You are on page 1of 16

6.

The Ditransitive construction: issues in syntactic and semantic representation28


6.1. Introduction
In chapters 3-5, I have presented an analysis of argument linking in terms of the representation of the situation denoted by the clause and the part of that situation profiled by the verb. Crucial elements of the analysis rest on the verbal profile: Linking Rule 1 is formulated in terms of the verbal profile, and the other linking rules are defined relative to the linking of Subject and Object roles in Linking Rule 1 (see 3.1). The precise profile of the verb, however, is a question of theoretical dispute in recent work on argument linking. The question arises because of the development of construction grammar, in which constructions such as argument linking constructions are represented as symbolic units in their own right, just as lexical items such as verbs are (Goldberg 1995). In other words, constructions have semantic representations as well. The question that arises is, what is are the independent contributions of the verb and the construction to the semantics of verb plus construction? The representation of the verbal profile hinges on this question. In this chapter, I will examine this question more closely, using the example of the English Ditransitive construction. I will argue that the specification of the individual contributions of verb and argument linking construction are a priori undecidable, and that the primitive unit of grammatical representation is the verb-specific construction, that is, an argument linking construction instantiated with a verb. In the verb-specific construction, we may specify the verbal profile in accordance with the linking rules given in 3.1. Consider the following two examples: (1) (2) Tess baked a cake. Tess baked Bill a cake.

Example 1 describes an act of creating a food item through a certain process. This is usually analyzed as the meaning of the verb bake in that sentence. Example 2 describes the same act of creation but also describes an intention to subsequently transfer possession of the cake to Bill. The question is, where does this additional component of meaning come from? How should example 2 be analyzed? The traditional approachtraditional in generative grammar and its offshoots, that isis the lexical rule approach. The additional component of meaning in 2 is analyzed as part of the meaning of the verb bake and other verbs of creation like it. Thus, bake in 2 does not mean the same thing as bake in 1, and so there must be a rule deriving bake in 2 from bake in 1, which allows it to be used with the appropriate meaning in the construction in 2. The lexical rule approach has been developed in greatest detail by Levin and Rappaport Hovav (see for example Levin and Rappaport Hovav 1995). A more recent approach, associated with the cognitive linguistic tradition, is the constructional approach. The additional component of meaning in 2 is analyzed not as a part of the verb meaning, but as a part of the meaning of the Ditransitive construction. The Ditransitive construction is an independently represented syntactic unit in its own right, and possesses its own semantics as well as its own syntax. In particular, the Ditransitive construction implies the presence of transfer of possession in some way or another in (almost) all of its occurrences, as in 3-5: (3) (4) (5) She gave him a book. She bequeathed him her art collection. She threw him the ball.

28An earlier version of 6.1-6.6 will appear as Croft (to appear).

134

8/2000

6 The Ditransitive construction: syntax and semantics

The constructional approach is central to cognitive linguistic models of syntax (Langacker 1987, 1991; Lakoff 1987; Fillmore & Kay 1993), but has been most vigorously defended by Goldberg (1995). Which approach gives us the right analysis of 2? In this paper, I will argue that it is neither, or perhaps both. The lexical rule-construction debate revolves around a false dichotomy. Although constructions exist, as cognitive linguists have argued, the relation between constructions (and their meanings) and their component words (and their meanings) is not as neat or as clear as lexical rule-construction debate implies. Language once again successfully resists the attempt of linguists to make it neat and clean.

6.2. Arguments against lexical rules


Goldberg presents a number of arguments against a lexical rule approach to argument structure alternations (Goldberg 1995:9-23). Goldbergs arguments are briefly summarized in the remainder of this section. (i) Lexical rules require the positing of implausible verb meanings. For example, the sentences in 6-7 would require a derived meaning of sneeze as X CAUSES Y to MOVE Z by sneezing and bake as X INTENDS to CAUSE Y to HAVE Z (Goldberg 1995:9). (6) (7) He sneezed the napkin off the table. Tess baked Bill a cake. [= 2]

However, we find no language in which there is a verb that means X INTENDS to CAUSE Y to HAVE Z by BAKING distinct from the verb meaning X BAKES Y (ibid., 10). (ii) Lexical rules are required because of the assumption in many syntactic theories that the verb determines how many and which kinds of complements will co-occur with it (Goldberg 1995:11; cf. Jackendoff 1997:557). This argument for lexical rules is circular. (iii) It is more parsimonious to attribute to constructions the semantic properties that would have to be assigned to different verb senses in the lexical rule approach. (iv) Semantic compositionality is affected by the constructions in which verbs (and other words) occur. Hence a version of semantic compositionality can be preserved by assuming that verb meanings combine with constructional meanings in a regular fashion. (v) A psycholinguistic experiment yielded different reaction times for verbs with different senses vs verbs with the same sense (in the constructional analysis) but occurring in different constructions (Carlson & Tanenhaus 1988). That is, there were significant differences in reaction times for the pair of sentences in 8a-b (distinct senses) than for the pair in 9a-b (same sense, different constructions): (8) (9) a. Bill set the alarm clock onto the shelf. b. Bill set the alarm clock for six. a. Bill loaded the truck onto the ship. b. Bill loaded the truck with bricks.

(vi) Syntactic bootstrapping in language acquisition is really acquisition of constructional meaning. The syntactic bootstrapping hypothesis is that children acquire verb meanings by exploiting their occurrence in syntactic constructions (Landau & Gleitman 1985). Goldberg argues that the reason that children can acquire verb meanings from the syntactic contexts in which they occur is that those syntactic contexts are constructions with meanings in their own right; the constructional meanings are what children are using in identifying the word meanings.

135

8/2000

6 The Ditransitive construction: syntax and semantics

Arguments (i)-(vi) are arguments in favor of a constructional analysis over a lexical rule analysis. Arguments (vii)-(x), given below, are based on arguments in favor of a lexical rule analysis over a syntactic transformational analysis. Goldberg argues in each case that the reasons for choosing a lexical rule analysis over a transformational analysis do not rule out a constructional analysis, and indeed there is sometimes evidence favoring a constructional analysis. (vii) What appear to be lexical restrictions on verbs argument structure are better described as constraints conferred by the construction itself. For example, in 10a-b the same verb can be either acceptable or unacceptable in the Ditransitive construction (Goldberg 1995:22, taken from Langacker 1991): (10) a. Joe cleared Sam a place on the floor. b. *Joe cleared Sam the floor.

Thus, one cannot say that there is a lexical rule (dis)allowing clear. Instead, the Ditransitive construction requires that the first object receive the second object; this is acceptable for a place on the floor, but not for the floor itself. (viii) Lexical rules allow for the restriction of idiosyncratic information to the lexicon. However, there is evidence that idiosyncratic information is associated with phrases and other syntactic units; in fact, this is the primary argument for the existence of constructions as independent units of grammatical knowledge (see e.g. Fillmore et al. 1988). Hence the idiosyncrasy of argument structure alternations is not an argument against the constructional analysis. (ix) In other languages, there are morphological affixes on verbs that differentiate verbs that occur in the two argument structure constructions, although in English the same verb occurs in both contexts. For example, compare the Russian verbs to the English translations in the following sentences (Croft 1991:201, from Comrie 1985:314): (11) Ivan se -jet p enic -u v pol -e John sow.IMPF -3SG.PRES wheat -ACC in field -LOC John sows wheat in the field. Ivan zase -iva -jet pol p enic -ej John DERIV - sow -IMPF -3SG.PRES field(ACC ) wheat -INST John sows the field with wheat.

(12)

The Russian sentences have an argument structure alternation equivalent to that found in their English translations, but sow in 12 is morphologically derived. Goldberg argues that a morphological affix plus the verb class it combines with is itself a construction (e.g. for the causative; cf. Goldberg 1995:61), so such examples can be handled in a constructional account as well. (x) Verb forms that are derived from argument structure alternations, such as the passive, can be the input to further word derivational processes, such as that which yields the adjectival passive. However, there is much evidence (see references in Goldberg 1995:23) that the line between lexical and syntactic rules is not at all clearcut, and so this does not count as an argument against a constructional analysis. On the whole, these arguments (along with arguments in the other construction grammar sources cited above) convincingly demonstrate that constructions are a necessary part of the model of syntactic representation for a speakers knowledge of their language, and that constructions have meanings. What is at issue here is the relationship between constructional meaning and verb meaning in argument structure constructions. In order to explore this issue, I turn to Goldbergs analysis of a specific construction, the Ditransitive construction.

136

8/2000

6 The Ditransitive construction: syntax and semantics

6.3. The grammatical representation of the Ditransitive construction


The English Ditransitive construction is usually described as [SBJ VERB OBJ1 O BJ 2], and is illustrated in its most common form in 13: (13) Julie gave Simon a book.

Goldberg argues that the Ditransitive construction contributes the meaning that the agentacts to cause transfer of an object to a recipient (Goldberg 1995:32). This meaning is not necessarily part of the meaning of the verb. As we saw with example 2 above, the basic meaning of bake does not involve transfer of possession; this semantic component is present only when bake occurs in the Ditransitive construction. Goldberg notes however that in the case of bake and other verbs of creation, it is in fact intended, not actual, transfer of possession that is contributed by the Ditransitive construction. Whereas it is an entailment of 13 that Simon comes into possession of the book, it is possible in 2 that Bill does not receive the cake. For example, Tess could bake Bill the cake, but then get angry at him and throw the cake away. Hence the contribution of the Ditransitive construction is slightly different for give and for bake. Goldberg resolves this problem by proposing that constructions can be polysemous. Polysemy is the phenomenon by which a single form has two meanings that are semantically related. Typically, one meaning is a historical extension of the other meaning. Synchronic polysemy, however, also implies that the speaker internally represents the two meanings as semantically related (Croft 1998b). For example, the Ditransitive construction has two related constructional meanings: actual transfer of possession and intended transfer of possession. The former is found with give, the latter with bake. There is good motivation for constructional polysemy. Other constructions have been found to be polysemous (e.g. Michaelis & Lambrecht 1996, Jackendoff 1997). Since constructions are essentially schematic, multi-word lexical items, one would expect to find similar properties among constructions as among lexical items, and lexical items are frequently, perhaps universally, polysemous. There seems to be no reason to posit two independent Ditransitive constructions, one with a meaning of actual transfer of possession, the other with a meaning of intended transfer of possession. Instead, it seems quite plausible to posit a single Ditransitive construction with two related constructional meanings. In fact, Goldberg lays out an analysis in which there are not two but six related constructional meanings for the Ditransitive, five of which are extensions of the first, central sense (Goldberg 1995:38, Figure 2.2). These meanings are associated with verb classes as given in 14: (14) A. Central Sense: agent successfully causes recipient to receive patient 1. Verbs that inherently signify acts of giving: give, pass, hand, serve, feed, 2. Verbs of instantaneous causation of ballistic motion:29 throw, toss, slap, kick, poke, fling, shoot, 3. Verbs of continuous causation in a deictically specified direction: bring, take, B. Conditions of satisfaction imply that agent causes recipient to receive patient 1. Verbs of giving with associated satisfaction conditions: guarantee, promise, owe, C. Agent causes recipient not to receive patient. 1. Verbs of refusal: refuse, deny

29In 4.5, I argue that ballistic motion verbs actually belong in Goldbergs subsense F, intended transfer of

possession. II will leave Goldbergs analysis as it is here, since it does not affect the arguments in this chapter.

137

8/2000

6 The Ditransitive construction: syntax and semantics

D. Agent acts to cause recipient to receive patient at some future point in time 1. Verbs of future transfer: leave, bequeath, allocate, reserve, grant, E. Agent enables recipient to receive patient 1. Verbs of permission: permit, allow F. Agent intends to cause recipient to receive patient 1. Verbs involved in scenes of creation: bake, make, build, cook, sew, knit, 2. Verbs of obtaining: get, grab, win, earn, A polysemy representation of the Ditransitive construction would have a single representation of the constructions morphosyntactic structure, linked to the range of meanings found in 14. Such a representation is given in 15 (I use the format of Langacker 1987, where [[form]/[meaning]] represents a conventional symbolic unit; and I abbreviate transfer of possession as XPoss): (15) [ [SBJ VERB OBJ 1 OBJ 2] / [actual XPoss] [conditional XPoss] [negative XPoss] [future XPoss] [enabling XPoss] [intended XPoss] ]

However, if we examine the Ditransitive construction more closely, it appears that the variation in the Ditransitive constructions meaning is not true polysemy. Each verbal semantic class is associated with only one sense of the Ditransitive construction. It is not an accident that the verbs found with Ditransitive sense E, agent enables recipient to receive patient are verbs of permission, or that the verbs found with sense C, agent causes recipient not to receive patient are verbs of refusal. That is, the modulation of the possessive relation specified by each constructional senseactual, enabling, and negative transfer of possessionmatches a semantic component of these verbs. Further evidence for this is the existence of at least one other sense of the Ditransitive construction, required by one class of verbs: (16) (17) a. That vase cost him $300. b. That painting set him back $500. G. Acquisition of goods causes recipient to no longer have possession of money. 1. Verbs of costing: cost, set back,

There may be other senses of the Ditransitive construction lurking in the lexicon of English. Moreover, even the verb classes that do not inherently specify a modulation of the transfer of possession relationship occur with only one sense of the polysemous Ditransitive construction: for example, verbs of creation are found only with the intended transfer of possession sense. Goldberg points out that this is evidence against a monosemous analysis of the Ditransitive construction, that is, an analysis such that the Ditransitive construction simply means transfer of possession without specifying the modulation of this event (Goldberg 1995:36-37). If the Ditransitive construction were truly polysemous, one might expect that the verb bring, for example, would be found with Ditransitive sense F, resulting in a meaning like X brings Z with the intention of causing Y to receive Z, or kick could also occur with Ditransitive sense C, resulting in a meaning like X kicks Z causing Y not to receive Z. But 138
8/2000

6 The Ditransitive construction: syntax and semantics

we do not. Instead, it seems that the different senses of the Ditransitive construction are very closely tied to the verb classes that each sense occurs with. In other words, a proper representation of the construction schema for each sense of the Ditransitive construction must specify which verb classes occur with it. That is, the proper representation of an English speakers knowledge of the Ditransitive construction is not the polysemous representation given in 15. Instead, there is a distinct syntactic schema for each constructional sense specifying the verb classes found with each subtype, and its corresponding meaning: (18) A. [ [SBJ GIVING .VERB OBJ1 OBJ2] / [actual XPoss] ] [ [SBJ BALL.MOT.VERB OBJ2 OBJ2] / [actual XPoss via ballistic motion] ] [ [SBJ DEIC.CAUS.VERB OBJ 1 OBJ 2] / [actual XPoss via deictic caused motion] ] B. [ [SBJ COND.GIVING .VERB OBJ1 OBJ 2] / [conditional XPoss] ] C. [ [SBJ REFUSE.VERB OBJ 1 OBJ 2] / [negative XPoss] ] D. [ [SBJ FUT .GIVING.VERB OBJ 1 OBJ 2] / [future XPoss] ] E. [ [SBJ PERMIT .VERB OBJ 1 OBJ2] / [enabling XPoss] ] F. [ [SBJ CREATE.VERB OBJ1 OBJ 2] / [intended XPoss after creation] ] [ [SBJ OBTAIN .VERB OBJ1 OBJ 2] / [intended XPoss after obtaining] ] G. [ [SBJ COST.VERB OBJ 1 OBJ2] / [depriving XPoss via paying] ] Of course, there is a family resemblance in both the form and the meaning of the Ditransitive constructions in 18: all involve a modulated transfer of possession, and all have two object phrases and a verb. But the verb classes used for each construction are mutually exclusive, and the modulation of the transfer of possession differs for each construction type. The semantic type of the second object NP also varies significantly: the latter two constructions of sense A require a mobile physical object NP, while senses C-E do not; sense G requires a price NP, etc. The subject and first object NPs are human NPs, of course. Just as semantically based verb classes are involved in constructional representations, so are semantically based noun classes. Adopting properly specified argument types will make even clearer how distinct each of the constructions in 18 actually are. It is not clear from Goldberg 1995 whether she adopts the constructional representations in 17 or in 18. On the one hand, Goldberg argues forcefully for constructional polysemy (Goldberg 1995:31-39), and accepts the need to account for which verb classes go with which sense of the Ditransitive construction (Goldberg, pers. comm.); these imply the representations in 18. On the other hand, the syntactic representations in Goldberg 1995 indicate only the general verb category V for the construction (Goldberg 1995:3, 53, 54, 55, 77, 91, 93, 96, 142, 145). In many of these representations, the meaning is expressed simply as CAUSE-RECEIVEPRED. Goldberg represents both sense A (actual transfer) and sense F (intended transfer) using only the general verb category V and its semantic counterpart PRED (Goldberg 1995:77), and in fact indicates that the representation [[V]/[PRED]] in sense F is inherited from the semantic representation of sense A. These descriptions imply the representations in 17. However, for the reasons given above, the representations in 18 are a more precise and accurate representation of an English speakers knowledge than those in 17. I will call the representations in 18 verb-class-specific constructions. In fact, other evidence that Goldberg presents indicates that even more specific representations than those in 18 are necessary. She points out that not every permission and refusal verb occurs in the Ditransitive construction (Goldberg 1995:130):

139

8/2000

6 The Ditransitive construction: syntax and semantics

(19) (20)

Sally permitted/allowed/*let/*enabled Bob a kiss. Sallly refused/denied/*prevented/*disallowed/*forbade him a kiss.

A representation of these two classes would in fact have to specify each verb that occurs in the Ditransitive construction: (21) (22) E. [[SBJ permit OBJ1 OBJ 2]/[enabling XPoss by permitting]] [[SBJ allow OBJ 1 OBJ 2]/[enabling XPoss by allowing]] C. [[SBJ refuse OBJ1 OBJ 2]/[negative XPoss by refusing]] [[SBJ deny OBJ 1 OBJ 2]/[negative XPoss by denying]]

I will call the constructions in 21-22 verb-specific constructions. There is also evidence that even the classes for which there appears to be a good case for only a schematic verb-class-specific construction must also include representations of verbspecific constructions. Goldberg points out that there are at least two anomalous examples of verbs used in the Ditransitive construction (Goldberg 1995:132), which would be represented as verb-specific constructions, as in 23b-24b: (23) (24) a. He forgave her her sins. b. [[SBJ forgive OBJ1 OBJ 2]/[forgive someone for something they did]] a. He envied the prince his fortune. b. [[SBJ envy OBJ1 OBJ 2]/[envy someone for something they possess]]

Goldberg points out that historically forgive meant to give or grant, and envy meant to give grudgingly or to refuse to give (ibid.), hence they began as members of verb classes normally found in the Ditransitive construction (namely classes A and A or C respectively). Their meanings changed, yet they are still allowed in the Ditransitive construction (although Goldberg suggests that they have become archaic and are being lost; ibid.). If sense A of the Ditransitive construction were represented only by a verb class schema as in 18, then the occurrence of the original forgive give, grant and envy give grudgingly, refuse to give in the Ditransitive construction could be predicted from the semantics of the basic verb entry and the semantics of the appropriate verb-class-specific construction in 18. That is, there would be no need for verb-specific constructions for these verbs. If this were true, then one would predict that once the verb meanings changed, the verbs could no longer occur in the Ditransitive construction. Instead, the verbs did continue to occur in the Ditransitive construction. This implies that for at least these two verbs, and probably many other verbs, the verb-specific constructions are autonomous (Bybee 1985), that is, independently represented in the mind, even though their syntax and semantics are predictable from the verb-class-specific construction. Their subsequent semantic divergence from the Ditransitive construction meaning is the constructional analog to lexical split, where for example something can be dirty without involving real dirt at all (Bybee 1985:88).

6.4. Constructions, lexical rules and verb-specific constructions


In the preceding section, I argued that the facts about the Ditransitive construction in English requires the postulation of verb-class-specific constructions and verb-specific constructions as part of an English speakers knowledge of his/her language. What is the consequence of this argument for the debate over lexical rules vs. constructions? Verb-specific constructions are simply more specific types of constructions. Constructions are popularly represented as abstract syntactic schemas without specific lexical content (except for obligatory inflections and function words). But they need not be that abstract. Verb-specific constructions are constructions but their schemas have specific lexical content for the verb. Verb-class-specific constructions are slightly more abstract, but still 140
8/2000

6 The Ditransitive construction: syntax and semantics

much less specific than a construction schema such as [SBJ V ERB OBJ 1 OBJ 2]. The existence of verb-specific constructions and verb-class-specific constructions do not preclude the existence of a more schematic construction, if the speaker has made such a generalization in his/her mental representation (see below). The existence and necessity of verb-specific constructions indicates that there is a false dichotomy underlying the lexical rule analysis and the abstract constructional analysis. One can account for lexical idiosyncrasy and constructional generality with verb-specific and verb-class-specific constructions. We now turn to the arguments given by Goldberg about the lexical rule vs. construction controversy, in order to support this point. Some of Goldbergs arguments are in favor of constructions in general, not critiques of a lexical rule analysis of argument structure patterns. Argument (viii) demonstrates the need for idiosyncratic information with respect to constructions, not just simple lexical items, and argument (x) cites the lack of a sharp boundary between lexical and syntactic processes. Although a model relying exclusively on lexical rules will find these facts problematic, the postulation of verb-specific constructions does not, since verb-specific constructions are constructions. Argument (vii) identifies a semantic constraint associated with the Ditransitive construction; but the grammar must still sanction the use of clear in 10a to mean clear with intention to transfer (metaphorical) possession with (at least) a verb-class-specific construction. Argument (vi) makes the case that what is called syntactic bootstrapping, where a child learns verb meanings via their syntax, is in fact constructional bootstrapping: the child learns syntactic constructions with their meanings, and uses the constructions to figure out the meanings of verbs, or to match verbs of a certain meaning with the appropriate syntactic constructions. Goldbergs interpretation of so-called syntactic bootstrapping is correct, I believe. Further evidence from language acquisition supports the initial learning of verbspecific constructions. Tomasello (1992) argues for the Verb Island Hypothesis: that children acquire verbs first in a single argument structure construction, and then learn to apply that verb to other constructions. More generally, I would propose that the acquisition of constructions proceeds inductively, with the child learning specific word-construction combinations and gradually generalizing. However, in the generalization process, it is not necessarily the case that the more specific instances are then dropped as independent grammatical representations, even when they are completely predictable from the more general constructions (see Langacker 1987:374). The historical evolution of forgive and envy also implies that redundant representation of verb-specific and verb-class-specific constructions is necessary. Goldberg argues that parsimony dictates assigning the meaning to the (schematic) construction. Goldberg now withdraws this argument (pers. comm.); however, since it is a widely accepted argument, I will discuss it here. Parsimony is not a criterion for mental representations; only psychological reality is (Croft 1998c). Moreover, Goldberg argued for only one sort of parsimony, storage parsimony. Storage parsimony involves minimizing redundant storage of grammatical information. The result of storage parsimony is computational complexity: language users must compute on line the meaning of a combination of a verb and a construction in which it occurs (this requires some preservation of compositionality, hence argument (iv)). Storage parsimony has long been the sine qua non of generative (and structuralist) analyses of grammar. But one could argue for computing parsimony instead: minimizing the computation involved in language use. Computation cannot be eliminated, of course; but nor can storage of knowledge in the storage parsimony approach. On a priori grounds one cannot argue for storage parsimony or computing parsimony (Croft 1998c). There is likely to be some storage redundancy and some computational complexity, though some cognitive psychologists suggest that there is probably more of the former (see references cited in Barsalou 1992:180). Linguistic evidence can place an upper bound on how abstract or general a mental representation must be; but they allow for less 141
8/2000

6 The Ditransitive construction: syntax and semantics

abstract and potentially redundantly represented storage of linguistic knowledge (Croft 1998c). In fact, as demonstrated above, closer examination of the linguistic facts almost always reveals idiosyncrasies that show that more specific representations are required than is usually thought. Goldberg argues in favor of redundancy in stored information for more schematic constructions. In comparing more schematic constructions to more specific instantiations, she allows for full-entry representations of the information inherited by the more specific construction from the more schematic construction (Goldberg 1995:74). Verb-specific constructions simply extend the full-entry principle down to constructions specific enough to specify particular verbs (or at least narrow semantic verb classes). Thus verb-specific constructions are compatible with Goldbergs general approach to the representation of constructional knowledge. In fact, the storage parsimony representation is not parsimonious in storage. In the storage parsimony representation, one would still have to specify which verb classes occur in which subsense of the Ditransitive construction. This specification is no different than specifying an additional meaning of the verb class members that is compatible with the semantics of the Ditransitive construction. Moreover, the storage parsimony representation would still fail for forgive and envy. Most of Goldbergs remaining arguments turn on the nature of verb entries: she argues against the necessity or desirability of multiple verb entries linked by lexical rule. In her argument (i), she argues that the derived verb meanings are implausible as verb meanings; one would not expect to find a distinct verb in another language with the derived verb meaning. This argument is slightly too strong. One does not find basic verb forms with the derived verb meanings. One does find morphologically derived verb forms with the derived verb meaning, derived by the addition of an applicative affix to the verb. Goldberg argues that the applicative affix should be represented as part of the construction, not as a derived verbal lexical entry with a derived meaning (argument (ix) above; cf. Goldberg 1995:61). This analysis will work for abstract constructions in some cases, but not in others. For example, the Russian verbs that participate in the locative alternation illustrated in examples 11-12 above require different derivational prefixes that must be lexically specified; that is, they must be represented as at least verb-specific constructions (see examples in Schupbach 1978).

6.5. Verb-specific constructions, verb entries, and analyzability


As a matter of fact, verb-specific constructions do not appear to be that different from multiple verb entries, each associated with the argument structure construction it occurs with, and linked together by lexical rules. What is the relationship between verb entries and verbspecific constructions? The chief difference is that a verb entry must have associated with it its specific contribution to the meaning of the constructional whole. Yet that contribution differs from verb to verb for the Ditransitive construction, even if we assume a single-entry model of verbal semantic representation as Goldberg does. The semantics of the combination of verb + Ditransitive construction can be divided into three components. The first is what Rappaport Hovav & Levin (1998:107) call the verbal constant: the core of meaning that differentiates verbs of the same semantic class such as throw, toss, kick etc. The second is the transfer of possession meaning that is associated with the verbs when they occur in the Ditransitive construction. The third component is what I called the modulation in 6.3: whether the transfer of possession is actual, conditional, intended, etc. The verbs that occur in the Ditransitive construction can be put into three groups depending on how many of the above semantic components the basic verbal meaning would include: 142
8/2000

6 The Ditransitive construction: syntax and semantics

Table 1. Componential analysis of Ditransitive verb class basic meanings. Verb class Group I inherent giving (A1) conditional giving (B1) future transfer (D1) Group II refusal (C1) permission (E1) costing (G1) Group III instantaneous ballistic motion (A2) deictic continuous causation (A3) creation (F1) obtaining (F2) Verbal Constant Modulation [actual] [conditional] [future] [negative] [enabling] [depriving] [actual] [actual] [intended] [intended] Transfer of Possession

When any verb from any of these classes occurs in the Ditransitive construction, all three semantic components are present. For Group I verb classes, occurrence in the Ditransitive construction does not add anything over and above the verbal meaning. For Group II verb classes, occurrence in the Ditransitive construction adds only transfer of possession. For Group III verb classes, occurrence in the Ditransitive construction adds both a modulation (actual or intended) as well as transfer of possession. The semantics of Group I verbs is a simple relation of specific meaning subsumed under a more schematic meaning (Goldberg 1995:60). The question then, is: do verbs of Groups II and III include the additional semantic components as part of derived verbal lexical entries, so that the process of semantic composition of verb + construction is simply subsumption? Or are they contributed only by the (verb-class-specific) Ditransitive construction, in which case a more complex semantic composition process is necessary? I would like to suggest that this question is not decidable by purely linguistic evidence. Speakers may or may not induce more schematic constructions, and they may analyze their component units in different ways. Consider the options available to an English speaker for the array of constructions in 18 and 21-24 with respect to both modulation and transfer of possession. An English speaker could abstract a single schematic Ditransitive construction [[SBJ DITR.V ERB OBJ 1 OBJ2]/[XPoss]] and analyze out fully derived verb meanings, along the lines of the Group I verbs. The construction could include transfer of possession meaning, except for forgive and envy, but the modulation would come from the derived verb entries. Or a speaker could abstract a set of Ditransitive constructions like the ones given in 17, each construction specifying transfer of possession and its modulation, combining with underived verbs, which would have to match the modulation of the particular construction for Groups I and II. The speaker need not induce any construction more abstract than the verb-class-specific constructions in 18, or the verb-specific constructions in 21-24. Even in that case, the same options for analyzing out a meaning for the verb are available. The speaker may analyze out the verb form with a single basic meaning; or the speaker may analyze out the verb form as possessing a meaning including the modulation and also the transfer of possession meaning that are specific to the argument structure construction.

143

8/2000

6 The Ditransitive construction: syntax and semantics

One might want to invoke a principle of syntagmatic parsimony avoid overlap in meaning between the verb and the construction. The same arguments against storage and computing parsimony apply to syntagmatic parsimony, of course. Even if we set these aside, however, storage parsimony fails to simplify the analysis. No matter how abstract or how specific is the meaning of the Ditransitive construction induced by the speaker, there will be one or more group of verbs whose meaning will overlap with that of the construction even in a single verb entry model. (I find the syntagmatically most parsimonious solution implausible, namely that the Ditransitive construction has no meaning with Group I verbs, only transfer of possession meaning with Group II verbs, and both transfer of possession and modulation meaning with Group III verbs.) In other words, the data does not unambiguously point to a single optimal solution for analyzing the form-meaning mapping for the family of Ditransitive constructions. Finally, we must ask ourselves, what is the verb meaning taken as basic by advocates of lexical rules or taken as the single verb meaning argued for by Goldberg? The basic or single verb meaning is actually the meaning of the verb when it occurs in some other argument structure construction such as the transitive construction. Verb meanings cannot be defined in pure isolation; they are only definable with respect to the construction(s) they occur in. The verb-specific and verb-class-specific constructions such as those in 18 and 21-25 are much closer to what a speaker actually hears and uses. Linguists cannot second-guess the sort of generalizations speakers make beyond these constructions. (Nor should we expect all speakers to form the same generalizations.) Only psycholinguistic experimentation might be able to establish the generalizations formed by individual speakers (Croft 1998c). Unfortunately, the psycholinguistic experiment conducted by Carlson & Tanenhaus 1988, discussed by Goldberg (argument (v) above), does not decide the issue. One need not assume that verb senses are either the same (leading to a shorter reaction time) or different (leading to a longer reaction time). It could be that there are degrees of difference between senses leading to differences in reaction times (cf. Goldberg 1995:230, fn. 13). For example, the difference between set [something somewhere] and set [a clock to a time] is greater than the difference between load [something somewhere] and load [something with something else]. The latter two share a semantic frame that the former two do not (see 4.2). Thus, the experimental results are compatible with autonomous derived verb senses or construction senses.

6.6. The mapping of form to meaning in language


In the preceding section, I suggested that the meaning of the verb-specific construction as a whole cannot be unambiguously factored out into the meaning of the more schematic construction and/or the meaning of the verb taked as an isolated lexical item. In fact, the relationship between (basic) verbal meaning and constructional meaning varies from class to class (see Table 1). For example, it follows from neither the semantics of verbs of creation nor the semantics of the Ditransitive construction (in all of its uses) that Tess baked Bill a cake involves only intended and not actual transfer of possession. In other cases there is semantic overlap. For example, it follows from both the semantics of verbs of inherent transfer of possession and the semantics of the Ditransitive construction that She gave Bill a book involves (actual) transfer of possession. What conclusions can we draw from these facts? Speakers are not exposed to verbs in isolation, nor are they exposed to schematic argument structure constructions without verbs in them. Actually occurring utterances are closest to verb-specific constructions. The process of language learning and (re)using is a process of analyzing the structure of utterances into its component parts. The parts will possess recurring meanings in most of their various contexts: bake will involve creating a food item, and the Ditransitive construction will involve transfer of possession. The recurrent aspects of meanings will sanction the use of bake and of the Ditransitive construction in new situations that the speaker wants to describe. But those recurrent meaning aspects do not give 144
8/2000

6 The Ditransitive construction: syntax and semantics

the full description of the meaning of the verb or the construction in each and every context of use (see Croft to appear, chapter 4). They do not even give the full description of the meaning in other conventionalized contexts, such as the Ditransitive construction with refuse, bequeath or forgive. This is in fact the great asset of language, its flexibility in communicationalthough it seems the great bane of natural language processing and of formal models of syntax and semantics. Linguists spill much ink arguing where the components of the meaning of a verb-specific construction lie. Do they belong in the verb in isolation, analyzed out from the construction? Or only in a more schematic construction? Or neatly divided between the two? But another perspective is more interesting. If linguists have so much difficulty analyzing the semantic contribution of each element of a fully specified construction, dont speakers have the same problem? In fact, there is evidence that they do. Consider first some simple morphological examples. The plural of child is children. It contains two etymological markers of plurality, -r and -en. Speakers apparently stopped identifying -r as a marker of plurality and added -en (this common process is called reinforcement in language change). That is, speakers reanalyzed the mapping between form and meaning, in part because the original formmeaning mapping is somewhat opaque. Similar cases are found in attested innovations such as feets and PFI initiative [PFI = Private Finance Initiative] (see Croft 2000:137-38). Moving closer to argument structure, we may consider the loss of governed oblique cases in Germanic languages (Croft 2000:121-24). That is, a verb that occurred in an argument structure construction with an oblique case argument, such as the dative or genitive, is found with an accusative (object) case argument instead. The speakers no longer identified the semantic contribution of the oblique case, which overlapped with that of the verb anyway. So they reanalyzed the form-function mapping so that the oblique case made no semantic contribution, and replaced the oblique case with the object case form. In other words, the form-function mapping of the component parts of constructions is not necessarily fully specified. Another way of putting it is that multiple analyses of the formfunction mapping may be available to the speaker (cf. Hankamer 1977). Part of the reason for this is that verb-specific and verb-class-specific constructions exist as part of a speakers grammatical knowledge, so the form-meaning mapping does not have to be fully specified. Verb-specific and verb-class-specific constructions are semantically partially analyzable but need not be completely so. Most of the time this is not a problem in communication, since the verb-specific construction specifies the meaning of the whole as well as its syntax. But the analysis of the form-function mapping can vary across speakers, and can result in variation and change across time. As linguists, we should accommodate this variation and indeterminacy in our analysis of argument structure constructions.

6.7. From the Ditransitive construction to the Directed Motion construction


The issues of syntactic representation discussed for the Ditransitive construction above also apply to other constructions. For example, similar arguments can be made for the Directed Motion (Path) construction discussed in 4.1; the example and semantic representation are repeated here as example 25 and Figure 1: (25) The bottle floated into the cave.

145

8/2000

6 The Ditransitive construction: syntax and semantics

Figure 1. English directed manner of motion

cave

exist travel with floating

bottle

The analysis represented in Figure 1 is that the verb meaning (the profile in black) is different from the meaning of the simple manner of motion verb, which is usually construed an undirected activity (see example 26 and Figure 2, also repeated from 4.1): (26) The bottle floated in the water.

Figure 2. Manner of motion

bottle t

move with floating

An alternative constructional analysis attempts to achieve storage parsimony and syntagmatic parsimony. In the alternative analysis, the verb meaning remains the same in both the Intranstive and Directed Motion constructions, and it is the construction that contributes the directional meaning to the clause in example 4. In the three-dimensional model, this analysis would be represented as an automatic construal of the undirected activity manner of motion verb as an accomplishment directed motion verb. However, this construal is not automatic, that is, it is not a general pragmatic or cognitive principle, but a conventional construal available specifically in English. It is not available in Spanish and other path-incorporating languages (see 4.1). In other words, even in the alternative analysis, it must be conventionally specified which verbs may occur in the Directed Motion construction. Hence there is no gain for storage parsimony in the alternative analysis. Likewise, there is no gain for syntagmatic parsimony either. There are at least two directed motion verbs that occur in the Directed Motion construction: the deictic motion verbs come and go. An analysis of the Directed Motion construction as including the path would not be syntagmatically parsimonious for come and go. On the other hand, it seems implausible to establish two Directed Motion constructions, one for come and go and the other for non-directed motion verbs such as float and screech.

146

8/2000

6 The Ditransitive construction: syntax and semantics

Yet the reason that we are inclined to analyze float, screech etc. as undirected activities but come and go as directed activities is not because of inherent properties of those situation types independent of the argument linking constructions they occur in, but because of their interpretation in other argument linking constructions, specifically the Intransitive construction, as in 27: (27) a. The bottle is floating. b. The brakes are screeching. c. The boys are coming.

Just as we saw for the verbs in the Ditransitive construction, all predicates occur only in specific argument linking constructions. The same is true for lexical aspect (chapter 2): all predicates occur only in specific tense-aspect constructions. In both cases, the semantics of the verb is a function of its occurrence in a particular construction. The primitive grammatical structure for argument-linking theory is the verb-specific construction. Analysis of the verbspecific construction into its semantic parts or categories is theoretically derivative and also developmentally secondary, as analysis is a process of schematization, that is, generalization over parts. This is the Radical Frame Semantic approach, referred to in 3.3.

6.8. The semantic representation of the Ditransitive construction


The Ditransitive construction poses a difficulty for the theory of argument linking advocated here. There is a tension between strictly respecting Linking Rule 1 (that the Object is the endpoint of the verbal profile), and preserving the constraint on nonbranching causal chains. Respecting Linking Rule 1 would require that the verbal profile end at both Objects. But the fact that the two Object referents play different roles in the situation leads to a branching causal chain, illustrated in Figure 3 for example 28a: (28) a. Sue gave Jack a book. b. gift donor recipient

Figure 3. Branching causal chain representation of the Ditransitive construction gain possession Jack change ownership

book

Sue t

cause transfer possession

147

8/2000

6 The Ditransitive construction: syntax and semantics

A nonbranching representation, however, would have to represent only one Object as the true endpoint of the causal chain, as in Figure 4: Figure 4. Nonbranching causal chain representation of the Ditransitive construction gain possession Jack change ownership

book

Sue t

cause transfer possession

In Croft (1991), I noted that other transfer verbs do not occur in the Ditransitive construction, and instead the item transferred is encoded as an antecedent oblique, as Figure 4 implies (Croft 1991:207; see also 5.2): (29) a. The dean presented the award to the valedictorian. b. The dean presented the valedictorian with the reward.

In the nonbranching causal chain analysis, the Object role in English would be polysemous between a true (endpoint) Objectthe recipientand an antecedent Obliquethe transferred item. Following Dryer (1986), I will call the true Object the Primary Object and the Oblique the Secondary Object. Although this is not an attractive analysis, it does have the advantage of preserving the principles underlying Linking Rule 1 and the nonbranching causal chain hypothesis. Such a solution appears necessary for languages such as Kinyarwanda and Tzotzil, whose applicative constructions code the derived antecedent referent without case marking (see Croft 1991:240-41). However, uniform treatment of the secondary object referent as antecedent gives rise to at least one anomalous case in English, again in the commercial transaction frame. There are argument structure constructions which allow profiling of just the equivalence relation between money and goods.: (30) (31) a. This lawnmower costs/is worth $25. b. goods money a. $25 will buy a lawnmower. b. money goods

Both patterns are allowed by the commercial transaction frame, because there is no directed transmission of force between money and goods.

148

8/2000

6 The Ditransitive construction: syntax and semantics

It turns out that both patterns also allow the expression of one of the transactors. Be worth allows the seller to be coded only with subsequent to , and hence is not problematic: (32) a. This lawnmower is worth $25 to me. [i.e. I would sell it for $25] b. Sbj goods Obj money to seller money buyer goods Buy encodes the transactor as the Primary Object, and if we assume that the Secondary Object is antecedent to the Primary Object, the analysis is not problematic: (33) a. $25 will buy you a lawnmower. [i.e. if you had $25, you could buy it] b. Sbj money SecObj goods PrimObj buyer money buyer goods Cost encodes the transactor as Primary Object as well. But if we assume that the Secondary Object is antecedent to the Primary Object, then a clash occurs with the commerical transaction frame. (34) a. This lawnmower will cost you $25. [i.e. you would have to pay $25 for it] b. Sbj goods SecObj money PrimObj buyer money buyer goods The boldface arrow goes in the opposite direction from what is expected, since the buyer transfers the money to the seller. It is possible that the money phrase is in fact a measure phrase, indicating the degree to which the buyer would be reduced in assets. In that case, the analysis of 35a would follow that of other measure expressions in resultative constructions (see 7.1). However, it is not obvious why the analysis of the monetary value phrase should be different in 35a than in all of the other commercial transaction phrases in which it occurs. This problem, and the broader problem of the semantic representation of Ditransitive constructions, will have to be addressed in a future version of this book. seller seller seller

149

8/2000

You might also like