[lkb] dummy qeq relations lead to big memory consumption during generation
Xuchen Yao
xuchen at coli.uni-saarland.de
Sun Aug 15 12:35:06 CEST 2010
Hello,
I woke up in the morning and found the LOGON/LKB generator hung on one
MRS generation and ate up to 14 GB of memory. The reason turned out to
be that I forgot to clean up the HCONS list in a modified MRS.
For instance, the original sentence is:
Corn, rice, fruit, cotton, hay, tobacco, and sugar are some of the crops
that are grown near old rivers.
My program replaced the coordination phrase with "what" to produce a
"what" question:
What are some of the crops that are grown near old rivers?
But it didn't remove all the qeq relations introduced by "Corn, rice,
fruit, cotton, hay, tobacco, and sugar". Then the generation time seemed
to take forever...
I'd be very interested to know why this happened: if a qeq relation "h1
qeq h2" doesn't exist in an MRS, how can it cause a huge memory
footprint? I thought they should be discarded automatically by LKB --
though it's my fault not to do it from my codes.;-)
The sample MRX file is attached in case someone's interested. Generation
can be invoked by:
(lkb::generate-from-mrs (mrs::read-single-mrs-xml-file "/tmp/t.xml"))
Xuchen
-------------- next part --------------
A non-text attachment was scrubbed...
Name: t.xml
Type: text/xml
Size: 8548 bytes
Desc: not available
URL: <http://lists.delph-in.net/archives/lkb/attachments/20100815/655e2c9e/attachment.xml>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: t.mrs
URL: <http://lists.delph-in.net/archives/lkb/attachments/20100815/655e2c9e/attachment.ksh>
More information about the lkb
mailing list