xiuzhizheng / zero-shot-semantic-parsing Goto Github PK

Code for the zero-shot semantic parser described in our EMNLP 2018 paper.

The structure mapper implementation is an extension of this code.

1. Install Miniconda2
2. Install Stanford CoreNLP:

$ wget http://nlp.stanford.edu/software/stanford-corenlp-full-2016-10-31.zip
$ unzip stanford-corenlp-full-2016-10-31.zip 

3. Install python dependencies:

$ conda install --file reqs_conda.txt
$ pip install -r reqs_pip.txt

To delexicalize data for all domains run and prepare cross domain splits use:

$ python src/py/zero_shot/preprocess.py 

To run one of the models implemented in the paper use:

$ sh scripts/MODEL.sh SPLIT

Where: MODEL is one of the following: zero_shot, cross_lex, cross_lex_rep, in_abstract, in_lex. SPLIT is either test (the original train/test split of the OVERNIGHT dataset), or dev (in this case the original train set is split to 80%/20% train/test sets).

To run all models use:

$ sh scripts/run_all.sh

Results are saved to /res folder. To print all results use:

$ python src/py/zero_shot/print_res.py