[developers] Using unbound variables in MRS

[Michael Wayne Goodman]

Hi again,

I've come across two instances where MRS -> DMRS conversion is lossy
due to the use of unbound variables (e.g. for dropped arguments). My
aim is not to point out a deficiency in DMRS, but rather to ask: what
are we allowed to do with unbound variables?

The first problem is when the unbound variables are co-indexed. In
Japanese, serial verb constructions are constrained such that they
share their arguments (e.g. ARG1 and ARG2), but Japanese frequently
drops both args, so we have a constraint between two unexpressed
individuals. The attached image shows the MRS for "tabe sugite ha
ikenai" ("don't overeat"), where both verbs taberu ("eat") and sugiru
("exceed") have "ARG1: i6 ARG2: i7". This constraint is lost is DMRS
because there are no nodes for i6 or i7, and thus no links for these
arguments. We could say that Jacy should have zero-pronouns for these
arguments, but that's exactly what the grammar has moved away from
(for other reasons).

The second, related problem is in expressing ICONS. Because ICONS can
occur on unexpressed individuals, they could be lost in DMRS. But
since it's unclear how ICONS would be expressed in DMRS at all, lets
set aside this problem for now.

Are we abusing these unbound variables by constraining them so, or is
this a legitimate use? Do others place constraints on them in their
grammars?

-- 
-Michael Wayne Goodman
-------------- next part --------------
A non-text attachment was scrubbed...
Name: tabesugite-wa-ikenai.png
Type: image/png
Size: 8410 bytes
Desc: not available
URL: <http://lists.delph-in.net/archives/developers/attachments/20140825/c1e1221b/attachment.png>