0

I've been theorizing if it's possible to create an automatic transliteration program between two different languages that use two different writing systems.

So, I was wondering if there are certain numerical parameters we could use to describe each phonetic sound. Letters of the first alphabet would have these numerical values, and we would be able to find their closest neighbors in the other alphabet.

The closest thing I could find were phonetic chart maps, which could be useful, but as I see it, they would give me boolean parameters, not numerical ones, so the "phonetic transliteration" would be much more inaccurate than it otherwise would be with numerical parameters.

Is it technically possible to describe phonetic sounds with numbers? If so, where can I read about it?

1
  • 1
    Writing systems rarely represent phonetic sounds accurately, so an automated transliteration program based on phonetics is doomed to fail. You can often easily transliterate one writing system to another (e.g., romanisation, which is frequently quite automatic), but unless both languages you’re dealing with happen to have perfectly phonetic orthographies, you’ll never be able to do it via phonetics. Jan 1 at 12:23

1 Answer 1

2

Physical sound can't be counted, it can only be measured. A digitized waveform is a vector of numbers that reproduces physical sound quite accurately, assuming you use standard sampling and bit rates. The concept of "between two languages" is essentially impossible / inaccessible if you are looking at speech from the physical perspective.

Linguistic units, i.e "sounds", can be counted and are a property of particular language. "θ" is a sound-unit of English, "ʕ" is a sound-unit of Arabic but not English, and so on. We already have a coding system for reducing speech to letter-sequences, where "thing" is represented as "θɪŋ". You can look up the symbols (and articulatory / auditory referents) here, and it will even tell you the Unicode numeric value for every symbol (there is a bit of further file fiddling depending on whether it is UTF-8, UTF-16 etc).

The main impediment is the reduction of pronunciations to "standard" transcriptions. There are dozens of ways to transcribe a given English word, which have nothing to do with pronunciation differences, for example "tame" might be [tem, te:m, tʰem, tʰe:m, teim, tɛɪm] and so on. There are two main approaches to transcription. One starts with an assumption of what the phonemes of English are, then fits words into some theory of phonemes and co-occurrence, the other starts afresh with each utterance and creates sequences base on how close a given recording is to a reference value (there are recordings on that page). In the latter case, there are very many squabbles over whether the vowel in a particular recording is "really" [e] or [ɛ]. The "we already know the answer" approach resolves this because there is a single vowel phoneme in question here, so you either decide that the vowel should be written as [ɛɪ], or as [e:] and so on, and for the most part people recognize that it is pointless to make claims for the superiority of one arbitrary convention over another.

For English, you can create transcriptional tables allowing you to map between [ɛɪ] and [e:]. Especially if you know the intended word such as "tame", transcriptional variation can be fairly easily tamed, if you have a sufficiently large corpus of transcriptions (not billions of words, but a few dozen that hit the range of phonemes in English). There are two impediments to transcriptional inter-translation. First, "English" is a language family, General American English and Londonderry English are not easily mutually intelligible, and many words are extremely different across dialects. Second, there are massive disagreements as to the phonemic analysis (the reduction of a narrow phonetic transcription that hugs the reference values of the IPA to a more compact representation of the "contrasts").

These are problems that will face most major languages of the world, thus a "transcriptional translation" between English and Spanish will also depends on the kind of Spanish you have in mind (Spanish is much less variable in terms of arbitrary transcriptional conventions). For most languages, you would have to start by first doing an extended analysis of phonetics and phonology, to discover what the surface phones and abstract phonemes are.

1
  • There are conservative dialects which preserve distinction between /eː/ and /ɛj/ (or /ɛɪ̯/). See here about pane–pain merger. In those dialects ⟨tame⟩ is /te:m/ [tʰe:m] and *⟨taim⟩ might be /tɛɪ̯m/ [tʰɛɪ̯m].
    – Arfrever
    Jan 2 at 0:22

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.