Spelling Iteration Log

Every phoneme spelling change made during Ingglish development: what we tried, what worked, what didn't, and why.

Summary of Changes

Sound	IPA	Spelling	Changes
my, time	/aɪ/	ai	ii → ie → ai
cow, out	/aʊ/	ou	ow → ou
go, show	/oʊ/	oh	o → oh
father, hot	/ɑ/	o	ah → o
thought, law	/ɔ/	aw	aw → o → aw
book, put	/ʊ/	u	uu → oo → u
too, blue	/uː/	oo	oo → uu → oo
arrow, carrot	/æɹ/	arr	aar → arr
but, cup	/ʌ/	uh	u → uh
schwa (about, sofa)	/ə/	a	u → a (AH0 only)

Diphthong Evolution

/aɪ/ (my, time): ii → ie → ai

Attempt 1: 'ii'

• Rationale: Double letter for long sound
• Problem: "fiit" looked like "feet", confusing for readers
• Verdict: ❌ Rejected - visual confusion

Attempt 2: 'ie'

• Rationale: Matches English words like "tie", "pie", "die"
• Problem: Still felt arbitrary, no international precedent
• Verdict: ⚠️ Better, but not ideal

Attempt 3: 'ai' (current)

• Rationale:
  • Directly represents IPA /aɪ/ - phonemically transparent
  • Pinyin uses 'ai' (standard romanization for Mandarin Chinese)
  • Italian and Vietnamese use 'ai'
  • You can "see" the a→i glide
  • English 'ai' words (rain, paint) use /eɪ/, so 'ai' is available
• Impact: More identical words than 'ii' or 'ie'. ('ei' looks better by raw dictionary count, but frequency analysis shows net -1 /M; gains are mostly rare German surnames like Bernstein and Alzheimer, essentially zero real-text impact.)
• Verdict: ✅ Adopted - international precedent + phonemic clarity

Examples:

• my → mai
• time → taim
• night → nait
• I → ai

/aʊ/ (cow, out): ow → ou

Attempt 1: 'ow'

• Rationale: Matches English "cow", "now", "how", "wow"
• Problem:
  • Words like "out", "loud", "sound" became "owt", "lowd", "sownd"
  • Looked unfamiliar despite matching some English patterns
  • Only "Regional" rating - just English uses 'ow' for this
• Verdict: ⚠️ Workable but not optimal

Attempt 2: 'ou' (current)

• Rationale:
  • Words like "out", "loud", "sound" become IDENTICAL to English
  • Dutch also uses 'ou' for this sound (oud = old)
  • Upgraded from "Regional" to "Common" rating
• Impact: Large frequency gain vs 'ow': out (3,965 /M), about (3,725 /M), our (1,308 /M), sound (141 /M), found (121 /M) are among the most common English words
• Trade-off: "cow" → "kou" looks less familiar
• Verdict: ✅ Adopted - identical common words outweigh unfamiliar rare words

Examples:

• out → out (identical!)
• our → ouer
• loud → loud (identical!)
• sound → sound (identical!)
• cow → kou
• house → hous

/oʊ/ (go, show): o → oh

Attempt 1: 'o'

• Rationale: Simple; 'o' is the conventional Latin-script letter for back vowels (though English /oʊ/ is a diphthong, unlike the pure /o/ of Spanish or Italian)
• Problem: We originally used 'ah' for /ɑ/ (hot → "haht", rock → "rahk"), but feedback was that 'ah' looked too foreign. Changing /ɑ/ to 'o' made those words natural (hot → "hot", rock → "rok") but meant 'o' was taken, creating a collision here.
• Verdict: ❌ Rejected - collision with /ɑ/ vowel

Attempt 2: 'oh' (current)

• Rationale:
  • English "oh!" already uses this for the exclamation
  • Distinguishes "go" (goh) from "cow" (kou) without ambiguity
  • Only option left after reserving 'o' for /ɑ/
• Trade-off: 'ow' gains show (501 /M), own (471 /M), throw (132 /M) but loses "oh" (3,374 /M), net -1,330 /M. 'oa' is even worse. Both introduce perceptual ambiguity (see attempt 3 below)
• Verdict: ✅ Adopted - necessary to avoid collision

Attempt 3: 'ow' (rejected)

• Rationale: Would make snow, throw, bowl, window identical to English. Gains show (501 /M), own (471 /M), throw (132 /M) but net -1,330 /M because "oh" alone is 3,374 /M
• Problem: ow is ambiguous in English: it represents both /oʊ/ (snow, throw) and /aʊ/ (cow, town). New combinations like bownz (bones) read as "bowns" and howm (home) sounds like it rhymes with "cow". Reintroduces the exact ambiguity ingglish is designed to eliminate.
• Verdict: ❌ Rejected - perceptual ambiguity despite no formal collisions

Examples:

• go → goh
• show → shoh
• hello → haloh

Vowel Evolution

/ɑ/ (father, hot): ah → o

Attempt 1: 'ah'

• Rationale: Phonetically accurate for open back vowel
• Problem:
  • "rock" → "rahk" looked strange
  • "hot" → "haht" unrecognizable
• Verdict: ❌ Rejected - words looked too foreign

Attempt 2: 'o' (current)

• Rationale:
  • "rock" → "rok" looks natural
  • "hot" → "hot" (identical!)
  • 'o' is the conventional Latin-script letter for back vowels across languages
• Impact: Massive frequency gain vs 'ah': hot (195 /M), got (222 /M), job (153 /M), lot (141 /M) are all high-frequency words that become identical
• Verdict: ✅ Adopted - familiar results

Examples:

• hot → hot (identical!)
• rock → rok
• father → fodher

/ɔ/ (thought, law): aw → o → aw

This vowel went through the most iteration.

Attempt 1: 'aw'

• Rationale: Matches English "law", "saw", "raw"
• Worked reasonably well
• Verdict: ⚠️ Acceptable

Attempt 2: 'o' (caught-cot merger)

• Rationale: Simplify by merging with /ɑ/, reflecting American pronunciation
• Problem:
  • Lost distinction for speakers who maintain the difference
  • Required changing /oʊ/ to 'oh' to avoid collision
• Verdict: ❌ Rejected - lost too much information

Attempt 3: 'aw' (current, reverted)

• Rationale:
  • "thought" → "thawt" is readable
  • "law" → "law" (identical!)
  • Preserves distinction for non-merged speakers
  • Common rating - matches English "law", "saw"
• Impact: Merging to 'o' would add +176 collision groups. 'au' avoids collisions but loses -555 /M: saw (413 /M), law (119 /M), lawyer (82 /M) outweigh the gains fault (107 /M), paul (97 /M), launch (20 /M).
• Verdict: ✅ Adopted - maintains distinction, familiar results

Examples:

• law → law (identical!)
• thought → thawt
• call → kawl

/ʊ/ and /uː/ Swap: oo ↔ uu (superseded)

Original:

• /ʊ/ (book) → 'uu'
• /uː/ (too) → 'oo'

Problem:

• "book" → "buuk" looked strange when English already has "book"
• "too" → "too" was identical, but the long sound had the shorter spelling

After Swap:

• /ʊ/ (book) → 'oo' - matches English "book", "good", "look"
• /uː/ (too) → 'uu' - longer sound gets longer spelling (Finnish pattern)

Impact: The original uu/oo assignment produces more identical spellings by raw count. But frequency tells the real story: the swap keeps "would" (1,813 /M), "good" (2,677 /M), "could" (1,475 /M), "should" (803 /M), "look" (1,038 /M), "book" (182 /M), "looking" (476 /M) as identical. These high-frequency words far outweigh the low-frequency words lost.

Verdict: ⚠️ Adopted then superseded - the /ʌ/→'uh' change later freed 'u' for /ʊ/ and 'oo' for /uː/; see chain change below

Later considered: 'eu' for /uː/ (rejected)

• Rationale: Would gain +19 /M (zeus 6 /M, neutral 4 /M, maneuver 3 /M) with minimal losses
• Problem: eu in English implies a /j/ onset: "feud" is /fjuːd/, "neural" is /njʊɹəl/. So meun (moon) reads as "mew-n" (two syllables), seun (soon) reads as "syoon", teu (too) reads as "tyoo". The mapping actively misleads English readers.
• Lesson: Identical word count isn't enough on its own. A shared spelling that reads as the wrong sound is worse than an unfamiliar spelling that reads correctly.
• Verdict: ❌ Rejected - perceptual ambiguity, and the +19 /M gain is negligible anyway

/ʌ/, /ʊ/, and /uː/ Chain: u/oo/uu → uh/u/oo

A three-way chain shift that resolved the 'uu' digraph and aligned all three vowels with more intuitive spellings.

Before (the oo/uu swap era):

• /ʌ/ (but) → 'u'
• /ʊ/ (book) → 'oo'
• /uː/ (too) → 'uu'

Problem:

• 'uu' has no precedent in English: "tuu", "thruu", "byuutafal" looked alien
• Most world languages use 'u' for /ʊ/, not 'oo'
• 'oo' is the natural English spelling for /uː/ (too, food, moon, cool)

After (current):

• /ʌ/ (but) → 'uh', the English interjection "uh" is exactly this sound
• /ʊ/ (book) → 'u', what most world languages use for this vowel
• /uː/ (too) → 'oo', matches English "too", "food", "moon", "cool"

Rationale:

• 'uh' is the intuitive English interjection for /ʌ/; everyone knows how "uh" sounds
• Freeing 'u' for /ʊ/ aligns with most world languages (Universal rating)
• 'oo' for /uː/ matches English conventions (too, food, moon, cool, school)
• Eliminates 'uu' entirely, no more unfamiliar digraphs

Impact:

• "too" → "too" (identical!), "food" → "food" (identical!), "moon" → "moon" (identical!), "school" → "skool", "blue" → "bloo", "you" → "yoo"
• "book" → "buk", "good" → "gud", "could" → "kud", "would" → "wud", "should" → "shud", "put" → "put" (identical!), "look" → "luk"
• "but" → "buht", "cup" → "kuhp", "love" → "luhv", "of" → "uhv"
• Loses some /ʊ/ identical words (book, good, could, would, should, look) but gains /uː/ identical words (too, food, moon) and eliminates the unfamiliar 'uu'

Verdict: ✅ Adopted: eliminates 'uu', aligns with world languages, 'oo' matches English conventions

Examples:

• but → buht
• cup → kuhp
• love → luhv
• book → buk
• good → gud
• could → kud
• too → too (identical!)
• food → food (identical!)
• school → skool
• beautiful → byootafal
• through → throo

/ə/ (about, sofa): u → a

The schwa (/ə/) is the most common vowel in English: it appears in nearly every multi-syllable word (about, the, beautiful, difficult, nation). This change only affects unstressed schwa (/ə/), not the stressed /ʌ/ STRUT vowel used in "but", "cup", "run" (those use 'uh').

Attempt 1: 'u'

• Rationale: Both /ə/ and /ʌ/ are represented as 'AH' in ARPAbet, so mapping all AH → 'u' was the simplest approach
• Problem:
  • "the" → "dhu" (unrecognizable; "the" is the most common English word)
  • "about" → "ubout" (lost the identical English spelling)
  • "hello" → "huloh" (the 'u' in the first syllable looked odd)
  • "nation" → "nayshun" (the '-un' ending felt wrong for a word ending in /-ən/)
  • "beautiful" → "byootufool" (confusing)
• Verdict: ❌ Rejected: schwa words looked too unfamiliar

Attempt 2: 'a' (current)

• Rationale:
  • "about" → "about" (identical!), "and" → "and" (identical!), "the" → "dha" (natural)
  • 'a' is phonetically close to schwa; many languages use 'a' for their neutral vowel
  • English already spells schwa as 'a' in the most common words: a, about, again, along, away, around, all identical in Ingglish
  • Stressed /ʌ/ stays separate ('uh'), so no collisions with STRUT words
• Impact:
  • 67.6× frequency-weighted improvement, the largest gain from any single change
  • Top gains: "a" (20,941 /M), "and" (13,733 /M), "about" (3,725 /M), "around" (1,428 /M)
  • Only +93 net collision groups (acceptable; most are low-frequency)
  • Losses cluster in predictable patterns: un- prefix (until→antil), up- prefix (upset→apset), -ful suffix (handful→handfal), -um suffix (museum→myoozeeam)
• Trade-off: AH0+R must remain 'ur' (not 'ar') to avoid collision with AA+R→'ar'. This is handled by a special R-colored vowel rule that overrides the schwa mapping before R.
• Verdict: ✅ Adopted: massive familiarity gain with minimal downside

Examples:

• about → about (identical!)
• and → and (identical!)
• the → dha
• again → agen (identical!)
• hello → haloh
• beautiful → byootafal
• difficult → difakalt
• nation → nayshan

R-Colored Vowel Evolution

R-colored vowels were added iteratively to fix collisions and improve readability. Collectively, the R-colored vowel rules (air, eer, ar, or, arr) produce a large frequency-weighted identical word gain and prevent 25 collision groups.

/æ/+R: aar → arr

Attempt 1: 'aar'

• Rationale: Double the vowel before R
• Problem: "arrow" → "aaroh" looked strange

Attempt 2: 'arr' (current)

• Rationale:
  • English already doubles consonants after short vowels (carrot, barrel, arrow)
  • "arrow" → "arroh" is more recognizable
  • Upgraded from "Novel" to "Common" rating
• Verdict: ✅ Adopted - matches English convention

Examples:

• arrow → arroh
• carrot → karrat
• barrel → barral

/ɛ/+R → 'air' (added)

Before: No special handling

• "air" → "er" (collision with "her")
• "there" → "dher" (collision with "the" + "her")

After: Look-ahead rule for /ɛ/ followed by R

• "air" → "air" (identical!)
• "there" → "dhair"
• "care" → "kair"

Verdict: ✅ Essential - eliminates collisions between /ɛ/+R and /ɝ/ words

/ɪ/+R → 'eer' (added)

Before: No special handling

• "beard" → "bird" (looks like the animal)
• "beer" → "bir" (unrecognizable)
• "fear" → "fir" (looks like the tree)

After: Look-ahead rule for /ɪ/ followed by R

• "beard" → "beerd"
• "beer" → "beer" (identical!)
• "fear" → "feer"
• "near" → "neer"

Verdict: ✅ Essential - eliminated confusing false cognates

/ɑ/+R → 'ar' (added)

Before: No special handling

• "star" → "stor" (collision with "store")

After: Look-ahead rule for /ɑ/ followed by R

• "star" → "star" (identical!)
• "car" → "kar"

Verdict: ✅ Essential - eliminated major collisions

/ɔ/+R → 'or' (added)

Before: No special handling

• "store" → "stawr" (confusing)

After: Look-ahead rule for /ɔ/ followed by R

• "store" → "stor"
• "more" → "mor"

Verdict: ✅ Essential - natural spellings

Lessons

1. Identical Words Are a Big Win (But Not Everything)

When a word is spelled identically in English and Ingglish (out→out, loud→loud, too→too, law→law), it provides maximum familiarity. We prioritize mappings that create more identical words, but not at the cost of creating collisions (different words with the same spelling).

Current status: 10,150 identical words (8.05% of 126,051 dictionary words). The schwa change (AH0 → 'a') alone produced the largest frequency-weighted gain of any change (67.6×). See Identical Words Analysis for frequency-weighted analysis of potential improvements.

2. International Precedent Matters

Spellings with support from multiple languages (like 'ai' from Pinyin/Italian/Vietnamese) are more defensible than purely English-based choices.

3. Collisions Must Be Fixed

R-colored vowel rules were essential to prevent words like "air" and "her" from both mapping to "er". Fixing collisions is more important than simplicity.

4. The "Novel" to "Common" Upgrade

Several changes were specifically made to upgrade ratings:

• 'arr' (was 'aar'): Novel → Common
• 'ou' (was 'ow'): Regional → Common
• 'u' for /ʊ/: Regional → Universal
• 'oo' for /uː/: Common

5. Reversion Is Okay

The /ɔ/ vowel went aw → o → aw. We weren't afraid to revert when a change didn't work out. The goal is the best final system, not sticking with early decisions.

6. Identical Word Count Can Mislead

A spelling that matches more English words is harmful if English readers pronounce those new combinations wrong. The correct test isn't "does this string match?" but "does an English reader naturally say this correctly?" See Design Decisions for examples.

7. Stress-Conditioned Splits Can Unlock Big Wins

The schwa change split AH into two spellings based on stress: AH0 (unstressed) → 'a', AH1/AH2 (stressed) → 'uh'. This required logic in the conversion function, not just a mapping table. When a single phoneme symbol covers two sounds that English speakers perceive as distinct (like /ə/ and /ʌ/), splitting by stress is worth considering. See Identical Words Analysis for other candidates.

Changes Not Made (Considered and Rejected)

Using 'au' for /aʊ/ instead of 'ou'

• Would match German/Dutch/Portuguese
• Rejected because 'ou' preserves high-frequency identical words (out 3,965 /M, about 3,725 /M, our 1,308 /M, sound 141 /M) that 'au' would lose

Using pure IPA-style spellings throughout

• Would be more internationally consistent
• Rejected because target audience is primarily English speakers

Removing R-colored vowel special handling

• Would simplify the system
• Rejected because it would lose high-frequency identical words (star, air, beer, store, etc.) and add 25 collision groups

Version History

For the complete git history, see:

git log --oneline --all --grep="spelling\|phoneme\|vowel\|diphthong"