Pieces of predicate transfer. Slightly edited transcript of a talk presented in Toulouse Orin Percus, University of Nantes / LLING EA 3827 07.06.2012 and at NYU in late 2011. Comments welcome!

The phenomenon of predicate transfer receives its most thorough discussion in a 1995 article by

Geoffrey Nunberg, and it is fair to associate predicate transfer with Nunberg. As I read the literature,

the phenomenon was actually discovered by Ivan Sag in a 1981 article; at the same time, Sag’s

discussion was inspired by Nunberg’s work at around that time, that later Nunberg article provides a

lot more argument, and the literature on predicate transfer pretty much circulates around Nunberg’s

work. In what follows, I’m not going to go much further than that Nunberg article, though there is

some important later literature on the topic.1 Mostly what I’ll do will be spelling out in a little more

detail than Nunberg does the motivation for Nunberg’s general point of view, which I accept (though

there are some differences in the specifics), and looking at it through the glasses of someone interested

in compositional semantics, asking some questions that it raises.

The phenomenon of predicate transfer.

Predicate transfer is what we find in examples like (1)-(4). In all of these examples, there is a

predicate that is associated with a meaning different from the one that is in some sense its normal one.

Take the well-known example in (1a). Parked is an expression that is designed to apply only to

vehicles, and so when we construct the VP is parked out back out of its pieces, we expect to get a

predicate that has only vehicles in its domain, and that characterizes a vehicle when that vehicle is

parked out back. But in fact that isn’t what we have in a sentence like I am parked out back. What we

have is apparently a predicate with people in its domain, and it characterizes a person when the vehicle

that person owns is parked out back, or something like that2 – that is, the VP behaves like am someone

whose vehicle is parked out back. Predicate transfer is the process that results in a new meaning for

the VP here. (Below I will sometimes use a “T” to signal a meaning transfer.)

(1) a. I am parked out back. (Nunberg)

b. My friend here is parked out back.

x: x is a vehicle. x is parked out back

=>T (something like)

x: x owns a unique (relevant) vehicle. The vehicle x owns is parked out back.

cf. (1’) a. My vehicle is parked out back.

b. I am someone whose vehicle is parked out back.

In (2) and (3) you have other examples of predicate transfer resulting in a new meaning for a

VP. The VP in (2), were neatly tied, has as its normal sense a predicate with items in its domain that

are capable of being tied, like laces; it characterizes those things when they are neatly tied. But in

Billy’s shoes were neatly tied, what we have is something that can apply to a pair of shoes, and

characterizes the pair of shoes when the laces belonging to it are neatly tied – that is, the VP behaves

like are an item of clothing whose laces are neatly tied. In (3), called comes to mean something like

be an organization whose staff called.

(2) Billy’s shoes were neatly tied. (Nunberg)

x: x is capable of being tied. x is neatly tied.

=>T

x: x is a pair of shoes (or ...). x’s laces (or...) are neatly tied.

1 Notably by Kleiber 1999, Recanati 2004, 2009, and most recently Asher 2011.

2 See Kleiber (1999: 146-147) for interesting discussion of the right paraphrase.

2

cf. (2’) a. The shoelaces of Billy’s shoes were neatly tied.

b. Billy’s shoes are an item of clothing whose laces are neatly tied.

(3) The hospital called.

x: ... . (Members of) x called.

=>T (something like)

x: x is an organization. (Members of) x’s staff called.

cf. (3’) a. The hospital staff called.

b. The hospital is an organization whose staff called.

Then in (4) you have an example of predicate transfer resulting in a new meaning for an NP.

Or maybe I should say the example, because examples of this kind seem not so easy to come by, and

it’s this kind of example that everyone discusses.3 When a waiter tells the person preparing the food

“The ham sandwich at Table 7 is getting restless,” he is talking about the person at Table 7 who

ordered a ham sandwich. Here, the predicate ham sandwich is transferring to mean something like

person who ordered a ham sandwich.4

(4) The ham sandwich at Table 7 is getting restless. (ham sandwich examples also

originate with Nunberg, cf. Nunberg 1979)

x. x is a ham sandwich

=>T (something like)

x. x ordered a ham sandwich.

Cf. (4’) The person who ordered a ham sandwich at Table 7 is getting restless.

We repeat: predicate transfer.

I talked about all these examples as examples where a predicate has a meaning different from what we

would naturally imagine to be its inherent meaning. But of course you might question this. There are

other ways in which we could see what is going on. In particular, you might wonder whether, in I am

parked out back, it’s not the VP that is changing its meaning but the subject – I in some way comes to

contribute my car. And similarly with the other examples.

That’s not what is happening. This is why I said that Sag (or maybe Nunberg) discovered

predicate transfer – facts like these were not themselves new, but it was Sag and Nunberg who

suggested specifically that these examples involved a change in meaning for the predicate. I am now

going to summarize some arguments that indeed sentences like (1)-(4) involve an altered meaning for

the predicate. Of course, this doesn’t mean that all examples that have ever been analyzed as predicate

transfer are predicate transfer – since there usually are other conceivable analyses, one has to check

these things case by case.5 But these examples seem to involve predicate transfer.

6

3 Ward’s variants involve orderers too. Kleiber (1999: 124) adds a French example that does not

involve a restaurant scenario: L’hépatite du 3 ne mange toujours pas.

4 Again, one can debate whether this is the best paraphrase. Another candidate is: destinee of a ham

sandwich order. 5 Ward 2004 argues that there is a relatively productive use of copular sentences like I am the pad thai

on which neither the NP pad thai nor the containing predicate be the pad thai transfers in the way that

we will be seeing, but rather the copula transfers. On this use, is means something like is matched

with. More precisely, this use presupposes the existence of a one-to-one correspondence between two

3

Let’s consider the example I am parked out back or My friend here is parked out back. Here

specifically is why we can’t say that in these sentences the subject has a transferred meaning, that I

comes to mean my car.7

One strong argument comes from coordination facts involving the predicate. I can say very

sensibly that I am parked out back and in a hurry to get home ((5a)), expressing with this sentence that

my car is parked out back and that I am in a hurry to get home. If I were transferring to my car here,

then we would expect to be able to say the same thing with My car is parked out back and in a hurry

to get home ((5b)), but that sentence is crazy – it implies that my car is in a hurry to get home. So at

least in this case where we have parked out back, it looks rather as though parked out back is

transferring to a property of people.

(5) a. I am [ parked out back and in a hurry to get home ].

b. ?? My car is [ parked out back and in a hurry to get home ].

This was a negative argument: it shows that it can’t be that I in examples like (5a) is

transferring. There is also a positive argument based on coordinations that there is predicate transfer

in examples like I am parked out back. (This argument is essentially due to Nunberg.) When we look

at coordinations we find the following pattern. On the one hand, we can conjoin parked out back in

these kinds of sentences with a lot of predicates that we would use to describe people but not cars.

This is what happened in the case we just saw: I am parked out back and in a hurry to get home is

fine, while My car is in a hurry to get home is crazy ((6)) On the other hand, we can’t conjoin with a

lot of other predicates that we would use to describe cars but not people. For example, I am parked

out back and particularly shiny is just as crazy as I am particularly shiny, while My car is particularly

shiny is fine ((6’)). The argument is just that it is pretty simple to account for this pattern if what is

happening in I am parked out back is that the subject stays a person but parked out back transfers to

denote a property of people. The contrasts then come down to the fact that for some reason

particularly shiny can’t transfer to describe a property of people whose cars are particularly shiny (and

that the conjoined predicate parked out back and particularly shiny can’t transfer either, a possibly

related fact that I will bring up again at the end).

(6) a. ?? (I am parked out back and) my car is in a hurry to get home.

b. (I am parked out back and) I am in a hurry to get home.

c. I am [ parked out back and in a hurry to get home ].

sets – a first set that includes one of the two objects referred to, and a second set that includes the

other. We use the sentence to say that the two objects referred to constitute one pair in this mapping.

While one could arrive at this effect using the mechanics we will see for VP transfer, there might be

evidence that Ward’s move is the right one, given that it is hard to see on the former view why ?? I am

a ham sandwich should differ from I am the ham sandwich.

6 See the closing remarks of this document for a very different alternative view that one could entertain

and that, at least at first glance, also seems to be compatible with the facts I will be discussing. 7Another possibility to consider is that the VP has the same meaning in (1) and in (i) below, that is,

that [[ parked out back ]] is just x. x OR A VEHICLE BELONGING TO x is parked out back. This looks

unlikely: the examples in (ii) suggest that be parked... doesn’t have both people and cars in its domain

(and if they worked, one might still wonder if this meaning was due to transfer).

(i) A red Volvo is parked out back. (red Volvos due to Asher 2011)

(ii) a. ? The red Volvo and I / my friend here are both parked out back.

b. ? I / my friend here and a red Volvo are parked right next to each other.

cf. (ii’) a. The red Volvo and my friend’s car are both parked out back.

b. My friend’s car and a red Volvo are parked right next to each other.

4

(6’) a. (I am parked out back and) my car is particularly shiny.

b. ?? (I am parked out back and) I am particularly shiny.

c. ?? I am parked out back and (am) particularly shiny.

For another argument against transfer of the subject in parked out back sentences, consider

contrasts like those in (7). Of all these people, only I am parked out back is fine, Of all these cars,

only I am parked out back is bizarre. This is easily accounted for if I can’t denote my car: the of-

phrase just specifies the domain of quantification. If I can denote my car, we have a mystery, as in

other sentences there is no problem with of all these cars when the thing only attaches to denotes a car.

(7) a. ?? Of all these cars, only I am parked out back.

cf. a’. Of all these cars, only mine is parked out back.

b. Of all these people, only I am parked out back.

I think it is reasonable to conclude from these considerations that the VP in I am parked out

back has transferred to denote a property of people, a property that a person has when his car is parked

out back. And similar arguments can be given for the other cases of “VP transfer” -- examples (8) and

(10) below parallel (5), examples (9) and (11) parallel (6).

(8) a. Billy’s shoes were neatly tied but too wide for his feet.

b. ?? Billy’s shoelaces were neatly tied but too wide for his feet.

(9) a. ?? (Billy’s shoes were neatly tied but) his shoelaces were too wide for his feet.

b. (Billy’s shoes were neatly tied but) his shoes were too wide for his feet.

c. Billy’s shoes were neatly tied but too wide for his feet.

(9’) a. (Billy’s shoes were neatly tied but) his shoelaces were frayed.

b. ?? (Billy’s shoes were neatly tied but) they were frayed.

c. ?? Billy’s shoes were neatly tied but frayed. (Nunberg)

(10) a. The hospital called and closed down for the holiday just afterwards.

b. ?? The hospital staff called and closed down for the holiday just afterwards.

(11) a. ?? (The hospital called and) the hospital staff closed down for the holiday just afterwards.

b. (The hospital called and) it closed down for the holiday just afterwards.

c. The hospital called and closed down for the holiday just afterwards.

(11’) a. (The hospital called and) the hospital staff was pleased that you were feeling better.

b. ?? (The hospital called and) it was pleased that you were feeling better.

c. ?? The hospital called and was pleased that you were feeling better.

Now for “NP transfer.” In the case of examples like the ham sandwich sentence, there are also

arguments that one can give that the transfer is happening at the level of the NP ham sandwich, not at

the level of the whole subject, and definitely not at the level of the VP.

As far as the possibility of transferring the whole subject is concerned, the basic point is that at

Table 7 in (4) clearly identifies the location of the sandwich orderer rather than the location of any

sandwich. For example, our waiter could use (4) even in a situation in which he is aware that there are

no concrete ham sandwiches yet at Table 7 and in which there might never be any; what counts is that

the orderer is there8. If the whole subject were to transfer, though, the meanings of the pieces of the

ham sandwich at Table 7 would be combining in the normal way, and we would expect at Table 7 to

identify the location of a sandwich since it would be modifying a meaning for ham sandwich that

8 Similarly, it is inappropriate if the orderer is at Table 1 and asked for one of the ham sandwiches

lying on Table 7 (or asked for a ham sandwich to be served over at Table 7).

5

describes sandwiches. So it looks as though the whole subject is not transferring. By contrast, if the

NP ham sandwich transfers so as to contribute a property of people – a property that a person has

when he is a ham sandwich orderer -- then it makes perfect sense that we arrive at a description of a

ham sandwich orderer at Table 7, since at Table 7 will modify this property of people.

(For honesty’s sake, I should note that I simplified a bit in making this argument that the

whole subject does not transfer, but I don’t think that the simplifications affect the argument. The

simplification is this. Until now, I have been speaking as though the contribution of ham sandwich

before it undergoes predicate transfer would be a predicate that describes actual physical ham

sandwiches that could be on one table or another. In that case, at Table 7 in combining with this

predicate, would be identifying the location of actual concrete ham sandwiches. But in fact, when we

consider ordering contexts, this is not always the right way to view the contribution of ham sandwich.

When I say the sentences in (12), I am clearly not talking about actual physical ham sandwiches. The

examples in (12’) seem to show moreover that, even in cases where I am not talking about actual

physical ham sandwiches, ham sandwich is ambiguous: it is natural to think that the descriptions here

denote different kinds of things in (12’a) and in (12’b) – a “sandwich kind” versus, for lack of a better

term, “a sandwich order” -- and thus that ham sandwich contributes a different predicate in the two

examples. Still, I am pretty sure that, on the right analysis of these uses of ham sandwich, they do not

describe people, and in that case the argument stands: without any further transfer to a property of

people, we would not expect at Table 7 in our original example to be specifying the location of the

sandwich orderer, which it does. Remember the “sandwich order” reading, by the way – it will come

up again.)

(12) a. I ordered a ham sandwich (that was impossible to make).

b. [The ham sandwich that I ordered] never arrived.

(12’) In a situation in which John and Bill both said “a ham sandwich with pickles, please” (and in

which the sandwiches never arrived) both of the following can be considered true.

a “sandwich kind” a. [The ham sandwich John ordered] was the same as the one Bill ordered.

(If their requests were granted, they would be getting the same kind of sandwich.)

a “sandwich order” b. [The ham sandwich John ordered] was different from the one Bill ordered.

(If their requests were granted, they would be getting different sandwiches.)

As far as the possibility of transferring the VP is concerned, there are further arguments that in

these cases the VP continues to denote a property of people, not of sandwiches, and that the DP

denotes a person. You can already imagine the evidence that comes from coordination and sentences

with only:

(13) ?? The ham sandwich at Table 7 is getting restless and tastes funny.

(We can’t conjoin the VP with predicates that describe sandwiches but not people.)

(14) a. ?? Of all the sandwiches here, only the one at Table 7 is getting restless.

b. Of all the customers here, only the ham sandwich at Table 7 is getting restless.

For variety, consider the facts concerning pronominal agreement. (Again, the argument to follow is

essentially due to Nunberg, though I have tried to sharpen the original argument.) In general, what

happens when a DP is underdetermined with respect to features is that pronominal agreement is

determined by the meaning of the DP. So if you take (15d), for example, what form of the relative and

other pronoun we have here varies according to whether we understand the task organizer in question

to be a man, woman or machine. Once we recognize this, we see by considering examples like (15’)

that the DP in ham sandwich sentences denotes a person.

6

(15) a. My friend here is proud of himself/ herself.

b. My friend here, who is very proud of himself/ herself, is Norwegian.

c. My task organizer is very efficient but has a problem with his/ her/ its screen/ finger.

d. My task organizer, who/ which is very efficient, has a problem with his/ her/ its screen.

(15’) (Only) the ham sandwich at Table 7 is still waiting for his/ her/ *its bill.

(In fact, if we didn’t already have evidence that our examples in (1) involve a transferred VP rather

than a transferred subject, we could build on this conclusion to show that. Parked out back sentences

pattern differently with respect to pronominal agreement from ham sandwich sentences. In sentences

like (16a,b), variants of our parked out back sentence, the pronominal agreement facts show that the

subject remains a human being rather than transferring to a car.9)

(16) a. My friend here is parked out back and has a problem with her/ *its wheels.

b. My friend here, who/ *which has a problem with her/ *its wheels, is parked out back.

cf. c. This car, which has a problem with its wheels, is parked out back.

Who cares?

Clearly, the phenomenon I have been talking about is not so productive, and very conventionalized.

The ways in which we transfer predicates are very limited, and probably the issue of what transfers we

have is somewhat like the issue of what words we have. As far as my own grammar of the moment is

concerned, my intuitions are that it is quite natural to talk about myself to talk about my car and where

it is parked (I am parked out back, addressed to the garage employee), less usual but not ridiculous to

talk about myself to talk about my house and its problems (I’m infested with termites or I’m leaking

just below the roof, addressed to someone competent with these things), and quite odd to talk about

myself to talk about my house and when it is built (I was built in 1920, addressed to the appropriate

kind of administrator). At the same time, I feel that, even in the odd cases, this is something that I

would probably do if I saw that other people did it.

Does the fact that predicate transfer is limited and conventionalized mean that we shouldn’t

talk about it? No, obviously not. When we have predicate transfer, we are presumably using

mechanisms that our linguistic competence makes available, putting together various ingredients to

arrive at the final product, even if our choice of when to use these mechanisms is determined by

convention. As Nunberg writes:

“What is conventionalized is not the mechanism involved, but only the particular

semantic domains the mechanism is allowed to operate on.” (Nunberg 1995: 119)

So it is interesting to ask: what exactly are we doing when we effect predicate transfer? I

would say that this is like looking at any other construction to inform ourselves about what kinds of

things we would like our theories of syntax and semantics to account for. To ignore predicate transfer

would be a little like ignoring the syntax and semantics of word formation because we often have to

learn what prefixes go together with what roots.

So that is a little apology for what I am going to do now, speculate about what pieces go into

predicate transfer. It applies much more generally though: even if some constructions are less

productive and more conventionalized than others and we don’t know why, this is no reason to

disregard them.

9 As a whole, then, these facts suggest that we can use agreement data as a diagnostic for whether

transfer of a DP meaning has taken place, as Nunberg supposes without real argument -- at least in

cases where the DP is underdetermined with respect to its features.

7

Now, the fact that these constructions are conventionalized does mean that it’s harder to draw

conclusions. For example, we would like to know what hard constraints there are on predicate

transfer, what kinds of new predicates we simply can’t create on the basis of an old one, but it’s hard

to determine this if the examples of predicate transfer that exist need to be learned. And it’s not

obvious that the examples that exist are all derived in the same way. In what follows, I will just start

from the few examples I have considered so far. And I will imagine that all of these examples have

something in common as far as their analysis. It could be that I am mixing apples and oranges,

especially by imagining NP transfer as similar to VP transfer – as I said, NP transfer seems a lot more

specialized, and we will see another difference. But so it goes.

Predicate transfer probably has a syntax.

How does predicate transfer happen? The initial observation is that probably there is some syntax

involved.

You will think immediately that there is syntax involved if you are the kind of person who

believes that purely interpretive operations that switch from one meaning to another should apply at

the level of the lexicon. (This is a matter of taste, I guess, but please read on even if you aren’t, as I

will come back later to some things I mention now.) Because suppose we tried to say that, in the case

of predicate transfer, some purely interpretive operation switches us from one meaning to another

while interpreting the sentence. Then we would be forced to say that sometimes this operation applies

at the level of a complex constituent, or alternatively that it operates on a trace or another variable, that

is, something without inherent “lexical content.”

To see that we might want it to occur at the level of a complex constituent, think of an

example like The [ham sandwich with mustard] at Table 7 is getting restless. Or, for that matter, I am

[ parked [out back] ] where out back does not describe the location of the person but rather the car and

so where it’s hard to see how to arrive at the right result by saying that parked is transferred alone in

such a way as to apply to people. To see that we could also explain predicate transfer by saying that

the operation occurs at the level of a trace, look what I did in (17c) to arrive at the transferred meaning

of parked out back. I imagined there that the value of a trace with respect to an assignment magically

changes to the object that is the unique relevant vehicle owned by the value of the trace. If this

happens to the trace that is the phrase-internal subject of parked out back, then, when we attach a

binder for the trace, we will get our transferred predicate.

(17) a. I am parked out back.

b. [[ parked out back ]] : x: x is a vehicle. x is parked out back

=>T

x: x owns a unique (relevant) vehicle. The vehicle x owns is parked out back.

P =>T x: There is a unique (relevant) object in dom(P) that x owns.

P(the unique object in dom(P) that x owns) = 1.

c. [[ t1 ]]g : g(1) =>T the unique (relevant) vehicle that g(1) owns (undefined if there is none)

z =>T the unique (relevant) vehicle that z owns

[[ Tt1 parked out back ]]

g (XP-internal subject)

is defined only if there is a unique (relevant) vehicle that g(1) owns.

Where defined: 1 iff the unique (relevant) vehicle that g(1) owns is parked out back.

8

[[ 1 Tt1 parked out back ]]

g

= x: There is a unique (relevant) vehicle that x owns.

The unique (relevant) vehicle that x owns is parked out back.

If you want to see other examples of cases where transfer would have to apply to a locally free

variable or a constituent that contains a locally free variable, you can look at the examples in (18)-(20),

which are cute and drawn from the literature. In the case of (19), for example, a simple way of

arriving at the transferred meaning of squeeze himself into a narrow space, on which it describes

individuals who squeezed their vehicles into a narrow space, would be to alter the meaning of the trace

in the way we did in (17c). A more complicated way would be to raise himself creating a predicate of

vehicles, and to alter that predicate.

(18) a. Everyone is parked next to his house.

x: x owns a unique (relevant) vehicle. the vehicle x owns is parked next to x’s house

b. Everyone [ 1 Tt1 parked next to his1 house ] Transfer as in (17c).

c. Everyone [ 1 t1 T[ parked next to his1 house ] ] Transfer as in (17b).

(19) a. Ringo (/ Everyone) squeezed himself into a narrow space. (Jackendoff)

x: x owns a unique (relevant) vehicle. x squeezes the vehicle x owns into...

b. Everyone [ 1 t1 squeezed Thimself1 into ... ] Transfer as in (17c).

c. Everyone [ 1 himself1 T[ 2 t1 squeezed t2 into ... ] ] Transfer as in (17b).

(20) Norman Mailer (/Everyone here) likes to read himself before going to sleep. (Fauconnier)

x: x is an author. x likes to read x’s work before going to sleep.

The real indication that predicate transfer involves some additional syntactic ingredients

comes from clues that point to the existence of a constituent with the untransferred meaning. In that

case, it looks as though we have a layer of structure that gives us the transferred meaning on top of a

layer of structure that gives us the normal untransferred meaning.

When it comes to “VP transfer,” the clue comes from sentences with ellipsis like (21). “I was

parked out back and that red Volvo was too,” I say to the garage employee. Most of us10

would

conclude from this that, somewhere around, there is a constituent that has the meaning of parked out

back, a predicate that applies to vehicles.11

(21) I was parked out back and that red Volvo was too. (based on Asher)

I was [ T [ parked out back ] ]

When it comes to “NP transfer,” we find a clue in another kind of anaphora that needs a syntactic

antecedent – maybe (the argument here is slipperier). Here I have in mind E-type-ish anaphora. To

see what I mean, start with the classic kind of contrast in (22). It’s been argued that what is

responsible for this contrast is that the anaphor we have here spells out a definite description that

includes material whose meaning is recoverable from the local context -- in (22a) we have woman, but

10

Maybe not all – see Asher 2011: 68, fn. 10.

11

To the extent that sentences like (ii a), fn. 7 are acceptable, one might speculate that this too is due to

a more complicated structure with ellipsis.

9

in (22b) we don’t. For concreteness (current analyses are more sophisticated, but this will do here),

we might say that in (22b) we copy the local material that we get by QR-ing a woman, and then leave

the determiner uninterpreted, arriving at something that means woman he is married to. It is important

to mention at this point that we find similar phenomena in examples like (22’) where the pronoun is it

rather than the usual “agreement” form that we find with the fully spelled out DP, so presumably we

would like a similar analysis for those. (This doesn’t correspond to the usual kinds of E-type anaphora

considered, that’s why I said “E-type-ish”.) With this in mind, now look at (23), from Ward’s 2004

article. Whether the it here is like the she of our first example or the it of our second, the diagnostic

we just established arguably tells us that, in examples like this, where it describes an actual filet

mignon, we have a constituent around with the untransferred meaning of filet mignon, a predicate that

characterizes filet(s) mignon(s).12

(22) a. Every man who is married to a woman thinks initially that she is the best woman he is

going to find.

b. # Every husband thinks initially that she is the best woman he is going to find.

(23) A copying analysis: (roughly Heim 1990)

[ Every [ man [ (who) 1 [ a woman [ 2 t1 is married to t2 ] ] ] ] ]

[ 1 t1 thinks that [ the [ a woman [ 2 t1 is married to t2 ] ] ] is ... ]

(22’) a. Every man who is married to a woman thinks initially that it is the best woman he is

going to find.

b. ?? Every husband thinks initially that it is the best woman he is going to find.

(24) Every filet mignon I’ve waited on tonight has said it was the best steak they had ever

eaten. (Ward 2004)

But what syntax?

So if there is a syntax to predicate transfer, what is it? What I said a little earlier suggests two

different directions, which I sketched in (25). Maybe there is something that combines with the trace

that serves as the phrase-internal subject of the predicate. (That’s Option 1.) Or maybe there is

something that combines with the predicate itself. (That’s Option 2.) Is there a reason to choose one

over the other?

(25) a. I am parked out back.

b. [ 1 [ t1 ’S-VEHICLE ] parked out back ] Option 1.

c. [ BE-SOMEONE-WHOSE-VEHICLE-IS-A-THING-THAT [parked out back] ] Option 2.

12

Not all such examples work : consider (i). Arguably, the oddness of (i) is due to an independent

constraint that concerns certain kinds of anaphors that recover descriptive material, and that disallows

anaphora in cases where the anaphor’s antecedent is too close somehow. We thus find contrasts like

(ii), for example. Possibly this constraint concerns “clitic-like” anaphors that get deaccented --

Francis Cornish noted that we can rescue examples like (i) by using that instead of it ((i’)).

(i) # Every filet mignon loved it.

(ii) a. # Every donkey loves one.

b. Every donkey loves another one.

(i’) Every filet mignon loved that.

10

I think that there probably is, and that we should choose Option 2 over Option 1 (which is

Nunberg’s view, in fact). And to indicate some of the considerations behind the choice I will make, I

want to consider a particular analysis along Option 1 lines, and to explain my misgivings about it.

A straw man along Option 1 lines.

The particular analysis I want to consider is exemplified in (26). It’s really the first thing that comes

to mind if we want to entertain an analysis where something combines with the trace. The idea is that

what combines with the trace is an item that denotes a function -- what function is determined in some

way by the context. So in the case of parked out back, for example, we would wind up with a

predicate that holds of x when f(x) is parked out back; f here might be a function that takes us from

owners of vehicles to vehicles they own.

(26) [ 1 [ F t1 ] parked out back ]

[[ F ]] = z: z is the owner of a unique (relevant) vehicle. the unique (relevant) vehicle z owns

Now here are my doubts. First of all, I think that there are problems with the syntactic side,

the idea that we are attaching the function to the trace. If we can attach a function to a trace, then why

can’t we attach it to the subject, which would be of the same type ((27)) ? And we can’t, apparently,

because we have seen that, in I am parked out back, I does not transfer to denote my car.

(27) * [ F I ] [ 1 t1 am parked out back ]

But beyond this, I have doubts about what this analysis claims about interpretation. This analysis says

that we create out of a predicate P a new predicate that holds of x when P holds of f(x), where f is a

function that in some way we recover from the context. And I’m not so happy about that either.

First consider the VP transfer cases. At first glance, this idea seems to be appropriate in the

sense that, in each case, there is a “natural” function that seems to do the job – natural in the sense that

there is a clear criterion for identifying the objects in the domain of the function and there is a simple

general description that we can give of how the function would work ((28)-(30)). For example, in a

situation where I am at a party and explaining where my car is, the function that comes to mind is one

whose domain is the set of people at the party who came in a car (or perhaps a car that they own) and

that maps these people to the car that they took to the party. I have the impression all the same that

this misses something, though, and that is that the functions that we use apparently have to be one-to-

one: two owners, the same car, and this kind of construction doesn’t work so well anymore. To

continue with the example: if my wife and I jointly own our car and came to the party in that car, I

wouldn’t use I am parked out back and My wife is parked out back interchangeably to express that the

car is parked out back. I don’t think I would use either one, as there is no natural way of

circumscribing the domain that includes one of us and excludes the other. I would use We are parked

out back, invoking a domain whose members are groups of people who together came in a car.13

13

Anna Szabolcsi pointed out that in principle it could follow from details of the workings of our

“pragmatic competence” that, of the various functions from people to vehicles, only the one-to-one

functions can be considered salient in the kind of situation I have described. The argument here

supposes that this is not the case, and thus depends in a certain sense on an idea about pragmatics.

One can criticize the argument on this basis: the argument is truly convincing only to the extent that it

is backed up by a theory of pragmatics which I am not providing. Similar criticisms can be levelled at

some of my later arguments.

11

(28) a. I am parked out back

b. [ 1 [ F t1 ] parked out back ]

[[ F ]]: John the car of John’s that he took to get to the party

Mary the car of Mary’s that she took to get to the party

...

z: z came to the party in a vehicle that z owns.

the vehicle z owns in which z came to the party

(29) a. Billy’s shoes are neatly tied (30) a. The hospital phoned.

b. 1 [ [ F t1 ] neatly tied ] b. 1 [ [ F t1 ] phoned ]

[[ F ]]: Billy’s shoes the laces of Billy’s shoes [[ F ]]: The hospital the staff of the hospital

Ilea’s shoes the laces of Ilea’s shoes The real estate agency its staff

... ...

z: z is a pair of shoes with laces. the laces of x z: z is an institution. the staff of z

Then, when it comes to NP transfer, this idea seems even less plausible. On the one hand, it is

hard to imagine a “natural” function that will do the job. As a case in point, look at the situation that I

sketched in the box below, where I have indicated customers and their orders, and imagine a waiter

saying (31a). If we want to maintain the view I sketched in the face of (31a), it looks like we have to

say that we allow a function that maps John and Bill to their sandwiches (or sandwich orders14

) but

Sally to her drink (or drink order). But how could we give a simple uniform description of a function

that does that?15

To make matters worse, look at (31b) and (31c). The function that we seem to need

to be salient for (31b), one that maps John to a ham sandwich but Bill to his mango lassi, we seem to

need to disregard when we change the example to (31c), for our waiter couldn’t use that to say that

John is getting restless. So something is funny. And finally, I wish to note that this approach does not

naturally capture our original intuition, that in ham sandwich sentences, ham sandwich is a predicate

that characterizes the orderers of ham sandwiches ((32b)). In order for the meaning that we derive

((32c)) to approach the paraphrase, we must be limited in our choice of [[F]] when we use the

sentence: we must be obliged to choose an [[F]] such that, for any x in its domain, if x ordered a ham

sandwich, then [[F]] yields for x a ham sandwich that x ordered. It is unclear why our choice would

be limited in this way, and in any event, as we just saw, that is not the kind of [[F]] that we seem to

need in order to account for (31b).

(31) a. The ham sandwich at Table 7 is a lot more restless than the one at Table 8, or, for that matter,

the Cabernet Sauvignon.

a’. Every ham sandwich is delighted with his order.

b. The mango lassi is getting restless. So is the ham sandwich at Table 7, and the roast beef with

french fries is threatening to leave without paying.

c. ?? ... So is the ham sandwich.

14

On the assumption – which I will continue to make use of -- that ham sandwich can contribute a

predicate of “sandwich orders” in the sense I alluded to earlier.

15

Here and below, I crucially assume that the waiter would be making use of the same function for all

instances of NP transfer in the sentences that he utters. This follows from some more general

assumptions that look reasonable: in examples like these, the functions that we consider salient when it

comes to interpreting the expressions that make reference to drinks are the same as those we consider

salient when it comes to interpreting the expressions that make reference to main dishes; in (31b,c),

whatever function is salient at the point when we interpret the first sentence remains salient when we

interpret the second. One could question these assumptions, however.

12

1 [ [ F t1 ] ham s./ mango l. ]

Table 7: John ham sandwich

Coca-Cola [[Fa]] : John h.s.

Bill h.s.

Table 8: Bill ham sandwich Sally C.S.

extra-large mango lassi

[[Fb,*c]]: John h.s.

Table 9: Sally roast beef with french fries Bill m.l.

Cabernet Sauvignon Sally r.b. with f.f.

(32) a. The ham sandwich at Table 7 is getting restless.

b. Our original paraphrase: x. x ordered a ham sandwich.

c. [[ 1 [ [ F t1 ] ham sandwich ] ]] = x: x dom([[F]]). [[F]](x) is a ham sandwich (order).

So what is the upshot of this discussion? First of all, that we don’t want an approach that

posits adjustments to a trace. But also more generally that we don’t want an approach that says that

we create out of a predicate P a new predicate that holds of x if a recoverable function F maps x to

something satisfying P. Maybe I was a little hasty (see the footnotes), but that’s my conclusion.

The semantics of predicate transfer.

So we are now heading in an Option 2 direction. Some additional structure around the original

predicate results in a new meaning. But what kind of new meaning precisely?

As far as I can see, we don’t need to give up on the idea that predicate transfer makes

reference to a salient function. We just need to give up on the idea that the function gets used in the

way we imagined on our Option 1 story. Suppose we take seriously the idea that the mechanisms for

predicate transfer derive for ham sandwich the kind of meaning we originally said, a predicate that

characterizes the orderers of a ham sandwich. Then here is a way of deriving this. (And in what

follows, I will crucially assume that ham sandwich can contribute a predicate of “sandwich orders” in

the sense I introduced earlier.)

We still say that predicate transfer involves reference to a salient function f. But predicate

transfer doesn’t create out of a predicate P a new predicate that holds of x when f takes x to something

satisfying P. Rather it holds of x when f takes something satisfying P to x. In the ham sandwich

example, there is such a salient function: a function from items that got ordered to their orderers. And

to say that this function takes a ham sandwich to x is to say that x ordered a ham sandwich. The

diagram in the box below should make this clear. Here I show you what this function does, with

reference to the restaurant scenario we considered, and if we look at the elements of the domain that

are ham sandwich orders, and follow the mapping, we find John and Bill, the two sandwich orderers.16

16

Note that a mapping like this is a function only if we consider John’s ham sandwich order to be a

different item from Bill’s ham sandwich order. But (12’b) showed that that is just the way sandwich

orders work (at least as long as John and Bill didn’t communicate that they wanted the very same

sandwich).

13

F: z: z is an item that a customer ordered. the customer who ordered z.

Fa: the h.s. that John ordered John

the Coca-Cola that John ordered John the ham sandwich orderers

the h.s. that Bill ordered Bill

the mango lassi that Bill ordered Bill

the roast beef that Sally ordered Sally

the Cabernet that Sally ordered Sally

So what we see here is that, when it comes to NP transfer cases, it might indeed be correct to

make reference to a salient function. So we can pursue the idea that the structure around the original

predicate contains an element that denotes a function. But the way this function gets used is that the

transferred predicate as a whole describes the value of that function rather than its argument.

Specifically, in the case of NP transfer, we create a new predicate that holds of x if x is the value of the

function for some item that satisfies the original predicate.

(33) a. The ham sandwich at Table 7 is getting restless.

b. What we had before:

[[Tham sandwich]] describes x if a certain salient function f maps

x to something that [[ham sandwich]] describes.

c. An alternative that also makes reference to a salient function:

[[Tham sandwich]] describes x if a certain salient function f maps

something that [[ham sandwich]] describes to x.

(34) First sketch of NP transfer

a. The ham sandwich at Table 7 is getting restless.

b. [ ... F ... [ ham sandwich ] ]

c. [[ (34b) ]] = x. There is some y such that [[F]](y) = x and y is a ham sandwich (order).

Of course, we can also imagine our VP transfer cases in this way. Since the functions that we

considered before were all one-to-one, we can say that there too the transferred predicate describes the

value of the function rather than the argument, but that the relevant functions in these cases are

mappings that go in the opposite direction, from cars to owners. The constituent that attaches to the

original predicate must be a little different, though, in that it seems to guarantee that the function does

not map two arguments to the same value, at least in the cases of interest to us. For example, it’s odd

to say I am parked out back but I am also parked down the street to say that I am the owner of a car

that is parked out back but I am also the owner of a car that is parked down the street. One way to

arrive at this effect is to say that, in the case of VP transfer, the additional structure around the

predicate creates a predicate that has x in its domain only if x is the value of the function for just one

item, and this transferred predicate holds of the unique item that the function takes to x ((35c’)).

(35) First sketch of “VP” transfer

a. I am parked out back.

b. [ ... F ... [ parked out back ] ]

14

c. One possibility:

[[ (35b) ]] is defined only if [[F]] is one-to-one.

When this condition is met, [[ (35b) ]] = x. The unique y s.t. [[F]](y) = x is parked out back.

[[F]]: z: z is a vehicle driven to the party by its owner. the owner of z.

the car of John’s that he took to get to the party John

the car of Bill’s that he took to get to the party Bill

...

c’. A variant:

[[ (35b) ]] = x: There is a unique y such that [[F]](y) = x.

The unique y s.t. [[F]](y) = x is parked out back.

(The variant doesn’t require the function to be one-to-one but insures that we will only be

concerned with the part of it that is one-to-one.)

Summarizing a bit, the facts we have looked at seem to point in the following direction (very

close to Nunberg’s position). We arrive at a transferred predicate by embedding the original predicate

in a larger structure, one of whose pieces is semantically a function – a function from cars to their

owners, for example, or a function from food orders to their orderers. And the transferred predicate

behaves in the following way: to say that something satisfies the transferred predicate is (i) to imagine

this thing as the value of that function, and (ii) to say that the original predicate applies to elements

that the function maps to that value. (VP transfer cases impose in addition the requirement that there

is only one such element.) And concretely, when it comes to the cases we have been looking at, the

transferred NP ham sandwich has a meaning as in (36a), while the transferred VP (be) parked out back

has a meaning as in (36b).

(36) Some meanings we have seen for transferred predicates

a. x. Some y such that [[F]](y) = x is a ham sandwich.

b. x: ...The unique y such that [[F]](y) = x is parked out back.

The syntax of predicate transfer (?)

Now we can finally move towards the question: what does that structure around the original predicate

look like? As I said, the use of predicate transfer is conventionalized, but the syntactic and semantic

pieced that go into creating it are pieces that our linguistic competence makes available and so pieces

that are presumably found elsewhere.

Once we look at things this way, we see what the ingredients might be that come together to

derive these meanings. Stare a little bit at the meanings in (36) and probably you will not be able to

resist seeing them as built up out of pieces of meaning in the way that I sketched in (37). After all, we

have seen most of these pieces before – the meanings of some and the, the meanings of the original

predicates, binders.

15

(37)

x

[[some]] y. [[F]](y) = x [[ ham sandwich ]]

[[the]] [[ parked out back ]]

x. Some y such that [[F]](y) = x is a ham sandwich.

x:... The y such that [[F]](y) = x is parked out back.

And this in turn suggests that the structure underlying predicate transfer is roughly of the kind in (38a),

with an unpronounced some or the, and a constituent that introduces the function that itself contains a

variable bound at the top. (I’ve included the variant in (38b) as well, since, if you are like Heim and

Kratzer and you think that binders come into existence via movement, then that variable must be a

trace of an uninterpreted element that moves to the top of the constituent and which we can identify

with PRO.)

(38) a.

SOME F... x1... ham sandwich

THE parked out back

or maybe

b. (cf Heim and Kratzer 1998)

PRO 1

SOME F... t1... ham sandwich

THE parked out back

Most of the pieces here are pretty recognizable, but you might have some difficulty with the

middle piece -- the one that introduces the function and the other object, let’s call it an internal

argument, and characterizes those things that the function maps to the internal argument.

y. [[F]](y) = [[a]]

...F...a...

So I would like to say a little more about that middle piece. I believe that it’s not unfamiliar at all.

There are other places where implicit functional relations play a role, and where it looks as though we

have a constituent just like that.

One place where we find them is with genitives when genitives are used predicatively, as in

That book is John’s or That’s John’s book! (when we assume that John has lots of books). In these

16

cases we are saying of a book that it bears a certain relation to John and to no one else. To see that the

sentence That book is John’s associates a book uniquely with John, note that a sentence like That book

is also Mary’s seems to contradict it, as shown in (39a). (As for the analysis of a sentence like That

book is also Mary’s, I would say that it should be viewed not as expressing two different associations

of the book with two different individuals, but rather as expressing one association of the book which

maps the book to the pair of Mary and someone else. The sentence says that the book is jointly

owned.) Or to take another example, imagine a professor who says That student is mine. He seems

rather possessive, unlike the professor who says My student came to see me, who doesn’t seem to

exclude co-advising. So it looks as though we have a constituent of just the kind we have seen where

the possessor, John, is treated as the value of the implicit function.

(39) Predicative genitives

a. That book is John’s!

i. – ?? Yes, and it’s also Mary’s.

ii. – √ No, it’s also Mary’s.

b. That’s John’s book!

[[John’s book]] = y. y is a book and [[F]](y) = John

...F...John... book

y. [[F]](y) = John y. y is a book

Another connection that I would like to emphasize is with the for-phrases that we find in For

every A there is a B sentences as analyzed by Rothstein. Take the sentence For every drop of rain that

falls, there is a flower that blooms. This sentence conveys that there are at least as many flowers that

bloom as there are drops of rain that fall. How does this happen? Rothstein argues that we have this

effect because for introduces an implicit function. The sentence says basically that, for every drop of

rain that you show me, I can find you a flower that blooms that this function maps to that raindrop.

Since a function can’t map the same flower to two different raindrops, this means that there are at least

as many flowers as raindrops. Now what precisely is the structure underlying this? Look at (41),

inspired by Rothstein’s analysis. The idea there is that the for-phrase is a constituent just like the one

we had earlier; every drop of rain starts out as its internal argument and QRs. What I’ve drawn in

(41c) is what you have just below the QR-ed quantifier. At a certain point in the structure, there is an

existential quantifier associated with the there-construction (about which there is more to say), but

basically, what you see is a structure that reflects quite transparently the way we would describe the

predicate that we are using to characterize every drop of rain.

(40) For-phrases in “Boolos sentences” (Rothstein 1995)

a. For every drop of rain that falls, there is a flower that blooms.

Paraphrase: There are at least as many flowers that bloom as there are drops of rain that fall.

b. For every Likud voter, there was one who voted Labour. (Rothstein 1995: 29)

Paraphrase: There were at least as many Labour voters as there were Likud voters.

17

(41) a. Every drop of rain [1 there is a flower that blooms for t1 ]

b. x. There is some y such that y is a flower that blooms and f(y) = x. (Rothstein)

c.

1 …

x.

isSOME (a) flower that blooms for F t1

[[some]] y. y is a fl. that blooms y. [[F]](y) = x

My suggestion, then, is that that less recognizable constituent in our structure is essentially just

a for-phrase. Why isn’t it pronounced? It isn’t in the same position where we would normally find a

pronounced for-phrase, that could have something to do with it. And if indeed its internal argument is

PRO, that could have something to do with it too. Besides, if Rothstein is correct, then there are a lot

of other places where we find a constituent that does what a for-phrase does, introducing a mapping

function and an argument, but where for is not pronounced. In fact, the majority of her paper is

devoted to these (cf. ((42)). So the fact that we have an unpronounced for-phrase in the case of

predicate transfer would in any event not be unusual. (There is one difference between Rothstein’s

examples and our examples that I wish to point out. In Rothstein’s examples, the function is at least

sometimes existentially quantified at the top. I don’t think it makes sense to say that the function that

maps Labour voters to Likud voters in (40b) is salient or recoverable in some way.)

(42) a. Mary opens the door every time the bell rings. (Rothstein 1995: 21)

b. Every time the bell rings [1 Mary opens the door for t1 ]

c. e. There is some e’ such that e’ is an event of Mary opening the door and f(e’) = e.

So, to summarize, I suggest that predicate transfer is built of the ingredients we see in (38),

where one of them is basically a for-phrase. Maybe not all of these are syntactically realized, but

some of them are, if we accept the argument from the ellipsis examples is that there is something other

than parked out back that makes it into a property of people. I suppose that one could take the

position that the ingredients that contribute definiteness and indefiniteness, the silent some and the, are

not syntactically realized. This is the position that you might be led to if you thought that the syntactic

units of natural language that behave this way are determiners, since the positions in which I have put

these ingredients don’t seem to correspond to determiner positions. In that case, what we really have

is a structure more like (43), and type shifting of just the kind that has been hypothesized by some for

determinerless languages. Indeed, depending on your other assumptions, you could take this

discussion of predicate transfer as motivating the claim that natural language makes available type-

shifting operations of this kind. (In (44) I have given myself as an exercise the task of accounting for

the Ward “E-type-ish anaphora” example on the type-shifting analysis, assuming that we resolve the

anaphor by copying a local constituent. It’s straightforward. It’s equally straightforward without

type-shifting, if we are willing to entertain analyses like (23): copying a determiner and leaving it

uninterpreted would yield the same result.)

18

(43)

for F t1 ham sandwich

y. [[F]](y) = g(1) y. y is a ham sandwich

type-shifting the unique y s.t. [[F]](y) = g(1) P. There is some y such that [[F]](y) = g(1) and P(y) = 1.

(44) a. Every filet mignon I’ve waited on tonight has said it was the best steak they had ever eaten.

b. every [ [ 1 [ [for F t1 ] filet mignon ] ] ... ]

[ 1 t1 has said that the was the best steak they1 had ever eaten ]

c. x:... x has said that the unique y s.t. [[F]](y) = x and y is a filet mignon was the best steak x

had ever eaten.

d. x:... x has said that the unique filet mignon that x ordered was the best steak x

had ever eaten.

Issues to investigate.

This was a first attempt but it opens lots of questions. A number of them concern the syntax.

To begin with, there seem to be syntactic constraints on predicate transfer. As things stand

now, we haven’t explained the original facts that motivated predicate transfer. Apparently, we don’t

like to apply predicate transfer to the coordination when we can apply the same operation to one of the

coordinates:

(45) a. ?? I am parked out back and particularly shiny. (= (6c))

b. * I am T[ parked out back and particularly shiny ]

But this doesn’t follow from anything. Why should this be? We find a similar constraint, I think,

when we look at our NP cases:

(46) a. The ham sandwich at Table 7 is getting restless.

(?? Situation: There are different kinds of ham sandwiches on different tables. John comes in

and says, “I would like one of the ham sandwiches at Table 7.” He waits for it at the counter.)

b. * The T[ ham sandwich at Table 7 ] is getting restless.

One possibility is that this is connected to the conventionalized nature of predicate transfer: we learn

something about the minimal units to which a transfer can apply, and then we don’t generalize. But I

think things are probably not so simple. The reason I think so is that we do generalize at least a little

bit. There is a productive side to predicate transfer, in that elements of the same semantic class seem

to transfer in the same ways. It isn’t only that parked out back can be transferred to contribute a

property of people -- all sorts of predicates that apply to conveyances can be transferred to contribute a

property of people. We can construct examples like I just got hit in the fender (based on Jackendoff), I

19

was rolling/ trotting/ galloping down Main Street at n miles an hour, and so forth and so on. So why

can’t we transfer a conjoined predicate that describes conveyances to contribute a property of people?

Similarly, we not only have parked out back transferring, we also have parked in front of the house,

parked by the supermarket, just parked, and so on transferring. In the terms of my discussion here, it

seems that, once we learn that there is a particular function that we can make use of in one context, we

feel pretty free about using it in other contexts as well. So it looks as though there is something

special about coordinations and some other complex constructions that blocks transfer, and this seems

to me to deserve investigation.

Another concrete syntactic question concerns the details of the structures I’ve been giving.

Before detailing the structures, I made a big deal of the fact that we didn’t want transfer of parked out

back to result from the attachment of something to a subject trace, because then we could allow the

possibility of transferring the subject, contrary to fact. But what I have proposed is something that

brings the possibility right back: we could attach our for-phrase to the subject, and THE to that, and

then we generate the wrong thing again. This was not very honest on my part. We now need

something to prevent that possibility. What? Could this just be syntax? Would attaching this material

to the subject mean that we would end up with a syntactic category that syntax doesn’t like in that

position?

I mentioned the risk of putting apples and oranges together in treating “NP transfer” cases and

“VP transfer” cases as basically alike. NP transfer seems to be a lot more specialized than VP transfer,

so we might wonder if the same kind of thing is really going on. This is not the only difference. My

impression is that, when it comes to VP transfer, only the “definite shift” is possible. For example (I

gave another one before), if I owned two cars both parked in a garage, I think it would be strange to

say I am parked on the second floor and I am also parked on the third floor to say that one of my cars

is parked on the second floor and another is parked on the third floor. But if the “existential shift” is

available, that is a surprise: the existential shift should allow us to use this to say that I am the owner

(or something like that) of a car on the second floor and I am also the owner (or...) of a car on the third

floor.17

But if the existential shift is not available, then we have another question: why not? I should

note here that Nunberg himself thought that existential VP transfer was available and used examples

like (48a) to show this. It seems to me that the existential force here is very likely due to something

else, the existential quantification over realizations of the subject that one frequently finds in sentences

with locatives where the subject is a mass term.18

One could say here that transfer yields a sentence

with the meaning of (48b) which also contains existential quantification (cf. (48c))

(48) a. I am in the Whitney. (Nunberg 1995: 113)

b. My work is in the Whitney.

c. Water is on the table.

Then there are other questions. What functions can we use in predicate transfer and what

functions can’t we use? As I said, predicate transfer has a conventionalized aspect to it so this is a

difficult question to investigate, but presumably there are constraints, and accordingly some potential

uses of predicate transfer will be possible and others will not. Constraints have been identified in the

case of the predicative genitive (Storto), and given what I have said it is natural to think that we would

find the same constraints in predicate transfer, but I have the impression that this is not the case: it’s

strange to say That shoelace is my left shoe’s or Those people are the hospital’s, for example,

suggesting that the predicative genitive is in certain ways more restrictive in the functions it can make

17

See, however, the reservations expressed by Anna Szabolcsi concerning this kind of argument,

mentioned in fn. 13. In presenting this material, I have found that speakers differ in their judgments

on this precise example. To account for the judgments of those who find it OK, one possibility could

be to say that also can be accompanied by a change in the function being referred to. 18

Our example The hospital called arguably provides another example of this.

20

use of. When it comes to predicate transfer, one speculation is that the function must be one that

introduces a sortal change, in the sense that we can feed the value for the function (a person, for

example) to predicates to which we can’t feed the argument of the function (a car, for example). (Note

that this doesn’t necessarily mean that the transferred predicate can apply to arguments that the

untransferred predicate could not apply to. When I use I’m just across the street to talk about my car,

the untransferred predicate just across the street admits as possible arguments all the things that the

untransferred predicate admits – namely, cars – and more.19

) In cases where a given function is

possible, there are also apparently pragmatic conditions that restrict its use. This is very clear in the

use of NP transfer, where the transferred NP seems to have to form part of a useful “cover” for the

purpose of the conversation: it is natural for the waiter to say The ham sandwich at Table 7 is getting

impatient, but it would be odd for a customer to use it.20

A guilty afterthought.21

Let me end by admitting that I cheated. I argued that there was predicate transfer in a variety of

examples. But did the arguments that I gave really show that? No, not really. I did show that in

sentences like I am parked out back the subject is not transferring. I also showed, I think, that in

sentences like The ham sandwich at Table 7 is getting impatient, the subject is not transferring, and

neither is the verb phrase. But I did not really show that the only way to handle these examples was to

say that the main predicate or the nominal predicate was transferring, and there is in fact another

possibility.

Consider sentences like I am parked out back. In sentences that we are used to, like My car is

parked out back, the meaning of the subject has a way of combining with the meaning of the predicate,

which I assume is functional application. Now, it is true that if I does not transfer, and is parked out

back doesn’t either, we are not going to account for the interpretation of this sentence if the subject

and the predicate combine in the same way. But who said that the only way in which subjects and

predicates can combine is by functional application? If there is another rule of semantic composition

that allows the two to combine as in (49c), then we can arrive at the same result without saying that

anything transfers.

(49) a. [[ My car is parked out back ]] standard

= [[ (is) parked out back ]] ( [[ my car ]] )

b. [[ I am parked out back ]]

≠ [[ (is) parked out back ]] ( [[ I ]] )

c. [[ I am parked out back ]] another possibility for composition?

= [[ (is) parked out back ]] (the unique y s.t. f(y) = [[ I ]] )

19 See Egg 2002 for other examples of this kind. Nicholas Asher brought up the question of whether

we can imagine the discourse as a “trigger” for predicate transfer in this example. My position at the

moment is that I’m just across the street exhibits an ambiguity that we resolve in the way in which we

resolve other ambiguities, taking the discourse into account, and that, if we have a preference for one

reading over another in “out-of-the-blue” contexts, this is to be related to other processing preferences

that we have, like (perhaps) a preference to posit simpler structures. 20

This is remarked on by Ward, and discussed at more length by Nunberg who talks of a more general

“noteworthiness” constraint on predicate transfer. Fabio Del Prete wondered whether pragmatic

constraints could also lie behind the impossibility of transferring a coordination as in (45b). 21

This final note was inspired by Asher 2011, even if Asher argues for a position different from the

alternative I bring up here. See Recanati 2009 for discussion that goes in this direction.

21

Note that, if we assume that there is a special composition rule that allows us to do this, this is

completely consistent with the conclusions we drew from the ellipsis examples, that there is a

predicate around with the untransferred meaning. But there would be no reason to assume that there is

also a predicate around with the transferred meaning, so no reason to assume that there is a further

syntactic layer! To account for a sentence like I was parked out back and that red Volvo was too, we

would simply say that the special composition rule occurs in the first conjunct but not in the second.

I will leave open the question of how different this option really is from the approach that I

have been sketching. There are two things I would like to note about it. One is that it requires, in a

sense, duplicating certain aspects of the composition rules that we have already – function application

is involved in (49c), and something like modification would be involved in the special additional

composition rule that we would have to add to take over from modification in expressions like ham

sandwich at Table 7. That is maybe a bit suspicious. The other is that, on this approach, we would

have to express in some way the fact that our special composition rule in (49c) can apply when the

main predicate is parked out back but not when it is parked out back and particularly shiny. It is hard

to see how we could distinguish these two by just looking at their denotations, so the constraint on the

application of the composition rule would have to make reference to syntactic information, which is

perhaps also a bit funny when it’s a matter of using a rule to combine meanings.

Asher, N. (2011) Lexical Meaning in Context: A Web of Words. Cambridge: Cambridge University Press. Beard, R. (1998) Derivation. In A. Spencer and A. Zwicky (ed.), The Handbook of Morphology. Oxford:

Blackwell.

Egg, M. (2002) Semantic construction for reinterpretation phenomena. Linguistics 40. Elbourne, P. (2005) Situations and Individuals. Cambridge, MA: MIT Press. Fauconnier, G. (1985) Mental Spaces. Cambridge, MA: MIT Press.

Heim, I. (1990) E-type pronouns and donkey anaphora. Linguistics and Philosophy 13. Heim, I. and A. Kratzer. (1998) Semantics in Generative Grammar. Malden, MA: Blackwell.

Jackendoff, R. (1992) Mme Tussaud meets the Binding Theory. Natural Language and Linguistic Theory 10.

Kleiber, G. (1999) Problèmes de Sémantique: La Polysémie en Questions. Villeneuve d’Ascq : Presses

Universitaires du Septentrion. Nunberg, G. (1979) The non-uniqueness of semantic solutions: polysemy. Linguistics and Philosophy 3. Nunberg, G. (1995) Transfers of meaning. Journal of Semantics 12.

Recanati, F. (2009) Compositionality, flexibility and context-dependence. In W. Hinzen, E. Machery and M. Werning (ed.), The Oxford Handbook of Compositionality. Oxford: Oxford University Press.

Recanati, F. (2004) Literal Meaning. Cambridge: Cambridge University Press. Rothstein, S. (1995) Adverbial quantification over events. Natural Language Semantics 3. Sag, I. (1981) Formal semantics and extra-linguistic context. In P. Cole (ed.), Radical Pragmatics, New

York: Academic Press. Storto, G. (2003) Possessives in Context. PhD dissertation, UCLA. Ward, G. (2004) Equatives and deferred reference. Language 80.