NER: Conditions

[TheCurle]

The Named Entity Recognition system works like thus:

Rules are registered via NER#addRule - these contain ExtractionRules, which eventually contain Extractors.
Extractors are functional interfaces, which define a single function - take in the full String, return a list of recognized entities.

Extractors like Between, After and Before work with these systems called Conditions.
Currently, because there are only three extractors that use them, Conditions are hardcoded to contain two Strings - from and to.

These hardcoded strings are shared between all current relevant Extractors, albeit the current implementation means that these Conditions must be queried from global state - until a full Query system is devised.

However, the issue at hand is about extending this system. Conditions can be used for a lot of things related to entity extraction - the Neural Engine supports things like categorization, so you could have a condition that searches for the most likely match for a generic topic, or a category of object, like Condition(category: "fruit").

This would be much better for design and construction than the current hardcoded / regex Extractions.