I am doing some research on a manuscript which I need to identify the language. My hypothesis is that it is written in phoneme by someone who does not understand the language. Spoken aloud, one that does understand it can guess context and meaning of sounds. Akin to Marcel Duchamp's LHOOQ paintings with its title written in "French phoneme"...

Is there such a thing as a database of phoneme's (x-axis) occurrence (y-axis) across many languages (old French, Occitan, old English, Gaelic, etc) ???

upsid or phoible are close but they do not show occurrence of individual phoneme for a given language, only total number of phoneme.

Any hints will be appreciated.


ps. I was able to get a distribution of phoneme for English and French. Processing corpus of text such as in Gutenberg.org would provide words in many languages but not the phonemes (or sound within the words)... syllables might be more appropriate that phoneme... but then there is no list of syllables for languages... or is there ???

  • 1
    A first pass attempt: wikipedia has phoneme inventories for many languages, you have to go to each language and get the table for the consonants and a separate one for the vowels. and do a lot of editing. I don't know of a a one-stop table for many languages.
    – Mitch
    May 25 '17 at 13:20
  • 1
    Wait...are you asking for the inventory, exactly those phonemes that appear, or for the frequency of such phonemes in the lexicon, or the frequency of all instances over a corpus (for each language)?
    – Mitch
    May 25 '17 at 19:21
  • Actually it's the frequency that I need. Ideally I'd have a list of say 20 phoneme (or different sound or syllables) for a given language (latin, Occitan, medieval French etc) and a frequency of occurrence of each of those phoneme. Plotted on a chart (x axis being the phoneme and y axis being frequency taken from any number of corpus in that language and sorted on decreasing frequency) this would give me a patern which I hope to compare against the (same) patern taken from the manuscript's phoneme frequency (whose language is unknown)
    – Aquila
    May 26 '17 at 11:08
  • 1
    Oh. So your real goal is to identify the language in a manuscript? A manuscript is already written, so how could you convert this into phonemes without knowing the language? If you start with text, just do your frequency analysis on the text itself; adding a layer will add translation error.
    – Mitch
    May 26 '17 at 12:54
  • 1
    all writing is an faulty approximation of sound, even shorthand stenography. If you encode the handwriting somehow, it is probably closer to typed writing rather than to the phonemic pronunciation.I think you're right to do a comparison of frequencies, but start with the simpler thing, the ascii text, before adding the whole extra complexity of pronunciation (which you admit is not one of your strengths). Also google 'MCMC cipher' for a nice way of doing such decoding (you'd try MCMC several times using corpora from different languages)
    – Mitch
    May 26 '17 at 17:21

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.