[Corpora-List] PS:minimal changes in a paragraph (based on a corpus it appeared) ... (2nd attempt (after first one was deleted))

[chris brew]

2011/8/9 Albretch Mueller <lbrtchx at gmail.com>

> > Oberlander, J. and Brew, C. [2000] Stochastic text generation.
> Philosophical
> > Transactions of the Royal Society of London, Series A, 358, 1373--1385.
> ~
>  thank you I gave a quick read to that article
> ~
> > ... goals are stated as macroscopic properties of the text
> ~
>  these goals to me would be like unique contextual pointers describing
> the text, which should be naturally describing properties of the texts
> themselves and the corpora they relate to. So, iterating through all
> the options in the corpora in an order fashion and issues relating to
> "similarity" (and therefore NLG) should be natural and directly
> relating to the texts' corpora, even if we talk about poetry
>

Yes. In theory this is correct. IF we have a good way of automatically
identifying whatever it is that poetry is, and we can work out a way of
efficiently exploring all the options that are available, and the options
that we explore give us sufficient freedom THEN we have a practical solution
to the NLG problem. The main point of the paper that Jon Oberlander and I
wrote was to draw attention to this theoretical possibility. A secondary
point was to invite consideration of the idea that there might also be
practical consequences, ones that would allow a richer
view of NLG than is usually entertained. That said, it is important to
realize that none of the three premises of the IF...THEN ... statement is
yet true.

- for premise 1, while Dmitri Genzel, Jacob Uszkoreit and Franz Och had a
nice paper in EMNLP last year on poetic translation (
http://static.googleusercontent.com/external_content/untrusted_dlcp/research.google.com/en/us/pubs/archive/36745.pdf),
the part of poetry that they address is just metre and rhyme, not content
and beauty.
- for premise 2, it is one thing to be able to articulate a procedure that
systematically explores a set of options, quite another to be able to do so
efficiently
- for premise 3, we don't know how best to formulate the options anyway.

So, in theory yes, in practice maybe one day, in a limited kind of way.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/corpora/attachments/20110809/0c41a0aa/attachment.htm>
-------------- next part --------------
_______________________________________________
UNSUBSCRIBE from this page: http://mailman.uib.no/options/corpora
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora