Closed ButteredGroove closed 5 years ago
Another thought. Is the repo missing the unsupervised alignment code? The README.md says:
The code of unsupervised aligning is adapted from this repository.
However, data/unsupervised_align is empty.
Thank you so much for telling me about this issue.
Yes, you are right. I just find out this repository does not include all the code and it is not the latest version. The sad thing is my computer collapsed last week and I lost some of the codes.
I will fix this issue in these two weeks since I need to rewrite some of the codes. I am so sorry for the inconvenience caused.
You're welcome. I'm sorry to hear about your computer, and I hope you restore it quickly.
No apologies necessary. I'm happy to help. I'll try it out when you make your update.
Hi Cartus, Hope all is well. How goes the update? Let me know if I can do anything to help.
Thanks for asking. Currently, I am rushing to meet the EMNLP deadline. I will continue fixing these bugs after the ddl. Sorry again. I will let you know as soon as I finish the update.
Hi Cartus, Hope all is well. How goes the update? Let me know if I can do anything to help.
Hi, I have updated the repo. I guess it can be run now.
I managed to get past the JAMR alignment step.
However, when I run ./align.sh I get:
I fixed this issue by making the eval directory myself:
mkdir -p unsupervised_align/eval
However, then I ran into these issues when I ran ./align.sh:
There are a couple of issues here:
The second point is easy enough to fix (I did
pip install nltk
), but not documented in the README.md. :)The first point I'm not sure about. My guess is it has something to do with MGIZA++, which generates a run.sh in the install directory. Unfortunately, AMR-Parser's README.md isn't specific regarding what to name the install folder for MGIZA++. I tried installing it in the mgiza subdirectory without success. I installed MGIZA++ thusly:
I haven't been able to find any reference to mgiza in the .sh nor .py scripts in data. But align.sh must want to do something with it.
What am I missing? Please help!
Another small correction: In the README.md the example arguments passed to train.py are train.pb.lemmas instead of train.txt.pb.lemmas.