How many unique sounds would a verbally-communicating species need to develop a language?

Question

A species with limited capacity for making different vowel/consonant sounds, pitches, vocalizations, and so on, has a limited capacity for the creation of different words in a compositional language. Even if making use of repetition or sound length in word syllables to account for that limited sound availability, there's only so far that that can go (especially without getting too long or tedious to the point of being less and less effective). If using compositional real-world languages as a model for the verbal language of intelligent species, how many unique sounds (phonemes) would be a rough "minimum" requirement for the effective evolution/development of a full, compositional language?

(This may potentially work better in some sort of linguistic stack exchange--if anyone knows a good place do let me know!)

Answer 1

I am not sure if it is the lowest number of sounds documented in a language, but Silbo gomero has a limited set of sounds.

According to different studies, Silbo Gomero has between 2 and 4 vowels and between 4 and 10 consonants. [...] It is a whistled form of a dialect of Canarian Spanish. [...] Silbo has many fewer phonemes than Spanish. This means that communication can be ambiguous at times. Context and word choice are important for effective communication.

Morse code used only two sounds to convey messages, and trained user could understand a message just by listening to the alternating sound of dot and lines being transmitted.

Answer 2

Everything on the internet is conveyed using bits that have only 2 possible states.

While I am not aware of any group of people that has ever natively spoken a binary language, people do use morse code to communicate.

Per Wikipedia, Ham radio operators can routinely send morse code at up to 60 words per minute.

According to John D Cook, Morse code uses an average of 4.53 symbols per English letter.

Furthermore this Stack Exchange post states that morse code operators can generate dots on the order of 50~60ms in duration.

While it's not the most efficient, a useable language could be constructed with as little as two distinguishable verbal states. This could either be two sounds, or one sound and its absence (like on-off-keying in radio). So, we can have a verbal version of morse code.

From the above, it certainly seems possible that a person could get used to generating/listening to something on the order of 20 short tones per second.

For example, one could have a language made entirely with just an "N" sound and gaps with no sound. To say a word the speaker generates a series of Ns and gaps with uniform timing (like 1/20 second per tone/gap).

To make things efficient, use the same principal as Huffman encoding. The most common words are shorter sequences, and more uncommon words are represented with longer sequences. If the 1000 most common words are on average represented with about 10 tones, then someone would be able to speak like 2 words per second.

Note that if the persons speaking can only make a limited set of sounds it makes sense to encode whole words rather than letters. This can increase efficiency a lot. Using English/morse code we might have 5 letters per word and 4.53 dots per letter (so 22.7 dots per word, as compared to 10 dots per word in the example above).

Answer 3

Many years ago, a linguist told me a story about a conference which had been set up to design a universal spoken language.

The attendees debated endlessly, gradually eliminating sounds which were unused by or unavailable to members with different cultural or linguistic backgrounds. They eventually reduced the list of candidate sounds to a mere two:

ee aw

So the answer is undoubtedly two, particularly if you are a donkey.

Answer 4

The only examples of languages we have are human ones.

Whilst there are systems like Morse code that use an extremely small number of distinct segments, they are still fundamentally encoding other languages, that use a larger system.

So, with that in mind we should only consider the size of phoneme inventories of natural human languages.

Central Rotokas has the smallest consonant inventory globally, with just 6 distinct consonant phonemes. It has 5 phonemically distinct vowel qualities which can occur in two distinct lengths (although it's unclear whether long vowels are best viewed as a sequence of two identical vowels, or a separate segment).

There are languages with just two phonemic vowels (and Proto-Chadic is sometimes reconstructed as only having had a single phonemic vowel), but languages with such few vowels typically have very large consonant inventories, so Central Rotokas probably either has the lowest or one of the lowest number of distinct phonemes at 11.

Generally, I'd expect the number of distinct sounds necessary for language to develop to be higher than the lowest number of sounds usable in language once developed (cf the existence of Morse code and the fact that no species with just two distinct sounds has developed language), so I'd expect a species that was able to evolve language to be able to produce more than 11 distinct sounds.

Humans can reliably produce one to two hundred distinct sounds or so (if trained to do so since birth as part of their native language), although of course most languages use far less, and so most individuals are unable to reliably produce and distinguish all those sounds.

So that gives us a range of around 11-200 for the number of distinct sounds needed for language to develop.

Answer 5

This is a subjective question to some extent. Two states is all that is required for a communication system, as has been mentioned by other answers. However, I think one has to consider efficiency in this mix. Here is my thought process:

• English has roughly 1 million words, according to a quick search.
• If we assume a fixed word size, like computer binary, then you'd need 20 bits to represent the number of words in the English language. That is, 2^20 = 1,048,576 unique outcomes.

This solution works, but efficiency is lacking. You could have a variable length word size of 2 states in order to represent various words, but structurally this would be hard to follow without pauses long enough to dictate the boundaries, which would introduce another point of inefficiency if your word sizes were too small (1 bit, 2 bit...etc). So, I think the solution with these assumptions is to find the number of sounds.

If we assume 1 million words, then our best bet for efficiency is to reduce the word size by increasing the number of sounds, but at a level that balances with the word size.

This by trial and error is roughly: 8 sounds with a word size of 7, providing 2,097,152 different states that can be represented.

One could gain more states if we do vary the word size, but the benefit seems minimal from a purely state based view: (8^1) + (8^2) + (8^3) + (8^4) + (8^5) + (8^6) + (8^7) = 8 + 64 + 512 + 4096 + 32768 + 262144 + 2097152 = 2,396,744

You could use this to vary the number of sounds, but I still argue that the need for more boundary pauses to account for a word size as small as 1 hurts the efficiency more than it helps.

There may be a more mathematical way of optimizing this problem, but this is my quick back of the napkin thought process. Overall, it is a question of optimization and physical ability to produce and hear the different sounds. Emotional conveyance can still be had with other sounds, or structural emotional words, or through loudness or other signals. There are many ways to convey such information, so you definitely have to think outside the box.

Note, my approach is from that of an engineer. I'm sure a linguist would be more appropriate, but hopefully my approach gives you some thought.

Answer 6

In theory, every message can be encoded as a string of 0's and 1's. So you could just map them to two vowel-only syllables like /o/ and /i/.

However, as another answer has pointed out, a sapient species probably wouldn't want to speak in binary for the same reason that we don't count in binary: It makes everything long and unwieldly. Like the current year having 11 digits (11111100111). Having more phonemes allows the same information to be encoded in fewer syllables. So, let's try to make that trade off.

Suppose a language has a “core” vocabulary on the order of 10 000 words, that we'd like to express in three or fewer syllables. To make this work, we'd need a minimum of 22 distinct syllables.

If we use a rigid consonant + vowel (CV) syllable structure, then we can make this work with 10 distinct phonemes. For example, 5 consonants and 5 vowels, giving 25 distinct syllables.

If we allow the slightly more complex syllable structure CVC, then 6 phonemes is the minimum. For example, you can have the three consonants /p/, /t/, /k/ and the three vowels /a/, /i/, /u/.

FWIW, the most phonetically minimalist language IRL is Rotokas or Pirahã, with 11 phonemes.

Answer 7

The length of sound in morse code or in a similar fashion sound / no sound pair is a terrible idea for organics. You may understand morse code by listening but it is hard to generate it without being ambiguous. An older person could talk slowly, or you may want to convey emotion.

I would argue you will need two vowels at least but to make it far better add in two consonants. For effective communication speed pair each vowel with a consonant and you will have 4 different states. You could throw a few more consonants in if you wish but I think with 4 sounds you could create not so boring words that could be spoken fast enough. You could also make it more alien to have a specific sound to end words allowing even faster communication by eliminating pause between words. alelakielelekeki.

Also I will oppose other answers that instead of encoding often used words to be shorter, you should encode critical words shorter. eli for warning, aki for stop, etc..

Answer 8

How do you define a "unique sound"? If you really just care about phoneme count, and not how those phonemes are composed, well... other answers have covered that. Plenty of human languages get by with remarkably few phonemes.

But, human language is also pretty exceptional among animal communication systems in that we can produce a wide range of different types of sounds, with different qualities, to make our phonemes. Can you get enough phonemes to be useful with the more restricted vocal abilities that you describe?

Yes, you can! And how do we know? Because whistle registers exist! Real-world humans can effectively communicate at normal speeds in a medium which has no distinctions in sound quality at all, with everything encoded in conventionalized (that is, phonemicized) patterns of volume and limited pitch modulation. So, if your species can produce one single kind of vocalization, as long as they can vary the pitch a bit and do it louder or quieter, and hold it different amounts of time, that is sufficient to put together enough distinct phonemic combinations to encode a complete human-like language.

Answer 9

Don't worry about it.

In a book I have out, there is a galactic community of several hundred alien species. None of them are able to correctly pronounce the phonology of the natural languages of any other race; there's the vocal range of humanity, another race that sounds like a goose honking and coughing, another that sound like a madman playing the flute, and so on. While machine translation is possible, reliably translating between any two of thousands of languages is an enormous burden, even with automation.

So at some point in the past, some linguists working together developed a conlang based on a few dozen syllables. Each species has its own way of pronouncing each syllable, and everyone who deals with other races wears an earpiece (or whatever serves as an earpiece) which transvocalizes the speaker's pronunciation of each spoken syllable into whatever the wearer uses.

With some species that have a limited sound vocabulary, they naturally take a bit longer getting each syllable out, but effective communication remains possible.

Answer 10

Consider a click-language.

Let's say your language is made up of language elements. Words, terms, particles; let's call them "tokens", like the AI researchers do.

And let's say we want to be able to represent roughly as many as written tokens as Japanese: so, a few thousand. We can, as the Japanese do, form new words either phonetically, by using the tokens associated with sounds, or by forming compounds of more complex tokens.

If we have only two possible types of click: click, and not-click, then for a thousand, we need ten clicks in a row.

But we don't need to use all ten. We could omit all click-sequences with more then two non-clicks, and have a space of three or more non-clicks mean "end of token" This loses us a lot of tokens, but is kind of needed so that people coming into the conversation partway, or losing track of the clicks, can re-synch. And that then means we can length-compress, so the most common tokens are shorter, and rarer tokens are longer. This also means we're no longer tied to ten clicks as a max length: we can have them any length we want.

But if we add a few more states, such as four different volumes or pitches of click, or clicks with different pincers, then you only need five clicks per token. With ten click-types, you only need three per token. And you don't then have to use non-clicks, which means you only need to use one clickless beat for a "space", and again you have re-synchonization if you lose track, and can length-compress.

We can also use some of the click sequences as context markers, switching between "token-sets" (like code pages, or alphabets!). That means that like Japanese writing, we can change to a numeric token-set rather than a speech token-set. Or shift the politeness register, or switch to jargon, or slang, or... whatever token-set.

And once you have these context markers, suddenly the language balloons, and word lengths get shorter, as with these multiple token sets, you can have more tokens with lower-numbers of clicks. But it's a balance between that saving, and the cost of the initial context marker. If it saves you one click per character, and costs you three clicks to say, then it's only a saving if you typically say more than three tokens from each token-set after a context marker.

So in theory all you need is click and not-click, but the language rapidly gets a LOT richer once you get more than a single sound.

For Morse code, receiving at 140 WPM (11.7 beeps/sec) is I believe the world record, which is about the 150 WPM of normal spoken speech. So a species evolved to use clicks should be able to communicate at a rate we'd consider normal, especially using some of the tricks above.

Answer 11

Zero.

Well. Depending on your definition of "language".

They don't need to use unique sounds, they could just use the sounds that are already around them.

The lyrebird is a fascinating animal, I highly recommend looking up some videos of them. Attenborough has a good one. It seems to be able to "record" sounds around it and then reproduce them almost perfectly at will, like an MP3 player.

In mating rituals, it reproduces as many sounds as possible to attract a female. They've been observed reproducing camera shutter sounds, toy ray-gun sounds, human speech, and various construction equipment, including hammers and chainsaws.

Suppose a society of humanoids with this remarkable ability of perfect audio recall and reproduction. They might not need to develop any unique sounds, and could instead simply communicate in terms of reference to existing things.

Big angry bear? Make a bear noise to warn people. Fresh vegetables to forage? Crunchy biting noises. I love you? A beating heart.

Think Darmok and Jalad, but with the world's biggest soundboard instead of folklore.

Answer 12

Theoretically, two, but an actual language using just 7 exists.

As numerous other answers correctly point out, you can theoretically create communications capable of expressing anything that any other language can using just two states and one of those sounds can be a meaningful absences of sound (which depending on how you define things, arguably means you need only one "sound").

However, as other answers point out, that is probably not overly efficient for organics.

A real language has been developed which uses only 7 distinct "items". It was designed so that those items can be encoded in multiple different ways including into 7 notes that can be made by instruments or 7 sounds easily made by humans. It is called solresol.

Notably, Solresol is a constructed language rather than a natural language (and has even less of a following than some other constructed languages such as Esperanto, but it does have a following). But its existence proves that an efficient, practical and rich language suitable for speech can be effectively made with 7 distinguishable sounds in the real world.