The problem is that you can't arrive at such a thing by ordinary exploratory coding without a strong conceptual framework guiding what you're doing.

Indeed, Winograd wrote a short but serious book explaining the linguistic and other theories behind shrdlu.

However, I don't think even a complete understanding of the points of his book would be sufficient to understand shrdlu's code.

> I don't think even a complete understanding of the points of his book would be sufficient to understand shrdlu's code

That sounds like a point against the use of Lisp. There's certainly quite a lot of handwritten case-specific parsing code:

https://github.com/stuartpb/shrdlu/blob/master/gramar

Perhaps the bit which looks most lispish is the dictionary: https://github.com/stuartpb/shrdlu/blob/master/dictio

(It also looks like this was written on one of those ancient systems that didn't support lowercase and only allowed 6-character filenames)

> (It also looks like this was written on one of those ancient systems that didn't support lowercase and only allowed 6-character filenames)

A limitation that also led to 'Schemer' dropping the R.

Could you tell me which of his publications you refer to here?