Wrapper for the parser by Zhang et al. (2021)

The source.

Docker

Docker image for a remote isanlp processor.

Local run

1. Download the weights into data/models_saved/model.pth and data/models_saved/xl_model.pth. Link 1 (Baidu), link 2 (Dropbox).
2. Configure the environment: source setup_environment.sh. If necessary, install the environment as a jupyter kernel: pip install ipykernel && python -m ipykernel install --name "rst_parser_zhang"
3. Use in the code:

from isanlp import PipelineCommon
from isanlp.processor_udpipe import ProcessorUDPipe
from isanlp_processor import ProcessorRST


ppl = PipelineCommon([
    #(ProcessorSpaCy(model_name='en_core_web_sm', morphology=True, parser=False, ner=False, delay_init=False),
    (ProcessorUDPipe(model_path='data/english-ewt-ud-2.5-191206.udpipe', parser=False),
     ['text'],
     {'tokens': 'tokens',
      'sentences': 'sentences'}),
    (ProcessorRST(),
     ['text', 'tokens', 'sentences'],
     {0: 'rst'})
])

some_text = "Brown fox jumped over the tree because it had to."
result = ppl(some_text)

>>> print(result['rst'][0])
id: 1
text: Brown fox jumped over the tree because it had to.
proba: 1.0
relation: Explanation
nuclearity: NS
left: Brown fox jumped over the tree
right: because it had to.
start: 0
end: 49