> it has been proposed (to try) to design machines that we could instruct in our native tongues. this would, admittedly, make the machines much more complicated

> it has been proposed (to try) to design machines that we could instruct in our native tongues. this would, admittedly, make the machines much more complicated, but, it was argued, by letting the machine carry a larger share of the burden, life would become easier for us. > I suspect that machines to be programmed in our native tongues - be it Dutch, English, American, French, German, or Swahili - are as damned difficult to make as they would be to use. Seems to me there's a intermediate step here that's worth mentioning: formal specification languages, such as Z notation and Event-B, which let us precisely describe our problem. [0] If you want a 'compiler' to be able to deal with a natural language request like Give me a C function that sorts an array of uint32_t, of arbitrary length, in place, you have two challenges: 1. Automatically map the natural language problem-description to a precise formal specification of the problem, in a language like Event-B. 2. Automatically map the formal spec to an efficient and correct implementation in C. (This is called refinement. For an example, see [1].) Formal specification languages are rarely used in today's software engineering, but it would make sense to use one in an automated system, to separate out these two very different challenges. (Needless to say, automating either of these stages is enormously challenging, and we're nowhere near accomplishing either of them.) Ideally, stage 1 would warn or error about the ambiguity in the example request: most likely the user wants an ascending ordering, but this should be made explicit in the request. Would such a system be useful? I think the answer is yes regardless of what precisely is meant. I don't share Dijkstra's concerns. We already know it's possible for humans to do effective software engineering, and the question here is really one of what an automated system would have to do to compete with that. Managers already make software happen using natural language! I don't think a comparison to attempts at spoken-word mathematics are all that relevant. A computer simulation of a skilled software engineer, including being a good conversationalist with requirements-engineering skills and the ability to absorb domain knowledge, would be useful by definition, and would put software engineers out of a job at a stroke. Would it be useful to have the weaker system Dijkstra describes, that 'compiles' a natural language specification? Seems clear enough that it would, as much human effort is currently spent doing this. Assuming the system works well, and isn't unworkably pedantic or obtuse, it would still be extremely valuable. > As a result of the educational trend away from intellectual discipline, the last decades have shown in the Western world a sharp decline of people's mastery of their own language Is this actually true, or is this one of those back in my day children were respectful things? [0] https://en.wikipedia.org/wiki/Specification_language https://en.wikipedia.org/wiki/Specification_language [1] See chapter 7 of this large PDF: https://eprints.soton.ac.uk/422225/1/Thesis.pdf https://eprints.soton.ac.uk/422225/1/Thesis.pdf