Cognitive approaches to grammar
Construction Grammar
Discussion
There are various differences between Construction Grammar, on the one hand, and other cognitive models like Cognitive Grammar and Radical Construction Grammar, on the other. We have already pointed out the possibly problematic postulation of constraints or generalizations, which cannot be assumed as independent objects in Cognitive Grammar, for example. Similarly, the fact that constructions are related to one another does not rest in Cognitive Grammar on the postulation of links as objects as is the case in Construction Grammar. Relatedness in Cognitive Grammar is a necessary corollary of the language-independent cognitive ability of categorization, which structures any type of information in terms of a standard (of comparison) and a target (of conceptualization), i.e. all information is stored in networks by default. There is no need to postulate links as independent objects in Cognitive Grammar.
Construction Grammar also makes use of abstract predicates (e.g. CAUSE, MOVE, see Figure 3 in Predictability and polysemy) and roles (both semantic, e.g. theme, and syntactic, e.g. SUBJ) but their nature is not detailed any further. That is, Construction Grammar, unlike Cognitive Grammar and Radical Construction Grammar, does not take a stance on their universality (or lack thereof).
Even more importantly, although constructions (like linguistic symbols in Cognitive Grammar) are defined as pairs of form and meaning, form is equated with syntax in Construction Grammar. That is, syntax is taken to symbolize (i.e. express) the semantic level, see Figure 3 in Predictability and polysemy. Langacker (2005a) objects to this analysis because it implies that syntax is treated as a level independent of semantics (as in modular approaches). In Cognitive Grammar what is (traditionally) called syntax simply amounts to a collection of (sufficiently) abstract schemas, i.e. assemblies of semantic and phonological structure. A pattern like SUBJ VERB OBJ OBL (instantiated for example by Pat sliced the carrots into the salad) cannot be taken as pure form in Cognitive Grammar. “Subject” and “object”, for example, refer to, respectively, the trajector and landmark of the relation designated by the verb and hence they profile concepts (although their phonological pole is maximally non-specific). Similarly, “verb” stands for a pairing of meaning and form. Its semantic pole designates a sequentially scanned relation and its phonological pole is non-specific (although we know, among other things, that most verbs usually take a final alveolar plosive segment when used to refer to the past, etc.). In sum, in Cognitive Grammar what symbolizes the semantic pole is the phonological pole, not syntax, because syntax in Cognitive Grammar is not pure form.
We can now turn to the issue of lexical (vs. constructional) polysemy, which also sets Construction Grammar apart from other cognitive models. As was pointed out, Construction Grammar aims to minimize lexical polysemy. Slice in (1) is analyzed as having only one meaning, roughly “cut with a sharp instrument”. The fact that slice can be used with a caused motion meaning, is attributed to the interaction between the verb and the caused motion construction. Construction Grammar does not assume that this type of interaction results in the verb’s acquiring a new meaning (i.e. in the verb being polysemous). More generally, such an analysis implies that Construction Grammar subscribes to a parsimonious view of grammar (i.e. “the best grammar is the shortest grammar”). A new (e.g. caused motion meaning) for slice is not postulated because this meaning can be retrieved from our knowledge that slice can be used in a caused motion construction. Even if we accepted this position as correct (i.e. cognitively plausible), at closer inspection there seems to be no clear theoretical advantage (in terms of parsimony) by reducing the number of word meanings. Such a move automatically implies an increase in the number of postulated constructions, whose inventory is potentially infinite. This is exactly analogous to what happens in lexicalist approaches, where the number of word meanings is potentially endless while the number of constructions is minimized. Further, Construction Grammar faces the same problem as do lexicalist approaches: how can we distinguish between constructions (i.e. word meanings in lexicalist approaches)? Are, for example, the prepositional dative construction (e.g. Sally sent a present to Chris) and the caused motion construction truly two different constructions?
Be that as it may, the more general point is that the parsimonious view of grammar is not necessarily descriptively, i.e. cognitively, adequate (i.e. it does not necessarily reflect how our mind works) and, hence, questions such as those concerning the distinction between constructions and lexemes may turn out to be vacuous. There is no compelling reason, within a cognitive model of language, to assume (a) that word meanings (vs. constructions) should be kept at a minimum (or vice versa), (b) that word meanings can neatly be separated from constructional meanings and (c) that (allegedly) different lexical meanings, or (allegedly) different constructional meanings for that matter, can be isolated from one another in clear-cut fashion.
In sum, the treatment of the relation between word meanings and constructions in Construction Grammar differs greatly from the one offered in Cognitive Grammar. Langacker (2005a) claims that in the case of the caused motion construction, which he incidentally conflates with the prepositional dative construction, a continuum should be recognized. Langacker states that the verb send (as in Sally sent a present to Chris) should be regarded as coinciding with the caused motion construction itself. In other words, send is regarded as having a caused motion meaning by default: the constructional meaning and the verbal meaning overlap completely. By contrast, in Goldberg’s (1995) famous example Sally sneezed the napkin off the table (i.e. Sally caused the napkin to fall off the table by sneezing), sneeze is not analyzed by Langacker (2005a) as a caused motion verb. Rather, as in Goldberg’s approach, the caused motion meaning is viewed as emerging from the merger of the verb sneeze with the caused motion construction: the caused motion meaning is imposed upon the verb by the construction and the verb has not (at least yet) developed a caused motion meaning. Between send and sneeze lies the verb kick (as in John kicked the football into the stadium): speakers may have difficulty in deciding whether it has a caused motion meaning or not.1
All in all, the Cognitive Grammar analysis of verbal vs. constructional meaning seems more consonant with truly cognitive assumptions about language. If one accepts that much in language is a matter of degree, then there is no a priori reason to assume the need for minimizing lexical polysemy, as does Construction Grammar. Still, the importance of Construction Grammar resides, among other things, in having brought to the fore the treatment of constructions (in the traditional sense, i.e. assemblies of two or more units) as “objects” of the linguistic system in their own right and in having stressed that they, like lexical items, can be polysemous. Of course, it remains to be seen to what extent analyses based on constructional polysemy networks can be regarded as psychologically plausible (cf. Langacker’s 2005b claim that networks may be more “diffuse” than is usually assumed; see also Sandra and Rice’s 1995 seminal paper on the psychological reality of network mapping and Broccias’s 2005a critique of network approaches like Evans and Tyler’s 2004).
1. The idea that the degree of overlap between verbal meaning and constructional meaning is a matter of degree is expanded on in Langacker (2005b). In this paper, he claims that “an element which regularly fills the verb position in a construction tends to develop a meaning congruent with that position”.
