"Sound–meaning association biases"

[jal]

"Sound–meaning association biases evidenced across thousands of languages"

An article and the actual study.

Your thoughts?


JAL

[Qwynegold]

I just read the article. I'm wondering how it is possible for many languages to retain these specific sounds in specific words. Shouldn't sound changes muddle things up completely?

"You" is unlikely to include sounds involving u, o, p, t, d, q, s, r and l.

This is surprising. Wasn't it a thing that Eurasian languages often have something like t, d or s in the word for you?

[Sumelic]

I just read the article. I'm wondering how it is possible for many languages to retain these specific sounds in specific words. Shouldn't sound changes muddle things up completely?

Well, words don't always retain the same meaning after a sound change. Semantic drift affects them; it seems plausible that one contributing factor is generalized sound-meaning analogies. It doesn't seem like it would have a very big effect in most cases, though. It's not the same thing, but something similar seems to be the re-creation of onomatopoeia (like "ba" for sheep in Greek) or momma/papa words. Although I think I've read some accounts of these kinds of words that postulate that they are actually preserved from sound changes, rather than re-created.

[Vijay]

Why does this article seem to have become so popular all of a sudden? I'm positive it's not new, and I doubt very much that any actual linguists contributed anything to the research beyond maybe data samples, yet I've seen people referring to it three times by now, twice on this forum.

[zompist]

Does anyone have access to the actual article? I'm curious how much of an effect they're finding.

If the effect is very large it would be a bit surprising that no one's noticed. Plus aren't there lots of obvious exceptions? (E.g. for 'nose' there's Mandarin bízi, Hungarian orr, Hebrew af, Swahili pua.) And if the effect is very small, it's hard to figure out what the causal mechanism would be. A weak constraint seems harder to explain than a strong one.

Also, how well did they test against chance? Some of these things are pretty broad— e.g. 'leaf' has b, p, or l. As a rough estimate, let's say an average language has 20 consonants and an average root has 2 consonants; then in a completely random language you'd get a word following the rule 30% of the time.

At the same time, I could see there being something deictic about, say, noses and tongues. A nasal sound for 'nose' is like a phonetic pointer. I wonder if word for 'lip' tend to have labials. An L for 'tongue' is also kind of satisfying— it's a sound that kind of draws attention to the tongue.

(Can't think of a reason why 'sand' would have S, or 'leaf' L. Again, it would be interesting to know how strong these correlations were.)

Finally, one more worry: they didn't just look directly at their 3600 languages, did they? If they were going for languages rather than families, they could severely distort the effects. E.g., there's a thousand Austronesian languages, and nearly that number of Niger-Congo languages. That's a lot of opportunity to create a pseudo-effect based on cognates.

[Richard W]

I just read the article. I'm wondering how it is possible for many languages to retain these specific sounds in specific words. Shouldn't sound changes muddle things up completely?

It's a statistical trend. Presumably words that don't conform are more likely to be replaced as the bearer of a particular meaning.

"You" is unlikely to include sounds involving u, o, p, t, d, q, s, r and l.

This is surprising. Wasn't it a thing that Eurasian languages often have something like t, d or s in the word for you?

Some words are corrected less readily than others.

[Rory]

Does anyone have access to the actual article? I'm curious how much of an effect they're finding.

You can access the article here.

I haven't read the article fully, just skimming, but I think I can help answer some of your questions.

If the effect is very large it would be a bit surprising that no one's noticed. Plus aren't there lots of obvious exceptions? (E.g. for 'nose' there's Mandarin bízi, Hungarian orr, Hebrew af, Swahili pua.) And if the effect is very small, it's hard to figure out what the causal mechanism would be. A weak constraint seems harder to explain than a strong one.

The supplemental information is available here. Tables S2 and S3 are the most relevant, which list RR, which is the ratio between the frequency of the phoneme in words relating to the concept and the frequency of the phoneme in other words. So, for "ash", the symbol "u" is listed, and RR=1.91, meaning that they found that "u" is 1.91 times more likely to appear in "ash" than to appear in some other word.

The highest RR value reported is 5.12 for small:[tS]. Next highest is tongue:[l], RR=2.77, and then sand:[s], RR=2.58.

Then there's the negative associations, with RR<0. The lowest is (1st person pronoun):[p], RR=0.18, and joint second is (1st person pronoun):[l], RR=0.19, (2nd person pronoun):[l], RR=0.19.

Interesting findings, I'm not quite sure what to make of it.

Also, how well did they test against chance? Some of these things are pretty broad— e.g. 'leaf' has b, p, or l. As a rough estimate, let's say an average language has 20 consonants and an average root has 2 consonants; then in a completely random language you'd get a word following the rule 30% of the time.

Monte Carlo simulations. I don't quite grok the method but it appears to take into account chance comparisons, as they're comparing the results with that of a randomly-generated sound-meaning correspondence.

At the same time, I could see there being something deictic about, say, noses and tongues. A nasal sound for 'nose' is like a phonetic pointer. I wonder if word for 'lip' tend to have labials. An L for 'tongue' is also kind of satisfying— it's a sound that kind of draws attention to the tongue.

(Can't think of a reason why 'sand' would have S, or 'leaf' L. Again, it would be interesting to know how strong these correlations were.)

Yeah. It's interesting that they don't offer any explanations for these (dis)associations, they just simply note that they exist. It's easy to come up with just-so stories for "lip" and "breast" and "nose", but the others are harder. Still, no-one complains when non-arbitrary lexemes are reported in signed languages, so why not spoken languages too? The causal mechanisms are the tough part, though.

Finally, one more worry: they didn't just look directly at their 3600 languages, did they? If they were going for languages rather than families, they could severely distort the effects. E.g., there's a thousand Austronesian languages, and nearly that number of Niger-Congo languages. That's a lot of opportunity to create a pseudo-effect based on cognates.

They classified their word lists into dialects within languages within lineages, and computed their statistics on a per-lineage basis.

Why does this article seem to have become so popular all of a sudden? I'm positive it's not new, and I doubt very much that any actual linguists contributed anything to the research beyond maybe data samples, yet I've seen people referring to it three times by now, twice on this forum.

Søren Wichmann and Harald Hammarström are among the authors. They are definitely "actual linguists", having active research agendas in language documentation. The other authors are Peter Stadler, a bioinformatician; Morten Christiansen, a psychologist of language; and Damian Blasi, a graduate student in evolutionary anthropology who focuses on linguistic issues. Also, the article was edited by Anne Cutler, perhaps the world's foremost psycholinguist.

I'm not surprised you've seen this article mentioned on this forum - it is a widely-publicized paper, and this is a forum of language enthusiasts.

[Vijay]

Søren Wichmann and Harald Hammarström are among the authors. They are definitely "actual linguists"

I don't doubt that they are. What I want to know is how much of a role they played in writing this paper. Did they even write any part of it, or are they simply listed as authors because they provided data or something?

I'm not surprised you've seen this article mentioned on this forum - it is a widely-publicized paper

Yes, but why? Why is this being so widely publicized now all of a sudden?

[Rory]

Søren Wichmann and Harald Hammarström are among the authors. They are definitely "actual linguists"

I don't doubt that they are. What I want to know is how much of a role they played in writing this paper. Did they even write any part of it, or are they simply listed as authors because they provided data or something?

I don't know how much of a role they played. However, I have reasons to believe that their presence on the paper is not just a matter of courtesy.

1. Adding people as authors just because they provided data, but provided no other input, is extremely rare in linguistics.
2. Even if such a thing happened, any academic who did not agree with the conclusions proposed by such a paper wouldn't allow their name to go on the paper. Having your name on a paper is an endorsement of its contents.
3. The acknowledgement section of the paper lists specific grants that Wichmann and Hammarström have obtained that supported their work on this project. If they just contributed already-constructed datasets, then they wouldn't have needed "support".
4. I share mutual friends with Wichmann on facebook, and I've seen him discussing the article with others and defending the main thesis.

I'm not surprised you've seen this article mentioned on this forum - it is a widely-publicized paper

Yes, but why? Why is this being so widely publicized now all of a sudden?

Because it's literally just been published?

[Richard W]

Does anyone have access to the actual article? I'm curious how much of an effect they're finding.

If the effect is very large it would be a bit surprising that no one's noticed. Plus aren't there lots of obvious exceptions? (E.g. for 'nose' there's Mandarin bízi, Hungarian orr, Hebrew af, Swahili pua.) And if the effect is very small, it's hard to figure out what the causal mechanism would be. A weak constraint seems harder to explain than a strong one.

Are you aware that the middle radical of Hebrew af is /n/? (OK, the derivative anaf 'to be angry' is weak for a synchronic claim, as opposed to a diachronic statement of fact.)

People have long noticed seemingly non-chance, non-genetic correlations, which are a heretical fact. The trick is to turn them into a testable hypothesis.

I'm now wondering what this result does to the Baxter-Manaster Ramer (punctuation?) relationship test. They'd already removed 'nose' from the list of meanings whose words were compared because of the likelihood of the corresponding words starting with /n/.

[Salmoneus]

Still, no-one complains when non-arbitrary lexemes are reported in signed languages, so why not spoken languages too? The causal mechanisms are the tough part, though.

The problem is that these alleged relationships ARE arbitrary. There's no non-arbitrary connection between a leaf and a lateral approximant, or between ash and a back rounded vowel. Indeed, some of these run contrary to the slight non-arbitrary connections we might suggest: the hissing /s/ for a barking dog, for instance, or /k/, the only common stop that ISN'T associated with closing the mouth, for biting.

The links they suggest (other than a few) are clearly arbitrary, which makes it very unlikely that they really are being selected for in some way, unless we want to suppose some sort of very opinionated god driving the the process. Random evolutionary processes should not be able to produce arbitrary non-coincidental correlations - it doesn't just challenge everything known about linguistics, it challenges everything known about science.

It's particularly peculiar when you have things like /z/ being associated with stars. The overwhelming majority of languages do not, of course, have a /z/ phoneme to begin with (only 14% of the languages in UPSID have it, for instance). So... there's something implanted in human brains that says "look, you probably shouldn't have a phoneme /z/, but once you do start phonemically distinguishing that sound you need to rush out and get yourself a word for 'star' that incorporates it!"

Those poor benighted souls who have a language without /z/, and who are forced to live their lives without the sort of word for 'star' that inwardly, unknown to them, they must desparately crave, and toward which the forces of evolution are mysteriously moving them, generation at a time! Oh, the unfulfillable longings and confusions that they must experience, knowing that something isn't quite right with their word for 'star' but not having the phonemes to improve it!

Now, what would be really interesting is if they could show that the opposite paradox also applied. That is, what about meanings that aren't universal, and particularly those that wouldn't have been applicable for humanoids in their evolutionary state (and location) of nature. Would the same process find associations with the words for "wallaby", "tundra", and "cart/chariot/car", I wonder? I strongly suspect that they would.... glorious phonemic predestinations implanted in our brains just waiting for the moment when science and exploration would yield them the opportunity for birth!

[Zaarin]

It's particularly peculiar when you have things like /z/ being associated with stars. The overwhelming majority of languages do not, of course, have a /z/ phoneme to begin with (only 14% of the languages in UPSID have it, for instance). So... there's something implanted in human brains that says "look, you probably shouldn't have a phoneme /z/, but once you do start phonemically distinguishing that sound you need to rush out and get yourself a word for 'star' that incorporates it!"

Those poor benighted souls who have a language without /z/, and who are forced to live their lives without the sort of word for 'star' that inwardly, unknown to them, they must desparately crave, and toward which the forces of evolution are mysteriously moving them, generation at a time! Oh, the unfulfillable longings and confusions that they must experience, knowing that something isn't quite right with their word for 'star' but not having the phonemes to improve it!

Tsk-tsk, the most natural word for star is obviously el.

[Sumelic]

Still, no-one complains when non-arbitrary lexemes are reported in signed languages, so why not spoken languages too? The causal mechanisms are the tough part, though.

The problem is that these alleged relationships ARE arbitrary. There's no non-arbitrary connection between a leaf and a lateral approximant, or between ash and a back rounded vowel. Indeed, some of these run contrary to the slight non-arbitrary connections we might suggest: the hissing /s/ for a barking dog, for instance, or /k/, the only common stop that ISN'T associated with closing the mouth, for biting.

The links they suggest (other than a few) are clearly arbitrary, which makes it very unlikely that they really are being selected for in some way, unless we want to suppose some sort of very opinionated god driving the the process. Random evolutionary processes should not be able to produce arbitrary non-coincidental correlations - it doesn't just challenge everything known about linguistics, it challenges everything known about science.

It's particularly peculiar when you have things like /z/ being associated with stars. The overwhelming majority of languages do not, of course, have a /z/ phoneme to begin with (only 14% of the languages in UPSID have it, for instance). So... there's something implanted in human brains that says "look, you probably shouldn't have a phoneme /z/, but once you do start phonemically distinguishing that sound you need to rush out and get yourself a word for 'star' that incorporates it!"

Those poor benighted souls who have a language without /z/, and who are forced to live their lives without the sort of word for 'star' that inwardly, unknown to them, they must desparately crave, and toward which the forces of evolution are mysteriously moving them, generation at a time! Oh, the unfulfillable longings and confusions that they must experience, knowing that something isn't quite right with their word for 'star' but not having the phonemes to improve it!

I'm also not sure about the validity of most of the conclusions, because of the arbitrariness you mention. I also wonder how the authors tried to avoid false positives considering the large scope of the study. But your last two paragraphs are just dumb mischaracterizations of statements made in the study. Nobody has said that "you need to rush out and get" a word for star that has /z/ or that people experience "longings" for such a word (most any native English speaker could disprove that straw man with simple introspection). The claim is that on average, there are more languages with a word for star containing the sound /z/ than we'd expect if the phonology of the word was entirely random.

[Vijay]

I don't know how much of a role they played. However, I have reasons to believe that their presence on the paper is not just a matter of courtesy.[...]

This isn't a linguistics paper, and it's not especially surprising to see linguists unilaterally defend people who work on natural language processing given how much linguistics departments depend on computational linguists for funding these days.

[Salmoneus]

Still, no-one complains when non-arbitrary lexemes are reported in signed languages, so why not spoken languages too? The causal mechanisms are the tough part, though.

The problem is that these alleged relationships ARE arbitrary. There's no non-arbitrary connection between a leaf and a lateral approximant, or between ash and a back rounded vowel. Indeed, some of these run contrary to the slight non-arbitrary connections we might suggest: the hissing /s/ for a barking dog, for instance, or /k/, the only common stop that ISN'T associated with closing the mouth, for biting.

The links they suggest (other than a few) are clearly arbitrary, which makes it very unlikely that they really are being selected for in some way, unless we want to suppose some sort of very opinionated god driving the the process. Random evolutionary processes should not be able to produce arbitrary non-coincidental correlations - it doesn't just challenge everything known about linguistics, it challenges everything known about science.

It's particularly peculiar when you have things like /z/ being associated with stars. The overwhelming majority of languages do not, of course, have a /z/ phoneme to begin with (only 14% of the languages in UPSID have it, for instance). So... there's something implanted in human brains that says "look, you probably shouldn't have a phoneme /z/, but once you do start phonemically distinguishing that sound you need to rush out and get yourself a word for 'star' that incorporates it!"

Those poor benighted souls who have a language without /z/, and who are forced to live their lives without the sort of word for 'star' that inwardly, unknown to them, they must desparately crave, and toward which the forces of evolution are mysteriously moving them, generation at a time! Oh, the unfulfillable longings and confusions that they must experience, knowing that something isn't quite right with their word for 'star' but not having the phonemes to improve it!

I'm also not sure about the validity of most of the conclusions, because of the arbitrariness you mention. I also wonder how the authors tried to avoid false positives considering the large scope of the study. But your last two paragraphs are just dumb mischaracterizations of statements made in the study. Nobody has said that "you need to rush out and get" a word for star that has /z/ or that people experience "longings" for such a word (most any native English speaker could disprove that straw man with simple introspection). The claim is that on average, there are more languages with a word for star containing the sound /z/ than we'd expect if the phonology of the word was entirely random.

...not sure why you feel the need to turn this into personal flaming. If you're not able to understand tone and context, that's really your problem, not mine.

[Richard W]

Still, no-one complains when non-arbitrary lexemes are reported in signed languages, so why not spoken languages too? The causal mechanisms are the tough part, though.

The problem is that these alleged relationships ARE arbitrary. There's no non-arbitrary connection between a leaf and a lateral approximant, or between ash and a back rounded vowel. Indeed, some of these run contrary to the slight non-arbitrary connections we might suggest: the hissing /s/ for a barking dog, for instance, or /k/, the only common stop that ISN'T associated with closing the mouth, for biting.

I can't help wondering if the /l/ for 'leaf' is associated with /l/ for 'tongue'.

What the study shows is word selection on conserved words - if a word isn't in the Swadesh list it won't be picked up by the survey.

It may be a just-so story, but I wonder if /u/ for 'ash' is related to ugh! for ashes mixed with water. The Swadesh 100 list doesn't have meanings like 'mud' and 'goo'.

At a wild guess, I wonder if /s/ could be for a dangerous animal - perhaps related to the widespread shhh. 'Dog' is the only dangerous animal on the Swadesh list, so we don't get /s/ for lion (Swahili simba, Sanskrit siṃha), let alone snake.

I wonder if /k/ is for fighting - 'bite', 'bone', 'horn', 'knee'. The lack of 'claw' and 'kill' is then striking. This wouldn't explain /k/ for 'ear'.

The links they suggest (other than a few) are clearly arbitrary, which makes it very unlikely that they really are being selected for in some way, unless we want to suppose some sort of very opinionated god driving the the process. Random evolutionary processes should not be able to produce arbitrary non-coincidental correlations - it doesn't just challenge everything known about linguistics, it challenges everything known about science.

Organisms are not modular. Word shape preferences may not have any relevance for natural selection of humans, but just be by-products of other effects. Sometimes, of course, dire drift occurs - humans suffer from scurvy because there wasn't sufficient selective pressure to stop us becoming vulnerable. Selection just doesn't notice some things.

It's particularly peculiar when you have things like /z/ being associated with stars. The overwhelming majority of languages do not, of course, have a /z/ phoneme to begin with (only 14% of the languages in UPSID have it, for instance). So... there's something implanted in human brains that says "look, you probably shouldn't have a phoneme /z/, but once you do start phonemically distinguishing that sound you need to rush out and get yourself a word for 'star' that incorporates it!"

Those poor benighted souls who have a language without /z/, and who are forced to live their lives without the sort of word for 'star' that inwardly, unknown to them, they must desparately crave, and toward which the forces of evolution are mysteriously moving them, generation at a time! Oh, the unfulfillable longings and confusions that they must experience, knowing that something isn't quite right with their word for 'star' but not having the phonemes to improve it!

Also add in the frustration of not being able to form onomatopoeic words like buzz and whizz!
So, maybe there is a latent tendency to use /z/ for apparently small non-terrestial objects like stars and flies (as in Beelzebub)

Now, what would be really interesting is if they could show that the opposite paradox also applied. That is, what about meanings that aren't universal, and particularly those that wouldn't have been applicable for humanoids in their evolutionary state (and location) of nature. Would the same process find associations with the words for "wallaby", "tundra", and "cart/chariot/car", I wonder? I strongly suspect that they would.... glorious phonemic predestinations implanted in our brains just waiting for the moment when science and exploration would yield them the opportunity for birth!

A true word spoken in jest.

[Richard W]

Nobody has said that "you need to rush out and get" a word for star that has /z/ or that people experience "longings" for such a word (most any native English speaker could disprove that straw man with simple introspection).

It doesn't work too well with my 'fly' comparison, but we do have zodiacal light. I don't think /z/ would have helped establish the word zodiac. Stabilisation of zenith is just barely conceivable.

Of course, we did briefly have the tenth planet Xena.

[M Mira]

Random thoughts: maybe /s/ for sand is reminiscent of the sound of sand's friction, while /z/ for star is the buzzing vibration for star's twinkling? Maybe?

[mèþru]

Stars and planets are generally seen as gods by ancient peoples. Do traditional gods buzz? (It sounds like gods are drugs. I should work on better wording.)

[Zaarin]

Random thoughts: maybe /s/ for sand is reminiscent of the sound of sand's friction, while /z/ for star is the buzzing vibration for star's twinkling? Maybe?

/s/ for sand makes sense--the hissing of wind in sand has a very sibilant sound. I'm not really feeling /z/ for stars personally. In fact, most of my conlangs have some combination of /i e l ʃ s t/ for star. (Though I have a somewhat quirky habit of avoiding phonemic voiced fractives...)

[Qwynegold]

I just read the article. I'm wondering how it is possible for many languages to retain these specific sounds in specific words. Shouldn't sound changes muddle things up completely?

Well, words don't always retain the same meaning after a sound change. Semantic drift affects them; it seems plausible that one contributing factor is generalized sound-meaning analogies. It doesn't seem like it would have a very big effect in most cases, though. It's not the same thing, but something similar seems to be the re-creation of onomatopoeia (like "ba" for sheep in Greek) or momma/papa words. Although I think I've read some accounts of these kinds of words that postulate that they are actually preserved from sound changes, rather than re-created.

Yeah, another question is why doesn't semantic drift further help muddle things up? But that's interesting about onomatopoeia. I've been meaning to ask what happens to onomatopoeia when sound changes turn them into something that isn't onomatopoetic anymore.

[Qwynegold]

"You" is unlikely to include sounds involving u, o, p, t, d, q, s, r and l.

This is surprising. Wasn't it a thing that Eurasian languages often have something like t, d or s in the word for you?

Some words are corrected less readily than others.

So these many languages that do have t/d/s are still too few to have an effect on the data?

[Richard W]

So these many languages that do have t/d/s are still too few to have an effect on the data?

As far as I can tell, each lineage (i.e. family) gets the same weighting.

I hope this is the case, because I noticed a whole bunch of Quecha dialects with luziru or similar in the ASJP Database for 'star'. As this word comes from Spanish lucero, I think those words have /s/ rather than /z/. There's a similar error with Basque izar 'star' in the database; this certainly doesn't contain a voiced fricative. On the other hand, all those East European and Turkish 'star' words containing <z> (in the database) probably do contain some sort of /z/.

[Rory]

I don't know how much of a role they played. However, I have reasons to believe that their presence on the paper is not just a matter of courtesy.[...]

This isn't a linguistics paper, and it's not especially surprising to see linguists unilaterally defend people who work on natural language processing given how much linguistics departments depend on computational linguists for funding these days.

I should know when to leave well alone, but...

You have no idea what you are talking about.

1. Many linguists actually are very quick to attack such work as linguistically ill-informed.
2. Many linguists dislike these kinds of approaches to the study of language, one reason being is that they are often done by non-linguists and can suggest that linguistics departments are "out of touch" or using outdated methods. Why would they want to champion something which could lead to their department being closed?
3. Computational linguistics and natural language processing are different things.
4. This paper is not a natural language processing paper.
5. Most higher education funding is awarded to faculty members, not entire departments.
6. None of the authors of this paper are at institutions or departments which are under economic stress. In fact, the Max Planck society is particularly well-funded (relative to other (psycho)linguistics departments, that is).

And while I hate to "pull rank" or argue from authority, I say this as someone who has a PhD in linguistics and who has published more than ten linguistics papers, some in computational linguistics.

But please, tell me more about how Søren Wichmann just got his name on this paper because he contributed a pre-existing dataset and nothing more.

[Curlyjimsam]

Linguistics articles in non-linguistics journals frequently do make errors which any linguist worth their salt would never make, but I'm not sure there are any of those here. They certainly seem to have tried to avoid the number one candidate error, of not balancing the sample by area or language family.

It strikes me as a touch odd that an article making such an important claim wasn't submitted to a proper linguistics journal, but that doesn't mean the article's wrong, and they might have good reasons for submitting it to this journal I don't know about.