Systemic Text Generation in Penman/Nigel

[Mann and Matthiessen 83] describes the Nigel systemic grammar created for use in the Penman text generation system. A systemic text generation system can be divided into three components:

1. systems: Each area of language production (i.e. count, determination, mood, quantification, tense, theme, voice, etc.) is treated by a separate system. Each system makes decisions based on information queries directed toward the ``environment'' (basically a collection of information about the input semantic text to be generated). The systems are intended to give principled, justified descriptions of the choices which the grammar offers.

2. grammar: The grammar is responsible for entering the correct systems based on the semantic input.

3. realizers: Realization operators build grammatical structure, constrain order, and associate features with grammatical functions.

In general, the systems are used to paradigmatically choose grammatical options, the realizers syntagmatically combine these decisions, while the grammar is the overall control mechanism that guides the whole process.

Systemic grammars are an excellent means of describing languages for use in text generation. They provide an explicit framework for discovering and recording the different choices associated with each grammatical feature, along with the reasons behind making one choice over another. On the negative side, interactions between systems are not explicitly represented. Realizers are used to combine the results of two separate systems, but the systems themselves are kept separate from one another.

This project could be used to implement a systemic grammar in a way that could overcome this problem. The ``grammar'' is equivalent to the initial rule instantiator, which could go to various ``systems'' of rules (i.e. one set of rules for planning attitudes might be used and another set for planning tense, etc.) based on the semantic input at each node. The ``realizer'' is simply the result of the structure builder for each rule selected. Furthermore, interactions between systems would be allowed using constraints and preconditions exactly as shown in this project.