Matching templates against a tree - any idea?

[Jon Fernquest]

Fernando Pereira wrote:

>[Matching templates against a tree]...has been the subject 
>of much research. For a recent textbook, check out 
>Franz Baader and Tobia Nipkow's "Term Rewriting and All That", 

Thanks for making the posting on term rewriting systems.

1. Do you think it's possible to make a rewriting system extension 
to an existing language like Python or Perl?

Everything seems to be in ML and ML extensions such as
Isabelle right now. It's so easy to cook up (or integrate 
components into) whole systems in Python 
(or Perl) and since so many people use them in their
day to day work modules have a greater chance of being used.

2. Are these term rewriting systems a broader category 
that would allow someone to implement for example different
linguistic tools?

What about non-roman script text processing?
The composition of the Burmese *orthographic* syllable 
from underlying glyphs cannot be described by a context free
grammar. There are even cross syllable 
dependencies/non-linearities, syllable n sometimes has 
glyphs in syllable n-1...and I assume this holds for all other
Indic scripts.

There are so many problems associated with these scripts
ranging from sorting to word-wrap to what will be chosen 
as an item in the dictionary, it would be nice to have a 
generalized framework in which to think about them.
Do you think rewriting systems (or discrimination nets/tries)
might be a viable approach?

There are so many different logical sets of tokens,
display tokens, ordering/sorting/spelling tokens (a,b,c,d,e..),
phonetic tokens,
transliteration into roman script, 
transliteration of English words into Burmese...
that have to be translated between in any application 
that features Burmese text (also applicable to Indian scripts in general)

Meanwhile, people type fonts into documents. 
For one character/glyph there are about 8 
different keys on the keyboard chosen based on context.
Everyone thinks that unicode is going to solve all their 
problems. Unicode is only an encoding.

I wrote sets of cascaded finite state transducers
to do these translations but maybe term rewriting 
rules are the solution?

Thanks again for the interesting posting.

Jon Fernquest
bayinnaung at hotmail.com
(on the road right now)