# IPA profiles — linguistic rationale

This document records, for every rule in the IPA profiles
(`ipa-eastern`, `ipa-western`), the primary linguistic source that
attests the rule and its classification (phonemic vs. allophonic /
morphophonological). It is the answer to "why does this transformation
exist, and why exactly this transformation?".

The IPA profiles target **Tier 1: orthography → broad phonemic**.
Rules that are lexicalized, morphophonological, or phonetic-level are
deferred to future narrow-phonetic profiles (`ipa-*-narrow`) and are
explicitly excluded here.

## Primary sources

| Citation | Full reference |
|---|---|
| **Dum-Tragut 2009** | Dum-Tragut, Jasmine. *Armenian: Modern Eastern Armenian*. John Benjamins. ISBN 978-90-272-3814-6. Phonology: pp. 12–55. |
| **JIPA 2024** | Seyfarth, S., Dolatian, H., Guekguezian, P., Kelly, N. & Toparlak, T. (2024). Armenian (Yerevan Eastern and Beirut Western). *Journal of the International Phonetic Association* 54(1): 445–478. DOI: [10.1017/S0025100323000130](https://doi.org/10.1017/S0025100323000130) (open access). |
| **Dolatian 2022** | Dolatian, Hossep. *Armenian Phonology and Phonetics*. Glottothèque lecture series. |
| **Vaux 1998** | Vaux, Bert. *The Phonology of Armenian*. Oxford University Press. ISBN 978-0-19-823661-6. Cited second-hand via Dum-Tragut and JIPA. |

## Eastern Armenian — rules and citations

### Phoneme inventory (KEEP — phonemic)

| Letter | Phoneme | Primary source |
|---|---|---|
| ա | /ɑ/ | Dum-Tragut 2009: 13 (Table 1.1); JIPA 2024: 446 |
| ի | /i/ | Dum-Tragut 2009: 13; JIPA 2024: 446 |
| ը | /ə/ | Dum-Tragut 2009: 13 (phonemic — definite article `-ը` and monosyllables) |
| է | /e/ | Dum-Tragut 2009: 13 |
| օ | /o/ | Dum-Tragut 2009: 13 |
| բ | /b/ | Dum-Tragut 2009: 17–18 (Table 1.2 — three-way contrast with minimal pairs) |
| դ | /d/ | ibid. |
| գ | /ɡ/ | ibid. |
| ձ | /d͡z/ | ibid. |
| ջ | /d͡ʒ/ | ibid. |
| պ | /p/ | ibid. (voiceless unaspirated) |
| տ | /t/ | ibid. |
| կ | /k/ | ibid. |
| ծ | /t͡s/ | ibid. |
| ճ | /t͡ʃ/ | ibid. |
| թ | /tʰ/ | ibid. (voiceless aspirated) |
| ք | /kʰ/ | ibid. |
| փ | /pʰ/ | ibid. |
| ց | /t͡sʰ/ | ibid. |
| չ | /t͡ʃʰ/ | ibid. |
| զ | /z/ | Dum-Tragut 2009: 19 (Table 1.3) |
| ժ | /ʒ/ | ibid. |
| խ | /χ/ | ibid. (voiceless uvular fricative) |
| ղ | /ʁ/ | ibid. (voiced uvular fricative) |
| հ | /h/ | ibid. |
| ս | /s/ | ibid. |
| շ | /ʃ/ | ibid. |
| վ | /v/ | ibid. |
| ֆ | /f/ | ibid. |
| լ | /l/ | ibid. |
| մ | /m/ | ibid. |
| ն | /n/ | ibid. |
| յ | /j/ | ibid. |
| ռ | /r/ | JIPA 2024: 447, 459 (alveolar trill — preserved in Yerevan EA) |
| ր | /ɾ/ | JIPA 2024: 447, 459 (alveolar tap) |
| ւ | /v/ | Dum-Tragut 2009: 12 (Classical glide merged with /v/ in MEA) |

### Context rules (KEEP — phonemic)

| Rule | Citation | Notes |
|---|---|---|
| ե → /je/ word-initial | Dum-Tragut 2009: 14 §1.1.1 point 1.a | Orthography-conditioned: ե vs. է distinguishes initial-glide from no-glide |
| ո → /vo/ word-initial | Dum-Tragut 2009: 16 §1.1.1 point 2.a | Same orthographic conditioning |
| ո + վ → /o/ word-initial | Dum-Tragut 2009: 16 (ով family exception) | Covers ով, ովքեր, ովևիցէ, plus phonotactic generalization to /ov/-initial loans |
| ու → /u/ digraph | Dum-Tragut 2009: 13, 18 (single grapheme) | |
| ե+վ / ե+ւ / և → /ev/ ~ /jev/ word-initial | Same rule as ե; reformed (1922) replaced classical ե+ւ with ե+վ | |

### Excluded (Tier 2 — phonetic / morphophonological)

| Rule | Why excluded | Source |
|---|---|---|
| ղ → [χ] before voiceless | Morphophonological in derived environments; orthography preserves underlying form | Dum-Tragut 2009: 22 ("the 'rules' … are NOT strict and general"); JIPA 2024: 458 |
| վ → [f] before voiceless | Lexically restricted to ~3 words: հարավ, ավտո, նավթ | Dum-Tragut 2009: 30 §1.2.1.f |
| -ությ- → -[ut͡sʰj]- palatalization | Not attested; Dum-Tragut transcribes [utʰjun] | Dum-Tragut 2009: 35; JIPA 2024 |
| Stress (ˈ) on rightmost non-schwa syllable | IPA convention but variable; many lexical exceptions | Dum-Tragut 2009: 47–49 §1.4.1; JIPA 2024: 466 |
| Schwa epenthesis in CC clusters | Phonetic-level; not in orthography | Dum-Tragut 2009: 30–35 §1.2.2; JIPA 2024: 464–465 |
| Word-final devoicing (մարդ → [mɑɾtʰ]) | Dum-Tragut: "in principle, no terminal devoicing"; lexicalized exception list | Dum-Tragut 2009: 22–23 §1.2.1.a; JIPA 2024: 451 |
| Post-nasal aspiration (անդամ → [ɑntʰɑm]) | Lexicalized | Dum-Tragut 2009: 23 §1.2.1.a |
| իա → /jɑ/ glide | Phonetic-level glide formation; not in primary sources as phonemic | (Wiktionary-only) |

## Standard Western Armenian — rules and citations

### Phoneme inventory (KEEP — phonemic)

Vowels and most fricatives/sonorants are identical to Eastern. The
Western-specific differences come from the W1 voicing collapse:

| Letter | Eastern | Standard Western | Citation |
|---|---|---|---|
| բ | /b/ | /pʰ/ | JIPA 2024: 447–448 (Table); Vaux 1998 (per JIPA); Baronian 2017; Dolatian 2022 slide 3 |
| գ | /ɡ/ | /kʰ/ | ibid. |
| դ | /d/ | /tʰ/ | ibid. |
| ձ | /d͡z/ | /t͡sʰ/ | ibid. |
| ջ | /d͡ʒ/ | /t͡ʃʰ/ | ibid. (merges with չ in Standard Western) |
| կ | /k/ | /ɡ/ | ibid. |
| պ | /p/ | /b/ | ibid. |
| տ | /t/ | /d/ | ibid. |
| ծ | /t͡s/ | /d͡z/ | ibid. |
| ճ | /t͡ʃ/ | /d͡ʒ/ | ibid. |
| թ | /tʰ/ | /tʰ/ | unchanged (aspirated retained) |
| ք | /kʰ/ | /kʰ/ | ibid. |
| փ | /pʰ/ | /pʰ/ | ibid. |
| ց | /t͡sʰ/ | /t͡sʰ/ | Standard Western retains aspiration (Beirut HD has /t͡s/ per JIPA 455) |
| չ | /t͡ʃʰ/ | /t͡ʃʰ/ | ibid. |

Note: Wiktionary's algorithm strips Western aspiration entirely on թ ք
փ ց չ. This is **not supported by the literature** — JIPA 2024:
447–448 and Dolatian 2022 confirm aspiration is retained in Standard
Western. Our profile retains aspiration. The Beirut variety (HD
speaker in JIPA 2024) unaspirated the affricates ց → [t͡s] and չ →
[t͡ʃ]; we keep them aspirated for parity with Standard Western
references (Bedrossian, Kouymjian).

Rhotic merger: Western has merged ր and ռ to a single rhotic /ɾ/ in
most varieties (JIPA 2024: 459 — "Beirut WA … has neutralized the
rhotic distinction").

### Context rules (KEEP — phonemic)

Same as Eastern: word-initial ե → /je/, word-initial ո → /vo/ with
ով family exception, ու digraph, ե+վ / ե+ւ / և sequences.

### Sequence rules (KEEP — phonotactic, phonemic)

| Sequence | Output | Citation |
|---|---|---|
| ստ | /st/ | JIPA 2024: 453 (sibilant + voiceless plosive cluster); Dum-Tragut 2009: 24 §1.2.1.b for the Eastern analogue |
| սպ | /sp/ | ibid. |
| սկ | /sk/ | ibid. |
| շտ | /ʃt/ | ibid. |
| շպ | /ʃp/ | ibid. |
| շկ | /ʃk/ | ibid. |

Rationale: Armenian (and most languages) disallow sibilant + voiced
stop clusters. The orthographic ս/շ + պ/տ/կ surface as voiceless
plosives despite the W1 voicing collapse on պ/տ/կ in standalone
contexts.

### Excluded (Tier 2 — phonetic / morphophonological)

| Rule | Why excluded | Source |
|---|---|---|
| Western progressive devoicing of voiced obstruents after voiceless | Morphophonological, derived environments only; Beirut-specific | JIPA 2024: 458–459 |
| Regressive ղ → [χ], վ → [f] | Same disposition as Eastern (Tier 2) | Dum-Tragut 2009: 22 |
| Adjacent-sibilant deaspiration (ափսէ → [ɑpse]) | Variable across speakers | JIPA 2024: 453 |
| յու → [ʏ] post-consonant | Documented as variable; engine lacks lookbehind for post-consonant context | Dolatian 2022 slide 6; JIPA 2024: 461 |
| ցնել causative schwa epenthesis | Subsumed under general schwa-epenthesis (Tier 2) | Dolatian 2020, 2021 |
| Classical էօ → /œ/ | Not attested in primary sources consulted | — |

## Wiktionary conformance

The fixture
[`tools/ipa-conformance/wiktionary-phonemic-testcases.json`](../tools/ipa-conformance/wiktionary-phonemic-testcases.json)
encodes Wiktionary's `phonemic_IPA` algorithm outputs as a comparison
baseline. Cases where our profile deliberately diverges from Wiktionary
are annotated with `east_divergence_reason` / `west_divergence_reason`
and categorized in `divergence_categories`.

To run the conformance check:

```bash
pnpm run ipa-conformance
```

Current state (v3.0.0):

- **Eastern**: 47/52 exact match, 5 expected divergences, 0 regressions
- **Western**: 9/23 exact match, 14 expected divergences, 0 regressions

All 19 divergences fall into one of the documented categories. The
script exits with code 1 only when a regression is detected (a
divergence without an annotated reason).

## Adding new rules

When adding a rule to either profile, follow this checklist:

1. Find a primary-source attestation in Dum-Tragut, JIPA, Vaux, or
   Dolatian. Note the page / section reference.
2. Classify it: phonemic (orthography → phoneme, predictable) vs.
   phonetic / lexicalized / morphophonological.
3. If Tier 1 (phonemic), implement in `src/standards/ipa/`. Add an
   inline citation in the rule's JSDoc.
4. If Tier 2+ (phonetic / lexical), do **not** add it to the broad
   phonemic profile. Document it here as a deferred rule and target a
   future `ipa-*-narrow` profile.
5. Add at least one test case in `__tests__/standards/ipa-*.test.ts`.
6. If Wiktionary applies the same rule, ensure the conformance fixture
   has a `*_divergence_reason` annotation if we deviate.