Notes
-------------------------------------------------------

And then some determiner-noun dischord:

#32 - det-noun mismatch in the subject 
Source: author
Vetted: f
Judgment: u
Phenom: wo, agr 
les fille mange les biscuits 
les       fille         mang.e   les      biscuits 
The.3PL   girl.3SG-fem  eats.3SG the.3PL cookies.3PL-masc 
'The girl eats the cookies' 

And some noun verb dischord:

#40 3PL subject with 2SG verb:
Source: author
Vetted: f
Judgment: u
Phenom: agr 
il manges le gâteau 
il         mang.es    le           gâteau
he.3SG     eat.2SG    the.3SG-masc cake.3SG-masc 
'you eat the cake' 

These exmaples illustrate agreement, not word order.  You also
need examples with noun-det order. 

  French does have auxiliary verbs, which appear before its
  complement.  Those complements are V's, although, I don't think they
  do any raising.  Just tense marking.

In order to get the clitics attaching to the auxiliary, it is probably
most convenient to say they take V complements rather than VP complements.
You'll have to say a little more to keep the auxiliaries from combing
with a V with clitics on it, though.    A first pass would be to disallow
the "clitic" rules from applying to the verb forms that the auxiliaries
take as complements.  It would be good to add examples to your test suite
illustrating this.

"Raising" means that the auxiliary is taking over the verb's arguments,
putting them on its own SUBJ/COMPS lists.

  Unfortunately, I wasn't able to implement any uses of auxiliaries in this lab.. 

Are they in the test suite, though?  (A quick glance says no.)

  I chose nouns of both genders, however, since I'm not using
  adjectives yet, I haven't had the opportunity to test this.

French determiners also agree in gender with the nouns they combine
with, so you should be able to test it.

  Case: French exhibits no case.  Argument roles are determined by
  word order, with the subject going before the verb, and the objects
  after.

French arguably has case in the free pronouns, like English (je v. moi).
An alternative analysis is that je etc are prefixes just like the object
clitics.  This helps with examples like "Moi, je mange ..."

  Sentential Negation: French achieves negation through the use of
  both inflection (I considered the -pas a suffix on the verb) and an
  adverb (ne), which goes before the verb.  Once I tackle the complex
  tenses, that use an auxiliary, I may revise this.

I think the better analysis is to take ne as a prefix and pas as an
adverb. (In fact, following Jong-Bok Kim's analysis, ne- is a prefix
that adds a negative adverb (pas, rien, point) to the comps list
of the verb, but like the negative addition LR from SWB 2003.  We
got that idea from him.)  The customization system isn't currently
providing this, though, so just going with ne- as a prefix and pas
as an adverb is a good approximation.  

As noted on GoPost, the problem with these examples seems to be
that the -pas forms are still [INFLECTED -].

#6 plur, inanimate, pronominal object 
Source: author
Vetted: f
Judgment: g
Phenom: pro-d
il les-mange 
il          les-mang.e             
he.3SG-masc it.3SG-fem-eats.3SG   
'he eats them' 

You're getting four parses of this example.  It seems that the les-prefix
can attach to the intransitive as well as the transitive form of manger.
In addition, you don't have agreement implemented, the non-2sg
rule is optional, so you get parses with and without it.  This means that
"tu les-mange" is parsing when it shouldn't.  

The first problem has to do with the fact that the object prefixes aren't
constraining the PNG of the object:

    verb-slot6_morph3_name=plur_inanimate_obj_pronoun
    verb-slot6_morph3_orth=les-
      verb-slot6_morph3_feat1_name=overt-arg
      verb-slot6_morph3_feat1_value=not-permitted
      verb-slot6_morph3_feat1_head=verb

These prefixs should say something about the person and number
of the object of the verb.  Without that, the not-permitted feature
isn't doing anything, and you are also parsing "Il les-mange les pommes".
(Which is arguably grammatical in spoken French, but should be included
in your test suite one way or the other.)

  I have not implemented any animate objects, and therefore, no pronouns. 

I don't see how this follows.  You don't need animate nouns in the
grammar in order to add the non-3rd-person object prefixes.

  French forms the past perfective using an auxiliary, avoir or etre,
  conjugated, and the past participle of the verb.  Since these are
  irregular verbs, I was not entirely clear on how to input them,
  since they're not really inflections.  I treated them as such,
  making them suffixes on an empty stem.  My next problem was how to
  define the past paeticiple, and then associate it with the
  construction.  I'm sure there's a way to do it, but I wasn't able to
  find it.

You *don't* want to posit empty stems.  What you should do here
is regularize the inflection on the auxiliaries in the morpheme-segmented
line so that "sont" becomes "etre-ent" or something like that.  

You can define the past participle by creating a new lexical rule for
verbs that adds the ending and constrains the FORM value to something
appropriate that you define on the other features page.  The auxiliaries
should then select for complements with this feature.  You can also
create a feature on the other features page to get the avoir/etre selection.
The idea there is you make a feature (say, AUX-TYPE) that is a syntactic
feature (on verbs; this is what syntactic features are in the customization
system right now).  Make the possible values of AUX-TYPE avoir and etre.
The etre verbs specify AUX-TYPE etre in their own HEAD value, and
etre selects for AUX-TYPE etre.  Similarly for avoir.  The last piece is to figure
out where to put the tense/aspect information associated with
passe-compose.  It could be in the auxiliary, in the past participle inflection,
or split between the two.  Laurie Poulson might have an opinion about this.

  My test suite now contains 171 sentences.  Of those, thanks to
  overgeneration, I got 282 parses.  At last glance, nothing that
  should not have parsed parsed.  I think the sources of
  overgeneration are as follows:

You don't need to estimate this.  Do Analyze | Overgeneration in 
[incr tsdb()] to find out for sure.  (I see that you have 5 ungrammatical
examples parsing.)  Likewise, please do Analyze | Coverage
to find out how many good examples are parsing at all.

Btw, "overgeneration" technically means parsing examples that shouldn't
parse/generating ungrammatical examples.  Extra parses for
good examples constitute spurious ambiguity.

  1) verbs that are underspecified for gender can therefore parse as
  masculine and feminine.  This includes basically all the personal
  pronouns..

I'm not sure how this would lead to any spurious ambiguity.

  Demonstratives: These are yet another determiner, also declining for
  both gender and number, ce(t), cette, ces.  Further reference to
  proximity can be achieved using the sufixes -ci and -là on the noun.
  Unfortunately, I was unable to implement these suffixes, and
  neglected to put any examples of these in my test suite.

It's fine not to implement the suffixes at this time, but the examples should
go into your test suite.

Regarding argument optionality, I'm not sure you need multiple entries
for manger ... it should be possible to have some verbs require the object
prefix in order to drop their object and others (like manger) to allow but
not require the object prefix in this case.

It also looks like your sg/pl suffixes aren't required ... I see edges for
lexical entries that haven't undergone one of these rules showing up as
daughters of phrase structure rules...  (Because the determiners are also
constraining PNG, but having extra edges is still going to give you extra
parses.)

