Beyond 'moo' and 'beep', Jev hears less sound-meaning link than people

Asked how much a word sounds like what it means, does Jev hear the same links people do?

result2,488 questions

Jev hears far less connection between how words sound and what they mean than people do: it puts 46% of words in the bottom two levels ("nothing in how it sounds connects to what it means"), where people put 7%, and averages 2.9 against people's 4.0 on a 1-7 scale. It agrees on the obvious ones (whoosh, moo, beep, oink) and ranks words moderately like people overall (0.69), but hears little in words like "blah" and "flee" that people rate as strongly iconic.

How to read this: Each dot is a word. Across: how much people say it sounds like what it means; up: how much Jev does. Dots below the diagonal are words where Jev hears less of a link than people.

2,488 words, about 10 raters each; 90% interval on the correlation [0.673, 0.709]; mean gap 90% interval [-1.07, -1.01].

In short

Jev hears a link between sound and meaning in far fewer words than people do, putting 46% of 2,488 words at the bottom two levels against 7%.
It agrees on sound effects like "whoosh" (6.9 vs 6.7) but misses subtler ones, giving "blah" 2.1 where people give 6.0.
The ranking still agrees moderately (0.69): Jev knows which words are more iconic than others, it just rates nearly all of them lower.

What the data shows

Shrek running meme: People: "flee" sounds like running away. Jev: it's just a word. — People: "flee" sounds like running away. Jev: it's just a word.

The obvious ones match: "whoosh" (6.9 for Jev, 6.7 for people), "moo", "beep", "oink", "hissing", "meow". Jev rates sound-effect words at the very top.
On average, Jev hears less: an average of 2.9 against people's 4.0, and 46% of words at the bottom two levels, where people put 7%.
Among the biggest gaps are words people hear as subtly iconic: "blah" (2.1 vs 6.0), "flee" (3.1 vs 6.0), "foolish", "jinx", "magnetic". People hear something in these; Jev mostly doesn't.
A few words Jev hears more: "alas", "crow", "jay" (for the two birds, Jev may be thinking of their call).
The ranking still agrees moderately: 0.69.

What it means, and what it doesn't

Jev knows which words are sound effects, but the softer layer of iconicity, the sense that "blah" sounds dull and "flee" sounds fast, is mostly missing. That's plausible for a model that has read about sounds but never heard one.

Some of the gap is the project's strict answer wording, and some of the biggest single misses look like noise on the human side, so the average gap is more trustworthy than any one word.

Caveats

The project's level wording. People rated on a 1-to-7 scale labeled only from "not iconic at all" to "very iconic", after instructions. Jev chose among seven described levels written for this project, whose bottom level is strict ("any other sound would do as well"). That wording may push Jev lower than people's plain scale pushes them.
Few raters per word, hand-picked examples. Each word has about 10 to 17 ratings, so single words are noisy. Some of the biggest "misses" look like rater noise: people rated "reader" 5.8 of 7 for sounding like its meaning. "Blah" and "flee" in the headline were picked by hand from the largest gaps among words the raters agreed on.
Reading, not hearing. Jev never hears a word; it answers from what text says about words. Iconicity is partly about sound itself, which text only describes.
The three best examples were removed. "Buzz", "hiss" and "boom" aren't asked, because the top level uses them as its examples.

Jev on this experiment

Would a person find it interesting to read?: Yes77%
Does it describe you?: Yes61%
Would you have predicted it?: Yes65%
How fair is the comparison?: The comparison is shaky
How much should a reader rely on it?: Moderately
Which caveat matters most?: Reading, not hearing93%

Why ask this

Some words sound like what they mean: "buzz", "moo", "whoosh". That's iconicity, and it's not limited to animal noises: in rating studies, people hear a hint of meaning in many ordinary words, such as "blah" or "flee".

A model that only reads text has never heard a word. Whether it still senses these links, or only knows the obvious sound-effect words, says something about what text carries.

How this was done

The people and the data

Winter and colleagues (2023) asked US English speakers to rate thousands of English words for iconicity on a 1-to-7 scale, from "not iconic at all" to "very iconic", and published every individual rating. This experiment uses 2,488 common words, spread across the range of ratings, each rated by about 10 to 17 people. The ratings are posted with the study without an explicit data license, so they're used here for private research only.

What Jev was asked

One question per word, with seven described answers written for this project:

How much does the word "whoosh" sound like what it means?
Any other sound would do as well: nothing in how it sounds connects to what it means · You can find a link between its sound and its meaning only by straining for one · One sound in it hints at its meaning, but you would notice only if someone pointed it out · Its sound suits its meaning, the way a short word suits something small, without imitating it · Its sound clearly fits its meaning, like a rough-sounding word for something rough · Part of it imitates what it means, such as a noise, a motion or a texture · Saying it imitates what it means, like "buzz", "hiss" or "boom"

Each was also asked with the levels reversed, and the two are averaged.

How it was measured

Jev's rating of each word next to people's average: the overall ranking (a rank correlation: 1 means the same order), the average level on each side, how many words each puts at the bottom, and the words with the biggest gaps.

Where these questions live

2,488 questions across 1 topic of the map. Each opens on the map with every question in it.

Word

Word Iconicity2,488

Every question

All 2,488 questions behind this result, the telling ones first: the examples the analysis points to, then the ones where Jev misses, biggest gap first.

Jev’s own answer

Showing 0 of 2,488