Yargy

Yargy is a Earley parser, that uses russian morphology for facts extraction process, and written in pure python

Install

Yargy supports both Python 2.7+ / 3.3+ versions including PyPy.

$ pip install yargy

Usage

from yargy import Parser, fact, rule
from yargy.predicates import gram
from yargy.pipelines import MorphPipeline


Person = fact(
    'Person',
    ['position', 'name']
)
Name = fact(
    'Name',
    ['first', 'last']
)


class PositionsPipeline(MorphPipeline):
    grammemes = {'Position'}
    keys = [
        'премьер министр',
        'президент'
    ]


NAME = rule(
    gram('Name').interpretation(
        Name.first.inflected()
    ),
    gram('Surn').interpretation(
        Name.last.inflected()
    )
).interpretation(
    Name
)
PERSON = rule(
    gram('Position').interpretation(
        Person.position.inflected()
    ),
    NAME.interpretation(
        Person.name
    )
).interpretation(
    Person
)


parser = Parser(PERSON, pipelines=[PositionsPipeline()])
text = '''
12 марта по приказу президента Владимира Путина ...
'''
for match in parser.findall(text):
    print(match.fact)

And in output you will see something like this:

Person(position='президент', name=Name(first='владимир', last='путин'))

For more examples, details on grammar syntax, predicates and pipelines see Yargy documentation.

License

Source code of yargy is distributed under MIT license (allows modification and commercial usage)

Changelog

0.9 (dev)

Major update required to support recursive grammar.

Library API changed including grammar DSL and parser API
Components updated: tokenizer, pipelines, parser, interpretation

0.6.0

Initial object interpretation support
Replaced custom tree-like struct used in combinator.resolve_matches method with IntervalTree
number_match and case_match labels now have understand same arguments as gnc_match label

0.5.1 - 0.5.3

Support match_all_disambiguation_forms argument in gnc_match label
New token types - ROMN for roman numbers, like XXI, EMAIL for emails and PHONE for phone numbers
New labels - and_ & or_
Implemented get_normalized_text function that returns normalized text from parsed tokens

0.5.0

Partial morphology disambiguation solving support (gnc_match label now accepts optional boolean argument solve_disambiguation, which when is True, reduces number of token forms in result match)
Rewrited labels, now they're function-based
Rewrited tokenizer's transform function for better extending
Tokenizer now adds different types of grammemes for different types of quotes (e.g. L-QUOTE for « quote)
Implemented DAWG-based pipeline, which shows better performance over dictionary-based pipeline

0.4.1 - 0.4.6

Reimplemented resolve_matches method in Combinator
[fix] Fixed error at parsing float range with comma as delimiter
[fix] Additional checks for terminal rule at reduce grammars method
[fix] Fixed requirements in setup.py
[fix] Tokenizer now correctly understands range values on Python2.x and PyPy platforms
[fix] Create new grammars with terminal rule instead of appending it to original one

0.4

Replaced shift-reduce parser with GLR parser, because it provides much more performance on multiple grammars (linear time with GLR vs. exponental time with shift-reduce parser).
Pipelines support
Added support for Python 2.7 & PyPy

0.3

Implemented gnc-match, in labels

0.2

Implemented gender-match, number-match, case-match labels
Replace is-title label with is-capitalized label, due to http://bugs.python.org/issue7008
Tokenizer now understands integer and float ranges (actually, two numbers separated by dash)

0.1

Initial release

stcoder/yargy