I found that researchers and scientifics are complicating themselves trying to make computers learn one language (english) but at the end there are many languages and pronunciations, so AI machines will not learn. The best option that works is to create a new language for humans, easy to understand for computers and phonetically easy to pronunciate and Ontologically ordered. That will be the beguinning, then apply AI concepts, environment, sensors, perceptive motion, hands motion, that way machines will learn from us, at some point in time they will learn to learn for themselves because lack of knowledge of things will be the trigger.

The general concept of learning is in general based on learning by example. Learning one language is learning one example of verbal communication. What you propose is to learn the general high level concepts and rules directly - this could be also possible but it is contrary to the principle of learning by examples and generalization

First, Is there a universally accepted definition of "plain language" that is unambiguous and straightforward? Second, there are those who advocate "plain" versions of different languages, "English" included. So, a plain version of which language? Sascha's observation on adaptation is also well made.

Human languages are for rationalizing - not being either rational, unambiguous, or particularly error-proof. I'm thinking in terms of languages which cannot express either puns or paradoxes.

There is definitely value in defining a precise and unambiguous language, a simplified subset of a spoken language, to be able to provide unambiguous instructions. I don't think that the language needs a unique description for a particular operation - think Perl.

Unambiguous transliteration and syntax would be nice. Perhaps I'm advocating for Esperanto or German rather than English. Or perhaps "Speedtalk", described by Robert Heinlein. (see http://www.zompist.com/kitlong.html#howmany)