[developers] FSC input to cheap

Prescott Klassen klassp at uw.edu
Mon Sep 26 22:25:55 CEST 2011


Dear PET Developers,

I would like to use a Feature Structure Chart to input token/POS to cheap,
however, my application currently uses the Stanford Parser and I would
prefer to use it instead of TNT to create a lattice of tokens. I believe TNT
and the Stanford Parser share a similar set of possible POS tags.

Could I simply "pretend" I am using TNT and create a lattice from the
Stanford Parser XML output similar to the sample on the PetInputFsc Wiki
Page--using a +TNT feature and f type="tnt" structure to wrap my +TAGS/+PRBS
feature pairs? Would I get ERG integration for free by doing this--simply
leveraging the work already done to integrate TNT POS mapping to lexical
items? Or perhaps I have misunderstood how TNT Token/POS is integrated with
the ERG.

I experimented a bit with this using the sample lattice and discovered that
if I defaulted to

<f name="+PRBS" org="list"><str>1.000000e+00</str></f>

for all of my tags it did not result in a parse using the command line:

cat alt_input.fsc | cheap -tok=fsc -default-les=all -cm -packing -mrs=new
-results=1 /home/prescott/delphin/erg/english.grm

Where alt_input.fsc is the sample lattice where all +TAGS/+PRBS lists
contain only one pair for each token and +PRBS is set to 1.000000e+00

Thank you,

Prescott Klassen
University of Washington
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.delph-in.net/archives/developers/attachments/20110926/95f972eb/attachment.html>


More information about the developers mailing list