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

>> I'm looking at the file testsuite to evaluate your testsuite,
but it seems to be missing some of the examples that you include in
your write up.  Not sure what happened there, but on a spot check
the example 

Source: author
Vetted: f
Judgment: g
Phenomena: {argument optionality}
Ikh es dem epl.
Ikh es dem epl.
1SG.NOM es;1SG DEF.ACC.M apple.
I eat the apple.

doesn't seem to be included in your skeletons/lab3/item file, either.


Source: {a:217}
Vetted: t
Judgment: g
Phenomena: {tam}
Ikh hob lib.
Ikh hob lib.
1SG.NOM have;1SG like;1SG.
I liked.

>> If this is really present, then the translation is wrong.  Also,
is this an example of argument optionality too?

Our grammar then appears to leave an ambiguous gap, where the verb has
a -n or -en suffix that is ambiguous between infinitive and past
participle interpretation when prefix ge- does not occur. This does
not occur all that often in Yiddish however, as the past participle
often -- but not always -- involves vowel mutation of the root (which
our morphological analyzer 'undoes').

>> You can address this with a "requires" constraint, stating that
the past participle verison of the -(e)n suffix requires the ge- prefix.

Since the Grammar Matrix does not support verbal particles, we only
implemented the prefix form of the perfective. We did this by firstly
creating a perfective aspect sub type (along with the other two
aspects described below), then by creating a position class that can
attach to any verb and an inflecting lex rules which handle the
following prefixes: on-, der-, far-, oys, op-, constraining the aspect
to be perfective.

>> If this is something you want to refine later (and this would
be optional), I would recommend looking to Antske's Germanic CLIMB
for inspiration on how to handle the separable-prefix verbs.

Source: {a:221}
Vetted: t
Judgment: g
Phenomena: {tam}
Er tut shraybn.
Er tu-t shrayb-n.
3SG.NOM.M do-3SG write-INF.
He regularly writes.

Even though ton is really more of a particle, we implemented it as an
auxiliary for now, in a similar way to nemen above, so that we could
parse these sentences. We will return to this and analyze it correctly
once we start hand editing the tdl.

>> Your gloss indicates that ton agrees with the subject.
Why do you think it's not an auxiliary?

It is possible to use nit in front of constituents other than the
verb. This still functions as sentential negation but serves to place
the focus on the particular constituent. Our grammar currently does
not cover nit occurring in a marked position. Examples of placing nit
in front of subject, direct object and adverb are below:

# alternative focus of sentential negation 
Source: author
Vetted: f
Judgment: g
Phenomena: {negation}
Nit der man leyent dos bukh.
Nit der man leyen-t dos bukh.
NEG DEF.NOM.M man read-1SG DEF.ACC.N book.
The man isn't reading the book.
# (=Someone else is reading the book.) 

>> What is the basis for calling this sentential negation, rather
than constituent negation?  (It could be, I'm just curious.)

ARGUMENT OPTIONALITY

>> What about subjects?

Yiddish has both promximate and distant demonstratives. When stressed,
the definite articles ('der', 'di', 'dos', & 'dem') function as
proximate demonstratives. Since this is a prosodic feature of the
language, we could not model this using the matrix.

>> You could make homophonous lexical entries...

This is the kind of error it would be nice to catch in the validation
of the questionnaire, however would be difficult without going down
the path of compiling the grammar.

>> That's what testsuites are for :)

The sentence 'Ikh trog dem epl.' gets two parses because dem is
ambiguous between a definite determiner and a possessive determiner.
We will have to work out how to constrain this if possible.

>> That sounds like legitimate ambiguity to me!  

The sentence 'Ikh es.' gives us two parses when it should just give
one. The spurious sentence does not have a VP but instead a V
immediately under the S, getting a full parse with the rule
basic-head-opt, instead of subj-head. We are not really sure why this
happens!

>> See GoPost.

As already mentioned, Yiddish uses AUX + verb constructions to
indicate tense. The auxiliary can either be the finite form of hobn
(to have) or zayn (to be), with the specific form being selected by
each verb, as illustrated below. This auxiliary selection is currently
not possible through the matrix.

>> Actually, I think you can handle this.  On the "Other features"
page, you can make a verby, syntactic feature (say "AUXTYPE" with
values indicating the different auxes).  Then define your verbs to
have a particular AUXTYPE and your auxes to constrain their
complements to have a particular value for AUXTYPE.

The two main auxiliary verbs in Yiddish (hobn and zayn) have irregular
inflections for person and number. This means that full paradigm of
their forms cannot be captured through inflectional rules and each
form must be enumerated manually. While the forms of the auxiliary can
be enumerated under each auxiliary lex type, no constraints can
currently be placed upon these forms, which we need in this case to
constrain person and number. If we were dealing with verbs, it
wouldn't be too bad manually creating separate types for each form,
however auxiliaries impose many more constraints, and duplicating
these for each entry would be both tedious and clunky -- far better to
save this for hand editing the TDL.

>> It seems like the main problem here is that the questionnaire
doesn't apply the functionality for creating hierarchies of lex
types to the auxiliary section.  Nor does it allow PNG constraints on
the subject to be part of an aux. definition.  So yeah, wait for tdl.

