informish

informish

Experiments with implementing inform7 style syntax with clojure and core.logic. My intent is to learn more about declarative logic, constraints, predicates, and natural language.

Reading list

Natural Language, Semantic Analysis and Interactive Fiction (Graham Nelson, 2005-06) http://inform7.com/learn/documents/WhitePaper.pdf

REPRESENTING GAME DIALOGUE AS EXPRESSIONS IN FIRST-ORDER LOGIC - Kaylen Wheeler http://ir.lib.uwo.ca/cgi/viewcontent.cgi?article=2646&context=etd

http://homepages.inf.ed.ac.uk/wadler/papers/propositions-as-types/propositions-as-types.pdf

Predicate dispatching: A unified theory of dispatch http://homes.cs.washington.edu/~mernst/pubs/dispatching-ecoop98-abstract.html

David Nolen, Pattern Matching & Predicate Dispatch http://vimeo.com/27860102

Natural Language Processing in Lisp: An Introduction to Computational Linguistics http://www.amazon.com/gp/product/0201178257/

Chris Martens's papers http://www.cs.cmu.edu/~cmartens/ http://www.cs.cmu.edu/~cmartens/njpls-may2014/njpls.pdf

logic proxies

I'd like to track logical values with a proxy, at least until I understand core.logic types better.

A cvar type tracks the constraints, dependencies, and relation facts that have been stated. Solving a cvar returns a dependent cvar. Solving can query logic if it has a finite solution domain (and filter with the predicate tests).

Some solving queries can be answered by the predicates alone, during a prepass phase where common constraint patterns have obvious answers.

(solve count (cvar))
;#<CVar #13number?(count #14)>
;attempting to solve count resulted in a dependent cvar constrained to a number.

(solve count (c- (cvar) empty?))
;0
;the empty? constraint provided a solution.

;the to-string of cvars shows a #uid, constraints, and the dependent inputs.

definite clause grammar (DCG)

DCG parses can find valid expressions from a sentance.

The solutions are converted to code containing literals and unsolved logic variables.

(parse '[:name of player])
;{(solve :name player) "john"}

(parse '[:name of first :siblings of thing])
;(solve :name (solve first (solve :siblings thing)))
;#<CVar #15(:name #16(first #17(:siblings #18)))>

constraint dependencies

When a logic query is used as an input for the value of a definition, it becomes a derivative. The dependencies are tracked.

Persistance

Cvars use unique identifiers.

constraints

Constraints are sets of predicates, when a logic variable has a finite solution space, the constraints filter.

'[The name of the culprit has a count of 5.]

;His name has no relations to query, so it's just an abstract with a #(= (count %) 5) constraint

'[The culprit's name is an entry in the list of the names of streets]

;Now his name is a constraint on a finite domain, and could be solved and converted to a literal.

constraints as types

Inform7 uses single inheritence for types. A sign is a kind of thing with a string called message. The message of a sign is usually "nothing".

Use a cvar for a type. Any cvars that have type relations would be a merge of cvars in order of relation precedence.

A safe is a kind of box with a number called combination.

Assets are a type of furniture, assets have a number called value.

A safe is a type of box, safes have a number called combination {:combination (cvar number?)}

A vault is a kind of safe, and a kind of asset. The combination of a vault is a number
above 100. {:value (cvar number?) :combination (cvar number? #(> (count %) 100))}