Lesson_14
What immediately comes to mind when we say "NLP" is a machine which can converse with a human and do so with human proficiency. "Mr. Data" from the Star Trek series is that kind of machine.
How do humans learn SEE (Standard Everyday English)? They do so with contributions from "significant others" like parents and siblings. They do so over a number of years of instruction, starting at age 1-2 years usually. The teaching therefore is very complex and time consuming. We should expect it to be no less demanding if we are teaching a robot like "IC".
The previous machine psychology lesson, /Language, referred to "stimulus freedom" and language. Succintly that means we can say anything. It is the NORMS of society which put reins on stimulus freedom as well as the constraints of formal logic which restrain us from talking nonsense. Both norms and logic entail the use of language in a CONTEXT. How do I respond to "Hi, I'm Jack"? If this is a dark alley of London England and I fear for Jack the Ripper, I will respond one way. If it is Jack, my friendly next door neighbour, I will respond another way.
LANGUAGE SCENARIOS
The NLP problem can be approached by analyzing the context for conversation (including monologues as appropriate) and setting out the various SCENARIOS for such conversation. The scenario of "purchasing stamps at the post office" carries with it a subset of SEE. The scenario of teaching mental arithmetic to your 10 year old child entails another subset of SEE. The number of scenarios in the life of a typical Canadian is limited. Despite "stimulus freedom", the restraints of NORMS and LOGIC limit what can be said without having others look askance at the verbal skills expressed.
While it is not likely I will know all of the subsets of language which can be evoked in a given scenario, I do know at least one. And I can articulate the algorithm or rules for the use of appropriate language in a given scenario. Parents, teachers, clinical psychologists and others continually refer to the subsets of SEE which are or are not appropriate and the rules which make them so.
Thus we could generate a comprehensive SEE rule book for a typical Canadian. It would set out the scenarios likely to be encountered in a typical Canadian life and the subsets of SEE which could be expressed according to well established norms. That is a huge task but it could be done. It is not made impossible because we don't know how to do it. We know how to do it.
IC LANGUAGE SCENARIOS
The language scenarios we want to generate are for a typical desk top pc which is connected to the present network of one billion email addresses worldwide. This pc has been dubbed "IC" for fun. The typical pc today has TITO (Text in/Text out) capabilities and that is how our language scenarios will be expressed initially.
Over time however, we can expect the pc to overcome its handicaps and acquire more peripherals which will bring it closer and closer to human equivalency on the input side of artificial sensation and artificial behaviour.
For example, S. Venkatraman is working on the problem of a general purpose vision system and some day we will have pc's with object recognizing capabilities comparable to those of a human. And at Blue Point Engineering we can purchase robotic arms and robotic legs for our desk top pc affordable to the hobbyist.
Thus over time our language scenarios for IC will expand greatly as it morphs into a humanoid robot and eventually walks off the desk top into society.
URLs for S. Venkatraman and Blue Point are at:
For now let's assume that our robot IC has the usual severe sensory and motor handicaps and only has TITO capabilities when it comes to the language scenarios we are going to develop. After some development however, many users of ICs NLP programming will be tempted to purchase readily available VIVO systems.
SUPERC ALGORITHM
An algorithm is just a rule or set of rules. IC needs rules at two levels:
What are the rules in SEE whereby ANY OF THOSE ONE BILLION USERS referred to above could TEACH, ie PROGRAM the desk top pc? The idea here is that if we can engage millions online in TEACHING (programming) a pc to converse in SEE, the task becomes plauisble, despite its enormity. And we can only engage these people if we make the use of C programming (or programming in another language) AS EASY AS USING GEOCITIES PAGEBUILDER.
Secondly, for those users who do not want to participate in teaching IC, what are the rules whereby they can engage IC in conversation?
IMP PROJECT
So that is our project for Introductory Machine Psychology. Before we go on, does anyone find any flaws in it? (1) We set out the scenarios for IC; (2) We write scripts for IC in SEE which are normative and logical for those scenarios; (3) We invite the participation of one billion users worldwide to TEACH SEE to IC;
(4) To enlist their participation we need a C super-language which was dubbed SUPERC, a language which covers the range of what C programming accomplishes, but is written in SEE; (5) Once SUPERC is developed we need a web site like the Geocities/Pagebuilder web site which allows any online user to logon free of charge and TEACH (program) IC in SUPERC.
MANY HANDS MAKE LIGHT WORK
This "agorithm" for developing NLP requires engaging millions of people worldwide as above who will serve as teachers and parents in a sense for IC. Its end product would be NLP scenario-by-scenario for a desk top pc: NLP which would meet the criterion of human equivalency in machine conversational ability. Are there any logical errors in the algorithm?