Families of Constructions: The English Resultatives [?]
Adele E. Goldberg and Ray Jackendoff
1. Introduction
There has been a trend in linguistic theory to attempt to explain syntactic distributions of argument structure in terms of very broad classifications based on aspectual classification of events (a la Vendler) and temporal relations among a sentence's subevents (Van Valin 1990; XXXX). This hypothesis has been greatly extended to proposals that attempt to derive all of syntactic argument structure from aspectual or temporal considerations (e.g. Tenny 198x, Borer 1994; [Ritter, 1996 #183]). While each of us is an advocate of semantics-based approaches to grammar, each of us has criticized particular instances of the aspectual-based approach on various occasions (Goldberg 2001a, to appear; Goldberg and Ackerman 2001; Jackendoff 1996). Instead, we have advocated an approach to argument structure based more directly on the intimate details of verb semantics.
Another growing trend in approaches to argument structure is to state generalizations only in a maximally general way. Reference to very general verb classes is often assumed to be sufficient for accounting for all argument structure patterns. While we acknowledge that verb classes are important, we advocate a return to the recognition that individual verbs and individual verbal distributions are often somewhat idiosyncratic (Grimshaw 1979; Pollard and Sag 1987; Boas XX). That is, narrow generalizations exist alongside certain broad generalizations; these narrow generalizations or minor patterns also require explanation.
A sterling example of the broad-based aspectual-based approach has recently been proposed by Malka Rappaport Hovav and Beth Levin (2001) (henceforth RH & L) to account for English resultatives. We choose this analysis to focus on because it represents in many ways the best that an aspectual approach has to offer. RH & L pay careful attention to the semantics of verb classes and have based their analysis on a wealth of attested data. Their approach also epitomizes the move away from syntacticocentric accounts and toward analyses that pivot on aspectual/semantic dimensions. That is, for many years prior to this article, the authors had advocated a more purely syntactic approach to resultatives themselves (e.g. Levin and Rappaport Hovav 1991, 1995, ...) Their recent paper acknowledges counterexamples to their previous proposal and recognizes that the distribution of resultative forms follows from semantic constraints, a position we and others have separately advocated for over a decade (Goldberg 1992, 1995; Jackendoff 1990, 1997a, 2002; see also Van Valin 1990; Weschler 1997, Pustejovsky 1991…).
However, HR&L take a different view of the semantic constraints from most previous ones, emphasizing a claim that temporal aspectual structure is pivotal. They also implicitly assume that the construction is completely productive. We think it's worth presenting a comparison from our point of view; we welcome their perspective on the comparison. Moreover, we think such a comparison is more generally useful as an example of the contrast between our respective overall views on argument structure and semantics.
Let's start by establishing some terminology.
• An identifying characteristic of a resultative sentence is an AP or PP that occupies the normal position of a verbal argument but is not subcategorized by the verb, for instance the italicized phrases in (1).
(1)a.Herman hammered the metal flat.
b.The critics laughed the play off the stage.
We will call this phrase the "resultative phrase" or "RP."
• One of the NPs in the sentence is normally understood as undergoing a change of state or a motion whose endpoint is expressed by the RP; we will call this the "host" of the RP.
• A resultative may contain a direct object, in which case the RP follows the object, as in (1); we will call such cases "transitive resultatives." Or a resultative may lack a direct object, in which case the RP is immediately after the verb, as in (2); we will call these "intransitive resultatives."
(2)Intransitive resultatives
a.The pond froze solid.[RP=AP]
b.Bill rolled out of the room.[RP=PP]
•In some transitive resultatives, the direct object can be selected by the verb in the absence of an RP. In others it cannot. We'll call the former cases "selected transitive resultatives" and the latter "unselected transitive resultatives."
(3)Selected transitive resultatives
a.The gardener watered the flowers flat.[RP=AP]
[cf. The gardener watered the flowers.]
b.Bill broke the bathtub into pieces.[RP=PP]
[cf. Bill broke the bathtub.]
(4)Unselected transitive resultatives
a.They drank the pub dry.[RP=AP]
[cf. *They drank the pub.]
b.The professor talked us into a stupor.[RP=PP]
[cf. *The professor talked us.]
•A special case of unselected transitive resultatives has a reflexive object that cannot alternate with other NPs. This is often called a "fake reflexive".
(5)Fake reflexive resultatives
a.We yelled ourselves hoarse.[RP=AP]
[cf. *We yelled ourselves; *We yelled Harry hoarse]
b.Harry coughed himself into insensibility.[RP=PP]
[cf. *Harry coughed himself; *Harry coughed us into insensibility]
With this terminology in mind, we can proceed.
2. RH & L's approach
In their earlier work, RH & L appear to be addressing essentially the following two questions about resultatives:
(6)a.Why, in transitive resultatives, is the host of the RP invariably the object? For example, in (1) the metal becomes flat, not Herman, and the play, not the critics, ends up off the stage.
b.Why are intransitive resultatives possible only with unaccusative verbs? For instance, freeze and roll in (2) are unaccusative. But the unergatives yell and cough cannot form intransitive resultatives (*We yelled hoarse; *Harry coughed into insensibility); rather they form fake reflexive resultatives as in (5). (Hence the term "fake reflexive.")
RH & L's answer to these two questions took the following form in their earlier work (Levin and Rappaport Hovav 1995):
(7)a.Direct Object Restriction (DOR): The host of an RP must be an underlying object.
b.Unaccusative Hypothesis: The surface subject of an unaccusative verb is an underlying object that undergoes NP movement.
Thus unergative intransitive verbs, which have no underlying object, cannot form resultatives.
RH & L 2001 advocate abandoning this position for two reasons. First, there are transitive resultatives whose subjects host the RP:
(8)a.Bill followed the road into the forest.
b.John danced mazurkas across the room.[1]
c.We drove Highway 5 from SD to SF.
Second, some intransitive verbs of manner of motion (9) (RH & L's (16)) and sound emission (10) (RH & L's (17)) occur in both intransitive and fake reflexive forms, suggesting (implausibly on their hypothesis) that they are simultaneously unaccusative and unergative. (11), based on RH & L's (21-23), presents absolute minimal pairs.
(9)a.The dogs padded ... up the path.
b.Walk yourself into a coma.
(10)a.The elevator creaked to the ground floor.
b.If the telephone bell rang, it could ring itself silly.
(11)a.She kicked (herself) free.
b.She wriggled (herself) free.
c.It wiggled (itself) loose.
As a consequence, RH & L abandon questions (6a,b) and instead ask the following two questions (pp. 774-5):
(12)What semantic factors determine the host of the RP in a transitive resultative?
(13)"What must the semantic relation be between the event denoted by the verb and the event represented by the result XP [i.e. by the RP -- AG & RJ] in order for the result XP and the verb to be combined without the mediation of a reflexive?"
Their answer to the first question, based roughly on Jackendoff 1990, is:
(14)a.A direct object can be host of the RP if it can be understood as the "recipient of force" ("patient" in the terms of Jackendoff 1990, Goldberg 1992, 1995).
b.[Auxiliary assumption] If the direct object in a resultative is not a recipient of force, the subject can be the host of the RP. (pg 786)
Since the direct objects in (8) cannot be understood as recipients of force, these examples are free to have the subject as host of the RP, according to RH & L.
Their answer to the second question is based on the following proposed Argument Realization principle (779):
(15)"Argument-per-subevent condition: There must be at least one argument XP in the syntax per subevent in the event structure."
The Argument Realization Principle is understood to only recognize subevents that are distinct temporally; i.e., two potential subevents are only considered distinct if they do not overlap temporally completely[2]. Thus the answer to (X) is that the reflexive is required when there exist two subevents that are not coextensive in time. And indeed, in padded up the path, the padding and the motion up the path must coincide temporally; while in walk yourself into a coma, the walking may entirely precede the going into a coma.[3] More generally, according to their proposal, predications involving one subevent may be intransitive or transitive, whereas predications involving two (temporally distinct) subevents are necessarily transitive.
As clarified in their Appendix B, RH & L classify the following examples as resultatives with simple event structure, i.e., resultatives in which the action denoted by the verb and the result denoted by the RP unfold at the same time. As they put it, "the happening in the world described by ..[the] resultative is not linguistically construed as two distinct events":[4]
(16) a the pond froze solidintransitive, change-of-state verb
b Robin danced out of the roomintransitive, non-change-of-state verb
c. The wise men followed the start out of B'hem.trans. with no “recipient of force” arg.
d. We pulled the crate out of the water.trans. With “recipient of force” arg.
The following examples are classified as having complex event structure in which two subevents need not be temporally dependent:
(17) a. The joggers ran the pavement thin.unsubcategorized DO, based on intrans. verb
b) We yelled ourselves hoarse.Reflexive DO, based on intrnas. verb
c)They drank the pub dry.Unsubcategorized DO, based on trans. verb
d. The cows ate themselves sickreflexive DO, based on trans. verb
e) The critics panned the play right out of town subcategorized DO, based on trans. verb
[we need to explain this—or else move it below—it sounds like a strongly positive aspect of their account—not clear it is true except as wishful thinking on their part] A crucial part of RH & L's vision of resultatives is present in both their earlier and current accounts: the principles governing the distribution of resultatives are taken to be totally general. Nothing special has to be said about resultatives to explain their distribution; the resultative is taken to be a natural consequence of principles independently necessary for other features of English. We believe this is in fact part of the general vision of language that underlies many approaches to argument structure based on aspectual structure. As it turns out, this is a fundamental way in which we differ from RH & L, as will be seen in a moment.
Before turning to our own approach, we pause briefly to challenge several key aspects of RH & L's account. The resolution of these challenges will turn out to highlight some of the virtues of our approach.
•We cannot make sense of RH & L's term in (X), "the event represented by the resultative XP." Neither an AP or a PP represents an event. Rather, the event they are trying to refer to is either (a) the host's change of state, terminating in the state represented by the RP, or, (b) if the RP represents a spatial path, the host's motion along that path. In a classic example like Bill sneezed the handkerchief off the table, there is no word in the sentence that expresses motion and that can therefore be used to construct the subevent in question. A hallmark of our own approach is that the event structure of resultatives must be constructed by adding semantic elements not contributed by the words of the sentence.
•We find (13), the "one-overt-syntactic-argument-per-subevent condition," to be empirically problematic. RH & L standardly treat causation and caused motion as separate subevents. Hence an expression of caused motion has two subevents and therefore, according to (15), requires an obligatorily transitive verb. However, contribute is semantically causal and yet can optionally appear intransitively (She contributed to his campaign). Verbs of bodily emission also allow the theme argument to be unexpressed (She sneezed onto her computer screen). Moreover in constructions such as the passive, causal verbs routinely appear without two overt arguments. Finally, in Korean, Japanese, Hindi, Hungarian and many other languages, any argument which is given and non focal may be left unexpressed; thus complex events are routinely expressed with none or only one overt argument.[5] Goldberg (2001, to appear) and Goldberg and Ackerman (2001) discuss these and other exceptions in some detail.[6] The one-argument-per-subevent condition is what is claimed to predict the presence of the “fake” object reflexives. Thus its empirical validity is crucial to the account.
•The recognition of the follow-type expressions in (8) is what motivated the move away from a syntactic account (in terms of the DOR) and toward an aspectual account. However, the generalization intended to account for the expressions in question is much too weak. As stated in (X), they claim that if resultatives apply to the “recipient of force”; if there is no “recipient of force” then the resultative can apply to any argument (pg 786). However, there are a multitude of transitive verbs in which the direct object is not a recipient of force including receive, own, sing (a song), and yet none of them allows a resultative phrase at all, either predicated of the subject or the object:
(18) a. *She received the package into the hall. (intended, “She moved into the hall in order to receive the package”)
b. *She owned the stock rich. (intended, “She became rich from owning the stock”)
c. *She sang songs hoarse. (intended, “She sang songs until she became hoarse).
Something more needs to be said about the follow-type expressions, beyond the fact that they do not involve a “recipient of force.”
•Though RH & L attempt to explain the distribution of resultatives and syntactic structure based on semantic principles, their account assumes that the examples in (16a-d) form some sort of natural class as distinct from the examples in (17). However, the examples in (16a-d) include intransitives (16a-b) as well as transitives (16c-d), verbs that have a “recipient of force” argument (16c) as well as verbs that don’t (16d), and resultative phrases that are predicated of the subject (16a,b,d) as well as resultatives that are predicated of the object argument (16c).
The class of expressions in (17) likewise fails to hang together on any particular dimension that distinguishes it from the set of examples in (16). Expressions (17a-d) involve unsubcategorized objects, but expression (17e), The critics panned the play right out of town, does not. All of the examples in (17) are transitive, but so are examples (16c, d). All of the examples in (17) are object-oriented resultatives, but so is (16d).
RH&L suggest that all of the examples in (16) are non-causal where as all of the examples in (17) are causal; however, it is unclear why example (16d, We pulled the crate out of the water) is considered non-causal; notice it is naturally paraphrased by “we caused the crate to move out of the water by pulling.”
In any case, in order to defend the idea that the temporal dimension explains the relationship between syntax and semantics, one would presumably want to claim that it correlates with some general syntactic dimension of variation. But we see that it correlates neither with inherent transitivity of the verb, nor with overt syntactic transitivity of the resultative expression, nor with whether the resultative is subject-oriented or object-oriented. In fact it only distinguishes “fake” object resultatives from intransitive resultatives, a generalization that accounts for only a proper subset of data from (16) and (17). Other formal attributes of resultatives, such as their overt transitivity are not addressed.
In fact, even the weak correlation between the proposed temporal dimension and whether a reflexive is required, is not without exception, as discussed below.
• RH & L assume that all intransitive resultatives involve events in which the action denoted by the verb and the resultant state denoted by the RP are temporally coextensive. The following examples raise doubts as to the empirical validity of the claim:
(19)A.She tripped into the ravine.
b.Milk spilled all over the floor.
It would seem that the tripping entirely precedes the movement into the ravine in (e) and that the spilling event likewise entirely precedes the movement of the milk all over the floor in (f). Conversely, there exist certain “fake” reflexive expressions that seem to entail a tight temporal connection between subevents. The example in (c) for example cannot be used to refer to a situation in which Alice ate a meal and later became sick, but can only be used to mean that Alice ate continuously until she reached a point of feeling sick (Goldberg 1995: 194):
c. Alice ate herself sick.
•RH & L assert (783, n. 15 that "the second subevent in a resultative's event structure must be telic." In point of fact, however, resultatives that express spatial motion of the host, the resultative is telic if and only if the RP is end-bounded:
(20)End-bounded spatial PPs:
a.Bill floated into the cave (*for hours [on non-repetitive reading]).
b.Bill pushed Harry off the sofa (*for hours [on non-repetitive reading]).
Non-end-bounded spatial PPs:
c.Bill floated down the river (for hours [non-repetitive]).
d.Bill pushed Harry through the forest (for hours [non-repetitive]).
While APs normally involve non-gradable adjectives denoting endstates of change (Goldberg 1991, 1995; Weschler XX), some APs do not, and when serving as RPs they create atelic resultatives:
(21)Non-end-bounded state of change denoted by an AP:
a.Bill heated the mixture hotter and hotter (for hours [non-repetitive]).
b.Bill hammered the metal flatter and flatter (for hours [non-repetitive]).
Jackendoff (1996) formalizes in detail the correlation between end-boundedness of spatial PPs and telicity of motion along the paths denoted by those PPs, and generalizes this analysis to non-end-bounded APs as well.
To summarize…
3. Our overall approach
The questions we asked about resultatives in our earliest work, and the questions we still think are most important, are these:
(22)a.What do resultatives mean?
b.How do they get their meaning, given that (as noted above), some of the meaning is not expressed by the words?
c.What licenses the RP and non-selected direct objects?
d.To what extent can the answers to (19a,b,c) be predicted on general principles, and to what extent must a speaker of English learn something special?
e.To what extent is the distribution of resultatives fully productive?
Note that RH & L's questions can be subsumed under these goals. In particular, the choice of host is at bottom an issue of meaning: which NP is asserted to change state or position? The possibility of construing a direct object as a recipient of force also is an issue of meaning.