One of the amazing things about learning language is that children rarely hear language sounds in ideal acoustic environments. Maybe other people are talking in the background, or the dishwasher is running, or the TV is on. Yet somehow children they learn words just the same. By the time we're adults, we've become experts at filtering out irrelevant sounds and patching together meaning out of the cacophony of everyday life.
As one example, listen to this short clip of me saying the word "dinosaur" three times.
I edited the "s" sound out of the first "dinosaur," so you can clearly hear me saying "dino_aur." The last "dinosaur" is obviously complete. But what about the middle "dinosaur," where I edited in a cough/sneeze right over where the "s" sound is supposed to be? Can you still hear the "s" in the background? Let's make this a poll:
Most adults believe they hear the "s" sound in cases like this, even if the sound has been edited out: the perceptual system adds in a sound where it doesn't exist. (Did I edit the sound out here? I'll keep that a mystery for now.) The effect, known as perceptual restoration, has been observed in children as young as five years old.
But what about younger children -- kids who are just beginning to learn language? Do they also exhibit perceptual restoration? It's a difficult question to study, since children who only know a few words aren't able to tell us what they hear with the precision needed (they can't read, so how can they tell us whether they heard an "s" sound?).
But Rochelle Newman did figure out an innovative way to study the phenomenon in very young children. At 24 months of age, toddlers understand a lot of words, but they still don't typically produce much. Even a three-word sentence is a rarity in children this young. Newman had parents of 24-month-olds sit with their children in their laps in front of two video monitors. One played a movie of a dog, while the other played a cat movie. Both animals basically stayed in place or groomed themselves for the duration of the experiment.
While the videos played, a voice repeated either "Kitty! Kitty! Kitty!" or "Doggie! Doggie! Doggie!" eight times. The video played for 10.7 seconds, then blanked out until the next repetition of the words. The words were either complete and intact, edited to include silence instead of the central consonant sounds, or edited to include a noise instead of the central consonant sounds (tt in kitty and gg in doggie). All toddlers heard all three types of words -- clear, silent, or noise. The time they spent watching each video was compared to the time they spent watching each of the videos when there was no sound. Here are the results:
So when they heard the complete, intact words, the toddlers looked at the matching videos significantly longer than they looked at the non-matching video (e.g. when "doggie" was spoken, they looked at the dog video longer than the cat video, and vice versa). When the central consonants were replaced with silence, they still looked longer at the matching videos, but for significantly less time than with the complete words. When noise was added in, there was actually no significant difference in looking time at the cat or the dog.
This is the opposite of what you'd expect if the toddlers were experiencing perceptual restoration! It suggests that they understand the words with silent gaps better than the words with noise added in. Remember, adults believe they hear the entire word when noise is added in, but not when a silent gap is edited into a recording.
Newman repeated the study with several variations: complete sentences instead of single words; more complex words and older (30-month-old) children; even carefully using dummy words like "dobby" and "kippy" as the basis for the edited-down versions to ensure that subtle hints of the edited letters weren't present in the sounds played for the children. In each case, the results were the same: Toddlers respond as if they do not experience the perceptual restoration effect.
There are many possible reasons why the effect isn't observed. Toddlers may not know the words as well as adults and older kids; the effect might require a very good knowledge of the words. They might simply be more open to new words, and thus treat the words with noise as if they are new words. They might be slower to process the words with noises. Or they might know the words just as well as older people, but not focus as much on this knowledge.
Whatever the explanation, this study suggests that perceptual restoration is acquired sometime between the second and fifth year of life. It may be a later add-on as kids get better with language, not one of the fundamental building blocks of language learning.
So, did I edit out that "s" sound in the middle word of my demo? Why don't you let me know what you think in the discussion? I'll let you know what I actually did after I hear a few of your guesses.
Newman, R.S. (2006). Perceptual restoration in toddlers. Perception & Psychophysics, 68(4), 625-642.
- Log in to post comments
There's a "s" sound at the end of the cough/sneeze that could be the spoken "s" or just the end of the sneeze sound. I can't tell. Interestingly, I hear the first word clearly as "dinothaur". Since you don't claim to have inserted the "th" sound, perhaps that is the better illustrator of perceptual restoration.
Karen, I hear it as "dinothaur", too. And maybe having the sound of a car door slam, instead of a sneeze with its inherent "s"-ness, would make the question of "can you hear an 's'?" more unambiguously answered. I hear an "s", but like Karen can't tell whether it's really there or part of the sneeze.
I can only guess that it's not surprising that we would hear the removed /s/ as /voiceless th/ if the entire sound is edited out as the surrounding sounds still have the co-articulation of the /s/ going on right? But the high frequency markers of the /s/ is gone, so we insert the /voiceless th/ in it's place, which is a noisier sound with a wider bandwidth right?
Well anyway, I think in the second dionsaur the /s/ is there, but then again, I am not a toddler so I'm probably putting it there. How could one tell? :)
I saw the visual version of this at some point in the past. Sorry, no reference. It was an animated GIF of, apparently, randomly moving lines. It was nearly impossible to attribute any order to the image... until you covered up the central section section diagonally with your finger. It was immediately apparent that it was a representation of a spinning wire cube (or something similar). Very dramatic.
I thought there was no 's', but now that I hear it a couple more times, I think there was an 's' there.
There was an e-mail going around a few years ago demonstrating that adults read similarly to this concept: we read the first & last letters to infer what a word is, and can read a sentence even if the middle letters are mixed up and/or incorrect. It makes perfect sense if you're familiar with a language...we can infer a lot with parts of a language we know.
(and yes, I heard "dinothaur" in that recording too. lol)
I've heard of this perceptual restoration before - again, someone edited out a spoken /s/ and replaced it with a sneeze.
However, coughs/sneezes seem to have some of the qualities of the /s/ sound (ch, as in ah-choo, is a unvoiced sibilant, like ess). Does perceptual restoration occur if you edit in a neutral, non-sibilant noise, like a beep? What if you edit in a person humming or making a nonsense sound?
@bg That email, while interestingly correct, was a hoax (there was no university study). It was probably a grad student in linguistics trying to prove his dissertation. The phenomenon does obviously exist, though.
This brings me to another point. In my short time as a language instructor (French) and as linguist (finishing my MA) I've noticed the remarkable degree to which English tolerates errors (in contrast to, say, French). This is only anecdotal, but I have noticed a strong tendency for English speakers to be able to understand even the most poorly constructed and, moreover, poorly articulated, broken English. The same is not true of French though. This could simply be cultural, though, as French has a history of normalization (L'Académie Française and L'Office de la Langue Française) that English does not.
For the record, though, I did hear [deɪnoʷθor] in the contaminated example.
I definitely hear something like "dinothaur" in the first recording as well. I vaguely hear something s-like in the second recording, but because of the "th" phenomenon, I'm going to go ahead and put my money on the guess that you edited out the 's' in the 2nd recording as well. It just doesn't sound convincingly enough like "dinosaur" to me to believe that the 's' sound is there, although it does sound like "dinothaur". My vote is in.
I didn't even hear "saur" on the second dinosaur, much less just the "s". Did you edit the whole thing out?
ONCE UPON A TIME, I was on a "College Bowl" team for a dorm at Michigan State University. The first question, called a pop-up, could be interrupted if you knew the answer, and indeed, the team that answered it got a shot at the bonus question.
The pop-up question began "The wars between Rome and Carth..." I hit my buzzer. I knew all I needed was an answer involving Hannibal, elephants, and the Punic Wars.
Alas. Instead of "Punic Wars," I thought of "Pubic Wars."
I could not for the life of me recall the correct phoneme. I was frantically going thru the alphabet in my head, "Pucic...Pudic..Pufic...Pugic" with increasing inner hysteria, when I ran out of time and the judge's buzzer sounded.
I erupted in laughter, and my teammates started pounding me in righteous indignation.
I realized later that I need not have been punished for my dirty mind. Had I simply said "Pu ic Wars," the judges would have filled in the phoneme on their own. But I never knew there was a name for this phonemonomon.
Thanks for filling in the gap.
You should've said Pubic Wars, and at least got the laugh out of it... :-)
I DID get the laugh!
Of course, I was laughing, not my teammates. Still gives me a giggle, now and again, and "Pubic Wars" would make a great title for, oh, say, a movie about marriage.
The sneeze sound seems to be comprised of t-s-ch, all unvoiced consonent sounds, so sure we fill in the blank with the softest least pronounced of these four, which is 's'.
It is well-known that you can average together many examples
of a sound, even when many are corrupted, and end up with a
pretty good copy, provided that the corruptions do not all
occur at the same point in time. I don't know at exactly what
level of representation in the brain the averaging takes
place in the perceptual system, but I have no doubt that
such averaging occurs. So what's interesting is that the
perceptual system substitutes the average from memory for
the incoming percept. The cough was several dB louder than
the word. I did not hear anything but a cough during that time.
I "heard" the -s- in the 2nd word the first time I listened, and the more often I listened, the less often it was there. Problem is, like #1, I hear the essiness (ha) of the sneeze. There are s-like overtones sitting in the sneeze, so my ear deconstructs that noise back into what should be there. At least, I think that's what's going on.
Then again, the more I listened, the less I heard it. Weird, that.
I'm pretty sure the 's' was edited from the second repetition too.
When you listen to the first version, the '_aur' makes it's own syllable, starting with the 'au' sound rather than the expected 's'. Keep listening until you don't hear the phantom 'th' to help isolate it.
This is audible again in the second version, but is blended out and becomes less obvious when the much longer deliberate 's' sound is preceding it in the third.
I do hear a slight 's' phonetic in the sneeze but it doesn't force me to change my perception of the syllable's starting point. This only happens on the complete and clean version.
I would expect to have the same perceptual change in the second version if it was the full word with a sneeze over the top. It seems likely (to me anyway) that adding an additional 'ch or 'sh' from the sneeze would make it more obvious, not less.
I hope that made sense.
Great stuff. Is it possible that there are individual differences among adults with regard to how perceptual restoration is achieved? Repair seems to draw on knowledge of the sound inventory and the lexicon, and so I wonder whether different hearers use these two knowledge sources the same way, or whether they use their strengths in one..