15

Conlangs are sorta infamous for taking a lot of phonemes to say things. This is mainly due to so many avoiding inflections and preferring agglutination when they want to tack on a lot of information.

Its not unheard of for this to be criticized. However, in the real world, do speakers really care about how long it takes to utter a sentence? There is a preference for more common words to be shorter in length than less common ones, but that's purely relative. English may see a six-phoneme word excessive, but this may not be so for a language that only allows CV syllables. In fact, in such a language its probably not too uncommon to see words that approach twice that length.

However, I've seen Japanese speakers claim the reason they like to omit so many noun phrases is because it would 'take too long to say things'. Keep in mind, this is a language where the word for 'I' is six phonemes in length (well, the most commonly used one anyway), and every noun phrase must come with a postposition that is up to 4 phonemes in length. Also, Japanese is agglutinating so its verb endings can easily get to be longer than the root verb itself.

I've sort of run into this myself in my own conlang. I decided to require all morphemes to be two syllables in length for reasons of parsing (why I'm doing this isn't relevant here). However, I have a hard time imaging this being very practical. I mean, all adpositions must be at minimum 4 phonemes in length (the only way to get down to 2 would be for both syllables to be a bare vowel). Pronouns must also have the same problem, and tense markers, and plural markers, and articles, and demonstratives, you get the idea. I've actually been leaning heavily into making the thing super inflecting just to keep the phonemes down, but with the limited phonotactics this can only go so far; the person and number marking on the verb alone takes up about half the possible morphemes. I've really been thinking that maybe I would have been better off making it monosyllabic like Chinese and the conlang Votgil.

Really though, would an actual user care? Like, would people really be bothered if all plural nouns were at minimum four syllables in length? I've found that sentences in this conlang wouldn't always be longer than their English counter parts, but that mostly comes from it omitting information like definiteness and plurals (I figured that the verb agreement alone could indicate number, at least on core arguments). Again though, in the real world, would anyone really care? There is a strong preference clearly for commonly occurring particles, like articles and adpositions, to be monosyllabic. They don't seem to mind two-syllable singular pronouns all too much though. Is there a language on this planet where monosyllabic morphemes are either rare (this isn't unusual for languages whose phonology only allows a few dozen, I'm more talking languages where such words aren't commonly occuring), or just straight don't exist at all? I find that highly unlikely.

Still, point is, is there really such a thing as 'too long winded'? Do people really care if their sentences take as long to utter as Lojban? Or those various Finnish-like conlangs out there where seemingly all words well exceed over a dozen phonemes. Obviously no one would speak a language where it takes literal minutes to utter a single noun. So clearly there must be an upper limit, but where is it? Would a conlang like mine annoy people? Would they care about Lojban or the Finnish-like conlangs (ignoring how utterly alien Lojban is of course)? What is the upper limit you normally see? Has anyone ever collected such data?

15
  • 3
    I don't think you can directly compare English "I" and Japanese "watashi" because English speakers need to use "I" a LOT more often - Japanese speakers would simply drop it whenever convenient. (Also, I think an average phoneme duration is probably shorter for Japanese, considering its very restricted sound inventory.)
    – jick
    Commented Apr 3 at 20:13
  • 11
    I could almost imagine a Japanese-speaking conlang designer asking "But are English speakers actually fine with saying 'I' every time they start a sentence about themselves? That sounds awfully inconvenient, but what do I know? Apparently English speakers are fine with that, but how far can I push this?"
    – jick
    Commented Apr 3 at 20:16
  • 6
    This is a little too tongue in cheek for an answer, but any language which takes too long to say "Look out! Tiger!" doesn't live very long.
    – Cort Ammon
    Commented Apr 4 at 2:05
  • 2
    @CortAmmon In Malay any sentence trying to fully say "Look out, Tiger!" would take 7 syllable at minimum. However when there is a tiger around people would shorten it to simply "T'gr!" (basically the word for Tiger takes 3 syllable: "Ha-ri-mau" however in an emergency people would just shout "Ri-mau").All these shortening is obviously ungrammatical and the word "Rimau" is not in the dictionary but is perfectly understandable by anyone speaking Malay (and I also believe Indonesian)
    – slebetman
    Commented Apr 5 at 9:27
  • 3
    Please don't vandalize your own posts. When you post here, you give Stack Exchange the right to distribute the content under CC-by SA 4.0. Any vandalism will be reverted.
    – greg-449
    Commented Apr 8 at 6:30

3 Answers 3

25

Yes, there does seem to be a limit. Spoken human languages don't like to take too much time, or too little time, to say something.

(I fully realize I've cited this paper far too many times on this site, but it keeps being relevant!)

Coupé, Oh, Dediu, and Pellegrino 2019 showed that, while languages encode different amount of information per syllable, and pronounce different numbers of syllables per second, the product of these two values is much more consistent across languages than either of them. In other words, languages that encode more information per syllable pronounce fewer syllables per second, and vice versa.

They speculate that this is a property of human communication in general, which evolutionary forces will push languages to match. If a language has very little information conveyed per syllable, people will start pronouncing more syllables per second to compensate—it's not a problem if one or two of those syllables get lost in transmission, since you have so many of them for redundancy. But if a language has a lot of information conveyed per syllable, people will start pronouncing fewer syllables per second—if any syllable gets lost in transmission, that's a big deal, so you have to be more careful with them.

12
  • 4
    To put it succinctly: the information rate is pretty consistent between spoken languages?
    – gidds
    Commented Apr 4 at 15:07
  • 1
    It's probably best not to think about how consistent your answer and my comment are with that, then :-)
    – gidds
    Commented Apr 4 at 15:44
  • 1
    To put it even more succinctly: language baud rate changes to meet human bandwidth.
    – jacob
    Commented Apr 4 at 16:35
  • 1
    @gidds It's a little bit contingent on what you count as "information". The mentioned paper bases it on number of distinct syllables (or rather the entropy of the syllables) when reading a provided pre-prepared text. I don't think they specifically looked at larger-scale semantic "information", like summarizing the conclusion of a paper or sufficiently describing how to properly assemble flat-pack furniture.
    – R.M.
    Commented Apr 4 at 17:44
  • 4
    @R.M. They did, a little bit—they cite an earlier study that tried to measure the "meaning density" (my terminology) of a language by getting several different experts to translate several different texts between languages, then averaging the ratio of syllables needed to express the same meaning in different languages. Their "information density" (conditional entropy on syllables) turns out to correlate extremely strongly with this, and is much cheaper to measure, which is why they use it.
    – Draconis
    Commented Apr 4 at 17:54
6

I cannot resist jumping in here. At first I thought this should be a comment, but it's really more of an answer, albeit an answer from a personal perspective, not literature based, and not necessarily human based (although I address that at the end).

I expect to lose reputation points because of this answer. Nevertheless it is fun, and it has allowed me to make a point about language design that is important. And which I think is relevant to the question, although perhaps in a slightly tangential or offbeat way.

===

I use speech recognition and voice control on my computer. I am using it right now to dictate this answer. I do this because of computeritis/RSI problems.

Voice control is useless without a good set of speech commands. Most of which I write myself.

(UI designers who think that accessibility is just about being able to click menu items and buttons by voice are just plain wrong. Better than nothing, but not good enough. At least not good enough to make me as productive as I used to be when I could type 16 hours a day without pain.)

I definitely and strongly emphasize that how long my speech commands take to utter is extremely important.

E.g. one of the first speech commands that I wrote on my very 1st day was "Go To Title" (of a Microsoft OneNote document where I was recording my progress). Within 15 or 20 minutes I shortened that to just "Title". This happens over and over and over again. I often start with a long less ambiguous command, and then provide a shorter alias command, keeping the original one in case I need to disambiguate.

Full Disclosure:at the moment what I actually say is "DRAGO Title". "DRAGO" is a prefix that I use to indicate that I am giving a voice command, and that I do not want what follows to be treated as dictation and entered into the OneNote page or the web browser text box that I'm entering into. My use of such prefixes is a bit unusual. The Dragon speech recognition system's standard way to do this to say "Start Command Mode" or "Start Dictation Mode", but I find those too slow to say. Indeed, apparently most people do: at the moment most people use "Start Normal Mode" or don't change from normal mode at all, and just speak a mixture of commands and dictation. You try to avoid saying things that can be mistaken as a command when you are dictating. Or hope that if such a problem does occur, it does not result in loss of data. Dragon added this normal mode later, in response to user demand, so apparently normal users objected to 4 or 5 syllable mode changing commands used frequently. My use of a speech command prefix is unusual, but not completely so, other users sometimes do similar things.

Why does it matter how long it takes me to say a speech command? Well, why do you like shorter keyboard shortcuts than longer? If you are somebody who can type or use a mouse, imagine how frustrating it would have to be if you had to type "Make this into a to do list item" rather than simply whatever keyboard shortcut your editor has. In EMACS org-mode, simply typing "[{space}]", 3 keys. Or a speech recognition user having to say "open square bracket spacebar close square bracket]" - that's pretty damned onerous. Time yourself doing it!

Moreover: the more you have to utter out loud, the more strain there is on your voice. That's inherent.

There's also a limitation of the Dragon speech recognition system: speech commands are pretty much limited to the amount of stuff that you can say on a single breath, in a single utterance. So if you want to have an open-ended speech command like "Italicize <...>" (or my "DRAGO Italicize <...>) the amount of stuff you can squeeze into the <...> parameter is already reduced by 3 or 5 syllables. One of the things I hope to do if I can ever change away from the proprietary Dragon speech recognition to open source system is fix this, to allow utterances to be spread across multiple breathes.

(You may ask why I don't use a shorter speech command prefix. Simply put: I've tried, but I haven't found any single syllable prefixes that don't have a high error rate. E.g. my 1st such prefix was "PUFF", a single syllable that amused me greatly ("Puff the magic Dragon..."). But it had too high an error rate. Especially given the sibilant Fs at the end. I've tried quite a few others, but "DRAGO" and "DRACO" are the best I found so far. You may ask why use a speech command prefix at all? Simply put, it works for me. One of my friends says I use the speech command prefix the same way EMACS uses esc-X, as opposed to the way VI and VIM I have distinct normal mode and insert mode. Some people like the EMACS approach, some people like VI/VIM)

You might say that this is irrelevant to a question about human linguistics. linguistics.stackexchange.com is about human linguistics, not computer speech user interfaces.

True, but:

(1) Most users of speech recognition and voice control on their computers use human language. Albeit in a stylized manner.

(There are people, with popular videos on YouTube, who use completely nonhuman voice control for their computers. One of my friends calls that the "squeak, gurgle, and grunt" style of voice control. But while that is a thing, and gets lots of views on YouTube, it is certainly not predominant amongst human users of voice control systems. Related: Speech recognition software is tuned for human language dictation. In theory your speech commands could be whistles or other sounds not part of normal human speech. I've read dissertations about this. But again, it's not popular in the marketplace.)

(2) Let me suggest that using voice control for your computer is not that different than providing instructions to other humans. Except... my wife sometimes accuses me of talking to her too abruptly, as if I was talking to a computer. But she also said this before I started using speech recognition. Nevertheless, saying things in small words, fewer syllables, efficiently, is often understood more clearly by other humans, as well as by computers. Perhaps not quite as short as for computer voice control: imagine yourself doing tech support over the phone. You might say "John, go to the title. Type..." rather than "John Title". But you certainly would not want to increase the length of what you were telling John to do 2 or 4 syllables, just because you have agglutinative modifiers.

You say "[you've] seen Japanese speakers claim the reason they like to omit so many noun phrases is because it would 'take too long to say things". Certainly, when I reduce a speech command from "highlight the current sentence" to "highlight", I'm omitting the object. When I say "title" rather than "go to title" I am omitting the verb. Conventional non-speech oriented GUI user interfaces that apply commands to the current selection use that as a way of omitting stuff that would take too long to type. Same thing applies to speech. On the other hand, when I use my speech command prefix and say "DRACO title" you might argue that I am leaving in the thing that I am asking to to the action, and omitting the verb and the place to which the action is being directed. But although I like using the speech command prefixes, if I could get rid of them I would love to.

Problem is, when you drop noun phrases or verb phrases or ... it often leads to ambiguity. Sometimes ambiguity is no problem, especially when you are talking to a person who might ask you to clarify. However ambiguity can be a big problem when you are giving a command to a computer, especially one that does not have an undo function. Very few speech recognition systems will ask you to clarify. Perhaps ChatGPT will change this. Back to humans: frequently humans do not ask you to clarify an ambiguous statement or command, they just make the wrong assumption. That can be unfortunate.

If fewer syllables, phonemes, and overall less time is required to clearly and unambiguously say something, whether to a computer or human, all the better. If you have to omit things (assuming they can be implicitly assumed, or is that assumed implicitly :-) ), it can lead to problems.

--

Besides: you are talking about conlangs. Can you imagine Klingons talking with long agglutinative nouns and verbs and other parts of speech? Unfortunately I am not fluent in Klingon. :-(

===

I expect to lose reputation points because of this answer. Nevertheless it is fun, and it has allowed me to make a point about language design that is important.

3
  • @Lambie: who is talking about "machine reproduction of what people say"? I was talking about what I say to a machine, which I believe is also related to what somebody might say giving commands to some other humans, or a trained animal. The OP's question was "Do languages have any sort of upper limit for how long it takes to say things?", and for my example the answer was "Yes, as much as you can say on one breath." Not true in general, but true for my example. Probably significant elsewhere. Could verify by experiments and observations in other situations.
    – Krazy Glew
    Commented Apr 8 at 19:33
  • Nevertheless, thanks for not downvoting me. I lurk on linguistics.se because I like languages, and to gain insight into design of voice control systems - what the human actually says, not the code that recognizes it. But when I saw this question, I could not resist...
    – Krazy Glew
    Commented Apr 8 at 19:38
  • I take your point. You like voice control systems. The problem is that the question doesn't work. If I said to you: What is the depth limit for scuba diving? There is a precise answer. But "languages" do not have limits about how long it takes to "say things". The question is naive. What is "say things" anyway? I can talk for an hour, is that "saying things'? Get my point? People can take a breath and keep talking.
    – Lambie
    Commented Apr 8 at 20:02
-1

I believe that when it comes to specific questions about whether the length of morphemes or sentences in a constructed language interests the user, it all depends on various factors such as the goals of the language creator, the target audience, and the communicative needs of the language. While some users may appreciate linguistic complexity and enjoy the challenge of mastering a language characterized by a large number of words, others may prefer languages where priorities are placed on simplicity and ease of use. The study service was used.

Your Answer

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