Symbolic flexibility and argument structure variation *.

Zhang, Ren

Abstract

Much work on the syntax-semantics interface has focused on verb frame alternations, the well-documented phenomenon that a verb is typically found in a variety of syntactic contexts. On the other hand, cognitive linguists have often focused on the flexibility of categories other than verbs in relation to such issues as metaphor, metonymy and polysemy. Drawing on the usage-based orientation of cognitive grammar, this article argues for a unified treatment of variation in the use of nouns and verbs under the general concept of symbolic flexibility. In this approach, typical cases of nominal metonymy (e.g., read Langacker) and unconventional verb use (e.g., Frank sneezed the napkin off the table) are both creative facets of a usage event that are rich in semantic value, the salient elements of which consist of both the conventional meaning and the contextual understanding of the noun or the verb. In addition, nominal metonymy and unconventional verb use are aspects of usage events that are simultaneously categorized by their conventional counterparts as well as their linguistic context in the form of constructional schemas. The network of categorizing relationships that results accounts for the semantic character of such usages as well as variation in argument expression associated with verbs. This article provides arguments against certain versions of construction grammar (e.g., Goldberg 1995) that treats verb meaning as invariant across syntactic patterns.

1. Introduction

A well-known challenge to linguistic theory is the tension between the relatively stable linguistic system and the dynamic and open-ended nature of language use. As an illustration of flexible language use, consider the following:

(1) a. Frank sneezed.

b. Frank sneezed the napkin off the table.

c. Pat kicked the wall.

d. Pat kicked Bob black and blue.

e. Pat kicked Bob the ball.

(2) a. Langacker visited China recently.

b. I asked my students to read Langacker.

(3) apple tree; oak tree; pine tree; palm tree; family tree

The data above might seem all too familiar, but they are deliberately brought together in an attempt to counteract the traditional tendency to treat them as having distinct status. Examples in (1) and the like show different argument expression patterns associated with a verb. These patterns have been termed diathesis or verb frame alternations, and have been a focus in both mainstream theories and constructional approaches (see e.g., Levin 1993; Van Hout 1998; Rappaport Hovav and Levin 1998; Mateu 2000; Jackendoff 1997; Goldberg 1995, 2005, inter alia), as they raise problems for the traditional verb-centered projectionist view of syntax (Chomsky 1981). Van Hout (1998) refers to the phenomenon as lexical-syntactic flexibility. By contrast, data in (2) and (3) exemplify variable uses of nouns and have been a central topic for many cognitive linguists. However, they typically don't catch the attention of syntacticians, for they apparently have no bearing on syntax. Lexical-syntactic flexibility should be expected under the framework of cognitive grammar (CG) (Langacker 1987, 1999; Taylor 2002) that treats lexicon and syntax as shading into one another and assumes a usage-based approach to linguistic knowledge. In this article, I draw attention to certain parallels between (1) and (2) from a cognitive grammar perspective and propose to treat them both in terms of SYMBOLIC FLEXIBILITY (symbolic in the sense of cognitive grammar). I outline a cognitive grammar approach to examples like (2) and argue that basically the same approach can be extended to variation in argument expression as exemplified in (1). Certain complexities associated with (1) can be attributed to the nature of the symbolic unit that is being flexibly used, rather than any distinct mechanism.

Regarding the data in (1), I want to raise some specific issues. First, how does the conventional value of the verb relate to the conceptualization associated with the whole sentence? In the case of (1b), for example, is Frank construed as a 'sneezer' or a causer or both? There seems to be a strong intuition that both construals are highly salient and may be allowed at the same time, which is problematic for lexical approaches (cf., Pinker 1989; Levin and Rappaport Hovav 1988; Rappaport Hovav and Levin 1998) as well as "pure" constructional approaches such as Goldberg (1995), and Goldberg and Jackendoff (2004). In addition, examples like (1b) are unconventional in that the conventional value of the verb does not easily support its occurrence in the construction. But there are restrictions on the options open to a speaker's creativity, as shown in the following (cf., Boas 2003):

(4) ??Frank yawned/exhaled/snorted/wheezed the napkin off the table.

How does cognitive grammar handle the potential restrictions? Thirdly, perhaps the most difficult issue involved in alternate verb usage patterns is the change in the number or type of arguments conventionally associated with the verb. For instance, an intransitive verb could take a "fake" object argument, as with sneeze in (1b). Additional examples below illustrate the same issue (cf., Jackendoff 1990; Goldberg 1995):

(5) They laughed the poor guy out of the room.

(6) Pauline smiled her thanks.

(7) We ate our way across the U.S.

(8) Sam rinsed/cleaned the soap out of her eyes.

(9) In the last Star Trek episode, there was a woman who could think people into a different galaxy.

Finally, is there a principled way to account for the fact that these examples sound more or less creative and may be even not acceptable to some speakers (cf., Boas 2003)?

In this article, I suggest that cognitive grammar offers a solid basis for exploring these issues. In particular, I propose that novel verb uses in verb frame alternations can be understood as the categorizing relationship of extension between the conventional verb and its contextual use. The relation established shows the asymmetrical character of the reference point relation (Langacker 1993, 1999). In (1b) for example, the event expressed by sneeze serves as the prototype and a reference point in relation to the force affecting the napkin. In Langacker's (2003) terms, the force can be apprehended as the sneezing event. I show how the approach extends to other cases of novel usage patterns and its consequences for argument expression.

The analysis makes explicit the role of frame and encyclopedic knowledge in the semantics of argument structure variation. This emphasis finds support in several recent works. For instance, both Verspoor (1997) and Boas (2003) argue for the importance of encyclopedic knowledge in licensing novel instances of argument structure constructions. While Verspoor (1997) invokes a theory of discourse coherence (Lascarides and Asher 1993), Boas (2003) in particular goes to considerable detail in elaborating the role of frame semantic information in licensing various types of resultative constructions. However, my proposal differs from these works in its adherence to the stringent "minimalist" requirements of cognitive grammar (see Section 2) and in its elucidations of cognitive operations underlying novel usage patterns.

In the next section, I present the relevant assumptions from cognitive grammar, introducing basic concepts and stressing its usage-based approach toward grammatical knowledge. In Section 3, I briefly outline a cognitive grammar perspective on the nature of nominal metonymy as exemplified by (2). In addition, I demonstrate how the approach can be extended to novel verb use in the caused-motion construction as in (1b). The consequences of the approach are discussed in Section 4, which focuses on the more general issues regarding the semantics of novel usage events, verb meaning and variation in argument expression. The strengths of the approach are demonstrated with detailed references to the influential construction-based approach presented in Goldberg (1995).

2. The cognitive grammar perspective on grammatical knowledge

Several central ideas in the cognitive grammar's approach to grammatical knowledge bear on the issue of lexical-syntactic flexibility or symbolic variation (see Langacker 1987, 1999 for a full account). First, as a foundational tenet, grammatical knowledge is believed to consist of a structured inventory of symbolic units that are abstracted from and serve to sanction actual usage events. Symbolic units, which can be characterized at different levels of abstraction or schematicity, are entrenched cognitive routines that automatically and conventionally associate phonological structures with semantic structures. For instance, many utterances express in a uniform formal pattern the idea of causing an entity to move along a specified path:

(10) a. Bill set the alarm clock onto the shelf.

b. Bill loaded the truck onto the ship.

c. Jack threw the ball out of the window.

Through repeated use, the pattern gets entrenched as a constructional schema, which is an internally complex schematic unit in the mentally represented grammar. Langacker (2003: 56) proposes the following representation of the schema (cf., Goldberg 1995):

(11) [ILLUSTRATION OMITTED]

Following Langacker's notation, square boxes indicate units and rounded boxes (cf., [17] below) indicate usage events. Dashed lines indicate correspondences. In (11), I added the syntactic category glosses as shorthand for the phonological pole of each constituent. The prepositional phrase profiles the path of motion with respect to a Landmark, which is not indicated in (11). The Landmark can either be the source or the goal of the path (cf., [10b], [10c]). The constructional schema (11) abstracted from usage instances serves to sanction further usage events as either conventional or unconventional. Conversely, aspects of a usage event (such as an utterance expressing caused motion) activate a set of interrelated sanctioning units for their categorization, in a fashion as conceived in spreading/interactive activation models generally associated with connectionism. The activation depends on factors such as accessibility, contextual priming and degree of overlap with the target (Langacker 1987, 1999, 2003). An underlying idea is that grammar is not conceived as a constructive or generative device, and therefore cognitive grammar does not aim at absolute predictability (Langacker 1987: 47-55). Turning to (1b) again, the question becomes how this novel pattern accords with and hence is categorized by the schema (11).

Second, the semantic pole of a symbolic unit, termed predication, is characterized relative to rich knowledge configurations called cognitive domains, which support the conceptual profile designated by the unit, as indicated in bold type in (11). In addition, the profile of a minimal unit determines its grammatical category. Profiling is one of the major construal operations, grouped under focal adjustments in Langacker (1987), which could be performed on a situation. This means that a concept is a conceptualization of a particular situation, which is one of the central tenets of cognitive linguistics. On this view, a profile and its supporting domain constitute a unified whole which defines the semantic value of an expression. Note that a domain typically serves as the base of multiple concept profiles. For instance, FINGER supports not only the concept of KNUCKLE, but also the idea of FINGERTIP. Conversely, a single concept is often profiled in several domains simultaneously, as is the case with the concept HUMAN BEING, which presupposes such domains as physical objects, living things and mind (Croft 2002; Croft and Cruse 2004). In other words, this set of domains, termed domain matrix, provides the context for the full understanding of the semantic unit. Langacker (1988: 56) cites the case of KNIFE as participating in a potentially open-ended domain matrix, including space, cutting, silverware, measurement, games, entertainment, history, and countless others, some of which could be more highly ranked and more likely to be activated (hence called primary domains) than others, based on the criteria of centrality such as intrinsicness and conventionality. Thus, in the encyclopedic semantics of cognitive grammar, "lexical items provide partially conventionalized path of access into the systems of knowledge they invoke" (Langacker 1998: 660). The encyclopedic stance of cognitive grammar (Langacker 1987, 1997; Croft and Cruse 2004) is supported by several other approaches (e.g., Fillmore 1985; Jackendoff 1983, 1997, 2002; Pustejovsky 1995).

Third, the syntagmatic interaction or the combinatory relationship among meaningful symbolic units, called valence (Langacker 1987, 1988), crucially presupposes correspondences (indicated as dashed lines in [11]) between elements or substructures in component units, which can be interpreted as instructions for superimposing or unifying the corresponding substructures. Given the profile/domain distinction in the rich encyclopedic semantics of cognitive grammar, one could imagine various ways to establish potential correspondence links and therefore potential combinatorial relations. Information involved in valence relations can be very marginal and context-dependent, as in the case of the expression alligator shoes (cited in Taylor 1995), hence supporting the encyclopedic view of meaning (cf., Langacker 1987: 156-157; Downing 1977). As will be shown later, encyclopedic and even marginal information are also productively involved in novel instances of argument structure constructions.

Fourth, symbolic units represented in the mental grammar make up a gradation of schematicity and complexity, showing no clear-cut boundary between the lexicon and the syntax. This is a position shared by all construction-based models (for discussion, see Goldberg 1995; Croft and Cruse 2004; Langacker 1987, 1999, 2003). On this conception, syntactic structures are not "projected" from the lexicon. Depending on the degree of entrenchment, flexible uses of verbs could affect both the lexicon and the syntax, resulting in augmented lexical and constructional units (cf., Goldberg 1995; Croft 2003; Langacker 2003). In a generative conception, by contrast, one either enriches the lexicon, as advocated by Rappaport Hovav and Levin (1998), for example, or the syntax, in order to generate novel usage patterns.

A final important property of cognitive grammar is the central role assigned to the cognitive process of categorization in both grammatical organization and the utilization of symbolic resources (Langacker 1987, 1999; Taylor 2002). The symbolic units in the grammar are related to each other through the categorizing relationships of elaboration/instantiation and extension. Furthermore, each pole of a unit may itself be a complex category, with interrelated members accounting for semantic and phonological variants. Thus the mentally represented knowledge of a grammar is conceived of as massive intricate networks, in the same way general knowledge is organized (Hudson 2001). In addition, the network is usage-based in that it represents abstraction from and entrenchment of usage patterns and is susceptible to pressures of language use (Langacker 1987, 1999, 2003). Categorizing relationships are based on the general cognitive ability of making comparisons according to the degree of resemblance between a standard and a target, which defines the schema-instance relation and the prototype-extension relation. As mentioned above, a symbolic unit is typically associated with rich knowledge structures. These knowledge structures can serve as a basis for variable categorization. The flexible use of a unit depends in part on the way the usage can be variously categorized in reference to established units. Langacker (2003) discusses the limiting case where an uncategorized lexeme lacks inherent profile, which makes it easy to perceive its similarities with a wide range of semantic units. This accounts for its relatively free distribution as either a noun or a verb.

3. Comparison asymmetry and symbolic flexibility

As noted in the preceding section, symbolic flexibility in cognitive grammar partly amounts to variable categorization of symbolic units, which, depending on the level of entrenchment, could result in a huge network of linked units in the mentally represented grammar. Well-known research in this area concerns lexical polysemy (cf., [3]) (see Langacker 1987; Taylor 1995, 2002 and references therein), which is often described in terms of schematic networks of categorizing relationships. For instance, the various conceptions associated with tree in (3) can be represented in a schematic network (see Langacker 1987: 378-381 for discussion):

(12) [ILLUSTRATION OMITTED]

In the above, dashed arrows represent extensions from prototypes. Schemas are abstractions over instances, as indicated by arrows pointing toward instances.

3.1. Nominal metonymy

In cognitive grammar, metaphor and metonymy can be treated as modes of semantic extension, even though they have not been a central focus in this framework as they are in the more general cognitive linguistic literature (cf., Langacker 1993, 1999). Studies on metaphor and metonymy within the cognitive-linguistic tradition have typically focused on the conceptual mechanisms involved in the interaction of two semantic structures, which are discussed largely independently of the symbolic system. Looking more strictly in cognitive grammar terms, metaphor and metonymy are usages that must be assessed by the units in the symbolic system in terms of categorizing relationships. In the following, I develop this perspective in greater detail and propose a unified analysis of certain cases of nominal and verbal metonymy as important cases of symbolic flexibility, though I will not try to integrate the huge literature on metaphor and metonymy. For present purposes, I also leave out discussion of more subtle cases of nominal metonymy involving situational context (cf., Croft 2002).

Let us turn to a typical example of nominal metonymy, such as read Langacker in (2b), where Langacker is understood to designate scholarly output from Langacker. This is a case of semantic extension, which assigns an additional value to the expression. The distinct values of the expression can be seen in the following contrast, which is based on a well-known test for polysemy (for discussion, see Taylor 1995, 2002):

(13) a. I asked my students to read Langacker, which was important for their research.

b. *I asked my students to read Langacker, who visited China recently.

The extended value of the word Langacker is not something constructed by the linguistic system per se, but results from the speaker's utilization of symbolic, psychological and contextual resources (cf., Langacker 1999). Thus the particular use of the word constitutes one facet of the usage event exemplified in (2b), which is formally represented as follows:

(14) ((WORKS BY LANGACKER)/[langacker])

In the above, I have followed Langacker in using square brackets to indicate unit status and parentheses to indicate non-unit structures associated with the usage event. What (14) shows then is the speaker's creative association of a phonological unit (indicated here by orthography) with an intended conception, understood in its contextual detail as evoking the domains of linguistics and scholarly research. In the case of (2b), the question becomes how the contextual usage (14) is assessed and thereby sanctioned by the linguistic system. The issue breaks into two components. Firstly, (14) must be assessed by the conventional unit (15) where Langacker designates a unique individual:

(15) [[LANGACKER]/[langacker]]

The semantic pole indicated in (15) is shorthand for a wealth of encyclopedic information, including the primary domain of living thing/mind (for "humans") as well as secondary domains such as linguistics and scholarly research (cf., Croft 2002). Secondly, (14) must simultaneously be assessed by the symbolic unit (16), which is activated by the use of the verb read in (2b):

(16) [[READ (tr) (lm)]/[read]]

Specifically, (14) must be categorized by a component of (16) for combinatory purposes, either the Trajector (tr) or the Landmark (lm) schematically specified by the relational predication READ.

By virtue of a conceptual correspondence between the intended conception in (14) and the Landmark (lm) in (16), whose schematic content is specified by READ as TEXT, the semantic pole of (14) can be readily categorized as an instance of the TEXT and hence elaborates the latter. Regarding the assessment of (14) with respect to the symbolic unit (15), one could also detect points of resemblance on the basis of the correspondence between LANGACKER in (15) and the "author" component in (14). In addition, the domains supporting the intended conception in (14), namely, linguistics and scholarly research, happen to correspond to the secondary domains in (15). These correspondences serve as the basis for the extension between (15) and (14). Their shared phonological material only encourages the extension (see Langacker 1987: 387). The two simultaneous aspects of categorization, which link the usage (14) to the linguistic system, are sketched in (17):

(17) [ILLUSTRATION OMITTED]

The double categorization of the usage indicated in (17) is crucial in the analysis. On this account, metonymy is a facet of a usage event that motivates and is categorized by a component in the constructional meaning. Put differently, the metonymic interpretation, as with word meaning in general, is closely intertwined with the construction serving as its linguistic context (Taylor 2003; cf., Gleitman 1990). On the other hand, the usage conflicts in its specifications with the conventional unit the speaker decides to use. The conflict is resolved by an extension from the unit to the usage. Note that the "extension" here should be understood not in "derivational" terms, but as a kind of categorizing relationship in which a usage instance is partially compatible with the specifications of a prototype, the relationship which Langacker (1987) calls partial schematicity.

The extension indicated in (17) can be separately represented as (18), using the bracket notation:

(18) ([[LANGACKER]/[langacker]] [right arrow] ((WORKS BY LANGACKER)/[langacker]))

Note that the categorizing relationship as a whole is included in parentheses, showing that the extension is not part of entrenched routines. However, repeated occurrences of the same kind of extension, as exemplified by read Proust/Shakespeare/Chomsky, etc., allow the emergence of a schema by which the concept of a famous person associated with the general domain of creative activity is related by extension to the concept of the person's creative output. Note that there is some directionality in the extension: it's hard to imagine the reverse extension in which [WORKS BY LANGACKER] serves as the prototype for the extension to [LANGACKER]. This poses a problem to the nature of comparison based on similarity (i.e., extension), which is in principle symmetrical (cf., Langacker 1987: 379-380). The similarity between A and B motivates the bidirectional relationship A [left and right arrow] B, as with "lime [left and right arrow] lemon" (Bowdle and Medin 2001). However, cognitive psychologists have long noted certain comparison asymmetries, as manifested in the preference of North Korea is similar to China over China is similar to North Korea (cited by Bowdle and Medin 2001; cf., Tversky 1977). Langacker (1987: 379) suggests in vague terms perceivable prototypicality as among the factors underlying the directionality of extension. In Langacker (1993, 1999, 2001), however, nominal metonymy is argued to be a manifestation of the cognitive ability of accessing one entity (the target) by first accessing a related and more salient entity as its reference point (for the relevance of reference point to other aspects of grammar, see Van Hock 1997; Taylor 1996; Kumashiro and Langacker 2003). It stands to reason that the directionality of extension follows from the prominence asymmetry between the reference point and its target. The prototype in the categorizing relation of extension, being both prominent and related to the target, qualifies as the reference point for its target (for similar arguments, see Bowdle and Medin 2001). Thus the relation of extension from a perceivable prototype to an instance coincides with that of reference point and target. On the other hand, bidirectional extension results from the lack of prominence asymmetry between the two entities.

Croft (2002) has proposed an account of metonymy in cognitive grammar terms, certain elements of which are however at odds with the spirit of cognitive grammar. Assuming encyclopedic semantics, Croft locates metonymy in the combination of noun and predicate. On this analysis, the semantic pole of the unit (15), i.e., [LANGACKER], is encyclopedic both in its domain matrix (as mentioned earlier) and in its profile. This means that works by Langacker is part of the profile [LANGACKER]. Croft (2002: 187) suggests that the semantics of the predicate highlights the relevant aspect of the "encyclopedic profile", i.e., Langacker's works. In addition, Croft (2002: 192) proposes domain highlighting that is induced by the predicate. Thus the secondary domains of [LANGACKER], i.e., linguistics and research, are highlighted when it combines with the predicate. The essence of this analysis is summarized in (18a):

(18) a. [[READ (tr) (lm)]/[read]] x [[LANGACKER]/[langacker]] = "highlighting" [??] ([READ (tr) (WORKS BY LANGACKER)]/[read langacker])

I share with Croft the emphasis on the encyclopedic view of meaning. In my analysis, the secondary domains of [LANGACKER] are part of the basis for the extension in (17). In addition, Croft's insight on the relevance of combination to metonymy corresponds to the simultaneity of the two categorizing relationships in (17), which links the lexical extension to the partially schematic unit in (16). The major difference between (17) and (18) lies in the "derivational" or processual flavor of the latter. In (18), the metonymic usage appears to be a constructive output of grammatical units via certain additional mechanisms. However, as mentioned in Section 2, instead of being a constructive device, cognitive grammar conceives of grammar as an inventory of units serving to categorize usage events (Langacker 1987, 1999). This kind of "declarative" conception of grammar is also found in construction grammar and word grammar (see Hudson 2001). In my analysis, instead of invoking the "procedural" mechanism of highlighting to derive the semantics of metonymy, all that is needed are categorizing relationships between conventional units and metonymic usage. Furthermore, Croft's idea of "encyclopedic profile" is a contradiction in terms from the perspective of cognitive grammar. Meaning is encyclopedic precisely in that encyclopedic domain matrix support concept profiles, not in profiles themselves, even though shift may occur between profile and domain information. The proper analysis in cognitive grammar is to treat "Langacker's works" as information in the secondary domains supporting [LANGACKER], not part of the profile itself. I suspect the notion of "encyclopedic profile" is an artifact of the processual conception of metonymy.

Different predictions regarding the semantic value of metonymy follow from (17) and (18). According to the processual analysis in (18), the metonymic expression read Langacker should highlight the specific type concept WORKS BY LANGACKER in the domains of linguistics and research. On this analysis, read Langacker is basically treated as having the same value as read works by Langacker. (1) But this does not capture the intuition that "Langacker" as a unique individual is also saliently implicated in the meaning of the expression. As Panther and Thornburg (2002: 282) have argued, both the source and the target of metonymy are mentally present. Turning to the analysis in (17), the simultaneous presence of source and target easily falls out from the two simultaneously present categorizing relationships. The metonymic usage (14), i.e., ((WORKS BY LANGACKER)/[langacker]), which is categorized as an instance of the lm of READ, must at the same time activate the conventional unit (15), i.e., [[LANGACKER]/[langacker]]. This implies that the usage (14) simultaneously activates the different domains associated with those units as well. In the case where (14) gets entrenched a unit so that Langacker becomes polysemic, the activation of (14) by a particular usage will most likely activate (15) as well. This is expected on the spreading activation model of the network that easily allows multiple parallel interpretations of units (see Lamb 2004 for a similar treatment). The status of (15) as a reference point for (14), as argued earlier, only adds to the likelihood of simultaneous activation. This analysis also applies to the rich value associated with metaphorical extension. For instance, star in the sense of [CELEBRITY] may also suggest the mental presence of the concept of celestial body (Langacker 1987: 386).

3.2. Quirky verbs in caused-motion constructions

The general outlook emerging from the discussion so far is that nominal metonymy is a manifestation of symbolic flexibility in language use, which can be categorized in terms of a network of categorizing relationships. Semantic properties of the usage pattern are captured by the character of the network. Let us now consider the cases of symbolic flexibility associated with verbs, where the conventional value of a verb does not support its occurrence in the construction. I first work through the famous case of the caused-motion construction featuring the verb sneeze as in (1b) (cf., Goldberg 1995; Goldberg and Jackendoff 2004; Pustejovsky 1995; Fauconnier and Turner 1996; Verspoor 1997; van der Leek 2000, Boas 2003, and references therein), and then discuss the consequences of the analysis in the next section, with respect to a much wider range of examples.

From the perspective of cognitive grammar, (1b) itself is a usage event whose acceptability hinges on the way it is categorized in all its facets by speakers with reference to their internally represented linguistic system. The categorization will most likely vary among speakers. If every speaker categorized its semantic elements and their relations as instantiating the semantic pole of the constructional schema in (11), nobody would doubt its acceptability nor get any sense of creativity associated with this usage.

The crucial point lies in the way the verb sneeze is used in (1b). Let us suppose that for speakers who accept (1b), sneeze is used to construe a specific kind of causal process, one that has sneezing as an external cause or source of force, in the primary domain of causal event model (Langacker 1991). This construal is compatible with the intended conceptualization of caused motion associated with (1b) as a whole. In other words, what I suggest here is that sneeze is not used in its conventional value. Rather the usage value of the word is closely connected to its linguistic context, as noted earlier (Taylor 2003; see Section 3.2). This aspect of the usage event, if sanctioned by the linguistic system, can be categorized by the schema of the caused-motion verb in (11) as an elaboration. The construal in the use of the word sneeze can be diagramed as follows:

(19) [ILLUSTRATION OMITTED]

The profiled causal process is indicated by the bolded double arrow, which corresponds to the backgrounded causal potential associated with the sneezing event. Another element in this construal is the correspondence of the trajector of the causal processes with that of the sneezing process, as indicated in (19). It is important to note that the semantic structure associated with the contextual use of the word sneeze is included in a rounded box in (19), indicating that it is not a conventionalized value of the word. Obviously, the conventional value of the symbolic unit sneeze differs considerably from (19) in both its profile and domain, as given in (20) (Langacker 2003: 56):

(20) [ILLUSTRATION OMITTED]

As can be seen in (20), the word sneeze conventionally profiles the emission of sound, air, etc., as indicated by the bolded arcs, with a potential force transmission in the base, as indicated by the double arrow. The primary domain that is implicated is physiological reaction. In addition, given the potential force transmission in the base, the secondary domain of causal event model may be involved as well, if only tangentially. What is "quirky" about the verb in a usage event like (1b) lies with the conceptual tension between (19) and (20). In other words, the use of the word sneeze in (1b) typically activates the symbolic unit (20), which however sharply diverges from the intended conceptualization in (19).

The problem could be resolved if (20) and (19) could be connected by a categorizing relationship of extension. I suggest that such an extension is indeed plausible, based on the perceivable correspondences between the profile in (20) and the base of (19), the secondary domain of (20) and the primary domain of (19). One could abstract from the different profiling in (20) and (19) to arrive at a conception that is schematic for both, which serves as the basis for extension. Once the verb sneeze is associated with the semantic structure in (19) by extension, it can be categorized by the caused-motion verb schema in (11) as an elaborative instance. These relations are diagramed in (21), which incorporates both the lexical categorization of the use of sneeze and its categorization by the caused-motion verb schema in (11):

(21) [ILLUSTRATION OMITTED]

The parallels between (21) and (17) should be apparent, a prominent feature of both being the double categorization of a facet of usage event. What makes (21) more complicated is related to the nature of the usage that is involved, namely, the process predication associated with verbs. Thus general modes of lexical categorization unify the flexible use of both nouns and verbs, which should be expected from the usage-based orientation of cognitive grammar.

There is another cognitive motivation for the extension illustrated in (21), which again parallels (17). Given that the conventional value of sneeze in (20) corresponds to the external cause or source of force for caused motion, as indicated in (21), one cannot fail to note a certain salience asymmetry between the external cause and the caused motion: the physical process of sneezing is intrinsically more salient than the resultant but virtually invisible force transmission that impacts an entity. Owing to this cognitive asymmetry, a conceptualizer could evoke a caused-motion event by first focusing on its salient external cause. Thus the prototype value of sneeze qualifies as the reference point for its extension, the target in the categorizing relationship, as in the case of nominal metonymy discussed earlier. As in the nominal case, one can hardly imagine the reverse direction in categorization, namely, having (19) as the prototype that categorizes (20).

4. Explaining the character of quirky verbs

The analysis of the quirky verb sneeze in the caused-motion construction presented above is based on the double categorization of the verb as a facet of the usage event, which resolves the conceptual tension between the contextual use of the verb and the symbolic unit that is most likely to be activated. The extension involved in the categorization, which is indicated in the lower half of (21), is exactly the source of creativity and unconventionality hinted in Section 1. As could be expected, the extension, even though cognitively plausible, could to lead to different reactions from different language users. For some speakers, the extension is not possible, due to the highly divergent profile configurations between the standard and the target. For speakers who accept (1b), the extension helps produce a creative instance of the constructional schema (11). Note that the categorizing relationships in (21) as a whole are put within a rounded box, showing that these linked relationships have not entered into the grammar as entrenched cognitive routines. This implies a low probability of occurrence for (1b) (cf., Boas 2003).

This analysis has a number of additional implications that need to be spelled out, which will provide a perspective on the specific questions raised in Section 1 and the character of quirky verbs in usage events. The proposed analysis should be favored over alternatives to the extent that aspects of an unconventional usage event, which proved problematic to other accounts, fall into place under the present analysis. Spelling out the implications will necessarily bring in discussion of more illustrative examples, which hopefully will help consolidate the analysis.

4.1. The semantics of unconventional usage events

Obviously, the sneeze-type of caused-motion in (1b) is just one instance of many unconventional usage events that can be characterized in terms of a conceptual gap between the conventional value of the verb and its constructional context. (Note: nominal metonymy evinces the same sort of gap, though involving nouns.) Examples (5-9) are all cases that can be looked at this way. Another prominent class is intransitive motion construction (Talmy 1985; Croft 1991; Goldberg 1995; Goldberg and Jackendoff 2004):

(22) a. The boat floated into the cave.

b. The ball rolled down the hill.

c. The train screeched into the station.

d. The knee replacement candidate groaned up the stairs.

The verbs in the above examples designate manner of motion or sound emission. The use of these verbs in motion construction is unconventional to the extent that the entrenched value of these verbs as manifested in intransitive construction (cf., The man groaned.) has no intrinsic connection with motion. So the question is, where is the motion component encoded?

To return to (1b) first, the analysis in the preceding section goes a long way toward capturing the intuition regarding the multiple facets of its semantic value mentioned in Section 1, in particular, both "sneezing" and "caused motion" are salient elements in the overall construal associated with the utterance. In addition, the subject can be interpreted as both a 'sneezer' and a 'causer'. These different facets stem from the discrepancy between the conventional value of the verb and the intended conceptualization associated with the utterance. Interpreting the analysis in (21) in terms of interaction between linguistic system and usage events, the utterance (1b) activates the constructional schema (11) for its categorization, but one facet of (1b), i.e., the contextual use of the verb sneeze indicated in (19), activates both the caused-motion verb schema and the conventional unit (20) as its prototype and reference point, again paralleling the case of nominal metonymy (cf., [17]). The analysis claims that the unit in (20) is not directly categorized by the caused-motion verb schema (cf., Langacker 2003), but it still contributes to the semantic value due to the simultaneous activation by (19). Note however that it is difficult (if not impossible) to incorporate (20) into (11), i.e., the composite value of (1b) expressing caused motion, due to highly divergent profile configurations. This should be expected on the network conception of semantic value according to which the conceptual import of a usage event is not propositional but a set of interrelated semantic units (cf., Langacker 1987).

While both (20) and the caused-motion verb schema are argued to be activated by the contextual use of the verb sneeze indicated in (19), for speakers who accept (1b), the caused-motion verb schema obviously takes priority over (20), as it is directly reflected in the grammatical structure of (1b) and serves to give grammatical import to the extension in (21). The double interpretation of the subject as both a 'sneezer' and a 'causer' mentioned earlier is also a result of the network of relationships in (21), with some elements linked by correspondences. By virtue of the priority of the caused-motion verb schema, the 'causer' status takes precedence over the 'sneezer' status. In other words, the constructional meaning contains the role of causer-trajector as part of the caused-motion profile, which however corresponds to the sneezer-trajector in the base of (19).

A central claim of this analysis is that the contextual use of the verb sneeze in (1b) is associated with a conceptualization that diverges from its conventional value. In other words, the meaning of the verb changes according to its constructional context. This should be expected, since a symbolic unit in cognitive grammar is regarded as a complex category, whose form and meaning typically vary in language use. As an indication of the variation, an adverb modifying sneeze must be construable as modifying caused motion as a whole, as seen in the following contrast:

(23) a. Frank suddenly sneezed.

b. (?) Frank suddenly sneezed the napkin off the table.

In addition, certain otherwise expected modifiers are ruled out, once the verb is associated with an extension. This is illustrated on the verb laugh:

(24) a. They laughed loudly.

b. *They laughed the poor guy out of the room loudly.

c. *They loudly laughed the poor guy out of the room.

Note that the change in verb meaning is induced by the pressures of language use, and follows general patterns of categorization. The categorizing relationship motivating lexical changes is a pattern of usage that will not necessarily become part of the grammar unless through entrenchment. Thus the nature of lexical changes is treated differently from accounts invoking lexical rules. What Goldberg (1995) sees as the problem of "implausible verb senses" only arises if grammar is kept entirely separate from usage.

Other instances of unconventional usage events listed in (5)-(9) and (22) can be treated in the same fashion. Consider (22a) that expresses the motion of the boat in a sailing manner. Under the encyclopedic view of meaning, there are overlapping structures between FLOAT and MOVE. Floating is also a salient manner accompanying the motion of a boat. Therefore, the contextual use of the verb float can be taken to mean "move in a floating manner", which is categorized by the conventional value of the word as an extension. As in the sneeze case, this use of the verb activates both the conventional unit (float as a manner verb) and a motion verb schema. Slobin (2000: 132) observes that in this kind of usage speakers "seem to conceive of manner and directed motion as a single conceptual event, making it difficult to have a mental image of one without the other". This kind of conceptual effect is nicely captured by the simultaneous activation of different units. While Slobin's remarks are made in the context of a discussion of motion events in the so-called satellite-framed language (cf., Slobin 2003), the preceding discussion demonstrates the same kind of double salience effect in nominal metonymy, unconventional caused-motion patterns, and arguably other cases in (5)-(9) and (22). Consider for example the apparent inseparability of 'smiling' and 'expressing gratitude' in (6), which has nothing to do with motion. What is at issue then may be a much more general phenomenon related to symbolic flexibility.

4.2. Against invariant verb meaning

Looking again at (21), since the conventional unit (20) is always activated by the contextual use of the verb sneeze, there is an intuitive appeal to treat it as the actual value associated with the use of the verb. On this treatment, there is no semantic extension. Indeed, many accounts have treated verb meaning as invariant. Langacker (2003) argues against this approach from the usage-based perspective and emphasizes the inseparability of lexemes and constructions (cf., Gleitman 1990; Taylor 2003), a point that has come up earlier in my analysis of metonymy and novel use of verbs. Langacker suggests that the use of sneeze in the construction induces its construal as a caused-motion predicate. From the present perspective, this construal is equivalent to the posited usage in (19) that describes the contextual use of the verb. However, on Langacker's analysis, the conventional unit (20) is directly categorized by the caused-motion verb schema as an extension. In abbreviated form, his analysis can be represented as ([caused-motion verb] [right arrow] [[SNEEZE]/[sneeze]]). This analysis risks being misunderstood as saying that the conventional unit (20) directly participates in the construction, and hence implies no change in meaning, though this does not appear to be Langacker's position. It is also unclear how such a categorizing relationship is established in the first place, given that the two structures involved are very different in their profile/domain specifications. By contrast, the analysis in (21) retains the usage-based spirit but makes explicit the missing link in the categorizing relationship between the conventional unit and the caused-motion verb schema.

Several constructional approaches have treated verb meaning as invariant when a verb combines with a construction (Jackendoff 1990; Goldberg 1995; Goldberg and Jackendoff 2004). In the view of Goldberg (1995), systematic differences in meaning between syntactic patterns featuring the same verb are directly attributed to semantics of the relevant syntactic pattern (i.e., construction). While Goldberg does recognize lexical and morphological polysemy, in practice she adopts a minimalist lexical semantics in treating the grammatical combination of lexemes and constructions. Goldberg and Jackendoff (2004: 534) assert that "the verb does not change its meaning so as to license these extra arguments". On this view, the sneeze type of example in (1b) would be treated differently as comprising a verbal subevent and a constructional subevent. Using the notation of Goldberg and Jackendoff (2005), the semantics of (1b) can be represented as follows:

(25) Syntax: Frank sneezed the napkin off the table.

Semantics: FRANK CAUSE [NAPKIN GO OFF TABLE] (constructional subevent)

MEANS: FRANK SNEEZE (verbal subevent)

On this analysis, the lexeme sneeze just means SNEEZE, either in (1a) or (1b), and it is incorporated into the meaning of the utterance as a "means" component, yielding the meaning "Frank caused the napkin to move off the table by sneezing". Similarly, Jackendoff (1990: 213-217) explicates the conceptual structure of the way-construction (e.g., Bill belched his way out of the restaurant.) by demoting the meaning of the verb to a subordinate conceptual clause, as in Bill went out of the restaurant belching/ by belching (see Goldberg 1995 on means/manner readings). Note that the "means" interpretation of the verb captures part of the intuition regarding the meaning of such usages. In (21), the same fact is represented as the conceptual correspondence between the conventional unit (20) and the base of the usage (19). Nevertheless, there are several arguments weighing against the analysis in (25) (see also Croft 2003; Langacker 2003).

First, it is far from settled that the conventional value (understood as encyclopedic) of a verb remains constant across syntactic patterns. Referring to the belch example mentioned above, Goldberg and Jackendoff (2004) argue that the meaning of the verb is the same as in Bill belched loudly. They take the invariant verb meaning to be in accord with speakers' native intuition. However, there are other aspects of intuition that are left out of this treatment. For one thing, the adverb loudly, which is otherwise commensurate with the conventional value of belch, cannot comfortably appear in the way construction (?? Bill loudly belched his way out of the restaurant) if the verb is interpreted as the means. (2) This fact can be easily explained if the verb is associated with a different profile/domain configuration in the way-construction (see also [23], [24]). Furthermore, there are other semantic differences associated with where the verb is located. (3) Van der Leek (2000) notes the semantic differences in the following:

(26) a. Joe kicked the dog into the bathroom.

b. Joe made the dog move into the bathroom by kicking it.

As van der Leek points out, (26a) differs from (26b) in that kicking happened there with the direction of the bathroom in mind. In fact, I believe Goldberg has recognized this contrast as part of a series of constraints on direct causation (cf., Fodor 1970). In particular, she observes that "the path of motion must be completely determined by the causal force" (Goldberg 1995: 173). However, if kick is understood only as a means in (26a) as in (26b), what motivates the additional constraint on the means in (26a)? Again the determined path constraint follows straightforwardly from the semantic extension of the verb, i.e., kick being categorized as a caused-motion verb.

Second, on the assumption of invariant verb meaning, the semantics of an unconventional usage event is treated as a "fiat" proposition in the form of a conceptual paraphrase (Jackendoff 1990), with verb meaning treated as a "means" component or other circumstantial elements, as shown in (25). This approach leaves no room for richer semantic value discussed earlier, such as the salience of both "sneeze" and "caused-motion" in (1b) that parallels double salience in nominal metonymy. The "flat" treatment of the meaning of (1b) has another consequence. According to Goldberg (1995) (see also Goldberg and Jackendoff 2004), verb meaning and associated participant roles are to be "fused" with construction meaning and associated argument roles respectively. Goldberg is not explicit as to the actual fusion process. While Goldberg recognizes the importance of general world knowledge in verbal semantics and argument structure, she does not make explicit use of that knowledge in the fusion process. Given the value of (1b) as a proposition with a "means" component, it is not clear in what way the 'sneezer' role in the "means" component is compatible with the causer role in the constructional event, how sneeze can be interpreted as a means and how it can unify with CAUSE-MOVE. In a similar vein, Mateu (2000) raises criticisms regarding the nature of conflation in Goldberg's analysis. For instance, regarding the way construction (e.g., John pushed his way into the bus), Mateu points out that Goldberg does not address how the relation between push and the postulated construction meaning CREATE-MOVE is established. (4) To make explicit these conflation processes, one has to describe in detail how categorizing relationships are established.

Third, the assumption of invariant verb meaning causes a problem for the symbolic thesis of cognitive grammar, which is also a basic tenet of construction grammar. Under the analysis in (25), a central element (the verb) in the grammatical structure does not symbolize the central element in the conceptualization. If this can be maintained, it does seem to constitute an argument for the strong autonomy thesis or the modular conception of grammar and therefore undermines the basis of sign-based theories. A modular grammar could allow a set of correspondence rules that deal with apparent idiosyncratic mappings between syntax and semantics (e.g., Jackendoff 1990, 2002), and these correspondence rules appear to approximate the notion of construction in cognitive linguistics. However, given the view of mapping assumed in such theories, these rules are at best weakly symbolic. By contrast, the symbolic thesis remains intact if the verb is recategorized in the usage event to fit into the larger constructional context.

Fourth, the view of fixed verb meaning leads to the fragmentation of a particular construction or constructional schema (in the terminology of cognitive grammar) that may not be adequately motivated. On this view, the occurrence of different semantic classes of verbs in a syntactic pattern correlates with different types of constructions, since different verb classes apparently combine with the constructional subevent in different ways, including the relations of means, result and precondition, among others (Goldberg 1995, 1997; Goldberg and Jackendoff 2004). Goldberg and Jackendoff do not see any prospect of unifying the various types of verbal subevent. Thus the sneeze type of utterance in (1b) cannot be licensed or sanctioned by the constructional schema in (11), since on this analysis sneeze is not a caused-motion verb. Instead, another construction is needed for this purpose whose semantic pole includes a "means" component. This is represented approximately as follows:

(27) [[X/NP][CAUSE-MOVE/][Y/NP][PATH/PP][MEANS/V]]

Example (27) differs from (11) in its stipulation of a null phonological pole for the caused-motion verb, and in the additional "means" component symbolized by the verb. Apart from the arbitrary nature of (27) in its symbolic structure as discussed above, it also raises the question regarding the relation between (11) and (27), given that they share the same syntactic form. (5) In terms of conceptual content, lexemes should be able to activate a much greater store of encyclopedic knowledge than grammatical patterns, since the latter necessarily entails selective activation of possible lexical knowledge. This means that lexemes rather than constructions may have a richer basis for variable categorization and so are more likely to be associated with different conceptualizations. (6)

Finally, the view of invariant verb meaning is inconsistent with the other assumptions adopted by Goldberg and other cognitive linguists, for example, the lexicon-syntax continuum and general acceptance of polysemy as the norm. In addition, it is also at odds with Goldberg's (1995) argument for "constructional bootstrapping" in the acquisition of verb meaning, a view that shows the inseparability of lexemes (including semantic and distributional properties) and constructions (Langacker 2003; Taylor 2003; Croft 2001; see also Section 3). If the two postulated constructions (11) and (27) share the same form but differ in constructional meaning, how does a child or even an adult listener know that sneeze just refers to the means, not the caused-motion event?

4.3. Restrictions on the verb

It is suggested in Section 4.1 that the verb in (5)-(9) and (22) can be treated consistently as being associated with semantic extension in the usage event, with the reference point relation underlying the asymmetry in the extension. As can be seen from those examples, semantic extension occurs in different types of constructions, including caused motion, ditransitive, transitive and motion constructions. In cognitive grammar terms, one could abstract the commonality shared by the different instances of extension into a schema that captures general restrictions on the semantic extension of the verb in usage events. In other words, the disparate instances of extension in verbal semantics have allowed a verb extension schema to emerge, with a certain degree of type frequency (see Croft and Cruse 2004). The schema is diagrammed in (28), which can be seen as a schematic abstraction of (21) and other instances of extension:

(28) [ILLUSTRATION OMITTED]

A prominent topic in studying novel argument structure patterns concerns the kind of verbs that are allowed in such patterns (see e.g., Goldberg 1995; Jackendoff 1990). Goldberg (1995, 1997, 1998) in particular has proposed a variety of possible relations between verbs and constructions, including, for instance, subtype, means, result, manner, etc., but it is not clear how the possible relations are constrained more generally, as it should be on this approach, by the supposed overriding power of constructional meaning. From the present perspective, the possible restrictions on the verb come down to the question of what kind of verb best serves as the standard in the categorizing relationship of extension indicated in (28).

First, the conventional value of the verb must relate to the target of extension as a salient reference point. This entails that the relational profile of the verb must construe a salient element in the domain of the target profile. This restriction subsumes the various relations between the verb and the construction mentioned above, especially means, result, manner and precondition. Goldberg (1995: 65) mentions the possible relevance of metonymy in these relations, but she stops short of an explicit formulation. Let's look at how the restriction works on some of the examples. The verb yawn could well construe a happening accompanying a motion event, but such circumstantial reading lacks sufficient salience to qualify as the reference point for caused motion. Furthermore, in our encyclopedic conception, yawning is not associated with forceful expulsion of air, in contrast to sneezing. So there does not seem to be any notion of force and causal potential in the conception of yawn. Thus the verb cannot be categorized as a caused-motion verb, witness (4). It should be noted from (28) that the reference point for a process need not be another process necessarily linked by the correspondence of force, as is the case in (21). Any circumstantial process construed as a salient element in a scenario can also serve as a reference point to a target process. A case in point is (7), where "eating" must be construed as a salient property of a long travel in order for the sentence to be felicitous. Likewise, nonsalient circumstantial element does not qualify as the process reference point. This is illustrated in the following (Goldberg 1995; Jackendoff 1990; Goldberg and Jackendoff 2004):

(29) a. *The bird chirped out of the cage.

b. *The car honked down the road.

c. *The dog barked out of the room.

d. The boat burned into the cave.

e. *Bill hid/crouched his way into the room.

f. Bill blushed his way to New York.

Arguably, the problem with these examples all comes down to whether the verb construes a sufficiently salient event that could evoke the targeted event within a single scenario. For sound emission verbs such as (29a-29c), Goldberg and Jackendoff (2004) (cf., Goldberg 1995) suggest that the verbal event must relate to the constructional event by the relation of "result" (cf., [22c], [22d]). They therefore propose a separate construction labeled sound emission path resultative. On the present analysis, the requirement for the "result" relation can be interpreted as the salience of the sound in the scenario of a motion event, which is a more general requirement stipulated in the verb extension schema in (28). Put differently, any sound accompanying the motion is not necessarily salient. Interestingly, the salience of sound emission verbs can be augmented in the way-construction, as in the case of (7). Consider the following (Goldberg and Jackendoff 2004: 541):

(30) a. The car honked its way down the road.

b. The dog barked its way out of the room.

Though I leave open the question why the way-construction appears to augment the salience of a circumstantial element, (7) the salience of sound emission in (30), as in the case of (7), should be rather obvious. This captures the contrast between (30) and (29b and 29c), noted by Goldberg and Jackendoff (2004).

Second, the conventional value of the verb must be conceptually linked to the extended value via the domain of the extension, as indicated by the correspondence links in the lower half of the diagram in (28). The correspondence of the trajectors is a necessary part of the conceptual linkage. This requirement forms the basis of the categorizing relationship, which is always built on conceptual correspondences. It also follows from the reference point ability motivating the choice of the prototype for the extension, since the reference point must show conceptual connectivity with its target (cf., Van Hoek 1997). It should be understood that the conceptual connectivity is a matter of degree. The sneeze-type of example in (1b) represents a case of strong connectivity wherein part of the base of the verb (i.e., the "force" element) directly corresponds to the profile in the extended value, as indicated in (21). For another example, consider an interesting case of correspondence based on very specific encyclopedic structures. The verb think is used in (9), owing to the abstract supernatural force transmission in its encyclopedic meaning, which corresponds to the profile of caused-motion in the extended value. This example also shows that the force need not be physical impact that causes motion. Invisible or psychological force could well have the same effect, as seen in the word clap in They clapped the pop star onto the stage. (8) The same remark readily applies to the verbs rinse, laugh in (8) and (5). Note that all these cases of strong connectivity correlate with the "means" reading of the verb. Another case represents weaker connectivity: the profile in the extended value does not find a correspondent in the prototype, though the base of the extended value is still linked to the conventional unit, as indicated in (28). Examples of this include the manner interpretation of the verb in (7), (22), and (30).

The condition on correspondence rules out cases where the conventional value of the verb contradicts the profile of the extended value, unless the contradiction is overridden by some other conceptual link (see below). This is illustrated by the following:

(31) a. *Sam convinced Bob into the room.

b. *He struck the ball across the field. (cf., He hit the ball across the field.)

Goldberg (1995) stipulates additional conditions to cover such cases, which are brought together in the following:

(32) a. No cognitive decision can mediate between the causing event and the entailed motion. (Goldberg 1995: 167)

b. If the action denoted by the verb implies an effect other than motion, then a path of motion cannot be specified. (Goldberg 1995: 170)

According to Goldberg, the verb in (31a) entails cognitive decision on the part of the entity denoted by the direct object whereas the verb in (31b) implies an effect other than motion. However, the verbs in these sentences share a more general property: they both entail a resultative change of state in the entity denoted by the direct object, either involving the intelligence (31a) or the physical state (31b) of the entity. In other words, the change of state is specified as part of the profile of the verb, which can sometimes be expressed:

(33) a. Sam convinced Bob that he should go into the room. (~Bob believed he should go into the room.)

b. He struck the ball fiat.

This shows that the profile of the verb contradicts the caused-motion profile in the extended value. In addition, such verbs cannot assume manner readings in (31). In such cases, one could argue, the conventional semantic pole of the verb is activated, which however cannot sanction the novel uses in (31).

Goldberg (1995) noted a somewhat different case:

(34) Sam carefully broke the eggs into the bowl.

Here the conventional profile of the verb implies a change in physical state associated with the direct object the eggs, which does not correspond to the caused-motion profile. But this contradiction is overridden by an entrenched scenario of an individual breaking an egg and causing it to move into a container. In cognitive grammar, such encyclopedic knowledge can be invoked as part of the symbolic resources in syntagmatic combination. Specifically, the Landmark (i.e., "the eggs") of break specifies the process of breaking in greater detail, adding a motion component in the base. This is consistent with Lascarides and Asher's (1993) theory of discourse in commonsense entailment (DICE), according to which world knowledge overrides lexical specifications and assigns new values (cf., Lascarides and Copestake 1998; Verspoor 1997). In other words, there is arguably a conceptual link between egg-breaking and egg-moving.

4.4. Variation in argument expression

The surface hallmark of novel verb usages discussed in this article is argument structure variation, in the sense that the nonconventionally used verbs either take additional arguments typically not associated with their lexical specifications or lose some of their "default" arguments. Traditionally, argument structure is taken to be a lexical property associated with verbs in the lexicon that is entirely separate from language use. Under such an assumption, variation in argument expression constitutes a central research question on the so-called lexicon-syntax interface. From the present perspective, any attempt to capture such variation has been hampered by its completely grammar-internal and "universalist" perspective. By contrast, variation in the type and number of arguments is easily captured by the usage-based orientation and the theory of categorization in cognitive grammar.

On the analysis proposed in this article, variation in argument expression is a complication related to the nature of the categorizing relationship that links a conventional verb unit and its contextual use, since the semantic arguments of the verb are necessarily involved in the categorization, as well as the process expressed by the verb. For instance, the case focused on in earlier discussion involves argument addition, since the intransitive verb sneezed takes a "fake" object the napkin in (1b), as is the case with the verbs laugh and smile in (5) and (6). In addition, as suggested earlier, the conventional 'sneezer' role associated with the verb is displaced in syntax by a 'causer' role, which is directly symbolized by the subject of the sentence, even though the subject is capable of double interpretation. These changes are a function of the semantic extension outlined earlier, which recategorizes both the process and its participants. On this analysis, the additional argument is provided by the construction, as in Goldberg's (1995) approach, in particular, the caused-motion verb schema included in (11). Where I depart from Goldberg is the proposed recategorization of the verb in the constructional context. As can be seen in the analysis in (21), the additional argument is also provided by the extended value of the verb.

A greater difficulty is posed by cases involving argument reduction, as with the verbs eat, rinse, and think in (7), (8) and (9) respectively, which are normally used transitively. The verb usage in these examples shows the suppression of the Landmark schematically specified by the verb. For example, in (8), repeated below as (35), the direct object the soap is not an argument specified by the conventional value of the verb rinse/ clean.

(35) Sam rinsed/cleaned the soap out of her eyes.

As Goldberg (1995) notes, (26) does not entail 'Sam rinsed/cleaned the soap'. Conceptually, what Sam rinsed/cleaned are 'her eyes', which happens to be coded in (35) as the Landmark of a motion event. On the approach of Goldberg (1995), the suppression of verbal arguments is attributed to the cutting function of constructions (see Goldberg 1995: 57f), but it is not made clear why constructions have such a function and how cutting is constrained. By contrast, the present approach again regards argument reduction as a matter of recategorization of the verb in usage events. For instance, in order for the verb rinse/clean to be categorized by extension as an instance of caused-motion verb in (35), its conventional Landmark must be suppressed in the grammar. Consider the following example:

(36) *Sam rinsed her eyes the soup out of her eyes.

The same can be said for the verb eat in the way-construction, whose conventional Landmark cannot be expressed:

(37) *We ate hot dogs our way across the U.S. (Jackendoff 1990: 212)

The verb usages in (36) and (37) can only activate the conventional value of the verb rinse and eat, not the caused-motion verb schema. The conventional units that are activated cannot sanction the argument structure of these sentences. Note that the subject argument is not suppressed, owing to the conceptual correspondence between the Trajector of the conventional value and that of the extended value, as indicated in (21) and (28). In this connection, one may ask why in (34) the verb break does not lose its Patient argument the eggs. A simple answer is that the Patient argument in this case happens to correspond to the Landmark in the extended value, and therefore need not be suppressed. Therefore, variation in argument expression is a result of the attempt to minimize any possible conflict between the conventional value and the extended value.

Argument reduction raises a question for the simultaneous activation by the verb usage discussed in Section 4.1. The idea is that the particular use of the verb in (35) should activate both the conventional unit rinse and the caused-motion verb schema. Now, in cognitive grammar, part of the semantic value of a verb is the schematic participants specified by the verb, which are to be elaborated by nominal expressions. In cases where a conventional participant in the semantic value of a verb is not directly expressed in the grammar, would the verb still be successfully activated? In (35), the linguistic context primes the conventional unit rinse (cf., Section 2), since its expected Landmark appears elsewhere in the sentence. With the use of the verb eat in (7), the situational context implicit in the sentence helps the activation of the conventional verb unit (Taylor 2002). It is well known that eat can be used intransitively in certain situations (see Goldberg 2005 for discourse conditions on argument omission). Under the network conception of linguistic knowledge, the picture that emerges is a division of labor in language use, which gives priority to constructional schemas rather than more distantly related units such as the prototype for the extension of a verb.

5. Concluding remarks

The various ways in which verbs are found in alternate argument structure patterns present a challenge to the traditional boundaries between lexicon and syntax as well as linguistic system and language use. This has become one of the thorny issues in mainstream linguistic theories. The approach advocated in this article adopts the usage-based perspective of cognitive grammar and takes more seriously the flexibility of language use, which can have repercussions on the linguistic system. On the present approach, lexical-syntactic flexibility in argument expression results from the unconventional use of a verb by semantic extension, which helps the particular verb usage to be sanctioned by a constructional schema. The asymmetrical character of the extension is motivated by the cognitive ability of invoking reference point. The mechanisms behind flexible verb use are of the same kind as those underlying nominal metonymy, both of which instantiate symbolic flexibility and manipulate the general ability of organizing knowledge in terms of categorizing relationships. The multiple facets in the value of a usage event, be it metonymy or verb frame alternation, are easily captured by the network of relationships. Based on a number of disparate constructional instances showing the same kind of extension (i.e., type frequency), an abstract verb extension schema is extracted and gets entrenched as a piece of grammatical knowledge. The extracted schema captures the general restrictions on the kind of verb that can be extended, hence constraining the space of a speaker's creativity.

Received 3 July 2003

Revised version received

6 July 2004

Nanjing University

References

Boas, Hans C. (2003). A Constructional Approach to Resultatives. Stanford: CSLI.

Bowdle, Brian F.; and Medin, Douglas L. (2001). Reference-point reasoning and comparison asymmetries. Paper presented at the 23rd Annual Conference of the Cognitive Science Society, Edinburgh, Scotland.

Chomsky, Noam (1981). Lectures on Government and Binding. Dordrecht: Foils.

Croft, William (1991). Syntactic Categories and Grammatical Relations. Chicago: University of Chicago Press.

--(2001). Radical Construction Grammar: Syntactic Theory in Typological Perspective. Oxford: Oxford University Press.

--(2002). The role of domains in the interpretation of metaphors and metonymies. In Metaphor and Metonymy in Comparison and Contrast, Rene Dirven and Ralf Porings (eds.), 161-205. Berlin: Mouton de Gruyter.

--(2003). Lexical rules vs. constructions: a false dichotomy. In Motivation in Language, Hubert Cuyckens, Thomas Berg, Rene Dirven, and Klaus-Uwe Panther (eds.), 49-68. Amsterdam: Benjamins.

Croft, William; and Cruse, Allan D. (2004). Cognitive Linguistics. Cambridge: Cambridge University Press.

Downing, Pamela (1977). On the creation and use of English compound nouns. Language 53, 810-842.

Fauconnier, Gilles; and Turner, Mark (1996). Blending as a central process of grammar. In Conceptual Structure, Discourse and Language, Adele Goldberg (ed.), 113-129. Stanford: CSLI.

Fillmore, Charles (1985). Frames and the semantics of understanding. Quaderni di semantica 6, 222-254.

Fodor, Jerold A. (1970). Three feasons for not deriving kill from cause to die. Linguistic Inquiry 1, 429-438.

Gleitman, Lila (1990). The structural sources of verb meaning. Language Acquisition 1, 3-55.

Goldberg, Adele E. (1995). Constructions: A Construction Grammar Approach to Argument Structure. Chicago: University of Chicago Press.

--(1997). The relationships between verbs and constructions. In Lexical and Syntactical Constructions and the Construction of Meaning, Marjolijn Verspoor and Eve Sweetser (eds.), 383-398. Amsterdam: Benjamins.

--(1998). Semantic principles of predication. In Discourse and Cognition, Jean-Pierre Koenig (ed.), 41-55. Stanford: CSLI.

--(2005). Constructions, lexical semantics and the correspondence principle: accounting for generalizations and subregularities in the realization of arguments. In The Syntax of Aspect, N. Erteschik-Shir and T. Rappaport (eds.), 215-236. Oxford: Oxford University Press.

Goldberg, Adele E.; and Jackendoff, Ray (2004). The English resultative as a family of constructions. Language 80(3), 532-568.

Hudson, R. (2001). Language as a cognitive network. In A Cognitive Approach to the Verb: Morphological and Constructional Perspectives, H. G. Simonsen and R. T. Endresen (eds.), 49-72. Berlin: Mouton de Gruyter.

Jackendoff, Ray (1983). Semantics and Cognition. Cambridge, MA: MIT Press.

--(1990). Semantic Structure. Cambridge, MA: MIT Press.

--(1997). The Architecture of Language Faculty. Cambridge, MA: MIT Press.

--(2002). Foundations of Language. New York: Oxford University Press.

Kumashiro, Toshiyuki; and Langacker, Ronald (2003). Double-subject and complex-predicate constructions. Cognitive Linguistics 14(1), 1-45.

Langacker, Ronald W. (1987). Foundations of Cognitive Grammar, vol. 1. Stanford: Stanford University Press.

--(1988). The nature of grammatical valence. In Topics in Cognitive Linguistics, B. Rudzka-Ostyn (ed.), 91-125. Amsterdam: Benjamins.

--(1991). Foundations of Cognitive Grammar, vol. 2. Stanford: Stanford University Press.

--(1993). Reference-point constructions. Cognitive Linguistics 4, 1-38.

--(1997). The contextual basis of cognitive semantics. In Language and Conceptualization, Jan Nuyts and Eric Pederson (eds.), 229-252, Cambridge: Cambridge University Press.

--(1998). Indeterminacy in semantics and grammar. In Estudios de Linguistica Cognitiva II, Jose Luis Cifuentes Honrubia (ed.), 649-672. Alicante: Universidad de Alicante.

--(1999). Grammar and Conceptualization. Berlin: Mouton de Gruyter.

--(2001). Dynamicity in grammar. Axiomathes 12, 733.

--(2003). Construction grammars: cognitive, radical, and less so. Paper presented at the International Cognitive Linguistics Conference, Universidad de La Rioja.

Lascarides, Alex; and Copestake, Ann (1998). Pragmatics and word meaning. Journal of Linguistics 34(2), 387-414.

--; and Asher, Nicholas (1993). Temporal interpretation, discourse relations and common-sense entailment. Linguistics and Philosophy 16(5), 437-493.

Levin, Beth (1993). English Verb Classes and Alternations. Chicago: University of Chicago Press.

Levin, Beth; and Rappaport Hovav, M. (1988). Lexical subordination. In Papers from the 24th Meeting of the Chicago Linguistics Society, 275-289. Chicago: CLS.

Mateu, J. (2000). Path and telicity in idiomatic constructions: a lexical-syntactic approach to the Way construction. Paper presented at the 2000 ESSLLI Workshop on Paths and Telicity in Event Structure, August, University of Birmingham.

Panther, Klaus-Uwe; and Thornburg, Linda (2002). The role of metaphor and metonymy in-er nominals. In Metaphor and Metonymy in Comparison and Contrast, Rene Dirven and Ralf Porings (eds.), 279-319. Berlin: Mouton de Gruyter.

Pinker, Steven (1989). Learnability and Cognition. Cambridge, MA: MIT Press.

Pustejovsky, James (1995). The Generative Lexicon. Cambridge, MA: MIT Press.

Rappaport Hovav, M.; and Levin, Beth (1998). Building verb meanings. In The Projection of Arguments, Miriam Butt and Wilhelm Geuder (eds.), 97-134. Stanford: CSLI.

Slobin, Dan I. (2000). Verbalized events: A dynamic approach to linguistic relativity and determinism. In Evidence for Linguistic Relativity, S. Niemeier and R. Dirven (eds.), 107-138. Amsterdam: Benjamins.

--(2003). Language and thought online: cognitive consequences of linguistic relativity. In Language in Mind, Dedre Genter and Susan Goldin-Meadow (eds.), 157-191. Cambridge, MA: MIT Press.

Talmy, Leonard (1985). Lexicalization patterns: semantic structure in lexical forms. In Language Typology and Syntactic Description, vol. 3, Tim Shopen (ed.), 57-149. New York: Cambridge University Press.

Taylor, John R. (1995). Linguistic Categorization: Prototypes in Linguistic Theory. Oxford: Clarendon Press.

--(1996). Possessives in English: An Exploration in Cognitive Grammar. Oxford: Clarendon Press.

--(2002). Cognitive Grammar. Oxford: Oxford University Press.

--(2003). Meaning and context. In Motivation in Language, Hubert Cuyckens, Thomas Berg, Rene Dirven, and Klaus-Uwe Panther (eds.), 27-48. Amsterdam: Benjamins.

Tversky, A. (1977). Features of similarity. Psychological Review 84, 327-352.

Van Hoek, Karen (1997). Conceptual Structure and Anaphora. Chicago: University of Chicago Press.

Van Hout, Angeliek (1998). Event Semantics of Verb Frame Alternations. New York: Garland.

Verspoor, C. M. (1997). Contextually dependent lexical semantics. Unpublished doctoral dissertation, Center for Cognitive Science, University of Edinburgh.

van der Leek, Frederike (2000). Caused-motion and the 'bottom-up' role of grammar. In Constructions in Cognitive Linguistics, Ad Foolen and Frederike van der Leek (eds.), 301-331. Amsterdam: Benjamin.

Notes

* Correspondence address: English Department, School of Foreign Studies, Nanjing University, Nanjing, China 210093. E-mail: ren.zhang@utoronto.ca.

(1.) This observation is inspired by a reviewer, who suggested the contrast between read Langacker and read Langacker's works as a problem for Croft (2002). However, Croft's analysis could easily capture the contrast by suggesting that metonymy results in a type concept (WORKS BY LANGACKER), not a grounded concept (LANGACKER'S WORKS).

(2.) The sentence seems to be acceptable with the manner interpretation of the verb (cf., Bill belched his way noisily out of the restaurant (Jackendoff 1990)).

(3.) Mateu (2000) similarly criticizes as a stipulation the conceptual paraphrases suggested by Jackendoff (1990) that subordinate the main verb to a conceptual by-phrase.

(4.) Mateu (2000) uses these criticisms to argue for his syntactic approach. It's beyond the scope of this article to examine his approach in detail, but I have shown that the criticisms can be properly taken care of within the framework of cognitive grammar.

(5.) Constructional polysemy cannot be the answer here, as (27) is just more specific than (11).

(6.) See Croft (2003) and Van der Leek (2000) for criticisms of postulated constructional senses.

(7.) A likely reason seems to be related to the affectedness of the 'way' by the verbal event, assuming that the word is not a meaningless marker even in the manner interpretation, contra Goldberg (1995).

(8.) I owe this example to an anonymous reviewer.