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

 I feel like I wasn't as thorough as I could have been, mostly because
 I spent a long time catching up with lab 6.  I know I said I would
 write up those changes here, but I don't have time now. However, I put
 a lot of effort into it, and am pleased with the results, so I will
 send you a writeup of those changes separately.

Please do :-)

 In the first sentence, the complementizer `ukuthi' precedes the clause
 it's attached to. In the second sentence, the question particle `na'
 follows the clause.  To solve this problem, I created two subtypes of
 complementizer-lex-item: qpart-lex-item and emb-comp-lex-item. The
 question particle is the first type, other complementizers are the
 second.  I then made two subtypes of comp-head-phrase:
 qpart-head-phrase and embedded-comp-head-phrase, each of which
 constrain their DTR to be of the corresponding lexical
 type. qpart-head-phrase is head-final and MC +;
 embedded-comp-head-phrase is head-initial and MC -. Problem solved!

You could probably get away with just adding the qpart-head-phrase
and letting your existing head-comp phrase take the complementizer
daughter.  Clever to use MC to constrain things though.  Your
qpart-head-phrase could say that the daughter is MC + and HEAD comp,
and then not need to mention the lexical type directly (generally
not a good idea).

 All of this is further complicated by the fact that imperatives can
 appear with an overt subject:

 u-Sipho gijim-a
 C1a-Sipho run-fv
 `Sipho, run!'

Are you sure this is a subject, and not a vocative?

 Imperatives agree with the number of their subject, which is marked by
 a suffix on the verb. To accomplish this I just made two subtypes of
 imperative-verb-lex-rule, one constrained to be SG and the other
 constrained to be PL. Notice that these do not have tense markers or
 any other agreement marking. To get this working, I had to make my
 imperative rule a lexeme-to-word rule, because the only other
 lexeme-to-word rule I had was the subject concord rule. This bypasses
 tense marking and object concords as well.  The problem is that
 imperatives can in fact take an object concord:

 yi-shay-a in-ja
 OC9-hit-FV C9-dog
 `Hit the dog!'

Why not have the imperatives rule apply *after* the object marking
rule?  That is, they can take the same sort of DTR as the tense rules
can, but they're ltow, so their 'outputs' can be 'inputs' to the
tense and subj concord rules?

; embedded clauses are always propositions.
clause-comp-verb-lex := clausal-second-arg-trans-lex-item & verb-lex &
			takes-oc &
  [ SYNSEM.LOCAL.CAT.VAL.COMPS < [ LOCAL [CAT [HEAD comp,
					       VAL  [ SPR < >,
						      SUBJ < >,
						      COMPS < >,
						      SPEC < > ],
					       MC - ], 
					  CONT.HOOK.INDEX.SF prop ] ] > ] .


Your semantics for your embedded clauses are slightly broken, because
this type doesn't relate the COMPS element to anything on the
ARG-ST.  The linking type (clausal-second-arg-trans-lex-item)
states the relationship between the semantic roles and the ARG-ST
positions, but if your subtype doesn't connect to ARG-ST, the chain
is broken.