I'm making this since now I think an 'abjad' isn't what I actually had in mind. I was thinking of a system where the vowels don't have to be written, because they're always obvious. Essentially, my idea was more in line with a compression algorthm than a true abjad. Here's a link to the previous question if you want to read over my original idea:

Tips for making a conlang that uses an abjad?

My plan was to make a language that perfectly fit a writing system that only wrote consonants. The vowels could always be determined some other way. Here are the techniques I came up with:

  1. The phonotactics often make it obvious what the syllable structure is. Basically, approximants and non-nasal plosives can only occur in onsets. Non-nasal plosives can only be followed by a vowel or approximant. Working out all possible 2 and 3 consonant roots (conflating all non-nasal plosives, approximants, and everything else into one for each of them), I've found that for all possible combinations, there's only at most 2 possible syllable structures, which themselves are distinguished by the number of vowels used. For instance, a sequence like ky could be one of kaya or kya. The vowel template required by its part of speech would itself make it obvious how many vowels there are.

  2. The vowels to be used could be inferred by part of speech, but also declension/conjugation. The latte isn't really necessary, but I would guess that having only one possible template per part of speech would quickly get repetitive. Besides, having multiple declensions is useful for derivation, and all I would have to do is make sure the articles use different consonants for each gender.

Of course, this puts a limit on what features the language can use; obviously stem changes are not an option. As I mentioned before, you can't have conjugations like English speak->spoke. This of course puts a severe limit on the morphology strategies that could be used. This thing may need to be a highly agglutinating language where every possible affix includes at least one consonant if not more.

Of course, in real compression, different compression strategies often work better for certain data than others. A good example of this is the original pokemon games, which used one of two compression strategies for its sprites, because which worked better differed based on the sprite in question. Also, compression often isn't always perfect; you often simply can't perfectly replicate the original data based on its compressed form. Youtube's compression is infamous for this since it lowers the quality of audio. If you want perfect reconstruction of the original data, often times it just needs to have very specific qualities.

The closest I've heard of this in the real world are languages that only have CV syllables but write with an alphabet. Some chose to not write the most common vowel, since the limited syllable structure makes it obvious when a vowel is omitted. You only need to indicate what the vowel is. This is essentially similar to the concept of a null morpheme in a paradigm. On that note, perhaps null morphemes would also be good for this language? Assume a noun is neuter unless specified otherwise. Assume perfective aspect if no other marker is present. Assume singular if the plural marker isn't present. Assume a third person singular pronoun if none other is present. You get the idea.

This obviously isn't a lot though, so I would need more. Also, there's the problem with not being able to ever use stem changes, which obviously puts quite a limit on the morphology. Obviously, it would have to make use solely of agglutination, and maybe zero derivation (other than things like 'to ktb' and 'the ktb' having different vowels for the sake of minimizing repetition). I've also had someone recommend having words be really broad in meaning to help get the most out of all possible 2 and 3 consonant words. What other things can I do though, and what else may I have to worry about?

edit: Before anyone says it, yes I know this would make it hard to incorporate loan words. I don't see that as an issue for this language though. Besides, I could always just literally translate proper nouns literally based on their etymology if I really need to.

  • Are you aware of Devanagari? The writing system you imagine looks pretty close to the system employed Devanagari: There is in inherent default vowel (short a in case of Devanagari) to every consonant, and only other vowels are explicitly written. Some devices for absence of vowels are implemented (virama, ligatures, special treatment for post-vocalic nasals and r's).
    – Sir Cornflakes
    Commented Jan 3 at 14:14
  • The other direction to look at are some shorthand systems and what tricks they invented to speed up writing.
    – Sir Cornflakes
    Commented Jan 3 at 14:25
  • Well, I do know about Cree Syllabics, where they mark vowels by rotating the starting consonant. I did try to make a system like that once, but I found it limiting. All possible rotations of any one symbol had to represent the same consonant, so this made it sorta hard to come up with characters. Also, all possible characters could not be symetrical in any direction due to how vowels are marked (so no l or -). Also, the rotation thing limits you to just 4 vowels (which works fine for the language it was made for), or 3 if you want codas.
    – user6046
    Commented Jan 4 at 0:16
  • Looking up shorthand, it seems rather similar to what I had in mind. Some forms of shorthand only write consonants (Arabic itself ironically started out as a form of shorthand). It also likes to make use of abbreviations, which is something I know Tagalog does for its articles/prepositions. Shorthand though seems to mostly rely on using using characters that can be written as quickly as possible. I'm not really interested in doing that (I'm more concerned with saving space on paper rather then increasing writing speed), but the abbreviation thing is something I could consider.
    – user6046
    Commented Jan 4 at 2:53
  • Also, I forgot to mention that yes, I know what devanagari systems are. Though I was trying to avoid using super elaborate characters like you see in syllabraries.
    – user6046
    Commented Jan 4 at 2:53

1 Answer 1


Assume: compression requires patterns.

If vowel change rules are absolutely consistent and productive, then you could perhaps match a vowel "naturally" with its parent consonant, then the spoken language would alter the vowel for grammatical purposes which are not recorded because reasons.

So the recorded language might be "lossy compression" so to speak.

e.g. "M" is an "E"-class letter, and "K" and "R" are "A"-class letters, and let's pretend that final vowels are elided, so when you see

MRK "to throw"

You naturally read it "MERAK".

But if someone was speaking about events in the past, he might have said "MERIK", "threw". That information is lost in transcribing MRK, so the reader hopes he has some context to retrieve the verb tense.

Something a little less lossy would require some sort of vowel letters or notation, and it assumes that the LAST vowel to be used is to be REUSED until you explicitly write a new one. So:



Unless you have some exceptions that always allow certain consonant clusters to not be blown apart (like doubled consonants above).

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.