Maybe latin is better suited to do such a thing.
Check out this site:
http://www.csse.monash.edu.au/~damian/papers/HTML/Perligata.html

A natural language is too much efort for all the above reasons, at this time. However another related idea would be to produce an IDR for an existing language that actualy helps to debug your code rather than just telling you the contents of variables and which line broke in execution. How you do this might take a lot of thought and experimentation as well as discussion.
One point on the natural language. I would forget parsing ALL the words and just pick salient ones from a sentence, if you don't already and word order of cause. this will expand the number of ways to write the same sentence.
Make a = 21
Make a varable equivilent to 21
Assign 21 to a variable called a

Also a vay to go is to use a tree constructto find what is being wanted in a sentence again with the salient words that brance and brance till an action is arrived at.
Syntax and gammer checking will be very important so a word out of context can be caught early in the sentance construction. I like the method used in Test automation stratagies with a high degree of abstraction, a keyword is selected from a list and this then populates a new list of possible 'next' words based on the growing context etc etc.

Good luck.
You might try Python as it has a clean grammer compared to Perl.