A module to normalise non-standard words in text

pip install normalise==0.1.8


Title Logo

A module for normalising text.


This module takes a text as input, and returns it in a normalised form, ie. expands all word tokens deemed not to be of a standard type. Non-standard words (NSWs) are detected, classified and expanded. Examples of NSWs that are normalised include:

  • Numbers - percentages, dates, currency amounts, ranges, telephone numbers.
  • Abbreviations and acronyms.
  • Web addresses and hashtags.

Table of Contents

  1. Installation
  2. Usage
  3. Customise to your variety
  4. Input your own abbreviation dictionary
  5. Execute normalise from the command line
  6. Example
  7. Authors
  8. License
  9. Acknowledgements

1. Installation

normalise requires Python 3.

To install the module (on Windows, Mac OS X, Linux, etc.) and to ensure that you have the latest version of pip and setuptools:

$ pip install --upgrade pip setuptools

$ pip install normalise

If pip installation fails, you can try easy_install normalise.

normalise has several nltk data dependencies. Install these by running the following python commands:

import nltk
for dependency in ("brown", "names", "wordnet", "averaged_perceptron_tagger", "universal_tagset"):

2. Usage

Your input text can be a list of words, or a string.

To normalise your text, use the normalise function. This will return the text with NSWs replaced by their expansions:

text = ["On", "the", "28", "Apr.", "2010", ",", "Dr.", "Banks", "bought", "a", "chair", "for", "£35", "."]

normalise(text, verbose=True)

 'twenty-eighth of',
 'twenty ten',
 'thirty five pounds',

verbose=True displays the stages of the normalisation process, so you can monitor its progress. To turn this off, use verbose=False.

If your input is a string, you can use our basic tokenizer. For best results, input your own custom tokenizer.

from normalise import tokenize_basic
normalise(text, tokenizer=tokenize_basic, verbose=True)

In order to see a list of all NSWs in your text, along with their index, tags, and expansion, use the list_NSWs function:

from normalise import list_NSWs

({3: ('Apr.', 'ALPHA', 'EXPN', 'April'),
  6: ('Dr.', 'ALPHA', 'EXPN', 'Doctor')},
 {2: ('28', 'NUMB', 'NORD', 'twenty-eighth of'),
  4: ('2010', 'NUMB', 'NYER', 'twenty ten'),
  12: ('£35', 'NUMB', 'MONEY', 'thirty five pounds')}

i. Customise to your variety

In order to customise normalisation to your variety of English, use variety="BrE" for British English, or variety="AmE" for American English:

text = ["On", "10/04", ",", "he", "went", "to", "the", "seaside", "."]

normalise(text, variety="BrE")
Out: ['On', 'the tenth of April', ',', 'he', 'went', 'to', 'the', 'seaside', '.']

normalise(text, variety="AmE")
Out: ['On', 'the fourth of October', ',', 'he', 'went', 'to', 'the', 'seaside', '.']

If a variety is not specified, our default is British English.

ii. Input your own abbreviation dictionary

Although our system aims to be domain-general, users can input their own dictionary of abbreviations in order to tailor to a specific domain. This can be done using the keyword argument user_abbrevs={}:

my_abbreviations = {"bdrm": "bedroom",
                    "KT": "kitchen",
                    "wndw": "window",
                    "ONO": "or near offer"}

text = ["4bdrm", "house", "for", "sale", ",", "£459k", "ONO"]

normalise(text, user_abbrevs=my_abbreviations)

['four bedroom',
 'four hundred and fifty nine thousand pounds',
 'or near offer']

iii. Execute normalise from the command line

From the command line, you can normalise text from a given .txt file. Use the command normalise /path/to/your-file.txt. This will print the normalised output, as well as save it to a separate file "your-file_normalised.txt" in the same directory as the original text.

To specify the variety as American English, use --AmE (default is British English). For a verbose output, use --V:

$ normalise /path/to/your_file.txt --AmE --V

3. Example

A further example demonstrating the expansion of more types of NSW (including abbreviations, spelling mistakes, percentage ranges, currency):

text = ["On", "the", "13", "Feb.", "2007", ",", "Theresa", "May", "MP", "announced",
"on", "ITV", "News", "that", "the", "rate", "of", "childhod", "obesity", "had", "risen",
"from", "7.3-9.6%", "in", "just", "3", "years", ",", "costing", "the", "Gov.", "£20m", "."]

normalise(text, verbose=True)

 'thirteenth of',
 'two thousand and seven',
 'M P',
 'I T V',
 'seven point three to nine point six percent',
 'twenty million pounds',

4. Authors

Our system is described in detail in Emma Flint, Elliot Ford, Olivia Thomas, Andrew Caines & Paula Buttery (2016) - A Text Normalisation System for Non-Standard Words.

5. License

This project is licensed under the terms of the GNU General Public License version 3.0 or later.

Please see LICENSE.txt for more information.

6. Acknowledgements

This project builds on the work described in Sproat et al (2001).

We would like to thank Andrew Caines and Paula Buttery for supervising us during this project.

The font used for the logo was Anita Semi-Square by Gustavo Paz. License: Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)