I. Let us say we have a syllabary of n symbols.

II. Let us have three ways or methods of transliterating a text written using the symbols of the syllabary:

  1. The first method considers the syllabary as consisting of n−1 syllabograms and 1 non-space delimiter.

  2. The second method has the same parameters as the first one, but every syllabogram can be considered as a complete coda-less syllable, with 0 to 5 possible codas. That is, a syllabogram X can be read as X or X1, X2, X3, X4, X5 (that is up to 6 alliterating syllables) The overall frequency F(non-X) of X representing X1, X2, X3, X4 or X5 is considerably lower than the frequency F(X) of X representing the simplex X. Thus, its value is known.

  3. The third method differs from the first one in that it considers the above-mentioned delimiter-like character as a (non-empty) wild card character representing up to m different consonants. Now, while in the first and second methods, there is a clear delimiter showing word boundaries, as mentioned above, in this method, there is none, because the symbol used for delimiting words in the other two methods is here used just as another sound (a consonant) either in the middle or at the end of a word. Hence, whoever tries to interpret or read the text in this way has to guess where words begin and where they end. The segmentation is thus arbitrary.

More specific examples of all three methods:

  1. a-me-li-ka | I | bi a-me-li-ka i bi
  2. a-me-li-ka | I | bi a-me-li-ka i(s) bi(g) (America is big or America, I be! :-) )
  3. a-me-li-ka | I | bi a me-li ka-bi(n) bi (A merry cabin bee etc. ;-) )

Now, how can I quantify which of the three methods leads to more uncertainty, and by how much?

The first one, at least, seems pretty straightforward and leads to minimum uncertainty (in comparison to the latter two). So, my question really is, how can I compare no. 2 and no. 3?

Many thanks for any advice or help!


EDIT: A more specific example then:

Let us have two competing scholars proposing two competing hypotheses.

Both of them have the same text before them, written using the same syllabic script. The two scholar differ, however, as to how the script should be transliterated/read.

1. Scholar X assumes one of the characters marks word boundaries. Scholar Y assumes the very same character is a wild card character that can represent up to 15 different consonants.

2. Scholar X assumes the underlying language is Language X. Scholar Y assumes the underlying language is Language Y (different from X).

3. Scholar X segments the text into words using what he believes are word delimiters, checks the vocabulary of Language X to see if it contains such words. Scholar Y goes through the vocabulary of Language Y to see if they can spot any of the words in the text.

Intuitively, it seems scholar X's method is more rigorous and does not allow too many alternatives. Scholar Y's method, on the other hand, appears to provide so many alternatives that, to me it seems it's almost impossible not to find at least something resembling Language Y.

Intuition aside, are there any ways to (dis)prove that scholar X is more likely to be right than scholar Y?

  • 1
    Can we assume we have a corpus for the underlying language? – WavesWashSands Jul 21 '18 at 17:38
  • @WavesWashSands Not in the sense of balanced data. We do have a certain, say, lexicon of lemmas and word-forms for Language X, and we may try to get some information regarding their frequencies, but the source texts are too unbalanced in many respects. As for Language Y, we have a huge lexicon of lemmas and word-forms (by a few orders, perhaps), but we have no information regarding their frequencies. We could only speculate based on closely related languages Y_A, Y_B and maybe a few others, if that makes any sense. ;-) But for the sake of simplicity, we may assume no knowledge of frequencies. – Pavel Jetušek Jul 22 '18 at 1:37
  • 1
    Well, darn :/ I was going to suggest a way of doing it that would only be possible with good corpora for both languages. Guess that isn't possible! Out of curiosity, is this a real problem that you're working on? (Just curious, you don't have to reveal the details) – WavesWashSands Jul 22 '18 at 7:45
  • 1
    Well, it actually is a real problem. There really are two competing hypotheses, very much like the ones of Scholar X and Y's. In the real world, X proposes what happens to be the mainstream view, while Y is considered an eccentric, to put it mildly. My goal is to come up with a way that can somehow compare the plausibility of their approaches without having to resort to, or, at least, heavily rely on, the properties of Languages X or Y. In other words, I'd like to be able to decide whether the arbitrariness in Y's way of segmenting the text + their use of the wild-card has a... – Pavel Jetušek Jul 22 '18 at 10:22
  • 1
    ...bigger and more negative impact on the validity of Y's ultimate interpretation of the text (that is, trying to segment the text until it makes some sense to them) than X's uncertainty as to whether, occasionally, a particular syllabogram represents a CVR syllable rather than the simple (and much more common) CV syllable (R represents 2 nasals, 2 liquids and one sibilant, that is 5 different consonants altogether). Also, X's language uses rhyming (C1V-C2V) sequences for C1C2V syllables. To simplify things, I guess we can assume similar word-length frequencies in both languages too. – Pavel Jetušek Jul 22 '18 at 10:32

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.