----------------------------------------------
Comments

When I load your grammar, there are four warnings in the
LKB top window (about redefined types).  Be sure to watch
that window for error messages.

	I had to change around my noun declension heirarcy a little
 to make them compatable with the determiner inflections. Mainly,
 I needed to add a type specific for just gender and number
 without class information or const/infl information. I also made
 the noun declensions lexeme-to-lexeme inflections to add the 
 determiner suffix. This created problems since, the nouns can
 appear with out the determiner suffix.

This seems to be another argument for distinguishing morphosyntax
from morphophonology/morphoorthography. The declension classes don't
have any import beyond the choice of appropriate suffixes, right?
In an ideal system, I believe that the lkb wouldn't know anything
about this, and an external system would take care of selecting
the right series of suffixes. The lkb would still need to know about
the case/gender etc. information, of course.


	I solved this by adding a rule that made these lexemes any
 num+case lexeme to word. This rule also added the discourse
 information of [ COG-ST type-id ] since when a noun appears
 without the determiner suffix it gets an indefinite interpretation.

This sounds like the best solution to me.

	For the argument optionality, I'm not exactly sure how it
 works in Icelandic. As far as I can tell, I think it has the
 object drop for certain verbs, similar to English, because I
 know it doesn't fit the other options. However, I was unable to
 find a concrete example of this. So I wrote a rule for this, but
 was unable to get the LKB to load the grammar correctly for this .
 The rule is in the icelandic.tdl file, but commented out so that
 the grammar loads.

The problematic thing wasn't a rule but a lexical type (opt-trans-verb-lex).
There were two things wrong with it:  A missing comma after [ CASE acc ],
and the fact that ARG-ST was spelled ARG-S (true for transitive verb
lex as well --- if you don't fix this, your MRSs for sentences with
transitive verbs are likely to be broken, since the linking to semantic
arguments is through the ARG-ST ... but if you have ARG-S instead,
the COMPS and SUBJ aren't linked to ARG-ST, just ARG-S).  I debugged
the former problem by noting the first error message in the LKB Top window,
which gave a position number to go to (goto-char, in emacs).

Once you instantiate the optional complement rule, you'll find that
it's happily applying to your regular transitive verbs, until you
say [OPT -] on the complements of those.  It's also worth noting that
your transitive-verb-lex and opt-trans-verb-lex share a lot of constraints
which could be moved to a supertype.


	For adjectives, I had to add another set of lexical rules.
 The adjectives have to agree with the nouns they modify in case,
 number and gender. This was added to the lexical rules by putting
 constraints on the MOD valuse for each rule. There is also a
 distinction in Icelandic between weak declension and strong
 declension of adjectives. This is a hard distincition to implement
 so I have just added in the weak declension for the moment, which
 will get more coverage for my test suite.

What is the difference between the two for Icelandic?

 note: I have the baseline tsdb in the tar file, but when I tried
 to run it on the new grammar, I was only able to come up with
 a null test suite, so I have not included that in the submission.

That's because you need to update your tsdb/skeletons/Index.lisp file
in order to tell tsdb++ where to find the skeleton. 


 ;helst := adverb-lex &
 ; [ STEM "helst" >,
 ;  SYNSEM.LKEYS.KEYREL.PRED "_gladly_r_rel" ].

This is missing the < after STEM.

Where can the adverbs appear? How will you constrain them to show
up only in the right spots?

Please post questions to EPost (and apply the 10 minute rule!).  As
you can see, there's quite a delay in my getting to the grading, so
it's better to ask the questions as they come up (even if it's midnight
Sunday night), as I'll get to them much sooner on EPost.