Ryan Mattson
May 8, 2005
LING567
Lab 5 writeup

Finnish matrix declarative clauses are in the typical SVO word order that is
often seen in Finnish. A couple of examples:

1a) kissat nukkuvat
    cat-pl sleep-3p-pl
    cats sleep

 b) minA syon      tomaattia
    i	 eat-1p-sg tomato-part
    i eat tomatoes

These sentences are parsable with the current version of the grammar.

Matrix interrogatives, on the otherhand, change the word order (typically) and
add a particle to the end of the question word that marks the sentence as
a question:

2a) nukkuvatko       kissat?
    sleep-3p-pl-ques cat-pl
    do cats sleep?

 b) syOtkO         minA tomaattia?
    eat-1p-sg-ques i    tomato-part
    I eat tomatoes? (I am eating tomatoes?)

The order here changes to VSO and the "ko/kO" question particle (possibly a
clitic?) is being added to the verb in order to mark the sentence as a
question. This construction is currently not working in the matrix, however,
but will probably be added soon (according to recent message board discussion).

Embedded declaratives look very similar to matrix declaratives:

3a) opiskelija  sanoo     ettA minA nAytAn     apinalta
    student-nom say-3p-sg comp i    look-1p-sg monkey-ablat
    the student says that i look like a monkey

The object here is the complementizer phrase 'that i look like a monkey.'

Embedded interrogatives change the word order of the embedded clause, as might
be expected based on the matrix interrogative syntax:

4a) opiskelija  tietAA     mitA      minA syOn
    student-nom know-3p-sg what-part i    eat-1p-sg
    the student knows what i eat

The wh-word 'mitA' (the partitive form of what) is in the front of the
embedded clause (which exists without a complementizer, and I believe it is
ungrammatical to put a complementizer in), creating (in the embedded clause)
an apparent OSV word order.

The current coverage of the grammar is on the matrix declaratives (as noted
above) and the embedded declaratives.

The embedded declarative coverage, however, is currently overgenerating with
respect to the attachment of the S node of the embedded phrase (connecting
to 'ettA', the complementizer) and the attachment of the entire embedded
declarative as the object of the verb. This overgeneration stems from the 
head-complement rule not being able to choose between the S-nodes with or
without clausal semantics.

I attempted to limit the head complement rule to choose nodes with clausal
semantics by using the same type that was specified in the roots.tdl file
(which limited possible roots to only those with clausal semantics), but
I was given a unification error when attempting to reload my grammar if
I specified this type as the type of the complementizers complement and the
non-head-dtr of the head-complement rule. I'm unsure where to go from here.

The semantic information of the clause embedding verbs constrains the CONT.MSG
of the complement to be of type 'message.'

REAL-TIME UPDATE: I realized I hadn't constrained that CONT.MSG feature to
type 'message' and when I did I eliminated two of the overgenerated structures
in the embedded declaratives.

From what I can tell with my still-growing understanding of MRS representations,
correct semantic representations are being given to matrix declarative clauses,
but the overgeneration of the embedded declaratives may be causing spurious semantic
representations to be assigned to those--but these will disappear once I resolve
the overgeneration.

I'm unsure, however, of whether or not I covered all of my semantic bases. Looking at the
indexed MRS representations, I see ones that seem to make sense, however I'm wary of how
much I dealt with the semantics in this assignment. (Given that 2/5 of the writeup
questions ask about semantics, I'm not confident I put 2/5 of my energy into semantics
in this assignment.)