Double Objects Again∗

SIGRID BECK
UNIVERSITY OF CONNECTICUT AT STORRS

KYLE JOHNSON
UNIVERSITY OF MASSACHUSETTS AT AMHERST

AUGUST 2002

1 The double object and NP+PP frames

In (1a) is a sentence that illustrates what we’ll call the double object frame; and in
(1b) is illustrated what we’ll call the NP+PP frame.
(1) a. Satoshi sent Thilo the Schwa¨bische Wo¨rterbuch.
b. Satoshi sent the Schwa¨bische Wo¨rterbuch to Thilo.
Many have entertained the notion that there is a rule that relates sentences such as
these. This is suggested by the fact that it is possible to learn that a newly coined
verb licenses one of them and automatically know that it licenses the other. Marantz
(1984) argues for the existence of such a rule in this way, noting that once one has
learned of the new verb shin by exposure to (2a), the grammaticality of (2b) is
also learned.
(2) a. Thilo shinned the ball to Satoshi.
b. Thilo shinned Satoshi the ball.
This is explained if there is a rule that ties the double object frame together with
the NP+PP frame, making it sufficient to know that a verb licenses one if it
licenses the other.
Frequently, the rule involved has been taken to be syntactic in nature. See,
among many others, Fillmore (1965), Oehrle (1976), Baker (1988), and Larson
(1988). The leading idea under this view is that the two frames are simply
different surface manifestations of the same underlying structure. Typically, this
approach posits that the NP+PP frame represents that underlying structure from
which the double object frame is transformationally derived.
There is evidence, however, that the two frames instead have different un-
derlying structures, and are not related by transformation. This evidence, then,
∗
Our thanks to Thilo Goetz and Satoshi Tomioka for the examples.

1
2 DOUBLE OBJECTS AGAIN

suggests that the rule relating (1a) and (1b) must be found somewhere else, a
likely candidate being the lexicon. We report here an argument from Kayne
(1984b) to this end that is based on Nominalizations.
With certain restrictions, deverbal nominalizations allow the object of the
verb to surface as either the genitive of the resulting NP, or inside an of -phrase.
(3) examine the problem →
(4) a. the examination of the problem
b. the problem’s examination
But this is only possible if the “object” of the verb is its logical object, i.e., its
argument.1 It is not possible if it is instead the subject of a small clause, for example:
(5) believe Thilo handsome →
(6) a. * the belief of Thilo handsome
b. * Thilo’s belief handsome
In this way, then, Nominalizations can be used to determine whether the NPs that
follow a verb are that verb’s argument or not. On this basis, Kayne argued that
the first NP following a verb in the double object frame is not that verb’s
argument because in Nominalizations it behaves like believe and not like
examine:
(7) present Satoshi a ball →
(8) a. * the presentation of Satoshi a ball
b. * Satoshi’s presentation a ball
In the NP+PP frame, by contrast, the first NP does behave like the verb’s argument:
(9) present the ball to Satoshi →
(10) a. the presentation of the ball to Satoshi
b. the ball’s presentation to Satoshi
He proposes that the two NPs following a verb in the double object frame are em-
bedded in a small clause in much the same way that the two phrases following
believe in (5) are. Just as Thilo in (5) is the subject of a small clause, then,
Kayne’s proposal is that Satoshi is the subject of a small clause in (7).
If we adopt a Larsonian,2 binary-branching, representation of VPs, we can
express the different structures that Kayne’s proposal would give to the double
ob- ject and NP+PP frames as follows. The NP+PP frame would place the two
comple- ments in the Specifier and Complement positions of a phrase headed by
the verb,
1
The paradigm supporting this generalization originates with Ross (1974).
2
See Larson (1988) and Larson (1990).
 SIGRID BECK & KYLE 3

and this VP will be embedded in another phrase (“vP”) whose head has no phono-
logical reflex.3 There is controversy about which argument goes into which position
inside VP — let’s take the position that the NP occupies the Specifier position, as
in (11).4
(11) vP

v VP
NP
VJ
the ball
V PP
present to Satoshi
From this underlying representation, the surface form is produced by moving the
verb through the position occupied by “v” and into a higher I0 position. In
addition, the “object” NP is moved into a position that determines its Case — a
position located between the surface position of the verb and vP. There are
presently many differing proposals about how to operationalize this system; any
of them will do for our purposes. For concreteness, we will let the object NP
adjoin to vP, as in Chomsky (1995), and simply call the position in which the verb
surfaces: I0. From (11), then, the surface representation of the NP+PP frame will
be (12).
(12) IP

I vP

present1 NP2 vP

the ball NP vJ
tsubject v VP
t1 NP VJ

t2 V PP
t1 to Satoshi
3
For the existence and nature of vP, see Hale and Keyser (1993, 1997) Kratzer (1994) and much
subsequent work.
4
This is how Larson (1988) would do it; for arguments to the contrary, see Takano (1996). Either
view is consonant with our conclusions.
4 DOUBLE OBJECTS AGAIN

This is a standard way of representing two-object verbs, and the NP+PP frame is
simply one of these.
By contrast, under Kayne’s proposal the double object frame has a small
clause hidden in it, which gives it an underlying representation like (13).
(13) vP

v VP

V XP

present NP XJ
Satoshi
X NP

the ball
“X” is the silent head of the small clause. From the underlying (13), the surface
representation is produced from the same verb movement + object movement that
operates to form (12):
(14) IP

I vP

present1 NP2 vJ
Satoshi v VP
t1 V XP

t1 NP XJ
t2 X NP

the ball
A complete theory would link the many other unique properties of the double object
frame to the existence of this small clause.5 But that’s not what we will do here.
5
The most famous of which are: the scope rigidity of the two objects (see Barss and Lasnik
(1986), Richards (1997) and Bruening (2001)); its behavior in particle constructions (see Kayne
(1984a), Dikken (1992) and Johnson (1991)); the inability of the first object to Heavy NP Shift
(see Pesetsky (1982) and Larson (1988)); and the patterns of Case marking allowed (see Zaenen,
Maling, and Thra´insson (1985), Collins and Thra´insson (1993)).
 SIGRID BECK & KYLE 5

In addition to syntactic contrasts such as these, there are semantic reasons for
believing that the double object frame and the NP+PP frame are not mere surface
variants of the same sentence. There are sometimes truth conditional differences
between a double object frame sentence and the corresponding NP+PP frame sen-
tence. Green (1974) argued that these differences form a pattern, moreover, which
we suggest gives a clue as to the identity of “X” heading the small clause in the
double object frame.
Green observes several differences between these frames. 6 She argues that
the double object frame always has a component to its meaning that is not
necessar- ily found in the NP+PP frame. She argues that every double object
frame includes a possession, or HAVE, component to its meaning.7 Very roughly,
the meanings of (1a) and (2a) might be expressed with (15).
(15) a. Satoshi’s sending [the Wo¨rterbuch] CAUSE [BECOME [Thilo
HAVE the Wo¨rterbuch]]
b. Thilo’s shinning [the ball] CAUSE [BECOME[Satoshi HAVE the
ball]]
By contrast, the NP+PP frame does not necessarily have the HAVE component in
its meaning. The meanings of (1b) and (2b), for instance, might be expressed with
(16).8
(16) a. Satoshi’s sending [the Wo¨rterbuch] CAUSE [BECOME [the
Wo¨rter- buch is AT Thilo]]
b. Thilo’s shinning [the ball] CAUSE [BECOME [the ball is
AT Satoshi]]
Green’s conclusion is built on the claim that the meaning of every double
object frame has properties that derive from the presence of HAVE, and that these
6
It is not our purpose here to examine all of these differences. Among those we will ignore is
the issue of telicity, which is what we believe is behind the oft-noted difference in teach someone
something, which entails that someone learnt something, and teach something to someone, which
doesn’t. (See Oehrle (1976)).
7
We must worry about such cases as deny and spare. A sentence like (1) can be shoved into
this pattern only by allowing NOT-HAVE, as in (2), to be part of the generalization.
(1) Thilo denied Satoshi the victory.
(2) Thilo’s denying CAUSE [BECOME [Satoshi NOT-HAVE the victory]]
Or, preferably, we can fold these cases into one of the other classes of verbs that license two NPs
in their VP but which are plausibly not members of the double object frame (e.g., “Satoshi elected
Thilo class president,” or “The book cost Thilo 60 euro.”)
8
We express here the locative meaning to has in this context with (is) AT.
6 DOUBLE OBJECTS AGAIN

properties are not always present in the meanings of the NP+PP frame. The contrast
in (17) is one such difference.
(17) a. Satoshi sent the Schwa¨bische Wo¨rterbuch to Tu¨bingen.
b. # Satoshi sent Tu¨bingen the Schwa¨bische Wo¨rterbuch.
Because the indirect object (i.e., the PP) in the NP+PP frame denotes a location,
it is free to refer to inanimate places as well as to animate ones. But because the
double object frame makes the indirect object (i.e., first NP) the subject of a
HAVE relation, it is confined to referring to objects that can be possessors. The
oddness of (17b) derives from imputing Tu¨bingen with this ability. All motion
verbs with the double object and NP+PP frames show this kind of contrast.9
Another contrast pointing in the same direction can be seen in (18).
(18) a. Thilo cooked spa¨tzle10 for Satoshi.
b. Thilo cooked Satoshi spa¨tzle.
The indirect object in (18a) has a significantly wider range of roles than does the
indirect object in (18b). It’s possible to understand (18a) to describe a situation in
which Thilo cooked spa¨tzle in place of Satoshi — perhaps Satoshi doesn’t
know the recipe,11 for example, but is supposed to bring some to the potluck. But
that meaning is absent in (18b), which can only mean that Thilo cooked
spa¨tzle for Satoshi to have. This follows if the meaning of cook in the double
object frame is as in (19a), whereas it is something like (19b) when in the NP+PP
frame.
(19) a. Thilo’s cooking spa¨tzle CAUSE [BECOME [Satoshi HAVE spa¨tzle]]
b. Thilo’s cooking spa¨tzle CAUSE [BECOME [EXIST(spa¨tzle)]]
for the benefit of Satoshi
Note that in (19b), for Satoshi is a kind of modifier, and therefore has a considerably
different status than does the argument Satoshi in (19a). This difference in the
double object and the NP+PP frame is reproduced in every benefactive verb that
has both these frames.12
These two sources of evidence can be tied together if “X” is the source of
the constant HAVE part to the meanings we have seen in the double object frame.
Thus, the double object frame for send and cook would be as in (20).
9
Motion verbs of this sort include kick, throw, mail, ship, hurl, push, roll, toss, kick, drag, . . . .
10
The pasta indigenous to southern Germany.
11
Beat 250 gramms of flour, 2 eggs, 4 ounces of water and some salt to a smooth dough. Spread
a portion onto a wet board and scrape thin slices off into a pot of boiling water with a long knife.
Spa¨tzle are done when they rise to the surface. Take out immediately and put on a warm
platter. Repeat until all the dough is used. Serves 4.
12
A list that includes bake, boil, fry, knit, sew, cut, make, build, fashion, . . . .
 SIGRID BECK & KYLE 7

(20) vP

v VP

V
HAVEP

send NP HAVE
J

cook
HAVE NP
Thilo
Satoshi
the Wo¨rterbuch
spa¨tzle
By contrast, the underlying representation for the NP+PP frames for these verbs
would be as in (21).
(21) a. vP

v VP

NP
VJ
the
Wo¨rterbuch V PP

send to Thilo
b. vP

vP PP

v VP for Satoshi

V NP

cook the spa¨tzle

In (21b) we have adjoined the modifier for Satoshi to vP. (We return to the
position of the for-phrase in section 3.)13
In this paper, we will produce new evidence that largely confirms this pic-
ture of the difference between the double object and NP+PP frames. The evidence
comes from the behaviour of the adverb again. Section 2 sketches an analysis of
again from von Stechow (1995, 1996), which is applied to the double object and
13
The literature is rife with alternatives, but almost all of them share with (21) what is crucial for
us: that verb and its arguments form a constituent underlyingly that does not include the modifier.
(Interestingly, Larson (1988) is one of the rare counterexamples to this.)
8 DOUBLE OBJECTS AGAIN

NP+PP frames in section 3. While the results confirm that the double object frame
has a small clause in it headed by HAVE that is absent from the NP+PP frame, they
will drive us to a more complex representation than (20). Furthermore, we will
see that the NP+PP frame also sometimes holds a hidden small clause in it. This
means that the central difference in these frames devolves to what kind of small
clause they can house: only the double object frame has HAVE.

2 A structural theory of again

In this section, we introduce our proposed test for the internal make-up of the
pred- icates in section 1: the adverb again. The theory we discuss is von Stechow
(1995, 1996), which shows that the different readings of again are a probe into
the syntac- tic and semantic composition of predicates.

2.1 A basic case: open

The basic fact to be captured by a theory of again is the ambiguity of data like (22),
whose two possible readings are paraphrased in (23).
(22) Sally opened the door again.
(23) a. Sally opened the door, and she had done that before.
(repetitive)
b. Sally opened the door, and the door had been open before.
(restitutive)
On both interpretations, what makes the sentence in (22) appropriate is some previ-
ous eventuality. On the repetitive reading, that event has to be a previous opening
of the door by Sally. On the restitutive reading, by contrast, that event is the
door’s being open.
The repetitive reading in (22a) is the straightforward one. For this read-
ing, we assume an interpretation of again as given in (24) (compare e.g. Stechow
(1996), Fabricius-Hansen (2001) for recent discussion and references). Again
oper- ates on a property of events, and indicates repetition of events characterized
by that property. More precisely, again expresses a relation between a property of
events and an event. It presupposes that there was a previous event that has the
property, and asserts that the property is true of the event.14
14
Like Stechow, we assume a standard framework of compositional translation into a formal
language. The logic we use makes use of event variables (cf. Davidson (1967)). The semantic
type of events is <i>.
 SIGRID BECK & KYLE 9

(24)

1 iff P(e) & eJ[ eJ <e & P(eJ) ]

 0 iff P(e) & ∃eJ[ eJ <e & P(eJ) ]
[ again] (P<i,t>)(e)  ¬ ∃
= undefined otherwise
The input to again on the repetitive reading is the predicate of events given (ap-
proximately) by (25): openings of the door by Sally. This can be derived from the
sentence in (22) straightforwardly, as we will see in detail below.
(25) λe.Sally opened the door(e)
More challenging is the restitutive reading. Under a structural theory of
again, it is assumed that again always has the meaning in (24), i.e. indicates repe-
tition. The difference between the two readings is in what event is repeated. On
the restitutive reading, only the result state of the opening of the door is repeated;
thus again operates on the predicate of events in (26).
(26) λe.opene(the door)
The problem is to derive the property of events in (26) from the syntactic struc-
ture of (22). Stechow (following earlier suggestions e.g. by McCawley (1968))
proposes a decomposition analysis of the verb open, which is reflected in the syn-
tactic structure. Thus, the sentence in (27a) has the underlying structure in (27b),
where open is decomposed into the adjective open, and a phonologically empty
verb (=V*) which contributes a CAUSE BECOME component to the meaning of
the sentence (cf. (28)). The surface structure is given in (27c).15 The sentence is
in- terpreted compositionally to derive the predicate of events in (29a), which
captures the intuitive truth conditions of (27a); (29b) is a paraphrase.16
(27) a. Sally opened the door.
b. vP

NP vJ
Sally v VP
V* AP

A NP

open the door

15
In this respect, his proposal mirrors Hale and Keyser (1993, 1997).
16
We assume standard interpretations of the expressions CAUSE and BECOME in these formulas
— compare Lewis (1973), Dowty (1979) and von Stechow (1996).
10 DOUBLE OBJECTS AGAIN

c. vP

NP vJ
Sally
v VP

V* AP
V*+open1 A NP

t1 the door
(28) [ V*] = λp λx λe ∃P[Pe(x) & ∃eJ[BECOMEe (p) & CAUSE(eJ)(e) ]
J

(29) a. λe P[Pe(Sally) & eJ[BECOMEe (λeJJ.opene (the door)) &

J JJ
∃ ∃
CAUSE(eJ)(e) ]]
b. There was an action of Sally’s that caused the door to become open.
The assumption that we have the structure in (27) for the verb open gives
us a straightforward way of deriving the ambiguity. Again has two different
adjunction sites, and consequently two different constituents it can modify. The
two structures for the ambiguous example (22) are given in (30). (30a) can
straightforwardly be interpreted as in (31) — the desired restitutive reading. (30b)
is interpreted as in (32), a more complete derivation of the repetitive reading.
(30) a. vP

NP vJ

Sally v VP

V* AP

AP AdvP

A NP again

open the door

 SIGRID BECK & KYLE 1

b. vP

vP
AdvP

NP vJ again

Sally v VP
V* AP

A NP

open the door

(31) a. λe P[Pe(S) & eJ[BECOMEe (λeJJ.againe (λeJJJ.opene (the door))) &

J JJ JJJ
∃ ∃
CAUSE(eJ)(e)]]
b. There was an action of Sally’s that caused the door to become once
more open.

(32) a. λe.againe(λeJ P[Pe (S) & eJJ[BECOMEe (λeJJJ.opene (the door)) &
J JJ JJJ
∃ ∃
CAUSE(eJJ)(eJ)]])
b. Once more, there was an action of Sally’s that caused the door to
become open.
The ambiguity is thus purely structural in nature. Again operates on prop-
erties of events, and always indicates repetition. Therefore, all properties of events
that can, intuitively, be modified by again must be available as the denotations of
syntactic constituents that again attaches to.

2.2 again is a test of syntactic structure

It might be thought to be rather costly to introduce the whole decomposition struc-

ture in (27) into the syntax for the purposes of deriving the ambiguity in (22). In-
deed, other researchers have argued that result states are available on a conceptual
level, and that again has a reading in which it indicates repetition of those result
states rather than of the property of events that it modifies (see e.g. Fabricius-
Hansen (2001)) for a recent proposal in this spirit, and references therein.). Under
such an analysis, again would no longer be a detector of constituents that denote
result states, since the result states need not be the meaning of any constituent in
the syntactic structure. The ambiguity in (22) would be lexical (two
interpretations of again) rather than structural.
12 DOUBLE OBJECTS AGAIN

Stechow’s most important argument for the syntactic solution is the fact
that the availability of the restitutive reading hinges on syntactic context. Since
we will use again as an indicator of syntactic structure, we will briefly review this
argument. It comes from the German word order facts illustrated by (33). A
restitutive reading of wieder (=again) is only possible when wieder follows the
direct object. If wieder precedes the object, only the repetitive reading is
available.
(33) a. (weil) Satoshi die wieder o¨ffnete.
Tu¨r
(because) Satoshi the door opened
again (repetitive and restitutive readings)

b. (weil) Satoshi wieder die o¨ffnete.

Tu¨r
(because) Satoshi the door
again opened (repetitive reading only)
Stechow’s explanation for these facts is, in a nutshell, that the direct object in
Ger- man obligatorily moves to a fairly high position in the structure (SpecAgrO,
for Stechow; vP for us). That position is higher than the element in the structure
that expresses the CAUSE BECOME meaning component. Thus when wieder
precedes, and is higher than, the direct object, it must take scope over CAUSE
BECOME. On the other hand, when wieder follows the direct object, it may be
above or below CAUSE BECOME. The relevant structures are given below. The
empty verb is re- sponsible for CAUSE BECOME. In (34), wieder can adjoin
either below (to the small clause) or above the verb (to the vP), as indicated. In
(35), word order tells us that the sister of wieder includes the VP and the verb.
Thus the example can only be repetitive.
(34) vP

NP1 vJ
die v VP
Tu¨r
AdvP VJ

wieder AP V*

AdvP AP

wieder AJ
NP A

t1 offnete
 SIGRID BECK & KYLE 1

(35) vP

AdvP vP
wieder
NP1 vJ
die v VP
Tu¨r
AP V*

AJ
NP A

t1 offnete
The data show that a proper analysis of restitutive wieder has to rely on syntactic
structure. Conceptual availability of a result state is not sufficient for the existence
of a restitutive reading.
The German facts can be reproduced in English. While (36a) is
ambiguous, (36b) with again preceding the finite verb, is unambiguously
repetitive.
(36) a. Thilo opened the door again. (repetitive and restitutive)
b. Thilo again opened the door. (repetitive only)
Stechow’s structural explanation extends straightforwardly to English. In (36a), as
we saw above, two adjunction sites for again are possible. Again could adjoin to the
Small Clause (resulting in a restitutive reading) or to the VP (resulting in a repetitive
reading, the adjunction site being above CAUSE BECOME):17
17
For simplicity, in this, and all following representations for which we give an interpretation, the
semantically vacuous movements that form the surface representation will be ignored.
14 DOUBLE OBJECTS AGAIN

(37) vP

NP vJ
Thilo v VP

VP AdvP

V* AP again

AP AdvP

AJ again

A NP

open the door

On the other hand, it is clear in (36b) that the adjunction site of again must include
VP:
(38) IP

AdvP IP

again I vP
NP vJ

Thilo v VP

V* AP

AJ

A NP

open the door

We conclude that there is support for the structural nature of Stechow’s analysis
in both German and English.18 His theory implies that the properties of events that
again operates on are directly provided as the denotations of constituents in the
18
English data also furnish support for Stechow’s account of certain counterexamples to the
gen- eralization just described. There are some verbs whose sentences allow wieder to precede the
object and still give a restitutive reading. One of these is verlassen:
 SIGRID BECK & KYLE 1

syntax. Again adjoins to those constituents. The word order facts support this.
Additional evidence that favours a structural theory over its alternatives has been
brought forth in Beck and Snyder (2001). In what follows we use this analysis of
again to gather information about the predicates from section 1.

3 Two-object verbs and again

This section will apply again as a test for constituent structure and for semantic
composition to the constructions from section 1. We first look at double object
constructions, and then compare them to the corresponding NP+PP frames.

3.1 Double object constructions and again

(39) is ambiguous between the repetitive reading in (40a) and the restitutive reading
in (40b).
(39) Thilo gave Satoshi the map again.
(40) a. Thilo gave Satoshi the map, and that had happened before.
b. Thilo gave Satoshi the map, and Satoshi had had the map before.
According to the structural theory of again, again must operate on the two predi-
cates of events given (roughly) in (41) for these readings.
(41) a. λe.Thilo gave Satoshi the map(e)
b. λe.havee(the map)(Satoshi)
The restitutive reading, in particular, requires there to be a constituent whose
mean- ing is (41b). This leads us to suggest the interpretation in (42) for (39) on
that reading.
(1) Als
Anna wieder das verließ, war es dunkel.
Haus
When Anna the house left, was it dark
again
‘When Ann left the house again, it was dark’
(repetitive and restitutive readings)
(Stechow 1996, section
6)
His account of these counterexamples is that the verbs in question allow objects to surface in a
lower position, and this allows wieder to be adjoined to the constituent that denotes the result state and
still precede the object. Interestingly, in the cases where our judgments are clear, these verbs in
English do not allow the restitutive reading when again precedes the verb:
(2) Anna again left the house. (repetitive reading only)
This is expected on Stechow’s account, as his solution involves manipulating the relative positions
of the object and adverb and in the English data it is the relative positions of verb and adverb that
matter.
1 DOUBLE OBJECTS

(42) a. λe[givee(the map)(Thilo) & eJ[BECOMEe (λeJJ.againe (λeJJJ.havee

J JJ JJJ
∃
(the map)(Satoshi))) & CAUSE(eJ)(e)]
b. Thilo’s giving of the map caused Satoshi to come to once more
have the map.
A fully specified interpretation for a double object give example without again
therefore looks as in (43).
(43) a. Thilo gave Satoshi the map.
b. λe[givee(the map)(Thilo) & eJ[BECOMEe (λeJJ.havee (the map)
J JJ
∃
(Satoshi)) & CAUSE(eJ)(e)]
c. Thilo’s giving of the map caused Satoshi to come to have the map.
According to section 1, the interpretation in (43b) would have to be derived
from the structure in (44).
(44) vP

NP vJ

Thilo v VP
V HAVEP
J
give NP HAVE

Satoshi HAVE NP

the map
However, note that the map would have to serve as an argument of both HAVE
and give in (43b). So we have to establish an anaphoric connection between these
ar- gument slots. And second, note that a source of the CAUSE BECOME
component has to be found.
For the first of these needs, we suggest changing the representation in (44)
so that it has a silent argument in it which we will designate with “pro.” This
silent argument must be c-commanded by the map, and so we will make it the
argument of HAVE, and make the map the argument of give. This means that the
underlying structure of the VP of a double object give example is now:
 SIGRID BECK & KYLE 1

(45) vP

NP vJ
Thilo
v VP

NP1 VJ
the map
V HAVEP

give NP HAVE
J

Satoshi HAVE NP

pro1
From this representation, the surface order is achieved in the way outlined in section
1: by moving the verb into a position outside vP and moving the “direct object” –
here Satoshi – into the position licensing accusative Case marking.19
Note that this representation preserves Kayne’s conclusion that the first NP
of the double object construction is not an argument of the verb. Even with these
changes, this NP is the subject of an embedded small clause, just as it was on the
simpler representation we examined in section 1. Therefore, the fact that the double
object frame cannot be mapped into Nominalizations still follows.
With respect to introducing the CAUSE BECOME meaning into the
double object frame, one obvious way to achieve this would be to introduce a
phonolog- ically empty element into the structure that is responsible for CAUSE
BECOME
— analogous to the case of open. This direction would be in line with Stechow’s
approach.
But there is reason to pursue a different approach. Snyder (2001) provides
evidence from language variation and language acquisition in favour of relating
double object constructions to resultatives. He proposes that both are governed by
the same parameter of grammar, which explains the correlations which he found
among them. Interestingly, the interpretation of resultatives like (46) on the follow-
ing page require the same CAUSE BECOME component, without there being any
counterpart in the structure. Translating Stechow’s (1995) proposal into our
syntax, (46a) would have the structure in (47):
19
In this representation the map does not have structural Accusative Case, and behaves
something like an indirect object as a consequence. Movement of Satoshi past the map therefore
works along the same lines that raising past an indirect object does (i.e., “Satoshi seems to Thilo to
have the map.”)
1 DOUBLE OBJECTS

(46) a. Satoshi hammered the sword flat.

b. Satoshi’s hammering the sword caused the sword to become flat.

(47) vP

NP vJ
Satoshi
v VP

NP1 VJ
the sword V AP
hammered PRO1 AJ

flat
Given the standard mechanisms of compositional interpretation, the interpretation
of (47) would crash at the point where we try to combine the Small Clause with
the verb hammer, due to a type mismatch: the Small Clause is a property of events
(type <i,t>) and the verb is of type <e,<e,<i,t>>>, requiring an individual
object. Stechow proposes the following interpretation principle for resultatives to
resolve this mismatch:
(48) If α =[γV βSmall Clause] and βJ is of type <i,t> and
γJ is of type <e,...<e,<i,t>>> (an n-place predicate),
then αJ = λx1 . . . λxn λ e.γJe(x1). . . (xn) & eJ [BECOMEe (βJ) & J
∃
CAUSE(eJ)(e)]
Compositional interpretation then proceeds as indicated in (49), and we derive an
intuitively appropriate interpretation for (46a). See Stechow (1995) for more details
and discussion.
(49) a. [V hammered [AP PRO1 flat] ] →
J

λx λy λe.hammere(x)(y) & eJ[BECOMEe (λeJJ.flate (x1)) &

J JJ
∃
CAUSE(eJ)(e)]

b. [vP Satoshi v [VP [the sword]1 [V hammered [AP PRO1 flat] ]]]]→
J

λe.hammere(the sword)(Satoshi) & eJ[BECOMEe (λeJJ.flate J JJ

∃
(the sword)) & CAUSE(eJ)(e)]

c. Satoshi’s hammering the sword caused it to become flat.

SIGRID BECK & KYLE 1

Our suggestion is that the interpretation of (43) proceeds in a manner essentially

analogous to (46a). The structure we assume is given in (50), and the crucial steps
in the interpretation in (51). The principle in (48) allows us to combine the verb
give with the small clause headed by HAVE.
(50) a. Thilo gave Satoshi the map.
b. vP

NP vJ
Thilo v VP
NP1 VJ

the map V HAVEP

gave NP HAVE
J

Satoshi HAVE NP

pro1
(51) a. [ gave [HAVEP Satoshi HAVE pro1 ]] →
λy λx λe [givee(y)(x) & eJ[BECOMEe (λeJJ.havee (x1)(Satoshi))
J JJ
∃
& CAUSE(eJ)(e)]

b. [ Thilo [ [the map] [ gave [HAVEP Satoshi HAVE pro1 ]]]] →

λe[givee(the map)(Thilo) & eJ [BECOMEe (λeJJ.havee (the map)
J JJ

(Satoshi)) & CAUSE(eJ)(e)] ∃

Both the double object and resultative constructions depend on the interpretation
principle in (48) and this is how we propose to capture the connection between
these constructions. (See also Beck and Snyder (2001).) Snyder’s observation also
extends to certain NP+PP frames — we will come back to this shortly.
Note that with (50) as the structure for double object give, again can find
an adjunction site that yields a resitutive reading (the small clause) as well as ad-
junction sites that yield the repetitive reading (for example vP). Thus we make the
desired predictions. When again is adjoined to the small clause, as in (52), the in-
terpretation in (53) results; and when it adjoins to vP, as in (54), the interpretation
in (55) is derived.
(52) Thilo [VPthe map1 [V give [HAVEP [Satoshi HAVE pro1] again ] ]]
J

(53) a. λe[givee(the map)(Thilo) & ∃eJ[BECOMEe J

(λeJJ.againe (λeJJJ.havee (the map)(Satoshi))) & CAUSE(eJ)(e)]

JJ JJ
2 DOUBLE OBJECTS

b. Thilo’s giving of the map caused Satoshi to come to once more have
the map.

(54) [vP [ Thilo [VPthe map1 [V give [HAVEP Satoshi HAVE pro1]]] again ]
J

(55) a. λe.againe(λeJ[givee (the map)(Thilo) & ∃eJJ[BECOMEe

J JJ

(λeJJJ.havee (the map)(Satoshi)) & CAUSE(eJJ)(eJ)])

JJJ

b. Once more, Thilo’s giving of the map caused Satoshi to come to have
the map.
As a first conclusion, we see that the existence of a restitutive reading and
the nature of that reading confirm the suggestions made in section 1 regarding the
structure of double object constructions. The proposed small clause is available
as the constituent that denotes the result state, and its head predicate HAVE yields
the intuitively correct restitutive reading. We have added an empty pronoun to the
proposed structure to capture the dual role of the second object in the argument
structure of the sentence.
The other types of double object verbs mentioned above have restitutive
readings parallel to (39), (see (56) and (57)) as would be expected from the
analysis sketched in section 1.
(56) benefactives:
Thilo sewed Satoshi a flag again.
a. Thilo sewed Satoshi a flag, and that had happened before.
b. Thilow sewed Satoshi a flag, and Satoshi had had a flag before.
(57) motion verbs:
Thilo sent Satoshi the map again.
a. Thilo sent Satoshi the map, and that had happened before.
b. Thilo sent Satoshi the map, and Satoshi had had the map before.
All types of double object verbs have restitutive readings in the double object
con- struction, and that reading is the same across these verbs: the first object is
restored to possession of the second object. This confirms the existence of a small
clause with a head predicate HAVE in all these data: again, on the restitutive reading,
takes as its semantic argument a property of events of the possession type, and
must mod- ify the corresponding constituent in the syntax. We conclude that the
combination of the analysis of double object constructions sketched in section 1,
and Stechow’s theory of again makes precisely the desired predictions.
 SIGRID BECK & KYLE 2

3.2 The NP+PP frame and again

We now turn to the NP+PP frame of the same verbs. The relevant facts are listed
below.
(58) give:
Thilo gave the map to Satoshi again.
a. Thilo gave the map to Satoshi, and that had happened before.
b. Thilo gave the map to Satoshi, and Satoshi had had the map before.
(59) benefactives:
Thilo sewed a flag again for Satoshi.20
a. Thilo sewed a flag for Satoshi, and that had happened before.
b. Thilo sewed a flag for Satoshi, and there had been a flag before.
(60) motion verbs:
Thilo sent the plane to New York again.
a. Thilo sent the plane to New York, and that had happened before.
b. Thilo sent the plane to New York, and the plane had been in New
York before.
Observe that the result states that restitutive again detects in these data are not
uni- formly states of possession. In the benefactive case, it is a state of existence;
and in the case of motion verbs, an object is restored to a location. This variation
as such confirms the claim from section 1 that the semantics of the NP+PP frame
is not identical to the double object frame, and in particular, does not uniformly
involve HAVE. Restitutive again sharpens our intuitions about the semantic
differences be- tween the two frames.
What more does restitutive again tell us about the syntax and semantics
of the NP+PP frame? The fact that restitutive readings are possible with these
predicates shows that the following result states must be available as the meanings
of syntactic constituents:
(61) λe.havee(the map)(Satoshi)
(62) λe.∃x[flage(x)]
(63) λe.ATe(New York)(the plane)
Let’s look at motion verbs first. We propose that the result states are the
denotations of the PPs in the structures of these predicates, since the crucial infor-
mation about the result state comes from the preposition. This is seen more
clearly
20
For the reason again precedes the for-phrase in this example, see the discussion surrounding
(79).
 2 DOUBLE OBJECTS

in the following paradigm (In each case, the sentence in (a) has the restitutive
read- ing in (b)).
(64) a. Satoshi pushed the cookies under the bed again.
b. Satoshi pushed the cookies under the bed, and the cookies had been
under the bed before.
(65) a. Satoshi kicked the ball onto the field again.
b. Satoshi kicked the ball onto the field, and the ball had been on the
field before.
(66) a. Thilo threw the ball behind the fence again.
b. Thilo threw the ball behind the fence, and the ball had been behind
the fence before.
Given these observations, a complete semantic interpretation for a send NP+PP
frame example must look as in (67).
(67) a. Thilo sent the plane to New York.
b. λe[sende (the plane)(Thilo) & eJ[BECOMEe (λeJJ.ATe (New York)
J JJ
∃
(the plane)) & CAUSE(eJ)(e)]
c. Thilo’s sending the plane caused the plane to come to be in New
York.
Turning to the compositional derivation of this interpretation, note that the PP in
example (67), according to standard assumptions, would have the denotation in
(68). Combination with the meaning of the NP the plane would yield the desired
result state (63). Once more, though, we need the referent of that NP to be the
argument of send as well.
(68) λx λe.ATe(New York)(x)
Note moreover that as in the double object frame, a source for the CAUSE BE-
COME component in the interpretation of these predicates is needed. We suggest
to derive (67) from the structure in (69). (69) is an exact match of the structure
discussed in section 1, with the exception that we have introduced an anaphoric
element (in this example, for the referent of the plane). In this case, we propose to
put this anaphoric item in the Specifier of PP, as shown.21
21
As in the double object frame, we’ve represented the silent argument in (69) with pro, though
in this case it could easily be PRO, as in Heim and Kratzer (1998).
 SIGRID BECK & KYLE 2

(69) vP

NP vJ
Thilo
v VP

NP1 VJ
the plane
V PP

send NP PJ
pro1 P NP
to New York
The derivation is quite parallel to the derivation given for the double object con-
struction above. The special interpretation principle (48) is needed to combine send
and the PP to NY; the rest is canonical.
(70) [ send [PPpro1 [to New York]]] →
λy λx λe[sende (y)(x) & eJ[BECOMEe (λeJJ.ATe (New York)(x1)) &
J JJ
∃
CAUSE(eJ)(e)]
This structure offers adjunction sites for again that yield the restitutive and repeti-
tive readings. When again adjoins to PP, as in (71), the restitutive reading results
(see (72)); and when it adjoins to vP, as in (73), the repetitive reading emerges
(see (74)).
(71) Thilo [VPthe plane1 [V send [PP[pro1 to New York] again ] ]]
J

(72) λe[sende (the plane)(Thilo) & ∃eJ[BECOMEe (λeJJ.againe J JJ

(λeJJJ.ATe (New York)(the plane))) & CAUSE(eJ)(e)]

JJJ

(73) [vP Thilo [ [VPthe plane1 [V send [PPpro1 to New York]]]] again ]
J

(74) λe.againe (λeJ[sende (the plane)(Thilo) & eJJ[BECOMEe (λeJJJ.ATe (New

J JJ JJJ
∃
York)(the plane)) & CAUSE(eJJ)(eJ)])
One may balk at the unconventional claim in (71) that again can adjoin to a PP.
But note that this PP is a small clause,22 and as such has the same status as the
bracketed PP in (75).
(75) Satoshi talked [PPThilo into a stupor].
22
Treating argument PPs like these as small clauses embedded within VPs is a view defended on
independent grounds by Pesetsky (1995).
2 DOUBLE OBJECTS

This PP can host again as the restitutive reading in (76) shows.

(76) a. Satoshi talked [PP[Thilo into a stupor] again].
b. Satoshi’s talking brought it about that Thilo was once more in a stu-
por.
Using the interpretation principle in (48) to introduce the CAUSE+BECOME part
to the meanings of these frames is in keeping with Snyder’s results, as these par-
ticular NP+PP frames were included in the group of constructions that he found to
correlate with resultatives.23
The case of benefactives like (52) looks a little different. Let’s first make
sure that the restitutive reading of (52) is what we claim it is, and is indeed
weaker than the restitutive reading of the corresponding double object frame.
Consider the following scenario.
(77) Satoshi’s favourite little cousin has a tree house, which used to
include its own, exciting and exotic flag. The flag has vanished
under mysterious circumstances, and the cousin is devastated.
Satoshi would like to make her a new flag, but he can’t sew.
Satoshi’s friend Thilo, ever helpful, . . .
a.sewed a flag again for Satoshi.
b. # sewed Satoshi a flag again.
The sentence in (77a) is appropriate, and according to our claims, its presupposi-
tions are met (i.e. there used to be a flag). The sentence in (77b), however, is not
appropriate, comfirming the stronger presuppositions we assume for it (i.e.
Satoshi used to have a flag), which are not met in this context. As expected, then,
the resti- tutive readings for these two frames pattern with the contrasts we saw in
section 1 (see (18)).
The question is, then, how to derive the restitutive reading of the NP+PP
frame for these verbs. We assume, once more following section 1, that the for+PP is
some kind of adjunct modifier. It does not figure into the derivation of the restitutive
reading at all. Thus we expect that the sentence with the for+PP has the same type
of restitutive reading as the same sentence without the for+PP, and this is indeed
the case:
(78) a. Thilo sewed a flag again.
b. Thilo sewed a flag, and there had been a flag before.
The adjunct status of the for-phrase is confirmed by the fact that (79), where
again falls outside the constituent containing the for-phrase, allows only a repetitive
read- ing.
23
Snyder investigates to-datives and put-locatives, but not for-benefactives.
 SIGRID BECK & KYLE 2

(79) Thilo sewed a flag for Satoshi again.

The lexical semantics of sew, and its decomposition in the syntax, is thus
responsi- ble for the restitutive reading. We will not reflect upon the fine structure of
creation verbs any further here.
Finally, a comment on our first example in this subsection: a sentence with
give (see (58). In this case, so far as our intuitions can tell, the restitutive reading
of the NP+PP frame is indistinguishable from the restitutive reading for the double
object frame. We predict that the derivation of the NP+PP frame follows the steps
of the send-example in (54). The predicate of events that again modifies, then, is
the one in (80), while our intuitions were described with (81).
(80) λe.ATe(the map)(Satoshi)
(81) λe.HAVEe(the map)(Satoshi)
So far as we can tell, (80) is indistinguishable from (81): that is, the map being with
Satoshi is indistinguishable from the map being in Satoshi’s possession.

4 Conclusions

Restitutive again is a detector of result state denoting constituents and a clue to their
meanings. The again-test confirms Kayne’s conjecture that double object
construc- tions have a small clause constituent in them, and identifies that head
as HAVE. It confirms that the corresponding NP+PP constructions are different,
and are not transformationally related to the double object frame. Specifically, the
lower predi- cate is not in general HAVE. It is provided by the preposition in the
case of motion verbs, and by the lexical result state of creation verbs.
The fact that those result states have to be available for 2-object verbs es-
tablishes the need for some semantic glue that relates the result state to the event
expressed by the verb (giving, sending etc.). We have speculated that the same
interpretation principle that is at work in resultative constructions is at work here
as well. This ties in with Snyder’s (2001) evidence that these constructions are
governed by the same grammatical parameter as resultatives.
It is interesting that the evidence gained from a closer look at the syntax of
these constructions, converges with the evidence gained from a closer look at
their semantics, particularly in view of the fact that the resulting analysis is not
what a na¨ıve approach to either syntax or semantics would suggest.
The differences between the semantic and syntactic evidence are also inter-
esting. In order to make the interpretation of these syntactic representations trans-
parent, we have had to make several unconventional claims about their structure.
We have argued that in the NP+PP frame of motion verbs, the PP actually consti-
2 DOUBLE OBJECTS

tutes a small clause. And we have concluded that in this small clause, as well as in
the HAVE small clause in the double object frame, there is a silent argument that is
bound to the object of the higher verb. Putting an empty pronoun in the Specifier
of PP, as we do for the NP+PP frame, has a precedent in Heim and Kratzer (1998).
It’s also needed in the analysis of resultatives like (82) (compare von Stechow
(1995)).
(82) Thilo ground the spa¨tzle into the counter.
However, putting a silent pronoun in object position, as we’ve done in the double
object frame, is, so far as we know, novel. In neither case are we aware of a
theory of empty categories that would allow silent pronouns in these positions. We
suggest that such a theory is called for.

References

Baker, Mark. 1988. Incorporation. Chicago, Illinois: University of Chicago Press.

Barss, Andrew, and Howard Lasnik. 1986. A note on anaphora and double
objects.
Linguistic Inquiry 17:347–354.
Beck, Sigrid, and William Snyder. 2001. The resultative parameter and restitutive
‘again’. In Auditur vox sapientiae: a festschrift for Arnim von Stechow, ed.
Caroline Fery and Wolfgang Sternefeld, 48–69. Berlin: Akademie Verlag.
Bruening, Benjamin. 2001. QR obeys superiority: frozen scope and ACD. Linguis-
tic Inquiry 32:233–273.
Chomsky, Noam. 1995. The minimalist program. Cambridge, Massachusetts: MIT
Press.
Collins, Chris, and Hoskuldur Thra´insson. 1993. Object shift in double object
con- structions and the theory of case. In Papers on case and agreement II,
ed. Colin Phillips, 131–174. Cambridge, Massachusetts: Department of
Lin- guistics and Philosophy.
Davidson, Donald. 1967. The logical form of action sentences. In The logic of
decision and action, ed. Nicholas Rescher, 81–95. Pittsburgh: Pittsburgh
University Press.
Dikken, Marcel den. 1992. Particles. Dordrecht, The Netherlands: Holland Institute
of Generative Linguistics.
Dowty, David R. 1979. Word meaning and Montague grammar, volume 7. Dor-
drecht, The Netherlands: Kluwer Academic Publishers.
Fabricius-Hansen, Catherine. 2001. Wi(e)der and again(st). In Auditur vox sapien-
tiae: a festschrift for Arnim von Stechow, ed. Caroline Fery and Wolfgang
Sternefeld, 101–130. Berlin: Akademie Verlag.
SIGRID BECK & KYLE 2

Fillmore, Charles. 1965. Indirect object constructions in English and the ordering
of transformations. The Hague: Mouton and Company.
Green, Georgia M. 1974. Semantics and syntactic regularity. Bloomington, Indi-
ana: Indiana University Press.
Hale, Ken, and Samuel Jay Keyser. 1997. On the complex nature of simple pred-
icates. In Complex predicates, ed. Alex Alsina, Joan Bresnan, and Peter
Sells, 29–66. Stanford: Center for the Study of Language and Information.
Hale, Kenneth, and Samuel Jay Keyser. 1993. On argument structure and the lex-
ical expression of syntactic relations. In The view from building 20, ed.
Kenneth Hale and Samuel Jay Keyser, 53–110. Cambridge, Massachusetts:
MIT Press.
Heim, Irene, and Angelika Kratzer. 1998. Semantics in generative grammar.
Malden, MA: Blackwell.
Johnson, Kyle. 1991. Object positions. Natural Language and Linguistic Theory
9:577–636.
Kayne, Richard. 1984a. Principles of particle constructions. In Grammatical rep-
resentation, ed. Jacqueline Gue´ron, Hans-Georg Obenauer, and Jean-
Yves Pollock, 101–140. Dordrecht, The Netherlands: Foris Publications.
Kayne, Richard. 1984b. Unambiguous paths. In Connectedness and binary
branch- ing, 129–164. Dordrecht: Foris Publications.
Kratzer, Angelika. 1994. On external arguments. In Functional projections, ed.
Elena Benedicto and Jeff Runner, 103–130. Amherst: GLSA.
Larson, Richard. 1988. On the double object construction. Linguistic Inquiry
19:335–392.
Larson, Richard. 1990. Double objects revisited: reply to Jackendoff. Linguistic
Inquiry 21:589–632.
Lewis, David. 1973. Causation. The Journal of Philosophy 556–567.
Marantz, Alec. 1984. On the nature of grammatical relations. Cambridge, Mas-
sachusetts: MIT Press.
McCawley, James D. 1968. The role of semantics in a grammar. In Universals
in linguistic theory, ed. Emmon Bach and Richard Harms, 124–169. New
York: Holt, Rinehart and Winston.
Oehrle, Richard. 1976. The grammatical status of the English dative alternation.
Doctoral Dissertation, Massachusetts Institute of Technology.
Pesetsky, David. 1982. Paths and categories. Doctoral Dissertation, Massachusetts
Institute of Technology.
Pesetsky, David. 1995. Zero syntax. Cambridge, Massachusetts: MIT Press.
2 DOUBLE OBJECTS

Richards, Norvin W., III. 1997. What moves where when in which language?
Doc- toral Dissertation, Massachusetts Institute of Technology, Cambridge,
Mas- sachusetts.
Ross, John. 1974. Nouniness. In Three dimensions of linguistic theory, ed. O. Fu-
jimura, 137–257. Tokyo: TEC Corporation.
Snyder, William. 2001. On the nature of syntactic variation: evidence from complex
predicates and complex word-formation. Language 77:324–342.
von Stechow, Arnim. 1995. Lexical decomposition in syntax. In The lexicon in
the organization of language, ed. U. Egli, P. E. Pause, Ch. Schwarze, Arnim
von Stechow, and G. Wienold, 81–118. Amsterdam and Philadelphia: John
Benjamins Publishing Company.
von Stechow, Arnim. 1996. The different readings of wieder “again”: A
structural account. Journal of Semantics 13:87–138.
Takano, Yuji. 1996. Movement and parametric variation. Doctoral Dissertation,
University of California, Irvine.
Zaenen, Annie, Joan Maling, and Hoskuldur Thra´insson. 1985. Case and
gram- matical functions: the Icelandic passive. Natural Language and
Linguistic Theory 3:441–483.