IPA vs phonetic respelling

There are two ways to tell someone how a word sounds.

Respelling rewrites it using ordinary letters and the reader’s existing spelling habits: pruh-NUN-see-AY-shun.

Phonetic transcription rewrites it in a notation where each symbol has one fixed value: /prəˌnʌnsiˈeɪʃən/.

The first is easier to read. The second is the only one that survives contact with a second language.

Why respelling feels better

It requires nothing. A reader who has never heard of phonetics can parse pruh-NUN-see-AY-shun immediately, and the capitalisation makes the stress obvious without a diacritic.

For a monolingual English dictionary aimed at native speakers, this is genuinely the better choice, and Merriam-Webster’s use of it is not laziness. The readers all share the same spelling conventions, so the system works.

Why it breaks

Respelling has one fatal property: it encodes the sound using the reader’s spelling system, which means it can only express sounds the reader’s language already has, and only to readers with the same spelling intuitions.

Three failure modes.

1. Readers do not share conventions

ay is /eɪ/ to an English speaker, /aj/ to a German speaker, /ɛ/ to a French speaker. oo is /uː/ in English and two separate vowels in Dutch. Every respelling silently assumes a reader.

An English respelling of a French word is a set of instructions written in a code the learner may not hold.

2. New sounds are unrepresentable

This is the serious one. If a sound does not exist in the reader’s language, there are no letters for it and the respelling substitutes the nearest thing — which is precisely the error you were trying to prevent.

Target	English respelling	What it makes the learner say	Actual
French tu	“too”	/tu/ — a different word (tout)	/ty/
French rouge	“roozh”	English /r/	/ʁuʒ/
German ich	“ikh”	/k/ or /ʃ/	/ɪç/
Spanish perro	“peh-rroh”	English /ɹ/	/ˈpero/
Mandarin xī	“shee”	/ʃ/	/ɕi/
Danish rød	“rurdh”	approximately nothing	/ʁœːð/

Every row teaches a wrong pronunciation with complete confidence. The respelling is not merely imprecise — it actively installs the error the learner was going to make anyway, and then endorses it.

3. It cannot represent what it does not distinguish

English respelling has no way to write vowel length independently of quality, so Finnish tuli and tuuli come out identical. No way to write tone, so all four Mandarin ma are “mah”. No way to write gemination, so Italian nono and nonno collapse.

These are not edge cases. They are the central difficulties of those languages.

Romanisation is a third thing

A frequent confusion: pinyin, Hepburn, RTGS and the rest are romanisations, not phonetic transcriptions. They are writing systems for the language, using Latin letters, with conventional values that are internally consistent and often unlike English.

Pinyin q is /tɕʰ/. Pinyin x is /ɕ/. Pinyin zh is /ʈʂ/. An English speaker reading pinyin as English produces something well outside the target, and this is one of the most common sources of persistent Mandarin pronunciation error.

Thai RTGS is worse for learners, because it marks neither tone nor vowel length — which are the two features that matter most. A learner relying on it is missing the load-bearing information entirely. This is the transliteration versus transcription distinction, and it is not academic.

Romanisations are useful for typing and for reading. They are not pronunciation guides for outsiders.

What IPA buys

One symbol, one sound, independent of any language’s spelling. /y/ is the same vowel in French, German, Turkish, Finnish and Mandarin. /ʁ/ is the same consonant in French, German, Danish and Hebrew.

That gives you three things respelling cannot:

Cross-language comparability. A learner who has learned /ʁ/ for French already has it for German. The knowledge transfers because the notation does.

Precision about what is wrong. “Your /ʁ/ is coming out as /ɹ/” names both the target and the error. “Your R is a bit off” names neither.

Representability of the unfamiliar. Anything a human language does, IPA can write — including the sounds your language lacks, which are exactly the sounds you need to be told about.

The honest costs

It has to be learned. Not all of it — about twenty symbols covers most of what a learner needs, and how to read the IPA is the twenty-minute version. But it is not zero.

It looks intimidating. /ˈɔːlsəʊ/ frightens people who would happily read AWL-soh.

Conventions still vary. Dictionaries transcribe the same English differently — some use /ɪ/ and /iː/, others /i/ and /iː/, and American and British traditions diverge on several vowels. The symbols are standard; the transcription conventions built on them are not entirely.

It is a pain to type. No IPA keyboard exists on any normal device, which is why teachers using a shared document almost never include phonetic transcription, and why pronunciation work rarely survives the end of a lesson.

What to use when

Monolingual dictionary for native speakers   → respelling
Learner dictionary                           → IPA
Teaching a sound your student's L1 lacks     → IPA, always
Quick note between speakers of one language  → respelling is fine
Anything involving tone or length            → IPA
Typing a language on a keyboard              → romanisation

The rule of thumb: respelling is fine when the target sound already exists in the reader’s language. The moment it does not — which is the moment that actually matters in pronunciation teaching — respelling stops being imprecise and starts being wrong.

The typing problem is the practical obstacle, and it is why phonetic transcription is absent from most lesson materials despite everyone agreeing it is useful. Generating it automatically rather than hunting symbols in a picker is one of the things Teachee does with lesson sentences — the transcription appears with the transcript and translation, so the notation costs nothing to include.