Templated Open Information Extraction

• Introduction
• Prerequisites
• Installation
• Running
• Output format
• Examples
• Users of template-oie

Introduction

Extracts templated Open Information Extraction: allowing for diverse, non-contiguous, multi-word predicates, while keeping the arguments short and useful for downstream applications.

For example, given the sentence:

Under the agreement with the House and Senate leaders , the minimum wage would rise from the current $ 3.35 an hour to $ 4.25 an hour.

One of the extractions is:

• Under {A0} {A1} would rise from {A2} to {A3}

    A0:	the agreement
    A1:	the minimum wage
    A2:	$ 3.35 an hour
    A3:	$ 4.25 an hour
  

Note that that the head of the predicate (rise) is also identified.

Prerequisites

1. python 2.7
2. pip 9.x

Installation

1. Install required packages:
   

pip install -r ./requirements.txt

2. Download spaCy English models:
   

python -m spacy download en

Running

Usage:
    prop_extraction --in=INPUT_FILE --out=OUTPUT_FILE [--id]

Extract propositions from a given input file, output is produced in separate output file.
If both in and out paramaters are directories, the script will iterate over all *.txt files in the input directory and
output to *.prop files in output directory.

Options:
   --in=INPUT_FILE      The input file, each sentence in a separate line
   --out=OUTPUT_FILE    The output file, Each extraction in a tab separated line, each consisting of original sentence,
   predicate template, lemmatized predicate template,argument name, argument value, ...
   --id                 Indicate that the input file is composed of sentence-id \t sentence, and copy this id in the output.

Output format

Each extraction is presented in a tab separated line, consisting of:

1. Original sentence
   Words are separated by a single space, chunks by double spaces.
2. Index of the main predicate's chunk
3. Predicate template
4. Lemmatized predicate
5. Argument1 name
6. Argument1 value
7. ...

Examples

See the example folder for the output over more than 3K sentences from the news domain.

python ./prop_extraction.py --in=../examples/sentences.txt --out=../examples/sentences.prop

Users of template-oie

The following projects make use of template-oie:

• Open Knowledge Representation
• Chirps