Nell/Prolog

CSA: I did a lot of research on Prolog and Prolog-related topics. The bottom line is that it's probably not directly relevant for our project: what we really need is an incremental planner or a truth maintenance system, not a backtracking solver. Prolog doesn't come with any magic bullets to answer the "now what goals have their preconditions satisfied" question, which is at the core of a UNIVERSE-style storytelling system.

Datalog might still be interesting as a data model for "avatar services".

Interesting Prologs

• Datalog
• Logtalk -- the object oriented features provide a good encapsulation mechanism for a prolog-style language. It also supports "event-driven programming"; I can't tell if that would help one construct a truth maintenance system (ie, track dependencies) or not.
• JScriptLog: implementation of ISO-Standard PROLOG in JavaScript. Seems pretty capable, but the code is not the cleanest -- no attempt at using prototypes or object-orientation. Parser is apparently weak.
• Yield Prolog: an embedding of Prolog control flow into JavaScript using the yield operator. (See https://github.com/cscott/jsshaper for an implementation of yield for non-Firefox browsers.) Contains a compiler to translate 'standard' prolog into yield-using JavaScript (or Python, etc). The parser is original DEC-10 PROLOG and the compiler is also written in prolog, translated into yield-JavaScript, so the implementation must be fairly complete. I'm not 100% convinced it handles the various cut cases correctly. Also, modules in prolog are a mess (see Logtalk for a better solution), although there is some sort of module system implemented.
  • Would be interesting to port Logtalk to run on top of Yield Prolog, although I'm not sure how useful the result would be. Seems like a rather esoteric tower to build software for kids with.

Truth Maintenance / Planners

• STRIPS

Storytelling

• UNIVERSE
  • 1984 paper, xxx, 19xx paper
  • Blog posts comparing it to MINSTREL

Natural Language Processing

Prolog and DCGs (Prolog's version of the Monad) are a pretty good way to tackle some natural language processing tasks.