[developers] Qs: How to check the feature structure under S
oe at csli.Stanford.EDU
Mon Feb 12 22:15:44 CET 2007
> 1) when looking at the feature structure for "S" from the parse chart
> (by clicking "Show chart"), I found it successfully contains all the
> ORTH values in its feature.
> 2) However, looking at the same feature structure from the pop-up
> parse tree (by right-clicking and selecting "Show enlarged tree"), I
> just found it lacking some of the values in the ORTH feature. The
> missing ORTH values seem to involve inflectional change. For example,
> "the dog barks" fails to include "barks" value in the ORTH.
> Which should I look at?
to our embarrassment, this appears to be a bug in the LKB software: it
would seem that the effects of applying an inflectional rule fail to be
recorded appropriately. the distinction between the AVMs you see from
the chart window and those off the parse trees is subtle: when parsing,
the system eliminates (`restricts') the value of ARGS on each new edge.
recall that ARGS records the immediate daughters, and that information
is available in the parse chart itself. hence, cutting out ARGS does
not loose any information but facilitates better efficiency in parsing.
when constructing the parse tree for each result, on the other hand, it
is desirable to provide the full AVM, including ARGS. to achieve this,
the system actually rebuilds the derivation, i.e. goes back to the full
AVMs of all lexical entries and rules involved and then replays all the
unifications. obviously, when replaying an inflectional rule, it will
be important to remember what the ORTH value (computed from the %suffix
annotation of the rule) should be, as that too has to be replayed. at
present, the LKB fails to do this latter bit.
for our current exercise, please rely on the AVMs that you can access
from the parse chart window. i will try releasing a corrected version
of the LKB early this week.
best - oe
+++ Universitetet i Oslo (IFI); Boks 1080 Blindern; 0316 Oslo; (+47) 2284 0125
+++ CSLI Stanford; Ventura Hall; Stanford, CA 94305; (+1 650) 723 0515
+++ --- oe at csli.stanford.edu; oe at ifi.uio.no; stephan at oepen.net ---
More information about the developers