Minimalist approach to chunks

I’ve written up an analysis of a minimalist design for chunks as a baseline to compare with when it comes to evaluating additional features for the chunks rule language. A minimalist design is much more plausible in respect to a rule engine implemented as a pulsed neural network, and would limit chunk property values to just names, i.e. dispensing with other types such as numbers, booleans, dates, strings and lists. 

All of these can be instead modelled as chunks:
Lists as a sequence of chunks with a property for the list item and another property for the successor chunk
Booleans as chunks that denote true or false
Numbers as a sequence of chunks where each chunk has a single digit and a successor chunk
String literals as a similar sequence of characters
Dates as a chunk with properties for the year, month, day, etc. which in turn are modelled as for numbers
My analysis suggests that a minimalist design would need a means to indicate a) when a condition property must not match the target chunk’s property value, and b) that a given condition chunk must not match the target chunk.

I also indicate how operations over sets of matching chunks can be implemented, taking into account the constraint that module buffers can only hold a single chunk.

A minimalist approach would complicate manual development of declarative and procedural knowledge, but may actually be better in respect to machine learning! That is something to be investigated over the next few months.

For more details, see: https://github.com/w3c/cogai/blob/master/minimalist.md <https://github.com/w3c/cogai/blob/master/minimalist.md>
Dave Raggett <dsr@w3.org> http://www.w3.org/People/Raggett
W3C Data Activity Lead & W3C champion for the Web of things