must be glad that A. given the same perspective.

wish.

12

Shifting contextual grounds and in particulu shifts in �hich possible worlds are relevandy

access

ible for the evaluation of attirude sentences and other intensional sentences are discussed in a seminal paper by David Lewis (1979� see also his 1996 paper on knowledge claims.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

couch at a 25% discount. But because of the semantics of want in (45), this question carries the presupposition that there are worlds in DOX*(John, w) where the couch is bought at a 2 5% discount. This is not the case for the old DOX*(John. w), so John will have to revise his opinions about what is possible. DOX*(John, w) will be widened to include worlds where the couch is bought at a 25% discount. Within this newly widened modal base, John's preferences (which have not changed) will single out the worlds where he does buy the couch at that discount as the most desirable worlds So, perhaps a UE analysis of want is possible after all, as long as we pay attention to the shifting grounds of context. If p entails q, a wa nts p will entail a wants q, as long as it is assumed that q has the required relation to the set of worlds DOX*(a, w). q must be an open issue with respect to what a believes will happen no matter how s/he chooses to act. I£: as we have done, these conditions are formalized as presuppositions of want-statements, then we have that a wants p Strawson-entails a wants q. In a context where q does not satisfy these conditions, the assertion of a wants q may nevertheless be appropriate, as long as the context can be adjusted so as to then satisfy the conditions. This is what happens in the couch example. But inferences are checked with respect to a constant context, and so this kind of example does not provide counter-evidence against the claim that want is Strawson-UE.12 We can now move on to the factive attitudes. We need to establish, using the tools developed so far, that sorry is DE in the relevant sense and that glad is not.

1.22 NPI Licensing.

Strawson Entailment, and Context Dependence

Glad

Now, let's talk about glad. Based on our prior discussion of want, we can perhaps start with the following idea: (so) [glad;] f,g(p) (a) (w) is defined only if (i) DOX(a, w) � p (ii) DOX(a, w) � fi (a, w) (iii) fi ( a, w) = n p ¥: 0 (iv) fi (a, w) - p =I= 0 If defined, [glad;] f•S(p) (a) (w) = True 'Vw ' E maxg;(o,w) (fi(a , w) ) : w' E p

iff

The first definedness condition (i) here implements the £activity ofglad. One can only be glad that p if one thinks that p is true. We have a want­ style truth-condition: all the most desirable worlds among the relevant worlds are p-worlds. For this to make sense, the set of worlds f(a, w), among which a's preference are reported, must not be equated with DOX(a, w). Otherwise, you would be glad about anything that you believe to be true. So, again we impose a presupposition that among the worlds in f(a, w), there are some p-worlds and some non p-worlds, just as we did for want in (45). These conditions in (iii) and (iv) have the immediate effect that

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

There are two immediate problems with this discussion. It can't really be want that is involved in the semantics ofglad and sony. We saw that a want­ claim is really only appropriate vis-a-vis matters that are still open in a certain sense. Glad and sony, on the other hand, are, of course, factive predicates. They are attitudes towards matters that are now settled. What must be involved in their semantics is a notion of desire that is not restricted to open matters. In fact, Heim suggests that the factive attitudes like glad and sony involve counterfactuality. This prima facie astounding claim ('factive verbs involve counterfactualityl?') can be supported as follows. Both a is glad that p and a is sony that p presuppose that a believes that p (and often indicate that p is in fact true). The attitude that is expressed by such sentences compare the world as a believes it to be to the world as a believes it would be ifp were not true. The element of counterfactuality is therefore forced to be present by the £activity of these predicates. The other immediate worry about Kadmon & Landman's discussion is that they often talk as if attitUdes are relations to focts, whereas most semantic treatments see them as propositional attitudes. Some of the persuasiveness of their argumentation may come from the particular way in which they talk about facts. I will come back to this point in a little while.

Kai von Fintel

I23

If I'm glad he bought a car, then it clearly agrees with my wishes that he bought a car. What ought I to feel, then, about his buying a Honda? Well, the purchase of the Honda has certainly satisfied my wish that he buy a car. Hence, I can easily be glad that he bought it, qua car, i.e. in as much as it 53tisfies the wish. This explains the fact that speakers often judge that we have here DE on a constant perspective.. However, note that I am not necessaril y FORCED to be glad that he bought the Honda, because buying a Honda is not by any means required for satisfying my wish. My wish could be satisfied in another way, for example, by buying a Toyota . . . Hence, glad is not DE.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

some counterfactual worlds (some non p-worlds) must be in the modal base, must be relevant alternatives. To prevent the modal base from totally going off into counterfactual space, we require it to properly contain the subject's doxastic alternatives (condition (ii) ). How is the modal base selected? How many 'counterfactual· worlds' are relevant? The de6.nedness conditions (i)-(iv) in (so) provide a lower bound on the extent of the modal base of worlds compared by a glad-statement. Can we say more about upper bounds? I can think of two plausible ways one might go about selecting a proper modal base for a glad-statement. (i) Start with DOX(a, w), add to it those non p-worlds that are most similar to w, then add any worlds not already in DOX(a, w) that are more similar to w than the most similar non p-worlds. The glad-statement would then amount to the claim that the worlds that best fit the subject's preferences are p-worlds even if one takes into consideration those non p-worlds that are most similar to the evaluation world. (ii) Many if not all evaluative attitude sentences involve attitudes towards the outcome of actions or events in the past. Perhaps to figure out which worlds are relevant to a glad that p­ statement, we should go back in time to a point where it was not yet determined that p would happen or turn out true. From that point on, go forward and collect all worlds that could have developed out of that situation, most importantly any world that might have resulted from 'deciding' that not p. For all I know, this second procedure is not importantly distinct from the first. Perhaps, what would he used in finding those non p-worlds that are most similar to w is exactly this method of going back into the past to a point where it was not yet sure that p. I will leave this aspect of the semantics of attitudes under­ developed. I can dimly see an elaboration of the analysis that dynamically updates the domain of quantification for attitude ascriptions in the manner just sketched. But exploring this would lead us too far astray here. We will concentrate on the question of the monotonicity behavior of the attitude predicates. . K.admon & Landman are concerned with showing that glad is not DE, even in a sophisticated context-dependent semantics (they are answering a concern raised by Linebarger). So, they add the following discussion:

12.4 NPI Licensing, Strawson

Entailment,

and Context Dependence

While I think Kadmon & Landman are right when they say that glad is not DE, I think that they nevertheless have the wrong semantics for glad. What is wrong about it is the following inference, which they implicitly endorse in the preceding quote:

(5 1) a wants p p

(a knows p) :. a

is glad that p

(52)

If defined, [glad;] f• g(p) (a) (w) = True iff DOX(a , w)
The new analysis demands that among the compared worlds the ones that are compatible with the subject's beliefs (which are all p-worlds and which for all the subject believes could be the actual world) are better (according to the subject's preferences in the actual world) than all the relevant 1 3 One could consider another way of addressing my concern. a non-monotonic analysis with the following truth-condition (�erything else stays the same): (i) fi (o:, w) fi(o:, w) - p This analysis d that not just the actual world (or rather any world that for all the subject believes could be the actual world) �ut !U relevant p-worlds be more desirable than � relevant non p-world. This is stronger than my proposal in (sz). I think that my proposal is correct. Assume .again that I want to buy a Honda Civic, because certain Honda Civics would Dl2Ximally satisfy my needs for a car. I end up buying one of those Honda Civics. I will be glad that I bought a Honda Civic, even if there arc relevant (but counterfactual) worlds where I buy an awful Honda Civic which are less desirable than worlds where I buy an adequate Toyota Corolla.

�a:J/o.w)

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

I think that's invalid. I want to buy a Honda Civic, because a Honda Civic would maximally satisfy my needs for a car. I buy a Honda Civic, which turns out to be a real lemon. It is still true that I desire a Honda Civic, just not the one I actually have. So, I'm not glad· I bought a Honda Civic. I wanted to, I did, I still think it was a good idea, but I'm not glad I did. In more abstract terms: according to (so), in order to be glad that p it is enough that the most desirable relevant worlds are p-worlds and that the actual world (or at least any world compatible with the subject's beliefs) is a p-world, but it is not necessary that the actual world is one of those p­ worlds which are among the most desirable worlds. This seems insufficient to me. Kadmon & Landman may actually agree with me, since they like to talk about facts rather than propositions. What they say is that if A is a fact and I want A, then I must be glad about A. For facts that may. in fact be right (even though I don't know what an analysis running on facts would look like). I would like to suggest that the semantics in (so) be replaced with the following semantics, which is still an upward entailing analysis:13

Kai

von Fintel I 2 S

non p-worlds in the modal base. Note that the main result that Kadmon & Landman care about still holds: glad is not DE, in fact it is UE. For me to be glad that I bought a Honda, the actual world (or rather any world that for all I believe tnay be the actual world) will have to be better than any relevant world in which I don't buy a Honda (including worlds where I buy a Toyota and worlds where I don't buy a car at all). Hence, I will be glad that I bought a car. That's enough about glad for the time being. We will soon come back to how one should deal with apparently coherent sequences such as Fm not glad that I bought a car, but at least Fm glad that I bought a Honda Civic or Fm not glad

(example due to the anonymous reviewer). Sorry

. We tum to sony (and regret, and with suitable adjustments surprised, amazed, etc.). We need to find a semantics that is Strawson-DE and that will allow us to deal with the apparent counter-examples like (3 I ). There are some obvious candidates to consider. Two that I will discuss here are the 'mirrors' of the two meanings we explored for glad. Here they are (we assume the same definedness conditions as for glad):

(53) If defined, [sony;t·S(p) (a) (w) = True iff 'Vw' E m� ; (o,w} { fi (a, w) ) : w ' .
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

she died but Fm glad she died peacefully/in her sleep/with a clear conscience

12.6 NPI Licensing, Strawson Entailment, and Context Dependence

Kadmon & Landman argue that sorry is downward entailing, which would mean we should pick (53) as the correct analysis. Here's their argument. They consider three examples and argue as .follows: (s s} rm glad he bought a car (s 6} fm sorry he bought a car (57} He bought a Honda. .

.

Now consider (s 6). If I'm sorry he bought a car, I clearly wish he had bought no car What ought I to feel, then, about his buying a Honda? I ought to be sorry about it, qua car In fact, and this is where (56} differs from (55� I MUST be sorry about it, qua car That is because refraining from buying a Honda is an absolute requirement for satisfying my wish. I cannot prefer for my wish to be satisfied in 'another way'. Hence, sorry is DE (on a constant perspective). The crucial difference should be clear: the wish associated with (ss} places no particular demand on the truth value of (57), whereas the wish associated with (56} requires it to be false. .

.

.

(s 8) [ wish;] f,g(p) (a) (w) is defined only if (i) DOX{a, w) n p = 0 {ii) DOX{a, w) � fi(a, w) (iii) fi(a, w) n p =F 0 (iv) fi(a, w) - p =F 0 If defined, [ wish;] f•g{p) (a) (w) = True iff Vw ' E maxg;(o,w) (fi (a, w)) : w ' E p ·

The meaning in (ii) is the mirror of the meaning for g/aJ mentioned in £n. 1 3. It demands that all relevant non p-worlds be better than �y relevant p-world. not just the ones that the subject believes might be the acwal world. Since this is the mirror of the rejected meaning for glaJ, it must be rejected for the same reason. According to this meaning. to be sorry that I bought a Honda, non­ Honda worlds don't just have to be better than the acwal Honda-world but better than any Honda I might have bought within the realm of additional counterfactual worlds in the modal base. That is too strong. It's enough that I loathe the Honda I actually bought

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

As far as I understand, this argument goes as follows: I'm sorry that he bought a car entails I wish that he had not bought a car, which in tum entails I wish that he had not bought a Honda, which in tum entails I'm sorry that he bought a Honda. Thus by transitivity, I'm sorry that he bought a car entails I'm sorry that he bought a Honda. QED The argument thus relies on two assumptions. One is that sorry that p is equivalent to wish that not p. The other is that wish is upward entailing. (Note by the way that Kadmon & Landman have switched from relating sorry to want to the more adequate relationship between sony and wish). If we adopt the following semantics for wish and pair it off with the meaning for sony in (53), we have what this argument needs:

Kai von Fintel 12.7 According to this analysis, wish is a counterfactual attitude: by (i), p is presupposed to be false in the subject's belief worlds. But the relevant modal base also contains p-worlds. The quanti£cational claim is that among the relevant worlds the ones that are best according to the subject's preferences are all p-worlds. (I leave it as an exercise for the reader to establish that this semantics indeed makes wish that TWt p and sony that p equivalent). ·

Curveball #1: Iatridou's conditional semantics for wish15

(59)

wishes that p: that p.

a

a

thinks that

if p were the case, s/he would be glad

This

representation is supposed to be a paraphrase that captures the meaning of wish that p as it would be delivered by an explicit compositional semantics. There is no implied claim that at some abstract syntactic level wish-sentences have the structure of (59). Here are some noteworthy properties of this analysis. First, note that the counterfactual conditional is embedded under think. Iatridou does this to explain why the counterfactual presupposition (that p is not true) does not (automatically) project out of the attitude context. Wish-statements pre­ suppose that the subject of the wish believes the complement to be contrary-to-fact; they don't necessarily presuppose that the speaker shares that belie£ In my proposal in (58), this 'subject-orientation' of the 15

For people not familiar with baseball: a curvehall is a hall -thrown in a particularly unexpected confounding way designed to catch the hitter off guard. I use the tenn here in one of its colloquial meanings. roughly a problem coming at one &om an unexpcctcd direction. 16 I have adjusted lanidou's formulation somewhat, hut the difference is not relevant here. and

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

While there is nothing obviously implausible about Kadmon & Landman's idea that sorry that p and wish that not p are equivalent, there is one reason to think that the semantics of wish is a little more involved (whether that would carry over to sorry is another issue). Iatridou (1998) describes how wish-statements in English and even more so their correspondents in other languages share properties of counterfactual conditionals. Roughly, English wish that p is often expressed in other languages as would want that p. Iatridou shows that the verbal mood/tense/aspect morphology found in the complement p is identical to that found in the antecedents of counterfactual conditionals, while the morphology found on predicate would want Qexicalized in English as wish) is identical to that found in the consequent of counterfactual conditionals. She proposes the following representation.:16

uS NPI Licensing, Strawson Entailment, and Context Dependence

17 Let me mention on the side that latridou's proposal is reminiscent of Pesetsky's (1991: ch. 10) ideas about the analysis of complement if-causes in English, as in the following: (i) Emma would like it if Orestes were here. Pesetsky builds on prior work by Williams (1974= especially 1 57-64), Steriade (1981), and Pullum (1987� Examples similar to (i) include also: (under one of its two readings) (ii) Emma would be happy if Orestes were here. Emma would have liked Orestes to be here. Karina Wilkinson (pc to Pullum) had noticed that these if-causes cannot contain NPis, in contrast to normal conditional antecedents: (iii) *I would like it if anyone were to ask me about the painting. {Pesetsky's (z 3 s). p. 178) If anyone were to ask me about the painting. I would like iL Pesetsky explains the absence of NP!s in complement if-causes by saying that there is a Post-LF rule that turns the if�clause into a conditional antecedent but simultaneously copies its 'non-quantifica­ tional content' (anything other than the ingredient that makes it into a conditional antecedent) into the complement position of the attitude predicate. (iv) Emma would like it if Orestes were here. -> If Orestes were here, Emma would like it that Orestes is here. NPis are of course not licensed in J:!OSitive attitude complements: (v) *Emma likes it that anyone came to visit her. So, again it is the double role of these complements that prevents them from licensing NPis. For the cases discussed by Pesetsky, this is a pressing concern since it is an if-clause that fails to license NPis (while it is a fact that 'normal' if-clauses do license NPis� For latridou's case, there may be a more direct solution: maybe wish doesn't involve a counterfactual antecedent at all ·

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

counterfactual presupposition is directly encoded in condition (i). A simpler paraphrase (Ifp were true, a would be glad that p) would not capture this fact. The second noteworthy feature of the paraphrase in (59) is the fact that p occurs twice in it. I think that this move is necessary for two related reasons. Treating p as just a counterfactual antecedent in the meaning of wish would not really capture the meaning of the constrUction. a wishes that p means more than ifp were the case, a would be happy. The simple conditional would be trUe if the subject would be happy no matter whether p or not p. But that of course is not enough for the wish-claim to be trUe. Not only does the subject have to be happy but she has to prefer p to not p. The double role of p is also beneficial in that it prevents wish from displaying the same monotonicity behavior as counterfactual conditionals. As we will see in section 4, there is good reason to think that conditional antecedents are DE environments. But then an analysis of a wishes that p as meaning if p were the case, a would be happy would predict that the complement of wish is also a DE environment. That is of course ludicrous. If anything, it should be a UE environment (at least that is what Kadmon & Landman and I assume). Iatridou's paraphrase makes wish non-monotonic and thus correctly predicts that it will not license NPis in its complement.17 In Iatridou's proposal, a wishes that p is equated with a thinks that ifp were the case, s/he would be glad that p. Under natural assumptions about belief­ ascriptions, counterfactual conditionals, and the semantics of glad (as we analyzed it earlier), we would have the claim that for any world w'

Kai von Fintel 12.9

3·4

Shifting contexts

After these detailed ruminations about the semantics of various attitude predicates (a discussion which seems so interesting that one has to be glad that it became necessary in our ongoing quest for a theory ofNPI licensing), we will now tackle some of the examples that are prima facie problematic for the monotonic analyses. These will have to be analyzed as involving shifting contexts. The following sequences are both coherent: (6o) Sandy is glad that Robin bought a car, but Sandy is sorry/not glad that Robin bought a Honda. (6z) Sandy is sorry that Robin bought a car, but Sandy is glad/not sorry that Robin bought a Honda. ·

Using the monotonic analyses for glad and sorry that we developed in the previous section, the coherence of (6o) is straightforward. Assume that we have an ordering of worlds where the best worlds are ones where Robin buys a Toyota, in the next best worlds Robin buys a Honda, and in the worst worlds Robin doesn't buy any car. Then, one should be glad that Robin bought a car (because any world where she doesn't is worse than the actual world; even though she could have done better by choosing a different car, her choice was good enough to beat worlds without a car). And one should be sorry abou�her buying a Honda, because in the best worlds she buys a Toyota. ' " A simila.r (also non-monotonic) meaning for wish is sketched by Heim (r99z: zos}. 'john wishes you wut gone means "John thinks dut ifyou wut gone he woulJ be in a more desirable world than he is in Mutug you are not gone" ·.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

compatible with what a believes in the actual world (and keep in mind that a must believe that p is not true), all of the p-worlds w" most similar to w' are such that w" is better (according to a's preferences) than any relevant non p-world.18 latridou's discussion .is very interesting, but if she is correct it would disturb the delicate balance of my edifice. It makes wish that p non­ monotonic, which threatens the downward monotonicity of sorry that p, at least under the assumption that Kadmon & Landman were right in claiming the two constructions to be equivalent I could of course give up that assumption, but I can't see any good direct arguments against it. The other course of action is to hope that the morphosemantic facts that latridou wants to explain can be captured without making wish-sentences into covert counterfactual conditionals. But that can only remain a hope for now.

IJO

NPI

Licensing, Strawson Entailment, and Context Dependence

Direct argwnents against non-monotonic analysis?

In the case of want, we had reasons

independent of NPI licensing to reject the non-monotonic account. We argued that problexnatic cases depend on a change in the modal base, in particular a widening of the possibilities considered. We saw that it is quite incoherent to attribute the relevant attitudes 'in one breath' as it were:

(49)

(i)

(ii)

!!John doesn't want to buy this couch but he wants to buy this couch at a 2 5% discount. !!John wants to buy this couch at a 25% discount but he doesn't want to buy this couch.

Can we produce similarly· convincing examples in the case of Reversing the sequence does seem to introduce some oddness:

(62)

?Sandy is glad that Robin bought a Honda but Sandy is sorry that Robin bought a car.

sor:ry?

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The coherence ·of (6r) is not so straightforward. To be sorry that Robin bought a car means that the most desirable relevant worlds are non-car worlds, and thus a fortiori non-Honda worlds. Then it can't be that the actual Honda-world is better than any non-Honda world. What we have to say is that there is a shift in which worlds are being considered. The idea is that for the evaluation of the second sentence only worlds in which Robin actually buys a car are compared. Then, all the second sentence would claim is that the actual Honda-world is better than any of the worlds in which Robin buys a car other than a Honda. That is a claim which is compatible with the claim expressed by the first sentence with respect to a larger set of worlds. Linebarger (1987: 369) in fact had the pertinent intuition when she observed that such sequences involve 'covert con­ ditionalization'. She noted: john may regret that he assaulted a fellow patron in a restaurant because he was arrested and fined, but it may also be true that given that he did so, he is glad that it was the obnoxious George Smith that he assaulted.' The claim then is that the coherence of (6r) does not point to the necessity of a non-monotonic semantics for sorry (etc.) but is merely a reflection of the shiftable nature of context. Specifically, from one constituent sentence to the other the modal base function has changed. As natural as this is, the usual conception of validity of inferences is formulated to check truth-preservation with respect to a constant context. And in the view taken here, (61) does not involve a constant context and so its coherence does not furnish an argument against the monotonicity of sorry with respect to a constant context.

Kai von

Fintel

I3I

(63) Bill and Mary came => Bill came. I! I'm glad that Bill and Mary came. But I'm sorry that Bill came. !! I'm sorry that Bill came. But I'm glad that Bill and Mary came. Better: I'm sorry that Bill came. But I'm glad he came with Mary. The effect of focus can also be contravened by using the expression qua which is quite often employed by Kadmon & Landman. (64) IISandy is sorry that Robin bought a car, but Sandy is glad/not sorry that Robin bought a Honda qua car. I guess the idea would be that Sandy is glad Robin bought a Honda qua car signals that the gladness is not one that concerns the choice of a Honda over possible alternative brands.19 We then do have some reasons to think that the coherence of (61) does not follow as a matter of course from a non-monotonic semantics for sorry but depends on contextually signaled narrowing of the domain of quantification of the monotonic operator sorry. A closer look at the contribution of focus

We actually need to consider carefully the contribution of focus to the . interpretation of our attitude statements. If focus induces a change in the modal base, all is to the good: changes in the modal base disqualify the relevant patterns as potential counter-examples to downward inferences in a constant contexL But if the contribution of focus is more in�ernal to the semantics of the attitude ascriptions, for example if focus is taken as an internal restriction on :the very same modal base that the broader 1 9 It would be intriguing to study the semantics and pragmatics of qua. I illuminating to see what it does to the interpretation of adjectives: (i) John is large, qua man, but quite sm.all, qua basketball player.

think

it might also be

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

According to our story, the narrower attitude needs to be understood as implicitly conditionalized. In the good example in (61), the first sentence furnishes the proposition that Robin bought a car, which can then be used as the background for the second sentence. In (62), this natural sequence is disturbed. It also seems to me that focus structure plays a role. (61) is naturally read as having narrow focus on Honda, evoking contrasts such as Toyota, Mercedes, etc. And Linebarger's example above has a cleft He is glad that it was the obnoxious George Smith that be assaulted. So, the focus structure of the second sentence may support an interpretation of the attitude against the back­ ground of an understood conditionalization. If these supporting clues are removed, we do not seem to get sequences that are quite as coherent:

I 32. NPI

Licensing. Sttawson Entailment, and ConteXt Dependence

,

20 The first author to draw attention to the phenomenon was Dretske (1972., 1975).

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

statement was sensitive to, focus-sensitivity would make the semantics non­ monotonic. Let's think about this. That there is an effect of focus structure on the interpretation of attitude ascriptions is quite well known.20 This phenomenon of 'association with focus' is one that attitude ascriptions share with many other constructions (especially quantificational ones). The now widely accepted story of how this phenomenon works involves a semantics of focus based on alternatives. Let us assume that for any expression a, we compute not only its usual denotation [a] but also its focus denotation [a]F, which will be a set of denotations which are focus-alternatives to [a]. The principal effect of focus is to introduce a set of alternatives to the focused item. This can then be passed on 'up the tree' and can lead to sets of alternatives for bigger expressions. The focus on Honda in (61) first evokes a set of relevant contrasts X to Honda. Higher up what we get are alternative propositions ofthe form 'Robin bought a X'. For our purposes, what will suffice is reference to the union of this set of propositions. What we would have as the relevant object is the set ofworlds in which Robin buys a car of one of the relevant brands. So, by the time we have interpreted the complement Robin bought a [Honda Civic]p, we have as its ordinary interpretation the proposition that Robin bought a Honda Civic and as its focus interpretation the proposition that Robin bought a car. Within modem theories of focus semantics, there are now two main lines of research about what to do with these two propositions: (i) focus structure induces a presupposition about the context (Rooth 1992, 1996), (ii) focus stnicture can be accessed directly by the semantics of other operators (Krifka 1991a). The presuppositional theory will fit in nicely with my analysis of why affective attitudes license NPis. The narrow focus in the complement would merely signal the presence in the context of the proposition that Robin bought a car; a condition that is clearly satisfied. That this proposition is also taken as an implicit restriction on what worlds are relevant to the second attitude ascription in (61) would not be attributed directly to ·the focus·· semantics. Focus would just be a defeasible signal. The theory that employs reference to focus alternatives directly in the semantics of focus-sensitive operators would however clash with my theory. Here is how such a semantics may look like: (6s) Focus is used directly by the semantics [sorry;tha�] f·S(a) (w) is defined only if (i) DOX(a, w)� [p](, g (ii) DOX(a, w) � fi(a, w) (iii) fi ( a w)n[p] f•s =I= 0

K.ai von

Fintel

133

(iv) fi (a , w)- [ p] f, g =f 0

If defined, [sony;tha�] f·S ( a) (w) = True

iff 'v'w ' E IllaXg ; (a,w) (fi( a , w)) n U [ p]i'8 )w' E [ p] f, g

Curveball #2: the DE-ness of focus-sensitive only

Danny Fox (pc) asks what happens when we try to determine whether only is Strawson-DE, once we move beyond the simple cases of only plus proper name that were discussed in section 2. Consider: (66) There only was any precipitation in (MEDford]F· (67) There only was precipitation in (MEDford]f. (There was rain in Medford.) . ". There only was

rain

in [MEDford]F.

We can take (66) as a representative of a much larger class of cases. Only is a cross-categorial operator that (like negation and conjunction) can combine with many kinds of expressions. As discussed in von Fintel ( 1997), a popular story (developed by Rooth 1985) is to reduce all such uses of only to a base case where only acts as a propositional modifier, a kind of generalized negation with respect to an understood set of alternative propositions. (66) would claim that among �e relevant alternative propositions the only true one is that there was precipitation in Medford. The focus structure of the sister proposition of only (which I call its 'prejacent', following medieval semantic terminology) gives clues about the set of relevant alternatives. With the focus on Medford in (66), we have a signal that only propositions

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Under the second proposal. there is no change in the contextual parameters. Instead, internal to the semantics of sony, we directly employ the focus denotation of the complement sentence by intersecting the contextually supplied set of relevant worlds with the set of worlds evoked by the focus structure of the complement. If we adopt this analysis, the example in (61) will not involve context shift and thus would he a crucial counter-example against the thesis that sony is DE. To maintain our story, then, we have to come down on the side of the purely presuppositional theory of focus structure (further discussion of the attraction of the indirect theory of association with focus can he found in von Fintel 1 994). Examples where focus effects a narrowing of the set of worlds compared by an attitude predicate are to he seen as examples where a contextual parameter is shifted. Whether a predicate is DE can only he judged in contexts where no such context shift occurs.

I 34

NPI Licensing. Strawson Entailment, and Context Dependence

about the weather in places that are relevant alternatives to Medford count as negated by only. As we see in (66), these uses of only also license NPis in their prejacent. And as we see in (67), we also observe Strawson-DEness. So, there does not seem to be any problem. There is one worry, however. We need to make sure that the semantics of only is stated in a way that delivers these facts correctly. And here, it is easy to go astray. In von Fintel (1997), for example, essentially the following two entries for propositional only are given as options between which one may choose freely:

What I was dealing with at that point was the problem of how to make sense of the claim that the only true proposition in the set of relevant alternatives is the one that it rained in Medford. Clearly, we don't want this to make the impossible claim that none of the entailments of that proposition are true. So, one possibility (the one in (a) ) is the claim that only makes is weakened to the claim that the only true propositions in the set of relevant alternatives are the proposition that it rained in Medford and any of the propositions entailed by that one. The other possibility, the one in (b), is to say that propositions entailed by the prejacent are not legitimate alternatives, that only sets of propositions C will be accepted that do not contain any entailments of the prejacent. At the time, I did not see any empirical reason to choose one option over the other, and I proceeded to adopt (b) for concreteness. This was exactly the wrong choice as we can see now. Adopting (b), we get a machinery that predicts that only will not license. downward inferences, within the same set of alternatives C (because C is devoid of any propositions entailed by the prejacent). The semantics in (a) correctly gives only the required Strawson-DE property. In this story as well, vre need to say that the focus structure of the prejacent does not force a different set of alternatives C on us. If for the premise in (67) only propositions of the form 'there was precipitation in X' are allowed in C, and for the conclusion only propositions of the form 'there was rain in X' are allowed in C, there would have to be two different

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(68) For all sets of relevant alternative propositions C, propositions p, r, worlds w: a. [only] (C) (p) (w) is defined only if p(w) = True If defined, [only] (C) (p) (w) = True iff 'Vr E C: r(w) = True => (p => r) b. [only] (C) (p) (w) is defined only if p(w) = True and (ii) - d r E C: p � r If defined, [only] (C) (p) (w) = True iff 'Vr E C: r(w) = True => (p = r)

Kai von Fintel

I3s

sets C of alternatives, and thus no constant context within which to assess the downward inference. Much of these complications would be avoided if we treated cases like (66) as not involving propositional only but a logical form where the associate of only (here, the proper name Medfort!) forms a constituent with only. The logical structure would be: Only Medford is such that there was rain in it. Then, the simpler semantics discussed in section 2 would be all we need. But this move is one I cannot consider any further here.

Where we are

We have reached a point where we are exploring the idea that the notion of downward entailment that NPI licensing is sensitive to has two important properties: (i) we need Strawson Entailment, because presuppositions carried by the conclusion in downward inferences don't seem to disrupt NPI licensing; (ii) contextual parameters need to be kept constant even if in a natural conversation they would normally evolve in a certain way. With these tools in hand, we tum now to the semantics of conditionals.

4 ANTE CEDE NTS O F C O NDITI O NALS

We will go through the same kind of dialectic as before. NPis are licensed where there is reason to think downward entailingness does not hold. A context-savvy semantics may get around that. But first a red herring needs to be taken care o£

41 ·

the problem

NPis are licensed in the antecedent of conditionals:2 1 (7o)

a. c.

d.

IfJohn subscribes to any newspaper, he is probably well informed. Ifhe has mr told a lie, he must go to confession. Ifyou had left any later, you would have missed the plane.

This fact is problematic from the point of view of the Fauconnier-Ladusaw generalization, since co�ditional antecedents are not obviously downward 21

Partee (1992) shows that the licensing of NPI.s in if-clauses is not some dumb mistake of the restricts a non-univenal q�tifier, where it is uncontroversial that there is no downward monotonicity, NPI.s are not allowed: (i) #Sometimes, if a man feecb a dog any bones, it bites him.

grammar. If the if-cl�use

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

3·5

1 36

NPI

Licensing, Sttawson Entailment, and Context Dependence

monotone contexts. In fact, in the modem semantic and philosophical literature on conditionals, it is now taken for granted that conditionals are not monotonic in their antecedent, in that they are claimed to not validate the inference pattern known as Strengthening the Antecedent. Some spectacular failures of this pattern are as follows:

(7o)

If I strike this match. it will light. � If I dip this match into water and strike it. it will light. b. IfJohn stole the earr:ings, he must go to jail. � Iflohn stole the earrings and then shot himself. he must go to jail. c. If kangaroos had no tails, they would topple over. � If kangaroos had no tails but used crutches, they would topple over. a.

(71) If lohn subscribes to a newspaper, he must be well informed. � If John subscribes to a newspaper that he can't read, he must be well informed. Kadmon & Landman suggest that conditionals are downward monotone after all, as long as we keep the context constant for the whole stretch of the argument. The same idea is advocated by Katz (1991). None of these authors addresses the fact that the standard Stalnaker-Lewis analysis (and Kratzer's variations, which those authors primarily refer to) actually does claim to keep the value of contextually supplied parameters constant.

The standard non-monotonic semantics for conditionals Let us see how the standard non-monotonic analysis of conditionals might go. We will formulate it in t.erms by now familiar to us. Conditionals are seen by Kratzer as the result of combining a modal operator with· a restrictive if-clause, which is analyzed as narrowing down the modal base that the modal operator is sensitive to. The ordering source is used to select a particular subset from the set of worlds in the modal base that the

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Heim (r984) proposed to weaken Ladusaw's analysis to only demand a 'limited' kind of downward monotonicity� The idea was to keep more of the environment constant. Instead of checking for entailment between two arbitrary antecedents, let them only differ in the place of the NPL Kadmon & Landman (1993) show that this move is not enough. The failure of Strengthening the Antecedent extends to cases where only the position of the crucial NP is manipulated:

Kai von Fintel I 3 7

Such a sentence claims that among the worlds in the modal base (assigned to the evaluation world) in which the antecedent is true, the ones that are best according to the relevant ordering source are all worlds in which the consequent is true.22 zz Formulating the Stalnaker-Lewis semantics in terms of selecting the maxitmlly best worlds is only possible under what Lewis calls the Limit Assumption, which Lewis in fact rejects. Stalnaker, on the other hand, defends the assumption against Lewis arguments by saying that in acrual practice, in acrual natural language semantics and in acrual modal/conditional reasoning. the assumption is eminently reasonable. Kratzer is persll2rled by Lewis' evidence and does not make the Limit Assumption; hence her semantics for modals is more convoluted than what I present here. I side with Stalnaker, not the least because it makes life easier. For discussion, see Lewis (1973) and Stalnaker (1984= ch. 7, esp. 14o-2.� Further arguments aga inst the Limit Assumption can be found in Herzberger (1979) and Pollock (1976: 1 8-2.0). Further arguments for the Limit Assumption can be found in Warmbrod (1982.� An assumption that I do not share with Stalnaker is the Uniqueness Assumption: that for any p. the set of maximally best p-worlds in the modal base is in fact a single p-world. Stalnaker uses this assumption to justify the Conditional Excluded Middle, the inference &om not (ifp, would q) to ifp, would not tJ· To deflect worries �t no realistic conteXt will supply an ordering source (similarity measure in the case of counterfacnials) that is sharp enough to distinguish p-worlds so finely as to single out one p-world as the best one, Stalnaker sketches an alternative employing supervaluation. My own preferred approach to the Excluded Middle is to say that those modal operators that validate this inference carry a presupposition of homogeneity-they presuppose that all p-worlds behave uniformly with respect to the truth of 1/· This approach is sketched in von Fintel (1997� '

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

antecedent is true in. The idea is that different conditional constructions differ in what exactly their modal base and their ordering source is. For example, counterfactuals (involving the modal would) take as their modal base the entire set of possible worlds and as their ordering source a set of propositions that is 'totally realistic' (taken together these propositions uniquely describe the evaluation world). Such an ordering source will thus single out from the worlds in the modal base those that are most similar to the evaluation world, most similar according to the particular description of the evaluation world encoded in the ordering source. In Lewis ( I 98 I), it is shown that this treatment of counterfactuals makes exactly the same predictions about the logical behavior of counterfactuals as the familiar Stalnaker-Lewis analyses (modulo possible refinements). Other conditionals may take as their modal base a set of propositions encoding the current state of knowledge in the utterance situation and use as their ordering source assumptions about the stereotypical course of events. This would make them 'epistemic conditionals'. And so on. The general schema for the interpretation of conditionals according to this view can be detected in this proposal for the semantics of the universal modal would as restricted by an if-clause:

I38

NPI

Licensing, Strawson Entailment, and Context Dependence

Consider then: {73)

If I

f;

had struck this match, it would have lit. had dipped this match into water and struck it, it would have

If I

lit.

{74) If the USA threw its weapons into the sea tomorrow, there would be war; but if the USA and the other nuclear powers all threw their weapons into the sea tomorrow, there would be peace. This

speaker simultaneously asserts a counterfactual conditional and the negation of a counterfactual conditional derived from it by Strengthening the Antecedent. Lewis deliberately put this example in the form of a single run-on sentence, with the counterfactuals conjoined by semicolons and but. This is meant to ensure that the context stays constant throughout. There are two options within our current project of trying to salvage the Fauconnier-Ladusaw account of NPI licensing: {i) We could try to argue against the standard non-monotonic account of conditionals. .Kadmon & Landman and Katz wish to pursue the first line of .analysis; but as mentioned above, they mistakenly assume that Kratzer's semantics is a monotonic analysis and so they don't actually go very far. (ii) We could try to make something out of the fact that the non-monotonic semantics involves a 'superlative' ingredient (this strategy was suggested to me by Angelika Kratzer). Let us pursue the second option for a moment. 4.2

I

Supedatives and NPI licensing

Superlatives license NPls: (75)

Emma

is the tallest girl to ever win the dance contest.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

According to the non-monotonic account, this inference is semantically invalid. The premise merely claims that the most highly ranked worlds in which I strike this match are such that it lights. No claim is made about the most highly ranked worlds in which I first dip this match into water and then strike it. The reason for the invalidity of (73) becomes perceptible because it is natural to assume that the most highly ranked worlds in which I strike this match are ones where I make sure that the match is dry. Quite importantly, the account maintains that the crucial examples are cases where the context remains relevantly the same throughout the examples. Lewis attempts to demonstrate this with the following kind of example:

.K.ai von

Fintel I 39

But again, we are disappointed to find that superlatives are not straight­ forwardly DE: (76) Emma is the tallest girl in her class. fo. Emma is the tallest girl in her class to have learned the alphabet. _

Superlatives are, however, Strawson-DE: (77) Emma has learned the alphabet. Emma is the tallest girl in her class. ::::? Emma is the tallest girl in her class to have learned the alphabet.

(78) The largest mammals are over 100 feet long. (Angelika Kratzer, pc) The largest mice are over 1 oo feet long. What is going on? The only discussion of NPI licensing in superlatives that I am aware of is Hoeksema (1986b), where he notes that superlatives have a limited kind of downward monotonicity, an idea that seems very similar to the idea of Strawson Entailment. Let me try to spell out how a story might go. We need a meaning for the superlative morpheme that fits into the structure a is the P-est Q. Assume that _the definite determiner here is idle or is interpreted as part of a unit together with the superlative morpheme. Observe that a is the P-est Q presupposes that a is a Q. It then asserts that among the other Qs all are P to a smaller degree than a. (79) [the ; . . -est] (P) (Q) (a) is defined only if Q (a) = True If defined, [the . . . -est] (P) (Q) (a) = True iff Vx � a : (Q(x) = True _. td P (x) (d) < t d P (a) (d)) For this to work Q needs to be a normal one-place predicate like girl in her class or girl in her class who has learned the alphabet.23 The predicate that the superlative attaches to must be a gradable ·one, which I treat here as a relation between an individual x and a degree d.24 By this semantics, it is invalid to infer downward in the position of Q, since the conclusion may suffer from presupposition failure. But the inference is Strawson-valid, since under the assumption that a satisfies 23 I will not discuss the intcrcsc:lng fact that relative clauses under a superlative operator are often realized as infinitivttl relative clauses. 2• There is, of course. relevant literature on comparatives and gradable predicates and so on. which I entirely ignore here. My modest aim is to present an analysis that is just plausible enough for our purposes.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

This behavior is clearly visible only when the superlative predicate is used in an otherwise simple sentence. When a superlative is used inside an argument definite description, a DE inference will not go through.

1 40 NPI

Licensing. Strawson Entailment, and Context Dependence

the subset property, the truth of the superlative premise guarantees the truth of the superlative conclusion. Now, (79) treats the superlative as a predicate-creating operator taking both a comparison predicate (P) and a domain predicate (Q) as its arguments. When the superlative predicate is used as the predicate inside a definite description, for example when the tallest girl is used as the subject of a sentence, the Strawson-DEness of the superlative will be obliterated. Consider for example:

The inference in (8o) is not even Strawson-valid. If the conclusion of (8o) has a presupposition it is that there is a girl in Emma's class that is taller than any other girl in Emma's class. This presupposition will not rescue the inference in (8o). What would be needed is the assumption that the tallest girl in Emma's class is the same as the tallest girl in Emma's school But this assumption is not available as a presupposition. The reason why the superlative licenses NPis is then that the superlative morpheme as defined in (79) is Strawson-DE. The fact that once such a structure is further embedded (as it is when it used as the restriction of a deftnite description operator), downward inferences are not licensed does not change the local Strawson-DEness of the superlative morpheme. This property only 'shines through' in an unadulterated way when the super­ lative predicate is used predicatively, without being embedded in further structure. Now what about conditionals?

My explanation for NPI licensing by superlatives depended on teasing apart the contribution of the superlative morpheme from that of the higher environment (including the deftnite description operator). This is of course legitimate because we can observe the superlative morpheme at work in an unadulterated fashion in predicative uses, as in (75). To carry this kind of story over to conditionals, one would have to claim that at a level relevant to the determination of NPI licensing, we are dealing not with simple modal operators like would qut with complex structures of the form the closest P-worlds. In other words, what is a paraphrase in the usual semantics· for conditionals must be turned into a hypothesis about the relevant syntactic structure of conditionals. This is a project that I ftnd implausible.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(8o) The tallest girl in Emma's school is over four feet tall fo. The tallest girl in Emma's class is over four feet tall

Kai von

4· 3

Fintel 141

A Monotonic semantics for conditionals?

(8 1) Admissible Modal Horizons A function D from worlds to sets of worlds is an admissible modal horizon with respect to the ordering source g iff for any world w, Vw"(w" :$: g(w) w' -+ w" E D (w) ) . (82) [woultl;]0·g (if p) (q) (w) is defined only if {i) Di is adm.issible"·with respect to gi (ii) Di (w) n p =/: 0 (p is compatible with the modal horizon) (8 3) If defined, [woultl;]0·g (if p) (q) (w) = True iff Vw' E Di (w ) n p: q(w) = True.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Our remaining hope then is that conditionals are in fact monotonic and that the examples that are supposed to show the invalidity of Strengthening the Ai:ttecedent involve context shifts. Authors who claim that this would be the correct analysis are not too hard to find. Approaches that give a monotonic semantics and explain non­ monotonic behavior by appeals to pragmatics are very popular for indicative conditionals. Accounts in a roughly Gricean vein (Grice 1967, 1989) continue to be refined; see especially Jackson's work (1979, 1984, 1987, 1 990). Other pragmatically informed analyses include Veltman (1986) and McCawley (1993: 5 48£f). Stalnaker himself defines a notion of 'reason­ able inference' in his paper on indicative conditionals (Stalnaker 197 5 ) which bears some resemblance to what I will develop here. Pragmatic approaches to the interpretation of subjunctive conditionals are advocated in some form or another by Warmbrod (198 1a, b, 1983), Wright (1983), and Lowe (1990, 199 5). None of these works quite goes where I go in my paper 'Counterfactuals in a Dynamic Context' (von Fintel 2ooo). They employ pragmatic considerations to explain away counterexamples. Instead, I admit that the counterexamples are genuine but deny that they force us to adopt a static non-monotonic semantics. Rather, the source of non-monotonicity is in the dynamics of domains of quantification. I refer the interested reader to that paper for details. The idea of my dynamic analysis is that there is an ever-widening 'modal horizon', which we will model by a function from worlds to sets of worlds, i.e. an accessibility function. We will use the ordering source parameter for a slightly different purpose than iii Kratzer's system: it will be used to make sure that the evolution of the modal horizon is such that it always forms a well-behaved 'Lewis-sphere' around the evaluation world. The domain of worlds that a modal/ conditional operator quantifies over is given by intersecting its antecedent with the modal horizon.

142 NPI

Licensing. Strawson Entailment, and Context Dependence

Evidence for context shift There is some evidence that the counter-examples to SA in fact involve such context shifts. Consider the clear contrast between Lewis example and a variant due to Irene Heim: '

(74) If the USA threw its weapons into the sea tomorrow, there would be war; but if the USA and the other nuclear powers all threw their (84)

weapons into the sea tomorrow, there would be peace. ??If all nuclear powers threw their weapons into the sea tomorrow, there would be peace; but if the USA threw its weapons into the sea tomorrow, there would be war.

In (84), the two counterfactuals claimed to be consistent by Lewis are reversed in their order and · the sequence does not work as before. The reason seems intuitively clear: once we consider as contextually relevant worlds where all nuclear powers abandon their weapons, we can't ignore them when considering what would happen if the USA disarmed itself We

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Counterfactual sentences carry a presupposition about the context: the context has to be well behaved in the sense of providing an admissible modal horizon. The counterfactoal is further only defined for worlds to which the modal horizon assigns a set of accessible worlds that is compatible with the proposition expressed by the antecedent. (In effect, counterfactuals thus carry an existence presupposition with respect to their domain of quantification.) In cases where the context does not already provide a modal horizon that obeys the compatibility condition, a new context will have to be created to satisfy the presupposition of the counterfactoal. A natural procedure for repairing the context would of course be to minimal1y expand the modal horizon so as to assign antecedent worlds to any evaluation world. In my paper 'Counterfactuals in a Dynamic Context', this procedure of enlarging the modal horizon in a constrained way is formalized in a fully dynamic semantics. Here, we can stick with the static presuppositional format. The semantics given here is Strawson-DE. Under the assu mption that the modal horizon is wide enough to be compatible with both p and p & r, the inference from if p, would q to if p & r, would q will be valid. The same inference is not valid without making this assumption. Thus, if p & r is not compatible with the initially selected modal horizon, the presupposition of the conclusion would not be satisfied. A constrained change of the context would then occur, selecting a slightly larger modal horizon. But this may in fact remove the justification for holding the initial conditional.

Kai von

Fintel 143

(8 5)

A:.

B:

If the USA threw its weapons into the sea tomorrow, there would be war; but if the USA and the other nuclear powers all threw their weapons into the sea tomorrow, there would be peace. But that means that if the USA threw: its weapons into the sea tomorrow, there wouldn't NECESSARILY be war.26

zs What I mean by 'no longer true' is not that the objective facts have changed. It is the parameters of the discowse that have changed so that the proposition expressed by the first counterfactual in the initial context can no longer be expressed by the same linguistic expression in the new context. Compare the W:t that the claim that Frana is hcct!goMI may be true in a context where it is preceded by Italy luu 1M sluzpe ofa boot, but cease to be true in a later context where the standards of precision have been sharpened. 26 Note that the stress on �assarily is required. B cannot say (i) or (ii): (Q But that means that if the USA threw its weapons into the sea tomorrow, there would Nar be war. (ii) But that means that it is not TRUE that if the USA threw its weapons into the sea tomorrow, there. would be war. . The reason for this is investigated in von Fintel (1997� The idea is that bare conditionals obey the Excluded Middle and that therefore negating them either has a very strong meaning or needs to be done by using an explicit operator that does not obey the Excluded Middle.

.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

seem to be in need of an account that keeps track of what possibilities have been considered and doesn't allow succeeding counterfactuals to ignore those possibilities. An account according to which the context remains constant throughout these examples would not expect a contrast between these two orders. As mentioned before, Lewis takes the coherence of (74) as a sign that the context does not shift. Similarly, Edgington (1995: 252£) presents the following scenario: 'a piece of masonry falls from the cornice of a building, narrowly missing a worker. The foreman says: "If you had been standing a foot to the left, you would have been killed; but if you had {also) been wearing your hard hat, you would have been alright".' Edgington says, quite correctly, that the building foreman's remarks constitute 'a single, pointful piece of discourse'. One can easily read them as a shrewd way of putting the suggestion that the worker should wear her hard hat at all times. The fact that (74) and Edgington's example are 'single pointful pieces of . discourse' argues against attempts at dismissing them as cases of illicit equivocation. But there is no argument here against the idea that the context can and does change over the course of simple pointful discourses. The proper diagnosis would seem to be that over the course of (74) the modal horizon properly expands, but that over the course of {84) it cannot shrink. This asymmetry is unexpected if one maintains there is no context change. Note also that if someone utters (74), someone else can then rejoin that the initial conditional is 'no longer' true:25

144 NPI

Licensing. Strawson Entailment, and Context Dependence

B': But that means that if the USA threw its weapons into the sea tomorrow, there might NOT be war. This is unexpected under the standard static approach. If we go back to the simpler antecedent, the domain of quantification should shrink back to the closest worlds where just the USA disarms, ignoring the far-fetched worlds where all nuclear powers become meek. But that doesn't seem to happen.27 The Strawson-validity of strengthening the antecedent

s

CONCLUSION

We have explored the prospects of the Fauconnier-Ladusaw approach to NPI licensing. With judicious choices in the semantics of particul?Ily problematic constructions and with specific assumptions about what kind r�

Note that B ' seems to rely on an inference &om ifp and r, rvould q to ifp, might q. This pattern is the standard system, but is valid in mine.

invalid in

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

According to the analysis given in (8 1)-(8 3), Strengthening the Antecedent is not a truth-preserving inference; it will not be classically valid. This is a prediction that is confirmed by the counter-examples to the pattern that we saw. Moving to a strengthened antecedent may lead to failure of the presupposition that the current modal horizon is compatible with the antecedent. So, from a premise that is true with respect to a particular context, we might move to a conclusion that puts conditions on the context that are not satisfied. The immediate result would be presupposition failure. But in the normal case, the context may be adjusted so as to assume a slightly wider modal horizon. But then the truth of the antecedent would have to be reassessed with respect to the new context. Nevertheless, SA is Strawson-valid in my system. This means that under the additional assumption that the presuppositions of the conclusion are satisfied, the downward inference will be truth-preserving. The additional assumption is essentially one that ensures that the antecedent is not so novel or bizarre as to fall outside the current modal horizon. If the modal horizon is already wide enough to accommodate the strengthened antecedent, SA will be safe. If this is the correct analysis for conditionals, the Fauconnier-Ladusaw analysis of NPI licensing can be maintained. Conversely, if the Fauconnier­ Ladusaw analysis of NPI licensing is attractive, we have reason to adopt the Strawson-monotonic semantics for conditionals.

Kai von Fintel I4S

These two environments-conditionals and relative clauses headed by universal quanti­ fiers-represent particularly good cases of licensing by downward entailingness, and particularly unconvincing cases of licensing by NI [= negative implicatures (KvF)). They are unconvincing because they license NPis with no 'conscious' negative implicature, although NPis are unacceptable in environments where the possibility of NOT P is absent (p. I78).

Linebarger herself thus admits that conditionals do provide prima facie support for the Fauconnier-Ladusaw analysis over hers (although she does not discuss the problem of the apparent failure of Strengthening the Antecedent, save for a reference to Heim's paper). The one interfering factor that she mentions can be illustrated as follows: ·

(86)

a. #If you drink any water, you'll feel a whole lot better. b. #If you think Bob had any fun, you should have seen Fred!

As noted by Lakoff (1969), NPis are unacceptable in conditionals used as promises as in (a) or in conditionals in which the possibility that the antecedent is false is remote as in (b). What exactly the actual nature of the additional conditions on NPI licensing are must remain an open question here. Acknowledgements As usual, I stand on the shoulders of pioneers, here in particular: Bill Ladusaw, Larry Hom, Angelika Kratzer, Irene Heim, David Lewis, Nirit Kadmon, and Fred Landman. This paper is in llWlY respects an elaboration of ideas found in Katz {I 99 I} and discussed in a MIT se!lWltics seminar presentatiOn in the spring of 1994 by Irene Heim. The latter was a memorable occasion which inspired my paper 'Coun�rfactuals in a Dynamic Context'. The present paper is an offspring of a subsection on NPI licensing in conditional antecedents in an early version of that paper. Thanks to my colleagues Irene Heim and Sabine latridou for their encouragement and help. The work reported in this . paper was

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

of entailment notion we are after, one can in fact stick to the Fauconnier­ Ladusaw idea. The project is, of course, only as attractive as the particular semantic analyses that it depends on. The most striking aspect of the project is that it requires checking of inferences only in cases where the contextual parameters are not affected by the assertion of the test sentences. In as much as this is not a particularly common situation, especially from the point of view of dynamic semantics, the account, if true, would be quite surprising. We must also note that the algebraic condition on NPI licensing can only be a necessary condition. As shown by Linebarger, there are effects that cannot easily be accounted for on a pure Fauconnier-Ladusaw-style analysis. Linebarger (1991) discusses conditional antecedents as licensing environments for NPis and concludes:

I46

NPI

Licensing. Strawson Entailment, and Context Dependence

first presented informally to the MIT LF-Reading Group, after which Danny Fox asked a question which led to the material in the final subsection of section 34- The paper was then presented at the second annual conference on Sinn und Betkutung in Berlin (December I997� I distinctly recall helpful and encouraging remarks from Graham Katz, Rob van Rooy, and Amim von Stechow, which they may, ofcourse, deny in retrospect. Thanks also to an anonymous reviewer for the journal of Semantics. I wish to dedicate this paper to the memory ofJim McCawley.

VON FINTEL Department of Linguistics and Philosophy Massachusetts Institute of Technology E39-245, 77 Massachusetts Avenue CAmbridge, MA OZ1J9, USA e-mail:[email protected] http://web.mit.edu/fintel/

Received: 07·09·98

KAI

Final version received: I 8.04-99

Asher, Nicholas (I987� 'A typology for attitude verbs and their anaphoric properties', Linguistics and Philosophy, Io, 1 27-89. Atlas, Jay (I993), 'The importance of being only: testing the neo-Gricean versus nco-Entailment paradigms', journal of Semantics, 10, 30I-I8. Atlas, Jay (I996), ' "Only" noun phrases, pseudo-negative generalized quantifiers, negative polarity items, and mono­ tonicity', Journal of Sem�Jntia, I], .2.65-32.8. Corcoran, John (I973), 'Meanings of implication', Dililogos, 2.5, 59-76. Dretske, Fred (I972.), 'Contrastive state­ ments', Philosophical Review, 8 1 , 41137· Dretske, Fred (I975), 'The Content of Knowledge', in Bruce Freed, A. Marras, & P. Maynard (eds� Forms of Repre­ sentation, North-Holland, Amsterdam. 77-93 · Edgington, Dorothy (I995� 'On condi­ tionals', Mind, 104. .2.35-32.9. Fauconnier, Giles (I975), 'PolaritY and the scale principle', Chicago Linguistics Society, I I, I 88-f)9Fauconnier, Giles (I979), 'Implication reversal in a natural language', in Franz

Guenthner & S. J. Schmidt (eds), Formal SemiJntics and Pragmatics for Natural Languages, ReideL Dordrecht, 2.89-30.2.. von FinteL Kai (I994), 'Restrictions on quantifier domains', Ph.D. dissertation, Graduate Student Linguistics Asso­ ciation (GLSA), University of Massachu­ setts, Amherst. von FinteL Kai (I997), 'Bare plurals, bare conditionals, and only', Journal of Semantics, 14. I-56. von FinteL Kai (2.ooo), 'Counterfactuals in a dynatnic context', to appear in Michael Keystowicz (ed.), Ken Hale: A Life in Language, Cambridge, MA: MIT Press. Giannikidou, Anastasia (I 997), 'The landscape of polarity . items', Ph.D. dissertation, Rijksuniversiteit Groningen. Grice, Paul (1967), 'Logic and conversation', MS.

Grice, Paul (I989), Studies in the Way of Words, Harvard University Press, Cambridge, MA. Heim, Irene (I984 'A note on polarity sensitivity and downward entailingness', NELS, I4. 98-Icry. Heim, Irene (I992.), 'Presupposition projec­ tion and the semantics of attitude verbs', Journal ofSnruJntics, 9, I83-2.2.I. Herzberger, Hans (1979), 'Counterfactuals

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

REFERENCES

Kai

•.

'Any', Linguistics and Philosophy, 16, 353-422. Katz, Graham (199I), 'The downward entailingness of conditionals and adversatives', FLSM, 2, 217-43. Klima, Edward (I 964� 'Negation in English', in Jerry Fodor & J. J. Katz (eds), The Structure of LAnguage, Prentice-� Englewood Cliffs, 246-323. Kratzer, Angelika (198I), 'The notional category of modality', in H J. Eikmeyer & H Rieser (eds), Words, Worlds, and Contexts: New Approaches in Word Semantics, de Gmyter, Berlin, 3 8-'74Krifka, Manfred (1990� 'Polarity phenom­ ena and alternative semantics', in Martin Stokhof & L Torenvliet (eds), Proceedings of the Seventh Amsterdam Colloquium, I'TI.I, Amsterdam, 277-302. Kri£ka, Manfred (1991a), 'A compositional semantics for multiple focus construc­ tions', SALT t, 127-58. Kri£ka, Manfred (I 99 Ib), 'Some remarks on polarity items', in Dietmar Zaefferer (ed.), Semantic Universals and Universal Semantics, Foris, Berlin, 1 5o-89. Kri£ka, Manfred (1994), 'The seznantics and pragmatics of weak and strong polarity items in assertions', Semantics and Linguistic Theory (SAL1) 4, 195-219. Kri£ka, Manfred (1995), 'The seznantics and pragmatics of polarity items', Linguistic Analysis, 25, 209-57. Ladusaw, William (I979). 'Polarity sen­ sitivity as inherent scope relations', Ph.D. dissertatio� University of Texas at Austin. Ladusaw, William (198oa� 'Affective or, £active verbs, and negative polarity irems', Chicago Linguistics Society, 16. Lad�w. William (198ob), 'On the notion "affective" in the analysis of negative polarity items, Journal of Linguistic R.Gearch, I, 1-23. Lakof( Robin (1969� 'Some reasons why there can't be any some-any rule', Language, 45· I.cwis, David (1973), Counterfaauals, Blackwea Oxford.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

and consistency',Journal ofPhilosophy. 76, 83-8. Hoeksema, Jack (1983), 'Negative polarity and the comparative', Natural lAnguage and Linguistic Theory. I, 403-34Hoeksema, Jack (1986a), 'Monotonicity phenomena in natural language', Linguistic Analysis, 16, 25-40. Hoeksema, Jack (1986b), 'Monotonie en Superlatieven', in C. Hoppenbrouwers (ed.� Proeven van Taalwetenschaap, Tabu, Groningen, 38-49. Horn, Larry (I969), 'A presuppositional approach to only and even', Proceedings of the Chicago Linguistic Society {CLS), 5, 98-107. Horn, Larry (1979), 'Only, even, and con­ ventional implicature', MS {LSA Paper, Los Angeles). Horn, Larry (I996), 'Exclusive company: only and the dynamics of vertical inference', Journal of Semantics, I 3, I-40. Horn, Larry (I997), 'All John's children are as bald as the King of France: existential impon and the geometry of opposition', Proceedings ofthe Chicago Linguistic Society {CLS), 33· latridou, Sabine (1998), 'The grammatical ingredients of counterfactuality', MS, MIT. Israe� Michael ( 1996), 'Polarity sensitivity as lexical semantics', Linguistics and Philosophy. 19, 6, 61g-66. Jackson, Eric (1995), 'Weak and strong negative polarity items: licensing and intervention', Linguistic Analysis, 25, 1 8 1-208. Jackson, Frank (1979), 'On assertion and indicative conditionals', Philosophical Review, 88, 565-89. Jackson, Frank (1984), 'Two theories of indicative conditionals', Australasian Journal ofPhilosophy, 62, 67-76. Jackson, Frank (1987), Conditionals, Blackwea Oxford. Jackson, Frank ( 1990), 'Classifying condi­ tionals', Analysis, so, 2, 134-47. Kadmon, Nirit & Landman, Fred (1993),

von Fintel 147

I48 NPI Licensing, Strawson Entailment, and Context Dependence

Lewis, David {I 979), 'Scorekeeping in a

language game', Journal of Philosophical Logic, 8 , 3 3 9- S9· Lewis, David (I 98 I), 'Ordering semantics and premise semantics for counter­ factuals', Journal of Philosophic Logic. Io, 2 I 7-34Lewis, David {I996), 'Elusive knowledge', Australasian journal of Philosophy, 74. 549-79· Linebarger, Marcia (I 980), 'The grammar of negative polarity', Ph.D. dissertation. .

Linguists Have Always Wanted to Know about Logic but Were Ashamed to Ask,

Chicago University Press, Chicago, {second edn. much revised). Partee, Barbara {I 992), 'On the "scope of negation" and polarity sensitivity', in Eva Hajicova {ed.), Functional Description of

Language: Proceedings of the Conference, Prague, 24-27 November 1992. Faculty of

Mathematics and Physics, Chatles Uni­ versity, Prague. Pesetsky, David { I 991� 'Zero syntax, vol 2',

MS, MIT.

Pollock, John (I 976), Subjunctive Reasoning, Reidel, Dordrecht. Pullum, Geoffrey (1987), 'Implications of English extraposed irrealis clauses', ESCOL, 26o-7o. Rooth, Mats ( 1 98 5� 'Association with focus', Ph.D. dissertation. Univ�rsity of Massachusetts at Amherst.

Zwarts, Frans (1995� 'Nonveridical con­ texts', Linguistic Analysis, 2.5, 2.86-3 12.. Zwarts, Frans (I 997), 'Three types of polar­ ity', in Fritz Hamm & Erhard Hinrichs (eds), Plurality and Quantification, Kluwer, Dordrecht.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

MIT

Linebarger, Marcia {I987), 'Negative polarity and grammatical repre­ sentation', Linguistics and Philosophy, 10, 325-87. {I99I), 'Negative Linebarger, Marcia polarity as linguistic evidence', Chicago Linguistic Society, 27, 2, I 65-88. Lowe, E. J. {I990), 'Conditionals, context, and transitivity', Analysis, so, 2, So-7. Lowe, E. J. {I 99S). 'The truth about counterfactuals', Philosophical Quarterly, 45 . 4 I - 59· McCawley, James {I99J), Everything that

Rooth, Mats (I992), 'A theory of focus interpretation', Natural Language Seman­ tics, I, 7S- 1 16. Rooth, Mats (I996), 'Focus', in Shalom Lappin (ed.), The Handbook of Con­ temporary Semantic Theory, Blackwell, Oxford, 271-97· Stalnaker, Robert (I975), 'Indicative conditionals', Philosophia, 5, 269-86. Stalnaker, Robert (I984), Inquiry, MIT Press, Cambridge, MA. Steriade, Donca (198 I), 'Complement if, MS, MIT. Strawson. P. F. (I952), Introduction to Logical Theory, Methuen. London. Veltman, Frank (I986), 'Data semantics and the pragmatics of indicative condi­ tionals', in Elizabeth Traugott, Alice ter Meulen,Judy Reilly, & Charles Ferguson (eds), On Conditionals, Cambridge Uni­ versity Press, Cambridge, I47-68. Warmbrod, Ken (198 Ia), 'Counterfactuals and substitution of equivalent ante­ cedents', Journal of Philosophical Lagic, 10, 267-89. Warmbrod, Ken {I98Ib), 'An indexical theory of conditionals', Dialogue, Cana­ dian Philosophical Review, 20, 4. 644-64Warmbrod, Ken (I982), 'A defense of the limit assumption', Philosophical Studies, 42, 5 3-66. Warmbrod, Ken (I983), 'Epistemic condi­ tionals', Pacific Philosophical Quarterly, 249-65· Williams, Edwin (1974), 'Rule ordering in ' syntax , Ph.D. dissertation, MIT. van der Wouden, Ton (I994), 'Negative contexts', Ph.D. dissertation, University of Groningen. Wright, Crispin (I98 3� 'Keeping track of Nozick', Analysis, 43, 3 , I 34-40.

]011.mal ofSmuzlllia

16: 149-:104

© Oxford Univenity Press

1999

Fast 'Almost' and the Visibility Parameter for

Functional Adverbs IRENE RAPP a n d ARNIM VON STECHOW University of Tuebingen

Abstract

1 WHAT THIS PAPER I S AB O UT The proponents of strong lexicalism hold the view that 'words' are formed in the lexicon and are opaque for the syntax {Dowty 1979; DiSciullo &Williams 1 987;]ackendoff 1 990; and many others). Modular morphology, on the other hand, offers the possibility that at least some words are . partially formed in the syntax (Baker 1 988; Borer 1 988; Hale & Keyser 1 994; Chomsky 1 995; and many others). In Stechow (199 5 , 1 996} it has been argued that facts observed. with German wieder 'again' cannot be accounted for within the framework' of strong lexicalism and therefore favour the second position. The authors of this paper warit to check these claims by investigating the behaviour of the German adverb fast 'almost', the other classical adverb to which the method of decomposition has been applied in

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Since the years of generative semantics, it has been claimed that the adverbs again and access to different parts of verbal meanings. These adverbs - which we call D(ecomposition)-adverbs - are therefore classical witnesses for any decompositional approach to lexical structures. Stechow (1995, 1996) has shown that positional facts observed with German wieder 'again' can be explained by the assumption that words are partially formed in the syntax. The wieder-data give evidence for a syntactic decomposition of verbs into VOICE + BECOME + RESULT STATE. In this paper we investigate the behaviour of German fast 'almost'. We show that fast can intervene between VOICE and VP, but that it does not have access to the result state of telic verbs. In other words, fast cannot have narrow scope with respect to BECOME in a decomposition structure. To account for this difference between wieder and fast, we assume a Visibility Parameter classifying D-adverbs; only wieder but not fast has the capacity of seeing the result XP of a decomposition. However, there seems to be cross-linguistic and inner linguistic variation with respect to the Visibility Parameter. According to the literature, English almost is able to modify the result state of a telic verb. Furthermore, the configuration BECOME +fast + Result XP in German seems to be accepted by a number of speakers as welL We conclude that the Visibility Parameter is a lexical property of D-adverbs which varies across languages and dialects.

almost have

1 50 Vtsibility Parameter for Functional Adverbs

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

the linguistic literature (c£ McCawley 1971). Most adverbs cannot look into lexical decomposition structures, but some can. In order to have a suitable name, let us call these D-adverbs (= decomposition adverbs). Again and almost are the prototypical representatives of this class. The method adopted is to investigate whether fast has access to different aspects of verbal meaning depending on its position. Following Dowty (1979), resultative verbs and prepositions are decomposed into BECOME + XP, where XP is the result state. Agentive resultative verbs contain an additional AGENT part. Other thematic relations connecting the subject and the event are HOLDER and cause. If the syntax tells us that an adverb like wieder or fast must have wide scope with respect to BECOME, then we expect other readings than in a configuration where fast can have narrow scope with respect to BECOME. In other words, the motivation for decomposition is syntactic and semantic. The idea is that certain function projections in the classical VP have a meaning and hence scopally interact with adverbs. The wieder facts support this view of the syntax rather well and so we would expect a similar behaviour for fast. But quite surprisingly, we have discovered no convincing evidence thatfast could have a narrow scope with respect to BECOME in verbs for which decomposition has been assumed. In our dialect, the configuration BECOME + fast + XP is not possible. On the other hand, the wieder-facts strongly suggest that the configuration BECOME + wieder + XP exists. Thus, the two adverbs behave differently in (our) German. However, the facts reported about English suggest that in English, the . configuration BECOME + almost + XP is a possible one. Furthermore, other speakers accept BECOME + fast + XP readings for German.So there must be cross-linguistic and inner linguistic variation. The parameter accounting for the difference in distribution may be that fast can only be attached to a phrase with a morphologically visible head. The XPs in a decomposition structure are syntactically somehow degenerated. Most adverbs do not see this projection but some do, most notably wieder. For some speakers of German,fast. can also modify these XPs. Hence we have to classify these adverbs with respect to their capacity of 'seeing' the result XPs of a decomposition, and we speak of the Visibility Parameter for Functional Adverbs. The term 'functional' is used in (Ernst 1998). Functional adverbs form a small class among the adverbs. They have in common that they do not �elect a particular syntactic category. In principle, they could be adjoined to every projection, provided we can interpret them there. Apart from topicalization, they never move. The Visibility Parameter is a lexico-syntactic property. It varies across dialects and across languages.

Irene Rapp and Amim von Stechow

151

At this juncture, we have to admit that we have not mvestigated the variation among speakers systematically. The following (introspective) data are based on the idiolects of the two authors. The structure of the paper is as follows. In section 2 we list some of the examples which in the literature have motivated decomposition in the syntax. In particular, we will comment on the wieder facts of German. Section 3 gives an introduction into our logical forms (LFs). Section 4 introduces classical data from English and contrasting ones from German. Section 5 introduces the semantics forfast 'almost' which is assumed in this article. Section 6 comments on the so-called counterfactual reading offast. We assume that counterfactual fast attaches to the aspect phrase (AspP). Interestingly, as inner fast, counterfactual fast can only modify projections with a visible head. Section 7 systematically investigates the 'inner' reading of fast for verbs from all the different aspectual classes of the Vendler classification. This section argues that the configuration BECOME +fast + XP does not exist in our dialect. But there is a reading AGENT + fast + BECOME + XP, the so-called scalar one. According to our generalization, this can be explained if we assume that only the verbal head is morphologically visible forfast. Section 8 shows that there are no semantic reasons for the absence of resultativefast: if the result state is expressed by a visible lexical head,fast can modify it. Section 9 deals with problematic cases of negation. Fast is a positive polarity item and has wide scope with respect to negation (NEG), and the negation is outside VP. It follows that no inner reading of fast + NEG should be possible, contrary to the facts. We will argue thatfast + nicht can form a complex particle synonymous with kaum 'barely'. This adverb can occur at a scalar position. Section 10 is a remark on the structure of the lexicon, and section I I contains the conclusion. The conclusion points out that the theory presented here may be regarded as a semantically motivated version ofDistributed Morphology (c£ Marantz I 997). ·

DE COMPOSITION IN THE SYNTAX: WIEDER 'AGAIN'

We will frequendy compare the adverb fast with the adverb wieder 'again'. Hence it is convenient to repeat the data that motivated the decomposition approach developed in Stechow (1995, 1 996), which is assumed as basically sound for the purposes of this paper. (repetitive/restitutive) ( 1) weil Ali die Tiir wi�er offnete because Ali the door again opened (2) weil Ali wieder die Tiir offnete (only repetitive) because Ali again the door opened

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

2

I S2·

VISibility Parameter for Functional Adverbs

( I ) is ambiguous between a repetitive and a restitutive reading: Ali has done

the opening before vs. Ali brought it about that the door was open again, but he had not opened the door before. (2), on the other hand, is unambiguously repetitive. In other words, the position preceding the direct object has a disambiguating effect, which has to be explained. Here are the two meanings represented in an extensional typed language in style of Zimmermann ( I993a).

(3 )

repetitive reading: the doorw >.y againw(e) >.e[AGew{x) & BECOMEew(>.s.OPENsw(y) )] b. restitutive reading: Ali Ax the doorw >.y AGew(x) & BECOMEew(>.s.againw(s)(>.s.OPENsw(y))) a.

Ali Ax

(4)

a.

BECOME is of type (s, (i, (1r, t))), i the type of eventualities and/or times and 1r = (s, (i, t)). II BECOME ll(w)(e)(P) is only defined, if P(w) is not defined for any part of e. If defined: II BECOME ll(w)(e)(P) = I iff 3s1: S1 > <e & P(w)(s1) = o & 3s2: e > < S2 & P(w)(s2) = I, where > < means 'abuts from the left side'. b. AG{ENT) is of type (s, (i, (e, t))). ll AGENT ll(w)(e)(x) = I iff x is the agent of event/action e in w. c. again is of type (s, (i, (1r, t))). Let P be a set of eventualities and let w and e_ be a world and an eventuality, respectively. II again ll(w)(e)(P) is de.fined only if 3e'[ IIMAXII(P)(w)(e') = I & e' < e]. Where defined, II again ll(w)(e)(P) = I iff P(w)(e) = I.

BECOME is more or less defined as in Dowty ( 1979). The difference is that we work with a partial interval semantics adapted for events. The AG{ENT) role corresponds to Dowty's DO, and · the MAX-operator assumed in the definiens of again means that the event argument is a maximal P-event (c£ Stecho'! I996). The explanation of the wieder facts given in Stechow ( 1995, I996) can be read from the following tree. The accomplishment ojfnete 'opened' is decomposed into AGENT + BECOME + OPEN (with the head at the right hand). The restitutive position is under BECOME:

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The notation omits many brackets and writes the argument in front of its functor in order to be closer to the surface (c£ Cresswell I 97 3 ). The relevant meaning rules presupposed in the formulas are these:

Irene Rapp and Arnim von Stechow I s 3

(5)

AgrOP

A.e.AgrOP

the doorw die Tiir

Ay.AgrO'

againew

wieder

AgrO

i..e. VoiceP

(repetitive)

NP� r--X

M.Voice'

ISUJ

Voi�

& AGew(x)

�

N:>..w.XP

� l wieder >..s.XP (restitutive) �X PENsw JBJ
V B ECOMEew

agai;;ew

The repettttve pos1t10n of the adverb is above Voice-P, whose head expresses agency, and the repetitive position is under BECOME. The accusative case is assigned in the checking domain of AgrO. 1 If the 1 We are no longer so sure that the AgrO position is correct. Indefinite objectS may ocasionally appear under wittkr and the sentenc� have a restitutive reading nevertheless. G. Jager gave as the following example: {i) Fritz oflnete dann wieder ein.Fenster. Fritz opened then again a wioaow This suggests that the AgrO position is under the subject of-Voice. In any cast, non-contrastive definite NPs are scrambled out of the Voice-P. Thus, for the examples with definite objects, we may consider AgrO as a scrambling position. For convenience, we leave the structure. lu far as we can see, no distortion of the facts discussed arises from this assumption.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

r-------__

VoiceP

I 54

VISibility Parameter for Functional Adverbs

2 Manfred Bierwisch objects that the theory cannot explain the repetitive/restitutive ambiguities in examples such as (i) weil die Griinen wieder im Parlament sind 'because the Greens are in Parliament again' The ambiguity is discussed in Stechow (I996, fn. 2.8), although no final answer is given. There, the ambiguity was resolved by contextualization. We can paraphrase the repetitive reading as 'Now we have another period with the Greens in Parliament'. This is a repetition and does not imply that the Greens were not in Parliament between the two legislation periods. The restitutive reading comes out automatically in view of the MAX-operator contained in the semantics of again. Now, it is exactly the MAX-operator that causes the break between two successions of a state. There is a very trivial solution to the objection. Assume that the MAX-Operator is not present in stressed again, i.e. we have a slight polysemy. Represent the meaning of stressed again as AGAIN. Then the ambiguity of the sentence is represented as: restitutive (ii) againw(pres)(in parliament(the Greens)) AGAINw(pres)(in parliament(the Greens)) repennve For countabie, i.e. telic, events, we could always work with the simpler operator AGAIN. Bierwisch makes a similar point with the sentence (iii) weil der Wind das Rad wieder drehte 'because the wind turned (= made rotate) the wheel again' The objection is that there is no operator in the decomposition with which again could scopally interact. It is true that causation-of-motion verbs are not represented in our systetn. A tentative analysis is this: (iv) e(VP the wind (v- causerw ..\w(XP the wheel rotateew]]) cause is a relation of type (si,((s, t), (e, t))) with the following semantics: (v) II cause ll(w)(t)(p)(x) = I iff there is a property P: CAUSE,w(p)(P(x)) = I CAUSE is Dowty's operator, which roughly expresses causal dependency. The details are rather complicated and are given in D� (I979� section l.J.7· It is an empirical question, whether the relation cause should be labelled as Voice. Those who can passivize the example classify the operator as Voice. Note that the formula contains a free time/state variable. Therefore, it can be hooked to time by tense. The repetitive/restitutive ambiguity now has the following analysis: repetitive (vi) again...(past)..\t3 e(VP the wind (v• causerw Aw(xP the wheel rotatecw)]] restitutive 3 e(VP the wind (V' caaseposr Aw(againw(e)..\e(xP the wheel rotatecw)]]) w

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

adverb precedes the direct object, it must be located higher than AGENT. Hence, we only have the repetitive reading. This is the disambiguation effect induced by the position. If the adverb follows the object, we do not know whether it is above or below AGENT. Hence, we can have more than one reading. Aside from the two positions indicated in the tree, the adverb may have scope between AGENT and BECOME. It is not clear whether this reading is ever realized. For a discussion, see Paslawska (1998). In the literature, it has been proposed to treat the inner readings of adverbs by means of meaning postulates. Dowty (1979), Fabricius-Hansen {1983), Stechow (1995, 1996), and Zimmermann {1993b) have disputed the soundness of the relevant postulates. They have argued that the method is not viable on principled grounds. Even, if meaning postulates could do the job, it is not all clear, how the positional effect observed with wieder could be explained by postulates. Hence, the wieder facts strongly support a decompositional approach to the VP.2

Irene Rapp and Arnim von Stechow I s S

3

C O MMENTS O N O U R L F s

(6)

Stechow

(1996)

Hale & Keyser

(1994=

72

)

VoiceP

�

VP

" ·

vOICe•

NP 'SUJ

r----_ _

VP

�v

�X

10BJ

O PEN

iifTnete

BECOME

V01ce AG

NP�

Ali

----'..

V

VP

NP ....... v ,...-. .. the door y AP A open

Apart from the position of the object, for which we assume an AP internal position, there is a one-to-one match between the trees: We have to assume

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

We assume a framework of transparent LFs (= TLF; c£ Beck I 996; Stechow 1996). The idea is that the variables and the 'logical glue' are written at the positions where they belong. The interpretation can be read in a one-to-one way from the tree. The architectUre of the grammar is, details aside, the same as in the Minimalist Program. Like Chomsky (1995), we will write the inflected form at one particular node, and we will assume that it is lexically inserted there. One important feature of the theory is that parts of the meaning of the verb may be located elsewhere in the tree, viz. in morphologically invisible heads. For instance, the finite form offnete .'opened' is associated with the stative property OPEN. The BECOME part and the AGENT part are located at different functional heads whose presence has to be guaranteed by an appropriate checking mechanism (c£ section ro). Other methods of lexical insertion are compatible with this approach. For instance, one could bring the heads together by means of head movement (c£ Baker 1988) and would then replace the complex head by a morphological form (c£ Stechow 1996). Case is checked at appropriate positions. For convenience we are assuming the functional projections AgrO and TP, but we do not commit ourselves to the exact categorial nature of these projections. Note that TP is identical to AgrS and AgrO could be an adjunction site. The only thing that matters is that case is not assigned VP-intemally-with qualifications to be discussed below. Relating our structures to other proposals found in the literature is a rather straightforward matter. As an illustration we compare our Voice-P with the VP deep structure which Hale & Keyser (1994) assume for transitive resultative verbs.

I s 6 VISibility Parameter for Functional Adverbs

the meaning BECOME for the lower empty V and the meaning AGENT for the higher empty V. This map seems to be in accordance with the semantics assumed by Hale and Keyser for these verbs: (7) Hale & Keyser's (1994) semantics (c£ p. 68 f£) The subject of the higher VP (Ali) is the agent of an event e1, which causes a change e.z, which implicates a state s, of which the lower NP e.z) & the door > (e.z s) (the door) is the theme: Ali > (e1 -+

-+

(8) A 1LF for Hale

&

Keyser (1994)

A.y.V'(e)

�

V

B ECOM Eew

l.. w l.. s . XP

open5w(Y)

A final remark on the information AGENT: it is located in the head of a functional projection which Hale & Keyser call V and which we, following Kratzer (1994), call Voice. Kratzer argues that the subject of Voice-P is not an argument of the verb. Rather, she claims, a separate functional head introduces it. This is thought as an explanation of Marantz's (1984) claim �t the subject is not theta marked by the verb. We follow Kratzer in this respect, but we add the qualification that some verbs have their subject as an argument (see section 7·3)· Chomsky (1995) explicitly builds on Hale & Keyser (1994). Therefore, our findings should be also valid for that approach.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The paraphrase rather suggests a 'classical' decomposition, which makes use of the predicate CAUSE in the style of McCawley (1971) or Dowty (1979), where we find something like 'CAUSE BECOME VP'. In our event approach, however, the predicate CAUSE seems unnecessary because we have the predicate AGENT of an action, and the decomposition 'x is an AGENT of action e, which is a BECOMing with the resultant state that y is OPEN' seems intuitively correct. Hence, if we fill the empty V nodes of Hale & Keyser in the way indicated, we obtain a precise semantics for their tree, which can be read off from the following structure:

Irene Rapp and Amim von Stechow 1 s 7 4

FAST: FIRST DATA

(9) John almost killed Harry a. almost(ACT(John) CAUSE BECOME dead(Harry)) counterfactual: b. ACT(John) almost(CAUSE BECOME dead(Harry)) scalar: sc c. ACT(John) CAUSE BECOME almost(dead(Harry)) resultative: res

cf

According to McCawley, the first reading is that John almost did something that would have had the effect of Harry's dying. In this case almost has wide scope with respect to the information expressed by the verb. We call this reading the outer or counterfactual reading of almost. By contrast, both the second and the third reading are inner readings: almost only modifies a part of the verb meaning. The scalar reading means that John did something which almost had the effect of Harry's dying. Under the resultative reading John did something which had the effect of Harry's becoming .almost not alive. Our intuition tells us that there is a clear difference between the outer and the inner readings: In contrast to the inner readings the outer reading means that the described event does not take place at all The situation is less clear with the two inner readings. They seem to describe rather similar events: In the case of John almost killed Harry both inner readings mean that the act of killing was somehow initiated but not completed. Let us now have a look at an example in German: (xo)

a.

weil David fast seinen Hasen erwiirgt h.atte OKcf *sc *res because David almost his rabbit strangled had (Subjunctive II)

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

One should recall that our decomposition structure for the Voice-P provides three positions for adverbs which give rise to different meanings: the position above AGENT, the one between AGENT and BECOME, and the position under BECOME. The intermediate position does not seem to be important for wieder, but the position under BECOME is crucial for explaining the restitutive reading. One would expect that the three positions play an equally important role for explaining the behaviour offast 'almost'. This is assumed for English in Morgan (I 969) and McCawley (I 97 I) and for German in Bierwisch (I997). Our own research, however, suggests that German behaves differently. In particular,Jast does not seem to figure under BECOME. We start the discussion of the data by reminding the . reader of the classical account given in McCawley (I 97 I).

IS8 VISibility Parameter for Functional Adverbs

b. weil David seinen Hasen fast erwiirgt h.atte3 OKcf sc *res because David his rabbit almost strangled had Subjunctive IT) c. ??weil David fast seinen Hasen erwiirgte ? cf *sc *res because David almost his rabbit strangled d. weil David seinen Hasen fast erwiirgte *cf OKsc/res because David almost his rabbit strangled

�

First, we observe that there is no inner reading

r. We can explain the positional effect by assummg that the event argument is introduced at the aspect node and that AgrOP dominates AspP. Under these assumptions, any position above AgrO leads to a counterfactual reading. The connection between the outer position offast and the subjunctive can be captured by the generalization that fast only attaches to projections that have a lexical head, i.e. N, A, V, P (c£ section 6). We have to warn the reader at this point that the notion 'lexical head' depends on the theory that we will assume here, namely a son of Distributed Morphology in The scalar reading arises when the Subjunctive D marks Er sagte, class er seirien Hasen fast erwiirgt hitte. He said that he his hare almost strangulated had We owe this observation to Manfred Bierwisch. We owe this point to Satoshi Tomioka. 3

(i)

•

indirect

discourse:

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

iffast precedes the direct object. Second, in German the counterfactual reading requires the sub­ junctive. Most speakers of German do not accept examples like (roc), where the position offast indicates a counterfactual reading and the indicative is used. Hence the indicative seems to be restricted to the inner readings. On the other hand, the subjunctive examples never have an inner reading: Independently of the position offast, (roa) and (rob) only have a counter­ factual meaning. Finally, in German, as in English, it is not really clear whether there is a scalar and a resultative reading, or just one inner reading. The positional effect helps us only to distinguish the outer reading from the inner ones, but it does not differentiate between the two inner readings.4 We resume the discussion of this section in the following way. Like wieder,Jast exhibits a positional effect that has to be explained. Whereas it is quite clear what the outer reading means, namely that the event almost occurs, it is not so clear what the inner reading(s) are. Do we have two inner readings, the scalar one as opposed to the resultative one as McCawley assutnes, or do we have only one inner reading? Furthermore, is there any connection between the outer reading and the subjunctive? These are the questions that we will investigate next. Our answers are the following:

Irene Rapp and Amim von Stechow

I S9

the sense of Halle & Marantz {I993)· For instance, BECOME-verbs will have the structUre [v BECOME XP], and the lexical head will be the phonetically invisible 'light verb' BECOME, and not the lexical root XP. 2. There is only one inner reading forfast It is always scalar. Unlike wieder, fast does not occur in the resultative position, ie. fast always has wide scope with respect to BECOME, as far as verbal decomposition is concerned (c£ section 7). However, in order to discuss the data properly, we need a precise semantics offast which will be given in section S·

FAST

The meaning rule to which we adhere closely follows Sadock (I98 I), with a modification to be discussed in a moment. ( 1 1 ) Core meaning: fast is of type (s, ((s, t), t)). Let w be any world: F(fast)(w)(p) = I iff (a) and (b) hold. a. There is a world w' which is almost not different &om w and p(w') = I. b. p(w) = o. This meaning rule contains a negation and the prediction therefore is that the adverb scopally interacts with quantifiers. The following pair of sentences illustrates the scope interaction:

( 1 2)

a.

Fast aile Pflanzen waren vertrocknet. Almost all plants were dry b. Aile Pflanzen waren fast vertrocknet. All plants were almost ·dry

If the meaning rule is correct, the first sentence entails that not all plants were dry and the second sentence entails that no plant was completely dry. These truth conditions are incompatible. Now, the negative entailment has been disputed in the literature. Sadock ( 198 I) claims that only condition (a) of ( I I) belongs to the content offast 'almost', whereas condition (b) is an implicature. An example given to support the claim is the following one:

{I3)

Not only did Bill almost swim the English C�el; swim it.

in

fact he did

' Sadock obviously presupposes that 'not only cp; in fact '1/J' implies cp and '1/J'. In this construction, '1/J is a reinforcement of cp. Sadock admits that sentence (I 3)

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

s THE C ORE MEAN I N G O F

1 6o Visibility Parameter for Functional Adverbs is odd, explaining this through the strong conversational implicature carried ' ' by 'alniost cp , viz. 'not cp , which is almost not cancellable. We are not sure how much the not only . . . in fact . . . construction says about the compatibility of the two argument sentences. If we consider the 'nicht nur . . . sondem . . . ' frame, ie. the German equivalent of'not only . . . but . . . ', we certainly can have incompatible sentences as arguments: {I4)

a.

b.

Sie ist nicht nur krank, sondem sie ist tot. She is not only sick, but she is dead #Sie ist krank und sie ist tot. She is sick and she is dead

is higher on a scale along some appropriate dimension. 'Sick' and 'dead' are scalar in some sense, but the two properties are incompatible. Thus, Sadock's examples (and others not discussed here) are not really counter­ examples to our meaning rule. Atlas {I 984) gives another argument intended to demonstrate that 'almost' cannot imply a negation. If this were so, he claims, then the sentence Almost all swans are almost white would be synonymous with Some swans are white. Here is his proof { I s) Atlas' proof Almost all swans are almost white => Almost all swans are not white => Not all swans are not white => Some swans are white It seems to us that this argument is fallacious inasmuch as it illicitly uses the principle: 'p entails q; therefore -i p entails -,q'. This is best seen by instantiating 'all swans' by the two swans a and b. Then the first line of the argument becomes equivalent with the Hrst line of the following argument, which reconstructs Atlas' proof in a somewhat different order: ( I 6) Atlas' proof reconstructed Almost [almost (a is white) & almost (b is white)) => [-. almost {a is white) V -.almost (b is white)] {semantics of almost plus De Morgan's rule) {fallacy!) => [-. -. a is white V ..., ..., b is white) (double negation) ¢:> [ a is white V b is �hite) (semantics of some) => Some swans are white Our meaning rule (I I) treats fast as a modal operator with a counter­ factual touch. which is due to the inherent negation. Iffast modilles degrees,

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Nicht nur . . . sondem auch 'not only . . . but' requires that the second conjunct

Irene

R.app and Amim von Stechow r6r

we have the impression that the meaning is purely extensional expressing a distance between two degrees: 'degree d is very near to d' but it is not identical to d'.' (r 7) Manfred is almost 6 feet tall 'The degree of Manfred's tallness is very near to 6 feet but it is not equal to 6 feet'

(18) a. Ich wiege jetzt fast 75 Kilo. I weigh now almost 7 5 kilos b. Leider wiege ich noch nicht 75 Kilo. Unfortunately weigh I not yet 75 kilos In other words, the context has to tell us whether 'nicht 75 Kilo' means 'not at most 75 kg' or 'not a least 75 kg', i.e. the context has to give us information about the direction from which we approach the target degree, from 'below'-the unmarked case-or from 'above'. To be concrete, the representations of McCawley's examples in our logical language would be the following formulae: (19) John almost killed Harry. a. almost(w)(>.w3 e[AGENTew(John) & BECOMEew >.w>.sDEADsw(Harry))) b. 3 e[AGENTew{Jdhn) & almost(w)(>.w.BECOMEew >.w>.sDEADsw(Harry))] c. 3e[AGENTew(John) & BECOMEew >.w>.s.almost(w)(>.wDEADsw(Harry))]

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

It seems to us that this reading can be rendered by means of the modal phrase 'Our world is very similar to a world where Manfred is 6 feet tall, though he is not 6 feet tall'. Thus, we see no need to replace the modal meaning by extensional meanings. If there were arguments in favour of some fast meanings being extensional, this would not change much, as far as we can see. As usual, the notion of similarity involves much vagueness. As Sadock and Atlas correctly notice, global metaphysical similarity will not always yield the correct result. In many cases, phenomenological similarity is the correct notion ('France is almost hexagonal'). Another point should be mentioned. In the literature, we often find the claim that the negation of a degree implies the negation of all higher degrees. Thus, if we negate sentence (17), we thereby claim that Manfred has no degree of tallness which is greater than 6 feet. The following example (due to Kaspar 1987) suggests that this inference is pragmatic. If I, weighing 76 kilos and trying to loose weight, made the statement .(1 8), I would speak the truth.

I 62.

VISibility Parameter for Functional Adverbs

We will explain the details of the notation in the course of the following discussion. 6 THE P OSITIONAL EFFECT AND THE PRETERITE/SUBJUNCTIVE CONTRAST

In section 4 we noticed thatfast shows a positional effect similar to the one

TP

�

nominative

T

r-----__ � Agn)P

T

Agn)'

accusative

�

AspP

�

VoiceP � Voice' NP VP

Agn)

Asp

Voice

AgrO and TP are higher than the aspect node. Hence, iffast precedes the direct object and/or the subject, it has the aspect node in its scope. We get the counterfactual reading that requires the subjunctive:

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

observed with wieder. For both adverbs, the position preceding the direct object has a disambiguating effect It yields a repetitive reading for wieder and a counterfactual reading for fast. For wieder, this positional effect was explained in the following way: As AgrOP is above Voice-P, an adverb preceding the direct object must be located higher than AGENT. Hence, we only get the repetitive reading of wieder. However, this explanation is not valid forfast. Iffast precedes the direct object, we do not get a seal� reading that includes the AGENT-relation. Instead we rather obtain a counter­ factual reading, which denies the event proper. This effect can be captured in the following way. We assume that the crucial functional node responsible for the counterfactual/scalar distinction is the aspect node, which introduces the event argument and relates it to the reference time. Wide scope of fast with respect to that node means that no event of VP-type occurs. Now, the positional effect is explained if we assume the following hierarchy of the projections above Voice:

Irene Rapp and Amim von Stechow 163

(2r)

*weil David fast seinen Hasen erwiirgte because David almost his rabbit strangled b. *weil fast David seinen Hasen erwiirgte · because almost David his rabbit strangled c. *weil fast David einschlief because almost David in slept a.

·

Note that this positional effect is quite independent of the semantic properties of the verb. It can be observed not only with telic verbs as those in (2I), but also with atelic verbs (activities and states): *weil die Manner fast das Lied grolten/hassten fKgegrolt/gehasst h.atten) because the men almost the song bellowed/hated (Subj. II) b. *weil fast die Manner das Lied grolten/hassten fKgegrolt/gehasst h.atten) because almost the men the song bellowed/hated (Subj. II) c. *weil fast Lisa rannte because almost Lisa ran a.

We conclude that in contrast to wieder the positional effect existing withfast does not tell us anything about the decomposition of Voice-P. However, it gives us some important insights into the hierarchical structure of the projections dominating Voice-P. Next, we have to say something about the fact that the 'counterfactual' reading offast always requires the subjunctive II in German. It never goes together with the preterit, in other words, for (23b) only a scalar interpretation would be possible which is however excluded by the verb trejfen 'meet': (23)

Gestem hatte sie mich fast getroffen. Yesterday had she me almost met b. *Gescem craf sie mich fast. c. ?Gescem hat sie mich fast getroffen. a.

(Subjunctive II) (Preterit) (Perfect Indicative)

(2 3b) should not be confused with (23c), a perfect indicative construction, which is (more or less) grunmatically correct. We assume that in (23c) hat is simply taken as a substitute for hiitte. The construction is used synony­ mously with the subjunctive II. On the other hand, (23b) is definitely odd. In order to explain the contrast, we will invoke the following stipulation:

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(22)

164 Visibility Parameter for Functional Adverbs

(24) The fast-generalization: Fast cannot attach to a non-lexical functional projection.

(25)

a. (weil) sie mich gestem traf b. past1 -Xr[Y(r) & 3e[e � r & meetwe(me)(she)]] = Y{past1} & 3 e[e � past1 & meetwe(me)(she)]

We assume that the simple tenses past and present have a deictic interpretation, ie. r, 'the reference time', is a free variable of type i (the type of times/events} whose value is determined by a contextually given assignment (c£ Partee 1 973; Heim 1 994; Kratzer 1998). In this paper we shall not distinguish between times and states. Furthermore, times and events will have the same logical type.

(26) Simple tenses: past; and pres; are symbols of type i II pasti II� = g(pas�). if g(pas�) is before TUc, the time of the utterance at c. Otherwise, g{pastJ is undefined II presi II� = g(presi), if g(presJ is not before TUc, the time of the utterance at c. Otherwise, g(presi) is undefmed. Y(r) means that r is in yesterday. We leave the exact formulation of the meaning rule to the reader. The information '3e[e s; r & . . . ' is an 'aspect' in the sense of Klein (1 994): Aspects relate events (in Klein's terminology, 'times of the situation') to the reference time. 'e � r' is to be read as 'the time of e is included in r'. This aspect is called Perfective, whereas the converse relation is the Imperfective. If we are given a simple preterit

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

It will turn out that the functional projections are TP, AspP and Voice-P. AgrS is not investigated It would certainly qualify as a functional projection as well. In section 7 we will show that the fast­ generalization is not only valid for counterfactual fast. It can also be used to describe a number of phenomena occurring with inner fast. As a matter of fact, the fast-generalization seems to be the outcome of the setting of a parameter which is responsible for the inner linguistic and cross-linguistic variation observed with fast 'almost'. Whereas (24) seems to hold for many speakers of German (including the two authors), it is not at all valid for English. It would, however, be a welcome result if all our restrictions on fast could be explained by one principle. To explain the sense of the principle for counterfactualfast, consider our treatment of sentences in the preterite:

Irene Rapp and Amim von Stechow z6s

sentence, the aspect is introduced by a phonetically invisible sem.:intic operation. For convenience, we will assume the following LF for example (25)· TP

� r----� � Ar.T'

paslj

�pP

T

Y(r gestem

Asp PV(r)

The LF neglects the details of surface syntax, e.g. movement of the finite verb to Comp and movement of the subject and the object to case positions. It is important to note that simple tenses are indexicals and that the AspP must therefore be a property of times. The LF is very similar to the ones adopted in Kratzer (1998). The main difference is that Kratzer assumes that tenses are deictic terms without referential indices. We treat tenses as sorted variables. It follows that our account is not really compatible with the standard assumption that simple tenses denote relations between the utterance time and the reference time (cf. Klein 1994). Tenses are like pronouns. Hence we will never have a scope interaction between simple tenses and quantifiers or negation, a welcome prediction. The meaning of the Perfective operator is this: (28) The Peifective PV is a symbol of type (i, ((i, t), t)). II PV ll(t)(p) = I iff 3 e: e s; t & p(e) = 1. It should be obvious by now that the tree (27) determines the formula (25b) in a straightforward way. Since the head of Asp is phonetically empty, our stipulation (24) prevents fast from being attached to AspP. Hence the ungrammaticality of (23b). Note that there is no semantic reason why (23b) should be ungrammatical. If the restriction (24) did not hold, the sentence would express the following statement: (29) almostw Aw (Y(past1) & 3e[e s; past1 & meetwe(me)(she)]) This

is precisely the meaning of the English sentence

(3o) Yesterday she almost met me.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

).e.VP meetew(me)(she) sie mich traf

166 VISibility Parameter for functional Adverbs

Given the present framework, we would have to say that English allows for an attachment of almost to an AspP with empty head. Or we could say that the perfective aspect is somehow contained in the meaning of verbs in the simple past, though it is not so clear how this idea could be implemented formally. Returning to the perfect examples in (23), it is now obvious what our account of their grammaticality is. The auxiliary haben/sein are lexical expressions of the Perfect aspect, which localizes an event in the past with respect to the reference time. This time,fast can attach to the AspP. In other �ords, (23a) and (23c) have the following TIF:

�r r------T �

ru m fast

Aw.AuxP

�

>.. e .VoiceP

� VoiceP

Y(e)& gestem

ux PERF (r) hat!hatte

meet ew<me)(she) sie mich getroffen

The tree e�presses the statement (32) almost(w)(.Xw3e(e < presi & Y(e) & meetwe(me)(she)]).

In plain English the formula means that the actual world is such that there

almost is an event e before the time of utterance, where e is a meeting with her as subject and me as object. The meaning of the Perfect operator is this: (3 3)

The Petfoct

PERF is a symbol of type (i, ((i, t), t)). II PERF ll(r)(p) = I iff 3 e: t(e) < r & p(e)

=

1.

To prevent confusion, several comments on the structure (3 I} are in order. The first remark concerns the location of the PERF-operator. Perhaps, the place of this information is the participial morphology (c£ Musan I998). In that case, the auxiliary would be semantically empty. If one took that route, one would have to make sure that the participle morphology is 'above'

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

TP

Irene Rapp and Amim von Stechow I 67

Voice-P. Then we would have more or less the same system as the present one. For our purposes, that particular choice has no great import. Second, note that the semantic tense in the structUre is pres1• This seems at odds with the fact that the subjunctive auxiliary hiitte is a subjunctive preterit. One would therefore expect a past tense for the sentence. The expectation, however, is deceptive. It has been observed in Abusch (1993) and Stechow {I995) that subjunctive IT forms of auxiliaries very often require a semantic present. For instance, miis.ste 'ought' is such a form: (34) *Als Jochen jung war, miisste er mehr arbeiten. When Jochen was young, he ought to work harder (35) Heute/morgen hitte meine Mutter ihren 98. Geburtstag gefeiert. Today/tomorrow had my mother her 98th birthday celebrated. The next comment concerns the different positions of gestem 'yesterday' in the two trees (27) and {3 1), respectively. Because of its meaning, yesterday can only localise a time in the past. In (27), the reference time is a past time. Therefore, yesterday can qualify that time. In (3 I), the reference time is not before TU. If yesterday would speak about this reference time, a contra­ diction would arise. A prior event e is introduced by the PERF-operator. Hence, yesterday has to be predicated of e and the adverb must be located in the AspP. This seems to lead to an incompatibility with surface syntax: fast ..1lmost" attaches to the left AspP and gestern 'yesterday' is inside of AspP. Therefore, fast should precede gestern at the surface if we make the standard assumption that adverbs do not move at LF. We observe, however, the opposite: At the surface, gestern precedes fast. The answer to this puzzle is that gestern is a definite term, and like other definite terms, it has to scramble out of the VP at surface syntax for reasons that are not yet fully understood but which have to do with the information structure of the sentence (c£ Diesing I992). A more accurate analysis of the adverb gestern 'yesterday' is something like [Pf IN yesterday), where IN is an abstract preposition expressing the locating function of the name. Ifgestern 'yesterday' is a deictic term of type i, it does not scopally interact with the PERF operator, and we could represent the meaning of (32) equivalently as: (36) Y At[almost(w) Aw3e[e < pres1 & IN(t)(e) & meetwe(me)(she)JJ This time Y is semantically reconstructed to the position t by A-conversion. Clearly, this formula is compatible with the surface syntax observed. Note that the approach is compatible with the principle that adverbs do not move at LF: We do not move the temporal adverb INM(e), but we move the object of this PP, i.e. the name Yesterday.. The statement IN(t)(e) remains in

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The form hiitte/wiire certainly can have the semantic present:

168 Visibility Parameter for Functional Adverbs

situ. One would have. to motivate the scrambling approach by further facts, but this is not the task of this paper. The description given so far does not yet account for the somewhat marginal status of (23c). Perhaps, the generalization is that an entirely counterfactual statement like 'there could have been an event . . . ' should be marked by the subjunctive. At present, we do not have an elegant way of describing this.

In the last section we noticed that the preterite always yields an inner reading. In contrast to the outer reading, inner fast does not deny the existence of an event but modifies part of the verb meaning. As a consequence, the adverb cannot be above a projection equal to or higher than AspP. Our question concerns the precise location of 'inner' fast. In the case of telic verbs there are three possible positions under AspP: (37)

VoiceP

_____-\

Vo ice'

sci-fast

� �;�

sc2-fast

VP

,...--_ v

)!:._

res-fast XP

BECOME

The position under BECOME would yield a resultative reading, the two positions above BECOME scalar readings (scr and scz). Whereas scr fast modifies the entire meaning of the verb, two different semantic decom­ positions are necessary for sczfast and resfast. sczfast requires the decomposition of the classical VP into Voice and VP, resfast requires the decomposition of VP into V and the resultative XP. Hence, to justify the decomposition in (37) we have to show that it is really indispensable to assume both the sc2- and the res-position for fast. The problem with verbs like erwiirgen 'strangle' is that it is hard to distinguish the three readings intuitively. Let us have another look at example (rod), repeated here as (3 8):

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

7 FAS T WITH PRETERITE: THE INNER READING

Irene Rapp and Amim von Stechow 169

(3 8) weil David seinen Hasen fast erwiirgte because David his rabbit almost strangled

*cf OKsc/res

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The resultative reading means that the rabbit reaches the state of almost being dead. The two scalar readings mean that there is an event that almost causes a result state of the rabbit's being dead. The difference between them is that sci includes the AGENT-relation and sc2 excludes it Let us first consider the difference between the two scalar readings. In our opinion, it is intuitively convincing that the VOICE-Relation is never modified by inner fast. For the sake of simplicity, we will show this with atelic VOICE-verbs. As they do not possess a result state, there is no risk of confusing the scalar readings with a resultative one. We will argue that scalar fast intervenes between VOICE and VP, hence defend a decomposition along the lines of Kratzer (1994). Second, we have to think about the need of a resultative position forfast. As we said, verbs like erwiirgen 'strangle' make it difficult to distinguish between the scalar reading(s) and the resultative one. This difficulty is due to the punctuality of their change of state: if David performs an action such that the rabbit is almost dying that means that the rabbit is almost dead. Hence, our intuition does not tell us where fast is located in the case of causative achievements. It can however be proved that fast is not a result state modifier by considering other telic verbs like the non-causative achievement einschlafen 'sleep in' and the gradual change of state verb essen 'eat'. These verbs will show thatfast has to be above BECOME. However, one must keep in mind that there are no semantic reasons against a result modification. As soon as the change of state predicate is split into copula + AP, fast has access to the result state represented by the AP. Our procedure will be to examine different semantic verb classes and to show which .arguments they provide against resjast/scr fast and in favour of sc2jast. In section 7.1 we examine state- . and activity-verbs. We will argue against the scr -position: fast always intervenes between Voice and VP. One the other hand, the behaviour of the adjectival one-place states will show that resjast is possible if the result state is represented by an visible lexical category. In section 7.2 we look at Subject Result-verbs like einschlafen ('fall asleep'). We will argue against the res-position of fast: Facts from temporal modification-the compatibility with adverbs of duration-suggest that fast is in front of the endre VP. In section 7·3 we examine Subject Object Result-verbs like erreichen ('reach') and verlieren ('lose'). As before, temporal modification suggests that fast modifies the entire VP. Interestingly enough, we do not have a

170 Visibility Parameter for Functional Adverbs

7.1 .

Verbs without result state (activities and statives)

There are two groups of verbs without BECOME in their decomposition: activities and statives (Vendler I 9 5 7; Dowty I 979; and many others). Innerfast means that the state/activity takes place to a certain degree: (39) Statives a. weil Max seine Lehrerin fast hasste because Max his teacher almost hated b. weil sie fast krank war because she almost sick was (40) Activities a. weil sie den Schlager fast grolten because they the song almost bellowed b. weil das Pferd fast galoppierte because the horse almost galloped (39a) means that there is a state of almost hating, and (4oa) means that there is an activity of almost bellowing. Again we ask what the exact position of inner fast is. Before we can address this issue, we have to think about the structural representation of verbs expressing states or activities. It is evident that they do not have a result phrase, but it is less clear whether they possess a Voice-P. According to Kratzer (1994), we assume that Voice contains the relations that connect the event with the subject, i.e. AGENT or HOLDER Furthermore, Voice checks· the accusative, which is realised in SpecAgrO. As a consequence, verbs with Voice are never unaccusative (Stechow I 996 p. 1 o s f£). If we assume that the AGENT-subject of activities is introduced by Voice, we expect them to be (at least potential) accusative markers. This

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

positional effect here. This can be explained if two-pla�e states embed Voice-P and AgrOP under BECOME (c£ Stechow 1996). In section 74 we discuss incremental theme verbs. As their change of state is not punctual, they allow us to distinguish clearly between the resultative and the scalar reading. Here, inner fast never has a resultative reading. Only the position above BECOME seems to be possible. Section 7·5 treats resultative verbs of motion. Here, innerfast modifies the direction or the manner of motion. Again, result modification is excluded. Our conclusion will be the following. Inner fast only appears directly above VP. The resultative position and the position above Voice-P are both excluded. However, the copula + adjective constructions which allow for resjast show that there are no semantic reasons for the lack of result modification. Hence, this possibility must be excluded structurally.

Irene Rapp and Amim von Stechow 171

7.1.1

One-place statives

We represent agentless one-place states expressed by adjectives in the following way: (4 1)

One place statives

she sie

A.y.AgrS'

(---AgrS-__

TP

�

A.,.VP

past1

r---v

AP

� A.w .AP

almost(w) fast

0 war

s"ck I wr(y)

Jtrank

These verbs are syntactically unaccusative because their only argument is embedded under the adjectival state (c£ Rapp 1 997). Note that we do not

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

expectation is borne out by the facts: activities are never unaccusative. One­ place activities like tanzen 'dance' are syntactically unergative, two-place activities like schlagen 'hit' are transitive. In contrast, one-place states are syntactically unaccusative: They are expressed by adjectival predicates (Rapp 1 997:6o). We conclude that they do not possess a Voice-P. The most delicate case is represented by transitive two-place states. Do they possess a Voice-P which introduces a HOLDER-Relation towards the event described by the VP, or is the accusative checked by some other · feature? Whether the subject of a two-place stative is actUally introduced by the HOLDER-Relation seems to be an empirical matter. In our opinion, the fast-data show that at least for psychological statives like hassen 'hate' a decomposition into Voice and VP makes sense. We will proceed as follows. First, we examine the innerfast-modification of the unaccusative one-place statives. We show that fast can be a VP­ modifier as well as an AP-modifier. Second, we show that in the case of activities inner fast intervenes between Voice and VP. As psychological statives behave like activities with inner fast, we assume that they are also decomposed into Voice and VP.

172 VISibility Parameter for Functional Adverbs distinguish between states and times in this construction since we see no linguistic need to do so, at least not for the time being. In (41) it is assumed thatfast modifies the AP. However, a modification of the copula VP should not be excluded. With the copula sein the two modifications do not differ in meaning. But if we look at other copulas we notice that there are indeed two readings:

(42) weil die Milch fast warm blieb (ambiguous) because the milk almost warm remained The AP-modification means that the milk remained in the state of

(43) weil die Milch fast warm geblieben ware because the milk almost warm remained were We assume the following semantics for remain:

(44)

REMAIN is of type (s, (i. (1r, t))), i the type of eventives and/or times

and 1r

=

(s, (i, t)).

II REMAIN ll(w)(s)(P) is only defined, ifP(w) is defmed for any part of s and s is longer than a moment. If defined: II

REMAIN ll(w)(s)(P)

3 Sz: s > <

Sz

=

& P(w)(s2)

I iff 3s1: S 1 > < s & P(w)(s.) = I & I & 'v' s'[s' � s P(w)(s') = I ]

=

-+

Here are the formulae representing the two readings discussed

(45)

a.

b

past 1 >. r[REMAINrw(-XwA5.almost(w)(>.w.warD1ws(the milk)))] past1 >.r[almost(w}(REMAINrw(>.w>.s.wartllws(the milk)))]

It is obvious that gradable adjectives like warm need a more elaborate representation. The standard treatment is to represent them as expressing relations between individuals and degrees/extents (c£ e.g. Seuren I984; Stechow I 984). This issue is ignored in this paper. We conclude that inner fast exhibits a scope interaction with the copula bleiben 'remain', although this verb has an almost trivial meaning. Now consider the change of state­ copula werden 'become':

(46)

weil die Nordsee fast warm wurde b. 3 e[e � past1 & a.

BECOMEew(>.w>.s.almost(w)(-Xw.warttlws(the North Sea)))]

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

being almost warm; hence it is not necessary to assume it had ever been really warm. The VP-modification means that the milk almost remained in the state of being warm: in this case it should have been warm before. The most likely state of affairs described by that reading iS that the milk became slighdy cooler. The second reading is somewhat marginal, because it corresponds to what we have called the counterfactual reading. It would be better expressed by a subjunctive sentence:

Irene Rapp and Amim von Stechow

173

7. 1 .2.

Activities and p�chological statives

Kratzer (1994) decomposes activities into AGENT + VP, and AGENT is located in the head of Voice-P. If we adopt this idea for our analysis, a sentence like

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

c. ? 3 e[e � past1 & almost(w)(BECOMEew{AwAs.wanDws(the North Sea)))] d. # almost(w)(3e[e � past1 & BECOMEew{AwAs.wanDws(the North Sea})]} Reading (46d) is what we have called the counterfactual reading. As was already noted, it has to be expressed by means of the subjunctive. Reading (46c) is possible but not very plausible: We do not know how to describe an event that is an almost becoming of something. The most likely inter­ pretation is that (46c) describes a state that is very close to a change of state. Hence, in this reading inner fast turns a change of state predicate into a state. In (46b) fast modifies the adjectival result state. Evidently, reading (46b) is the most prominent one if we consider the result state to be gradable as in (46a). Consider two other examples: {47) a. weil der Junge fast dunkelrot wurde because the boy almost darkred became b. weil der Junge fast rot wurde because the boy almost red became In (47a) the gradable adjective yields an AP-modifying interpretation for fast. By contrast, the unmarked interpretation for fast in (47b) is the VP­ modifying one. This analysis is supported by facts from temporal modifica­ tion. (47a), describing a change of state, allows for frame adverbials, (47b), describing a state, allows for an adverb of duration: (48) a. weil der Junge in 10 Sekunden/*10 Minuten lang fast dunkelrot wurde because the boy in 10 seconds/*for 10 minutes almost dark red became b. wei! der Junge 10 Minuten langl*in 10 Sekunden fast rot wurde because the boy for 10 minutes/*in 10 seconds almost red became We conclude that the unaccusative copula + adjective constructions provide two positions for innerfast: One of them is the AP-modifying one, the other one the VP-modifying one. Hence, with a change of state copula like werden 'become' we get a resultative reading as well as a sc2-reading, where the latter has marginal status. In section 7.2 we will show that the resultative position is excluded as soon as the result state is part of a decomposition structure.

174 Vtsibility Parameter for Functional Adverbs

(49) weil Rosinante galoppierte because Rosinante galloped would have the following s-structure: (s o)

NP� ___---\ A.,.TP

R05fnante

i.. ,.AspP

past I

/----_

PV{r)

i..0 • VoiceP

ce

� ��

galopplerte

AG

we{x)

The question we want to address is this: does the behaviour offast give us any evidence that justifies this kind of decomposition? The answer will be that we need the decomposition in order to express the plausible scalar reading. .We will see, however, that our argument will not be very strong. Consider the following sentence: (s I) weil Rosinante fast galoppierte because Rosinante almost galloped Assuming Kratzer's decomposition, we have three different positions for fast that are sources for potential different readings: TP ( 5 2)

� �

NP

i..x.TP

Roslnante

i.. ,.AspP

past1

��� i..

cf.rast

�

VoiceP

& PV(r)

�

almost(w) sc l .f

�

i.. ., .VoiceP

�& . �

Vo ce 1 AGENT.,.(xl

VP

almost(w ) sc2.fast

A.,.:VP

gallopm�(e) galopp1erte

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

VP GALLOPINGw{e)

Irene Rapp and Amim von Stechow 175 We know that 'counterfactual' fast requires the subjunctive, so the highest post subject position is ruled out. But we have two scalar positions, which are labelled as sc1 and sc2, respectively:

(53)

& almost(w)(Aw[AGENTew(Rosinante) & GALLOPINGw(e)])] 3 e[e � past1 & AGENTew(Rosinante) & almost(w)(Aw.GALLOPINGw(e))]

sci. ? 3e[e � past1 sc2.

(54)

weil sie das Lied fast grolten sc1 . ?3e[e � past1 & sc2.

almost(w)(Aw[AGENTew(sie) & BELLOWINGw(e)(the song)])] 3e[e � past1 & AGENTew(sie) & almost(w)(.Xw-BELLOWINGw(e)(the song))]

Among these two positions, only sc2 makes sense: In our intuition the AGENT-relation is not modified by fast. In Stechow (1 996) it has been noted that VP-modification of activities makes little sense with wieder in most cases. Stechow (1 996) is reluctant, however, to exclude the inner modification altogether. A possible text that might require that representation is this:

(5 5)

Erst sangen die Eltem den Schlager. First sang the parents the song Spater sangen die Kinder ihn wieder. Later the childen sang it again

It is not necessary that the children had sung the song before; the song had been sung before by someone else. If so, not onlyfast but also wieder requires the decomposition assumed here for activities.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The reading sc2 makes sense: Rosinante performed an action that was almost a galloping. It is not so clear, however, that the first reading exists: Here we have an event of which it is unclear whether Rosinante is its agent. For instance, the event may be a galloping done by some other horse but it may nearly have been done by Rosinante. Clearly, this is not what we are after. If we would not decompose the VP, only the first reading would be available for 'inner' fast. This may be a point in favour of Kratzer's decomposition, but the argument is not very strong, due to the vagueness of our meaning for fast. In any case, Kratzer's decomposition allows us to have a reasonable representation of the scalar reading, and that will be enough for the present purposes. Exactly the same reasoning holds for two-place activities. The following sentence also offers two positions for inner fast:

176 Visibility Parameter for Functional Adverbs Now let us consider transitive statives. Psychological verbs like hassen 'hate', glauben 'believe' behave like activities. First, we have strong intuitions that inner fast does not include the agent relation:

(56)

weil Fritz seine Lehrerin fast hasste because Fritz his teacher almost hated

Second, it is possible to invent a context where wieder allows for an inner reading:

(57)

Hence we assume that the subject of psychological statives is introduced by Voice and that it bears the HOLDER relation to the event described by the

VP.

The situation is different for locative statives like umgeben 'surround'. As these verbs are not gradable, there is no evidence from fast for or against decomposition. However, wied'er shows us that we should not decompose these verbs. Though they allow for repetitive wieder, inner wieder is excluded:

(5 8 )

a.

b.

1 920 umgaben diese Bretter einen Schatz. I 920 surrounded these planks a treasure Jetzt umgeben sie ihn wieder.

Now surround they it again. I 920 umgaben einige Bretter den Schatz. I 920 surrounded some planks the treasure. jetzt umgibt eine Eisenwand ihn wieder. Now surround an iron wall it again.

We conclude that

umgeben

(59) I I AROUND

ll(w)(t)(b)(a)

'surround' is not decomposed into Voice and

VP. It is simply represented by the predicate AROUND embedded under VP: = I

iff a is around b in w at

t.

According to Rapp ( I997: I44 £), there is an important difference between psychological and locative statives, which is proved by the fact that only the former allow for the passive and the imperative. The prese.nce or absence of Voice could be responsible for this differences as well. Hence, the fast- and wieder-data support this classification of atelic verbs given in Rapp ( I 997) for independent reasons.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Vor der Oktoberrevolution glaubten die Bauem in Russland an ein Leben nach dem Tode. Before the October revolution believed the farmers in Russia in a life after the death Heute glauben ihre Kinder wieder daran. Today believe their children again in it

Irene Rapp and Amim von Stechow

177

The upshot of the discussion of activities and statives is that their behaviour under modification byfast does not provide a strong support for a decomposition along the lines indicated in Dowty (1979) or Kratzer (1994). However, Kratzer's decomposition of the VP into Voice + VP allows us to get the intuitively correct scalar reading sc2 for inner fast. 7.2

Subject Result-verbs

Subject Result-verbs are one-place verbs. Their subject is the only argument of the embedded state: Hence, they are syntactically unaccusative (c£ Rapp 1997). Used with the preterite, fast yields an inner reading: ·

The following structUre shows the different positions for inner fast: (61)

VP

�

almost( w) scl .fast

A. •v.VP

·

�

Asw·XP

BECOMECW

�

almost(w) sc2.fast

XP is the category describing the resultant state. Our question is whether fast is above BECOME or directly above the result state. Due to the vague semantics o£Jast, it is not clear whether and how the two representations differ in meaning: ·

(62)

almost(w) Aw.BECOMEew �sAw.ASLEEPsw(the boy) 'e is almost a becoming with the result that the boy is asleep' b. BECOMEew ASAW. almost(w) Aw.ASLEEPsw(the boy) 'e is a becoming with the result that the boy is almost asleep' a.

The first reading presumably expresses a state, namely the state which is almost an event of falling-asleep, so the use of the event variable e might be somewhat confusing, but we have to use it, beeause the predicate BECOME requires an event as its argument. The second reading expresses an event with a graded result, an almost being asleep. In her discussion of verbs of change,

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(6o) weil der Junge fast einschlief because the boy almost in slept

I

78 Visibility Parameter for Functional Adverbs

Fabricius-Hansen (1989) assumes in fact that the two readings exist, and she speaks of semi-telic verbs. If this is so, our theory certainly can express this. There may be a way to decide which of the two readings exists. The second reading is an accomplishment and should therefore not be modifiable by adverbs like 'for two minutes' which say that the event should take place at any subinterval of two minutes, an impossibility (c£ Dowty 1 979). On the other hand, statives are not compatible with adverbs like 'in two minutes'. The following contrast suggests that the modification by fast changes an accomplishment into a stative: (63)

weil der Junge *zwei Minuten l�Kin zwei Minuten einschlief because the boy *for two minutes/°Kin two minutes in slept b. weil der Junge OKzwei Minuten langl*in zwei Minuten fast einschlief because the boy OKfor two minutes/*in two minutes almost in slept c. weil der Junge OKzwei Minuten langl*in zwei Minuten fast eingeschlafen ware because the boy OKfor two minutes/*in two minutes almost in slept were a.

(64)

Der Bailon platzte fast. The balloon burst almost b. Das Seil riss fast. The rope broke �ost a.

Intuitively, the only reading of the sentences in (64) is a stative one. Again we conclude thatfast is above BECQME and under Asp. So our official LFs for the subjunctive and indicative sentence with inner fast are these:

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(6 3 a, b) show that innerfast changes the aspectual class of a predicate: It turns a change of state predicate into a stative predicate. This is explained if we assume that fast is located immediately above BECOME. In (62a) the fast­ modification creates a stative interpretation: The boy is in a state which is very near to a falling asleep, but actually no change of state takes place at all. It is important to note that inner fast and counterfactual fast do not really differ in meaning with non-causative BECOME-verbs: Both (63b) and (63c) describe a state that is very close to falling asleep. This counterfactual touch of innerfast is due to the specific semantic properties of Subject Result-verbs: If we assume thatfast is a VP-modifier, there is no part of the verb meaning above fast. The question remains whether (63b) and (63c) differ in logical form. For the time being, we will assume that they do: in (6Jc)fast is under the Asp-node whereas the subjunctive requires that it is located above it. Finally, consider some Subject Result-verbs which allow for inner fast though a result modification is excluded:

Irene Rapp and � von Stechow 179

Subjunctive inner fast

(65) Indicative inner fast

TP

AspP

/---._

PV(r}

�

i..-T'

past)

" di� m

catJve fast..

r---T---

i.w.VP

�

�XP

..--.. NP ASLEEPsw y einschlief

BECOME,w

�

subjunctive fastw

ASLEEP5w elngeschlafen

We conclude that innerfast has to be above BECOME with Subject Result­ verbs, but we would like to point out an interesting fact. Again, we stress the fact that the position between BECOME and the result phrase XP is accessible forfast if XP has a genuine phrasal character. As we have shown in 7-I.I, a modification of the result phrase is possibl� as soon as the resultative predicate is split into copula + adjective: (66)

weil die Milch fast kiih1 wurde because the milk almost cool became b. 7weil die Milch fast abkiihlte because the milk almost down cooled a.

(resultative) (scalar)

While (66a) has a resultative meaning, (66b) is somehow odd, although the verb abkuhlen should mean exactly the same in terms of decomposition. We do not understand the sentence very well, which is due to the fact that we do not know what a state is which is an almost becoming cooL Under the perspective of decomposition in the syntax, this contrast is rather unexpected, but we find a similar contrast with adverbs expressing repetition. (67)

weil die Milch wieder abkiihlte (repetitive/restitutive) because the milk again (wieder) down cooled b. weil die Milch emeut abkiihlte (repetitive/*restitutive) because the mil!c again (emeut) down cooled c. weil die Milch Wieder kiih1 wurde . (repetitive/restitutive) because the milk again (wieder) cool became d. weil die Milch emeut kiih1 wurde (repetitive/restitutive) because the milk again (emeut) cool became a.

·

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

NP Y

x8o VISibility Parameter for Functional Adverbs

7· 3

Subject Object Result-verbs

Certain verbs describe a change of state that concerns both their subject and their object: erreichen 'reach', vergessen 'forget', eifahren 'learn'. Subject and object are both arguments of a stative under BECOME. In Rapp (1997) it is claimed that the embedded tWo-place state is either locative (erreichen 'reach') or psychological (vergessen 'forget'). Let us take the locative erreichen as an example of the behaviour of Subject Object Result-verbs withfast. We assume the following representation:

(68)

a.

b.

weil der Soldat die Stadt erreichte because the soldier the town reached 3 e e � past1 & BECOMEew (.Xs.XwLOCsw(town)(soldier))

LOC is an abstract predicate meaning that someone is located at some place:

(69)

LOC is of type (s, (i, (e, (e, t)))). II LOC ll(w)(t)(b)(a) = I iff a is located at b in w at t.

The subject of verbs of location is not introduced by a voice head. For Hale & Keyser ( I 994), verbs of these kinds are of the category P of (two place) prepositions. In many cases, this assumption even makes sense morphologically. We do not want to commit ourselves to this view, but our analysis is compatible With it. As the Subject Object Result-verbs do not have · a Voice-P, inner fast should have a counterfactual touch iffast is located higher than VP. This prediction is borne out by the facts. As in the preceding section, it is ·

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The adverbs wieder and erneut mean exactly the same, but erneut cannot look into the decomposition of a verb: While (67a) allows for a restitutive rea4ing, in (67b) no modification of the result phrase is possible. As soon as the result phrase is syntactically represented by an AP, it can be modified not only by wieder, but also by erneut: Both (67c) and (67d) have a repetitive and a restitutive reading. Stechow {I996) takes this as evidence that there must be a distinction between 'full' lexical heads and functional heads which are used for the decomposition of verbs. Functional heads are the natural borders of the 'word' and most adverbs cannot be located inside of them. wieder is a notorious exception. In some sense, it behaves like a morpheme. The literature about English suggests that almost behaves similarly, but fast does not. It behaves like an 'ordinary' adverb and cannot look into a word. In a way, this is a disappointing result, because our hope was that fast would reveal something about the nature of decomposition.

Irene Rapp and Arnim von Stechow I 8 I

(74)

VP

�

BECOME

AgrOP

� accusative AgrO' •.

X

�grO

...--... Subject X' .......

X

Objeci

.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

difficult to distinguish between the counterfactual reading expressed by the subjunctive and the indicative inner reading on intuitive grounds. As before, we assume, however, that the two have different LFs: (7o) a. weil der Soldat fast die Stadt erreichte (stative reading) 3e e � past1 & almost(w)(.Xw.BECOMEew(.Xs.XwLOCsw(t)(s))) b. weil der Soldat fast die Stadt erreicht hatte (counterfactual reading} almost(w)(.Xw.3e(e �. past1 & BECOMEew(.Xs.XwLOCsw(t)(s))]) In theory, we could expect a resultative modification under fast if we consider the copula werden {BECOME) plus a PP: (71) *Der Leutnant liisst den Soldat fast auf den Gipfel werden The lieutenant lets the soldier almost on the peak become Since lassen 'let' is an ECM-verb, one would expect the sentence to be grammatical. Hencefast should be able to modify the PP aufden Gipfel. But for reasons that are not well understood, the copula werden has many restrictions to its complements. It does not tolerate locative or directional PPs. For many pertinent observations, see Vendler (1957). An interesting property instantiated by Subject Object Result-verbs is that they do not show a positional effect with fast/wieder: (72) a. weil der Soldat wieder die Stadt erreichte (restitutive/repetitive) because the soldier again the town reached . b. weil der Soldat die Stadt wieder erreichte (restitutive/repetitive) because the soldier the town again reached (stative reading) (73) a. weil der Soldat fast die Stadt erreichte because the soldier almost the town reached b. weil der Soldat die Stadt fast erreichte (stative reading} because the soldier the town almost reached fast and wieder always allow for an inner reading, independently of their position. This can be explained by the assumption that Subject Object Result-verbs license the accusative under BECOME (c£ Stechow 1996):

I 82 Visibility Parameter for Functional Adverbs structure explains why there is no positional effect. Both adverbs can have an inner reading with the accusative position in their scope: Innerfast dominates the VP, inner wieder dominates AgrOP. We assume that in (72) and (73) the a-examples show the canonical word order. In the b-examples, the direct object is scrambled. Our analysis depends, of course, on assumptions concerning case assign­ ment: The accusative position is higher than the phrase which introduces the semantic subject. This assumption is not compatible with Kratzer (1994) but it seems to be in agreement with work in the minimalist program (c£ Chomsky I 992). We conclude that the lack of a positional effect with Subject Object Result-verbs is due to the fact that they license accusative under BECOME. In contrast to fast, inner wieder can be attached to the result node PP giving rise to the restitutive reading. This complementary distribution of the two adverbs follows from the assumption that fast cannot look into a decomposition, but wieder can.

This

Incremental theme verbs

In this section we treat a subcategory of change of state verbs, incremental

theme verbs. Their change of state is not punctual but gradual Dowty following Krifka (1989), gives the following definition for incremental theme verbs: 'If x is part of y, then if a telic predicate maps y (as Theme) onto event e, it must map x onto an event e' which is part of e.' With incremental verbs, the change of state of an object implies that the object can be divided into smaller parts which undergo the change one after the other: If you eat a cake in five minutes, you eat part of the cake in every minute. Note that there are not only incremental Object Result-verbs like miihen 'mow', trinken 'drink', but also incremental Subject Result-verbs like verdampfen 'evaporate', which are syntactically unaccusative (c£ Rapp 1997). A better term for some of these is 'decremental', since the object is not only gradually affected but disappears. We do not distinguish between mere­ mentality and decrementality, though a more careful analysis may reveal that we actually should do this (c£ Stechow 1997). Now, incremental theme verbs give us good evidence that fast is not a result modifier. Recall that in the case of punctual change of state verbs like enviirgen 'strangle' we did nc;>t have strong intuitions whetherfast was scalar _ or resultative. With incremental verbs, the situation is different. As · the change of state is gradual, a modification of the VP is intuitively different from a modification of the result phrase. Take the eating of a cake as an example. The VP-modifying scalar reading would mean that the change of

{I99I:567),

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

7·4

Irene Rapp and Amim von Stechow I 8 3

state did not take place at all in the described manner. The resultative reading would mean that it was not finished. Hence, incremental verbs give us a chance to distinguish clearly between scalar and resultative fast. To formalize the incremental verbs we use the predicate DEV: {75) Er .miht den Rasen. AGew{he) & MOWINGw{e) & DEVew {the lawn){-\y,\s-\w.MOWNws{y))

{76) Das Wasser verdampft. DEVew {the water)(-\y-\s-\w.EVAPORATEDws{y)) And this is the semantics for DEV, which makes precise the informal notation used in Rapp (1997): (77)

DEV is of type (s, (i, (a, (1r, t))))), 1r of type (e, (s, (i, t))), a of type e. IIDEVII(w)(e)(x)(P) is only defined, if P(w)(e')(x) is not defined for any e' that is a proper part of e. If defined: II DEV ll(w)(e){x)(P) = I iff a. 3 £If{e) = x & for any initial part e' and e" of e with e" a part of e': £{e") is a part of £{e'), and b. for any initial part e' of e: II BECOME II(P(f{e'))) = I].

DEV requires a 'structured property of states' as argument It is a relation between an event and a property + individual. This is encoded in a schonfinkelized manner, i.e. DEV first applies to the property and then to the argument of the property. The presupposition tells us that there is a change of state for every part of the object. As usual, the result is not defined for the time of the change of state. In order to see how this works, let us evaluate (76): II DEVew {the water){-\y-\s-\w.EVAPORATEDws{y)) II = I iff 3 £If{e) = the water & 'v'e'[e' is initial part of e & e" is initial part of e & e" a part of e' -+ f{e") is a part of f{e')] & 'v'e'[e' is initial part of e -+ II BECOME ll(w)(e')(-\s-\w.EVAPORATEDws(f{e')) = I]]]

This means that for any initial segment e' of the event e, the water part £{e') which corresponds to the segment e' has not evaporated at the beginning of

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

For the purpose of this paper, we C
1 84 VJSibility Parameter for Functional Adverbs

e', but it has evaporated at the end of e'. At the end of the entire event e, the entire water f{e) has evaporated. This is exactly Krifka's semantics of incrementalityIdecrementality. The LF which expresses the formula is slighdy more complicated than the one for achievements: DEV does not simply embed a result phrase XP, but a structured result phrase, i.e. the object has to be extracted and adjoined to the XP:

A superficial look at the structure may suggest that the adjunction of the object y to its own host is semantically vacuous because it reduces to what we had before the adjunction by means of >.-conversion. But this is not the way we evaluate the LF: y is the first argument ofDEV while the abstract AyAs.AwXP is the second one. We find exactly the same distribution offast and wieder as in the last section: fast can only attach to VP, but it cannot modify the lowest XP. wieder, on the other hand, may occur in both positions, i.e. we have a resultative ('restitutive') reading for wieder but not for fast. ew

{79)

a.

b.

c.

weil das Wasser fast verdampfte almostw >.w[DEV (the water)(>.y>.s>.w.EVAPORATEDws{y))] (scalar) *DEVew (the water)(>.y>.s>.w.almostw[UVAPORATEDws(y)]) (*resultative) weil das Wasser wieder verdampfte againe -\[DEVew (the water)(>.y>.s>.w.EVAPORATEDws(y))] (repetitive) OKDEVew {the water)(>.y>.s>.w.agaills[-\.EVAPORATEDws(Y)]} (restituti�e) ew

·

(So)

a.

b. c.

These facts corroborate our conjecture that fast cannot look into the decomposed VP, while wieder can. Transitive incremental verbs behave exactly alike. We recall first that wieder behaves in the usual way:

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Asp PV(r)

Irene Rapp and Amim von Stechow r 8 5

(8 1)

Melina mihte den Rasen wieder. Melina mowed the lawn again b. Melina mihte wieder den Rasen. Melina mowed again the lawn a.

(repetitive/restitutive) (only repetitive)

The first reading is expressed with wieder modifying Voice-P, the second is obtained by attaching wieder to the result phrase XP. There may be intermediate readings, but these do not concern us here. As we would expect, fast does not have the resultative reading, e.g. (82)

a.

(only scalar) (OK with Subjunctive ll)

It follows from the previous discussion that (82) must have the representation (83) AGew(M.) & ahnostw(-\w(MOWINGw(e) & DEVew(th-L)(-\y-\s-\wMOWNsw(y))]), while the representation (84) AGew(M.) & MOWINGw(e) & DEVew(th-L)(-\y-\s-\w.ahnostw(-\w(MOWNsw(y)])) is doubtful on semantical grounds, because it implies that no part of the lawn is mown after the action. It is very hard to imagine such a scenario and certainly the sentence does not mean that. Anyhow, it seems to be safe to assume that a resultative reading is excluded for syntactic reasons. A remark on the structure is in order. With Rapp (1997) we assume that the lexical information of verbs of this class is represented in two positions: It describes the result and the manner of the action. If we neglect the 5

Manfred Bietwisch remarks that (82.b) has a scalar reading in the following conteXt: Melina mahte fast den Rasen, so niedrig harte sie die Rader eingestellt. Melina almost mowed the laWn, so deep had she adjusted the wheels We think this means that the car that Melina was driving was almost mowing the lawn. It seems to us that this is interpreted in analogy to locative verbs of touching such as: (ii) Ihre Haare berii.hrten fast seine Wange. Her hair almost touched his cheek. In section 7.1.2. we said that th�'are not decomposed: a rwQ place stative XP is embedded under a semantically empty light verb V. For instance. the 6m sentence is analysed as: (ill) almost(w)[VP v [xP her hair touch,... his Cheek ]] We have to make sure that accusative can assigned inside VP. This is the intuition. But a direct analysis of Bierwisch's sentence seems not possible in our system. (i)

·

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Melina mihte den Rasen fast. Melina mowed the lawn almost b. *Melina mihte fast den Rasen.5 Melina mowed almost the lawn

186 VISibility Parameter for Functional Adverbs

semantic strUcturing of the XP, the syntactic structure of the Voice-P is

this:

(8s) [voice-P Melina [VP [VP [XP the lawn MOWN) DEV) MOWING) AG)

fast must be attached to the higher VP. An attachment to the lower VP is

(86)

??Rafi all den Kuchen fast. Ra£i ate the cake almost (only scalar) b. Ra£i fraB den Kuchen fast. Ra£i devoured the cake almost a.

The examples in (86) cannot mean that the cake was almost eaten/ devoured. To get a scalar reading, it must be possible to imagine a kind of graduation. In (86b), a scale is opened by the verb fressen 'devour' which denotes the eating activity performed by animals. So. there is eating, and eating in a hasty, uncontrolled manner. This is the scale. By contrast, it is hard to open a scale with (86a): Being eaten is one of the things we expect to happen to a cake. Of course, the same reasoning holds for miihen 'mow'. If we accept (82a) with a scalar reading, we are forced to imagine a scale being activated by the verb. This could be the case if Melina is a goat eating grass in a way that amounts to a mowing. Hence, it is not always easy to produce a scalar reading, but still it is the only one available for inner fast. The lack of resultative fast must be a lexical/syntactic property of our variant of German. According to Dowty (1979: s B), English almost can be a result modifier of incremental verbs. He assumes that the sentence John · almost painted a picture, has two readings: '(a) John had the intention of painting a picture but changed his mind and did nothing at all, or (b) John did begin work on the picture and he almost but not quite finished it.' None of these readings is possible for a German preterite sentence. As we observed in section 6, in German the (a)-reading requires the subjunctive. If we assume that the fast-generalization holds for German, but not · for English, both phenomena are explained. Not only resultativefast/almost, but also counterfactualjast/almost used with the preterit would attach to a non­ visible functional head.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

contradictory, because a MOWING that doesn't generate an object MOWN is not a MOWING. Thus, there is a close connection between the manner meaning and the result meaning, which our theory has not formalized. Some comments on the scalar reading are helpful here. A scalar reading is only available if the verb opens a scale. Consider the following contrast:

Irene Rapp and Arnim von Stechow 1 87 7. 5 Verbs of motion

(87a) shows the usual wieder-ambiguity. There is a repetitive (89a), a restitutive (89b), and an intermediate reading (89c)-the latter was discussed in section 7.1.2: (89)

a� Ae[AGew{x) & rtlllw(e) & (repetitive) BECOMEew{A5Aw(L0Csw(the house)(x)))] b. AGew(x) & runw(e) & BECOMEew{AsAw(a�(A5LOCsw(the house)(x))]) (restitutive) c. AGew{x) & againew Ae(run w(e) & BECOMEew{AsAw(LOCsw(the house)(x))]) a.

Note that the subject of the directional PP is controlled by the subject of Voice-P. The control is indicated by the choice of the same variable.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Verbs of motion can be causative (schieben 'push') or non-causative (sich bewegen 'move'). Syntactically, the causative verbs are transitive, the non­ causative verbs are unaccusative in English and very often reflexive in German. We have already shown the positional effects of transitive verbs with fast and wieder in detaiL For the sake of simplicity, we will therefore focus on non-causative verbs of motion in this section, but our considera­ tions will apply to the causative verbs as well. Let us forget about the fact that verbs of motion are incremental in some sense as well: The motion gradually affects a path. The notion of path has been analysed in great detail in Cresswell (1978). We could analyse the affection of the path by a special version of the BECOME-operator along the lines of the following paragraph, and we might, following Egg (1993), call this operator CHANGE. For the purpose of our investigation, the special nature of this operator is immaterial, however. So in order to simplify things, let us follow Dowty (1979) and analyse directional prepositions and decomposed verbs of motion by means of our version of the BECOME-operator. Interestingly enough, wieder andfast show a special positional effect with verbs of motion: (87) a. weil sie wieder zum Haus rannte because she again to the house ran b. weil sie zum Haus wieder rannte because she to the house again ran (88) a. weil sie fast zum Haus rannte because she almost to the house ran b. weil sie zum Haus fast rannte because she to the house almost ran

1 8 8 VISibility Parameter for functional Adverbs

As soon as wieder follows the directional phrase, there is only one reading,

which is distinct from the three other ones. (87b) means that only the er component was repeated:

mann

(9o) AGew(x)

&

� �e[l'Wlw(e)] &

BECOMEew(�s�w(LOCsw(the house)(x)))

(91)

AGew{x) & rullw(e) & Aw.almostw(BECOMEew{�s�w(LOCsw(the house)(x)))] b. AGew{x) & Aw.almostw[l'Wlw(e)] & BECOMEew(�s�w(LOCsw(the house)(x))) a.

(88a) does not have a resultative reading, however. The following LF is excluded: (92) *AGew{x)

&

rullw(e) &

BECOMEew(�s�w(almostw[Aw.LOCsw(the house)(x))])

We conclude that, as usual, fast does not have access to the XP expressing the result state. Depending on its position, it modifies either the entire directional PP or the manner-VP: VoiceP

VP

� � � PP

·

almost,. fast

, llow.PP

almo5tw fast

�

P+
>..sw·XP

�x·

)(

�NP

X+ loc LOC.., zu

the house dem Haus

·

VP

w.VP RUNNING... renn t

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

There is a similar positional effect for fast. (88a) means that she did a running that was almost into the house. This is a direction that does not lead quite into the house, but almost. (88b) means that she did an almost running, perhaps a very quick walking, into the house. We assume that in the first case the adverb is adjoined to the directional PP, in the second to the manner-VP:

Irene Rapp and Arnim von Stechow I 89 The tree shows that verbs of motion with an open directional offer two scalar positions for fast. It is interesting to observe that fast does not have access to the directional information if it is expressed by decomposition, i.e. if the directional is part of the lexical meaning of the verb. The verb besteigen 'climb' illustrates the point: (94) ?Der Junge bestieg den Hiigel fast. The boy climbed the mountain almost

(95) Der Junge erstiirm.te den Hiigel fast. The boy ran up the hill almost The difference between (94) and (95) is that the latter contains a gradable MANNER-component, whereas the former does not. The following tree shows that it is the MANNER-component of ersturmen that gives us the position for inner fast: VoiceP

VP

& AGwe( X)

� � *wmoo� � VP

pp

*fast

P+di BECOMEew

almostw fast

>-w.PP

.VP STORMINGw(e) erstiinnt

.X P

---" � X'

X

,....---...

ONSW

NP

the mountain den Berg

While this representation correctly represents the meaning, we cannot be entirely happy with it. BECOME is classified as a preposition P and should be visible for fast, contrary to the facts. Now, verbs like ersturmen must be

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(94) means neither that the boy did a large part of a climbing of the hill nor that he followed a path which does not lead quite up to the top of the hill If the sentence means anything at all, then that the boy did something that was almost a climbing of the hill This VP-modifying reading is more likely if we use a verb like ersturmen 'run up', which induces a: scale:

190 VISibility Parameter for Functional Adverbs

7.6

Conclusion

In our dialect, fast cannot modify the result state if the latter is introduced by lexical decomposition. This means that fast cannot be adjoined to abstract functional phrases. There cannot be semantic grounds for this behaviour, because the result state can be modified if it is introduced by a projection with a lexically visible head. Recall that scalarfast does not attach to Voice-P, but to VP. wieder, on the other hand, behaves differently. It can attach to any functional projection which is an appropriate argument for that adverb. If thefast-generalization is responsible for the lack of sci {ast and res{ast in decomposition structures, we have to assume that only the verbal head V is morphologically visible-an interesting consequence for our decomposi­ tion structures. Our prediction is that those German speakers who allow for a counterfactual reading of fast with the preterit accept the resultative reading of inner fast as well. Both phenomena depend on the same parameter. We have not investigated the variation among German speakers systematically. However, the difference between English and (our) German can be explained by assuming that the fast-generalization holds for (our) German, but not for Englisli: English allows for counterfactual almost with the preterit as well as for resultative almost. Hence, the two phenomena seem to be closely related. It is an advantage of our theory that they can be captured by one single generalization.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

applicative verbs in some sense, ie. the directional preposition is incorpo­ rated into the verb (c£ Baker 1988). Hence it is not visible anymore. We will not attempt to give the correct analysis for the verb. Note that a representation without any decomposition would have the problem that there would be no manner component that could be modified byfast. Thus, the analysis might at least be partially correct. Whatever the correct approach is, besteigen 'climb' has exactly the same structural representation. Like ersturmen it disallows for a modification of the directional component. In contrast to erstiirmen the VP-component 'CLIMBING' does not open a scale. Hence, it cannot easily be modified by fast. The discussion of the verbs of motion points out once more thatfast can only attach to projections with a visible lexical head. Obviously, the VP­ component is such a projection: A fast-modification is possible, if the verb opens a scale. Resultative fast is excluded, because the result phrase XP is always given by decomposition. A fast-modification of the directional phrase is possible if its head is a morphologically visible preposition.

Irene Rapp and Amim von Stechow 191 8 LEXICALLY EXPRESSED COMPLETION 8.1

Fast as a modifier of result states?

We have argued that fast does not have access to a result state which is expressed by an XP in a decomposition structUre. Here are some apparent counterexamples:

We will argue that these constructions do not exhibit modifications of.an abstract result-XP contained in a decomposition. In (97) and (98), an adverb of completion is added to an incremental theme verb and a directional verb, respectively. The semantics of the adverb will make sure that a modification of Adv + VP or Adv + PP by fast yields more or less the same meaning as modification of the result state of a decomposition (8.2 and 8.3). In (99) and (1oo), the access to the result state is due to specific syntactic constructions, viz. the adjectival passive and the perfect (8.4 and 8.5). 8.2

fertiglganz/vollstandig 'completely'

As we have seen in the last section, incremental theme verbs do not allow for a result modification byfast. The situation is different if an adverb like Jertig 'ready' or ganz/vollstiindig 'entirely/completely' is added. While (1o1a) is unacceptable, {Ioih) is correct:

(1o1)

??Sie mahte den Rasen fast. She mowed the lawn almost b. Sie mahte den Rasen fast fertig/ganz/vollstandig. She mowed the lawn almost completely a.

The following example shows that fast has to precede the adverb of completion in order to achieve a "result" modification. Iffast follows the adverb, only a scalar reading is available:

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

{97) Er mahte den Rasen fast fertig. He mowed the lawn almost completely (98) Er schob den Wagen fast his nach Berlin. He pushed the car almost up to Berlin {99) Der Rasen ist fast gemaht. The lawn is almost mown (1 oo) Er hat den Rasen jetzt fast gemaht. He has the lawn now almost mown

192 VISibility Parameter for Functional Adverbs

(102)

*weil sie den Rasen ganz/vollst:andiglfertig fast mahte because she the lawn completely almost mowed

It follows thatfast modifies a projection to which an adverb of completion has been adjoined and the 'result' interpretation is produced by a combination of fast and the adverb. We will interpret adverbs of completion as modal operators. The idea is that an action with property P is completed iff there is no possible world where the action is a bigger P-action. (103) II COMPL ll(w)(e)(P) = 1 iff -, 3w'e' [e proper part ofe' & P(w')(e� = I ]

(1o4) AGr:w (she) & MOWINGw{e) & almostw Aw.COMPLr:w AwAeDEVew(the lawn)(AyAwAs.MOWNsw(y)) The formula can be paraphrased as: 'She is the agent of action e which is almost a complete mowing of the lawn.' Note that the COMPL-Operator is not adjoined to Voice-P: It only modifies the change of state component VP. The effect of adjoining both COMPL and fast to VP is that we get a 'pseudo-resultative' reading: If a change of state event is almost completed, the result state is almost reached. 8.3

his

'up t�' with incremental path verbs

Directional PPs are treated alike: bis 'up to/until' is analysed as the COMP­ operator for directionals. !jere are the relevant data. (105) a. Er schob den Wagen fast nach Berlin He pushed the car almost to Berlin b. Er schob den Wagen fast his nach Berlin. He pushed the car almost up to Berlin (Io6) a. Er lief fast nach Berlin. He walked almost to Berlin b. Er lief fast his nach Berlin He walked almost �P to Berlin.

(scalar) (pseudo-resultative) (scalar) (pseudo-resultative)

Once again, a result reading is only possible iffast precedes bis: (I 07) a. *weil er his fast nach Berlin rannte because he up almost to Berlin ran

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Note that this operator can only apply to change of state verbs. It may well be the case that adverbs of completion have been treated along these lines in the literature, but no such analyses is known to us. If we disregard tense, the formalization of (10Ih) is something like the following formula.

Irene Rapp and Arnim von Stechow i93 b. weil er his nach Berlin fast rannte because he up to Berlin almost ran

(scalar)

The distribution of the data follows from our analysis. For convenience, we indicate the formulas expressing the sentences in ( I o6): (1o8)

#AGew (x) & almostw[�w.BECOMEew(�w�sLOCsw(Berlin}(x}}] RUNNINGw(e) b. AGew (x) & almostw(�w.COMPLew �w�e.BECOMEew �w�sLOCsw(Berlin)(x) & RUNNINGw(e)) a.

(1o9) AGew (x) & (COMPLew �w�e.BECOMEew �w�sLOCsw(Berlin)(x)) & almostw �w.RUNNINGw(e) 8.4

Adjectival passives

In section 2 we briefly mentioned Kratzer's analysis of the adjectival passive (Kratzer I 994). The essential feature is that the adjectival head introduces the result state of an action, which is called target state by Kratzer. Iffast operates on an adjectival passive phrase, we obtain a property true of a state if the state is almost the target state of the action described by the VP. As it stands, Kratzer's analysis is not compatible with our more refined analysis of incremental verbs. So let us forget this for the time being and let us assume her simpler analysis of the verb 'to mow'. (uo) MOW (in the style of Kratzer (1994) is of type (s, (i, (e, t))). II MOW ll(w)(e)(y) = 1 iff y is mown in w at the target state of e. The LF of the adjectival participle in sentence (99) together with its fast­ modifier would then be this:

(1 1 1)

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The reader has to bear in mind that the 'scalarity' expressed by (roSa) concerns the meaning of the directional PP only, ie. the directional expresses an 'almost to Berlin' direction. If we grade the meaning of the manner component expressed by the verb of movement, we have to modify the verb alone and the adverb has to follow the directional PP as in (Io7b). And this is the formula expressing that reading:

194 VISibility Parameter for Functional Adverbs

P:au:K is Kratzer's perfectivizer which must not be confused with our PERF{ect)-operator, which is used in the next section.

(I 12) Kratzer's perfectivizer PERFK is of type ((i, t), (i, t)). I PERFK II(P)(s) = I iff 3e(s is the target state of e & P(e) = I ]

.

(I998): (I I 3) adjectival passive PERfS is of type (1r, 1r), where

= (s, (i, (e, t))). II P� II(P)(w)(x)(s) is only defined, if there is an preceding event e 1r

abutting s, such that a. II BECOME ll(w)(e)(.XwA5(P(w)(s)(x)]) = I iff P is an achievement/accomplishment b. II DEV ll(w)(e)(x)(.XX.Xw.Xs[P(w)(s)(x)]) = I iff P is an incremental pro_peey. If defined, II P.E!W I I(P)(w)(x)(s) = 1 iff (P)(w)(x)(s) = I .

Given this semantics, the sentence Der Rasen ist gemiih t The lawn is mown' would be represented by the following formula:

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Thus, (I I I) expresses the property which is true of a subject y in a word w at state s if s is almost the target state of an event e which is a mowing of y in w. If we predicate the AP of the subject 'the lawn', we have an analysis of sentence (99). Kratzer's analysis is compatible with our assumption thatfast cannot look into a decomposition, but it has some problematic features. To begin with, the meaning rule (I I2) presupposes that it makes sense to speak of target states of events. This is common practice in the literature, for instance, Parsons (I 990) speaks of culmination points of events, target states, etc. Events, however, are individuals of a certain kind, and we have to make sure that we consider only events for which the notion makes sense. Further- · more, there should be only one target state pro event. But why should the target state of the mowing of my lawn be that the grass is short? It could be very well that the gas tank of the lawn-mower is nearly empty, that I have finished my work etc. In other words, the notion does not make much sense without a lexical description. In fact, the lexical content of resultative verbs tells us. exactly what the target state is. The other problem with Kratzer's analysis is that it is not compatible with the more complicated analysis we gave for incremental verbs. To improve the situation, we assume that the adjectival perfectivizer PERF' gives us the lexically determined result state and presupposes that an event generating this state occurs just before. In an informal way, this point of view is defended in Rapp (I996). The following semantics is due to Stechow

Irene Rapp and Arnim von Stechow I9S ( r 14) P£RF'i(MOWN)(w)(pres1)(the lawn) This

presupposes:

3e[e > < g(pres1) & II DEV ll(w)(e)(the lawn){hlw.As[IJ MOWN ll(w)(s)(x)])

]

= r .

And it is true iff the lawn is mown in w at g(pres1). '><' stands for the abutting relation. Recall that we do not distinguish between times, states and events. For convenience we give the LF for sentence {99):

____----\

pres1

f..

�

NP� lawn

the der Rasen

T

f.. x.V'

r---v-

AP

� al�:�(w) f..w .AP(w)(s)(x)

0 1st

XP....A..-....., PERFK

MOWN gem8ht

If we carry out several ..\-conversions, the formula reduces to the following expression: (I I 6) almost(w)(..Xw[Pmu:s(MOWN)(w)(pres1)(the lawn)]) This

means that the lawn is almost mown at the present time. 8.5

Perfoct

The PERF-aspect is an existential quantifier which is expressed by an auxiliary (haben/sein). Hence we have a visible projection and fast can be attached to that. Here is the tree for the complex VP fast gemiiht hat in

( r oo):

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

{I I S)

196 VISibility Parameter for Functional Adverbs AspP

almost(w) fast

Asp PERF(r) bat

(u 8) almost(w)[>.w.PERF(pres1) (>.e[DEVew(the lawn)(>.y>._.MOWNsw(y)) & MOWINGw(e) & AGw(e)(he)])]

With respect to the contextually given assignment g, this means that the time g(pres1) is almost after an event which is a mowing of the lawn done by him. For the semantics to work, it is important that almost contains a negation. Otherwise, the relative scope between almost and PERF would not matter because PERF does not contain a world variable but it is merely an existential quantifier over events. It should be obvious that this analysis raises a number of questions. Among other things, it is not obvious that the PERF-information is located in the auxiliary and not in the participle. In that case, the auxiliary would be semantically empty. It could very well be that this alternative is the correct one (c£ Musan 1 998) The following topicalization seems to support the alternative:

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Recall that x is the trace of the subject and y is the trace of the object. AgrO is above AspP. The tree is complicated but it is not very different from the tree for the fmite verb in the simple past miihte 'mowed'. The main difference is that the aspect is expressed by a word and therefore has a projection visible forfast. The entire formula which expresses the meaning of sentence (10o) is the following one:

Irene Rapp and Arnim von Stecbow 197 (I I 9) Fast gemaht hat er den Rasen jetzt. Almost mown has he the lawn now If this sentence did express the proposition that the lawn is close to be mown. then the PERF-information has to be located in the participle head. But we are not so sure that the sentence really means that. A more careful analysis of participles has to be postponed to another occasion (c£ Stechow I998).

We will make the following points in this section. fast is a positive polarity item (PPI), ie. it cannot occur in the scope of a negation. The sentential negation is located above Voice-P. Therefore, fast + nicht should always trigger the subjunctive II, because it is outside Voice-P. There are counter­ examples to that prediction. We will argue that fast + nicht can also be a complex particle which is synonymous with kaum 'barely'. This particle can occur in a scalar position. The following data show that fast is a positive polarity item in the strict sense that it cannot occur in the scope of the sentential negation nicht: ( 120)

"'Er hasste sie nicht fast He hated her not almost b. *Sie schlief nicht fast ein She slept not almost in a.

According to Diesing (I992), the sentential negation (NEG) is located very high in the sentence, between INFL (our AgrS) and the classical VP (our AgrO). We will assume the following LF for negated sentences: {12I)

TP

r---'r

r---

NegP

,...--...

NEG

T

AspP

NEG is either an abstract negation, which is made visible by a 'cohesion' in the sense of Bech {I9SSIS7). ie. negative indefinite articles like kein, which have no negative meaning but are NPis. Or NEG is a visible adverb, ie, nicht. It follows that sentence {I22a} has the representation (122b): {I22)

sie den Schlager nicht grolten b. -,PV(Past1)(..\e[AGw(e)(they) & BELLOWINGew(the song]�) a.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

9 FAST AND NE GATION

198 VJSibility Parameter for Functional Adverbs

Deictic terms are scrambled past NEG in German (c£ Diesing I 992) The prediction following from these assumptions is that a fast above NEG should go in tandem with the counterfactUal reading and Subjunctive 11 Scalar readings and the preterit should be excluded in this configuration. There are many sentences £iJ1611ing this expectation: (I 2 3) a. Er h.atte den Zug fast nicht erreicht. He had the train almost not reached b. Ich ware gestem fast nicht nach Berlin gefahren. I were yesterday almost not to Berlin travelled The following sentences are apparent counterexamples to the prediction, however. {I24) a. Es regnete fast nicht/kaum. It rained almost not/barely b. Sie atmete fast nicht/kaum. She respired almost not/barely These sentences have no counterfactual touch and we therefore conclude that the negation adverb cannot be the sentential negation. In fact, we believe thatfast + nicht may form a complex adverb, which is synonymous with kaum 'barely': (us) fast + nicht may form a complex adverb synoymous with kaum 'barely'. I I ahnost + NEG ll(p)(w) = II barely ll(p)(w) = I iff a. 3w': w' is almost identical to w & p(w') = o, and b. p(w) = I . The LF of (124b) is the following structure: (126) TP .

�. .

r------_

T

AspP

;----____

PfV(r)

�

NP she sie

" . ' .oace

/.'---.. /--_

& AGENTow(x)

Ad\1 barely(w) rast nlcblikaum

�.... VP respire.., atmete

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

�

Irene Rapp and Amim von Stechow 199 Thoughfast + nicht and kaum are synonymous according to rule (rzs), there is a syntactic difference between the two. Iffast figures as a sentence adverb, it selects the Subjunctive II. kaum as a sentential adverb has a meaning different from fast + nicht, as can be seen from the non-synonymy of the following two statements:

(127)

Er b.atte

a sentential adverb, kaum means 'hardly', not 'barely'. Thus, a lexical ambiguity seems to be involved here, which makes the facts even more baflling.

As

1 0 WHAT IS A LEXICAL ENTRY? Any decompositional approach has the difficulty (and duty) to say what the lexical entry of decomposed verb (or other head) is. The most traditional and redundant view is that a lexical entry is a tree with the heads filled and with the NP/DP-arguments left open. Something like this is said in Hale & Keyser (1994), and this will be our official position. Thus the entry for the finite form ojfnete 'opened' is the following strUcture:

(u8)

VoiceP

� ���X

NP

Voice'

vOICe

VP

XP

NP

AG

V BECOME

3.sg.,past OPEN iiffnete

The representation differs fr�m traditional entries in terms of sub­ categorization features only in so far as we have more than one head. The structure above Voice-P should be deducible from general principles, such as case checking and the checking of the finite verb features.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

das Examen fast nicht bestanden. He had the exam almost not passed 'He almost did not pass the exam' b. Er b.atte das Examen kaum bestanden. He had the exam hardly passed 'He hardly would have passed the exam' a.

2.00

VISibility Parameter for Functional Adverbs II

THE VISIBILITY PARAMETER FOR ADVERBS

German

Adverb

projections seen

Our

wieder/again

TP

?

?

AspP

+

+

Voice-P

+

+

BECOME-P

+

+

II

XP

+

+

emeut 'again'

TP

?

?

AspP

+

?

Voice-P

+

?

BECOME-P

+

?

II

II

II

II

II

II

II

XP

fast/almost

TP

n

II

-

II

?

-

+

AspP

-

+

Voice-P

-

+

+

+

-

+

.

II

English

BECOME-P XP

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Almost is a classical witness for decomposition in English, but fast has no access to the result phrase XP in the decomposition. Our data behave consistently in this respect. We have proposed to treat the contrast by the principle that fast can only attach to projections that have a 'visible' head. The visibility cannot be an absolute notion, but has to be relativized to particular adverbs, to dialects, and to speakers. In the text, we have spoken of the fast-generalization, but the principle behind it seems to be more general, and the Visibility Parameter for Adverbs should capture it. The following list shows which adverb can see which projection:

Irene Rapp and Arnim von Stechow 201

'' And, following Dowty (1979),

as

P for the arulysis of directionals.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The chart shows that the head which carries BECOME is visible for fast 'almost' and hence must be a visible head. But an attachment to XP, Voice­ P, AspectP and TP seems not possible. English almost can attach to each of these projections and so can German wieder. The German adverb erneut means the same as wieder, but it cannot modify the result phrase XP. On the other hand, erneut must have access to Voice-P, because it expresses repetition. Every adjunction must make sense, semantically. If wieder does not attach to TP, then this is due to semantic reasons: wieder has a time argtiment and must be in the scope of tense. The V1Sibility Parameter for Functional Adverbs therefore means that functional projections are visible only for D-adverbs. Hence, we classify almost, wieder and again as D-adverbs. We must admit that erneut is somehow problematic for our classification: It attaches to Voice-P and AspP but obviously it cannot see the result XP. We have shown that German fast is definitely not a D-adverb. This classification accounts for the lack of a resultative reading and for the impossibility of counterfactual fast with the preterit in German. On the other hand, a projection is visible forfast if it has a lexical head. Hence, fast can modify the result phrase of copula+adjective constructions, even if the AP has the structure [AA XP] with an abstract A-head. As to the Subj unctive II, here we have the structure [VP AspP Auxiliaryv)· In this case, the lexical head V of the auxiliary enables fast to modify the VP. Presumably, the subjunctive auxiliary is semantically empty. Therefore this modification amounts to a modification of AspP, and we obtain the counterfactual reading. Our account has important consequences for decomposition. Iffast can attach to the BECOME- VP, then the head hosting BECOME must be 'visible'. In view of the fact that BECOME is an abstract morpheme, one would rather have expected that this head is invisible. It is, however, an empirical result thatfast can attach to BECOME-phrases. Therefore we have categorized BECOME as the lexical category V.6 The crucial theoretical assumption is that lexical categories always generate visible projections. On the other hand,Jast does not seem to attach to XP and Voice-P. These projections are only visible for D-adverbs. This result is not trivial, because is provides evidence for our decomposition. In particular, we must separate the XP from the BECOME-node and the AGENT/HOLDER argument from the arguments of the stative root, i.e. the XP. The latter is an old idea originating with Marantz (1 984) and spelled out in more precise semantic terms in Kratzer (1 994).

2.02 VISibility Parameter for Functional Adverbs

Acknowledgements

This research was carried out within the Sonderforschungsbereich 441 'Linguistische Datenstrukturen: Empirische und theoretische Grundlagen der Grammatikforschung'. The results were first presented at the conference 'Lexicon in Focus', August 1998, Wuppertal. We would like to thank the participants of the conference for constructive criticism. We also thank Manfred Bierwisch, Kjell Johan Sa:be, Satoshi Tomioka, and two anonymous referees for helpful comments.

IRENE

RAPP

Deutsches Seminar Universitiit Tiibingen Wilhelmstrasse so D-72074 Tuebi�gen Germany e-mail: [email protected]

Final

Received: 06.05.99 version received 17.12.99

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

One might dispute this kind of explanation and one might claim that the functional nodes in question are artefacts of the theory. But then one would have to explain the wieder-facts, the positional effects and the variation among speakers and languages. The visibility of BECOME corroborates recent proposals made in Distributed Morphology that classify BECOME as a 'light verb' v (e.g. Marantz I 997 or Alexiadou I 999). In Distributed Morphology the lexical roots have no lexical category (V, A. P, N) at all. The root projection is called LP, 'lexical projection', in that theory. However, L is not a lexical category. It simply indicates the presence of a root. LP is precisely our XP, which has no lexical category. Details aside, the theory developed here seems to be a variant of Distributive Morphology, though it has developed independently over the years, notably under the influence of Baker (1988). It is interesting to note that we came to this particular conception of the morphology/syntax interface by semantic considerations: We wanted to explain why adverbs seem to have different meanings at different syntactic positions. The explanation is that they scopally interact with functional heads that have a meaning. These heads are not manifest in phonetic substance but they are visible at LF and D-adverbs attach to them. As a final remark, we note that this paper builds on the decomposition of verbs assumed in Stechow {I996). It provides some further details, but the essential features of the analysis are the same. Furthermore, it shows that the verb classification given in Rapp (I 997.) can be fruitfully used to describe the fast- and wieder-data. The decompositions clearly recall for revisions and refinements, and other details like Case theory remain to be spelled out.

Irene Rapp and Amim von Stechow 203 VON STECHOW Seminarfor Sprachwissenschaft Univmitat Tubingtn WilhelmstrllSSe 1 13 D-72074 Tubingen Germany e-mail: [email protected] http://coletrane.sfi.nphil.uni-tuebingen.de/Lehrstuhl/lwme.html ARNIM

REFERENCES Cresswell, M J. (I973), Logic and Languages, Methuen, London. Cresswell, M J. (1978� 'Prepositions and points of view', Linguistics and Philosophy, 2: 1-41. Diesing, Molly {1992), Indefinites, MIT Press, Cambridge, MA. DiSciullo, Anna Maria & Williams, Edwin (1987), On the Definition of Word, MIT Press, Cambridge, MA. Dowty, David (1979), Word Meaning and Montague Grammar, Synthese Language Library, Reidel, Dordrecht. Dowty, David (199I), 'Miematic proto-roles and argument selection', Language, 67= 547-6I9. Egg. Markus (I 99 3), 'Aktionsart und Kompositionalitat: Zur kompositionel­ len Ableitung der Aktionsart komplexer Kategorien', Dissertation, University of Konstanz. Ernst, Thomas (1998), 'Chapter 4= Argu­ ments for right-adjunction', Indiana University, unpublished MS. Fabricius-Hansen, Cathrine (I 98 3 ), 'Wieder ein wieder? Zur Semantik von wieder', in R Bauerle, C. Schwarze, & A. von Stechow (eds), Meaning, Use, and Inter­ pretation ofLanguage, de Gruyter, Berlin. Fabricius-Hansen, Cathrine (1989), 'Tempus im indirekten Referat: In Tempus-Aspekt-Modus', in W. Abra­ ham & T. Janssen (eds), lm lexikalischen und grammatischm Formen in dtn gtrma­ nischen Sprachen, Niemeyer, Tiibingen. Hale, K. & Keyser, ]. (I994), 'On argument structure and the lexical expressions of syntactic relations', in K. Hale & S. J.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Abusch. Dorit (1993). 'Two theories of tense in intensional contexts', Pro­ ceedings of the Ninth Amsterdam Colloquium, Amsterdam, The Nether­ lands. Alexiadou, Artemis (1999), 'On the syntax of nominali'"'tion and possession: remarks on patterns of ergativity', Habi­ litationsschrift, University of Potsdam. Adas, J. D. (1984), 'Comparative adjectives and adverbials of degree: an introduction to radically radical pragmatics', Linguistics and Philosophy, 7: 347-77Baker, Mark (1988), Incorporation: A Theory of Grammatical Function Changing, Uni­ versity of Chicago Press, Chicago. Bech, Gunnar (1955/57), Studitn iiber das dtutsche verbum infinitum, Det Kongelige Danske Akademie av Videnskabeme, Copenhagen. Beck, Sigrid (I996), 'Wh-constructions and transparent Logical Form', Dissertation, Univ. of Tiibingen. Bierwisch. Manfred (I997). "'Almost", "Again": A Speech for a Lexicalist Approach', handout, Sinn und Bedeu­ tung. 5-7 December I997• Berlin. Borer, Hagit (1988), 'On the morphological parallelism between compounds and constructs', in G. Booj & J. v. Marie (eds), Yearbook of Morphology 1, Foris, Dordrecht. Chomsky, Noam (I992), 'A minimalist program for linguistic theory', ART Owuional Papm in Linguistics 1, MIT Press, Cambridge, MA. Chomsky, Noam ( 1995), The Minimalist Program, MIT Press, Cambridge, MA.

204 Vl.Sibility Parameter for Functional Adverbs Keyser (eds), The View from Bui/Jing,

MIT Press, Cambridge, MA.

Halle, Morris & Marantz, Alec (I993), 'Distributed morphology and the pieces of inflection', in K. Hale & S. J. Keyser (eds), The View from Bui/Jing zo, MIT Press, Cambridge, MA. Heim, Irene (I994), 'Comments on Abusch's theory of tense', City, MIT, MS.

MA.

Kaspar, Walter (I987� 'Semantik des KonJunktivs I I in Deklarativsatzen des Deutschen', in H Henne, H Sitta, & E. Wiegand (eds), Reihe Germanistische Linguistik, Niemeyer, Tiibingen. Klein, Wolfgang (I994), Time in Language, Roudedge, London and New York. Kratzer, Angelika (I994), 'The event argu­ ment and the semantics of voice', UMass, Amherst, MS. Kratzer, Angelika (I998), 'More structural analogies between pronouns and tenses', SALT VIII, MIT, May I998. Kri£ka, Manfred (I989); Nominalrefermz urul Zeitkonstitution, Studien zur Theo­ retischen Linguistik, WUhelm Fink, Munich. McCawley, J. D. (I97I), 'Prelexical syntax', in O'Brien (ed.), Report of the zzrul Rourult4bk Meeting on Linguistics arul Language Studies, Georgetown University Press. Marantz, Alec (I984), On the Nature of Grammatical Relations, MIT Press, Cam., bridge, MA. Marantz, Alec (I997), 'No escape from syntax: don't try a morphological analy­ sis in the privacy of your own lexicon', Cambridge, MA, MIT, MS. Morgan. J. (I969), 'On arguing about semantics', Papers in Linguistics, I: 49-70. Musan. Renate (I998), 'The core semantics of the present perfect', ZAS Papers in 149-179· Linguistics, I I : I I 3-4S· Zimmermann; Th. E. (I993b), 'Zu Risiken und Nebenwirkungen von Bedeutung­ Parsons, Terence (I990), Events in the Semantics ofEnglish: A Study in Subatomic spostulaten', Linguistische Berichte, 146: Semantics, MIT Press, Cambridge MA. 263-82.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Jackendo£( Ray (I990), 'Semantic struc­ tures', in J. Keyser (ed.), Cu"ent Studies in Linguistics, MIT Press, Cambridge,

Partee, B. (I973), 'Some analogies between tenses and pronouns in English', Journal ofPhilosophy,' 70: 6oi-g. Paslawska, Alla (I998), 'Transparente Mor­ phologic und Semantik eines deutschen Negationsaffixes', Linguistische Berichte, I 7S= 3 S 3-8s. Rapp, Irene (I996), 'Zustand? Passiv?­ Oberlegungen zum sogenannten Zustandspassiv', Zeitschrift for Sprach­ wissenschaft, IS3: 23 I-6S. Rapp, Irene (I997), Partizipien urul seman­ tische Struktur, Studien zur deutschen Grammatik, Stauffenburg Verlag Bri­ gitte Narr GmbH, Tiibingen. Sadock, Jerrold M (I98 I}, 'Almost', in Radical Pragmatics, Academic Press, New York. Seuren, Picter (I984), 'The comparative revisited', journal ofSemantics, 3· Stechow, Arnim von (I984), 'Comparing semantic theories of comparison',journal ofSemantics, 3: I-n. Stechow, Arnim von (I99S). 'Lexical decomposition in syntax', in U. Egli. P. E. Pause, S. Christoph, A. von Stechow, & G. Wienold (eds), Lexical Knowkdge in the Organisation of Lan­ guage, Benjamins, Amsterdam and Philadelphia. Stechow, Arnim von (I996), The different readings of wieder "again": a structural account',joumal ofSemantics, I 3: 87-I38. Stechow, Arnim von (I997� 'Verben des Entstehens (und Vergehens?)', talk pre­ sented at the DGfS meeting, Diisseldor£, 24 February I997· MS. Stechow, Arnim von (I998), 'Participles I I in German', City, to appear in the Proceedings of the Bergamo Conference on Tense and Mood Selection. Vendler, Zeno (I9S7). 'Verbs and times', Philosophical Review, 66: I4J-6o. Zimmermann, Th. E. (I993a), 'On the proper treatment of opacity in certain verbs', Natural Language Semantics, I:

© Oxford University Press

]otmlid of&mantia 16: �os

1999

IN MEMORIAM

The Editors report with great regret the death of two members of the Editorial Board of the Journal of Semanti�:

Megumi Kameyama

James D. McCawley who died suddenly on I o April I 999· Both made a great contribution to the Journal as well as to the field of natural language semantics. Megumi and Jim were both lovable people, great scholars, and great friends, whom we miss very much.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

who died after many months of severe illness on 2 3 January I999, and