The uniform algorithm is a generalization of the original Early deduction method presented in [Pereira and Warren1983] since it can be used for parsing and generation in both a uniform but task-oriented manner. The basic flow of control is a mixed top-down and bottom-up strategy. Prediction is performed top-down by the inference rule prediction. Passive-completion performs a bottom-up completion, where active-completion as well as scanning could be viewed as a kind of top-down completion.
The processing order of the elements of the body of a clause is partially data-driven. Therefore, the grammar itself is an important source of control, since it is the place which defines how a string or semantics expression is to be composed. The selection function we are using is able to dynamically make use of this knowledge. Thus, it depends on the grammar whether the selection is performed in a leftmost or head-driven way or whatever. Moreover, if the grammar also expresses a kind of left-to-right decomposition of semantic information, our algorithm would be leftmost also in the case of generation. On the other hand, if the phonological information has a head-oriented representation, our algorithm would be able to ``simulate'' a head-corner parser. Thus, although our algorithm is oriented towards efficiency it can handle a wide range of grammar theories.