So it turns out that pinyin can't be reversed back to Chinese characters. However, I keep seeing images like the ones below for different languages (the images below are for Hindi and Japanese, but there are others). Basically they show (in some cases) abugidas (consonant/vowel pairs), or just consonants/vowels separately like the Latin alphabet. So these are sound-based scripts, as opposed to Chinese which is image based. That makes sense then why pinyin wouldn't directly map.

But I'm basically wondering if there is a 1-to-1 mapping, then, between the romanized spelling of say a Japanese or Hindi word, vs. the spelling in the native script.

I'm asking because it seems that English isn't actually a sound-based language. Or more specifically, it is only partially sound-based. You can have two words sound the same, like brake (1 primary meaning) and break (2 primary meanings, break something, and take a break). So you have 1 phonological spelling with IPA that maps to 2 different English orthographic spellings, and a bunch of different meanings. This creates a problem similar to Pinyin, in that you have to disambiguate which spelling you are considering. Chinese has almost 100 meanings for /yi4/. You also have patterns like horse and worse which are the same orthographically but different phonologically.

I'm wondering if these other languages have this same problem. That is, if languages like Hindi, Japanese, even Inuktitut, or other languages that have their own script, yet are sound-based. That is, the pronunciation of the word can map to multiple symbols in that language. Or perhaps that's a benefit of these languages, that they have a direct 1-to-1 mapping between the romanized pronunciation and the native script.

The languages I'm particularly interested in are any with custom scripts (Sanskrit/Hindi, Japanese, Tamil, Inuktitut, Amharic, etc.), or even those with Latin script (Swahili, Malaysian, etc.).

enter image description here enter image description here

enter image description here

  • 2
    Serbo-Croatian can be easily transformed form Cyrillic to Latin and vice-versa
    – David
    Sep 6, 2019 at 8:35

4 Answers 4


In theory there could be a lossless representation, but, in practice, yes, they almost all have the same problem that no, it is not a 1:1 mapping nor otherwise trivial.

(In natural language processing and generally machine learning this task is known as transliteration, for example here, although in some sense it is de-transliteration.)

Transliterating from one script to another is usually lossy, a one-way function, because scripts do not have the same characters, and thus cannot fully and unambiguously represent other languages.

The use of digraphs is a good hint at possible ambiguity (1:2 mapping).

There is nothing special about converting Latinisations back to other scripts. For example, write how are you, Mr Thoreau? in Russian Cyrillic, and then try to convert it back by generic rules.

The exceptions - the only cases where it is mostly doable by rules rather than intelligence or statistical model - are for example conversions between two versions of the same language like Serbian Latin and Serbian Cyrillic.

Balkan fontBalkan font

Even there it is not 1:1 - see љ, њ џ - and there are caveats since the Latin text could contain URLs and so on that should remain in Latin, and the Cyrillic text could contain a Russian word with a character not in the mapping, or a Cyrillic URL.

Maybe there are Indic scripts that are structurally the same and essentially only stylistic variants of each other, similar to Antiqua and Fraktur, or Nuskhur and modern Georgian.


This touches the issue of transcription (may be and is often lossy) and transliteration (that aims at losslessness). The system shown in the question for Hindi is lossless as long as you are ignoring the ligatures which are optional in Hindi orthography and not noted in the transliteration. For Hindi, the use of the letter h in digraphs creates not ambiguity, because a consonant immediately followed by an h always fuses to the appropriate aspirated consonant.

The system for Hiragana is a transliteration for Hiragana, but unfortunately in the Japanese writing system there is more than just Hiragana.

Losslessness in transcription is typically a feature of scientific writing and rendering, practical transcriptions prefer a system that is more intuitive for the end user at the cost of some ambiguity.

  • Wondering if the Hindi system shown, the Hindi symbols actually represent those sounds, or that is just an approximation. Like the Hindi symbols mean something else.
    – Lance
    Sep 26, 2018 at 10:49
  • 2
    The Hindi sounds are quite accurately represented, except for the fact that in Hindi some short a's are silent. The dental sounds could also be represented with a bridge below (t̪, d̪, n̪), but given the context the simple IPA symbols are good enough. Sep 26, 2018 at 12:16
  • 3
    Note: there are ambiguities in the Hunterian system for Hindi. Consider औ → au → औ, अउ or ऐ → ai → ऐ, अइ
    – iacobo
    Sep 26, 2018 at 22:18

Here are some counter-examples to showcase where the mappings are not 1-1 in some of the scripts/romanisations you mention:

Hindi: Devanagari / Hunterian system

  • औ → au → औ, अउ
  • ऐ → ai → ऐ, अइ

Japanese: Hiragana / Hepburn romanisation

  • あな → ana → あな, あんあ
  • おお → ō → おう, おお
  • wouldn't あんあ be transliterated an'a? Feb 23, 2020 at 5:49
  • 1
    @AntonSherwood Apostrophes are used in Modified Hepburn to avoid this ambiguity, but not in traditional Hepburn. Sometimes hyphens were used, but it doesn't appear to have been consistent.
    – iacobo
    Apr 22, 2021 at 12:01

Some transliterations schemes were designed to be reversible, such as the Buckwalter Transliteration Scheme for Arabic: http://languagelog.ldc.upenn.edu/myl/ldc/morph/buckwalter.html.

Even this one fails for edge cases of rare characters that occasionally find their way into natural language text, such as foreign letters, archaic letters, and initial/medial/final forms that are used atypically.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.