Assumptions
The central thesis of Goldberg’s theory is that sentence-level constructions ‘themselves carry meaning, independently of the words in the sentence’ (Goldberg 1995: 1). According to this view, constructions are themselves the oretical primitives rather than ‘taxonomic epiphenomena’ (Chomsky 1991: 417), as we saw in the last chapter. Although Goldberg does not deny that word level units contribute a great deal to the meaning and structure of sentences (section 20.1.2), she argues that a purely ‘bottom-up’ or lexically driven model of grammar fails to provide the whole picture.
As Goldberg observes, the issue of argument structure alternations has received a considerable amount of attention in twentieth-century linguistics. We will look in more detail at argument structure alternations in the next section, but for the time being consider the examples in (1) and (2).
As these examples illustrate, the ditransitive verb bring can occur in two different construction types. Examples like (1a) and (2a) are called double object constructions (or dative shift constructions) because the verb is followed by two nominal objects. In examples (1b) and (2b), which we will call the prepositional construction(Goldberg 1995: 8), the indirect object (Lily or the table) is instead represented by a preposition phrase (PP). The point of interest here relates to the fact that while the prepositional construction allows the recipient to be either animate (1b) or inanimate (2b), the double object construction requires that it be animate (compare (1a) with (2a)). The issue that arises from this observation is how these differences are best captured in the model of the grammar. Goldberg argues that the most explanatory account associates these semantic restrictions directly with the grammatical construction itself, rather than stating the information in the lexical entries of individual verbs.
Before proceeding with the discussion of Goldberg’s theory, it is important to point out that her definition of a construction differs somewhat from the definition assumed by Langacker in his theory of Cognitive Grammar. Recall that Langacker defines a construction as any unit with a complex symbolic structure (a complex word, a phrase consisting of more than a single free morpheme or a sentence). Compare Goldberg’s definition:
In this definition, F stands for ‘form’ and S stands for ‘semantics’, so that represents a symbolic unit. The subscripts represent the symbolic link between form and meaning. Crucially, this definition of construction hinges on the issue of predictability, which in turn is related to compositionality, but in a different way from Langacker’s definition. If any aspect of either the form or the meaning of a unit cannot be shown to be predictable from the properties of its component parts, then it has the status of a construction in Goldberg’s model. It follows that both bound morphemes (like plural-s) and free morphemes (simplex words like cat) are constructions in Goldberg’s theory, while they do not have construction status in Langacker’s theory. For Goldberg, neither the form nor the meaning of a morpheme is predictable from its component parts, since it lacks compositional structure. It also follows from Goldberg’s definition of a construction that a complex word, phrase or sentence (which are all constructions in Langacker’s theory), will only count as a construction in Goldberg’s model if some aspect of its form or meaning cannot be predicted from its subparts.
Given that the central status of constructions blurs the boundaries between lexicon and syntax, Goldberg, like other cognitive linguists, assumes the lexicon-grammar continuum. Because Goldberg makes no distinction between simplex and complex symbolic units (since either kind may count as a construction) she refers to the lexicon-grammar continuum as the constructicon (the repository of constructions). Goldberg (1995: 5) also assumes that knowledge of language is represented as a ‘highly structured lattice of interrelated information’. This view is consonant with Langacker’s description of knowledge in terms of a structured inventory. Furthermore, Goldberg (1995: 5) assumes that ‘knowledge of language is knowledge’. In other words, in keeping with the Cognitive Commitment, she rejects the idea that knowledge of language is separate and distinct in nature from other kinds of knowledge and experience. Instead, like other cognitive linguists, Goldberg argues that the properties of language directly reflect human experience, conceptual organisation and construal. Finally, as we have already mentioned, Goldberg’s theory in part rests upon the theory of Construction Grammar that we explored in the last chapter. As we saw there, Construction Grammar is a monostratal generative model. While Goldberg’s model can also be described as monostratal in the sense that it does not involve transformations, it cannot be described as a generative model because it assumes the usage-based thesis. In these core respects, then, Goldberg’s construction grammar is a cognitive approach to grammar.
