Open interrogator opened 9 years ago
I would leave NLG aside for the moment. The work in NLG is very interesting and potentially very useful in whatever we do here. It can come in handy only after we have finished and then think how to bring in bits and pieces into what we have decided to model for parsing/annotation purposes. As such, NLG grammars, specifically Nigel Grammar, are computationally infeasible for parsing purposes [Bateman2008] so the will have to undergo anyway a transformation process [O'Donnell 'xx and Kasper 'xx] into a form which suits parsing tasks.
Pretty much my thoughts. I just think that we ought to be mindful that in some ideal world, code used for generation could/should be shared with code used for parsing.
Anyone else? We could close this issue if nobody thinks there's anything we really need from NLG.
I vote for closing the issue. In case someone shows up making a case for using particular pieces of existing code we can react on that. I like the idea of reusing existing solutions, but if none of us are aware of such concrete, re-usable solutions, let's not trail off in that direction.
@costezki @interrogator The Nigel grammar is not unfeasible for parsing. The only issue is that the notion of structure must be rethought from scratch. I have two papers on this topic. One is "Towards a description of symbolic maps" (http://www.aclweb.org/anthology/W14-4411) and the other is the one that is coming out this year after the ISFC 2015. I can share both with you guys if you are interested in this topic. @SigridK I agree with your pragmatical approach.
We have access to a lot of text generation material (code, books, etc.). We haven't yet spoken about whether or not, or which bits, might help with our work. Does anybody have a reference list for SFL text generation? Starting point for the uninitiated:
http://www.isfla.org/Systemics/Software/Generators.html