Coreference

This Gem provides coreference resolution for various languages such as English and Spanish.

The CorefGraph-en module provides an implementation of the Multi-Sieve Pass system for for Coreference Resolution system originally proposed by the Stanford NLP Group (Raghunathan et al., 2010; Lee et al., 2011) and (Lee et al., 2013). This system proposes a number of deterministic passes, ranging from high precision to higher recall, each dealing with a different manner in which coreference manifests itself in running text.

Although more sieves are available, in order to facilitate the integration of the coreference system for the 6 languages of OpeNER we have included here 4 sieves: Exact String Matching, Precise Constructs, Strict Head Match and Pronoun Match (the sieve nomenclature follows Lee et al (2013)). Furthermore, as it has been reported, this sieves are responsible for most of the performance in the Stanford system.

The implementation is a result of a collaboration between the IXA NLP (http://ixa.si.ehu.es) and LinguaMedia Groups (http://linguamedia.deusto.es).

Requirements

  • Ruby 1.9.2 or newer
  • Python 2.7 or newer
  • Pip 1.3.1 or newer

Installation

Installing as a regular Gem:

gem install opener-coreference-base

Using Bundler:

gem 'opener-coreference-base',
  :git    => '[email protected]:opener-project/coreference-base.git',
  :branch => 'master'

Using specific install:

gem install specific_install
gem specific_install opener-coreference-base \
   -l https://github.com/opener-project/coreference-base.git

Usage

To run the program execute:

coreference-base -l (de|en|es|fr|it|nl) -i input.kaf

Corefgraph will output KAF via standard output with the clusters added to the KAF input received. Note that for the full functionality of CorefGraph you will need to provide the elements with the heads of (at least) the Noun Phrases marked, as it can be seen in the treebank input examples in the resource/examples directory. If you do not provide heads, only Exact String Match will work properly, whereas Precise Constructs, Strict Head Match and Pronoun Match will not.

For a full explanation of how the Multi Sieve Pass system works see documentation in resources/doc.

Adapting CorefGraph-en to your language

There are a number of changes needed to be made to make CorefGraph works for other languages. Although we have try to keep the language dependent features to a minimum, you will still need to create some dictionaries for your own language and make some very minor changes in the code. Here is the list of very file in the Corefgraph module that needs to be changed. Every change except one (see below) to be done in the $project/core/corefgraph/resources directory:

  • dictionaries/$lang_determiners.py
  • dictionaries/$lang_pronouns.py
  • dictionaries/$lang_verbs.py
  • dictionaries/$lang_stopwords.py

  • dictionaries/$lang_temporals.py

  • tagset/$TAGSETNAME_pos.py

  • tagset/$TAGSETNAME_constituent.py

  • files/animate/$lang.animate.txt

  • files/animate/$lang.inanimate.txt

  • files/demonym/$lang.txt

  • files/gender/$lang.male.unigrams.txt

  • files/gender/$lang.female.unigrams.txt

  • files/gender/$lang.neutral.unigrams.txt

  • files/gender/$lang.namegender.combine.txt

  • files/gender/$lang.gender.data

  • files/number/$lang.plural.unigrams.txt

  • files/number/$lang.singular.unigrams.txt