[developers] Treebanking and training with FFT

[Petter Haugereid]

Hi everybody,

I have been trying over some days to make treebanking work with FFT.
Following instructions on the DELPH-IN site, I have given the commands
below, and I end up with a browser window with the items of the profile I
attempt to treebank. However, when I click on one of the items, I get an
error message "404 Not Found". Do any of you know what I am doing wrong?

Here are the commands (with full paths):
mkprof -s ~/logon/lingo/lkb/src/tsdb/home/mrs.2020.05.12/ /tmp/mrs-test
art -f -a '~/tools/ace-0.9.30/ace --disable-generalization -g
~/tools/ace-0.9.30/norwegian-small.dat -O' /tmp/mrs-test
~/acetools-x86-0.9.30/fftb -g ~/tools/ace-0.9.30/norwegian-small.dat
 --browser --webdir ~/acetools /tmp/mrs-test/

I am quite keen to get a statistical model for my grammar, so I have tried
to train a model from a small treebank which I have disambiguated with the
logon tool. When I try to train with LOGON, only get a lot of garbage
collection messages, and I eventually have to kill the process. When I try
to train with FFT with the following commands, I get the messages below:

mkprof -s ~/logon/lingo/lkb/src/tsdb/home/mrs.2020.05.12/ /tmp/mrs-test/
art -a '~/tools/ace-0.9.30/ace -g ~/tools/ace-0.9.30/norwegian-small.dat
-O' -f /tmp/mrs-test/
FFGRANDPARENT=0 ~/acetools-x86-0.9.30/ffmaster 1 mrs-test-gp0.mem &
FFGRANDPARENT=0 ~/acetools-x86-0.9.30/ffworker
~/tools/ace-0.9.30/norwegian-small.dat /tmp/mrs-test/
~/logon/lingo/lkb/src/tsdb/home/mrs.2020.05.12/ localhost

# loading /tmp/mrs-test/...
# loading /home/petter/logon/lingo/lkb/src/tsdb/home/mrs.2020.05.12/...
# loading gold
#  ... iid 1 -- gold tree 1 / 1 not in parse forest
#  ... iid 2 -- gold tree 1 / 1 not in parse forest
#  ... iid 3 -- gold tree 1 / 1 not in parse forest
...
#  ... iid 68 -- gold tree 1 / 1 not in parse forest
#  ... iid 69 -- gold tree 1 / 1 not in parse forest
# loaded 0 ambiguous feature forests with gold trees
# [1]+  Exit 255                FFGRANDPARENT=0
~/acetools-x86-0.9.30/ffmaster 1 mrs-test-gp0.mem
# Floating point exception (core dumped)

I tried the same commands with the ERG MRS treebank in LOGON, and I was
able to train a model with it. I suspect the reason I don't succeed, is
that I have treebanked with LOGON, while Dan has used FFT.

Here are links to
1) the MRS treebank
 https://www.dropbox.com/s/7mj53j644vwhbes/mrs.2020.05.12.tgz?dl=0
2) The Norwegian MRS items I have treebanked
https://www.dropbox.com/s/qfhuqwnxlz0e1pb/mrs.txt?dl=0
3) The Norsyg grammar (loading 'lkb/small-script' with the LKB,
'ace/config-small.tdl' with ACE is sufficient)
https://www.dropbox.com/s/rmoy6q40dvz1dxh/norsyg.20-05-13.tgz?dl=0
4) A compiled version of the grammar, compiled with ace-0.9.30
https://www.dropbox.com/s/cb0dq9omuhojlmv/norwegian-small.dat?dl=0

If someone can point me to what I am doing wrong, I would be very greatful!

Best,

Petter
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.delph-in.net/archives/developers/attachments/20200513/b3144d7f/attachment.html>