Jump to ContentJump to Main Navigation
Lexical Semantics, Syntax, and Event Structure$

Malka Rappaport Hovav, Edit Doron, and Ivy Sichel

Print publication date: 2010

Print ISBN-13: 9780199544325

Published to Oxford Scholarship Online: May 2010

DOI: 10.1093/acprof:oso/9780199544325.001.0001

Show Summary Details
Page of

PRINTED FROM OXFORD SCHOLARSHIP ONLINE (www.oxfordscholarship.com). (c) Copyright Oxford University Press, 2019. All Rights Reserved. An individual user may print out a PDF of a single chapter of a monograph in OSO for personal use.  Subscriber: null; date: 18 October 2019

Verbs, Constructions, and Semantic Frames *

Verbs, Constructions, and Semantic Frames *

(p.39) 3 Verbs, Constructions, and Semantic Frames*
Lexical Semantics, Syntax, and Event Structure

Adele E. Goldberg (Contributor Webpage)

Oxford University Press

Abstract and Keywords

In what ways can events combine to form a single predication? This chapter argues that the only constraint on the combination of events designated by a single verb is that the events must constitute a coherent semantic frame. A verb can designate subevents that are not causally related, and a verb can specify both manner and result. On the other hand, a combination of verb and construction can designate a one‐time predication that does not correspond to an established frame.

Keywords:   frame, predication, verb meaning, constructions, subevents, argument structure

In what ways can events combine to form a single predication? In the simple case, the question becomes, what combinations of subevents can a single verb refer to? This chapter argues that the only constraint on the combination of events designated by a single verb is that the events must constitute a coherent semantic frame. A verb can designate subevents that are not causally related, and a verb can specify both manner and result, but it must constitute an established semantic frame.

When a verb meaning is combined with a meaningful argument structure construction, the resulting combination also forms a single predication. This chapter emphasizes an important difference between the two cases, however: the combination of verb and construction can designate a one‐time predication that does not correspond to an established frame. We address the slippery issue of what can count as semantic frame in the sense of Fillmore (1975, 1977a,b, 1982a,b, 1985) or Lakoff (1987) in section 3.1.3.

3.1 Semantic frames: profile and background frame

As Fillmore (1977a) wrote, ‘Meanings are relativized to frames.’ He defined a frame to be an abstraction (an ‘idealization’) of a ‘coherent individuatable perception, memory, experience, action or object’ (Fillmore 1977b). To count as ‘coherent’ and ‘individuatable’, the situation or experience must be (p.40) construable as a unit. Frames are relevantly the same as Lakoff's (1987) idealized cognitive models (ICMs). See Petruck (1996) for an excellent overview of frame semantics.

As discussed below in section 3.2, each word sense evokes an established semantic frame. Within the frame, it is useful to distinguish a word sense's profile (Langacker 1987:118) from the rest of the frame, and we can refer to the non‐profiled aspect of a frame as the background frame (or base in Langacker's 1987 or Croft's 1991 terminology). A word's profile is what the word designates or asserts (if we may use ‘assert’ to describe non‐propositional meaning); its background frame is what is taken for granted or presupposed. Examples help clarify the point. Diameter's profile is the line that is designated by diameter; diameter also requires reference to a background frame that consists of a circle. The meaning of diameter thus refers to the frame consisting of a line through the center of a circle, where the line itself is singled out as the profile of the word (Langacker 1987:185). The background frame, particularly in the case of verbs, may involve complex events that are spread out over time; for example, the verb divorce presupposes a previous marriage event as part of its background frame, while it profiles or asserts a legal dissolution of the marriage.

  1. (1)

    1. a. A word sense's semantic frame (what the word ‘means’ or ‘evokes’) = profile + background frame

    2. b. A word sense's profile: what the word designates, asserts

    3. c. A word sense's background frame: what the word takes for granted, presupposes

A test for whether an aspect of meaning is profiled or part of the background frame is the traditional test for assertion vs. presupposition. Only background frames (or presuppositions) are constant under (non‐metalinguistic) negation. For example, the following negated sentences leave the background frames intact:

  1. (2) They didn't divorce. (They are still married.)

  2. (3) That's not the diameter. (It is the circumference.)

The test indicates that being married is part of the background frame of divorce, and a circle is part of the background frame of diameter. A background frame (or presupposition) can only be negated using metalinguistic negation, signaled by heavy lexical stress on the word that evokes that frame:

  1. (4) They didn't DIVORCE, they weren't even married.

  2. (5) That's not the DIAMETER, it's not even a circle!

(p.41) 3.2 Verbs

In the case of verbs, we can say that an (underived) verb sense corresponds to a semantic frame of predication where such a frame can be defined as follows:

A semantic frame of predication: a generalized, possibly complex state or event that constitutes a cultural unit.1 Certain aspects of the semantic frame are profiled; the rest constitutes the background frame.

Consider verbs that profile that a human is somehow taken away from a situation or place. That might seem to be a very specific scenario, unlikely to warrant more than one or two dedicated lexical items, but there are actually an abundance of English verbs for variations of this scenario. Consider the surely incomplete list shown in Table 3.1.

Several of these words imply that the removed person has somehow transgressed; they differ from one another in various ways, for example, in terms of which organization or place the person is removed from. To banish is to remove a person from society; to expel (in one sense of the word) is to remove from a school; to deport is to remove from a country; to fire is to remove a person from their job.

Lay off differs from fire in that there's no impugnation of the person who has to leave their job. A person who is blacklisted is barred from some type of work, often unfairly and for political reasons. People can be ousted only if they are political figures.

One way to view the relationships among this set of verbs is that all of them share the same basic profile (removing a person from a location), but each differs in the particulars of the rich background frames involved. As Fillmore (1977) and Langacker (1987) have discussed, reference to frames (again, bases for Langacker/Croft) can be used to illuminate meaning relationships

Table 3.1 Partial list of verbs that designate the removal of a person from a situation or place







cast out





lay off



(p.42) between words in this way. Fillmore has cited the examples of land vs. ground that also seem to share the same profile (area of solid earth) but differ in their background frames since the background frame of land involves sea and the background frame of ground involves air. Athlete vs. jock, lazy vs. laid back, washroom vs. restroom, and father vs. dad, are arguably other examples of words that share the same profile but differ in their background frames.

Conversely, other word senses share the same background frame but differ in what they profile. For example, lease and renter profile different aspects of the same semantic frame, as do teacher and student, soccer ball, and soccer net (cf. also Fillmore 1977, 1985).

3.3 Previously proposed constraints on a verb meaning

Are there constraints on what can serve as a semantic frame for a verb sense? Or is the internal complexity of a verb's meaning only constrained by whether the combination of subevents is viewed as a cultural unit? In sections 3.3.1 and 3.3.2 we consider previously proposed constraints, and conclude in section 3.3.3 that only a Conventional Frame constraint seems to be operative.

3.3.1 Exclusively causally related subevents?

One proposal for a constraint on the possible semantics of verbs comes from Croft (1991; cf. also 2005). He suggests that ‘a possible verb must have a continuous segment of the causal chain in the event ICM [idealized cognitive model, aka frame] as its profile and as its base’ (1991:20). That is, verbs are claimed to designate only simple events, or complex events in which the subevents are causally related, regardless of whether those events are part of the profile or background frame (base).

Distinct subevents

It is not always easy to decide what should count as distinct subevents in a single lexical item's designation. Should we construe sauté as involving a heat‐with‐a‐small‐amount‐of‐fat subevent and separate stirring subevent? We cannot use the fact that the situation can be described by a single word without begging the question we are aiming to address: can a single verb designate two causally unrelated subevents?

For present purposes, I adopt a conservative criterion for distinguishing subevents. A predicate involves two distinct subevents if and only if there are two independently describable aspects of what is designated by the predicate that do not entirely overlap temporally: (p.43)

  1. (6) Two events e1 and e2 are distinct subevents of an event E designated by a verb V, iff E → e1 & e2, and e1 is not completely within the temporal extent of e2.

According to this definition, heating and stirring do not constitute distinct subevents of sauté since they overlap temporally: the stirring is completely within the temporal duration of the heating (the stirring may continue beyond the heating, but it is no longer sautéing once the pan is removed from the heat).

Causal relationship

Another question arises as to what should count as a causal relationship. Does having a predisposition to get cancer cause one to get cancer? It is not clear that we should expect any categorical definition for ‘causation’ since such definitions are rarely available in any domain (Rosch 1975; Lakoff 1987; cf. also Croft 1991; Espenson 1991 for relevant discussion in the domain of causation in particular). The general issue of causation has been debated for centuries, and we are not likely to get to the bottom of it here. At the same time, it is possible to make some progress on the questions we set out to address by focusing on fairly clear cut cases.

In what follows, I will consider any event that is construed to be sufficient to lead to a new state or event to be a cause.

  1. (7) E1causese2 if and only if e1 is sufficient to lead to e2.

According to this definition, being predisposed to get cancer does not strictly speaking cause cancer because, while it may turn out to be a necessary condition, it is by hypothesis not a sufficient condition.

Causally related and profiled subevents

Predicates that designate both an activity and the endstate of that activity—accomplishments—satisfy the criteria of involving two subevents that are causally related (Dowty 1979; Vendler 1967). The activity and the resulting state count as two distinct subevents because the resulting state does not completely overlap temporally with the activity. The two subevents are causally related because the activity is sufficient to bring about the change of state. Examples include smash ‘to direct force on a rigid object with the result that the object breaks into many pieces’ and fill ‘to infuse until full’. This analysis of accomplishments is in accord with longstanding and widespread assumptions about this type of predicate (see e.g. Gruber 1965; McCawley 1968; Dowty 1979; Pustejovsky 1991; Grimshaw and Vikner 1993; Rappaport Hovav and Levin 1998).

(p.44) Accomplishments are not necessarily telic: they generally allow atelic aspectual uses (e.g. 8a), and even perfective uses do not necessarily result in an end‐of‐scale endstate (8b) (Filip, 2004, 2008; Goldberg and Jackendoff, 2004; Jackendoff, 1996; Levin, 2000; Levin and Rappaport Hovav, 2005; Rappaport Hovav, 2006; cf. Rappaport Hovav and Levin, this volume, for discussion and analysis of accomplishments in terms of scalarity):2

  1. (8)

    1. a. The cousins were filling the tub with sand for hours. (atelic use of fill; the tub may not be full)

    2. b. Dave cooled the pop‐tart so that Odessa could eat it. (the pop‐tart may not be cool, just cooler than it was).

We may leave the issue of how accomplishment predicates are best defined aside for present purposes. Clear cases in which verbal meanings involve non‐causally related subevents are discussed below.

Verbs that have non‐causally related subevents

Croft's (1991) proposed constraint cannot be correct as it stands, since there exist many verbs whose profiled event is not causally related to an event that is part of its background frame. Consider the verb return or the verb appeal as in (9a,b):

  1. (9)

    1. a. Obama and Clinton returned from the campaign trail to vote. www.opencongress.org/bill/110‐h2381/

    2. b. He appealed the verdict. mercury.websitewelcome.com/3

To return from a place presupposes that the place was travelled to previously, that it was not a point of origin. The subsequent return is not caused by the previous move; the two are not causally related. Appeal in (9) presupposes a complex background frame involving a trial which resulted in a verdict of culpability, and profiles a subsequent act of filing legal papers for the purpose of a retrial; the two subevents are not causally related: one does not cause the other, nor vice versa. At the same time, we have as part of our world knowledge the understanding that entities often move away and then back (p.45) again, legal decisions that result in verdicts of culpability may be retried; return and appeal give names to these complex frames of experience, profiling the movement back and filing of legal papers, respectively. Another example is double‐cross as in (10):

  1. (10) ‘Ruthless casino owner Willy Bank…double‐crossed Danny Ocean's friend and mentor Rueben.’ videoeta.com/movie/81764

Double‐cross profiles an event of betrayal following a state or event of understood cooperation. The betrayal is not caused by the state of trust, nor does the betrayal cause the state of trust. Instead the state of trust is part of the background frame that is presupposed in order for the profiled or asserted act to count as double‐crossing.

The verbs appeal and double‐cross (also divorce) profile one subevent while their background frames presuppose one or more other subevents, without a causal relation between them. Are there verbs that profile two subevents that are not causally related?

Verbs that have non‐causally related and profiled subevents

Verbs that profile two or more non‐causally related subevents are somewhat harder to find, yet candidates exist. For example, the cooking term, blanch, refers to immersing food, such as tomatoes, briefly in boiling water, then in cold water (in order to remove skin or heighten colour). Meat that is braised is first browned by being seared with a small amount of fat, and then cooked in moist heat. Two non‐causally related subevents are profiled by each of these verbs, insofar as either or both subevents may be negated by (non‐metalinguistic) negation:

  1. (11)

    1. a. He didn't blanch the tomatoes, he only dipped them in boiling/cold water.

    2. b. He didn't blanch the tomatoes, he peeled them with a knife.

  2. (12)

    1. a. She didn't braise the meat, she only steamed it/browned it.

    2. b. She didn't braise the meat, she roasted it.

These cases indicate that subevents involved in a lexical item's profile itself can be non‐causally related.

Croft (1991) had offered the example of ‘spinning and getting hot’ as an impossible meaning for a verb since the two subevents are not causally related. But such a meaning is only impossible if there is no semantic frame that relates these two events. If we can imagine some kind of superstitious ritual in which a ball is spun rapidly on a turntable in an oven until the ball bursts (the time until (p.46) bursting taken to indicate, for example, the length of a pregnancy), then it is not hard to imagine giving a name to his process, e.g. The guru hotspun the ball. In fact there is a verb used in pottery‐making, jiggering, which refers to bringing a shaped tool into contact with clay while the clay is spinning on a pottery wheel.

To summarize, many verbs designate causally linked subevents (smash, fill, etc.). Other verbs, however, do not involve a causal sequence of subevents. Some of these cases involve a sequence of subevents in which one or more subevent constitutes part of the background frame for another profiled subevent (e.g. return, appeal, double‐cross). The cooking terms blanch and braise profile two non‐causally related subevents.

3.3.2 Exclusively manner or result/change of location?

Levin and Rappaport Hovav (2006; Rappaport Hovav and Levin, this volume) suggest a different systematic sort of lexical gap: namely that the specification of both manner, and result or change of location by a single verb is disallowed. Examples of manner and result verbs (culled from Rappaport Hovav and Levin, this volume) are provided below:

Manner verbs (designating a non‐scalar change): amble, dance, flap, flutter, laugh, nibble, rub, rumble scribble, sweep, run, swim, scrub, tango, wipe

Result verbs (designating a scalar change): advance, ascend, arrive, break, clean, clear, cover, crack, depart, die, enter, empty, exit, faint, fill, freeze, kill, melt, open, leave, reach, recede, return

Rappaport Hovav and Levin (this volume) clarify that result verbs need not be telic; instead, the critical factor is that the predicate be scalar. There must be a single dimension with an ordered series of values in order for a predicate to be considered scalar. Scalarity underlies both change of state verbs and directed motion verbs in a straightforward way. Verbs that lexicalize scales with two points are also considered scalar; this allows achievement verbs (e.g. crack, arrive) to be assimilated to other result and change of location verbs. Activity predicates are designated as manner predicates, and are defined as dynamic verbs that designate non‐scalar change (lacking an ordering relation).

In principle a single verb could designate both a scalar change and a different non‐scalar relation. Rappaport Hovav and Levin argue that this does not happen.

Their suggested tests for manner and result are grammatical. Only manner predicates are said to allow unspecified and nonsubcategorized direct objects in non‐modal, non‐habitual sentences (13a,b). Result verbs are said to (p.47) obligatorily require their patient arguments (14a,b). Examples (13) and (14) are Rappaport Hovav and Levin's (this volume) examples 2 and 3:

  1. (13)

    1. a. Kim scrubbed all morning. (manner)

    2. b. Kim scrubbed her fingers raw.

  2. (14)

    1. a. *The toddler broke. (result)

    2. b. *The toddler broke his hands bloody.

As discussed in Goldberg (2005), although result‐oriented verbs tend to require their patient arguments except in restricted contexts, while manner‐oriented verbs need not, the correlation is not exceptionless.4 For example, kill and murder are generally regarded as result verbs and yet they can appear with an unspecified (15a) or non‐subcategorized (15b) object in the right contexts, which need not necessarily be modal or habitual. Also, non‐scalar activity (‘manner’) predicates sometimes require their subcategorized objects (16a, b) (see Goldberg 2005, for motivation for the tendency and for the exceptional cases).

  1. (15)

    1. a. The tiger killed again. (result)

    2. b. The young man murdered his way into the gang by killing the group's nemesis.

  2. (16)

    1. a. *She touched. (manner)

    2. b. *She touched her hands bloody.

Note that if the tests themselves are used to determine which verbs are result verbs and which are manner verbs, then the claim that a verb cannot be both manner and result would be rendered a tautology: a verb either can or cannot be used without a subcategorized object. Therefore in the following, we rely on an analysis of the semantics of the verbs involved to determine whether any verbs can encode both manner (a non‐scalar activity) and result (a directed change).

(p.48) It seems to be generally true that verbs of motion tend to divide between those that designate manner and those that designate a change of location (cf. also Talmy 1985). One can walk, run, jog in place and one can ascend, descend without specifying any particular manner. This may typically be true because the manner of motion and the direction of motion are generally very independent: one can walk uphill, downhill, sideways and in place.

However, the generalization is not exceptionless. Scale as in She scaled the mountain implies moving upward in a particular (full‐bodied) manner. The ski term, to schuss means to ski straight downhill (directed change of location) intentionally and very fast (manner).5

Climb would also seem to violate a constraint against manner and direction co‐occurring, since in its prototypical sense, it entails both directed motion (upward) and manner (clambering). RH&L note that climb can be used without directed motion (as a manner verb, 17a) or without manner (as a directed motion verb, 17b), and this is true. Still, its third and prototypical sense requires both these entailments (17c) (Fillmore 1982b; Jackendoff 1985). Example (17c) would be false if Kelly were simply hoisted up the tree via a rope without moving in a clambering manner or if she were to descend the tree.

  1. (17)

    1. a. The child climbed down the stairs.

    2. b. The plane climbed 1000 feet.

    3. c. Kelly climbed the tree. (RH&L, this volume, example 28)

RH&L (this volume) argue that the upward direction entailed by examples like (17c) is due to the choice of direct object argument. But upward motion is entailed by all transitive uses of climb (Jackendoff 1985). In (18a), adapted from RH&L 2007, the fence must be higher than knee deep so that upward motion is possible; the implication that she moves over the fence is only an implication, since the sentence is acceptable if she merely perches herself on top of the fence. Likewise, stairs can be ascended or descended, but if one climbs the stairs (18b), the motion must be upward. Non‐transitive uses of climb can imply motion in other overtly specified directions (e.g. 18c), but the upward entailment with transitive uses is a fact about climb. It is not a fact (p.49) about the transitive construction, since clearly, the transitive construction readily appears without any implication of upward motion (18d):

  1. (18)

    1. a. She climbed the fence in high heels.

      greylily.wordpress.com/2008/ 06/21/

    2. b. He climbed the stairs to his room.


    3. c. A tenth firefighter…climbed down the ladder.


    4. d. We descended the wall.


It seems manner and change of location are allowed to combine in certain terms because the two facets tend to co‐occur as a single culturally recognized unit. In these cases, the manner is often dependent on the type of change of location: one cannot schuss uphill; to move up a steep mountain, one's body is likely to move in particular ways (cf. scale, climb).

The constraint against designating both manner and result might appear to hold of verbs like write and scribble, where write requires that something contentful comes to exist while scribble designates a manner without specifying a result. But this analysis hinges on what counts as a result. Scribble does entail that some sort of written form is created, so perhaps this verb should count as designating both a manner and result. The difference between scrawl and jot down would seem to involve the fact that the former entails that the writing is done quickly and sloppily (manner), while the latter entails the writing is done quickly but without necessarily being sloppy (different manner); both scrawl and jot down also imply that written words were created, which would appear to be a result. Thus these verbs also appear to be counterexamples to a constraint against encoding both manner and result.

In fact, verbs of creation generally allow both manner and result, since the creation itself is type of result. The difference between manufacture and create can be attributed to the fact that manufacture entails something about the manner of creation: the entity is created by some sort of machinery or systematic division of labour. The differences among verbs of idea‐formation would also seem to involve differences in manner; for example, concoct, contrive, scheme, invent, conceive, hatch, dream up, formulate differ in whether the process takes time (concoct, scheme), whether the process is effortful (scheme) or not (dream up) among other more subtle distinctions. Verbs of cooking also would seem to often designate both a manner and a result. For example, the difference between sauté, roast, fry, and stew would seem to involve the manner of cooking and yet there is arguably a directed change as well, as the (p.50) concoction becomes sautéed, fried, or stewed. Each of these verbs pass the suggested test for manner verbs in that they may appear without their normal direct object arguments:

  1. (19) The harried housewife sautéd/roasted/fried/stewed herself sick.

Are these verbs in fact scalar? While gradability is not required for scalarity according to Rappaport Hovav and Levin's definition, since two‐point achievement verbs are defined to be scalar, verbs that are gradable would seem to necessarily be scalar. As food can be more or less fried, fry appears to be a counterexample to the claim of manner/result complementarity.

Perhaps additional criteria for counting as manner and result are required. Further clarification of these terms may be needed in order to evaluate the proposal fully (see Rappaport Hovav and Levin 2007, 2008 and this volume for discussion of some other possible counterexamples).

3.3.3 Verb meanings must evoke established semantic frames

We have seen that there appear to be exceptions to strong constraints on what can count as a verb meaning, at least as the constraints have currently been formulated. Yet there is, at least, a Conventional Frame constraint:

Conventional Frame constraint: For a situation to be labelled by a verb, the situation or experience may be hypothetical or historical and need not be directly experienced, but it is necessary that the situation or experience evoke a cultural unit that is familiar and relevant to those who use the word.

Clearly one need never be banished in order to use the word, banish. But speakers would not use the label unless they were, and expected those they were speaking with, to be familiar with the frame associated with banishing. That is, if a situation were wholly unfamiliar to speakers of a language, then there would be no frame for the situation and no corresponding label for the situation. The conventional frame constraint does not require stipulation, as it follows from principles of cooperative communication (e.g. Grice 1975).6

It appears that the only constraint is that a single verb's meaning cannot involve two or more subevents that are not related by a semantic frame, just as Fillmore (1977) had long ago proposed. Any semantic frame offers the potential for a lexical label. At the same time, the existence of a frame is not sufficient for the existence of a word meaning.

(p.51) 3.3.4 The existence of a frame does not entail that a verb exists to label it

Not all recurrent aspects of experience happen to be labelled. Although we are all aware that people are sometimes forcibly removed from bars, there is no verb that designates this action (although there is a word for the person who performs this action: bouncer). Many of these gaps do not appear to have a ready explanation. For example, while we have a verb dine that captures the complex event of eating out at a nice restaurant, we have no corresponding verb specifically for eating at a more casual restaurant.

We saw that individual verbs often (although not always) evoke causally linked subevents. But the existence of causally linked subevents is not sufficient for an event to be labelled by a verb, even if the complex situation is a familiar, regularly occurring one. For example, the subevents of feeling warm and opening a window to allow in a cool breeze do not form a scenario that is named by a single verb. (To air out is not such a verb because houses are not aired out because their occupants are warm). Some of us regularly miss deadlines, and yet there is no single lexical item that designates this failure. Different languages can be expected to have different lexical gaps, since the gaps are idiosyncratic (see e.g. Majid and Bowerman 2007).

Thus verb meanings correspond to semantic frames of predication, which designate generalized, possibly complex states or events that constitute cultural units. The subevents within a semantic frame need not be causally related, and at least occasionally designate both a manner and result. But the subevents must combine to designate a coherent, familiar situation or experience that constitutes a cultural unit. In what follows, we will see that combinations of verb and construction are not subject to the Conventional Frame constraint. But before we focus on combining verb with constructional meanings, we need to review the notion of constructional meaning.

3.4 Predications designated by combinations of verb and construction

There is a growing consensus that it is important to distinguish a verb's inherent or ‘core’ lexical semantics from the semantics associated with the grammatical structures in which the verb can occur (e.g. Jackendoff 1990; Goldberg 1989, 1992, 1995; Rappaport Hovav and Levin 1998, 1999). That is, simple sentence types are directly correlated with semantic structures. For example, in English we find the correspondences shown in Table 3.2.7


Table 3.2 English argument structure constructions


(Subj) V Obj1 Obj2



(Subj) V Obj Oblique



(Subj) V Obj Pred



(Subj) V Obj


Way construction

(Subji) V [possi way] Oblique


Linking generalizations on this view are statements about argument structure, constructions; as we've seen, individual verbs typically code much richer, more complex frame semantic meanings.

Each argument structure construction in Table 3.2 itself designates a general, very abstract semantic frame: transfer of something from one person to another, motion of something to a new location, causation of a state change, directed action and motion along a path. In fact, in the cases of argument structure constructions, the frames involved are basic and very commonly experienced; otherwise they would not be frequent enough to be considered argument structure constructions. Argument structure constructions are generalizations over multiple verbs; this insures that they are fairly general. As I put it in earlier work, ‘Simple clause constructions are associated directly with semantic structures that reflect scenes that are basic to human experience’ (Goldberg 1995:5).

Since the meanings in Table 3.2 are so abstract, it might seem that the semantic frames associated with constructional meanings are all profile and no background frame. In fact, these abstract predicates do contain little in the way of background frame, presumably because, again, constructional meanings arise from generalizing over many different verbs. So in many cases, the semantic frame consists only of the profiled relation. But this is not always the case. The way construction is used to convey the creation of a path and movement along that path (Goldberg 1995); in the case of this construction, only the motion is profiled—the creation of the path (implying motion despite difficulties or obstacles) is part of its background frame. This is evidenced by the fact that metalinguistic negation (involving particularly (p.53) stressed elements) is required to negate the presupposition that the motion was difficult in some way:8

  1. (20)

    1. a. #She didn't make her way into the room, there was a clear path ready for her.

    2. b. She didn't MAKE her WAY into the room, there was a clear path ready for her. (ok to negate the backgrounded creation of a path with metalinguistic negation)

Once we acknowledge that verbs and constructions can each convey meaning, a question arises as to whether there are any general semantic constraints on their combination.

3.4.1 Constraints on combinations of verb and construction

It is clear that the most common and prototypical case is one in which the verb and the construction do not designate two separate events. Rather the verb designates the same event that the construction designates, or the verb elaborates the constructional meaning. For example, if we assume that the ditransitive construction has roughly the meaning of transfer, ‘X CAUSES Y to RECEIVE Z’ then it is clear that the verb give lexically codes this meaning. The verbs hand and mail lexically elaborate, or further specify, this meaning. More interestingly for the present purposes are cases wherein the verb does not itself lexically designate the meaning associated with the construction, in which case we have two distinguishable events.9

Causally linked events

Events can be causally related by specifying the means, the result or the instrument involved in some act. A common pattern in English, Chinese, and Dutch is that the verb can code the means of achieving the act designated by the construction (Talmy 1985; Goldberg 1995). This is the case in each of the following attested examples: (p.54)

  1. (21)

    1. a. I literally had to close my eyes every time they kicked him the ball. www.extremeskins.com/forums/

    2. b. I actually had a moth go up my nose once. I…coughed him out of my mouth. bikeforums.net/archive/index.php/t‐292132

    3. c. He wrote his way to freedom. books.google.com/books?isbn=1593080417

Kicking is the means of achieving transfer; coughing is the means of achieving caused‐motion; and writing is the means of achieving metaphorical motion.

Pinker (1989) discusses the following example from Talmy (1985) (cf. also Croft 1991):

  1. (22) The bottle floated into the cave.

He notes that this sentence is not felicitous in the situation in which the bottle is carried into the cave in a bowl of water. It is only acceptable in the case that the floating is the means of the bottle moving into the cave.

Goldberg (1995:62) observes that verbs of sound emission are more felicitiously used in motion constructions when the sound is construed to be a result of the motion. Consider the contrast between (23a–b) and (24a–b):

  1. (23)

    1. a. another train screeched into the station. www.wunderland.com/WTS/Andy/EmptyCity/

    2. b. a tank rumbled down the street at a high rate of speed. books.google.com/books?isbn=0312980442

  2. (24)

    1. a. % The bird screeched out of its cage. (to mean that the bird happened to make a screeching sound as it flew out of its cage.)

    2. b. % Elena rumbled down the street. (to mean that her stomach rumbled as she walked down the street.)

Notice the same verbs, screech and rumble, are used in both (23 a–b) and (24 a–b), and yet the examples in (23) are fully acceptable to all speakers, while those in (24) are rejected by many (this is indicated by ‘%’). The difference is that in the examples in (23) the sound is caused by the motion, whereas in the examples in (24) the sound is not caused by the motion, but is simply a co‐occurring event.

Verbs may also designate other aspects of causal events such as the instrument (25):

  1. (25) Gilbert wristed the ball into the back of the net. www.cstv.com/sports/

    (the wrist is the instrument of the caused motion)

(p.55) Therefore, as we saw was the case with lexical accomplishment verbs, it is possible to combine two subevents into a single predication if a causal relation holds between the two subevents.

But as in the case of lexically specified subevents, the verbs are not necessarily causally related to constructional meanings. We now consider certain non‐causal relations.


If we assume that the ditransitive construction has roughly the meaning of transfer, i.e. ‘X (intends to) CAUSE Y to RECEIVE Z’ (e.g. Goldberg 1992b), then we find that this construction allows the verb to designate a precondition of transfer, namely, the creation or preparation of the transferred entity, as for example in (26):

  1. (26) Orlando baked his sister a cake. www.englishclub.com/young‐learners/

Here the preparation of the cake is a precondition for Orlando's transferring the cake to his sister.

Similarly, for a theme to move in a direction requires a precondition that the theme be free of physical restraints. In the following attested examples involving the caused motion construction, the verb designates the precondition of removing constraints that will enable motion; the construction designates caused motion.

  1. (27)

    1. a. He freed the prisoner into the crowd, as he had been ordered. www.angelfire.com/mo/savagegardener/

    2. b. The girl… unleashed the dog into the west slums. boards.stratics.com/php‐bin/arcpub/

These cases are reminiscent of the lexicalized verbs that encode both some sort of precondition and asserted event (e.g. return, appeal, double‐cross). Thus just as verbs may encode subevents related by a causal relation or subevents in which one serves as a precondition for another, so too, can combinations of verb and construction (see Goldberg 1998 for further parallels between verbs and more complex predications). In the case of the examples in (26) and (27), both subevents are asserted, as either can be negated without metalinguistic negation, as is shown in (28):

  1. (28) Orlando didn't bake his sister a cake, he microwaved it/he slept all morning.

    He didn't free the prisoner into the crowd, the prisoner was still in chains/he kept the prisoner in lock‐up.

(p.56) We now focus on an important difference between lexicalized verbs on the one hand, and verb + construction combinations on the other. We have seen that lexicalized verbs always evoke established semantic frames. In the following section, we observe that combinations of verb and construction can instead evoke novel events.

3.4.2 Frames, verbs, and constructions

Does the Conventional Frame constraint hold of combinations of verb and construction? That is, do novel combinations of verb and construction only designate (unlabelled but) pre‐existing semantic frames?

We have seen that the ditransitive construction can be combined with verbs of creation that do not themselves designate transfer. We know that what is transferred from one person to another is often created for that purpose; thus the creation of the transferred entity is a salient precondition within our frame semantic knowledge of transferring. At the same time, we need not have established frames that involve the combined semantics of specific verbs with argument structure constructions. For example, while it is arguably the case that we do have a frame of experience in which someone bakes something for someone else, it would be a stretch to say that we have an existing semantic frame that involves microwaving something for someone. And yet we can readily say both (29) and (30):

  1. (29) I baked her a loaf of homemade apple bread.


  2. (30) I microwaved her some leftover noodles.


That is, while the combination of an abstract meaning associated with a general verb class, together with an argument structure predication does seem to require an established semantic frame, the more specific meanings that arise from the combination of an argument structure construction and a specific verb need not. To take another example, we can be said to have a general frame of knowledge involving forces that cause motion; and we also can be said to have a specific frame in which strong winds blowing may cause movement; this more specific frame is evoked when blow is combined with the caused‐motion construction as in (31). At the same time, one would be hard pressed to claim that we have an established semantic frame that involves the idea that sneezing can cause motion and yet the attested example in (32) is also acceptable: (p.57)

  1. (31) It blew the beard right off of the Captain's chin.


  2. (32) She sneezed the tube right out of her nose!

    journals.aol.com/gosso23/my‐breast‐cancer‐story/ entries/2007/09/06/moms‐in‐the‐hospital/630

The idea that verbs can combine with constructions in truly novel ways is supported by the fact that one does not find unique simple morphemes that designate the requisite meanings. I'd venture to say that no language has a unique simple morpheme e.g., snope, that specifically means ‘to move by sneezing’ and no language has a unique simple morpheme e.g., micrim, to mean to intend to give something prepared by microwaving. It is in this sense that the meanings involved are ‘implausible’ verb meanings (Goldberg 1995, 2006).

The way construction for some speakers allows the verb to designate a co‐occurring activity that is not directly related to the action designated by the construction (see Levin and Rapoport 1988; Jackendoff 1990; Goldberg 1995 for discussion of this construction). For example,

  1. (33) He wheezed his way through all 3 grades.

    (about a child with asthma, reported by Kay Bock, heard on the WILL radio station, May 7, 2003)

The relationship between verb and construction in this case is that of simple co‐occurring activity. Since we don't find underived verbs in any language that convey both metaphorical motion and some unrelated activity such as wheezing, it is safe to say that the meaning of ‘metaphorical motion while wheezing’ is not an established semantic frame.

3.5 Conclusion

This chapter has explored the question of what constitutes a unitary semantic predication. It was argued that the constraints on what a verb can mean are dependent only on the notion of semantic frame (cf. Fillmore 1977). The subevents associated with a verb's meaning need not be causally related as has been proposed (cf. blanch, braise), and may also encode both manner and result as long as there exists a semantic frame that unites both meaning components (cf. schuss, scale, fry, and scribble).

The events designated by combinations of verb and argument structure construction are in some ways parallel to subevents within a lexical item's semantic frame. When the verb and argument structure construction profile distinct events, the verbal event and the event profiled by the argument (p.58) structure relation often stand in a causal relationship. However, other sorts of relationships including precondition and co‐occurring activity also sometimes hold.

The present chapter also focuses on an important difference between verbs and combinations of verb + construction. Verbs necessarily evoke established semantic frames. Constructions also evoke established semantic frames. On the other hand, while classes of verbs are related to argument structure constructions by general, abstract frames, particular verbs may be combined with argument structure constructions to designate novel events that do not evoke any pre‐existing semantic frame.


(*) I would like to thank Jean‐Pierre Koenig for discussion on this topic, and Sarah Berson, Malka Rappaport Hovav, Ray Jackendoff, and Beth Levin for very helpful comments on an earlier draft. Some of the issues addressed here are discussed in compact form in Goldberg (1995:61–5). An earlier analysis has also appeared as Goldberg (1998). The analysis here supersedes those analyses.

(1) For discussion and clarification of the notion of a ‘cultural unit’ see e.g. Enfield (2002).

(2) Further complicating our understanding of accomplishments, in Chinese and Thai the resultant state is often only an implication for many of utterances that would translate as involving entailed endstates in English (Singh 1991; Smith 1997; Talmy 2000; Koenig and Muansuwan 2002).

(3) In what follows, I attempt to use attested examples from the web where possible. At the same time, in order to avoid including irrelevant, lengthy and potentially distracting reading material, I have opted to shave the example sentences down, omitting continuations of various sorts. I have also taken the liberty of inserting periods and correcting spelling where appropriate. I also do include constructed examples when minimally different examples are important to the exposition, or of course, when the sentence in question is ungrammatical.

(4) The correlation also must be relativized to English, since many languages allow any argument to be unexpressed as long as it represents given and non‐contrastive information. For instance, all three arguments can be omitted in Chinese in the following conversation despite the fact that give is generally considered a result (or change of state) verb:

  1. (i.) gei3


    ‘[I] give [you] [some peach]’ (Chinese; Mok and Bryant, 2006)

The present discussion, however, concentrates on the extent to which the proposed constraint holds in English.

(5) Rappaport Hovav and Levin (personal communication, 6/18/08) argue that very fast manner is not a counterexample, because it only contributes an adverbial meaning to a scalar change without specifying a separate non‐scalar change. But the manner involved in schussing is not just speed, since a person cannot be said to schuss a mountain unless the skiing is done in a controlled manner; e.g. it is not schussing to head straight down fast because one loses control. Schuss also passes the suggested test for being a manner verb:

  1. (i.) The avid skiier schussed himself silly.

(6) At the same time, this is not to say that the conceptual categories named by words necessarily name preexisting categories for the learner at the outset of language learning. Recent research suggests that the labels serve to guide the learner's attention, enabling children to learn the conceptual categories that are used in the particular language(s) they are exposed to (Bowerman and Choi 2001).

(7) The form of such constructions abstracts over the linear order of constituents. For example, I assume the same ditransitive construction is involved when it is questioned, e.g. What did Pat give Chris? or clefted, e.g. It was a book that Pat gave Chris. I should also note that the constructional semantics given in Table 3.2 is somewhat oversimplified, since each formal pattern is typically polysemous (see Goldberg 1995, 2006 for discussion).

(8) There is a limited alternative sense of this construction in which the verb only designates a co‐occurring activity and no (possibly metaphorical) creation of a path is required (see Jackendoff 1990; Goldberg 1995).

(9) I do not rely in this case on the before‐mentioned criterion for determining distinct events. In particular, the events may be temporarily coextensive in some cases. It is clear we have distinguishable events once we recognize that one is designated by the verb, and another by the construction. Verb meaning is determined by what is common across uses of the same verb in different argument structure constructions; Argument structure meaning is determined by what is common across uses of the same argument structure construction across different verbs.