Phonomaton: Phonological rule engine

Comments

Metadata

• Language
• Date (format: yyyy-mm-dd)
• Family
• Code (Glottocode)
• Sources (format: Author (year))
• Implementation (name of person)
• Details (language phenomena depicted)
• Complexity (integer, 0 .. 5)

Preamble

• Phonemic inventory

  A line starting Inventory: and continuing with a space-separated list of IPA symbols constituting the phonemic inventory.

• Defined phonemes

  A line starting Define: and continuing with a space-separated list of capital letters (not including the special letters XVCSU) that the rules use as custom phonemes. A custom phoneme can be defined with a set of features enclosed in square brackets, such as A[+syllabic,+low]. Any omitted feature is treated as underspecified (equivalent to polarity ⊗ (\u2297)). As a convenience, you copy a set of features to the clipboard by clicking on a symbol in the IPA chart and then clicking on the features that are displayed.

• Redefined phonemes

  A line starting Redefine: and continuing with an IPA symbol representing a segment and then a set of features (positive or negative) enclosed in square brackets causes this segment to acquire the specified features.

• Disallowed patterns

  A line starting Disallow: and continuing with space-separated patterns causes any rule that would result in one of those patterns to fail to apply, even if the fault lies in the current representation, before the rule is applied. The patterns have the same syntax as the TARGET described below.

• Deleted features

  A line starting Delete: and continuing with a set of features (without polarity) enclosed in square brackets removes those features from consideration.

• Geometric features

  A line of the form Geometry: NAME = gfeature ... defines a new name as a shorthand for the given set of geometric features. Geometric features can be ordinary features, like labial, geometric features from previous Geometry definitions, or predefined geometric features. The predefined geometric features are:

  name	feature set
  cor	coronal anterior distributed strident
  dor	dorsal high low front back tense
  lab	labial labiodental round
  place	cor dor lab
  lar	constr_gl spread_gl voice

  We show below how to use geometric features in rules.

Phonological rules

• Rules are of the form RULENAME: TARGET ARROW CHANGE / ENVIRONMENT // OPTIONS . The Phonomaton executes rules in order, first on the underlying representation, and then on the results of previous rules (the "current representation"). Except as noted below, a rule that can target several places in the current representation applies longest-match (including the environment) first, left-to-right.
• The ENVIRONMENT section is optional; if it is omitted, so is the / that introduces it. If present, it must have a single underscore (_), preceded optionally by a PREVIOUS CONTEXT and followed optionally by a SUBSEQUENT CONTEXT.
• The TARGET and CONTEXT parts may be multiple "characters", where a "character" may be
  • any Unicode sequence representing a segment (see the table), optionally followed by one or more of these diacritics, which can also be entered via the keyboards:
    • [+nasal] (nasalized):   ̃ (\u0303)
    • [+syllabic] (syllabic):   ̩  (\u0329)
    • [-voice] (voiceless):   ̥  (\u0325)
    • [+constr_gl] (creaky-voiced): for vowels,   ̰   (\u0330); for consonants,   ʼ   (\u02bc)
    • [+spread_gl] (breathy-voiced): for vowels,   ̤   (\u0324); for consonants,   ʰ   (\u02b0)
    • [+distributed]:    ̪ (\u032a)
    • [+front]:    ̟   (\u031f)
    • [+back]:    ̠   (\u0320)
    • [+round]:    ʷ   (\u02b7)
    and at most one of these mutually exclusive tone marks:
    • [+tone_top]:   ̋   (\u030b)
    • [+tone_high]:   ́   (\u0301)
    • [+tone_mid]:   ̄   (\u0304)
    • [+tone_low]:   ̀   (\u0300)
    • [+tone_bottom]:   ̏   (\u030f)
    • [+tone_rising]:   ̌   (\u030c)
    • [+tone_falling]:   ̂   (\u0302)
    • [+tone_highRising]:   ᷄   (\u1dc4)
    • [+tone_lowRising]:   ᷅   (\u1dc5)
    • [+tone_highFalling]:   ᷇   (\u1dc7)
    • [+tone_lowFalling]:   ᷆   (\u1dc6)
    • [+tone_peaking]:   ᷈   (\u1dc8)
    • [+tone_dipping]:   ᷉   (\u1dc9)
    and at most one of these mutually exclusive secondary articulations:
    • labial ([+labial, +round]):   ʷ   (\u02b7)
    • palatal ([+dorsal, +high, +front]):   ʲ   (\u02b2)
    • velar ([+dorsal, +high, +back]):   ˠ   (\u02e0)
    • pharyngeal ([+dorsal, +low, +back]):   ˤ   (\u02e4)
    and/or this suprasegmental:
    • [+long] (long):  ː (\u02d0)
  • a bracketed, comma-separated list of FEATURES with polarity in brackets.
    • Available polarities are +, -, 0, ±, α, -α, γ, -γ, δ, -δ, and ∀.
    • The available FEATURES are listed in the table, along with tone marks and suprasegmentals. Example: [-syllabic, -long, αtone_top].
    • The polarity ⊗ (\u2297)in a component of TARGET or ENVIRONMENT means "must match exactly ⊗", which only applies to underspecified segments, which are only introduced as custom phonemes.
    • The polarity α, γ, or δ introduces a variable, whose value is either + or - for the scope of the rule; the corresponding negative polarity, such as -α, is then either - or +.
    • The polarity ∀ (\u2200) may only appear in the CHANGE and ENVIRONMENT of a rule. It must apply to a geometric feature (see above). In the ENVIRONMENT it collects the polarities of all the features that underly the geometric feature. In the CHANGE it asserts all those polarities. For example, V → [∀place] / _[∀place,+syllabic] causes vowels to assimilate in place of articulation with an immediately following vowel.
  • the shorthands V for "vowel", C for "consonant" X for "any consonant or vowel", S for "white space", and U for "subscript" (inserted for autosegmental phonology by an Associate rule).
  • any defined custom phoneme.
  • null, which can be written as ∅ (\u2205), Ø (\u00d8), 0, or ␀ (\u2400),
  • a Perl regex combining the items above. Use ^ to mark the start of the word and $ to mark the end of the word. Other Perl regex metacharacters, such as * and {, are also acceptable.
  • The symbol #, which marks the start of a line, the end of a line, the start or end of a fragment terminated by a space or punctuation.
  • The symbols < and >, which mark the start (respectively, end) of a line or a space-terminated word.
  • Punctuation symbols

    	=	enclitic
    \u21d2	⇒	proclitic
    \u2011	‑	affix symbol (not a minus sign)
    \u20c8	ˈ	stress
    \u27e8	⟨	left infix boundary
    \u27e9	⟩	right infix boundary
    \u2263	≣	juncture
    \u276c	❬	left phrase boundary
    \u276d	❭	right phrase boundary
    	~	reduplicant boundary

  • IPA marks for stress: ˈ \u( 02c8) and ˌ (\u02cc).
  • Ellipsis: A form like ❬...❭ is a shorthand for ❬[^❭]*?❭, and ❬...abc❭ is shorthand for ❬[^❭]*?a[^❭]*?b[^❭]*?c[^❭]*?❭. The brackets can be any pair of bracketing punctuation, particularly ❬❭ and ⟨⟩ . However, the brackets may not be ordinary parentheses '()' or brackets '{}'.
• ARROW may be written as ->, =>, or → (\u2192).
• CHANGE may be
  • null, as above
  • multiple Unicode characters
  • a set of features to apply to the TARGET (which must be a single segment). The feature set may use these polarities:
    • Positive and negative, which indicate a requirement.
    • ±, which means "any polarity of this feature is acceptable"
    • Variable polarities α, γ, and δ and their negations
    • ⊕ and ⊖, which indicate a preference to change polarity to + or - if the result is in the inventory.
• Within TARGET, a segment or a feature set may be followed by any of ¹ ² ³ ⁴ (these are superscript numerals: \u00b9, \u00b2, \u00b3, and \u2074, respectively) to label that segment or feature set. CHANGE may then use any of those labels to refer to the matched segment.
• The OPTIONS section is optional; if it is omitted, so is the // that introduces it. If present, it may have one or more of the following options, separated by spaces.
  • iterate: The rule applies repeatedly until it makes no further changes.
  • once: The rule applies only once in the current representation, even if it could apply in multiple places.
  • exclusive: The rule applies in all places in the usual order, but later applications are suppressed if they occur within the region matched by an earlier application.
• For example, the rule intervocalic lenition: [-continuant,-nasal] → [+continuant,±delayed_release,±distributed] / V_V // once modifies the left-most non-nasal consonant that is surrounded by vowels into a continuant version of that consonant, ignoring the features delayed_release and distributed.

Underlying representations

Morphological rules

• The phonological rules treat affixed forms, which are the results of applying morphological rules to underlying representations. So if there are 4 underlying representations and 3 morphological rules, there are 12 underlying forms for the phonological rules to treat.
• If there are morphological rules, the underlying representations are treated as "stems". The phonological rules also apply to the stems unless the line ExcludeStem is present in the morphological rules.
• If there are no morphological rules, the underlying representations are used directly.
• The morphological rules follow the same syntax as phonological rules.

Autosegmental phonology

• Tiers: The tier names are «seg» , «tone» , and «morph»
• Underlying representation
  • A line of the underlying representation may include multiple tiers; the tier names precede the contents. Sample: «seg» gaga da bala ma «tone» HL N LH HL
  • If there are multiple tiers, each must contain the same number of (space-separated) words.
  • The «seg» tier is the default; if no tier is mentioned in a rule, the rule applies to «seg».
  • The «tone» tier may include any of THMLBthmlbN, refering to top, high, medium, low, and bottom, then their floating versions, then "no assigned tone", as well as spaces separating the tones for each word.
  • The «morph» tier is intended to contain morphosyntactic feature sets that guide rules in the other tiers.
• Phonological rules
  • A rule may apply to a single tier or multiple tiers. If it applies to multiple tiers, it only takes effect if it triggers on all the tiers. Sample: deduplicate: «seg» V¹ → ∅ / ¹_ «tone» H → N / _ *H
  • Special autosegmental rules. These rules take parameters, some of which have default values, as indicated below.
    • Associate: --tiers=tone-seg --dir=L-R --features=[+syllabic] --endpoint=# --max=3
      All the values shown are the defaults. This rule indicates which tiers to associate, the direction of association within a chunk (either L-R or R-L), to which segments in tier2 to associate, how to separate the chunks, and how many elements of tier1 at most to associate with a segment in tier2. The result of each association is shown by a subscript linking an element of tier1 with and element of tier2. The special tone value N always associates with subscript 0.
    • Reassociate: --tiers=tone-seg --dir=L-R --features=[+syllabic] --endpoint=# --max=1 All the values shown are the defaults. This rule moves associations that no longer apply to later (if L-R) or earlier (if R-L) segments. An association no longer applies if a rule has changed its segment so it no longer satisfies the features parameter.
    • Spread: --tier=seg --dir=L-R --endpoint=#
      All the values shown are the defaults. This rule spreads associations in the given tier in the given direction, with a separator betwen chunks.
    • Realize: --tiers=tone-seg --null=tone_low
      Here, tone-seg is default, but there is no default for the null parameter. This rule uses tier1 to realize the associated values in tier2; unassociated segments acquire the default value.
• Morphological rules: These rules can also apply to multiple tiers.
• If a rule has multiple tiers, and any of the per-tier rules fails because its TARGET is not found with the given ENVIRONMENT, then none of the per-tier rules is executed, even though on their own they might be effective.

Disjunctive rules

• Within a tier, a rule can be conditioned on the current form satisfying a pattern. The format is if pattern then rule 1 else rule 2.
• The pattern follows the same syntax as TARGET.
• The rule components have the usual components: TARGET ARROW CHANGE / ENVIRONMENT // OPTIONS.
• If the pattern matches the current form, then rule 1 is executed; otherwise, rule 2 is executed.
• The chosen rule might have no effect, just like any ordinary rule, because its TARGET is not found with the given ENVIRONMENT.
• Although disjunctive rules are mostly intended for morphological rules, they are also valid for phonological rules.