Transformations

In early generative models, constructions in deep structure can be transformed in surface structure. For example, deep structure active sentences become passive in surface structure via the passivization transformation.

For all its theoretical attractions, a major problem with the transformational component of the model was that it was largely unconstrained. Transformations could not introduce new meaning-bearing elements, but they could move constituents (for example both NP2 and NP1 move in the passivization transformation), add elements (by) or, on occasions, delete them. In other words, they could do practically anything, which sat awkwardly in the context of a research programme aiming to capture the universal principles of grammar acquisition, which are purportedly simple and restricted in number. Later models have therefore set out to specify the constraints on transformations.

Transformations initially gave way to movement rules, and the label ‘Transformational-generative grammar’ (or ‘TG’) of the 1970s had become simply ‘Generative grammar’ by the 1980s. Deep and surface structure were renamed D- and S-structure respectively, and generativists talked of i-language (‘internal language’) and e-language (‘external language’) rather than competence and performance. Rules specifying grammaticality in individual constructions gave way to principles, which set out conditions on grammaticality applicable to all constructions in human language, and parameters which constrain their application according to their setting in a given language. As an example, the generalization that all phrases have a head is a principle, while the head parameter or head-directionality parameter determines the position of that head with respect to complements within its phrase.