The System Architecture of our YAG implementation is shown below.

The Core YAG

The Knowledge Representation Realizer  (surface-2) realizes a knowledge representation into an appropriate feature structure. Its input contains two parts: a semantic network that represents content, and a set of control features that provides supporting information
and optional syntactic constraints. Some of these control features are used by YAG to select the appropriate template, the remainder are used to select options within a template. 

The Feature Structure Realizer (surface-1) realizes a feature structure into a surface text. This feature structure specifies the template to be used along with other features and their values. In addition, this layer will use defaults to specify any missing values in a feature structure input. 

The Template Manager stores a template into a specified library, and retrieves template structures for the Knowledge Representation Realizer and the Feature Structure Realizer. If a template has multiple names, these names (template aliases) will be maintained by the
Template Manager. 
  

The Knowledge Base

There are two types of knowledge bases in YAG; domain dependent and domain independent. The domain dependent knowledge base includes the Mapping Table and Domain Template Libraries. The domain independent knowledge base includes the Syntactic
Template Library and the Lexicon. 

The Mapping Table is used to map a knowledge representation to its associated template. The Knowledge Representation Realizer will access a mapping table with selected control features to pick to an appropriate template when realizing a text from a knowledge representation. 

The Domain Template Libraries contain templates that are specific to a particular application. Developers can author their own templates when necessary by manually editing a text file that contains templates in any text editors. However, YAG also comes with TATTOO (The Template Authoring and Testing Tool) which is a program that allows a knowledge engineer to author and test templates in a graphical environment. 

A template can reuse existing templates (including ones in the Syntactic Template Library). These templates can serve applications that have special information presentation needs, such as those in scientific information and weather reports. 

The Syntactic Template Library contains templates that are used as a grammar of English, such as clause, noun-phrase, and pronoun template. Other templates can embed these templates to form more complex structures. They can be combined with templates from the Domain Template Library. 

The Lexicon contains word level information. Templates can access the lexicon directly with a template rule. YAG includes lexical functions to inflect a given verb according to verb features (e.g. tense, person, and aspect), and to generate the singular or plural form of a noun. Additional lexicon functions can be coded if required.