How do our brains process speech? - Gareth Gaskell

396,447 views ใƒป 2020-07-23

TED-Ed


์•„๋ž˜ ์˜๋ฌธ์ž๋ง‰์„ ๋”๋ธ”ํด๋ฆญํ•˜์‹œ๋ฉด ์˜์ƒ์ด ์žฌ์ƒ๋ฉ๋‹ˆ๋‹ค.

๋ฒˆ์—ญ: Soomin Kim ๊ฒ€ํ† : DK Kim
00:06
The average 20 year old knows between 27,000 and 52,000 different words.
0
6814
7231
ํ‰๊ท ์ ์ธ 20์‚ด ์„ฑ์ธ์€ 27,000 - 52,000 ๋‹จ์–ด ์ •๋„๋ฅผ ์••๋‹ˆ๋‹ค.
00:14
By age 60, that number averages between 35,000 and 56,000.
1
14045
6008
60์„ธ์—๋Š” ํ‰๊ท ์ ์œผ๋กœ 35,000 - 56,000 ๋‹จ์–ด๋ฅผ ์•Œ์ฃ .
00:20
Spoken out loud, most of these words last less than a second.
2
20053
4277
์ž… ๋ฐ–์œผ๋กœ ์†Œ๋ฆฌ๋‚ด์–ด ๋งํ•œ๋‹ค๋ฉด ๋Œ€๋ถ€๋ถ„ ์ฑ„ 1์ดˆ๋„ ๊ฑธ๋ฆฌ์ง€ ์•Š์Šต๋‹ˆ๋‹ค.
00:24
So with every word, the brain has a quick decision to make:
3
24330
4205
๊ทธ๋ž˜์„œ ๋ชจ๋“  ๋‹จ์–ด์— ๋Œ€ํ•˜์—ฌ ๋‡Œ๋Š” ์•„์ฃผ ๋น ๋ฅธ ๊ฒฐ์ •์„ ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
00:28
which of those thousands of options matches the signal?
4
28535
3700
์ˆ˜์ฒœ ๊ฐœ์˜ ์„ ํƒ ์‚ฌํ•ญ ์ค‘ ๊ณผ์—ฐ ์–ด๋–ค ๊ฒƒ์ด ์‹ ํ˜ธ์— ๋ถ€ํ•ฉํ• ๊นŒ์š”?
00:32
About 98% of the time, the brain chooses the correct word.
5
32235
4110
๋‡Œ๋Š” 98%์˜ ๊ฒฝ์šฐ์— ์˜ฌ๋ฐ”๋ฅธ ๋‹จ์–ด๋ฅผ ๊ณ ๋ฆ…๋‹ˆ๋‹ค.
00:36
But how?
6
36345
1130
๊ณผ์—ฐ ์–ด๋–ป๊ฒŒ ํ•˜๋Š” ๊ฑธ๊นŒ์š”?
00:37
Speech comprehension is different from reading comprehension,
7
37475
3640
๋ง์„ ์ดํ•ดํ•˜๊ธฐ๋Š” ์ฝ๊ณ  ์ดํ•ดํ•˜๊ธฐ์™€๋Š” ๋‹ค๋ฅด์ง€๋งŒ
00:41
but itโ€™s similar to sign language comprehensionโ€”
8
41115
3260
์ˆ˜ํ™”๋ฅผ ์ดํ•ดํ•˜๋Š” ๊ฒƒ๊ณผ๋Š” ๋น„์Šทํ•ฉ๋‹ˆ๋‹ค.
00:44
though spoken word recognition has been studied more than sign language.
9
44375
4486
์Œ์„ฑ ์–ธ์–ด๋Š” ์ˆ˜ํ™”๋ณด๋‹ค ๋” ๋งŽ์ด ์—ฐ๊ตฌ๋˜์—ˆ์ง€๋งŒ์š”.
00:48
The key to our ability to understand speech
10
48861
2560
์šฐ๋ฆฌ๊ฐ€ ๋ง์„ ์ดํ•ดํ•˜๋Š” ๋Šฅ๋ ฅ์˜ ๋น„๊ฒฐ์€
00:51
is the brainโ€™s role as a parallel processor,
11
51421
3270
๋ณ‘๋ ฌ ์ฒ˜๋ฆฌ๊ธฐ๋กœ์„œ ํ™œ๋™ํ•˜๋Š” ๋‡Œ๊ฐ€ ์žˆ๊ธฐ ๋•Œ๋ฌธ์ธ๋ฐ์š”,
00:54
meaning that it can do multiple different things at the same time.
12
54691
4000
๋‡Œ๊ฐ€ ์—ฌ๋Ÿฌ ๊ฐ€์ง€ ์ผ์„ ๋™์‹œ์— ์ฒ˜๋ฆฌํ•  ์ˆ˜ ์žˆ๋‹ค๋Š” ๋œป์ž…๋‹ˆ๋‹ค.
00:58
Most theories assume that each word we know
13
58691
2610
๋Œ€๋ถ€๋ถ„์˜ ํ•™์„ค๋“ค์€ ์šฐ๋ฆฌ๊ฐ€ ์•„๋Š” ๋ชจ๋“  ๋‹จ์–ด๊ฐ€
01:01
is represented by a separate processing unit that has just one job:
14
61301
4470
ํ•œ ๊ฐ€์ง€ ์ž‘์—…๋งŒ์„ ์ˆ˜ํ–‰ํ•˜๋Š” ๊ฐœ๋ณ„ ์ฒ˜๋ฆฌ๊ธฐ๋กœ ํ‘œํ˜„๋œ๋‹ค๊ณ  ๊ฐ€์ •ํ•ฉ๋‹ˆ๋‹ค.
01:05
to assess the likelihood of incoming speech matching that particular word.
15
65771
5160
๋“ค์–ด์˜ค๋Š” ๋ง์ด ๊ทธ ๋‹จ์–ด์™€ ์ผ์น˜ํ•˜๋Š”์ง€ ํ™•์ธํ•˜๋Š” ์ž‘์—…์ž…๋‹ˆ๋‹ค.
01:10
In the context of the brain, the processing unit that represents a word
16
70931
4208
๋‡Œ์—์„œ ๋ง์„ ๋‚˜ํƒ€๋‚ด๋Š” ์ฒ˜๋ฆฌ ์žฅ์น˜๋Š”
01:15
is likely a pattern of firing activity across a group of neurons
17
75139
4657
ํ•œ ๋ฌด๋ฆฌ์˜ ์‹ ๊ฒฝ ์„ธํฌ๊ฐ€ ํ™œ์„ฑํ™”๋˜๋Š” ์–‘์ƒ์ด๋ผ๊ณ  ๋ณผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
01:19
in the brainโ€™s cortex.
18
79796
1890
๋‡Œ์˜ ํ”ผ์งˆ ์•ˆ์—์„œ์š”.
01:21
When we hear the beginning of a word,
19
81686
1820
๋‹จ์–ด์˜ ์•ž๋ถ€๋ถ„์„ ๋“ฃ๊ธฐ ์‹œ์ž‘ํ•˜๋ฉด
01:23
several thousand such units may become active,
20
83506
3780
์ˆ˜์ฒœ ๊ฐœ์˜ ์‹ ๊ฒฝ ์žฅ์น˜ ๋ฌด๋ฆฌ๊ฐ€ ํ™œ์„ฑํ™”๋˜๋Š”๋ฐ,
01:27
because with just the beginning of a word,
21
87286
2066
๋‹จ์ง€ ๋‹จ์–ด์˜ ์•ž๋ถ€๋ถ„๋งŒ ๋“ค์€ ์ƒํƒœ์ด๋ฏ€๋กœ
01:29
there are many possible matches.
22
89352
2180
๋งŽ์€ ์„ ํƒ์ง€๊ฐ€ ์žˆ์„ ๊ฒƒ์ž…๋‹ˆ๋‹ค.
01:31
Then, as the word goes on, more and more units register
23
91532
4003
๋‹จ์–ด๊ฐ€ ๋งˆ์ € ๋ฐœ์Œ๋ ์ˆ˜๋ก ์ ์  ๋” ๋งŽ์€ ๋ฌด๋ฆฌ๋“ค์ด
01:35
that some vital piece of information is missing and lose activity.
24
95535
5131
์ค‘์š”ํ•œ ์ •๋ณด๊ฐ€ ์—†์Œ์„ ํ™•์ธํ•˜๊ณ  ๋น„ํ™œ์„ฑํ™”๋ฉ๋‹ˆ๋‹ค.
01:40
Possibly well before the end of the word,
25
100666
2460
์•„๋งˆ ๊ทธ ๋‹จ์–ด๊ฐ€ ๋๋‚˜๊ธฐ ํ•œ์ฐธ ์ „์—
01:43
just one firing pattern remains active, corresponding to one word.
26
103126
4964
ํ•˜๋‚˜์˜ ๋‹จ์–ด์— ๋Œ€์‘ํ•˜๋Š” ๋‹จ ํ•˜๋‚˜์˜ ํ™œ์„ฑํ™”๋œ ๋ฌด๋ฆฌ๋งŒ ๋‚จ๊ฒ ์ฃ .
01:48
This is called the "recognition point."
27
108090
2738
์ด๊ฒƒ์„ '์ธ์ง€ ์‹œ์ '์ด๋ผ๊ณ  ๋ถ€๋ฆ…๋‹ˆ๋‹ค.
01:50
In the process of honing in on one word,
28
110828
2820
ํ•œ ๋‹จ์–ด๋กœ ์ถ”๋ฆฌ๋Š” ๊ณผ์ •์—์„œ
01:53
the active units suppress the activity of others,
29
113648
3070
ํ™œ์„ฑํ™”๋œ ์žฅ์น˜๋“ค์ด ๋‹ค๋ฅธ ์žฅ์น˜๋“ค์˜ ํ™œ๋™์„ ์–ต์••ํ•จ์œผ๋กœ์จ
01:56
saving vital milliseconds.
30
116718
2120
๊ท€์ค‘ํ•œ ์ฒœ๋ถ„์˜ ๋ช‡ ์ดˆ๋ฅผ ์ ˆ์•ฝํ•˜๊ฒŒ ๋˜๋Š” ๊ฒƒ์ด์ฃ .
01:58
Most people can comprehend up to about 8 syllables per second.
31
118838
4797
๋ณดํ†ต ์‚ฌ๋žŒ์€ 1์ดˆ๋‹น ์•ฝ 8๊ฐœ์˜ ์Œ์ ˆ๊นŒ์ง€ ์ดํ•ดํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
02:03
Yet, the goal is not only to recognize the word,
32
123635
3330
ํ•˜์ง€๋งŒ ๋ชฉํ‘œ๋Š” ๊ทธ ๋‹จ์–ด๋ฅผ ์ธ์‹ํ•˜๋Š” ๊ฒƒ๋ฟ๋งŒ์ด ์•„๋‹ˆ๋ผ
02:06
but also to access its stored meaning.
33
126965
3450
๊ทธ ๋‹จ์–ด๊ฐ€ ๋‚ดํฌํ•˜๋Š” ์˜๋ฏธ๋ฅผ ์•Œ์•„์ฐจ๋ฆฌ๋Š” ๊ฒƒ๋„ ์žˆ์Šต๋‹ˆ๋‹ค.
02:10
The brain accesses many possible meanings at the same time,
34
130415
3780
๋‡Œ๋Š” ํ•œ ๋ฒˆ์— ๋งŽ์€ ์˜๋ฏธ๋“ค์„
02:14
before the word has been fully identified.
35
134195
2680
๊ทธ ๋‹จ์–ด๋ฅผ ๋‹ค ์•Œ์•„๋“ฃ๊ธฐ๋„ ์ „์— ๋”ฐ์ ธ๋ด…๋‹ˆ๋‹ค.
02:16
We know this from studies which show that even upon hearing a word fragmentโ€”
36
136875
5143
๋งŽ์€ ์—ฐ๊ตฌ์—์„œ ๋ฒํ˜€์ง„ ๊ฑด
์‹ฌ์ง€์–ด๋Š” ๋‹จ์–ด์˜ ์ผ๋ถ€, ์˜ˆ๋ฅผ ๋“ค์–ด 'cap-'์„ ๋“ฃ๋Š”๋‹ค๋ฉด
02:22
like "cap"โ€”
37
142018
1280
02:23
listeners will start to register multiple possible meanings,
38
143298
3500
์‚ฌ๋žŒ๋“ค์€ ๊ฐ€๋Šฅํ•œ ์—ฌ๋Ÿฌ ๊ฐ€์ง€ ์˜๋ฏธ๋“ค์„ ์ƒ๊ฐํ•˜๊ธฐ ์‹œ์ž‘ํ•˜๋Š”๋ฐ
02:26
like captain or capital, before the full word emerges.
39
146798
5172
captain์ด๋‚˜ capital์„ ๋‹จ์–ด๊ฐ€ ๋๋‚˜๊ธฐ๋„ ์ „์— ์˜ˆ์ธกํ–ˆ์ฃ .
02:31
This suggests that every time we hear a word
40
151970
3150
์ด๊ฒƒ์€ ์–ด๋–ค ๋‹จ์–ด๋ฅผ ๋“ค์„ ๋•Œ๋งˆ๋‹ค
02:35
thereโ€™s a brief explosion of meanings in our minds,
41
155120
3360
์šฐ๋ฆฌ ๋จธ๋ฆฟ์†์—์„  ์ˆœ์‹๊ฐ„์— ์˜๋ฏธ ๋‚˜์—ด์˜ ํ–ฅ์—ฐ์ด ํŽผ์ณ์ง€๊ณ ,
02:38
and by the recognition point the brain has settled on one interpretation.
42
158480
4811
๋งˆ์นจ๋‚ด ์ธ์ง€ ์‹œ์ ์— ๋„๋‹ฌํ•˜๋ฉด ๋‡Œ๊ฐ€ ํ•œ ๊ฐ€์ง€๋ฅผ ๊ณ ๋ฅด๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.
02:43
The recognition process moves more rapidly
43
163291
2930
์ธ์ง€ ๊ณผ์ •์ด ๋” ๋นจ๋ฆฌ ์›€์ง์ด๋Š” ๊ฒƒ์€
02:46
with a sentence that gives us context than in a random string of words.
44
166221
4600
๋งˆ๊ตฌ ๋‚˜์—ด๋œ ๋‹จ์–ด๋“ค์ด ์•„๋‹Œ ๋ฌธ๋งฅ์ด ์žˆ๋Š” ๋ฌธ์žฅ์ผ ๋•Œ์ž…๋‹ˆ๋‹ค.
02:50
Context also helps guide us towards the intended meaning of words
45
170821
4188
๋ฌธ๋งฅ์€ ์šฐ๋ฆฌ๊ฐ€ ๊ทธ ๋‹จ์–ด๊ฐ€ ์˜๋„ํ•œ ์˜๋ฏธ๋ฅผ ์ฐพ๋Š” ๊ฑธ ๋„์™€์ฃผ์ฃ .
02:55
with multiple interpretations, like "bat," or "crane,"
46
175009
4000
'bat'์™€ 'crane'์ฒ˜๋Ÿผ ๋‹ค์–‘ํ•œ ํ•ด์„์ด ๊ฐ€๋Šฅํ•˜๊ฑฐ๋‚˜
02:59
or in cases of homophones like "no" or "know."
47
179009
4000
'no'์™€ 'know' ๊ฐ™์€ ๋™์Œ์ด์˜์–ด์ธ ๊ฒฝ์šฐ์— ๋ง์ด์ฃ .
03:03
For multilingual people, the language they are listening to is another cue,
48
183009
4384
์—ฌ๋Ÿฌ ์–ธ์–ด๋ฅผ ๊ตฌ์‚ฌํ•˜๋Š” ์‚ฌ๋žŒ๋“ค์€ ๋“ฃ๊ณ  ์žˆ๋Š” ์–ธ์–ด๋ฅผ ๋‹จ์„œ๋กœ ์‚ผ์•„
03:07
used to eliminate potential words that donโ€™t match the language context.
49
187393
5313
ํ•ด๋‹น ์–ธ์–ด์˜ ๋ฌธ๋งฅ์— ๋งž์ง€ ์•Š๋Š” ๋‹จ์–ด๋“ค์„ ์ง€์›๋‹ˆ๋‹ค.
03:12
So, what about adding completely new words to this system?
50
192706
4000
๊ทธ๋Ÿฌ๋ฉด, ์™„์ „ํžˆ ์ƒˆ๋กœ์šด ๋‹จ์–ด๋“ค์„ ์ด ๊ณผ์ •์— ์ถ”๊ฐ€ํ•˜๋Š” ๊ฑด ์–ด๋–จ๊นŒ์š”?
03:16
Even as adults, we may come across a new word every few days.
51
196706
4000
์–ด๋ฅธ์ด ๋˜์–ด์„œ๋„ ์šฐ๋ฆฌ๋Š” ๋ฉฐ์น ๋งˆ๋‹ค ์ƒˆ๋กœ์šด ๋‹จ์–ด๋ฅผ ์ ‘ํ•˜๊ณค ํ•ฉ๋‹ˆ๋‹ค.
03:20
But if every word is represented as a fine-tuned pattern of activity
52
200706
4403
๋งŒ์•ฝ ๋ชจ๋“  ๋‹จ์–ด๋“ค์ด ์„ธ๋ฐ€ํ•œ ํ™œ์„ฑํ™” ํŒจํ„ด์œผ๋กœ์„œ
03:25
distributed over many neurons,
53
205109
2330
๋งŽ์€ ์‹ ๊ฒฝ ์„ธํฌ์— ํผ์ ธ ์žˆ๋‹ค๋ฉด
03:27
how do we prevent new words from overwriting old ones?
54
207439
4553
์ƒˆ๋กœ์šด ๋‹จ์–ด๋“ค์ด ๊ธฐ์กด ๋‹จ์–ด๋“ค์„ ๋ฎ์–ด์“ฐ๋Š” ๊ฑธ ์–ด๋–ป๊ฒŒ ๋ฐฉ์ง€ํ• ๊นŒ์š”?
03:31
We think that to avoid this problem,
55
211992
2330
์ด๋Ÿฐ ๋ฌธ์ œ๋ฅผ ๋ฐฉ์ง€ํ•˜๊ธฐ ์œ„ํ•ด
03:34
new words are initially stored in a part of the brain called the hippocampus,
56
214322
4763
์ƒˆ๋กœ์šด ๋‹จ์–ด๋“ค์€ ์ฒ˜์Œ์— ํ•ด๋งˆ๋ผ๊ณ  ๋ถ€๋ฅด๋Š” ๋‡Œ์˜ ํ•œ ๋ถ€๋ถ„์—,
03:39
well away from the main store of words in the cortex,
57
219085
3608
๋Œ€๋‡Œ ํ”ผ์งˆ์˜ ์ฃผ ๋‹จ์—ฌ ์ €์žฅ์†Œ๋กœ๋ถ€ํ„ฐ ๋ฉ€๋ฆฌ ์žˆ๋Š” ๊ณณ์— ์ €์žฅ๋œ๋‹ค๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค.
03:42
so they donโ€™t share neurons with others words.
58
222693
3370
๋‹ค๋ฅธ ๋‹จ์–ด๋“ค๊ณผ ๊ฐ™์€ ์‹ ๊ฒฝ์„ธํฌ๋ฅผ ์‚ฌ์šฉํ•˜์ง€ ์•Š๋„๋ก ๋ง์ด์ฃ .
03:46
Then, over multiple nights of sleep,
59
226063
3010
๊ทธ๋ฆฌ๊ณ  ๋ช‡ ๋ฐค ์ž ์„ ์ž๋ฉฐ
03:49
the new words gradually transfer over and interweave with old ones.
60
229073
5397
์ƒˆ๋กœ์šด ๋‹จ์–ด๋“ค์€ ์„œ์„œํžˆ ์˜ค๋ž˜๋œ ๋‹จ์–ด๋“ค๊ณผ ์„ž์ด๊ธฐ ์‹œ์ž‘ํ•ฉ๋‹ˆ๋‹ค.
03:54
Researchers think this gradual acquisition process
61
234470
3520
์—ฐ๊ตฌ์›๋“ค์€ ์ด ์ ์ง„์ ์ธ ์Šต๋“ ๊ณผ์ •์ด
03:57
helps avoid disrupting existing words.
62
237990
3364
์ด๋ฏธ ์•Œ๊ณ  ์žˆ๋˜ ๋‹จ์–ด๋“ค์„ ๋ฐฉํ•ดํ•˜์ง€ ์•Š๊ฒŒ ๋„์™€์ค€๋‹ค๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค.
04:01
So in the daytime,
63
241354
1420
๊ทธ๋ž˜์„œ ๋‚ฎ์—๋Š”
04:02
unconscious activity generates explosions of meaning as we chat away.
64
242774
4530
๋Œ€ํ™”๋ฅผ ๋‚˜๋ˆ„๋ฉด์„œ ๋ฌด์˜์‹์ ์œผ๋กœ ๋งŽ์€ ๋‹จ์–ด์˜ ์˜๋ฏธ๋ฅผ ์ƒ๊ฐํ•ด๋ƒ…๋‹ˆ๋‹ค.
04:07
At night, we rest, but our brains are busy integrating new knowledge
65
247304
5001
๋ฐค์— ์šฐ๋ฆฌ๋Š” ์‰ฌ์ง€๋งŒ ์šฐ๋ฆฌ ๋‡Œ๋Š” ์ƒˆ๋กœ์šด ์ •๋ณด๋ฅผ
04:12
into the word network.
66
252305
1820
๋‹จ์–ด ๋„คํŠธ์›Œํฌ์— ํ†ตํ•ฉ์‹œํ‚ค๋Š๋ผ ๋ฐ”์ฉ๋‹ˆ๋‹ค.
04:14
When we wake up, this process ensures that weโ€™re ready
67
254125
3530
์ด ๊ณผ์ •์€ ์šฐ๋ฆฌ๊ฐ€ ์•„์นจ์— ์ผ์–ด๋‚  ๋•Œ
04:17
for the ever-changing world of language.
68
257596
3100
๋งค์ผ ๋ณ€ํ™”ํ•˜๋Š” ์–ธ์–ด์˜ ์„ธ๊ณ„์— ๋Œ€์‘ํ•˜๋„๋ก ์ค€๋น„ํ•ด์ค๋‹ˆ๋‹ค.
์ด ์›น์‚ฌ์ดํŠธ ์ •๋ณด

์ด ์‚ฌ์ดํŠธ๋Š” ์˜์–ด ํ•™์Šต์— ์œ ์šฉํ•œ YouTube ๋™์˜์ƒ์„ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค. ์ „ ์„ธ๊ณ„ ์ตœ๊ณ ์˜ ์„ ์ƒ๋‹˜๋“ค์ด ๊ฐ€๋ฅด์น˜๋Š” ์˜์–ด ์ˆ˜์—…์„ ๋ณด๊ฒŒ ๋  ๊ฒƒ์ž…๋‹ˆ๋‹ค. ๊ฐ ๋™์˜์ƒ ํŽ˜์ด์ง€์— ํ‘œ์‹œ๋˜๋Š” ์˜์–ด ์ž๋ง‰์„ ๋”๋ธ” ํด๋ฆญํ•˜๋ฉด ๊ทธ๊ณณ์—์„œ ๋™์˜์ƒ์ด ์žฌ์ƒ๋ฉ๋‹ˆ๋‹ค. ๋น„๋””์˜ค ์žฌ์ƒ์— ๋งž์ถฐ ์ž๋ง‰์ด ์Šคํฌ๋กค๋ฉ๋‹ˆ๋‹ค. ์˜๊ฒฌ์ด๋‚˜ ์š”์ฒญ์ด ์žˆ๋Š” ๊ฒฝ์šฐ ์ด ๋ฌธ์˜ ์–‘์‹์„ ์‚ฌ์šฉํ•˜์—ฌ ๋ฌธ์˜ํ•˜์‹ญ์‹œ์˜ค.

https://forms.gle/WvT1wiN1qDtmnspy7