Wiki NLP Tools

Python package to perform language-agnostic tokenization.

Vision

• Researchers can start with a Wikipedia article (wikitext or HTML), strip syntax to leave just paragraphs of plaintext, and then further tokenize these sentences into sentences and words for input into models.
• This would be language-agnostic – i.e. the library would work equally well regardless of Wikipedia language. https://meta.wikimedia.org/wiki/List_of_Wikipedias
• This would be easily accessible – i.e. each component is a open-source, pip-installable Python library that is configurable but provides good default performance out-of-the-box that Wikimedia could use internally via PySpark UDFs on our cluster and external organizations/researchers could incorporate into their workflows.
• The connections between states are transparent – i.e. for any text extracted in word tokenization, that text can be connected directly back to the original wikitext or HTML that it was derived from.

Features

• Tokenize text to sentence and words across 300+ languages out-of-the-box
• Abbreviations can be used to improve performances
• Word-toknizer takes non-whitespace delimited languages into account during tokenization
• Input can be exactly reconstructed from the tokenization output

Installation

$ pip install mwtokenizer

Basic Usage

from mwtokenizer.tokenizer import Tokenizer
# initiate a tokenizer for "en" or English
tokenizer = Tokenizer(language_code = "en")
sample_text =  '''Have Moly and Co. made it to the shop near St. Michael's Church?? \n\t The address is written by Bohr Jr.   here!'''
print(list(tokenizer.sentence_tokenize(sample_text, use_abbreviation=True)))
'''
[output] ["Have Moly and Co. made it to the shop near St. Michael's Church?? \n\t ", 'The address is written by Bohr Jr.   here!']
'''
print(list(tokenizer.word_tokenize(text=sample_text, use_abbreviation=True)))
'''
[output] ['Have', ' ', 'Moly', ' ', 'and', ' ', 'Co.', ' ', 'made', ' ', 'it', ' ', 'to', ' ', 'the', ' ', 'shop', ' ', 'near', ' ', 'St.', ' ', "Michael's", ' ', 'Church', '??', ' \n\t ', 'The', ' ', 'address', ' ', 'is', ' ', 'written', ' ', 'by', ' ', 'Bohr', ' ', 'Jr.', '   ', 'here', '!']
'''

Project Information

• Licensing
• Repository
• Issue Tracker
• Contribution Guidelines
• Benchmarking
• Resource Generation: Abbreviations & Benchmarking data + Sentencepiece Corpus and Training