[11 December 2009]
One of my most pleasant memories is the discovery, some years ago, that while it is often rather boring to copy a program out of a book or off the Web and modify it to suit the needs at hand, it’s a lot more interesting if at the same time you can translate it from one programming language into another.
I discovered this when I needed a sort routine in Spitbol once; one of the Spitbol implementations we were using had a built-in sort, but the other didn’t. I copied a Shell sort, or possibly a Quicksort, from a Pascal or C textbook, and found that the task of translating the algorithm from one language into a different language with rather different control and data structures was much more rewarding and interesting than it had ever been to copy a Pascal or C program from a book into a file on my disk, compile it, and ‘play’ with it. It was non-trivial enough to give me a small feeling of accomplishment, and easy enough (the algorithm was right in front of me, after all, in executable form) not to cause serious troubles.
I suppose that most textbooks have no choice: they have to show you programs in the language they are teaching, and they can’t really assume the reader knows some other language. (When I started to learn Middle High German, the professor asked the participants in the class who had learned Greek in school, so he could know whether citing Greek parallels would be helpful. He remarked with a sigh that it was years since it had made any sense to ask who in the class had Hebrew.) But I learned a lot more about both Spitbol and the sorting algorithm in question when I did that translation than I ever had before.
This topic came to mind this morning because in my continuing work with Alloy and ACL2 I have been trying to rewrite a simple ACL2 example (a function, a few examples, and a couple simple theorems to prove that the function meets its requirements) into a roughly corresponding Alloy model. I think the exercise illuminates both Alloy and ACL2 (more on that later, after I’ve gotten this example to work and maybe done a few more, in both directions).
Similarly, I have learned a lot about Steve Pepper’s Italian Opera topic map by thinking about what a translation into other forms (SQL, Prolog, …) would look like; I expect to learn more about the topic map and about the technologies involved, when I push those translations further.
It’s funny, though: I am not sure I’ve ever seen anyone say explicitly that translating short programs from one language to another can be an interesting and rewarding experience for those learning the languages involved. Does no one else find it as helpful as I do? (Am I insufficiently lazy as a programmer?) Or is it too obvious to merit mention? Or is there just not a good single term for the practice?
Makes me wonder whether the double-translation method (by which Roger Ascham taught Greek to Elizabeth I) could be applied to programming and markup languages?
Just remember, translations are like sandpiles (and floating-point values): every time you move them, you get a little less sand and a little more dirt.
I think I remember a discussion something like from the heated time when OfficeOpen was coming up for approval.
Point: Look all the defect reports coming up against the (already approved) Open Document standard.
Reply: Yeah, but that standard has gone out for translation. Translators always raise a blizzard of requests for clarification.