# Minimum phonemes for a conlang?

I'm designing an alien race with limited points of articulation, and it got me thinking what is the minimum number of phonemes needed for a language to function? That would help me determine their physiology?

I've seen Is a Language with Two Phonemes Feasible? but that seems too limited, and the idea of speaking in binary doesn't mesh with my idea of this alien race.

• Not to make it eben more complicated, but are languages lacking of sound acceptable? Light, signs, magnetic or electrical fields to even forms of a biological RF or the like could be employed without any phonemes. The written word is just a different way of conveying a spoken language, so any form without sounds can be valid. – Trioxidane Mar 16 at 14:08
• You might also want to look at the Film Theorists YouTube channel, for the episode on "I am Groot" being a full language. – JDługosz Mar 16 at 14:58
• @JDługosz the Librarian from Discworld uses a similar language with only one word: "ook". It's also a programming language. (OK, the Librarian also occasionally uses "eek", but really the vast majority of his words are "ook") – VLAZ Mar 16 at 15:36
• Please note that Stack Exchange has a stack specifically focused on Constructed Languages. We embrace a wide variety of questions, but you would likely get a better answer over there. – JBH Mar 16 at 15:44
• Morse code has only 2 sounds (3 if you include the longer gap) and only 1 type of sound (only the duration differs) yet provides effective communication, albeit slower than speech as we know it. – Bohemian Mar 16 at 22:40

Depending on how you count phonemes, the "minimum number of phonemes needed for a language to function" is obviously either one or two.

In real-life linguistic descriptions, Polynesian languages feature small phonemic inventories; for example, Tahitian makes do with five vowels and nine consonants (/a/, /e/, /i/, /o/, /u/, /p/, /t/, /m/, /n/, /f/, /v/, /ʔ/, /h/, and /r/) and Hawaiian makes do with the same five vowels and only eight consonants (by having no use for /f/). The Hawaiian alphabet has only 13 letters: a, e, i, o, u, h, k, l, m, n, p, w, and ʻ (which is called ʻokina and is used to write the glottal stop /ʔ/). (And yes, that is the alphabetical order.) (Depending on who's counting, that can be ten vowels, because each of them can be short and long.)

Attentive readers have noticed that in the preceding paragraph I spoke about the number of phonemes in linguistic descriptions, and not in languages. Languages do not have phonemes; phonemes are elements of the models we make of languages. In the physical reality, all we have is (allo-)phones: phones are the physical realities, phonemes are the abstract models. To a large extent, the number of phonemes depends on the principles behind a specific language model.

To give an extreme example, how many vowels does Mandarin Chinese have? Well, that depends on who is counting. The minimum number I have seen is zero (Edwin Pulleyblank's model). The maximum is 25 (counting /a/, /ə/, /i/, /u/, /y/, each with one of the four tones or with the "zero" tone). Or, for a maybe better known example, how many vowels are there in classical Latin? Is it five, each having a long and a short version, or is it ten, or is it eleven? How many vowels are there in BBC English? The usual description (for the use of us foreigners who learn English) gives 19 (nineteen!) vowels and diphthongs, plus 4 quasi-diphthongs.

So for the purpose of creating a mock-up of an alien language, I would say that you cannot go wrong with three vowels (a central vowel /a/, a front unrounded vowel /e/, and a back rounded vowel /o/), two glides (/j/ and /w/) and a small set of consonants. Which consonants, that depends on the set of articulation points you accept; on whether you accept distinctions of voicing (as between /p/ and /b/ or between /k/ and /g/) or not; on whether you accept nasals or not; and so on.

• This is such a great answer, but it troubles me that the alien race is going to have no vowels and no ability to do glides. Slightly outside of the scope of the question, but is an even smaller inventory functional? – Pureferret Mar 16 at 23:24
• @Pureferret: That depends very much on what you mean when you say that their language has no vowels... (There is no human spoken language which does not have at least surface vowels, i.e., allophones which function as a syllable nucleus.) But practically speaking, you should play (for example in Excel or LibreOffice Calc) with various phonemic inventories and phonotactic rules, and see if you are satisfied with how many distinct one or two syllable roots can be built. Or generate sample words with Awkwords. – AlexP Mar 16 at 23:42
• the alien race have a pneumostome but don't produce sounds that way, instead they use their equivalent the equivalent of a radula to scrape and tap to make noises. I'll look at Awkwords and see if anything clicks – Pureferret Mar 17 at 12:01
• @Pureferret: You may want to look at this answer, concerning specifically the issue of representing an alien speech which uses an alien sound-producing mechanism. Basically, insted of directly representing the alien sounds, you can map the roles of alien sounds onto the roles of human sounds; sounds which can be sustained and constitute the nucleus of elementary phonation sequences map to human vowels, momentary sounds which cannot be sustained map to human stops etc. – AlexP Mar 17 at 12:19

## In theory, any number from 1 to infinite phonemes would work for a language.

Here is a run down of the lowest numbers, and how they would work

### Unary Language

Assuming you can pause between words (in a way that doesn't encode it's own data), unary works, although it likely very highly impractical:

The unary numeral system is the simplest numeral system to represent natural numbers:1 to represent a number N, a symbol representing 1 is repeated N times.2

In the unary system, the number 0 (zero) is represented by the empty string, that is, the absence of a symbol. Numbers 1, 2, 3, 4, 5, 6, ... are represented in unary as 1, 11, 111, 1111, 11111, 111111, ...

So in this language, 1 sound would reflect the first word, 11 the second word etc. The complication come when combining words in a sentence to reflect more complex concepts:

The sentence related to "1 11 111 1 111 1", could encode enough complication if, for example "1 11" and "1 111" encode more information than their parts alone.

Without pauses

### Binary & Ternary

As the Q&A you found, it goes over in good detail how binary works, as well touching on ternary.

you can even compress those words by frequency of usage simply by adopting the same trick that saved Unicode

As you can see there's a similar conception of encoding more than the sum of it's parts.

On ternary:

Tri-state let's you scrap the UTF-8 "extension bits", or gives you another symbol to play with (log(26)/log(3) now). Speed or compression gains either way.

This is the most 'efficient' number of phonemes, if we make an analogy in terms of radix economy, however as we see below there are other considerations. This is not a perfect analogy.

## More than 4 phonemes

With more than 4 phonemes, we can start to consider smaller words and even how combinations of sounds work together. One could safely assume, the most common words are also the shortest

For instance if you consider any combination of the 4 phonemes in a 3 phoneme word (assuming simple Onset, Nucleus and Coda), you get

$$\frac{4!}{(4-3!)} = 24, \text{words}$$

Is this enough for a language? Yes, in theory. Bear in mind that this assumes that, say a, s, d and f  represent our phonemes, that fff is a valid word, which may not actually be distinguishable from ff (note, two phoneme words aren't part of our calculation, but may and most likely will still exist in the language!).

How many words are required in a language to convey basic concepts? If it's more than 24, we will need more phonemes. Thankfully we have the Swadesh List for this very purpose, which claims that there are for example 100/207/35 words needed to convey basic concepts.

Given these restrictions, we can use wolframalpha to tell us the minimum phonemes to make enough 3-phoneme words to convey these concepts:

$$\frac{x!}{(x-3!)} > 100, \text{x = 5.7 or ~6 phonemes},$$

$$\frac{x!}{(x-3!)} > 207, \text{x = 6.9 or ~7 phonemes},$$

$$\frac{x!}{(x-3!)} > 35, \text{x = 4.3 or ~5 phonemes},$$

This gives a ballpark region for the minimum number, given the above constraints. Given that words longer and shorter than 3 phonemes can and do exist in languages, this is not an absolute limit. So you may have only 5 phonemes, which gives 60 3-phoneme words, and then 40 words with 1,2,4 or more phonemes. Or more minimally with 4 phonemes, 24 3-phoneme words, and at least 11 with 1,2,4 or more phonemes.

$$\frac{4!}{(4 - 3)!} + \frac{4!}{(4 - 2)!} + \frac{4!}{(4 - 1)!} = 24 + 12 + 4 = 40 \text{words}$$

As you can see, including the shorter words means we can round down instead of up with 4 phonemes to get the smallest list (35). Doing the same with 5 phonemes puts us 15 words short of 100, but including 4-phoneme words puts us 3 words short of the longer list. If we're assuming far fewer combinations are valid words, even longer phoneme combinations are needed.

### Conclusion

Any number of phonemes works, but depending on if you pick 1,2, 3 of more changes how you construct the language itself.

• If you combine complex inflections and stresses the info density increases vastly even without the potential addition of gestures/facial expressions. – A Rogue Ant. Mar 16 at 12:56
• @Rottweileronmarket-day. agreed, but then I'd say you've just got more phonemes on your hands (ears?). At least that's how I'd try and calculate it. e.g. A-stressed-F-S and A-F-S vary by one phoneme, because you're distinguishing between stressed and unstressed now. – Pureferret Mar 16 at 13:03
• Fair enough, that makes sense. – A Rogue Ant. Mar 16 at 13:10
• You're not thinking very "alien". You're assuming the mode of communication is human-like speech, when even humans exhibit at least two other modes (sign and whistling). – John O Mar 16 at 13:24
• You might want to mention somewhere in there that MORSE CODE is an example of a unary language. Morse code is simply signal, or no signal. Sound, or no sound. Yet it suffices to transmit complex messages. For that matter, so is this Internet that is conveying everything we discuss now. TCP/IP is asynchronous transmission of: signal or silence. Yet one can squeeze a full streaming movie through it. – PcMan Mar 16 at 13:39

Tones and inflection

Theoretically you need one.

A letter can have multiple meanings by just Inflection. It exists in many languages. A different inflection of the same phoneme can distinguish a word from another. With a language that is limited in phonemes it's likely they'll express the differences in other ways, of which inflection is one of the first that comes to mind. Yet you can even potentially just have them wistle.

You might think it's impossible or too difficult, but an alien race grown up with it can distinguish the differences of tones at a much deeper level. It is the difference between a tone deaf person or a highly skilled musician, only instead of learning they have thousands if not millions/billions of generations of evolution. They could distinguish such minute details in the tones that would seem impossible without electronics. This can mean that a single phoneme can in theory convey more information than all phonemes in the human languages.

Even if you require a human like language you can put inflection or tones into it. Depending if you want a human like language or completely alien the answer is two or one respectively.

• How is a phoneme with different tones and inflection different to just multiple phonemes? – Pureferret Mar 16 at 13:20
• @Trioxidane Phoneme: "any of the perceptually distinct units of sound in a specified language". If you can hear the difference, then by definition they are not the same phoneme. Thus the letter E with a g flat or e flat are two different phonemes. So is an E sustained for 500ms different from and E sustained for 520ms, if the speaker and listener can discern the difference between those two lengths. If they cannot perceive a difference, then they are the same phoneme. – PcMan Mar 16 at 13:29
• @PcMan: Nope, even if you can hear the difference it can still be the same phoneme. Only if you can hear the difference and you can find two words which are distinguished only by that difference, then you can say that you have two distinct phonemes. Perhaps in English a good example would be the difference between the clear [l] (as in lean) and the dark [ɫ] (as in all); they are clearly distinct, but there is no minimal pair of words where the distinction makes a difference (because the two allophones cannot appear in the same phonetic context). – AlexP Mar 16 at 13:50
• @AlexP that only makes this one thing different but is still relevant to my answer. As you allude to it's valid if you can distinct different meanings for it, which us the idea in my answer. So each is a different phoneme. – Trioxidane Mar 16 at 14:04
• @Trioxidane: I was replying yo PcMan... (But then phonemes are abstract entities anyway, and you can count, for example, in Greek, η, , and as three phonemes, or as one phoneme + one of three tones.) – AlexP Mar 16 at 14:10