Wednesday, February 15, 2017

The problem of dialogue as regards chatbots

Me: "So here's my plan."

Clare: "OK."

Me: "I've copied Peter Norvig's chapter 11 code for a Prolog interpreter in Lisp into a Notepad file. Unfortunately copying from PDF introduces all kinds of formatting errors, so it takes a while to get it into a compilable state."

Clare: "?"

Me: "I need to work through his main functions - principally to exactly understand his unification algorithm to start with. I may modify the code a bit, make it more transparent."

Clare: "??"

Me (oblivious): "And then I'm not sure I like the way he's storing clauses as plists on the clause head predicate. I think an assoc list would be clearer, if less efficient. Also, I need to do a few paper and pen examples to understand exactly the rationale for variable-renaming between clauses during resolution."

Clare: "???"

Me: "I'll get his standard Prolog depth-first search to work first of course. But I think it would be a good idea to explicitly break out the resolution procedure as a standalone function, and then parameterise the proof procedure by a range of possible search methods including breadth-first and best-first."

Clare: "????"

Me: "Of course, all this is just to get a series of tools in place for my speech-act planner, the one that's going to make my chatbot able to steer a conversation."

Clare: "?????"

Me: "I really think this could be a stellar app. I can see it on Google Play. The trouble is, 'in the knowledge lies the power' is just as relevant here and I might need to spend way too much time just data-filling various knowledge-bases, ... hmm, unless I can get it to safely learn from users. That's where Google does have an advantage ..."

[Pause]

Clare: "Is that Mr Bullfinch I see on the fat block?"

No comments:

Post a Comment

Comments are moderated. Keep it polite and no gratuitous links to your business website - we're not a billboard here.