Introduction

onf-parser is a lightweight pure Python library for parsing the OntoNotes Normal Form format (.onf -- cf. section 6.3).

Installation

Note that Python >=3.7 is required due to our dependency on dataclasses.

pip install onf-parser

Usage

There are three top-level functions:

from onf_parser import parse_files, parse_file, parse_file_string
# read a single file
sections = parse_file('ontonotes/some/file.onf')
# or parse a raw string
sections = parse_file_string(s)
# read all .onf files in a single directory
files = parse_file('ontonotes/')

For each file, a list of Section objects (which correspond to documents for the purposes of annotation) will be available:

for filepath, sections in files:
    for section in sections:
        coref_chains = section.chains
        for chain in coref_chains:
            print(chain.type)
            print(chain.id)
            print(chain.mentions)
            for mention in chain.mentions:
                print(mention.sentence_id)
                print(mention.tokens)
        for sentence in section.sentences:
            print(sentence.plain_sentence)
            print(sentence.plain_sentence.string)

            print(sentence.treebanked_sentence)
            print(sentence.treebanked_sentence.string)
            print(sentence.treebanked_sentence.tokens)

            print(sentence.speaker_information)
            print(sentence.speaker_information.name)
            print(sentence.speaker_information.start_time)
            print(sentence.speaker_information.stop_time)

            print(sentence.tree)
            print(sentence.tree.tree_string)

            print(sentence.leaves)
            for leaf in sentence.leaves:
                print(leaf.token)
                print(leaf.token_id)

                # NER
                print(leaf.name)
                print(leaf.name.type)
                print(leaf.name.token_id_range)
                print(leaf.name.tokens)

                # Coreference
                print(leaf.coref)
                print(leaf.coref.type)
                print(leaf.coref.token_id_range)
                print(leaf.coref.tokens)

                # WordNet sense
                print(leaf.sense)
                print(leaf.sense.label)

                # PropBank
                print(leaf.prop.label)
                print(leaf.prop)
                for arg_label, arg_spans in leaf.prop.args.items():
                    print(arg_label)
                    for arg_span in arg_spans:
                        print(arg_span)

See model classes for more information.