deletion of duplicate predicates

  • Raveesh

    Raveesh - 2009-10-15

    Hi Mike,

    In OpenCCG there is a piece of code that deletes duplicate predicates in a LF. HyloHelper.compactAndConvertNominals(cat.getLF(), index);
    Can you suggest what led to that functionality? I have a LF, which has Owner feature twice (at predicate level). More specifically, an utterance with theme-structures separated by a rheme-structure, as in

    the box-L+H* LH% is-H* LL% blue-L+H* LH%

    I know that an LF with duplicate features looks weird, but I guess my approach needs that.

    Was wondering if this could be avoided, or is there a way to not sort the features in LF…



  • Michael White

    Michael White - 2009-10-16

    Duplicate predications complicate realization, and I found them to be both unnecessary and likely errors.  When you really seem to want duplicates, I suspect it would be better to have them on different nominals.  For example, with MRS, they support bags for phrases like "big big car"; but with HLDS, each "big" would get its own nominal.  With your example, I don't see why the owner features would not be attached to the different sem heads.

  • Raveesh

    Raveesh - 2009-10-16

    I got your point. But my implementation - within the constraints of DotCCG platform - makes certain obvious things complicated. Unlike the Flights grammar I can't use LF valued syntactic features to easily distribute prosodic semantic features such as Ownership and pitch Accents. In my implementation Ownership is a semantic feature of boundary tones.  When a boundary combines with an accent bearing phrase, this feature is added to the LF of that phrase. Moreover, since boundaries are defined as of type s$\s$ the nominal to which Owner feature would thus be attached is always the main predicate ("is" in this case).  As a consequence, all the Owner feature from various boundaries goes to the main predicate. (This is what the OpenCCG platform offers). Up till now, we have been resorting to split this Ownership feature as ThemeOwner and RhemeOwner, and it seemed to make sense, and also works. But only to fail in the aforementioned utterance (multiple rheme/theme units). Hmm!

  • Michael White

    Michael White - 2009-10-16

    It seems to me that by attaching both the theme and rheme features to the main predicate, you're losing track of what's the theme and what's the rheme.

    One solution might be to upgrade dotccg to support lf-valued syn feats.


Get latest updates about Open Source Projects, Conferences and News.

Sign up for the SourceForge newsletter:

No, thanks