Calculating writing system efficiency with respect to reading ambiguity?

Question

I have been thinking of developing a software tool that would make it possible to calculate the efficiency of a particular writing system (attested rather than hypothetical) for a particular language (primarily attested, usually ancient and thus also extinct) with respect to the ambiguity of interpretation/reading, whether this ambiguity is caused by homography or any other "deficiencies".

Primarily, I would like to use the tool to find out which of the scripts or writing systems compared make it possible to represent a given [set of] language[s] in writing more unambiguously, especially, though not exclusively, with respect to phonetics and phonology.

Since we may not have access to native users of the scripts, any psycholinguistic testing would have to be carried out only indirectly, on a group of speakers of a different, but typologically similar, language and, perhaps, using romanization/transliteration to avoid having to teach them new graphemes and face literacy/familiarity issues.

Still, while I can imagine constructing tests like this, I would like to have a tool that could use as its imput relevant features of the particular phonology PLUS features of the script/writing system used to represent strings of text of the language. Its output would then be some kind of relative or absolute enumeration, a function whose arguments would be the...

1. number of phonological combinations a particular string can represent
2. their subset with respect to the given language's phonology and phonotactics

Or, more simply, how many (phonotactically permissible) texts can be transliterated by a particular string of text using the script, i.e. how many (phonotactically persmissible) texts can a particular string of text written in the script represent (as compared to another script). It follows that we would always have to apply the tool on existing text and thus, in fact, measure ambiguity of reading a particular text written in a particular script as compared to the same text being writen in another script.

Hopefully, the following example based on a syllabic script, originally fitted to write a different language, but later (imperfectly) adapted for Ancient Greek, will show better what I mean.

"Case Study": Linear B

The predominantly syllabic writing system of Linear B displays a lot of ambiguity with respect to the phonology and phonotactics of Mycenaean Greek language it was used for. Some examples include...

• no disambiguation between /r/ and /l/
• no voicing/aspiration disambiguation in most obstruents
• no marking of initial aspiration
• no marking of vowel length
• omission of sonorants and /s/ in syllable codas

etc.

On the other hand, the script as used by Greeks was capable of representating sonorant-free consonantal clusters by means of two consecutive syllabograms sharing the same vowel component (i.e. "rhyming"), and it delimited words/phrases by means of a vertical stroke between them. (For the sake of simplicity, I am deliberately ignoring non-phonetic graphemes, the ideograms, the script could employ to abbreviate or clarify the meaning.)

Hence, a string of two syllabograms such as <KA> and <RA> could, in theory, represent the following phonotactically permissible (though not necessarily attested) combinations (note that M stands for any of /m, n, s/ medially, and /m, n, r, l, s/ finally):

(s)K(V)(M)lV(M)

1. /gala/, /gla/, /galaM/, /glaM/, /gaMla/, /gaMlaM/
2. /kala/, /kla/, /kalaM/, /klaM/, /kaMla/, /kaMlaM/, /skala/, /skla/, /skalaM/, /sklaM/, /skaMla/, /skaMlaM/
3. /kʰala/, /kʰla/, /kʰalaM/, /kʰlaM/, /kʰaMla/, /kʰaMlaM/, /skʰala/, /skʰla/, /skʰalaM/, /skʰlaM/, /skʰaMla/, /skʰaMlaM/

(s)K(V)(M)rV(M)

4. /gara/, /gra/, /garaM/, /graM/, /gaMra/, /gaMraM/
5. /kara/, /kra/, /karaM/, /kraM/, /kaMra/, /kaMraM/, /skara/, /skra/, /skaraM/, /skraM/, /skaMra/, /skaMraM/
6. /kʰara/, /kʰra/, /kʰaraM/, /kʰraM/, /kʰaMra/, /kʰaMraM/, /skʰara/, /skʰra/, /skʰaraM/, /skʰraM/, /skʰaMra/, /skʰaMraM/

You can see now, that a simple <KA-RA> may have tens of possible interpretations. Clearly, reading Mycenaean clay tablets must have been very difficult at times, especially in case of lists of personal names.

Imagine now a three-word clause, each (graphical) word composed of two syllabograms and, in accord with the rules, delimited by the vertical stroke symbol, such as the following:

<KA-RA|KO-RO|KU-RU>

The number of possible interpretations increases dramatically, on the other hand, there are some morphosyntactic and semantic constraints and cues which reduce the tremendous ambiguity a bit.

What if we, however, unlike in the mainstream Mycenaean Greek reading, but like the fringe (pseudo-)Proto-Slavic reading, (1.) interpreted the vertical stroke not as a word delimiter, but a consonantal wildcard - one that is capable of representing up to a dozen different consonants. (2.) Consequently, word boundaries could be placed anywere in the string, between any two syllabograms. (3.) In addition to that, consonant clusters could no longer be resolved by means of consecutive rhyming syllabograms, and (4.) neither sonorants nor /s/ could be thought of in syllable codas.

While (3 & 4) would reduce the number of options, (1 & 2) would lead to their increase (X stands for consonantal wildcard, | stands for arbitrary segmentation done by the reader):

1. kalaXkoloXkulu
2. kalaXkoloXku lu
3. kalaXkoloX kulu
4. kalaXkolo Xkulu
5. kalaXko loXkulu
6. kalaX koloXkulu
7. kala XkoloXkulu
8. ka laXkoloXkulu
9. kalaXkoloX ku lu
10. kalaXkolo Xku lu
11. kalaXko loXku lu
12. kalaX koloXku lu
13. kala XkoloXku lu
14. ka laXkoloXku lu

etc. up to something like ha raž ho rož hu ru (the alternative, pseudo-Slavic, system also fails to distinguish /h/, /x/ and /k/ on the one hand, and /r/, /ř/, /l/ and /ľ/ on the other, along with other missing distinctions in other consonant rows).

Having described (in a simplified form) the two different systems intended for two different languages, I would like to ask:

Does the version of the system as used for Mycenaean Greek produce more phonotactically acceptable outcomes, leading to greater potential ambiguity, or is the Slabic system that may lead to greater potential ambiguity, producing generating/representing larger numbers of phonotactically acceptable readings?

Intuitively, I feel the mainstream view has a potential to produce less ambiguous results, but how can we quantify that? Once again, I should stress that this would only be about a potential ambiguity, which could only be confirmed or refuted by psycholinguistic experiments, as correctly pointed out by commentators.

Similarly, we can generalize the question and set the software tool to find out, whether reading a text T writing by means of writing system W1 leads to greater nubmers of possible readings than writing system W2, whether also representing the same language, or a different one completely.

Some Follow-up Remarks:

What are the most significant caveats never to be neglected? Some questions to be asked and parameters my tool should obviously take into account and probably include:

• the word/phrase delimiters (or any other punctuation):

  • Are there any?
  • Do they vary (un)systematically?

• the grapheme-to-phoneme mappings:

  • Is the system capable of representing ALL phonemes of the language paradigmatically?
  • If so, is it a simple 1-to-1 mapping?
  • If not, i.e. if multiple graphemes have to be employed to cover some phonemes, can any syntagmatic ambiguities arise (e.g. /ts/ as a biphonematic sequence or an affricate)?
  • If not, i.e. if one grapheme can ambiguously map onto a number of different phonemes (e.g. liquids indistinguishable in Linear B), what constraints can be observed or defined, if any?
  • Is the system capable of representing all of the phonemes syntagmatically, i.e. are any phonemes (un)systematically omitted/lost?
  • If so, what are the rules and constraints?

Others to be added...

Answer 1

The discussion of Linear B doesn't make any sense to me, so I will skip that part. Here is how I understand your question: you want to know what properties of a writing system optimize accurate identification of {the word / the phonetic sequence} of strings in language. There should of course be a provision whereby speakers of the language have some idea what a given word is, when they don't know the word (e.g. "squamulose", whose meaning many people don't know but could look up, or "Wahkiakum" which people may not be able to pronounce or spell based on hearing it said). The problem is that efficiency can only be computed by psycholinguistic testing of real people, in which case you have to ask "Who?" Hangul is terribly inefficient as a vehicle for representing Korean pronunciation, taking all humans as your subject pool, but limiting the subject pool to just speakers of Korean, it turns out to be quite efficient.

There are a few cases where there are actually competing writing systems, such as Cyrillic and Latin for Serbo-Croatian, or Lule Saami in Norway and Sweden. Generally speaking, you would always be comparing a hypothetical use of one writing system to an actual use of another writing system -- such as English spelling, vs. IPA transcription; or Cyrillic Russian vs. conventionalized Latin transliterations. The problem is, if you test Russian speakers on their ability to read actual Cyrillic Russian which they are familiar with, vs. a foreign transliteration system that they are not familiar with, you can't be sure whether you're measuring an effect of the writing system, or simple familiarity (I vote for the latter). This would also apply to the aforementioned cases of two writing systems for a language – you would have to control for familiarity, and I don't think there is a practical but valid way to conduct the experiment.

Your question refers to "phonemes", a concept which unfortunately is not all that well defined. In some theories, English flap is not a phoneme (it's claimed to not be underlyingly necessary), and in some theories it is (it merges two things which are phonemes, /t/ and /d/). So English spelling either does or does not paradigmatically represent all of the phonemes of English.

Although Russian has a phonological rule devoicing final obstruents, which changes one phoneme into another, the spelling system does not reflect that fact (same as in German). That means that the writing system is "suboptimal" in representing pronunciation, but it is actually optimal for word identification, since порок and порог are spelled differently. It is not clear whether you would consider this a case of one grapheme <г> mapping ambiguously to multiple phonemes: it certainly maps to multiple phonemes, but it is contextually predictable. The pronunciation of <г> in the genitive (as [v]) is predictable according to the rule "pronounce <г> as [v] in the genitive singular case ending of adjectives". Do you mean "referring to nothing besides preceding and following letters"? It's not obvious what you mean by "ambiguously map onto a number of different phonemes".

Answer 2

The problem is with formulating the task as reading ambiguity. Ambiguity is a function of context not simple homography. Spoken language is full of all sorts of ambiguity some of which is reflected in written language. Orthographies introduce new ambiguities and remove others. For instance, the example of final devoicing given in the other answer. It reduces potential ambiguity with respect to word derivation but introduces potential ambiguity with respect to word shape. But neither of these represent real ambiguities for actual speakers of the language.

What's worse, because the types of ambiguities and the corresponding disambiguation strategies will be different across languages, whatever measures you use for comparison will be highly suspect. How do you compare the impact of the unmarked word stress in Russian with the unmarked vowels in Arabic or the unmarked tone in Pinyin? They all have very specific consequences which cannot be captured by a single number.

You could look at it from the perspective of the learner (native or non-native) but even then, you need a much less precise number. How do you compare the precise transparency of e.g. French, Italian, Spanish and Greek orthographies? We can intuitively say that French and Greek are similar and Italian and Spanish are also similar. But from the perspective of a learner (or any other practical perspective I can think of), the precise ranking of Italian vs. Spanish is irrelevant.