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

  I have cheated and implemented 'courir après' (to run after) as
  single lexical items, rather than as a verb paradigm:

More straightforward would have been to implement it as a
verb taking a PP complement, selecting for a particular (semantically
empty P) with a particular FORM value.

  * added NUM pl and PER 3rd to my NP and N coordination rules, so now
    sentences like "le chat et la fille mange" no longer parse.  When
    I'd just added NUM pl, the sentence le chat et la fille
    mangent/mangez/mangeons was generating!

This probably isn't quite right, if non-3rd person pronouns can
appear as coordinands (moi et toi...)

  The one thing that did confuse me, though, was that the INV - is
  what made it work (I'd tried to phase in parts of the rule from the
  English). Of all the features there, INV - is the last one I'd think
  would have made the difference.

The int-cl rule looks for [ INV + ].  If INV is underspecified, then
int-cl will over apply.

  Unfortunately, it's still parsing un verre and les verres.  I don't
  know what else I can do to make 'du' stand out from the other
  determiners..

COG-ST?

  Of course, this only gives a totally accurate translation for "I
  chase you" since these forms, moi/toi, can only appear after a
  preposition (in this case, après).  Ça ne blesse pas moi, I think,
  is an anglicism, and would be more correctly translated 'Ça
  ne-me-blesse pas' but I don't know how to enforce case on a dropped
  object verbal prefix.  I tried, but it didn't seem to take:

It's not case you need to enforce.  You should be able to get this
by a) saying that the ordinary transitive verbs can't take pronominal
complements and b) adding an ARG2 pro-drop rule to acm.mtr.

  * "Do cats chase dogs" doesn't translate with the est-ce que
    question particle.  I poked through the English grammar to see if
    "do" was what added it, but it seemed to be higher up.

It looks like you don't have a trigger rule for est-ce que, so it's
not getting into the chart...

  * "The dogs are hungry" / "The dogs are in the park" - Both these
    give 0 generatiion results.  Comparing the MRS, it seems to be the
    COG-ST again.  Input has uniq+fam+act and I could't find where it
    picked it up.

In the English grammar, this is coming in from "the".

  * "ti-insegu-o" - wow, that's a lean input MRS.  I assume the first
    sg is in the conjugation of the verb, and second sg is a prefix of
    the verb, similar to French.  It would have required a transfer
    rule to suppress all the pronouns and determiners.

Actually, you would need a transfer rule to put the pronouns back in.
This is provided in acm.tdl ... you just needed to instantiate it in acm.mtr.

  "non mi-f-a male" - I avoided this mostly because of the "make" and
  "harm" rels.  I knew it would require a transfer rule.

Again, this was provided.  Did you not look in acm.tdl?

  Sorry I didn't go for the higher hanging fruit, but I had a bit of a
  scare today when I went back to check the Italian translations after
  polishing up the French all day, and _none_ of the Italian sentences
  translated!  I tried to reverse engineer what I'd done today, but
  then they magically started to translate again (I still think this
  grammar is sentient is likes to cause me undue stress!).  Anyway,
  it's made it a little gun shy to change things drastically!

This is where svn is your friend.  Check in a version of the grammar
that you like, and jot down its revision number.  Then play to your
heart's content, knowing you can always check out (i.e., back up to)
the working version.

