[pet] item traits

Peter Adolphs Peter.Adolphs at dfki.de
Thu Aug 20 18:23:25 CEST 2009


Hi!

I find the current encoding of item types confusing. On the one hand we have a
class hierarchy of tItem, tInputItem, tLexItem and tPhrasalItem, on the other
hand we have traits INPUT_TRAIT, INFL_TRAIT, LEX_TRAIT and SYNTAX_TRAIT. These
traits are used to flag both rules and items, but INPUT_TRAIT isn't used for
rules and INFL_TRAIT isn't used for items. It also seems that the item traits
just repeat the class hierarchy information, although tPhrasalItem also gets
LEX_TRAIT at times (namely as the output of 'inflectional rules', that is
lexical rules with morphographemic conditions, having INFL_TRAIT). And then it
seems that the only tests on item traits are whether an item is an input item or
not. In the other cases (e.g. in the ItemPrinters), the class information is
exploited.

So why do we need item traits? Couldn't we test for input items by checking the
item's type dynamically (dynamic_cast<tInputItem*>(i) == 0 or using typeid())?
Furthermore I don't see where LEX_TRAIT or SYNTAX_TRAIT for items is ever used.
What are they capturing anyway? 'Morphographology' vs syntax?

Or why do we need an item class hierarchy? For storing the surface string of
input items, stems for lexical items and the corresponding rule for phrasal
items or for deciding whether an item is an input, lexical or phrasal item in
the item printers? Could these be regarded as decorations of an item? (Another
case is a flag whether an item resulted from a robust PCFG rule application.)

Any opinions? (or decisions?)

Cheers,

Peter

-- 
Peter Adolphs
DFKI GmbH, Projektbuero Berlin, Alt-Moabit 91c, D-10559 Berlin
Telefon: 030 / 3949-1826
_______________________________________________________________

Deutsches Forschungszentrum fuer Kuenstliche Intelligenz GmbH
Firmensitz: Trippstadter Strasse 122, D-67663 Kaiserslautern

Geschaeftsfuehrung:
Prof. Dr. Dr. h.c. mult. Wolfgang Wahlster (Vorsitzender)
Dr. Walter Olthoff

Vorsitzender des Aufsichtsrats:
Prof. Dr. h.c. Hans A. Aukes

Amtsgericht Kaiserslautern, HRB 2313



More information about the pet mailing list