Which languages can you directly convert the spelling of the word into a "standard" pronunciation? From my understanding so far:

  • Chinese (through pinyin)
  • Hebrew (seem to have a rigid grammar for pronunciation if diacritical marks are present)
  • Sanskrit (same here)
  • Arabic (same here)

The languages which seem you cannot accurately generate the pronunciation from the spelling (without having a dictionary of individual words):

  • English
  • Tibetan (read there are too many edge cases, but not sure).

Could someone please help me identify the ones from this list which you can generate the pronunciation from some orthography? This way one can know which languages you have to collect a lexicon/pronunciation mapping for, vs. those that only require the lexicon without pronunciation. If nothing else, can you get "very reasonably close" to doing it, with only a few edge cases? (As opposed to English which has tons of edge cases).


  • Pali
  • Malayalam
  • Sinhala
  • Tamil
  • Telugu
  • Thai
  • Korean
  • Japanese (through some pinyin equivalent?)
  • Spanish
  • French
  • Finnish
  • Georgian
  • Armenian
  • Amharic
  • Inuktitut
  • Navajo
  • Latin
  • Greek (preferably Koine/Ancient greek)
  • Cherokee
  • Igbo
  • Yoruba
  • Xhosa
  • Hausa
  • Urdu
  • Punjabi
  • Coptic
  • Gothic
  • Old Norse
  • Old Irish
  • Russian
  • Vietnamese
  • Khmer

I would like to build a text to speech sort of thing (very basic), and knowing which languages I have to manually bite the bullet and collect individual words for would be helpful, as I'm sure linguists probably know this quickly.

While not all in the list necessarily need to have an answer, knowing what categories of languages from this list support this feature would help. For example, do all dravidian languages have the same feature of being + or - this feature?

By "standard pronunciation", I mean sort of like what you'd find in a general IPA transcription of some word, it doesn't have to be super perfect.

  • 1
    How would you count, say, Latin, where general rules can get it right 90% of the time but there are a handful of exceptions like huic and traicere? Where is the line drawn between "phonemic" and "not phonemic"? – Draconis Aug 19 '20 at 1:16
  • I guess knowing that it was 90%, that would at least tell me it would be useful to write out the rules manually instead of gathering every single word. Then you could just add the exceptions to the list of special words I guess. – Lance Pollard Aug 19 '20 at 1:54
  • The phonemic line would be drawn at basic IPA rather than adding all the fancy bells and whistles of IPA superscripts and subscript markers. So all the major vowel and consonant sounds, so lets say there's 100 total sounds, maybe even just 50 or less. A real general IPA. – Lance Pollard Aug 19 '20 at 1:57
  • Talking about phonemic scripts, specifically (ones which represent all phonemic distinctions of the language); which actual features are phonemic varies by language, and can absolutely include superscript and subscript markers. – Draconis Aug 19 '20 at 2:00
  • 1
    For example, Ancient Greek had a phonemic distinction between /t/ and /tʰ/, Latin had a phonemic distinction between /k/ and /kʷ/, etc. Which distinctions are indicated with separate symbols and which distinctions are indicated with superscripts and subscripts comes down to historical accidents, in most cases; there's no linguistic reason why ejectives should be marked with a diacritic and implosives with special symbols of their own, that's entirely historical coincidence. – Draconis Aug 19 '20 at 2:02

As a general guideline, it depends how old the orthography is, and how much evidence we have about the language.

There are very few ancient scripts that are completely accurate and unambiguous in representing the language's phonology. Sanskrit written in Devanāgarī is one of those very few exceptions, and even it has some quirks that have to be accounted for.

Latin, Greek, Coptic, and so on were all adapting foreign writing systems to fit their language—and this generally led to some distinctions being ignored. Greek, for example, doesn't indicate the difference between short /a/ and long /aː/, and Latin indicates that same difference only sporadically. These orthographies are sometimes called defective.

In some cases, usually for more-widely-studied languages and ones that have significant religious use, diacritics have been invented to fill in these gaps: it's possible to write the Greek distinction as ᾰ versus ᾱ. But even when the diacritic systems exist, they're generally seldom used.

For some languages, we simply don't have enough information to know. All surviving information about Gothic is written, so if there was a distinction never indicated in writing, the best we can do is guess. Were certain vowel alternations allophonic variation, or had the alternation fossilized? There's no way to be sure.

And then there are languages which have had a new orthography specifically engineered for them by linguists. These are engineered specifically to make the mapping from phonemes to graphemes (and vice versa) as easy as possible, and the newer they are, the less likely the language has changed significantly since their inception. The oldest orthography of this type is probably Hangeul (for Korean); modern orthographies for many American and African languages are also in this category.

You'll probably have the most success sticking to this last category. If you want to work with "defective" scripts, Johan Winge's thesis lays out a promising model, but this would be a lot of work for a language without lots of groundwork already done. And for the middle category, often the best you can do is a broad approximation; it's up to you if that's sufficient or not.

  • So generally newer languages are easier? Except for Sanskrit which is the exception, and other widely studied languages like Hebrew. – Lance Pollard Aug 19 '20 at 2:04
  • 2
    @LancePollard "Newer" here referring to the orthography, specifically, not the language as a whole. Having more resources matters too (Hebrew will be easier than Aramaic), but newer linguist-designed orthographies will almost always be easier than older ones. – Draconis Aug 19 '20 at 2:06
  • @LancePollard Rather than ‘newer’, it will generally be easier the shorter the time span between the invention/development of the script or spelling and the time of attestation. Classical Attic Greek is not new, but is fairly simple and deducible because the orthography hadn’t been settled for very long by the time attestations start appearing. – Janus Bahs Jacquet Aug 19 '20 at 22:58
  • @JanusBahsJacquet A good point, though the Phoenician script wasn't a perfect fit for Greek either (needs diacritics added in for vowel length, /h/, phonemic pitch accent, etc depending on dialect). I'd say Hangeul fits modern Korean "better" (in terms of this question) than the Euclidean alphabet fit Attic Greek, because of the method of its design. – Draconis Aug 19 '20 at 23:10
  • The Euclidean alphabet, cool! Learn something new everyday :) @JanusBahsJacquet I can't quite follow what you're saying, are you saying that the shorter the time between first speaking and first writing the language, the better? I am getting tripped up by the term attestation. – Lance Pollard Aug 20 '20 at 2:04

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