Most theories of generative grammar assume a componential model of grammatical knowledge. On such a model, each type of linguistic knowledge (syntax, semantics, and so on) makes up a separate component with general linking rules mapping information between components. Various assumptions lie behind this model of language structure. First is the notion that syntax is autonomous, operating independently from other components of linguistic knowledge. From this, it follows that syntax can be studied independently from semantics and other aspects of meaning. Second is the idea that complex structures are built out of discrete, atomic primitives in accordance with the combinatorial rules specific to each component. This reductionist perspective lends itself to the stronger hypothesis that constructions (more complex grammatical structures) are purely epiphenomenal; all of their properties can (or rather should) be "captured by the general rules of the grammatical components and their interfaces" (Croft & Cruse 2004: 228). This leads us to the third assumption: that grammatical knowledge is organized to favour economy. In other words, a speaker's grammar is a non-redundant system which contains only the smallest set of rules required to generate all the configurations of a given language. This is especially true of the most recent variant of generative grammar, the minimalist program, which originated in the early 1990s (see Chomsky 1993, 1995).
In contrast, theories of construction grammar assume that a speaker's grammatical knowledge is made up of constructions, defined as symbolic form-meaning pairings (or signs). These constructions form a structured inventory, represented as a hierarchical network. Lower-level constructions (more specialized linguistic patterns) inherit attributes from higher-level (more general, schematic) constructions. In a default inheritance model, the speaker's grammar is a redundant system where information is stored not only in the most schematic or highest-level constructions, but at all subsequent levels in the hierarchy. Again, this model of grammatical knowledge brings with it certain theoretical assumptions. For example, while, in componential models, separate components of the grammar interrelate via linking rules, in construction grammar, the form-function mapping is internal to the construction. From this, it follows that syntax cannot be studied in isolation and all aspects of meaning are required to give a full account of grammatical knowledge (see Croft 2001). In addition, this model of language structure anticipates that some constructions have properties which are not predictable from highly general patterns of correspondence; a construction can contain construction-specific form-meaning pairs and may even encode meaning which cannot be attributed to its component parts. Furthermore, while the generativist concern is to represent or create a maximally economical system, the constructionist concern is for reducing demands on processing (see Goldberg 1995: 74). A redundant system, in which information is stored in several places and can be easily located, is therefore regarded as the most efficient.
-  While most versions of construction grammar invoke default inheritance, the Construction Grammar of Kay & Fillmore (1999) is a complete inheritance model, in which information is represented non-redundantly at only the highest possible level in the constructional taxonomy. As Goldberg (2006: 214) notes, this version of construction grammar has developed somewhat separately, in that it more closely resembles the formalist theory HPSG (which is also sign-based). Unlike the others, it is not a usage-based theory, making no claims as to the psychological reality of a complete inheritance model (see Croft 2007a: 494).