Ask anyone whether they can hear the temperature of water being poured and the answer is a flat no. The physical attributes of pouring water do change with temperature, though, and those changes reach the ear. So researchers tested listeners on recordings of water being poured, and set a pre-trained the same job. Who was right about what people can do?
People classified it correctly, while insisting they could not. Listeners sorted recordings of pouring water by temperature, which makes this an skill: a lifetime of kettles, taps and glasses taught it without anyone noticing. A deep neural network separated the same physical characteristics, which confirms the difference really is in the sound. The authors call it a , one sense carrying what belongs to another.
Ask people whether they can hear how hot water is and they say no without hesitating. The physical attributes of pouring water do change with temperature, though, and those changes reach the ear. The open question is whether anybody makes use of them.
Listeners in this study sorted recordings of pouring water by temperature, while continuing to claim they could not. That gap between what people do and what they believe they can do is the signature of an implicitly acquired ability, built up by across a lifetime of kettles, taps and glasses.
A pre-trained deep neural network then got the same recordings, and it too picked out the physical characteristics underlying the effect. This matters for the interpretation. Had the cues not been separable by a machine, the human result would need explaining some other way. Since they were, the information really is sitting in the sound.
The authors describe this as a cross-modal correspondence: a property that belongs to one sense, temperature, carried reliably by another, hearing. Keep the claim modest, all the same. What is established is that the sorting works and that the acoustic cues exist. How accurate any one listener is, and which cue they lean on, is not settled here.
wat. A lifetime of pouring water taught your ears its temperature, and never once told you that it had.
wat wat
This is a clean case of a skill people have and deny having. The denial is the interesting half, because it shows how much perceptual learning runs below anything a person could report.
How do we know?
Wenger and colleagues tested listeners on recordings of water poured at different temperatures, and put the same recordings through a pre-trained deep neural network.
What is still unknown
What is established is that the sorting works and that the acoustic cues exist. How accurate any one listener is, and which cue they lean on, is not settled by this paper.
Learn something?
Where to next
The cat does not come when called. Does it at least know its name?hearing
What gets through while you are ignoring everyone else at a party?hearing
Can you catch somebody else’s fear by smelling it?senses