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

The morphemic representation of the pronouns are not attached
in SOV order.

>> That's fine --- there's no reason to expect them to be.
Affix order within words is independent of major constituent
order in sentences.

 however the challenge we saw there was that we don't currently have a
mechanism for attaching person-number-gender to a verb to specify
features on an NP that does not actually exist.

>> At the point where you are attaching this info to the verb, the
verb has no knowledge of whether the NP will be overtly realized.
The solution is the argument optionality library, which I think you
are digging into now.

(In particular, where the pronoun affixes are not in S-O order.)

>> You might need to create more position classes and work carefully
on their inputs -- rather than having one pc for all S affixes,
you would separate out the ones that are positioned differently.
Does that make sense?

Lakota has five different values of number agreement: singular, dual,
plural, collective and distributive. Collective and distributive can
be considered subtype of plural, especially considering that In many
situations, there is no distinction between them. Third person stative
verbs and agentive active verbs exhibit the phenomena. The dual is
only used in the first person.

>> Are the number values ever marked anywhere other than on verbs?
If so, does plural marking in one place agree with collective and distributive
in another?

In order to model this, we defined three types with the default super
type "number": single, dual, and plural. Two additional types were
created (collective and distributive), with "plural" as a super
type. With this arrangement, we can constrain a third person agent to
be plural, while also allowing for it to be distributive or collective
as well.

>> This sounds reasonable, though I can imagine data which 
would contradict it.  (i.e., a testable hypothesis!)

In our model, we chose to create three genders taking the default root
super type "gender": animate, inanimate, noncountable. Two additional
genders were created with the super type "animate": human and
nonhuman. With this arrangement, we can specify nouns as human, taking
an indefinite article with gender human while the verb agreement is
animate.

>> This sounds good, though it is perhaps a bit unusual to see a language
treating "noncountable" as a noun class along side animate, etc.   What
is the evidence that it's part of the same system?

Because the vast majority of verbs are either transitive and
intransitive, we decided to base our case choice for the model on
those. The intransitive accepts either of two cases, so split-s was an
appropriate fit for our model. The impersonal and ditransitive verbs
can be addressed at a later time in our tdl files.

>> You may have a split agreement system, but nothing in your IGT
indicates that there is any case in the language.  Case = variation
in form on NP arguments, depending on their relationship to the verb.

Lakota is not a direct-inverse language. There is no grammatical
gender, for example, on agentive arguments.

>> What does grammatical gender have to do with direct-inverse?
For that matter, what do you mean there's not grammatical gender
on agentive arguments?  The gender should be inherent to the argument.
Perhaps you mean that the affixes agreeing with (or standing for)
agentive arguments don't mark gender?

Except for emphatic uses, pronouns are not distinct
words in Lakota. They are morphological phenomenon, appearing as affixes on the verb.

>> If there are emphatic uses, then there are separate pronouns,
even if they are not used that often.

However, mass nouns only are syntactically relevant when dealing with negative indefinite articles,
which the customization system cannot specify.

>> This suggests that there is no agreement for the so-called "noncountable" gender,
and that suggests that it isn't really part of the gender system.  Also: the
customization system can create negative indefinite articles, and can restrict
them to occur with only certain types of nouns.  What we can't do is to restrict
to only occur in negative contexts.

The verbs in Lakota can be categorized in three different ways. Active instransitive, stative
instransitive and active transitive. As a split-s language, the intransitive verbs may take
either an agentive or patientive cased argument. Those cases satisfy active and stative
verbs respectively. Transitive verbs always take two arguments: one argument each in the
agentive and patientive case. There also exist special transitive verbs that take two
patientientive arguements.

Source: author
Vetted: f
Judgment: g
Phenomena: pn
igmú kiŋ wakhúwa
igmú kiŋ wa-khúwa
igmú kiŋ 1SgAgt-khuwá
cat DEF 1SG.AGT-chase
'I chase the cat'

Source: author
Vetted: f
Judgment: g
Phenomena: pn
igmú kiŋ makhúwa
igmú kiŋ ma-khúwa
igmú kiŋ 1SgPat-khuwá
cat DEF  1SG.PAT-chase
'The dog chases me'

>> There's no evidence of case here.  You want to answer that the language
has no case, and then handle this via three types of lexical rules for the three
types of verbs.  This will likely involve rules in the different PCs having
the same spelling in some cases.

We created three verb types for our Lakota model. One type for transitive verbs that requires
two arguments: agentive and patientive. We did not created a type for the transitive verb that
requires two patientive arguments, because the customization tool did not allow it (discussed below).
Because we had designated our language a split-s case system, the intransitive verb types were
easy to implement. We created an itr-stative type that has a patientive argument, and a itr-active
that has an agentive argument.

>> When would you want to "patientive" arguments?

Lakota has only one definte article (except for semantic differences). The indefinite articles
are quite complex, on the other hand. Factors including number, person, time, gender, and negation
all play a role in the indefinite article's agreement.

For our Lakota model, only one definite article was implemented. As described below, most of
the variables in the indefinite article form were not able to be accounted for in the model.
For this reason, we created a separate indefinite article for singular and plural, as all
genders within those are identical.

>> The customziation system aboslutely can handle determiners that agree with
the nouns they combine with in gender.  Number is also not surprising.  Person is odd:
do non-3rd person nouns combine with determiners?  Or are you talking about possessives
that indicate the person of the possessor?   I'm puzzled as to what "time" would be
on a determiner, still.

The grammar can handle the unspecified transitive, that would cover our patientive-patientive
transitive verbs, however it would also allow other combinations of argument cases that we do
not want to allow. Of course there would also be an issue of overgeneration. All of these verb
types can later be added to our grammar by editing the appropriate tdl file.

>> You can add constraints on case to the two arguments if you use the unspecified form.
However, I repeat, this is NOT a case system. I'm curious if the materials you are working
from refer to it as case. 

A second phenomena that the customization system was not able to cope with was the negative
polarity of Lakota's indefinite articles. To further complicate matters, articles taking on
the negative form occur in complementary distribution with a semantic "time" feature (real
versus hypothetical), existing only in positive polarity. The following examples demonstrate
the negative and time paradigms of the indefinite article:

šuŋkawakȟáŋ waŋ bluhá.
horse            IDF have-1SG.AGT
'I have a (real) horse.'

šuŋkawakȟáŋ waŋží luhá he?
horse            IDF.HYP have-2SG.AGT Q
'Do you have a horse?'

šuŋkawakȟáŋ waŋžíni bluhá šni.
horse            IDF.NEG have.1SG.AGT NEG
'I do not have a horse.'

>> Okay, "time" is a terrible term for this phenomenon.  Is that from your
source materials?  I think the best bet here is to treat these as all
basically the same (if they can combine with the same types of nouns, that
is) and overgenerate for now.


1SgAgt := pn-1sg-agt-noun-lex &
  [ STEM < "1SgAgt" >,
    SYNSEM.LKEYS.KEYREL.PRED "_1SG.AGT_n_rel" ].

>> You'll want to remove the choices that were creating these fake
pronouns from the choices file.

