Skip to main content

IPA tokeniser

Project description

A simple IPA tokeniser, as simple as in:

>>> from ipatok import tokenise
>>> tokenise('ˈtiːt͡ʃə')
['t', 'iː', 't͡ʃ', 'ə']
>>> tokenise('ʃːjeq͡χːʼjer')
['ʃː', 'j', 'e', 'q͡χːʼ', 'j', 'e', 'r']


tokenise(string, strict=False, replace=False, diphthongs=False, tones=False, unknown=False, merge=None) takes an IPA string and returns a list of tokens. A token usually consists of a single letter together with its accompanying diacritics. If two letters are connected by a tie bar, they are also considered a single token. Except for length markers, suprasegmentals are excluded from the output. Whitespace is also ignored. The function accepts the following keyword arguments:

  • strict: if set to True, the function ensures that string complies to the IPA spec (the 2015 revision); a ValueError is raised if it does not. If set to False (the default), the role of non-IPA characters is guessed based on their Unicode category (cf. the pitfalls section below).

  • replace: if set to True, the function replaces some common substitutes with their IPA-compliant counterparts, e.g. g → ɡ, ɫ → l̴, ʦ → t͡s. Refer to ipatok/data/replacements.tsv for a full list. If both strict and replace are set to True, replacing is done before checking for spec compliance.

  • diphthongs: if set to True, the function groups together non-syllabic vowels with their syllabic neighbours (e.g. aɪ̯ would form a single token). If set to False (the default), vowels are not tokenised together unless there is a connecting tie bar (e.g. a͡ɪ).

  • tones: if set to True, tone and word accents are included in the output (accent markers as diacritics and Chao letters as separate tokens). If set to False (the default), these are ignored.

  • unknown: if set to True, the output includes (as separate tokens) symbols that cannot be classified as letters, diacritics or suprasegmentals (e.g. -, /, $). If set to False (the default), such symbols are ignored. It does not have effect if strict is set to True.

  • merge: expects a str, str → bool function to be applied onto each pair of consecutive tokens; those for which the output is True are merged together. You can use this to, e.g., plug in your own diphthong detection algorithm:

    >>> tokenise(string, diphthongs=False, merge=custom_func)

tokenize is an alias for tokenise.

other functions

replace_digits_with_chao(string, inverse=False) takes an IPA string and replaces the digits 1-5 (also in superscript) with Chao tone letters. If inverse=True, smaller digits are converted into higher tones; otherwise, they are converted into lower tones (the default). Equal consecutive digits are collapsed into a single Chao letter (e.g. 55 → ˥).

>>> tokenise(replace_numbers_with_chao('ɕia⁵¹ɕyɛ²¹⁴'), tones=True)
['ɕ', 'i', 'a', '˥˩', 'ɕ', 'y', 'ɛ', '˨˩˦']

clusterise(string, strict=False, replace=False, diphthongs=False, tones=False, unknown=False, merge=None) takes an IPA string and lists its consonant and vowel clusters. The keyword arguments are identical as for tokenise:

>>> from ipatok import clusterise
>>> clusterise("kiaːltaːʃ")
['k', 'iaː', 'lt', 'aː', 'ʃ']

clusterize is an alias for clusterise.


When strict=True each symbol is looked up in the spec and there is no ambiguity as to how the input should be tokenised.

With strict=False IPA symbols are still handled correctly. A non-IPA symbol would be treated as follows:

  • if it is a non-modifier letter (e.g. Γ), it is considered a consonant;

  • if it is a modifier (e.g. ˀ) or a combining mark (e.g. ə̇), it is considered a diacritic;

  • if it is a modifier tone letter (e.g. ), it is considered a tone symbol;

  • if it is neither of those, it is considered an unknown symbol.

Regardless of the value of strict, whitespace characters and underscores are considered to be word boundaries, i.e. there would not be tokens grouping together symbols separated by these characters, even though the latter are not included in the output.


This is a standard Python 3 package without dependencies. It is offered at the Cheese Shop, so you can install it with pip:

pip install ipatok

or, alternatively, you can clone this repo (safe to delete afterwards) and do:

python test
python install

Of course, this could be happening within a virtualenv/venv as well.

other IPA packages

  • lingpy is a historical linguistics suite that includes an ipa2tokens function.

  • loanpy is another historical linguistics suite which uses ipatok as a dependency.

  • ipapy is a package for working with IPA strings.

  • ipalint provides a command-line tool for checking IPA datasets for errors and inconsistencies.

  • asjp provides functions for converting between IPA and ASJP.


MIT. Do as you please and praise the snake gods.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

ipatok-0.4.1.tar.gz (16.5 kB view hashes)

Uploaded source

Built Distribution

ipatok-0.4.1-py2.py3-none-any.whl (15.1 kB view hashes)

Uploaded py2 py3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page