Skip to main content

Library for dealing with sound changes

Project description


Codacy Badge Build Status codecov

alteruphono is a Python library for applying sound changes to phonetic and phonological representations, intended for use in simulations of language evolution.

Please remember that, while usable, alteruphono is a work-in-progress. The best documentation is currently to check the tests, and the library is not recommended for production usage.

Future improvements

  • Move from existing AST to a dictionary, mostly for speed and portability (even if it might be the code more verbose); should still be a frozen dictionary
  • Memoize parser.__call__() calls
  • Consider that, if a rule has alternatives, sound_classes, or other profilific rules in context, it might be necessary to perform a more complex merging and add back-references in post to what is matched in ante, which could potentially even mean different ASTs for forward and backward. This needs further and detailed investigation, or explicit exclusion of such rules (the user could always have the profilic rules in ante and post, manually doing what would be done here).
  • Use logging where appropriate
  • Allow different boundary symbols, including "^" and "$"
  • Add support for clusters/diphthongs
  • Add tone and other suprasegmental
  • Add custom features
  • research about kleene closures


In any standard Python environment, alteruphono can be installed with:

pip install alteruphono

How to use

Detailed documentation can be found in the library source code and will be published along with the paper accompanying the library; terser technical description is available at the end of this document. Consultation of the sound changes provided for testing purposes is also recommended.

For basic usage as a library, the .forward() and .backward() functions can be used as a wrapper for most common circumstances. In the examples below, a rule p > t / _ V (that is, /p/ turns into /t/ when followed by a vowel) is applied both in forward and backward direction to the /pate/ sound sequence; the .backward() function correctly returns the two possible proto-forms:

>>> import alteruphono
>>> alteruphono.forward("# p a t e #", "p > t / _ V")
['#', 't', 'a', 't', 'e', '#']
>>> alteruphono.backward("# p a t e #", "p > t / _ V")
[['#', 'p', 'a', 't', 'e', '#'], ['#', 'p', 'a', 'p', 'e', '#']]

A stand-alone command-line tool can be used to call these wrapper functions:

$ alteruphono forward '# p a t e #' 'p > t / _ V'
# t a t e #
$ alteruphono backward '# p a t e #' 'p > t / _ V'
['# p a t e #', '# p a p e #']


We are not exploring every detail of the formal grammar for annotating sound changes, such as the flexibility with spaces and tabulations or equivalent symbols for the arrows; for full information, interested parties can consult the reference PEG grammar and the source code.

AlteruPhono operates by applying ordered lists of sound changes to textual representation of sound sequences.

Sound changes are annotated in the A -> B / C syntax, whose constituents are for reference referred as "source" (A), "target" (B), and "context" (C), with the first two being mandatory; the other elements are named "arrow" and "slash". When applied to segment sequences, we refer to the original one as "ante" and to the resulting one (which might have been modified or not) as "post". So, with a rule "p -> b / _ a" applied to "pape":

  • p is the "source"
  • b is the "target"
  • _ a is the "context"
  • "pape" is the "ante (sequence)"
  • "bape" is the "post (sequence)"

Note that, if applied backwards, a rule will have a post sequence but potentially more than one ante sequence. If the rule above is applied backwards to the post sequence "bape", as explained in the backwards definition and given that we have no complementary information, the result is a set of ante sequences "pape" and "bape".

AlteruPhono operates on sound sequences expressed in standard CLDF/LingPy notation, derived for Cysouw work, i.e., as a string character string with tokens separated by single spaces. As such, a word like the English "chance" is represented not as "/tʃæns/" or /t͡ʃæns/, in proper IPA notation, but as "tʃ æ n s". While the notation might at first seem strange, it has proven its advantages with extensive work on linguistic databases, as it not only facilitates data entry and inspection, but also makes no assumptions about what constitutes a segment, no matter how obvious the segmentation might look to a linguist. On one had, being agnostic in terms of the segmentation allows the program to operate as a "dumb" machine, and on the other allows researchers to operate on different kinds of segmentation if suitable for their research, including treating whole syllables as segments. In order to facilitate the potentially tedious and error-prone task of manual segmentation, orthographic profiles can be used as in Lexibank.


While they are not enforced and in some cases are not needed, such as when the system operates as a glorified search&replace, alteruphono is designed to operate with three main catalogs: graphemes, features, and segment classes.

Graphemes are sequences of one or more textual characters where most characters are accepts (exceptions are...). While in most cases it will correspond to common transcription system such as the IPA, and in most case correspond to a single sound or phoneme, this is not enforced and sequence of characters (with the exception of a white-space, a tabulation, a forward slash, square and curly brackets, and an arrow) can be used to represent anything defined as a segment in a corresponding catalog. Also note that the slash notation of Lexibank is supported. The default catalog distributed with alteruphono is based on the BIPA system of clts.

Features are descriptors... Default is derived from BIPA descriptors, mostly articulatory, but we also incluse some distinctive feature systems.

It is not necessary for a grapheme catalog to specify the features that compose each grapheme, but this severly limits the kind of operations possible, particularly when modelling observed or plausible sound changes.

The default catalogs are derived from BIPA... such as in examle

Segment classes are just shorthards. The default distributed with AlteruPhono includes a number of shorthands common in the literature and mostly unambiguous


  • A grapheme is a sequence of one or more textual characters representing a segment, such as "a", "kʷʰ".

  • A bundle is an explicit listing of features and values, as defined in a reference, enclosed in square brackets, such as "[open,front,vowel]" or "[-vowel]". Features are separated by commas, with optional spaces, and may carry a specific value in the format feature=value with value being either a logical boolean ("true" or "false") or a numeric value; shorthands for "true" and "false" are defined as the operators "+" and "-"; if no "value" is provided, it defaults to "true" (so that [open,front,vowel] is internally translated to [open=true,front=true,vowel=true]). Note on back-references here (experimental)

  • A modifier is a bundle of feautes used to modify a basic value; for example, if "V" defines a segment class (see item below) of vowels, "V[long]" would restrict the set of matched segments to long vowels.

  • A segment-class is a short-hand to a bundle of features, as defined in a reference, intended to match one or more segments are expressed with one or more upper-case characters, such as "C" or "VL" (for [consonant] and [long,vowel], respectively, in the default). A segment class can have a modifier.

  • A marker is a single character non-segmental information. Defined markers are # for word-boundary, . for syllable break, + for morpheme boundary, stress marks and tone marks. Note that some markers, particularly suprasegmental features as stress and tone, in most cases will not be referred directly when writing rule, but by tiers. See section on tiers.

  • A focus is a special marker, represented by underscore, and used in context to indicate the position of the source and target. See reference when discuss contexts.

  • An alternative is a list of one or more segments (which tzype?) separated by a vertical bar, such "b|p". While in almost all cases of actual usage alternatives could be expressed by bundles (such "b|p" as "[plosive,bilabial]" in most inventories, using an alternative is in most cases preferable for legibility

  • A set is a list of alternative segments where the order is significant, expressed between curly brackets and separated by commas, such as {a,e}. The order is significant in the sense that, in the case of a corresponding set, elements will be matched by their index: if {a,e} is matched with {ɛ,i}, all /a/ will become /ɛ/ and all /e/ will become /i/ (note how, with standard IPA descriptors, it would not be possible to express such raising in a an unambiguos way)

  • A back-reference is a reference to a previously matched segment, expressed by the symbol @ and the numeric index for the segment, (such as @2 for referring to the second element, the vowel /a/, in the segment sequence "b a"). As such, back-references allow to carry identities: if "V s V" means any intervocalic "s" and "a s a" means only "s" between "a", "V s @1" means any "s" in intervocalic position where the two vowels are equal. Back-references can take modifier.


For version 2.0: - Implement mapper support in the automata (also with test cases) - Implement parentheses support in the grammar and automata (also with test cases) - Consider moving to ANTLR - For the grammar, consider removing direct sound match in segment, only using alternative (potentially renamed to expression and dealt with in an appropriate way) - don't collect a context, but left and right already in the AST (i.e., remove the position symbol)

- In Graphviz output
    - Accept a strng with a description (could be the output of the
    - Draw borders around `source`, `target`, and `context`
    - Add indices to sequences, at least optionally
    - Accept definitions of sound classes and IPA, at least in English

Old version

  • Use logging everywhere
  • Implement automatic, semi-automatic, and requested syllabification based on prosody strength
  • Implement both PEG grammars from separate repository
  • Add support for custom replacement functions (deciding on notation)


There are two basic elements: rules and sequences. A rule operates on a sequence, resulting in a single, potentially different, sequence in forwards direction, and in at least one, potentially different, sequence in backwards direction.

Following the conventions and practices of CLTS, CLDF, Lingpy, and orthographic profiles, the proposed notation operates on "strings", that is, text in Unicode characters representing a sequence of one or more segments separated by spaces. The most common segments are sounds as represented by Unicode glyphs, so that a transcription like /haʊs/ ("house" in English Received Pronounciation) is represented as "h a ʊ s", that is, not considering spaces, U+0068 (LATIN SMALL LETTER H), U+0061 (LATIN SMALL LETTER A), U+028A (LATIN SMALL LETTER UPSILON), and U+0073 (LATIN SMALL LETTER S). The usage of spaces might seem inconventient and even odds at first, but the convention has proven useful with years of experience of phonological transcription for computer-assisted treatment, as not only it makes no automatic assumption of what constitutes a segment (for example, allowing user to work with fully atomic syllables), but facilitates validation work.

A rule is a statement expressed in the A > B / C _ D notation, where C and D, both optional, express the preceding and following context. It is a shorthand to common notation, internally mapped to C A D > B A D. While A and B might expresses something different from historical evolution, such as correspondence, they are respectively named ante and post, and the rule can be real as "the sequence of segments A changes into the sequence of sounds B when preceded by C and followed by D". A, B, and C are referred as as "sequences", and are composed of one or more "segments". A "segment" is the basic, fundamental, atomic unit of a sequence.

Segments can be of X types:

  • sound segments, such as phonemes (like a or ʒ) or whatever is defined as an atomic segment by the used (for example, full-length syllables such as ba or ʈ͡ʂʰjou̯˨˩˦). In most cases, a phonetic or phonological transcription system such IPA or NAPA will be used; by default, the system operates on BIPA, which also facilitates normalization in terms of homoglyphs, etc.
  • A bundle of features, expressed as comma separated feature-values enclosed by square brackets, such as [consonant], referring to all consonants, [unrounded,open-mid,central,vowel], referring to all sounds matching this bundle of features (that is, ɜ and the same sound with modifiers), etc. Complex relationships and tiers allow to expressed between segments, as described later. By default, the system of descriptors used by BIPA is used.
  • Sound-classes, which are common short-hands for bundles of features, such as K for [consonant,velar] or R for "resonants" (defined internally as [consonant,-stop]). A default system, expressed in table X, is provided, and can be replaced, modified, or extended by the user. Sound-classes are expressed in all upper-case.
  • Back-references, used to refer to other segments in a sequence, which are expressed by the at-symbol (@) and a numeric index, such as @1 or @3 (1-based). These will are better explored in X.
  • Special segments related to sequences, which are
    • _ (underscore) for the "focus" in a context (from the name by Hartman 2003), that is, the position where ante and post sequences are found
    • # (hash) for word boundaries
    • . (dot) for syllable breaks

Sound segments, sound-classes, and back-references can carry a modifier, which is following bundle of features the modifies the value expressed or referred to. For example θ[voiced] is equivalent to ð, C[voiceless] would match only voiceless consonants, C[voiceless] ə @1[voiced] would match sequences of voiceless consonants, followed by a schwa, followed by the corresponding voiced consonant (thus matching sequences like p ə b and k ə g, but not p ə g).

Other non primitives include alternatives and sets.

How to cite

If you use alteruphono, please cite it as:

Tresoldi, Tiago (2020). Alteruphono, a tool for simulating sound changes. Version 0.3. Jena. Available at:

In BibTex:

  author = {Tresoldi, Tiago},
  title = {Alteruphono, a tool for simulating sound changes. Version 0.3.},
  howpublished = {\url{}},
  address = {Jena},
  year = {2020},


Tiago Tresoldi (

The author was supported during development by the ERC Grant #715618 for the project CALC (Computer-Assisted Language Comparison: Reconciling Computational and Classical Approaches in Historical Linguistics), led by Johann-Mattis List.

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

alteruphono-0.4.tar.gz (127.0 kB view hashes)

Uploaded source

Built Distribution

alteruphono-0.4-py3-none-any.whl (252.4 kB view hashes)

Uploaded py3

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Huawei Huawei PSF Sponsor Microsoft Microsoft PSF Sponsor NVIDIA NVIDIA PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page