Cognitive approaches to grammar
Construction Grammar
Networks
As in Cognitive Grammar, grammatical knowledge is represented in networks. Instead of categorizing relations, Construction Grammar makes use of inheritance links (as in computer science) to relate constructions to one another. Four major types of inheritance links are recognized: polysemy links, metaphorical extension links, subparts links and instance links (see Goldberg 1995: 74–81). Before briefly illustrating them, it should be observed that inheritance links are treated as objects in Construction Grammar (see Goldberg 1995: 74–75). This differs from Cognitive Grammar (and Radical Construction Grammar), where the only elements postulated in grammar are the semantic and phonological poles of a construction plus their symbolic links. Like other cognitive models, however, Construction Grammar allows for multiple inheritance, i.e. a construction can be categorized by two (or more) independently established constructions. For example, the intransitive motion construction (e.g. The car screeched around the corner) inherits from both the intransitive construction (since the intransitive motion construction is quite trivially intransitive) and the motion construction (since motion is predicated of one of the constructional elements, i.e. the subject, via an oblique phrase).
Polysemy links and metaphorical links roughly correspond to extension relations in Cognitive Grammar. Construction Grammar claims for example that the double object construction has a variety of senses which can be viewed as extensions (via polysemy links) from the central sense “[a]gent successfully causes recipient to receive patient” (see Goldberg 1995: Fig. 2 in Word classes and grammatical functions in Cognitive Grammar). A metaphorical link connects the caused motion construction and the resultative construction (see (1d) and (1e) in Predictability and polysemy) in that the latter is said to have originated from the former via the metaphorical construal of states (e.g. crazy in Chris drove Pat crazy) as locations. A subpart link is posited when a construction is “a proper subpart of another construction” (Goldberg 1995: 78). The intransitive motion construction is linked to the caused-motion construction via such a link since the former is a proper subpart of the latter (the only missing element in the intransitive motion construction is the “cause” argument; for a different view see Broccias 2003: Ch.5). Finally, instance links correspond to elaborative categorization in Cognitive Grammar. This obtains when a construction specifies another construction in more detail (e.g. Chris drove Pat crazy instantiates the schematic resultative construction).
It should be pointed out that Construction Grammar, at least in its 1995 version, also makes use of constraints to motivate the (alleged) impossibility of some patterns. Among them is the constraint (see Broccias 2003: 213–218 for a review of Goldberg’s 1995 constraints) that states that no intermediary time intervals are possible between the causing event and the resultant event in a resultative construction (as in Chris shot Pat dead, where death is said to necessarily occur immediately after the shooting event). Even if we accepted the descriptive correctness of this and other proposed constraints (but see Broccias 2003: Ch. 6), it remains to be established how they are represented in the grammar. One might claim that, in a truly cognitive model of grammar, constraints cannot be independent objects in the grammatical system. Constraints should simply amount to paraphrases of schemas which subsume a large amount of specific instances (see e.g. Taylor’s 2002: Ch. 5 discussion of phonological constraints as phonological schemas).1 Broccias (2003) argues, for example, that Goldberg’s (1995) constraints for the caused motion construction and the (transitive) resultative construction are just paraphrases of the conceptual import of the schema to which both construction types can be related, i.e. the billiard-ball schema.2
1. Taylor (2002) stresses that phonology is autonomous, i.e. it does not necessarily symbolize the semantic pole and it obeys constraints of its own. Despite this, phonological constraints are analyzed as phonological schemas. Their nature, therefore, is in no way different from that of constraints pertaining to symbolic units: both are schemas abstracted away from specific instances.
2. The billiard-ball model captures our conceptualization of events in terms of energetic interactions between two entities resulting in some change (of either state or location) of the affected entity, see Langacker (1991: 13).
