Cognitive (Construction) Grammar

Cognitive (Construction) Grammar RONALD W. LANGACKER* Abstract Goldberg overstates the di¤erences between Cognitive Grammar and Cog- nitive Construction Grammar. The former does not claim that a clause in- variably inherits its profile from the verb; it has merely been suggested that the latter’s preference for monosemy may have been pushed too far. The matter can only be addressed given a specific definition of what is meant in saying that a verb ‘‘has’’ a certain sense. Also, the schematic meanings proposed in Cognitive Grammar for basic grammatical notions do not imply a ‘‘reductionist’’ or ‘‘essentialist’’ view based on classical categorization. Instead they complement the characterization of these notions as ‘‘metage- neralizations over construction-specific categories’’, which otherwise begs the question of why the distributional patterns supporting such generaliza- tions should be observed in the first place. Keywords: Cognitive Grammar; categorization; construction; grammati- cal category; verb meaning. With Constructions at Work, Goldberg has once more raised and illumi- nated fundamental issues through an insightful blend of analysis, theoret- ical discussion, and varied empirical evidence. I particularly appreciate her treatments of island constraints and subject-auxiliary inversion, which clearly point the way to the resolution of these classic problems. Here, though, I consider just five pages of this important work, namely those comparing Cognitive Grammar (CG) and Cognitive Construction Gram- mar (CCxG). Cognitive Linguistics 20–1 (2009), 167–176 DOI 10.1515/COGL.2009.010 0936–5907/09/0020–01676 Walter de Gruyter * Department of Linguistics, University of California, San Diego. Author’s email: 3rlangacker@ucsd.edu4. In this passage (§10.6, pp. 220–225), Goldberg responds to my own comparison of the two frameworks (Langacker 2005a, 2005b), which be- gins with a long list of tenets that they share. The roster of disagreements is considerably shorter. Goldberg notes just four, which (in a di¤erent order) she states as follows (221): 1. The too-restrictive definition of construction in Goldberg (1995); 2. The alleged adoption by CCxG of autonomous syntax; 3. The fact that Goldberg (1995) allows the construction itself to be the profile determinant of the clause instead of universally requiring the verb to be the profile determinant; 4. The fact that RCxG [Radical Construction Grammar] and CCxG are non-reductionist and do not adopt Cognitive Grammar’s essentialist definitions of grammatical categories and functions. I have certain disagreements about this list of disagreements. They are worth discussing with the aim of bringing some basic issues into clearer focus. Point 1 can be disposed of quickly, as Goldberg no longer defines a construction as a ‘‘not strictly predictable pairing of form and func- tion’’ (224). Instead she allows that ‘‘facts about the actual use of lin- guistic expressions such as frequencies and individual patterns that are fully compositional are recorded alongside more traditional linguistic generalizations’’ (45). While the original position was ‘‘conservative methodologically—we know we must mentally represent a construction if there is anything not strictly predictable about it’’ (224)—the exclu- sion of fixed but regular expressions seems arbitrary from a psychological standpoint. We agree, then, that ‘‘the issue of whether a construction actually exists as an established psychological entity’’ has to be distin- guished from ‘‘the very di¤erent issue of whether one can prove its exis- tence to the satisfaction of other analysts’’ (Langacker 2005b: 143). Just to be perverse, however, let me add a further wrinkle that I noted as part of the same discussion: one could very well claim that ‘‘entrench- ment and conventionalization always result in some measure of idiosyn- crasy vis-a`-vis other constructions. It can be argued that unit status invariably narrows the range of interpretive options in subtle ways, or that evoking something as a prepackaged unit implies a kind of process- ing e‰ciency which makes it distinct from an otherwise equivalent non- unit structure’’ (142). In this case the non-predictability of constructions follows from their very nature. As a methodological requirement, though, it would be both impractical and superfluous. Point 2 is grossly overstated. I did not accuse CCxG of adopting auton- omous syntax, and I would never accuse Goldberg of such a crime. I 168 R. W. Langacker merely suggested that a few specific positions could be seen as vestiges of generative thinking, and that the attitude in regard to grammatical cate- gories and functions (point 4) results in CCxG being a less radical alter- native to it than is CG. More on the latter below. As for the former, one position reminiscent of classic generative thinking is precisely the exclusion from ‘‘the grammar’’ of expressions predictable on the basis of more general patterns—the ‘‘rule/list fallacy’’ (Langacker 1987: Ch. 1). That is no longer an issue. It does however exemplify a second posi- tion, namely Goldberg’s occasional appeal to ‘‘parsimony’’ in a way that recalls the transformational motto ‘‘the shortest grammar is the best grammar’’. A third position was generativity: ‘‘Construction Grammar is generative in the sense that it tries to account for the infinite number of expressions that are allowed by the grammar while attempting to account for the fact that an infinite number of expressions are ruled out or dis- allowed’’ (Goldberg 1995: 7). Precisely what is meant by ‘‘account for’’ and ‘‘(dis)allowed’’ is left unclear. I will not discuss this matter here (see Langacker 2005a: 158–160) except to note a certain resemblance to the characterization of a generative grammar as an algorithmic device enu- merating ‘‘all and only the grammatical sentences of a language’’, which are taken to be an infinite and well-defined set. If strictly interpreted, all three positions are at odds with the ‘‘usage-based’’ perspective (Barlow and Kemmer 2000; Bybee and Hopper 2001). I was not however suggest- ing that these were essential features of CCxG, but rather that they seem- ingly ran counter to its general spirit. Point 3 responds to my critique (Langacker 2005b) of how Goldberg treats the relation between verbs and constructions. For the most part our treatments agree—my critique was limited to one specific aspect of her account. I have little doubt that I failed to present my view on this matter as clearly as I should have. As it stands, however, point 3 is a seri- ous misrepresentation of my position. At issue is the relation between verb meaning and constructional mean- ing, and how many senses should be ascribed to a verb on the basis of its occurrence in di¤erent constructions. It was of course Goldberg (1995) who raised this important issue and showed—most strikingly with exam- ples like (1)—that the grammatical organization of a clause is not invar- iably determined by its lexical verb. (1) He sneezed the napkin o¤ the table. Instead, the construction itself may be responsible for factors like profil- ing, argument structure (trajector/landmark alignment), and even essen- tial conceptual content (e.g., the causation of motion). Since I explicitly accepted Goldberg’s analysis of this example, I can hardly be accused of Cognitive (Construction) Grammar 169 ‘‘universally requiring the verb to be the profile determinant’’. This erroneous accusation was also made in Goldberg and Jackendo¤ 2004 (pp. 533–534), where I was cited as one who assumes that the comple- ment structure is determined by the verb alone in sentences like those in (2): (2) a. I’ll fix you a drink. b. Fred watered the plants flat. c. Bill belched his way out of the restaurant. d. We’re twistin’ the night away. My critique of Goldberg 1995 focused on the evident attitude that most verbs have just a single construction-related sense, so that when a verb occurs in multiple structural frames, only one is congruent with its mean- ing. Among the examples considered were send, which appears in both the caused-motion and ditransitive constructions, and the caused-motion use of kick: (3) a. She sent a package to her uncle. b. She sent her uncle a package. c. He kicked the ball into the stands. Goldberg once objected (p.c.) to my positing multiple senses for send. A caused-motion sense of kick is used to illustrate the circularity of ‘‘posit- ing a new sense every time a new syntactic configuration is encountered and then using that sense to explain the existence of the syntactic config- uration’’ (1995: 12). It would also be rejected on grounds of parsimony: ‘‘the semantics of . . . the full expressions are di¤erent whenever a verb occurs in a di¤erent construction. But these di¤erences need not be attrib- uted to di¤erent verb senses; they are more parsimoniously attributed to the constructions themselves’’ (1995: 13). I merely suggested that this preference for monosemy might have been pushed too far. Since verbs are learned in the context of constructions, they should generally take on the value of the verb slot in those construc- tions. In the case of send, the caused-motion and ditransitive uses are both so frequent and familiar that denying the emergence of congruent senses would seem quite implausible. In the case of kick, I judged the caused-motion use to be familiar enough that a congruent sense seems likely to have some cognitive status. Thus send and sneeze lie at opposite extremes of the scale representing the extent to which a caused-motion sense is entrenched and conventionalized, with kick falling somewhere in between. In short, it should not be presumed that every verb used in a secondary pattern is analogous to sneeze in (1). Neither should it be pre- 170 R. W. Langacker sumed, of course, that every occurrence of a verb implies an established verb meaning congruent with the meaning of the clause containing it (point 3). This matter can hardly be resolved without addressing the more funda- mental issue of what it actually means to say that a verb ‘‘has’’ a certain sense. In the usage-based perspective of CG (Langacker 2000), a verb ‘‘has’’ a construction-congruent sense to the extent that a constructional subschema emerges in which it occupies the verb slot and is thus appre- hended as an instance of the type of verb the construction specifies. This does not su¤er from the circularity Goldberg objects to, since mere occur- rence in a construction is not taken as establishing the requisite sub- schema: it is posited only on grounds of entrenchment and conventionali- zation, which can in principle be determined empirically (e.g., through corpus investigation). Nor is there any inappropriate lack of parsimony, as the same subschema belongs to the networks of variants characterizing both the verb and the construction. It is not a matter of attributing the relevant semantic properties to either the verb or the construction consid- ered in isolation from the other. Instead, the interaction between the verb’s conceptual content and the construction-induced construal gives rise to what—when viewed from the standpoint of the verb—constitutes a construction-congruent meaning. I am not aware of ever having claimed or implied that the verb is ‘‘uni- versally’’ required ‘‘to be the profile determinant’’ of a clause. Goldberg must however be credited for focusing attention on this issue and showing me the need to clarify my thoughts in regard to it. In a subsequent work (Langacker in press) I have tried to be more precise about when or whether a verb’s occurrence in a construction is likely to induce a second- ary, construction-congruent sense. Very briefly, I adopt the term skewing for a discrepancy between the process profiled by a verb and the one designated by the clause as a whole. I then distinguish between skewing uses, skewing elements, and skewing constructions. Sneeze in (1) repre- sents a skewing use, where a verb is non-congruent with the construction it appears in. This is the situation in which a new, congruent sense is most likely to develop. Through occurrence in expressions like (3)(c), kick comes to be apprehended as a caused-motion verb. To the extent that this usage becomes entrenched and conventionalized, kick can be said to ‘‘have’’ a caused-motion sense, especially when—through loss of analyzability—this construal comes to be directly accessible (no longer being based on activation of the simple transitive sense). Only time will tell whether sneeze will ever develop a caused-motion sense in this manner (which is not impossible should force-dynamic sneezing become prevalent and culturally salient). Cognitive (Construction) Grammar 171 There is no comparable encouragement of a new verb sense when the discrepancy between verbal and clausal meaning is due to a skewing element or a skewing construction. An example of a skewing element is the passive participial morpheme, which imposes an alternate choice of trajector on the process profiled by the verb it combines with (Langacker 1982). In a passive like (4)(a) there is thus a discrepancy between the argument structure of destroy and that of the clause as a whole. This is not however a skewing use, since destroy is fully compatible with the verb slot in the passive construction. The skewing is due to the participial morpheme, not to any mismatch between the verb and the slot it fills; the construction-congruent sense is simply the basic sense, despite its discrep- ancy with the clause-level profile. And since the verb is used with its basic value, this provides no impetus toward a new verbal meaning. (4) a. The town was destroyed by a hurricane. b. The yard was swarming with locusts. Alternatively, the skewing can be e¤ected by the overall construction, rather than any specific element. For instance, the well-known construc- tion in (4)(b) selects as clause-level trajector the location of the verbal process, leaving the verb’s trajector to be specified periphrastically as the object of with. But since the construction itself imposes this discrepancy, the verb is once more being used with its basic value in conformity with the verb slot of the sanctioning constructional schema. Likewise for the expressions in (2)(b)–(d). (As for (2)(a), one can plausibly argue that fix does indeed have a ditransitive sense owing to conventional occurrence in the ditransitive subconstruction based on verbs of creation.) That leaves us with point 4, pertaining to the CG claim that basic grammatical notions like noun, verb, subject, and object have schematic conceptual characterizations valid for all instances. Once again, Goldberg overstates my limited critique and ascribes to me an extreme position I do not recognize as my own. I did not say that the eschewal of my concep- tual definitions by CCxG and RCxG constitutes ‘‘an endorsement of strongly autonomous syntax’’ (p. 221), but merely suggested that the re- luctance to take them seriously (widespread even in cognitive linguistics) represents a vestige of generative (as well as structuralist) thinking. And while I speak of ‘‘reducing’’ these notions to something more fundamen- tal, I do not think my views are properly described as radically ‘‘reduc- tionist’’ or ‘‘essentialist’’. Goldberg’s discussion implies that my position is essentialist in the sense of embracing the classical view of categories and reductionist in the sense of denying the possibility of emergent properties. Neither char- acterization is accurate. As for linguistic categories generally, I assume 172 R. W. Langacker that some form of prototype categorization is appropriate for basic no- tions like noun, verb, subject, and object. It is merely hypothesized that these particular categories are further unified by a schematic characteriza- tion valid for all instances. Indeed, their dual characterization in terms of a prototype (reflecting a conceptual archetype) and a schema (reflecting a basic cognitive ability) may well contribute to their fundamental role in language structure. And in speaking of ‘‘reducing’’ grammar to some- thing more fundamental, I never intended to foreclose the possibility of emergent properties (though I should of course have guarded more care- fully against this interpretation). I have always maintained that an over- arching schema (when one can be posited) does not by itself constitute the full description of a linguistic category. Its full characterization consists in the entire network of conventionally established variants (Langacker 1987: §10.1). As Goldberg notes, CCxG and RCxG emphasize the construction- specific properties of grammatical categories and relations. Through its usage-based stance, CG does so as well, with every construction implicitly defining lexical categories, semantic roles, and grammatical relations in terms of the elements that appear in it. The lexical categories defined in this manner are often referred to as ‘‘distributional classes’’. Now CG, CCxG, and RCxG agree that distributional classes do not provide the ba- sis for general characterizations of notions like noun, verb, subject, and object. Even in a single language, there may be no construction in which appear all and only those elements commonly recognized as nouns or verbs. And if subject or object is defined in a given language by participa- tion in certain grammatical constructions, this does not a¤ord a universal characterization, since the defining constructions di¤er from language to language. Goldberg subscribes to Croft’s position that labels like noun, verb, subject, and object are ‘‘metageneralizations over construction-specific categories’’, thus allowing for their ‘‘functional characterizations’’ (221). Croft says that ‘‘syntactic roles define regions in conceptual space that represent semantically related groupings of participant roles in events . . . The structure of the conceptual space reflects the hierarchies of implica- tional relations governing the participant role distribution patterns of these constructions across and within languages’’ (2001: 170–171). I rec- ognize both the validity and the relevance of this bottom-up approach, which o¤ers a way to accommodate the linguistic importance of these elemental grammatical notions while maintaining the firm basis of ob- served distribution. This is not at all inconsistent with my top-down char- acterization in terms of basic mental operations. Though speculative, I would argue that these characterizations are plausible from both the Cognitive (Construction) Grammar 173 cognitive and the linguistic standpoint. I suggest, moreover, that they (or something comparable) are needed in a full account. They play a role in explaining why the distributional patterns supporting the metageneraliza- tions—and indeed, the specific constructions they are based on—should be observed in the first place. I view the mental operations in question as being inherent in the conceptual archetypes and aspects of clausal organi- zation which anchor the target categories, and thus as being responsible for their emergence. Space permits only a brief synopsis of relevant considerations. First, the standard argument against the conceptual characterization of basic categories is fallacious, being based on erroneous assumptions (Lan- gacker 2005b: 122–123). Not only should we thus be open to the possibil- ity of schematic definitions, but arguably their existence ought to be the default expectation for such fundamental notions if grammar is seen as meaningful and meaning resides in conceptualization or cognitive pro- cessing. Second, the cognitive abilities invoked in the characterizations (conce