I'm interested in the idea of a conlang which could be used to create extremely information dense statements - such that, for example, a single word could be built up from morphemes to express a complete sentence. It would probably be a highly agglutinative language, similar to Eskimo-Aleut languages.

Are there successful* examples of conlangs which have been created with this principle in mind? I'm interested in seeing how other people might have applied this principle in creating another language.

*When I say successful: I'm aware of fictional examples like Speedtalk, where each phoneme is meaningful, and has incredibly complex and subtle distinctions. I don't consider that successful, because it's far too complex for anyone to reasonably learn. Japanese verbs can be informationally-dense in the way that I mean in terms of its inflectional morphology, e.g. the single word 行かせられたくなかった (ika-se-rare-ta-kuna-katta) can express the complete sentence "[I/he/she/they] didn't want to be made to go," using inflectional morphemes.

  • NB: I'm incredibly pleased to see that there exists a Conlangs SE. I hope that I can make a meaningful contribution through my question to support this site's beta. – Lou Aug 9 at 14:21
  • Relevant: Fictional language "SpeedTalk" – Jeff Zeitlin Aug 9 at 19:22
  • I did acknowledge Speedtalk in my OP :). I don't consider it to be a successful informationally-dense conlang for the mentioned reasons (in addition to the fact that it's not a real fleshed out conlang.) – Lou Aug 9 at 19:23
  • 1
    Wups, sorry; missed the reference to it in your question. :) – Jeff Zeitlin Aug 9 at 19:56
  • No worries at all! :) – Lou Aug 9 at 22:08

Answering this question is tricky. As you obtain higher and higher levels of information density, you have to sacrifice some naturalism or some simplicity in order to get there.

Marking the point where you've sacrificed too much naturalism or simplicity to be learnable is a judgment call.

Short answer:

Possibly guaspi, but it's obscure.

One famous example of this is Ithkuil. Parts of Ithkuil are reasonably naturalistic, such as its phonology, which is less complex than Ubykh or Chechen (which also has a large number of vowels). Other parts are not, such as the very fine-grained case distinctions or how meanings are organized into roots, which is more complex than the system of triliteral roots in Semitic languages that it is based on.

Is Ithkuil learnable? I really don't know. My guess is no but I don't have proof.

An example of an informationally-dense written language is Classical Chinese. Classical Chinese is not a conlang, but it was written centuries after it stopped being spoken and uses a heavily abbreviated style where nearly everything that's pragmatically inferable is dropped.

If you want to make an informationally dense conlang, Classical Chinese is a good starting point. It has pervasive zero marking (although you can optionally mark possessors and relative clauses with an overt marker) and the reconstructed roots are monosyllabic or sesquisyllabic.

One possible example of an intentionally information dense language is guaspi, a tonal derivative of lojban.

I don't know whether lojban is considered learnable or not and I don't know whether to consider guaspi successful or not, but efficiency is one of its stated goals.

Gua\spi is efficient. Words are short, and extensive defaults on articles and modal cases eliminate the majority of structure words.

In my personal opinion, so take it with a grain of salt, a human can speak "informal" lojban riddled with logic errors and this should count as speaking lojban.

Closing remarks:

Looking at some of these examples reveals an important design decision that creators of an informationally-dense language face: whether to allow the speaker to drop things that are pragmatically inferable and how to account for that when evaluating how well their language achieves its goals.

For instance, suppose a language has very specific tenses that are marked on all verbs or person marking on all verbs. Tense information is frequently recoverable from context, as are pronominal arguments. If information is recoverable from context, does the language get any points for encoding it directly, however efficiently it may do so?

| improve this answer | |
  • The other answers have pointed out eloquently why information density wouldn't work as a design philosophy for a conlang, but I'm accepting this answer because you've provided examples of conlangs which have nonetheless tried. – Lou Aug 23 at 14:19

There is a general trade-off between two aspects of any encoding or language: redundancy versus information density.

If you have an information-dense language, that means there won't be much redundancy (as every symbol has a distinct meaning). This makes for efficient communication in perfect conditions, but as soon as there is any noise (in the widest sense of 'noise'), communication could get disrupted. If any symbol is changed during transmission (ie misheard, or dropped), then the meaning of your message will change in a way that cannot be recovered. It would also require perfect production, ie no typos or other mistakes, as they would all change the meaning. Because there is no redundancy, you cannot even recognise a typo, as it would simply be another, different, word. That of course is in the extreme case only. Imagine you numbered all the words in your dictionary and would simply transmit a sequence of numbers.

Another aspect of information density is not in the language but the texts created in the language. Any text requires shared context between author and recipient for the receiver to be able to interpret the message. Anything not shared between the sender and receiver needs to be contained in the message itself. So even if your language itself is very information dense (or redundant), your messages don't have to be the same.

There is probably a natural limit for information density in human languages. If every letter you write has an impact on the meaning of your message, you need a huge cognitive effort to create it (either when learning the language, or when writing a text in it). And similarly, the receiver needs to spend a lot of effort into decoding/understanding your message. So while it is space efficient (you get away with shorter messages), the time efficiency is pretty poor. And if your message content gets lost in the process, you need to re-transmit, which adds to that as well.

For those reasons languages tend not to be on the extremes of density (or redundancy). And any conlangs that would go there, would be hard to use. So it doesn't really make sense to have this as a design goal for a language.

| improve this answer | |
  • To use an example, consider the sentence "This is atypical behaviour for Susan." Accidentally omit a single letter so it reads "This is typical behaviour for Susan" and you've completely inverted the message. – Keith Morrison Aug 12 at 17:57
  • I could see a dense language being spoken by alien androids for that reason, but then again they would probably prioritize a language that is easy to parse and non-ambiguous, not necessarily dense. – Domino Aug 12 at 20:33
  • @KeithMorrison You also only need a slight hesitation or different stress after the a to make it a typicial... – Oliver Mason Aug 13 at 11:08
  • @OliverMason, good point, and pronouncing the a as [ɑ] or [ə] (how I often pronounce the article a) instead of [e] would do the same thing in my dialect. – Keith Morrison Aug 13 at 17:06
  • This is a really good answer, and I would accept this as well if I could. You've explained elegantly why information density isn't a good criteria for a conlang. – Lou Aug 23 at 14:17

To use your example from Japanese: ika-se-rare-ta-kuna-katta Yes, all one word. A word of 10 syllables, 6 morphenes. What, precisely, is the advantage over "didn't want to be made to go" (8 syllables, and 8 morphenes)? It only works if you decide that "word" is the most basic unit to measure against, but "word" can be a very arbitrary concept.

In a 2010 paper, a study was done looking at the "information density" of speech; that is, given the same text translated into multiple languages, so everyone is conveying the same information, how long would it take a fluent speaker of that language to transmit that information, speaking at a normal cadence (and using multiple speakers to get an average, of course). As part of that process, one step was determining how much information was conveyed per syllable. They used Vietnamese as a baseline, arbitrarily giving it a value of 1.00. If a language conveyed the same information in more syllables than Vietnamese, it would be lower than 1 (you need more syllables to communicate the same information). English was 0.91. Japanese was 0.49. In other words, you needed nearly twice as many syllables to communicate the same thing in Japanese as you did in English, Mandarin (0.94), or obviously Vietnamese (1.00).

When you compare speaking rates (how fast those syllables are spoken), it turns out that most languages are about the same when it comes to transmitting information per time, between 0.9 and 1.1 (Vietnamese again being 1.00). Except Japanese, which although having easily the fastest speaking rate of the languages tested, because of the extremely low density of information per syllable, they had the lowest rate of transmission by far, at 0.74.

So Japanese is significantly less information dense than many other widely spoken languages.

You see the same thing in other agglutinative languages. Just because a given language can pack more information into a single word doesn't mean that it will necessarily be shorter to communicate. I work with Inuktitut translations all the time, and I just pulled up an example I've dealt with:

The Proponent intends to conduct a mineral exploration program including drilling, sampling, magnetic surveying, and mapping.

The Inuktitut translation is:

ᐱᓕᕆᔪᒪᔪᖅ ᐅᔭᕋᖕᓂᐊᕐᕕᒃᓴᖅᓯᐅᕈᒪᒐᒥ ᐱᓕᕆᓇᔭᖅᓱᓂ ᐃᑰᑕᖃᕐᓂᖅ, ᐲᔭᐃᖃᑕᕐᓂᖅ ᖃᐅᔨᓴᒐᒃᓴᓂᒃ, ᓇᐅᒃᑯᑦ ᓂᐱᖓᓂᖃᕐᓂᖏᓐᓂᒃ ᓄᓇᐅᑉ ᐃᓗᐊᒍᑦ ᐅᔭᕋᖏᑦ ᐊᒻᒪᓗ ᓄᓇᖑᓕᐅᓂᕐᒥᒃ.

Which transliterates as:

Pilirijumajuq ujarangniarviksaqsiurumagami pilirinajaqsuni ikuutaqarniq, piijaiqatarniq qaujisagaksanik, naukkut nipinganiqarninginnik nunaup iluagut ujarangit ammalu nunanguliunirmik.

Obviously the English is much more concise simply based on inspection even though it has more words. Now, an argument can be made that the English has some terms which might require a longer word to explain it in Inuktitut, so that's a fair argument. So here is a more "common language" example from the Government of Nunavut's website (https://livehealthy.gov.nu.ca/en/health-topics/injuries/preparing-hunt-land):

Telling someone at home where you are going. Even if you are going out on the land for just a few hours, you should tell at least one person at home the names of all riders and passengers, where you are going, and when you plan to come back. If you do not return in time, this person will be able to send for help.

And here is the Innuinaqtun version:

Uqaqlugu kimut humungauliqtutin. Aulaaqhimanahuaqtillutin ikituni ikaangnini, uqaqtukhauyutin atauhinaugumi inuk aimavingni kitkut aulaaqatiniatatin, humungauliktutin, humi utiqnahuaqtuninlu. Uttinngitkuvin mikhaatigun, tamna inuk ikayutikhangnik aullaqtittniaqtuq.

The Innuinaqtun is 21 words compared to the English 67. If, however, you look at the syllables, there's 109 compared to English's 73. English clearly has the higher information density per syllable. Just because the average word is longer doesn't make the language more information dense.

Part of the issue with Inuktitut, just as with Japanese, is that the number of possible different syllables is limited due to the phonotactics of the language, which means you have to use more syllables to compensate for the limited inventory you have. Standard Inuktitut has at most 210 possible syllables. Japanese has, depending on who you ask, about 400 possible syllables, so you could have that number of possible one-syllable words. With English phototactics, for "standard" English, you have about 316,800 possible one-syllable words (there is about 9,300 in use).

The reason for this difference is the number of consonant clusters allowed and the ridiculous number of vowels in most English dialects.

tl;dr summary: to have a more information dense and still practical language (so you allow some natural redundancy and not have so many near-homonyms that mininterpretation due to missing something is a constant problem), you need a language with a lot of possible syllables and phonotactics that allow those syllables to be used or combined into sufficiently many distinct morphenes, however you put those morphemes together or not in any kind of language, whether isolating or agglutinative.

| improve this answer | |
  • I really wanted to accept all three answers. Both yours and Oliver Mason's answers have explained eloquently why information density as a design goal is impractical, but I like your examples in particular. I accepted Gregory Nisbet's answer because it answered the direct question, but you answered the XY problem better. – Lou Aug 23 at 14:16

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.