Ryan Mattson
Ling 567
5/27/05
Lab 8 writeup

Coordination in Finnish (testsuite: lab8.items)

Finnish has very similar coordination to English--what Scott has termed
monopolysyndeton. An example of the monosyndeton structure is:

1) opiskelija  syO       lasi-a,    tomaatti-a  ja  auto-a
   student+Nom eat+3p+Sg glass+Part tomato+Part and car+Part
   The student eats glass, tomatoes, and cars

The polysyndeton structure of Finnish coordination looks like:

2) opiskelija  syO       lasi-a     ja  tomaatti-a  ja  auto-a
   student+Nom eat+3p+Sg glass+Part and tomato+Part and car+Part
   The student eats glass and tomatoes and cars

While polysyndeton structure tends to make one sound overly-exuberant
in English, it is perfectly acceptable in Finnish and will tend to be
used to separate coordinated phrases that are larger because of
multiple adjectives, etc.:

3) opiskelija  syO       iso-a    punainen-ta tomaatti-a  ja  pien-ta    sininen-ta 
   auto-a   ja  lasi-a
   student+Nom eat+3p+Sg big+Part red+Part    tomato+Part and small+Part blue+Part 
   car+Part and glass+Part
   The student eats big red tomatoes and small blue cars and glass

VP coordination acts in the same way:

4a) minA  kAvele-n   ja  nuku-n      ja  tanssi-n
    i+Nom walk+1p+Sg and sleep+1p+Sg and dance+1p+Sg
    I walk and sleep and dance

4b) minA  kAvele-n,  nuku-n      ja  tanssi-n
    i+Nom walk+1p+Sg sleep+1p+Sg and dance+1p+Sg
    I walk, sleep and dance

As does AdjP coordiation:

5a) syO-vAt-kO       opiskelija-t    hyvA-A    ja  iso-a    ja  punainen-ta tomaatti-a?
    eat+3p+Pl+Ques   student+Pl+Nom  good+Part and big+Part and red+Part    tomato+Part
    Do students eat good and big and red tomatoes?

5b) syO-vAt-kO     opiskelija-t    hyvA-A,   iso-a    ja  punainen-ta tomaati-a?
    eat+3p+Pl+Ques student+Pl+Nom good+Part big+Part and red+Part    tomato+Part
    Do students eat good big and red tomatoes?

Further:

6) syO-vAt-ko     opiskelijat    iso-a    auto-a   ja  tomaatti-a?
   eat+3p+Sg+Ques student+Pl+Nom big+Part car+Part and tomato+Part
   Do students eat big cars and tomatoes?

Example (6) should give two parses:

7a) syOvAtko opiskelijat [isoa autoa] ja  [tomaattia]
                         [big cars]   and [tomatoes]

7b) syOvAtko opiskelijat [isoa [autoa ja  tomaattia]]
                         [big  [cars  and tomatoes]]

Implementing Coordination, and Current Coverage

Implementing coordination was straightforward given the rules that Scott
included in coord.tdl. Because Finnish operates like English, I chose
the monopolysyndeton rules. The following rules are instantiations of
Scott's coordination framework for the NP-based coordination.

np-top-coord-rule := basic-np-top-coord-rule & monopoly-top-coord-rule &
[ SYNSEM.LOCAL.CAT.HEAD.CASE #case,
  NBR-DTR.SYNSEM.LOCAL.CAT.HEAD.CASE #case,
  BR-DTR.SYNSEM.LOCAL.CAT.HEAD.CASE #case ].

np-mid-coord-rule := basic-np-mid-coord-rule & monopoly-mid-coord-rule.

np-bottom-coord-rule := binary-bottom-coord-rule &
[ SYNSEM.LOCAL.CAT.HEAD noun,
  ARGS.REST.FIRST.SYNSEM.LOCAL.CAT.HEAD noun ].

The top rule requires the daughters to identify their (matching) case
(which, for the current implementation of the grammar, is usually
nominative) with the case of the node licensed by the top rule.

Subject NP coordination doesn't say anything about the PER matching,
however, and the strategy to fix this involves creating rules for
each of the PER combinations possible, such as:

second-non-first-coord := per-coord &
[ SYNSEM.LOCAL.COORD-REL [ C-ARG.PNG.PER second,
			   L-INDEX.PNG.PER second,
			   R-INDEX.PNG.PER non-first ]].

A current problem with this is that C-ARG.PNG.PER feature doesn't
seem to be identified with SYNSEM.LOCAL.CONT.HOOK.INDEX.PNG.PER
where the VP is constraining the PNG.PER value at. This results
in a large number of parses, as each rule is being allowed to
parse a coordinated subject phrase.

There is a problem in the semantics of NP coordination, however.
When NP objects are coordinated, only one relation gets mentioned
in the Indexed MRS report. Take the following sentence for example:

8) minA  syO-n     lasi-a     ja  tomaatti-a
   i+Nom eat+1p+Sg glass+Part and tomato+Part
   I eat glass and tomatoes

The Indexed MRS ends up looking like:

9) <h1, e2:SEMSORT:TENSE:ASPECT:MOOD,
    {h3:pronoun_n(x4:SEMSORT:BOOL:FIRST:SG),
     h5:pronoun_q(x4, h6, h7),
     h8:_eat_v(e2, x4, x9:SEMSORT:PERSON:NUMBER:BOOL),
     h10:_glass_n(x11:SEMSORT:PERSON:NUMBER:BOOL)},
    {h6 qeq h3}>

A similar thing is happening in coordinating VPs. The following
sentence illustrates:

10) minA  syO-n     lasi-a     ja  kAvele-n
    i+Nom eat+1p+Sg glass+Part and walk+1p+Sg
    I eat glass and walk

The Indexed MRS for (10) looks the same as (9).

Update: This problem was fixed when I ran through some lexical
rules and constrained their C-CONT [ RELS, HCONS ] to be <! !>.
The proper Indexed MRS representations appear to be showing
up now.

One problem that still happens