[matrix] need some advice on free word order parsing

[Emily M. Bender]

Dear Atlaf,

Thank you for using the Matrix.  Please keep posting questions here,
so we can all gain from increased understanding of Bangla and the
Matrix.

To answer your last question first, it is not expected that users
should understand all of the types defined in the Matrix in order
to begin developing a grammar.  While there is no thorough documentation
of the types that exist, I recommend following the lab exercises
from my grammar engineering class:

http://courses.washington.edu/ling567

... as well as posting questions here (and perusing the archives
of this list).  In addition, you should look carefully at the types
defined in bangla.tdl (by the customization script) as a starting
point. 

Regarding "pragmatic free word order", if you specify "free word
order in the customization script, you will get a series of phrase
structure rules that allow the major sentence constituents (Subject,
Verb, First complement, Second complement) to be realized in any
order, without any spurious extra parses.  This is achieved by the
following means:

1. We instantiate both head-final and head-initial versions of the
head-subj, head-1st-comp, and head-2nd-comp rules.

2. Neither the head-subj nor the head-comp rules require the other
rule to fire "first", such that subjects can attach before objects
and vice versa.

3. The head-2nd-comp rule picks of the second (rather than the first)
element of a COMPS list, in order to allow the complements to be realized
in either order.  (If you have heads which select for three complements,
you can add a head-3rd-comp rule on the model of the head-2nd-comp
rule, but we have not done this yet.)

4. In order to keep down spurious ambiguity (e.g., two parses of
a string in SVO order, where either the subject or the object attach
first), we require any arguments appearing to the right of the verb
to attach before any arguments appearing to the left.  This is tracked
through the feature ATTACH.  

Finally, I should note that we do not yet have any analysis of the
pragmatic effects of word order in free-word order languages.  That
should come eventually as part of a general study of information structure
in this context.

Emily

----- Forwarded message from ~Split~Fire~ <raptor167 at gmail.com> -----

Dear Madam

I am Altaf Mahmud, student of BRAC University, Dhaka, Bangladesh (Web:
www.bracuniversity.ac.bd). I currently doing some research on Bangla
Language Processing. 'Bangla' is a pragmatically free word order language,
and I want to build up a formalism for syntactic parsing in HPSG using LKB.
I go through some papers and books written on HPSG and LKB, and I downloaded
a matrix for pragmatically free word order. The matrix created about 1250
nodes! So, it will be a tedious process to explore the whole code. Would you
be kind enough to give some advice and any key idea about the matrix
construction (how did it handeled pragmatically free word order)? Do you
have any paper or technical report about free word order formalism in
matrix? And the final question is: Do I actually need to understand the
whole code constructed by the matrix to build up a full resource grammar on
Bangla language?

Thank You.

----- End forwarded message -----