Ryan Georgi
Ling 567

Lab 8 Writeup:

----======== PRELIMINARIES =======----

For this lab, (because of the "I Can Eat Glass Project"), a translation
of "I can eat glass, it doesn't hurt me" was rather easy to find.
Using:

http://www.geocities.com/nodotus/hbglass.html

I was able to find the following translation for Egyptian Arabic:

Ana momken aakol el-ezaz, we dah ma beyewgaaneash 

Seeing as how this appears to be a phonetic spelling rather than
transcription, I was able to figure out the probable MSA analogue
of the above:

?ana  ?-imkin          ?-akal  ?al-kizaz-a,   wala     da:  (y)-yuga-ni:
I     1SG-be-possible  1SG-eat the-glass-ACC, and-not  this  3MSG-hurt-me(OBJ)

Here's the closest my grammar comes to laying out modals:

yimkin             yi?ra      lkita:ba      da:
3msg-be-possible   3msg-read  the-book-ACC   this
'It's possible for him to read this book' (= possibility)

yi?dar             yi?ra      lkita:ba      da:
3msg-can           3msg-read  the-book-ACC  this
'He can read this book' (= ability/capacity)

(p. 230)

Taking into account the info provided by my grammar (and given that I've
been building on its info) the preferred parse I'm going to use for
"I can eat glass. It doesn't hurt me. Is:"

?-i?dar  ?-akal    ?al-kizaz
1SG-can  1SG-eat   the-glass
'I can eat glass'

la:   y-yuga-ni:
not   3MSG(it)-hurts-me
'It doesn't hurt me.'

--------======== NEGATION ===========---------------

As covered in previous labs, negation in my langauge is mainly handled
by the negative particle /la:/ preceding the VP. Though largely handled
by the matrix in the form of a neg-adv-lex (simple adverb that selects
its modifier as HEAD verb and marks itself as POSTEAD -).

Due partially to the negative particle, my grammar was overgenerating
considerably, as it was attaching to nearly everything HEAD verb, so I
wanted to constrain it to fewer phrases. Unfortunately, the best information
I can get from my grammar about whether la: attaches to S or VP is shown
in the top example that uses 'wala', where la: is cliticized to the 
coordinating 'wa.' Still, the position it is in could still be treated
either as attachingto the S or VP. Since I simply needed to pare the
overgeneration down, and my grammar provided no ungrammatical examples in
this area, I added the constraint to neg-adv-lex that its modifier's VAL.SUBJ
must be nonempty, which forces /la:/ to attach to VPs, rather than S's and
helps reduce the previously huge number of generated strings.

There are other ways to do negation in Arabic; some are included in my
test suite. Specifically, there is what appears to be a negative form
of the copula, /lays-/, such as in the following:

lays-tu		jundi:y-an
not-1SG		soldier-ACC
'I am not a soldier'

(p. 240)

As I haven't gotten to the positive copula yet, I haven't treated this
version of negation.

Lastly, the other largely used form of negation appears to be one used
in the given dialectal "It doesn't hurt me" at the top, /ma:/. ma: is
a particle used with the "subjunctive" verb form (as usual, not quite the
standard indo-european usage here) to indicate some form of unfinished
or hypothetical aspect. Since this statement seems to more closely fall
into this category, this is, strictly speaking, probably the preferred
treatment.

Unfortunately, this verb form (like most in Arabic) is not formed by
affixation, but rather an entire change of the vowel pattern of the verb.

Since I'm (a) not sure how to differentiate this hypothetical negative
in the semantics and (b) not quite ready to create a new lexical entry
for every one of the verbs to fit this paradigm, I'm going to just stick
with using la:, at least for this lab.

-----========== MODALS ==============------------

Contrary to what I'd previously thought about Arabic, it appears that
modals do exist, and are in fact quite regular, as seen in the
preliminaries section. To recap; arabic modals are auxiliary verbs that
show subject agreement with their argument verb. Thus, it was relatively
straightforward for me to create a modal-verb-lex inheriting from
verb-lex and trans-first-arg-raising-lex-item. To make sure it got
the right comps (as I was having difficulty with this), I made sure
all verb types other than the modal-verb-lex were AUX -, and modal
verb lex was AUX +. Furthermore, it specified the second item on its comps
list be HEAD verb, AUX -, and have COMPS null (so it will take a VP rather
than an unattached V).

Lastly, I thought this would've been handled already, but I was getting
very many spurious parses until I explicitly identified the COMPS element
of modal-verb-lex with the second arg-st element. This ensured that
the modal was only taking the VP as a complement, and not attaching
in any number of other ways.

Something I left intentionally ambiguous; my grammar only barely covers
modals, and from its coverage, I've surmised that the modal and main verb
must agree in, well, agreement; but it's not clear what happens when there
is an explicit subject. For instance:

y-i?dar   ?al-rajul-u  y-i?ra     ?al-kita:b-a    da:
3MSG-can  the-man-NOM  3MSG-read  the-book-ACC    this
'The man can read this book'

...would seem to follow from everything I know about the grammar at this point.
What's not clear is the grammaticality of the following:

y-i?dar   y-i?ra       ?al-rajul-u   ?al-kita:b-a   da:
3MSG-can  3MSG-read    the-man-NOM   the-book-ACC   this
? 'The man can read this book'

This version obeys all the rules for word order in the main verb, and
subjects get dropped all the time in Arabic; I will try to check with
a native speaker, but my grammar provides no insight into this, so I'm
leaving my grammar to produce both alternatives for now.

(To show understanding:) Were I to limit the grammar, I might consider
using the 'opt' feature to the head-opt phrases indicating their
subjects' "embededness" and constrain the second ARG-ST member of the modal 
to this head feature, guaranteeing the verbal complement would be 'missing'
its subject.


---==== WRAP-UP ======--------

"I can eat glass":
 -- ?-i?dar ?-akal  ?al-kizaz-a
    1SG-can 1SG-eat the-glass-ACC

 parses with the preferred parse, and generates only grammatical strings.
(?kizaz-a-n in this case isn't the same meaning of ability to eat glass
generically, but rather it would be some as yet undetermined (but somehow
more specific!) glass)

"It doesn't hurt me":
 -- la:  y-yuga-ni:
   NOT  3MSG-hurt-me(OBJ)

...again parses fine and generates back the string. It also includes
a la: y-yuga-ni:, ("?? Does it not hurt me?"), which I am unsure of
the grammaticality. (Note above that "it" is 3MSG, as there is no "neuter"
in Arabic and due to the prodrop, there's no explicit expression of the
pronoun.

--------
OVERALL COVERAGE

Coverage remains good (42.6), and due to the changes listed below (in "Other fixes"),
there is far less overgeneration, and the parses I am getting are largely
legitimately ambiguous (e.g. /t-/ is both the 2SG and 3FSG prefix), and limited
to generally only 2 or 3.

--------
OTHER FIXES

In addition to the work for this lab, I fixed quite a number of things
that were broken in Lab 7, such as changing my imperative rule to 
const-ltow-rule and adding INFLECTED - to verb-lex
(thanks for the email, I'd figured these out just the night before!)

Another fix; since I was using [ FORM imp ] for imperatives, and the modal-verb
type was going to take VP as an argument, I made sure that any subject-inflected
verb was [ FORM fin ] by adding that constraint to subj-agreement-lex-rule.

More bugfixes in my "finiteness" lex rules. The definite markers were previously
inheriting from infl-ltow-rule for some debugging I had done and not fixed,
instead of fnt_lex_rule, which constrains the input to the type cased_rule, the
output of the case inflection, ensuring that such inflection is mandatory.

I also realized that when I had written my obj_incorp_lex_rule, I'd written
it as infl-val-change-only-ltol-rule, and not fully constrained the VAL features.
As a result, it seemed that I was actually getting the SUBJ of some verbs
wiped out when this rule applied, resulting in all sorts of incorrect parses.

(Finally, I also added the clausal-second-arg-trans-lex-item to my 
embedding-verb-lex and no-hcons-lex-item to my noun-lex, fixing some 
of the issues I'd left broken in lab 7.)


--------======== STILL NOT WORKING =======----------

As I've probably made many references to, I still have
never had a chance to implement a form of the copula, which is 
all over my test suite, and one of my other forms of negation, as mentioned
above.

Another unresolved issue is that I have a large number of proper nouns
in my test suite; another simple fix, but one I just haven't had a chance
to cover in previous labs.

Finally, one thing I think might be interesting to cover that I use in
a few of my testing examples is the genitive case for posessives and
"construct" phrases.