In generative theories, grammatical knowledge is innately programmed and so all speakers begin life with the same capacity for language, otherwise known as Universal Grammar (UG). According to the principles and parameters framework, the acquisition of a particular language involves a process of parameter setting. Based upon the input received, the parameters of UG are set to reflect the rules of the language to which the child is exposed. Language learning is therefore a relatively simple process: "Experience is necessary only to fix the values of the parameters" (Roberts 2007: 21). In essence then, the end of language acquisition signals the end of (the potential for) variation in UG. From this, it follows that language change must be something that happens between generations, during the period of acquisition. On this model, once the speaker has acquired their language (through the setting of parameters) their grammar is no longer susceptible to change. However, what may change is their linguistic output. Consequently, subsequent generations of language learners may be exposed to a different input. On the basis of this input, these children may (through errors of abduction) acquire a different grammar, resulting in language change.
Given these theoretical assumptions, language change on the generative model has been characterized as both abrupt and random (see e.g. Lightfoot 1979, 1991, 1999). The abruptness of change follows as a direct consequence of the discreteness of parameters, since "a given parameter changes its value suddenly and irrevocably" (Roberts 2007: 292). Likewise, the image of change as a random 'walk' around the variation space defined by these parameters follows from the claim that change is confined to the acquisition process, with grammars constructed anew by the individuals of each generation (see Roberts 1993: 252). On this view, change is not directional, nor does it continue across generations of speakers. The generative approach therefore focuses more on the actuation of change (the initial reanalysis), rather than on its spread (across the grammar and across speakers). Recently, however, attempts have been made to incorporate the characteristics of directionality and gradualness into this theory of language change (see Roberts and Roussou 2003; Roberts 2010; van Gelderen 2004, 2010, 2011). On this view, reanalysis is still abrupt, but it is no longer associated with large catastrophic changes, or 'salutations. Instead, change occurs through a series of discrete micro-reanalyzes (involving micro-parameters), which create an apparent gradualness. Directionality is accounted for on the suggestion that some parameters have marked and unmarked values, based on a preference for simplicity or economy. For Roberts and Roussou (2003), this creates 'basins of attraction' within the parameter space, towards which grammars have a tendency to move.
In contrast, most versions of construction grammar are usage-based theories. On this model, humans are not innately programmed with grammatical knowledge; instead, all aspects of language are learned from the input (or rather, from the speaker's linguistic experiences). Both language learning and language change involve the speaker inductively generalizing over instances to form mental schemas (or constructions) which are represented in the language system. On a usage-based model then, constructions are simply conventionalized chunks of linguistic knowledge (Goldberg 2006). From this, it follows that the storage and organization of grammatical knowledge is dependent upon, and can change according to, patterns of use (Bybee 1985, 1995, 2006). Token frequency (the number of times a given instance is activated) results in the entrenchment of the instance as a unit in the language system. In contrast, type frequency (the activation of different types of instance) results in the entrenchment of a more basic schema; that is, an abstraction which stipulates characteristics shared by the different types of instance. In this way, changes to the language system are the result of general cognitive processes, such as categorization. This means that there is no need to confine language change to the acquisition process on this model; instead, a speaker's grammar can change throughout their lifetime.
Given these theoretical assumptions, language change is characterized as gradual and continual. As Goldberg (2006: 62) observes, "...we constantly parcel out meaning, form abstractions, and generalize over the instances we hear". From this, it follows that change can also be directional. For instance, we have seen that type frequency results in the entrenchment of a more schematic construction. This can, in turn, have repercussions for yet higher-order constructions in the taxonomy, as existing schemas become more abstract in order to accommodate (or sanction) the new lower-level constructions. This type of constructional change, known as schematization (or grammatical constructionalization), therefore proceeds upwards throughout the hierarchy, leading to the creation of new constructions or the reconfiguration of existing ones (Trousdale 2008: 55). As Traugott and Trousdale (2010a: 14) note, "constructionalization shifts the perspective [away from a focus purely on reanalysis] to incorporate more substantially analogy [including extension] and pattern alignment"
This cursory description highlights some of the key differences between generative and constructional models of language structure and language change. In doing so, it simplifies (and perhaps polarizes) the two approaches. Certainly, there are areas of overlap among these theories, as noted above in relation to the current analyses of parametric change as accounting for gradualness and directionality. As Roberts (2010: 70) notes, on this issue at least, "we are close to a convergence of views arising originally from quite distinct traditions of grammatical and historical analysis". Nevertheless, since this chapter compares a current constructional account of it-clefts to one drawing from the generative tradition of the 1990s, the background assumptions under comparison are more obviously at variance. In what follows, I limit myself to comments about generative grammar relevant to the time at which Ball (1991, 1994) was working.