|
From: Hitzeman, J. M. <hi...@mi...> - 2008-09-04 11:58:02
|
I am not convinced that changing "NR" to a path is a good idea, mainly
because PATHs seem directional, while LINKs seem positional, e.g., "the
road from Belmont to San Mateo" or even "walking towards Belmont" both
strike me as directional, while "the Hillsdale Mall in San Mateo" or
"San Mateo, near San Francisco" strike me as lacking movement and
therefore describe the position rather than the path. That's the set
of definitions I've been taking as an annotator, and it has helped me
to have those types of intuitions concerning paths v. links.
Janet
Janet Hitzeman, Ph.D.
Senior Scientist
The MITRE Corporation
781-271-8246
-----Original Message-----
From: spa...@li... on behalf of
Mani, Inderjeet
Sent: Wed 9/3/2008 9:08 PM
To: spa...@li...
Subject: [Spatialml-discussion] Proposed changes and clarifications
relatedto LINK tag
Dear SpatialMLers,
Several of us have been informally discussing changes to the LINK tag
in SpatialML, ever since a discussion on the subject that originated at
the LREC workshop in Marrakesh.
Specifically, in the LINK tag, we would like to remove the 'linkType'
attribute value NR, for "near". NR discretizes a distance metric, and
as such doesn't belong with the other RCC8-inspired link types. It
belongs instead in the PATH relation, where it can be hospitably
accommodated by extending the 'distance' attribute value. So, "Belmont,
near San Mateo" would have a PATH tag with a source and a target and
distance="NR".
We would also like to change the expansion of EC to be "external
connection", rather than "extended connection", since the former is the
correct RCC8 terminology. Likewise, DC should expand to "disconnected"
rather than "discrete connection".
Regarding the provenance of the links, here is a clarification
statement, that should perhaps be added to the guidelines. DC, EC, EQ,
and PO are from RCC8. IN is not RCC8, but collapses two RCC8 relations,
TPP and NTPP (tangential proper part and non-tangential proper part,
respectively). The reason for the collapsing is that it is often
difficult for annotators to decide whether the part's region touches or
doesn't touch the container's. Finally, we don't include the remaining
RCC8 inverse links TPPi and NTPPi from RCC8, since these can be
represented in annotation by swapping arguments, and are in addition
likely to confuse annotators.
These changes leave SpatialML with 5 link types: DC, EC, EQ, PO, and
IN. The result is not of course RCC8 (which has DC, EC, EQ, PO, TPP,
NTPP, TPPi and NTPPi), but nor is it RCC5 (which has DR, EQ, PO, PP and
PPi). While IN is in fact the PP "proper part" relation in RCC5, DC is
different from DR ("discrete from") -- the latter just means the two
regions don't overlap. Also, SpatialML has RCC8's EC as well, and
doesn't of course have RCC5's PPi. So, the 5 'linkType' attribute
values in SpatialML are a proper subset of the relations in RCC5 and
RCC8. I am not sure where these RCC "impoverishments" leave us in terms
of qualitative reasoning capabilities.
Any comments or suggestions are of course welcome.
Cheers,
Inderjeet.
-----------------------------------------------------------------------
--
This SF.Net email is sponsored by the Moblin Your Move Developer's
challenge
Build the coolest Linux based applications with Moblin SDK & win great
prizes
Grand prize is a trip for two to an Open Source event anywhere in the
world
http://moblin-contest.org/redirect.php?banner_id=100&url=/
_______________________________________________
Spatialml-discussion mailing list
Spa...@li...
https://lists.sourceforge.net/lists/listinfo/spatialml-discussion
|