What is the use or quality of the orthography-to-IPA mapping charts?

Question

In relation to How to build a robust transliteration scheme across languages? I am now confused about orthography-to-IPA mappings, such as for Turkish. When you see the orthograph like the letter a mapped to the IPA value /a/, what does this mean? Does it mean in the "general" case? In the ideal case? In a mainstream case? Because (and I don't know for Turkish, but for English)... Well, English's situation is the craziest I've seen. So my question is, how should I be using these orthograph-to-IPA mappings? As the gold truth? As a rough guideline? As an entrypoint to further research in some textbook? Why do they even put it there if it's not the gold truth?

I ask because I want to build a transliteration (or perhaps transcription? pending the above question) system for various languages. I want to result in the latin alphabet for pronunciation purposes, so I've been using these orthography-to-IPA tables (all the southeast asian and asian languages have simple ones, for example). But now I'm not sure what they represent. I would like to know what they represent, and what I should be doing instead, to capture such information to create a latin text version of some language/script for purposes of pronunciation.

Answer 1

Generally, Wikipedia articles like that list the phonemes of each language: the theoretical mental units, not the actual sounds pronounced. Each phoneme is then usually given a name that indicates its most common or easiest-to-write pronunciation. For example, the General American English phoneme /t/ can be realized as any of [t t̪ t̚ ɾ ʔ] and more, depending on the environment—but those all represent the same mental unit, and out of those, t is the easiest to write, so the phoneme is called /t/.

If you want a narrow phonetic transcription (writing down the exact sounds produced as accurately as possible), you'll have to dig deeper into the phonology of the language. A good reference grammar for a language should include an overview of this.

Answer 2

As a general rule, you should not accept Wikipedia claims about language as the definitive truth. Sometimes the claims are really false, sometimes they are reasonably true. You have to look at the totality of knowledge that we have about the language, so supplement your reading of Turkish with other data sources on the language. Since Wiki articles do not have "an" author and we usually do know know anything about the ideology or competence of the author, we cannot say what it means to write "e" or "a" in a table of sounds.

Generally speaking, information on written languages is driven by native orthography, so they are telling you what the letter "a" or "e" is supposed to sound like. Authors generally do not present information on notions like "phoneme" and "allophone", unless you get lucky and you are using a source with detailed phonetic transcriptions paired with spellings (Basbøll's Danish phonology is an example – the spelling system is a bit of a mess).

Since most reference works on literary languages are based on written standards, you should expect massive imprecision and indeterminacy w.r.t. pronunciation. A stellar example of that is Modern Mongolian, which is transliterated in a manner that is really deceptive (there's no vowel there, that's just a spelling device). There are a few sources of MM which give transcriptions, such as Svantesson et al. You need a speaker of Mongolian to know how "pizza" is pronounced (the orthographic vowel is actually pronounced, it's not a dummy vowel).

So, for literary languages, the status is mainly "it's about the native letters". It is also about dialect-unification (one spelling system for North Saami, hundreds of minutely-distinguished dialects). For languages reported on in descriptive monographs based on field work, for example Klamath, you have to study the author's ideology. Barker, who wrote the works on Klamath, was reasonably clear about what his letters "mean". Unfortunately, sometimes you don't have any basis for concluding anything about the meaning of "e" in a work. This is why in modern descriptive best-practices, written data are paired with sound recordings (which has substantial associated technical problems).