Five key features of RCG
Croft (2001: 362–3) states that RCG can be summed up in five key points, which we briefly summarise in this section.
Primitive units
Firstly, Croft assumes that the construction is the only primitive unit in the grammar, and may therefore be either simplex or complex in terms of structure and either specific or schematic in terms of meaning. However, only overt (which is to say fully substantive) constructions, such as independent words, can be recognised as atomic in Croft’s model. This means that grammatical categories (for example, word classes like noun and verb, or grammatical functions like subject and object) have no independent status, but are defined in relation to the constructions within which they occur. This explains why the relevant line in Table 20.4 is shaded: in RCG, word classes do not exist as primitive categories. This does not mean that words do not exist, but that words cannot be categorised into word classes that have any independent reality. Instead, words are just part of individual constructions. In this respect, the RCG model is diametrically opposed to the ‘words and rules’ model, where the words are the primitives and the constructions are epiphenomenal. In the RCG model, constructions are the primitives, and word classes, as they emerge from constructions, are epiphenomenal. From this perspective, it is to be expected that the types of word classes that we observe from one language to another might be significantly different, and because no universal word classes are posited, this cross-linguistic variation is not only unproblematic but predicted. Croft therefore argues against the traditional distributional approach to word classes (Chapter 14), which holds that they can be identified by morphological and syntactic properties. In support of this position, Croft (2001: 29) points out that some languages lack some of the relevant features that define the distributional approach (the lack of inflectional morphology in Vietnamese, for example), and that other languages might have the relevant features but reveal such different patterns of distribution that it is difficult to arrive at meaningful distributional criteria. Croft therefore argues against universal primitives, and also argues against the independent existence of word classes within any given language. Instead, Croft argues in favour of language-specific constructions, and in favour of construction-specific elements (grammatical subparts) and components (semantic subparts).
Syntactic relations and constituent structure
Secondly, the only syntactic relations admitted in the RCG model are the part whole relations that hold between the construction as a whole and the syntactic elements that fill it. In other words, the model does not recognise grammatical relations (grammatical functions) like subject and object as having any independent reality outside of individual constructions. Instead, to the extent that grammatical functions emerge from constructions, these also have the status of construction-specific epiphenomena. In this model, constituency is conceived in terms of grouping, where grammatical units are identified in terms of contiguity and prosodic unity, and heads receive a semantic characterisation as primary information bearing units or PIBUs (Croft 2001: 258). Croft adopts Langacker’s account of relationships between heads and dependents in terms of semantic valence and in terms of instantiation (Fillmore and Kay 1993), which is a property of constructions that links semantic components to their syntactic counterparts or elements.
Symbolic relations
Thirdly, the form and the meaning of a construction are linked in RCG by symbolic relations, in Langacker’s sense of the term. In other words, each construction as a whole is a form-meaning pairing in the same way that each lexical item is a form-meaning pairing in the conventional view of the lexicon. As we have seen, this is a defining feature of constructional approaches (including Langacker’s Cognitive Grammar approach).
Functional prototypes
Croft’s fourth point relates to how RCG describes typological generalisation and variation. In Croft’s model, both are characterised in terms of categorisation and in terms of how function is linguistically encoded. In other words, cross linguistic similarities and differences are described in terms of functional typo logical prototypes: while referring expressions relate to OBJECTS, attributive expressions relate to PROPERTIES and predicative constructions relate to ACTIONS (Croft 2001: 87). Of course, OBJECTS, PROPERTIES and ACTIONS are semantic or conceptual categories, and these prototypes underlie the parts of speech in the world’s languages. However, RCG does not specify the boundaries of these categories, which may vary from one language to another (Croft 2001: 103).
Explaining linguistic universals
Finally, RCG explains linguistic universals (linguistic generalisations, in Croft’s terms), not by assuming of a set of universal grammatical primitives, but by assuming a universal conceptual space. In this respect, the RCG approach, which inherits much from functional typology (Croft 2003), reflects one of the core assumptions of cognitive approaches to grammar: cross-linguistic patterns of grammatical structure, such that they exist, are motivated by meaning, which in turn emerges from conceptual structure.
As we mentioned in Chapter 3, many typologists adopt some version of a semantic map model. A semantic map is a language-specific typological pattern which rests upon a universal conceptual space or system of knowledge. Croft defines conceptual space as follows:
Conceptual space represents a universal structure of conceptual knowledge for communication in human beings. (Croft 2001: 105)
The categories defined by constructions in human languages may vary from one language to the next, but they are mapped onto a common conceptual space, which represents a common cognitive heritage, indeed the geography of the human mind. (Croft 2003: 139)
To take a concrete example, recall our discussion of case-marking systems from Chapter 17, where the subject of a transitive verb is labelled A (for AGENT), the object of a transitive verb is labelled O (for object) and the subject of an intransitive verb is labelled S (for subject). We saw that a case system need only distinguish A and O (the subject and object of a transitive sentence), since S and A cannot co-occur (a sentence cannot simultaneously be transitive and intransitive), and S and O do not co-occur (an intransitive sentence does not have an object). The conceptual space that represents these participants is rep resented in Figure 20.8. This diagram represents the universal conceptual space that underlies language-specific patterns for marking these participants morphologically.
We saw in Chapter 17 that if a language marks S and A in the same way but marks O differently, this is a nominative/accusative case system (for example, German).In contrast, if a language marks the intransitive subject S and the object O in the same way (absolutive) but marks the transitive subject A differently (ergative), this is an ergative/absolutive system (for example, Basque). The semantic maps for these two systems are represented in Figure 20.9.
Although this overview of RCG is necessarily brief, it should be clear why Croft describes his model as radical. This approach questions basic assumptions that have defined theoretical and descriptive linguistics throughout the history of the discipline, such as the existence of word classes and grammatical functions. From this perspective, what many linguists think of as the building blocks of language (its grammatical units) are epiphenomenal. In the place of these cross-linguistic universals, the RCG model emphasises the universality of the conceptual system and explains typological patterns on this basis.
