[developers] Questions from Korea - Yong-hun Lee

[Emily M. Bender]

Following Francis's lead and replying on-list (with permission):

Dear Yong-hun,


2011/7/8 이 용훈 <yleeuiuc at hanmail.net>
>
> Dear professor:
>
>
>
> Hello, Professor! Thak you for your answer. I also found the website for your Grammar Engineering classes. I downloaded two grammars (English and Italian) in your 2009 class and ran a translation system. Two grammars work well. However, since these two grammars are a little big (although they are smaller than ERG or KRG), I have some problems to find out which parts of the grammar are necessary for the MT.
>
>

For a LOGON-based MT system, what you need are two grammars
that  can both handle MRS representations --- producing them in
parsing on the source side and generating from them as input on the
target side. In addition, each grammar will need a semi.vpm file
to handle the variable property mapping.  Finally,  you need a third
grammar, called the transfer grammar, that takes the MRSs produced
by the source grammar and produces MRSs in the form expect
by the target grammar.  In the simple example in from the exercise
in my 567 class, the transfer grammar is in fact embedded in the
source-side grammar, but this is not how the larger scale systems do
it.

Can you say a bit more about what you're trying to find in the
grammars?  It's not like some grammar rules are required for
translation and some are not.  Arguably, there are types in the
Matrix-based grammars that are not required for translation since
they are not used in any instantiated rules or lexical entries
in those small grammars.  On the other hand, a lot of what the
Matrix does is to facilitate the creation of grammars that produce
well-formed MRSs.


>
> I found another set of two grammars (English and Esperanto) in your 2005 class (http://lingo.stanford.edu/courses/05/ge/). I think that these two grammars are much simpler than two grammars (English and Italian) in your 2009 class. Could I get the answer for the "Exercise 8" in your 2005 class? If it is possible, it will be very helpful to me. I found that only monolingual translation is possible along with the 'mt.gz' package in the 2005 class. If I get the answers for the "Exercise 8" in your 2005 class, I will compare the answers with the original files and find out what are the differences.
>
>

That one is not my class :-)  It was taught by Dan Flickinger
and Stephan Oepen.  They may still have the answer keys. It's
also possible that the software may have changed a bit in
the intervening 6 years.

>From a second message:



2011/7/9 이 용훈 <yleeuiuc at hanmail.net>
>
> Dear Professor:
>
>
>
> I am very sorry to bother you. I examined two grammars (English and Esperanto) in the exercise of your 2005 class, and found that the transfer rules are included in both grammars (in the 'types.tdl'). I followed the directions and ran the MT system as follows.
>
>
>
> [1] I added the lexical entries for the following words: pretty, boy, and sleep
> [2] I ran the English Grammar, parsed the sentence 'the pretty boy sleeps', and got the correct parse tree and the MRS structure.
> [3] I ran another emacs window and uploaded the Esperanto Grammar.
> [4] I ran [Options]-[Expand Menu] in the English Grammar window.
> [5] I parsed the Esperanto sentence 'la bela knabo dormas.'
> [6] I ran [Rephrase] in the Esperanto Grammar window.
>
>
>
> However, I have no following messages in the *common-lisp* window.
>
>
>
> [12:05:04] translate(): read 1 MRS as generator input.
> [12:05:04] translate(): processing MRS # 0 (4 EPs).
> [12:05:04] translate(): 4 generation results.
>
>
>
> I got only the Esperanto sentences with different inflections in the Esperanto Grammar window. That's my problem. Please let me know how to solve this problem.
>
>
>
> Sincerely Yours,
>
>
>
> Yong-hun Lee
>
>

Comparing the steps you took to the steps outlined here:
http://courses.washington.edu/ling567/lab9.html

it looks like you might have skipped this one:

7. In the "target" LKB, select Generate | Start server.

Still, it's surprising that you got  Esperanto output (instead
of just no output).  Perhaps you selected "Generate"
instead of "Rephrase"?

Emily



--
Emily M. Bender
Associate Professor
Department of Linguistics
Check out CLMA on facebook! http://www.facebook.com/uwclma