---------------------------------------------------------------

 Currently, the grammar does not distinguish the two negation markers
 "bu4" and "mei2". A constraint needs to be added to enforce agreement
 of negation markers and verbs (much like the agreement of classifiers
 and nouns).

My understanding of this phenomenon is that it has to do with 
an aspectual distinction, and so would best be implemented in the
context of an analysis of aspect.

 I also noticed that negated sentences can be incorrectly interpreted
 as imperatives, even when the subject is not 2nd person. The grammar
 requires the subject of imperatives to be 2nd person, but this
 restriction doesn't seem to be working if the matrix verb is
 negated. For example, the following sentence is interpreted as both a
 declarative and an imperative:

This sounds like a matrix bug:  The scopal adverb is the semantic
head, so the HOOK of the adj-head phrase is coming from the adj daughter,
but it's not saying anything about the HOOK.XARG.  Thanks for pointing
this out.

 There is still a problem with possessives that I have not figured out
 how to fix: Possessives can appear either before or after the modified
 noun and with or without the possessor SUBJ. All three of these
 sentences produce parses, but two are ungrammatical:

    wo3 kan4jian4 ni3 de lao3shi1
    1SG see 2SG GEN teacher
    'I see your teacher.'

    *wo3 kan4jian4 lao3shi1 ni3 de

    *wo3 kan4jian4 de lao3shi1

That last one is actually grammatical as a relative clause, isn't
it? ;-)  (You'd want different semantics, though.)

It sounds like you need to constrain the head-mod rule to require
that the modifier daughter's SUBJ to be empty.  Also, you'll want
to make 'de' POSTHEAD -.  

 Sentences with classifiers parse correctly. For example:

    na4 ben3 shu1 hao3
    that CLF book good
    'That book is good.'

 But generating produces one good parse and one ungrammatical one (with
 a doubled demonstrative and no classifier):

    na4 ben3 shu1 hao3

    *na4 na4 shu1 hao3

 I haven't figured out what is causing this.

The classifier is putting an underspecified relation (predsort) into
the semantics.  You want it to contribute *no* relation.  The problem
is probably that it's inheriting ultimately from single-rel-lex-item,
and you don't want that :-).



