The uniform tabular algorithm and the item sharing method have been fully implemented in Common Lisp and tested with several small constraint-based grammars. The incremental monitoring and revision strategy as well as the method for generation of paraphrases have also been fully implemented as straightforward extensions of the uniform tabular algorithm.
The overall structure of the model's architecture can graphically be represented as follows:
Figure 1.1: General control flow of the new model
To give a first flavor how the model works, a brief description of the flow of control follows: We assume that the conceptual system (see section 5.1 for more details) communicates via semantic expressions with the uniform algorithm. The uniform algorithm consists of a control logic (the inference rules) and an agenda mechanism. This agenda mechanism maintains two different agendas - one for generation and one for parsing. New items are first stored in one of the agendas according to a given priority (depending on whether parsing or generation is performed). Computed items are stored in different item sets, where it is assumed that passive items (which correspond to completely successfully processed substructures) are shared during generation and parsing. The reason why only shared passive items are considered will be given in section 4.15. Monitoring, revision, and paraphrasing is performed by means of specific methods. These methods are triggered (or activated) by the conceptual system but are interleaved with normal processing which is indicated by the dotted arrow.