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

 Malayalam contains overt personal pronouns, and I have lexical entries for
 I, you, he, she, and I added one for who.
 These have all been specified to have a COG-ST value of activ-or-more.

We're not handling wh- questions in this class.  Just adding the
pronoun 'who' won't be enough to get the semantics right...

 this and that [ii and aa] are demonstrative determiners, so I assigned
 them COG-ST active+fam.

I assume you mean demonstrative adjectives, since that's how you
describe them below?  But in your grammar, they're still determiners.
You'll need to follow the directions from this lab (lab 6) for
demonstrative determiners in order to get the right semantics for these.

 I don't think that there are any verb which can't drop their subject in
 Malayalam, so I didn't add any restrictions to this.

I don't think that lexical exceptions to subj pro-drop are wide-spread.
There might be some somewhere, since there seem to be lexical exceptions 
to just about anything, but still...

 The lexical entry for want-FUT was changed to derive from
 transitive-subjdat-objopt-verb-lex.

Do you get the impression that the possibility of dropping objects is
lexically specific?  It doesn't sound that way from the rest of your
description.

 (Just out of curiosity - if I generate, and I have a rule that this verb can
 have an optional argument, should I expect the grammar to generate sentences
 with all the missing pronouns?)

Not with the current semantic representations.  The overt pronouns
are contributing a pronoun_n_rel, while the so-called "dropped" pronouns
don't correspond to any relation in the MRS.

 The adjectives are inflected to match the noun they follow, even though the
 form of the noun may not change according to its gender.

Right --- gender is usually an inherent property of nouns.

 While I was reading about Malayalam, it was often stated that Malayalam
 doesn't really have adjectives.  There is a small class or words, including
 good which act as adjectives.  There is also a mechanism by which nouns are
 'adjectivalised' by adding the relative participle of unatax (be), ulalaa.
 There is also the relative participle of aakukua, aaya.

I'd say that Malayalam does have adjectives, just that most of them
are derived from nouns.  It would be interesting to try to write
the lexical rules for this, especially to try to figure out the right
semantic representation (given that we can't change relations or throw
them away).

 Transitive verbs can take either a NOM or ACC subject.
 The subject should be ACC case if it is HUMAN +, or ANIM +.  Otherwise the
 object is NOM.  I modified my case types to include a super-type nom+acc, and
 made case nom and acc subtypes of that.  Eventually, I will have to implement
 the agreement rules which can create the correct agreement between the object
 noun and its verb when the noun is an object.

I think this is why I thought that Malayalam might be a split-ergative
language.  What happens to the object when the subject is marked with
accusative?  It seems like you're going to want lexical rules which
fill in the case requirements on the verbs, along with constraints
on some feature like HUMAN or ANIM to make sure only the right nouns
appear.

 cold := basic-adverb-lex &
   [ STEM < "thanuppundu" >,
     SYNSEM.LKEYS.KEYREL.PRED "_cold_r_rel" ] .

I'm surprised to see something glossed as 'cold' tagged as an adverb.
Does this really modify verbs?