Mental Spaces and Constructional Meaning

Claudia Brugman

Center for Research in Language

0. Introduction

Sentences like those in (1) suggest that the main verb HAVE can be used as a verb of depiction of an irrealis situation or of prediction of a future situation:1
1.  a.  The play has him lonely and old when he dies.
    b.  "Dearborn has [Henry Miller] die on June 7,
        Fergusen on June 4."2
    c.  Jeane Dixon has Dan Quayle winning the 
        nomination in 1996.
    d.  Wayne Walker has the A's finishing outside of
        first in the division.
However, in this paper I will show that the "predictive" or "depictive" semantics of these sentences actually follow from more general principles for the interpretation of individuals and relations--and do so, for the most part, invisibly (in the sense of Fauconnier 1990). In this study I will employ the basic concepts and constructs of two independent traditions of linguistic description: that of the Mental Spaces framework (Fauconnier, 1985, 1990, to appear), as well as those the lexically- and construction-based theory of syntactic and semantic competence (in the sense of Fillmore and Kay in prep., Fillmore, Kay, and O'Connor 1988). I propose to show that the general semantic properties are associated with the constructions at issue as lexical properties of its head. Those constructional semantic properties, plus the "invisible" aspects of meaning which are recovered via pragmatic and real-world knowledge, account for the interpretation of these examples in a completely straightforward and elegant way. I will also discuss a construction which shares the lexical head and many of the semantic properties but differs in its syntactic properties. By contrast with the constructions exemplified in (1), this Extraposition-type construction has semantics which necessarily appeals to the creation of another space. I will thus present the conclusion that a specific complementation pattern of a verb may be a space-builder in the sense of Fauconnier 1985.

The paper is structured as follows. In section 1, I will outline the basic analysis of the four main constructions headed by the English verb HAVE. I will mention those properties which must be assigned to the constructions independent of the creation of another space. This overview is designed merely to introduce the notion that HAVE heads four constructions which show regular and productive semantic differences and have morpholexical constraints on the form of their complements corresponding to those semantic differences. In section 2, I show that a construction which appears to have a distinct semantic interpretation--with the "depictive" or "predictive" semantics exemplified in (1)--is actually just an instance of one of the four basic constructions with an added mental space. In section 3, the final data section, I will discuss a the Extraposition-type construction, whose interpretation requires the creation of a daughter space. I will conclude, in section 4, with a summary and some speculations as to the larger significance of these findings.

1. The Semantics of HAVE-Constructions

1.1. Introduction: Foundational constructs

1.1.1. Mental Spaces

Fauconnier (1985, 1988, to appear) and others have elaborated a theory of the partial possible worlds which speakers construct when talking about (or hearing about) the entities and relations of perceived or imagined worlds. These partial models, called Mental Spaces, are not specifically linguistic in nature. Rather they are a manifestation of general cognitive abilities. Mental spaces may be representations of the speaker's reality, or may be fictional or intensional, or may reflect past or future states of the "real" world.

This approach provides an account of why example (2) does not express a contradiction:

2.  In Len's painting, the girl with blue eyes has green eyes.
We understand this sentence to include reference to a girl with blue eyes "in reality" -- the R-space; to identify her counterpart in L (the partial depiction of the world "Len's painting"); and to assert about that counterpart that she has green eyes.

Example (2) contains what Fauconnier (1985) calls a "space-builder": the phrase In Len's painting identifies the existence, and the character, of a mental space distinct from R, the speaker's reality. But sentences requiring the construction of a space distinct from R need not contain a space-builder. Linguistic signals to construct a new space may be contained earlier in some stretch of discourse, or may not appear at all. (3) can be interpreted accurately without benefit of a space-builder:

3.  Humphrey Bogart gave up Mary Astor without 
    a shred of remorse.
While there is no explicit signal to do so, hearers can use their knowledge of the profession of Humphrey Bogart and Mary Astor to induce that this sentence expresses a statement about an event which takes place in a film, between their counterparts in that film-space, i.e. the characters they portray. A single sentence may contain expressions which refer to entities or relations both in R and in the daughter space -- in (2) and (3), the painting-space and the film-space respectively. It is unremarkable for speakers and hearers to "cross" from one space to another, and back, within their production or comprehension of a single sentence. This fact makes the analysis of sentences like those in (1) exceedingly simple, once we have a clear understanding of the space-independent, constructional semantics imposed on them by the verbal head HAVE.

1.1.2. Construction Grammar

The constructions which are at issue comprise a proper subset of those headed by English polyseme HAVE. They form a natural subset, since they have superficially identical structural properties and differ in well-defined ways as to their semantic properties.

I am using the expression "construction" in accordance with the usage of Fillmore and Kay (in prep.) and Fillmore, Kay and O'Connor (1988). In this usage, a construction is any pairing of formal properties with semantic properties (see also Lakoff 1986:467, and Langacker 1987:58 on the "symbolic unit"). For the examples discussed here, a grammatical construction may be thought of as equivalent to a lexical entry for its head, HAVE: one can describe the formal conditions of cooccurrence between HAVE and its complements, and one can also describe a frame-based semantics which the use of each such construction invokes. The latter is what I mean by its constructional semantics. The construction also specifies semantic properties which must hold of its complements individually (such as their semantic roles and aspectual properties of the predicational complement). Both kinds of semantic information will be relevant to the description of these data, and both are included in a lexically-headed grammatical construction as conceived by Fillmore et al.

In Brugman 1988 I argue that most HAVE-constructions fall into one of four classes. I will not recapitulate those arguments here; rather I will describe the four classes which together I will call the "basic" constructions, and I will argue further that the "depictive" and "predictive" semantics of the sentences in (1) can be derived from the "basic" constructions. These "basic" constructions are those exemplified in (4). They all share certain syntactic properties which may be schematized as in (5).3

4 . a.  I had my baby kissed by the President.
    b.  I had him climbing the walls.
    c.  Albany has an express bus running to it.
    d.  I had him bring chips to the party.
    e.  I have $5 in my pocket.
    g.  "A neighbor had at least two wives pass away 
         before anyone thought anything of it."
    h.  I had him angry the minute I walked in 
         the door.
5. [S [ NP1] [VP HAVE [ NP2 ] [ XP ] ] ]

Throughout the following discussion, I will refer to the subject noun phrase as NP1, and its referent as NP1'; similarly the postverbal nominal is NP2, and its referent NP2'. "XP" as usual is a variable over all major phrase types, any of which can appear in the third (or "predicational") complement position in all four readings; in this paper, I shall concentrate on sentences with a VP in this position. The situation or state of affairs which the syntactic sequence NP2 + XP denotes will be notated as S. In all the examples to be considered here, NP2 is construed as the notional subject of XP.

The sentences in (4) all express relationships between an individual ( NP1' ) and some situation ( S ) an event, an action, or a state.4

1.2. The basic constructions

In Brugman 1988 I argued that the sentences in (4a - l) fall into four classes, distinguishable ultimately on semantic grounds. The semantic differences are constructional--that is, there are differences of interpretation at the sentence level--and concomitant, or consequential, differences in the cooccurrence constraints placed on the complements of HAVE.

I have called the four construction types Causative, Resultant Event, Attributive, and Affecting Event. These names for the lexical entries are intended to be labels for the constructional semantics of each type. In the next four subsections, I will outline the semantic constraints on the complements in each construction and briefly talk about the constructional semantics of each.

1.2.1. Causative

The use of HAVE which is most easily isolated is the one I have labeled the Causative, exemplified in (6):

6.  a.  I had the children make dinner last night.
    b.  I had the President kiss my baby.  (cf. 4a)
    c.  I had him bring chips to the party.   (= 4e)
    d.  She has children come to her house every
         Sunday.  ( = 4f)
An often-noted semantic property of this causative construction is that it is generally used to describe a situation of indirect, and often verbal, causation: even when the causation is verbal, it usually must refer to an indirect verbal (persuasive) causation rather than an order (cf. Rader 1981). It follows from the fact that this construction denotes an act of persuasive causation that NP1' must be human. The other important semantic constraint, on the aspectual structure of the XP complement, is best discussed in the context of the next construction.

1.2.2. Resultant Event

Often confused with the Causative is the Resultant Event reading of HAVE. Examples like (7) show this semantic type, in which the existence, presence, action, appearance, or demeanor of NP1' results in the situation denoted by the sequence NP2 + XP. The resulting situation need not be intended, nor indeed desired, by NP1':
7.   [Despite my best efforts,] I had him angry the
      minute I walked in the door. 
Notice that the sort of nondeliberate instigation emphasized here by the sentence adverbial is not compatible with the semantics of the Causative construction5:
8.  #[Without knowing it,] I had the children wash 
    the dishes.
Since NP1' is not an Agent, naturally NP1 need not be Animate:
9.  The boss' foul mood had the steno pool 
     quaking in their high-heels.
Another difference in the semantic conditions placed on the complements of these two constructions concerns the aspectual type of the predicational complement. Given that a human can be the referent of the subject of either construction, but a nonhuman can be NP1' only of the Resultant Event, the following contrast is highly suggestive:
10.   a.  I had the children wash the dishes last night.
      b.  I had the children washing the dishes in no 
11.   a.  *My glowering countenance had the children 
          wash the dishes last night.
      b.   My glowering countenance had the children 
           washing the dishes in no time.
This contrast in acceptability indicates that the Causative construction requires a perfective XP complement, while the Resultant Event reading requires an imperfective XP. When XP is a perfective VP, this condition may be manifested in bare-stem infinitive morphology, and when it is imperfective, the VP will have the morphology of the present participle. (For other forms, see Brugman 1988.) The distribution of temporal adverbials is correlated with the aspectual type of the XP constituent, as shown by (10a) and (10b).

There is another formal difference between the Causative and Resultant Event constructions: under many circumstances, having NP2 coreferential with NP1 in the Causative construction renders a pragmatically marked sentence, whereas a corresponding Resultant Event sentence is unremarkable (example (12b) is an example of Resultant Event, while (13) is a Causative):

12.   a.   I had the children get out of bed promptly.
      b.   I had myself out of bed promptly.
13.    (cf.: #I had myself get out of bed promptly.)
We can readily see, then, that identity of formal properties at the first order gives way to some semantically-based differences in the conditions on the complements and modifiers of the two constructions. The similarities in their constructional meanings are also accounted for, since in both cases the XP denotes a situation resulting from some other situation, a fact which can be notated by means of giving it the same Semantic Role assignment, Resultative, in both cases.

Two other HAVE-constructions, the "Affecting Event" and "Attributive/Existential", show similarities and differences corresponding to those which characterize the Causative and Resultant Event constructions.

1.2.3. The Affecting Event

The Affecting Event construction, exemplified in (14), expresses a situation in which some event is seen as affecting NP1' (the optional phrase on me forces a "malefactive" interpretation, which is a subcase of this reading and hence precludes the Causative reading):
14.  I had my dog die (on me).
There appears to be a strong preference for Animate subjects in this construction. Note the oddness of (15):
15.  *?Main Street had four dogs die at its busiest 
       intersection last month.
This is odd even if one can imagine, for example, Main Street being affected by traffic jams at that intersection resulting from the accidents. Inanimate subjects are much more likely to occur in the Attributive construction, as I will show below.

As is the case for the Causative construction, having an Animate subject seems to cooccur with the requirement that the XP denote a Perfective situation. This accounts for the fact that sentences which most clearly provide this reading are those in which XP is a bare infinitive VP, since those are always perfective. However, the Affecting Event construction differs from the Causative in that there is a substantial difference in their frame semantics, since the XP denotes a circumstantial rather than a resulting state of affairs. There is also a difference in the Semantic Role assignment of the NP1' constituent, which is respectively Patient and Agent.

1.2.4. The Attributive

Sentences like
16.   a.  I have keloid tissue on my back.
      b.  He has a fly resting on his nose.
exemplify the construction I call "Attributive". In this construction, the state of affairs S presented as (if it were) a property of NP1'. This state of affairs is often stative, as it is in (14), but in fact may be any imperfective situation (as in (17), where XP denotes an event type repeated regularly):

17. Albany has an express bus running to it.

This construction may just as readily be used to impose an attribution to NP1' (as in (16b) ) as it does to express an objective property of NP1', as in (16a). The attribution, therefore, may impose a particular point of view on a situation (cf. Fillmore 1976).

As the examples suggest, the differences between this construction and the Affecting Event reading are twofold and interconnected: there is an Animacy requirement on NP1' in the Affecting Event construction which does not hold for the subject of the Attributive construction, and the XP complement must be Perfective in the Affecting Event construction and imperfective in the Attributive construction. Assuming the basic distinctions presented in section 1 to be correct, I will now talk about sentences which, with the addition of space-creation, can easily be assimilated to these four basic constructions.

2. The "depictive" cases: HAVE-constructions involving cross-space connectors

2.1. Introduction

In this section, I will consider examples of HAVE-sentences in which S holds in some mental space other than the origin space R (the "speaker's reality"), and NP1' brings about that situation (as it holds in that space) or else the situation is treated as a property of NP1'. The examples I will consider first are given in (18 - 27):
18.  Wayne Walker has the A's finishing outside 
       of first in their division.
19.  Jeane Dixon has Quayle winning in '96.  
20.  "Dearborn has [Henry Miller] die on June 7,
       Fergusen on June 4."
21.  Imelda's count of the votes had Ferdinand 
       being the winner.
22.  "Another scenario had thirtysomething 
       returning as a TV movie."
23.  "Lefebvre had Canseco scoring all the way." 
24.  LaRue had me bringing chips to the party.
25.  John Sayles has Happy Felsch throw 
      the 1927 World Series.
26.  The movie has him dying in the end.
27.  In The Maltese Falcon, Raymond Chandler 
      had Philip Marlowe give up Brigid 
      O'Shaughnessy without an iota of remorse.
These sentences appear to manifest a constructional semantics in which HAVE is used to convey either the prediction of a situation in a future world (as in (18) and (19) ), or where the (real-world) facts are subject to dispute or confirmation (as in (20 - 22) ), where the "world" is a declarative one (as in (24) ) or a fictional one (as in (25 - 27) ). One striking difference of these examples from those discussed in sec. 1 is that, in those earlier examples, HAVE is factive--that is, the situation S is entailed in the origin space. In the sentences (18 - 27), the corresponding situation need not, or is not known to, hold in the origin space. My claim is that, despite the superficial semantic similarities among this set of sentences and its obvious semantic distinction from the four constructions discussed in sec. 1, these do not constitute a fifth construction type. Rather, they are all instances of one or another of the four constructions plus the creation of another space.

There are both descriptive and explanatory reasons for proposing this solution, as compared to one which simply identifies a fifth HAVE-construction. We have seen one of them just above: despite the shared semantic property that S is irrealis, the kind of non-realis situation denoted varies, although there is no formal signal of that variation. (Clearly this aspect of the interpretation comes down to the kind of space created, which is a product of pragmatic principles arising from knowledge about the referents of the complements.) In this discussion I will also show that these "depictive" uses impose the same correlated constraints on complements that the four basic constructions have; also, constructional semantic differences among "depictive" sentences apart from the one just noted are explained, since these sentences really instantiate different constructions. In other words, the "depictive" uses have the syntactic and semantic properties of the four basic HAVE-constructions, with the additional condition on interpretation that they involve the building of a daughter space.

The fact that HAVE-constructions are so easily accessible to cross-space readings--readings in which a daughter space is created-- follows naturally from the semantics of those constructions: in so far as the semantics of some construction is compatible with the semantics of a space-crossing reading, that reading will be a layer of interpretation superimposed on the constructional semantics. If indeed the space-crossing aspect of the meaning is independent of the construction, it follows that all four HAVE-constructions should be able to have space-crossing instantiations; as I will show in sec. 2.4, this is in fact the case.

2.2. The constructed spaces of "depictive" sentences

In each of the examples (18-27), NP1 refers either to a created space or to a person (or institution) which can create a space. This space (no matter whether it is named directly or evoked via its "author")6 most often falls into one of the following general categories: a fictional text, a theory, or a future or hypothetical space corresponding to the origin space. Following Fauconnier 1985, I will use M to refer to any non-origin space evoked by these sentences, and R to refer to the origin space. I will call the reading in which both NP1' and S hold in R the R-space reading, and the one in which S holds only in M the M-space reading. In a sentence with an M-space reading, NP2' and XP' may or may not have counterparts in R, but in these examples NP2' and XP', and the relations between them, by definition all exist in the M-space, whatever its nature may be. In other words, NP1 evokes another space either directly or indirectly, and the situation S holds in that space.

As noted above, nothing about the sentences (18 - 26), besides pragmatic knowledge about the referent of NP1, signals that S holds in a space other than R. Most of these sentences have a strongly preferred reading involving a space M. However, sentences like (23) are ambiguous over a reading where the embedded predicate holds in R (this being the most contextually neutral reading) or in some other space. (23) was uttered by a radio announcer to describe the then third-base coach of the Oakland A's waving then right-fielder Jose Canseco into home, but actually Canseco only reached third base on the play: only by extralinguistic knowledge can we identify (23) as instantiating the "depictive" reading rather than an R-space Resultant Event reading. In Lefebvre's intention space, Canseco's counterpart reached home, doing so as a result of Lefebvre's instigating act.

Other spaces include some familiar ones: (19) is readily interpreted as involving a future hypothetical (or "predictive") space because of our encyclopedic knowledge about NP1'. ( (18) has the same predictive interpretation, Wayne Walker being a sportscaster. ) (24) can describe a situation in which the department secretary has constructed a future intended world in which everyone is to bring an assigned food. That S does not necessarily hold in R is again shown by a possible continuation of (24):

24'.  LaRue had me bringing chips to the party,
       but since everyone's on a diet I've brought 
       rice cakes instead.
The examples discussed so far and (25) and (27) involve a NP1 which denotes an Animate participant, an "author" of a space in the sense that he or she is the creator of a non-origin space in which S holds. The other examples, (21), (22), (26), and (28), show that NP1 can denote a "text" which corresponds directly to the space.
28.  The 1967 theory had COMP hanging from S.
It should be obvious that one important property of use of these "depictive" sentences is that they usually occur in a context of explicit or implicit contrast with a situation holding in R, or in another non-origin space, holding of the counterparts to NP2' and XP' in the M-space. (24') provides such an environment, where the contrast is between M (LaRue's intention space) and R. Similarly, (28) could be continued as in (28'):
28'.  The 1967 theory had COMP hanging from S,
       but by 1968 it hung from S'.

2.3. The properties of "depictive" sentences

All the examples (18 - 27) describe situations in which, in some space M, S holds as a result of the action, situation, or disposition of some individual (the counterpart of NP1') in R. As noted, some of those examples have Animate ("author") and some have Inanimate ("text") subjects. What is interesting and important for our purposes is that the properties of the four basic constructions hold for the examples we have here; that is, the examples with Animate subjects have either Perfective or Imperfective predicational complements while those with Inanimate subjects may properly have only imperfective predicational complements. This follows from the general constraint that the Resultant Event constructions may have an animate or inanimate subject while the Causative may only have an Animate subject, and that these are correlated with, respectively, imperfective and perfective predicational complements. Therefore, the examples we have seen so far are all most readily interpreted as examples of the Causative or the Resultant Event constructions with the understanding that the situation described holds in the space M, and that that space itself is invoked by NP1.

Let us look at this in more detail. In the two examples with bare-stem infinitive embedded VPs, i.e. (25) and (27), NP1 denotes an "author" who (our encyclopedic knowledge tells us) creates a text in which S holds. (25), on the intended interpretation, describes a situation in which the mention of John Sayles evokes the film Eight Men Out in virtue of his being its director and having had some control over the content of the story. John Sayles is therefore the "author", in my sense, of the text which builds the film space. In the film, Happy Felsch throws the 1927 World Series. In R, Charlie Sheen plays Happy Felsch. The ID connector allows us to use the counterpart in R to refer to any individual in M. The text may be named explicitly, as it is in (27); the in-phrase is a space-builder, in the terminology of Fauconnier 1985. However, we can get an equivalent reading of the sentence without the space-building in-phrase, assuming some knowledge of Raymond Chandler (or perhaps merely knowledge that the other two NPs denote fictional characters). In either case, I am identifying NP1 as an Agent of a caused situation. Needless to say, the kind of causing that exists in the real-world situation described in (27) or (25) is different from the causing act described using a sentence like (10a), I had the children wash the dishes (on its R-space reading). This is a consequence of the differences in the properties of NP2 in such cases; neither Philip Marlowe nor (these days) Happy Felsch is subject to being acted on in R. However, given the fact that the R-space HAVE-Causative is not appropriately used to describe either direct manipulative causation or causation from a direct order, this construction is natural for use to describe causation that takes place in virtue of a fictional world being created.

Let us now look at the other examples, those with an imperfective embedded VP. In these sentences, NP1' may be Animate or Inanimate, but either way, it is a situation, or a disposition, or an event which triggers the resulting event denoted by NP2 + XP. As we noted for the R-space examples of the Resultant Event construction, the Subject referent may bring about the event deliberately or inadvertently. That situation holds as well in the M-space sentences (19), (21), and (24):

19.  Jeane Dixon has Quayle winning in '96.  
21.  Imelda's count of the votes had Ferdinand 
      being the winner.
24.  LaRue had me bringing chips to the party.
In (19), Jeane Dixon brings about the prediction-space situation by her psychic powers and irrespective of her political leanings, so (19) describes a (non-realis) situation which results from a disposition (i.e. a power) of Dixon's. Similarly in (21): Imelda just counts the votes and from that action Ferdinand becomes the winner (in a non-realis space). Notice with this example and with (24) that the text which evokes the space may be purely declarative and created solely for the purposes of establishing that XP' holds of NP2'. Another example of this type is (29):
29.  The first-base umpire had him holding up,
      while the home-plate umpire had him 
      swinging through.  
The intended interpretation here is that there is a batter at home plate and according to the first-base umpire the batter stopped swinging before he broke his wrists (which means it doesn't count as a strike) while according to the home-plate umpire he executed a complete swing. Since the call "ball" or "strike" depends on the determination of the swing, the umpires here have a dual role: they observe and judge the situation in R, then in their individual belief spaces make a determination of the action of the batter. (Note that one event in R corresponds to one event in each of the two belief spaces and that those two events are incompatible in any single space.) In other words, the observation and judgment of the umpire bring about the determination in his belief space. So this is an example of the Resultant Event reading across spaces. Just as we saw in section 1, the Causative and Resultant Event readings have semantic properties in common, and there are situations which could be accurately and appropriately described using either of them, such as (24) and (24' '):
24.     La Rue had me bringing chips to the party 
24.''   La Rue had me bring chips to the party 
        [ , but I won't be able to come at all].
(The parenthetical forces the M-space reading since the situation S cannot hold in R.)

Interestingly, however, the M-space uses of the Causative and Resultant Event constructions do not so strictly adhere to the Animacy conditions as do those constructions describing situations holding in R.

One counterexample to the Animacy constraint on NP1 of Causatives occurs in sentences like (30):

30.  a.  The movie version has him die at the end.
     b.  ?The theory has INFL attach to S.
These are less than perfect for many speakers, but they are not as bad for anyone I know as (12a), repeated here as (31):
31.  *My glowering countenance had the children 
        wash the dishes. 
There are two possible explanations for this difference between M-space and R-space instances of the Causative. One is that a metonymy of work-for-author is operating here: since NP\d\u in these cases will always denote a work of an author, they will always provide an "author" causer of the event. An unattractive property of this solution is that the work-for-author metonymy appears rarely if at all outside this construction, so this solution lacks generality, and seems ad hoc.7

A more plausible explanation is that--whether over all or just in the M-space instance--the correct level of generalization is not with Animacy but the the directness of causation (which I have hinted at in the preceding discussions of the semantic differences between these two). Perhaps if the "author" or "text" requires that S hold in M, either can serve as subject referent of a Causative construction, whereas if the properties of the "author" or "text" merely bring S about, then that situation is compatible semantically with the Resultant Event construction.

2.4. The other HAVE-constructions and their M-space instances

If the "depictive" sentences we have seen so far can merely be assimilated under the Causative or Resultant Event construction, my analysis predicts that the Attributive and Affecting Event HAVE-constructions should also have M-space instances. This in fact is so: the Attributive reading is quite common with "text" subjects and somewhat less so with "author" subjects. The Affecting Event construction, which appears to be limited in use to Animate subjects, also finds some instances in M-space sentences. That reading is obviously pragmatically marked, probably because it is used to describe situations which are just rarer in the world. So both the existence and the distribution of these two falls out of my analysis in a way that positing a separate "depictive" construction will not do.

2.4.1. The Attributive case

Here are some examples of the Attributive construction with a cross-space reading:

32. a.  "The Standard-Theory way of representing a 
         PP would just have PP immediately 
	     dominating P and NP."
    b.  "He doesn't have himself participating in 
         the [rape and murder], he has himself 
         observing the situation."
    c.  "A folk tale in deaf culture has a man and 
         a woman sitting at different tables in a 
         restaurant.. . ."
In the examples (32), the situation S is presented as an attribute of NP1', who exists in R. (32b), for instance, was uttered by an attorney in defense of his client, who was claiming diminished capacity. The speaker claims that in the client's belief space, he is observing but not participating in the crime. (32a) says that the standard theory has the property that the mentioned situation holds in that theory.

2.4.2. A word about ambiguity

Recall my mention in sec. 1 that while the formal differences in the HAVE-constructions could be traced directly back to differences in their semantic requirements, there are semantic differences which are not formally notated in the description I have given here. In fact, many sentences are ambiguous along one or both of two dimensions. As noted in sec. 1, many individual sentences are perceived as ambiguous between two R-space readings, an example being I had the book stolen. This is because the morphosyntactic properties which distinguish constructions are not sufficient to identify the constructional semantics (Semantic Role assignment is one formal, though invisible, means of distinguishing them). The other dimension of ambiguity is along the axis of space-building. Many examples we have seen are readily given either an R-space reading or an M-space reading. But it is also possible to find a sentence which is ambiguous over two M-space readings, in a way entirely analogous with the two R-space readings available. However, sometimes it is very hard to disambiguate two M-space readings, while disambiguation between two R-space readings is often much easier in context.

The question arises whether there is actual ambiguity or mere vagueness in these examples. It is hard to decide, in some cases, whether a sentence is an example of the Attributive or the Resultant Event reading, especially when NP\d\u denotes a text. Is there really a difference in the situation holding within the world created as opposed to the situation in which the state of affairs named is a property of that text?

There do seem to be specific cases in which one reading and not the other is intended. The surrounding context of the attested (32b), for example, suggested that it was the reading in which the perceived state of affairs was a property of NP\d\u' rather than a result of some intention. The apparent vagueness or indistinguishability of the two readings in many cases follows from the lack of morphosyntactic distinguishability plus the nature of the daughter spaces themselves: since the texts are constructed by some author, we can either see them as constructed with S a property (in the "Attributive" use) or as constructed such that S follows from other properties (which would give it a "Resultant State" construal). Similarly, when NP1' is an Author, we can either see the situations holding in his belief or intent space as properties of his, or as something that holds (in that space) as a consequence of his intentional beliefs. (It is possible that the degree to which the two readings can be distinguished by the hearer is a function of the kind of M-space created.) This "Necker-style" ambiguity, in which both readings are compatible with the situation described, has been discussed in Norvig 1988, and can be found in sentences which do not involve multiple spaces.

2.4.3. The Affecting Event reading

Finally there is the Affecting Event reading to consider. It should be possible to get sentences with this reading in which S holds in M. Recall that in this construction NP1' is (portrayed as) being affected as a result of the S. Novelists occasionally report the situation in which they are taken aback by some act of the characters they have created, and this would qualify the sentence, if NP1' denotes the Author, as being of the Affecting Event type:

33.  a.  In The Maltese Falcon, Raymond Chandler 
          had Philip Marlowe give up Brigid 
          O'Shaughnessy without an iota of
          remorse.  = (25)
     b.  In my dream last night I had a pack of angry 
          wolves chase my little dog Fluffy. . .
Of course (33b) also has a reading in which NP\d\u names an individual in M who is the counterpart of the speaker in R.

3. The "Extraposition" Construction

The "extraposition" HAVE-construction must be distinguished from the ones discussed in sec. 1 on both syntactic and semantic grounds. The important part of the semantic difference, for current consideration, is that it can only be used when NP2 denotes a space-builder and the embedded predication holds in the invoked space.

The syntactic properties of this construction can be schematized as in (34), and exemplified in (35):

34.  [S [ NP1 ] [VP HAVE it [S' that [S NP2  XP ] ] ] ] 

35.  Rumor has it that Reagan's aides wanted to 
       invoke the 25th amendment.
Of course, Rumor has it is the most common and most idiomatic instance of the schema in (34). However, (34) does not simply describe the constituency of the idiom Rumor has it that S: the sentences in (36) (which seem odd for some speakers but unremarkably good for others) instantiate the same syntactic schema.
36.   a.  The grapevine has it that Nolan Ryan is
          pitching next week.
      b.   The scuttlebutt around town has it that they 
           had to get married.
      c.  "A new joke has it that after 450 years, the 
           next member of the royal family to be 
           executed in the Tower of London will be 
          Princess Michael of Kent."  
      d.  "A new theory has it that the US wanted to 
           keep the Contadora contract from being 
      e.  "The description [of verb gapping] had it 
           that a verb in a non-final conjunct gets 
           omitted. . ."
For at least some speakers, then, (34) gives the syntactic specifications of a productive construction form; it is apparent that there are concomitant semantic requirements on the lexeme filling the NP1 slot, as I will discuss briefly below.

What makes this construction interesting is that it has only an M-space reading. That is, while (37) has available both an R-space reading (in which, in R, the children clean their rooms as a result of hearing the story) and an M-space reading (in which in the story about the rampaging mother, the children clean their rooms), (37') has only the M-space reading.

37.  The story about the rampaging mother had the 
      children cleaning their rooms immediately.
37'. The story about the rampaging mother had 
      it that the children cleaned their rooms 
The examples so far have been ambiguous over the two readings, Resultant Event and Attributive. In fact neither of the other readings is available, since for most speakers, no Agent or Patient participant can be the referent of NP1:
38.  ?Glenn Dickey had it that Nolan Ryan was 
       pitching this week.
Some speakers may find this sentence acceptable on the reading that Glenn Dickey metonymically refers to his column. (He is a columnist for the San Francisco Examiner.) But in general, an "author" NP1 cannot appear in this construction if there is no actual text--that is, this construction with an "author" NP1 cannot be used to convey a belief or intention.

While this construction appears to be roughly synonymous to the basic Attributive or Resultant Event in its M-space construal, its circumstances of use are rather different. The "extraposed" version seems to be used when the veracity (in R) of the embedded predication is in doubt: that is, its implication is that the situation described in the embedded clause, holding in M, does not hold in R. (It also occurs with NP1 denoting a text which is assumed not to hold in R, as a joke.) As I noted in sec. 2, this is a frequent but not necessary feature of the use of the basic constructions in M-space readings. This amounts to a restriction on the semantics of the nouns that can appear in NP1 position; the NP must refer to a text whose veracity (or accuracy in describing a situation in R) is subject to disconfirmation or at least dispute. Note the oddness of (39):

39.  #The truth has it that the earth is round.
And notice the pragmatic effect of conveying the speaker's disbelief in the veracity of the Bible in example (40):
40.  The Bible has it that woman was made 
       from man.
Thus this "Extraposition" HAVE-construction is similar to the "basic" uses discussed in sec. 1 in that a specific complementation pattern is associated with a frame semantics just like two of the ones in the basic construction, the Resultant Event and the Attributive. However, it has an additional and unique condition on it that the situation holds only in some M-space and that space is invoked, generally by direct reference, by NP1 (and hence the referential possibilities of NP1 will constrain the possibility of this reading). Because it is only within this syntactic form for HAVE that an M-space reading is required, we can say that this (partially lexically-filled) syntactic pattern is a space-builder.

4. Summary and conclusions

To summarize, I have shown the following: that an apparently distinct construction, headed by HAVE, whose semantics is in the business of expressing a prediction or a depiction, actually is an unremarkable conjunction of independent sets of interpretive principles. For the examples discussed in sec. 2, the principles exist at two levels: first is the lexical level in which one of four readings of the polysemous lexeme HAVE is invoked and the other is at a much higher level of conceptual-semantic organization, that of mental space construction. The apparent idiosyncrasy of using HAVE as a verb of prediction dissolves when one realizes that a daughter space may be set up, with or without an explicit signal to the hearer to do so (as noted in Fauconnier 1985, 1990).

I have also shown that there is also a construction which is dedicated precisely to the depiction of some situation in M, and whose syntactic properties are unique to this reading of HAVE and are compatible only with that reading.

The case of HAVE is interesting both for Construction Grammar and for the theory of Mental Spaces. For Construction Grammar it shows that a multifunctional lexical network can have a varied internal landscape. In the case presented here, one lexical entry in the complex of HAVE must make explicit reference to the building of a space, and to the fact that some state of affairs holds in that space. On the other hand, the basic HAVE-constructions discussed here must be semantically described so that their usage is compatible with M-space readings, but not confined to them. The interesting question remains as to how to notate the difference between the two sorts of construction.

The moral of the story for the students of mental space phenomena is that a mostly skeletal construction such as HAVE-Extraposition may criterially include a space-builder. We should investigate further the question of how common it is for a sentential skeleton to require that one of its complements be a space-builder.

1This paper is a revised and condensed version of Brugman to appear. Part of the writing of this paper was done under support from NEH fellowship no. FA-30259-91, while I was in residence at CRL. Thanks go to M. Catherine O'Connor, Michele Emanatian, Robert Kluender, and Mark Turner for useful comments. All errors are my own.

2Throughout this paper I will surround attested examples with double quotation marks.

3This paper is not concerned with syntactic structure except as it defines the subset of HAVE-constructions I consider here. There may in fact be a "small clause" constituent sister to HAVE, within the matrix VP, which dominates NP\d\u and XP, rendering the structure in (i):

(i) [S [ NP1 [ [VP HAVE [? [ NP2 ] [ XP ] ] ] ]

However, nothing in this paper would be affected if such an additional constituent were revealed. Cf. Stowell 1981 and Rothstein 1984 for a conclusion different from mine, and Williams 1980 and Alsina 1992 for arguments in favor of the constituent structure adopted here.

4I am not using this term in the sense in which it is used in Situation Theory.

5The symbol `#' here means `unacceptable on the intended reading'.

6"Author" and "text" are purely convenient notations for the two participant types and should not be misconstrued as, e.g., Semantic Role assignments. By "author" I will mean the person with predictive powers, the author (director, etc.) of a fictional narrative work, the maker of a theory, and the person in authority to create states-of-affairs by declaration (as a baseball umpire or the person who gives out assignments at the departmental pot luck). The "text" will correspondingly (and respectively) be the future world, the narrative work, the theory, or the declared state of affairs.

7The author-for-works metonymy (as mentioned in Fauconnier 1985, inter alia) is much more common; one way to analyze examples like (25) is that they utilize this metonymic principle.


Alsina, Alex. 1992. The monoclausality of causatives: evidence from Romance. Paper delivered at the Annual Meeting of the Linguistic Society of America.

Brugman, C. 1988. The syntax and semantics of HAVE and its complements. Unpublished doctoral dissertation, University of California at Berkeley.

Brugman, C. to appear. Mental spaces and constructional meaning. In G. Fauconnier and E. Sweetser, eds., Mental Spaces and Grammar.

Fauconnier, G. 1985. Mental Spaces: Aspects of Meaning Construction in Natural Language. Cambridge, Mass.: MIT Press (Bradford Books).

Fauconnier, G. 1990. Invisible meaning. In K. Hall et al., eds., Proceedings of the Sixteenth Annual Meeting of the Berkeley Linguistics Society, 390-404. Berkeley Linguistics Society.

Fauconnier, G. 1991. Roles and Values: The Case of French Copula Constructions. In C. Georgopoulos & R. Ishihara, eds., Interdisciplinary Approaches to Language: Essays in Honor of S.-Y. Kuroda. Dordrecht: Kluwer Academic Publishers.

Fauconnier, G. to appear. Analogical Counterfactuals. In G. Fauconnier and E. Sweetser, eds., Mental Spaces and Grammar.

Fillmore, Ch. 1976. Frame Semantics and the Nature of Language. Annals of the New York Academy of Sciences, 20-32.

Fillmore, Ch., and P. Kay. in preparation. On Grammatical Constructions. ms., Dept. of Linguistics, University of California at Berkeley.

Fillmore, Ch., P. Kay, and M. C. O'Connor. 1988. Regularity and Idiomaticity in Grammatical Constructions: The Case of Let Alone. Language 64:501-538.

Langacker, R. 1987. Foundations of Cognitive Grammar, vol. 1. Stanford, California: Stanford University Press.

Lakoff, G. 1987. Women, Fire, and Dangerous Things. Chicago: University of Chicago Press.

Norvig, P. 1988. Interpretation under Ambiguity. In S. Axmaker, et al., eds., Proceedings of the Fifteenth Annual Meeting of the Berkeley Linguistics Society. Berkeley Linguistics Society.

Rothstein, Susan. 1984. The syntactic forms of predication. Doctoral Dissertation, MIT. Distributed by Indiana University Linguistics Club.

Stowell, T. 1981. Origins of Phrase Structure. Unpublished Doctoral dissertation, MIT.

Williams, E. 1980. Predication. Linguistic Inquiry 11:203-238.

[CRL Newsletter Home Page] [CRL Home Page]
Center for Research in Language
CRL Newsletter
Article 7-1