Phon Tiers (%xmodsyl, %xphosyl, %xphoaln, %xphoint)

Status: Reference Last updated: 2026-07-31 09:51 EDT

The Phon extension tiers provide syllable-level phonological annotation, segmental alignment between target and actual IPA, and per-phone time intervals. They are produced by the Phon application. As of Phon 4.0.0-beta.9 (2026-06-25), Phon reads and writes CHAT natively (see “Data quality notes” below for what this means for existing corpus files).

The authority for these formats is upstream, not this chapter. Phon generates these tiers, so Phon specifies them; the current specification comes from Phon’s maintainer. This chapter and the E735-E746 error specs are chatter’s implementation OF that specification, not a substitute for it. Where chatter disagrees with it, the disagreement is a finding to take upstream rather than a rule to settle here.

chatter parses and validates all four tiers as first-class CHAT tiers.

The x prefix. Phon emits these tiers with a leading x (%xmodsyl, %xphosyl, %xphoaln, %xphoint) to mark them as extension tiers. The grammar accepts both the x-prefixed names and the historical non-x names (%modsyl, %phosyl, %phoaln, %phoint); the parser and validator key off the tier kind, not the literal prefix. The canonical serialized form is the x-prefixed name.

The four tiers

%xmodsyl, %xphosyl, and %xphoaln are word-aligned to their source tier(s) with single ASCII spaces. %xphoint uses / (space-slash-space) as its word separator because single spaces already separate the phone and bullet tokens inside each word.

Tier formats

%xmodsyl / %xphosyl, syllabification

A word is one or more phone:CODE units concatenated with no internal whitespace; words are separated by single spaces. The phone is one IPA phone (IPA length is written with the modifier letter ː, U+02D0, never an ASCII colon, so the : separator is unambiguous). A leading stress marker (ˈ primary, ˌ secondary) is part of the phone it precedes.

Pause fillers. Phon keeps every word-aligned phonology tier in index lockstep with the main tier: when the main tier carries a pause, the pause token ((.), (..), (...), or numeric (x.x)) is mirrored at the same word position on %mod, %pho, %xmodsyl, and %xphosyl (and as a (..)↔(..) pair on %xphoaln). A pause filler is a valid word on the syllabification tiers; it carries no phone:CODE structure and must mirror the same pause token as the source-tier word at its position. Numeric pauses ((1.5), minutes:seconds (1:02.5)) are accepted on the same footing as the three untimed forms, per Greg Hedlund’s spec.

The constituent code is one character. The legal codes are O N C L R E A D U:

The remaining Phon SyllableConstituentType mnemonics, B (boundary), S (stress), W (word boundary), T (tone), are not emitted on these tiers: boundary, stress, and tone need no per-phone marker.

*CHI:	I want three .
%mod:	aɪ wɑnt θri
%xmodsyl:	a:Dɪ:D w:Oɑ:Nn:Ct:C θ:Oɹ:Oi:N
%pho:	aɪ wɑn fwi
%xphosyl:	a:Dɪ:D w:Oɑ:Nn:C f:Ow:Oi:N

%xphoaln, phone alignment

A word is one or more comma-separated pairs; a pair is model↔actual (↔ is U+2194). Either side may be ∅ (U+2205, empty set): ∅ on the left is an epenthesis (a phone produced but not targeted); ∅ on the right is a deletion. Both sides are never ∅ at once.

*CHI:	the best .
%mod:	ðə bɛst
%pho:	ðə bɛs
%xphoaln:	ð↔ð,ə↔ə b↔b,ɛ↔ɛ,s↔s,t↔∅

The alignment lists segments (phones). Suprasegmental stress (ˈ/ˌ) that may appear on the %mod/%pho word is therefore not part of the alignment pairs; the reconstruction checks below compare modulo those stress markers.

%xphoint, per-phone intervals

%xphoint gives the time segmentation of each individual phone on %pho, effectively phone-level bullets analogous to the word-level timing on %wor. Groups (one per %pho word) are separated by /. Within a group, each phone is followed by a CLAN time-alignment bullet: the byte 0x15 (NAK), the interval start_end, then 0x15.

*CHI:	I want . •0_500•
%pho:	aɪ wɑnt
%xphoint:	aɪ •0_250• / w •250_320• ɑ •320_400• n •400_460• t •460_500•

(Bullets are shown as • above; in the file they are the 0x15 byte.)

Validation

These checks run by default. Pass --suppress xphon to silence the entire Phon %x validation surface, or suppress an individual code. (The historical --check-xphon opt-in flag is now a deprecated no-op: the checks it used to gate are on by default.)

Word-count cross-checks:

• %xmodsyl ↔ %mod: E725, %xphosyl ↔ %pho: E726. Always a strict equality: inter-word pauses are mirrored onto both tiers by construction (spec §1 rule 4), so the counts must match exactly.
• %xphoaln ↔ %mod: E727, ↔ %pho: E728. NOT strict equality: a pause word present on only one of %mod/%pho forms its own %xphoaln alignment word (its other side entirely ∅, e.g. (..)↔∅) and consumes a word slot only on the tier bearing the pause (spec §2 rule 5). E727/E728 compare each tier’s word count against %xphoaln’s count after excluding the one-sided pause words that do not consume a slot on that tier, not against %xphoaln’s raw word count.

Content checks:

See Alignment Architecture for the word-count implementation.

Parsing strategy

• %xmodsyl / %xphosyl: stored as flat word strings (talkbank-model::dependent_tier::phon::SylTier), consistent with how %pho and %mod store flat phone words. The validator tokenizes each word into typed phone:CODE units (PositionCode) to apply the content rules above; the IPA characters themselves stay verbatim for exact round-trip.
• %xphoaln: each word is parsed into a Vec<AlignmentPair>, where AlignmentPair { source, target } carries one model↔actual mapping (None is ∅).
• %xphoint: parsed into typed groups of (phone, bullet) pairs (XphointTier / XphointGroup / PhoneInterval), reusing the same 0x15 bullet machinery as %wor.

Deep phonological analysis is Phon’s domain; chatter parses the structure that validation needs and keeps the IPA content verbatim.

Phon XML source format

In Phon’s native XML format, phonological data is stored as structured elements:

<ipaTarget>
  <pho>
    <pw>
      <ph scType="onset"><base>θ</base></ph>
      <ph scType="nucleus"><base>ɹ</base></ph>
      <ph scType="nucleus"><base>i</base></ph>
    </pw>
  </pho>
</ipaTarget>

Each <pw> (phonological word) element contains <ph> elements with syllable constituent types (scType). The <alignment> element provides phone-level mappings between target and actual using index-based <pm> (phone map) entries.

Data quality notes

A small percentage of Phon corpus XML records have an orthography↔IPA word-count mismatch: the number of <pw> elements in <ipaTarget> / <ipaActual> differs from the number of <w> elements in <orthography>. This is expected in child phonology data: children may produce extra syllables, partial words, or over-productions relative to the target.

For current counts on a local CHILDES/TalkBank data tree, run:

python3 scripts/analysis/scan_phon_mismatches.py /path/to/data

A subset of existing corpus CHAT files handle this discrepancy inconsistently between tiers:

1. %mod/%pho map IPA words to orthography words one to one; extras are silently dropped.
2. %xmodsyl/%xphosyl/%xphoaln carry the full IPA word set, undropped.

This produces CHAT files where %xmodsyl may have more words than %mod, triggering the E725-E728 word-count errors. As of Phon 4.0.0-beta.9 (2026-06-25), Phon reads and writes CHAT natively; we have not yet seen output from that native export and do not know whether it reproduces this inconsistency. Files already affected remain in the corpora and must continue to validate against the behavior above.