phonomorph-researcher

name	phonomorph-researcher
description	Research and document phonetic transformations (e.g., [p] -> [b]) for the PhonoMorph atlas. Use to find linguistic evidence, examples, and scholarly sources to fill an empty cell in the sound matrix.

PhonoMorph Researcher

Research and document phonetic transformations between two IPA symbols.

1. Expert Research Protocol

1. Internal Knowledge First: Rely on your deep training in historical linguistics.
   • Vowels: Check for Raising/Lowering, Fronting/Backing (Umlaut), Rounding/Unrounding, and Centralization (Reduction).
   • Consonants: Check for Lenition scales (Plosive -> Fricative -> Approximant -> Zero), Fortition, Palatalization, and Place Assimilation.
2. Directional Check: If you find [A] -> [B], check if the inverse [B] -> [A] is a documented, regular shift. Document both if valid.
3. Targeted Search: Use google_web_search with specific parameters:
   • Academic Deep Dive: Use site:scholar.google.com, site:books.google.com, and site:cyberleninka.ru.
   • Search Operators: Use quotes for exact shifts (e.g., "ʁ to l" or "uvular fricative" to "l") and the OR operator for variations.
   • Linguistic Repositories: Look for papers in JSTOR, Project MUSE, or university repositories (e.g., site:edu "phonetic shift") found via Google.
   • CyberLeninka Specialization: For Slavic or Eurasian shifts, CyberLeninka is a primary source.
4. Wikification: Ensure sound names and language names match their standard Wikipedia titles for easy linking.
5. No generic language placeholders: Each languageExamples[].language must name a specific language ("Modern Greek", "Yoruba", "Old English"). Never write "Various languages", "Multiple families", or similar. If you cannot name at least one specific language with attested examples, output {"unattested": true} and move on.
6. Fast Failure: If no regular, historical phonetic shift exists after searching these deep sources, mark as unattested and move on. It is better to skip than to fill a cell with speculative or generic content.

2. Data Population

File Path: public/data/transformations/{fromId}_to_{toId}.json

Strict Data Schema:

{
  "fromId": "string",
  "toId": "string",
  "preamble": "string (popular science intro)",
  "phoneticEffects": "string (technical terms)",
  "languageExamples": [
    {
      "language": "string",
      "languageFamily": "string (optional)",
      "examples": [
        { "from": "string", "to": "string", "note": "string (optional context)" }
      ]
    }
  ],
  "certainty": 5,
  "commonality": 5,
  "sources": ["string (citations)"],
  "tags": ["string (e.g., Romance, Indo-European, Lenition)"],
  "related": [
    { "fromId": "string", "toId": "string", "label": "string", "type": "chain | branch" }
  ]
}

Certainty rubric — apply ALL gates

• 5: A specific historical period or dialect is named (e.g., "Old English to Middle English", "Caribbean Spanish") AND at least one citation directly documents this specific shift (not a generic reference work).
• 4: The shift is well-attested in the literature AND you have a specific (non-generic) citation, but it is not period-specific.
• 3: Your only citations are generic reference works (Ladefoged & Maddieson 1996; Campbell 2013; Hock 1991; Crystal) — OR — plausible from phonetic principles AND you can name at least one specific language, but you have no citation at all.
• 2: Speculative or marginal — strongly prefer {"unattested": true} instead.
• 1: Do not produce. If you would assign 1, output {"unattested": true}.

If you are about to write certainty: 5 with only Ladefoged & Maddieson (1996), Campbell (2013), or Hock (1991) as your citation, downgrade to 3.

2.5. Source Citation Rules — STRICT

Citations are the most fragile part of LLM-generated content. Hallucinated URLs and books degrade the atlas. Follow these rules.

1. URLs: Cite a URL ONLY if it was returned by your google_web_search tool in this response. Do NOT cite URLs you "recall" from training — fabricated Wikipedia and archive.org URLs are the dominant error mode (e.g., Stopping_(phonetics), Consonantalization, Vocalization_(phonetics) — none of these articles exist). If grounding returned no URL for the specific shift, omit URLs entirely and cite via book/article only.

2. Books: Use the format Author, A. (Year). Full Title. Publisher. — the Year is mandatory; never write (n.d.). Never use abbreviated forms like Wells: Accents of English or topic phrases like Phonological history of Ancient Greek and Spanish. If you cannot recall the publisher and full title with confidence, do not cite that book. Verify book titles against your training knowledge before citing — Alkire & Rosen (2010). Romance Philology. is wrong; the actual book is Romance Languages: A Historical Introduction.

3. Articles: Author, A. (Year). Title. Journal Name, vol(num), pages. — Year is mandatory. If you cannot find the year, do not cite that article.

4. Specific over generic: Prefer citations that directly discuss the shift you are documenting. Reference works (Ladefoged & Maddieson 1996; Campbell 2013; Hock 1991) are acceptable but should not be the only citation when a more specific source exists. If a reference work is your only citation, set certainty=3 (not 4 or 5).

5. No placeholder text or non-academic sources: Never write "Source via search", "verify before merge", "research snippets", "Cited in search result N", (n.d.), or any phrase that signals "I have no real citation." Never cite Reddit, Quora, Stack Exchange, Wikipedia, languagehat.com, Scribd, Calaméo, or Internet Archive — these are not peer-reviewed sources. If you have no real academic citation, mark {"unattested": true}.

6. No duplicate citations within a single transformation's sources array.

3. Automation

After saving the JSON file (or modifying unattested in index.json), always rebuild the data index to bundle the new metadata:

npm run rebuild-index

Check git status to ensure public/data/index.json and the new transformation file are both staged for the next batch commit.