A command-line wrapper to query textual entailment and paraphrasing systems. Currently wraps the EXCITEMENT Open Platform web demo.
Given two text fragments called 'Text' and 'Hypothesis', Textual Entailment Recognition is the task of determining whether the meaning of the Hypothesis is entailed (can be inferred) from the Text.
To install, run:
python setup.py install
If pip fails to install lxml, you might need to run sudo apt-get install libxml2-dev libxslt1-dev.
Usage summary:
usage: excitement [-h] [-m model] [-o out_file] test_file
Specify your test set as a comma-separated file of text, hypothesis.
John is in love with Mary,John likes Mary
John is in the park,John is not in his house
John is tall,John is a man
The output format is a comma-separated file of text, hypothesis, entailment relationship, and confidence score. If an output file is not specified, excitement will print to the console.
John is in love with Mary,John likes Mary,Entailment,0.9813405993
John is in the park,John is not in his house,Entailment,0.240675551189
John is tall,John is a man,Entailment,0.994980828692
You can specify one of these models using the --model flag:
| Code | Model |
|---|---|
| edit-distance | ALG:EditDistance COMP:FixedWeightLemma |
| edit-distance-wordnet | ALG:EditDistance COMP:FixedWeightLemma RES: WordNet |
| maxent-verbocean | ALG:MaxEntClassification COMP:TreeSkeleton RES:VerbOcean,TreePattern |
| maxent-wordnet | ALG:MaxEntClassification COMP:TreeSkeleton RES:WordNet,TreePattern |
| maxent-all | ALG:MaxEntClassification COMP:TreeSkeleton RES:WordNet,VerbOcean,TreePattern |
| biutee | ALG:BIUTEE RES:WordNet,CatVar,BAP |
| pieda | ALG:P1EDA RES:Paraphrase Table |
The default model is maxent-all.