I already had the definite article ("e" or "n") from the last assignment.  I
have assigned it the PRED value def_q_rel.  I've added the indefinite
article ("me"), and assigned it the PRED value indef_q_rel.  Armenian does
have demonstratives, with the usual proximal/distal/remote semantics, but
they are apparently simply adjectives rather than determiners, since they
require that the definite article be used as well.  I have lexical entries
for these demonstratives, with proximal+dem_q_rel, distal+dem_q_rel, and
remote+dem_q_rel as the PRED values, but they're commented out until we do
adjectives.

(By the way, I've also added all the vocabulary items you asked us to find
as preparation for this lab, so lexicon.tdl is a bit big.  The determiners
are down at the bottom.)

My various references say that the article can be omitted in Armenian (and
the meaning recovered from context, presumably).  This means that both of
the following sentences are OK (recall that "ke" is the present/imperfect
tense particle):

   e mard e girkh ke karday
   "The man reads the book"

   e mard girkh ke karday
   "The man reads book"

In addition, it appears that articles can be used with all classes of nouns,
including proper names.  Therefore, the glosses of both of the following
sentences are OK (note that they don't actually parse, since I didn't put
"Aram" in the lexicon):

   Aram ke phrrshtay
   "Aram sneezes"

   N Aram ke phrrshtay
   "The Aram sneezes"

Again, the meaning must be recovered from context.

These facts have two results in my grammar.  First, because determinerless
NPs can apparently receive either definite or indefinite readings depending
on context, the PRED I used in that rule is simply non+demonstrative_q_rel.
Second, since determiners are optional with all nouns including proper names,
I did not need to define two noun classes as described in the lab.

I couldn't get my grammar to produce multiple sentences from a single parsed
sentence in an interesting way.  It will overgenerate any sentence with the
definite article, but only because it has two forms ("e" and "n"), and my
grammar doesn't have the (morphological?) smarts to distinguish the two.  So:

   e kovu ke merrni
   "the cow dies"

...will generate:

   e kovu ke merrni
   n kovu ke merrni

Another kind of overgeneration would occur in a language in which, for
example, leaving off the determiner in a noun phrase is semantically
equivalent to using the indefinite article.  In that case, after parsing a
sentence with an indefinite noun phrase, we would expect to see a generated
sentence with the indefinite article, and another with.  Both would be
consistent with the semantics of the parsed sentence.

I had expected, in fact, to see generated sentences like this when I omitted
the Armenian article.  Since my determinerless NP rule marks NPs without
determiners as non-demonstrative, and both the definite and indefinite PRED
values are derived from that, I thought that when I parsed a sentence like:

   kovu ke merrni
   "cow dies"

...I would get back three sentences:

   kovu ke merrni
   "cow dies"

   e kovu ke merrni
   "the cow dies"

   me kovu ke merrni
   "a cow dies"

However, that's not what happens -- only the first sentence is generated.
Still, with a small tweak, it's easy to make my grammar overgenerate in
this way.  If the PRED in the determinerless NP rule is changed to
indef_q_rel, and then *either* of the two following sentences is parsed:

   kovu ke merrni
   "cow dies"

   me kovu ke merrni
   "a cow dies"

...*both* of the sentences will be generated, since their semantics are
identical.