On Thu, Nov 15, 2012 at 8:21 PM, Alexander Mikhailov <[email protected]> wrote: > http://www.jsoftware.com/jwiki/AlexMikhailov/Parsing > > There are multiple questions regarding J style. For example, it seems like it > would be awfully inefficient to try to solve parsing problem in parallel - > avoiding the loop. On the other hand, in highly parallel environment with > grammars good enough (without errors and ambiguities) it could be interesting.
Note that Context Free Grammars tend to be ambiguous, in traditional programming environments we typically want a Parsing Expression Grammar. And we probably need "cloud sized resources" to adequately tackle natural language parsing. Note also that J [the language] has no problem with serial processing, it's J [the implementation] which has a performance bias. These are related issues -- we could imagine a transforming operation in J (analogous to f.) which lifts unnecessary conditionals out of loops, to make serial processing more efficient. [And we could also imagine a transforming operation for the purpose of deploying something written in J into a GPU, or whatever.] We do not currently have that, but it's doable -- the trick to getting started would be to start small. > Another is how to accumulate the output into a tree. Recent discussions > showed that there definitely many ways to represent trees... There are also many different kinds of trees. -- Raul ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
