The language subsystem will be defined in a future paper in detail due to its complexity and wealth of phenomena to be explained. Here we will only provide a sketch of it's functionality.
"Words" in the language subsystem are treated as "labels" which denote specific entities in the SBR system:
- "Nouns" are labels denoting sets of objects with a "type" attribute identical to the noun or of a sub-type. Sub-type relationships are stored as hypothesis from the "scientific reasoning subsystem".
- "Adjectives" are labels denoting specific attribute values of objects.
- "Verbs" are labels denoting Active Tasks (sub-plans or tasks with task decompositions).
- "Prepositions" are labels denoting "relations" be-tween objects.
- "Conjunctions" are labels denoting "relations" be-tween plans.
Important operations in the language subsystem include the conversion of language to and from scenes and scripts. Similar operations have already been implemented with certain success as part of natural language processing research:
- "Comprehension": Language -> Script:
Transform a piece of language into a "minimal" scene or script, using scripts and abstraction from the training material.
- "Generation": Script -> Language:
Create a language representation of the script that will induce on the listener‘s side a "similar" script as the base script. We envision an iterative algorithm for this task that would use the Language -> Script operation above in order to create scenes supposedly produced on the listener‘s side. The algorithm would then iteratively modify and refine the language used in order to minimize the distance delta between the intended scenes and the scenes produced on the listener side.