a puzzle language & mind - linguistics at...

1

Language & MindHONR 218L

Development of Speech Perception

A Puzzle

• Korean speakers use the sounds [r] and [l]e.g. Korea

Seoul• Korean babies hear the difference between [ra]

and [la] … they don’t know Korean yet

• Korean adults know Korean … but they havedifficulty hearing the [ra] vs. [la] contrast

Speech Perception

A speaker of a language is aprisoner of its ‘alphabet’ ofsounds

Cross-language Differences

R L

R L

2

Developmental Questions

• How does the native/non-native differenceemerge?

• What is the nature of the ‘learning’ that isinvolved?

• When does the native/non-native differenceemerge?





Possibility #1: Adding Features

• Children learn the feature contrasts of theirlanguage

• Children learn gradually, adding featuresover the course of development

Roman Jakobson, 1896-1982

Possibility #1: Predictions

• Poor discriminators at birth• Better and better with age


3

English VOT Perception

To Test Children

Not so easy!

High AmplitudeSucking

Reality Check for Possibility #1:

• Infants’ show Categorical Perception ofspeech sounds - at 2 months and earlier

• Discriminate a wide range of speechcontrasts

• Discriminate Non-Native speech contrastse.g., Japanese babies discriminate r-le.g., Canadian babies discriminate d-D

Universal Listeners

• Infants may be able to discriminate allspeech contrasts from the languages of theworld!

How can they do this?

• Innate speech-processing capacity?• General properties of auditory system?

4

What About Non-Humans?

• Chinchillas show categorical perception ofvoicing contrasts!

Connecting Hearing & Speaking

Auditory [ba] + Visual [ga] = Perceptual [da]

McGurkEffect

Connecting Hearing & Speaking

Auditory [ba] + Visual [ga] = Perceptual [da]

Infants start to connect auditoryand visual information early

• Infants knowconnectionbetween visualand auditoryspeech stimuli

• Mix and match[a] vs. [i]

5





When does Change Occur?

• About 10 months

Janet Werker

U. of British Columbia

When does Change Occur?

• Hindi and Salishcontrasts testedon English kids

Janet Werker

U. of British Columbia Conditioned Headturn Procedure





6

Possibility #2: Maintenance & Loss

• Infants maintain features they hearbeing used in their language

• They lose all others

Patricia KuhlUniversity of Washington

Possibility #2: Maintenance & Loss

• Anotherschematic:

Possibility #2: Predictions

• Loss of discriminability should bepermanent and absolute

Possibility #2: Reality Check

• Loss of discriminability IS NEITHERpermanent NOR absolute

7

Training improves adultperformance

Some non-native contrasts are easyfor adults to discriminate

Adults perform better at non-native contrasts if they think thesounds are not language sounds!

We seem to have different “ears” forlanguage sounds vs. other sounds

8





Possibility #3: FunctionalReorganization

• Changes in performance withdevelopment do not reflect changes inthe hard-wiring of the brain

• They reflect the operation ofpostperceptual processes that allow theinfant to attend to only phoneticinformation that distinguishes meaning

Janet WerkerUniversity of British Columbia

Possibility #3: FunctionalReorganization

• Schematically (from Phillips 2001)

What 1-year-olds can’t do

• Recognize minimal pairs while relatingthem to real words– “bear” versus “pair”– “Piglet” versus *“Biglet”

• More Werker experiments

9

Word Learning

• Stager &Werker 1997

‘bih’ vs. ‘dih’and‘lif’ vs. ‘neem’

Word learning results

• Exp 2 vs 4

Why 1-year-olds Fail on MinimalPairs

• They fail specifically when the task requires word-learning

• They do know the sounds• But they fail to use the detail needed for minimal

pairs to store words in memory• What is going on?

Swingley & Aslin, 2002• 14-month olds did recognize mispronunciations of familiar

words

10

Werker et al. 2002More…

• Word-learning is very hard for younger children,so detail is initially missed when they first learnwords

• Many exposures are needed to learn detailed wordforms at early stages of word-learning

• Success on the Werker/Stager task seems to berelated to the vocabulary spurt, rapid growth invocabulary after ~50 words

One-Year Olds Again

• One-year olds know the surface soundpatterns of the language

• One-year olds do not yet know whichsounds are used contrastively in thelanguage…

• …and which sounds simply reflect variationthat does not affect meaning

• One-year olds need to learn contrasts

We need another level!

11

Same or different?

light liedtight tiedsite sighedlife liveknife knive(s)lice liesdice dies

Some people have this system:

light lajt lied lajd

tight tajt tied tajd

site sajt sighed sajd

life lajf live lajv

knife najf knive(s) najvz

lice lajs lies lajz

dice dajs dies dajz

Some people have this one:

light l√jt lied lajd

tight t√jt tied tajd

site s√jt sighed sajd

life l√jf live lajv

knife n√jf knive(s) najvz

lice l√js lies lajz

dice d√js dies dajz

What’s the pattern?

light l√jt lied lajd

tight t√jt tied tajd

site s√jt sighed sajd

life l√jf live lajv

knife n√jf knive(s) najvz

lice l√js lies lajz

dice d√js dies dajz

12


t d

t d

t d

f v

f v

s z

s z

voicelessalveolarstop

voicelesslabiodentalfricativevoicelessalveopalatalfricative

voicedalveolarstop

voicedlabiodentalfricativevoicedalveopalatalfricative


t d

t d

t d

f v

f v

s z

s z

voicelessalveolarstop

voicelesslabiodentalfricativevoicelessalveopalatalfricative

voicedalveolarstop

voicedlabiodentalfricativevoicedalveopalatalfricative

So these speakers have a rule ...

Before a voiceless consonantaj _ √j

Isn’t it just two sets of words?

13

l√js lajzd√js dajzl√jf lajzs√jt sajdl√jt lajd

Isn’t it just two sets of words?

l√js lajzd√js dajzl√jf lajzs√jt sajdl√jt lajd

Nope, it’s a rule ...

s t i gh t

s t i e d

[stajd]Nope, it’s a rule ...

[st√jt]

Two “levels” of speech sounds

The sounds you “store” in your head

14



/lajt/

/lajd/


The sounds you actually produce



[l√jt]

[lajd]


/lajt/

/lajd/

“phonemes”

TerminologyTerminology

15


[l√jt]

[lajd]

TerminologyTerminology

“phones”[√j]

/aj/

[aj]We call the phonesWe call the phonesallophonesallophones of the of thephonemephoneme

1 phoneme; more than 1 phone

So:

• In some dialects of English, thephoneme /aj/ has two allophones: [aj]and [√j].

• The allophone [√j] occurs wheneverthe phoneme precedes a voicelesssound

Or, more succinctly:

aj

16


aj

This phoneme


aj_

becomes


aj_√j

this sound(allophone)


aj_√j/

where?

17


aj_√j/____

immediately before


aj_√j/____[-voice]

a voiceless sound

The whole rule:

aj_√j/____[-voice]

Another rule:

t_d/V____V

18

• “sit” [sˆt]• “sitter” [sˆd\r]• “heat” [hit]• “heater” [hid\r]• “at” [æt]• “attic” [ædˆk]

Another rule:

• “attack”• “atone”• “determine”• “detect”

What about these?

[sˆd\r] [\tæk][hid\r] [\ton][ædˆk] [dit´kt]

Is there a pattern?

[sˆd\r] [\tæk][hid\r] [\ton][ædˆk] [dit´kt]

Is there a pattern?

19

So we need a slight revision

t_d/V____V

1 phoneme

/t/

(sound storedin the brain)

[d]

1 phoneme 2 allophones

/t/

[t]sound(s) actually produced

/t,f,p.../What do you have

in your head?

sounds(phonemes)

20

rules

/t,f,p.../What do you have

in your head?

rules

/t/

rules[d][t]

Points to note:

• Sequence becomes “easier to say”BUT• This process is a specific rule of a particular

dialect of English

In what sense a specific rule?

• doesn’t apply to all instances of “t” betweenvowels

• isn’t a part of the grammar of other dialectsof English

• is only one way to make sequences ofvowels and voiceless consonants easier tosay

21

Moral:

The rules that we discover are often“natural” in that one can find an explanationfor many of them in terms of ease ofarticulation, but they are notinevitable/innate: they are specific rules ofparticular dialects or languages, and had tobe learned.

A way to think about phonemes

• You have a “mental dictionary” where youstore the words you have learned.

• What information does it have to contain foreach word?

• At least:• the meaning of the word• the sound of the word

A way to think about phonemes

• How much detail do you have to rememberabout the sound of each word?

If you can predict something by arule, you don’t have to remember

it

Just remember:

•• the the rulerule•• the things that the things that cancan’’tt be be

predictedpredicted

22

/aj/

rules

[aj][√j]

Allophonic differences ignoredby hearers

/aj/

rules

[aj][√j]


/aj/

rules

[aj][√j]

rules

/aj/


• Voiceless stops /p, t, k/

• Aspirated at start of syllable; unaspirated after [s]• 6 month olds easily distinguish bottom 2 rows; 1 year olds

do not (adults aren’t great either)

pitspitspitbit

tackstackstackdack

Varying Pronunciations

23

Languages can differ in whatis predictable

Korean has [l] & [r]

[rupi] “ruby”[kiri] “road”[saram] “person”[ir}mi] “name”[ratio] “radio”[mul] “water”[pal] “big”[s\ul] “Seoul”[ilkop] “seven”[ipalsa] “barber”



But [r] doesn’t show up everywhere ...


24



[r] is alwaysin front of avowel

And nor does [l] ...




[l] is neverin front of avowel

/l/

rules

So: Korean has only 1 liquid phoneme

(Koreans don(Koreans don’’t have t have to remember if a wordto remember if a wordhas has [l][l] or or [r][r]))

25

/l/

rules

[r][l]

... but that phoneme has 2 allophones

/l/

rules

[r][l]

rules

So in Korean, [l] and [r] are “the same”

/l/

So Korean works like this:

1 phoneme

/l/

(sound storedin the brain)

26

[r]

1 phoneme 2 allophones

/l/

[l]sound(s) actually produced

While English works like this:

(sounds storedin the brain)

2 phonemes

/l//r/ [l]

2 phonemes 2 phones

[r]

/l//r/

sound(s) actually produced

27

Even more schematically:Stored Produced

English/l//r/

[l]

Even more schematically:Stored Produced

English

Korean

/l//r/

[l]

/l/[r][l][r]

Minimal Pairs• In English, [r] and [l] can occur in the same position in a

wordrake lakeramp lamprim limbripper ripple

• In English, [r] and [l] can be used to mark a meaningcontrast

• In English, /r/ and /l/ are two phonemes

Minimal Pairs

• Korean works differently

28





[r] is alwaysin front of avowel



[l] is neverin front of avowel

Minimal Pairs

• Korean works differently• [r] and [l] are two allophones of a single

phoneme in Korean• It’s impossible to create minimal pairs

which contrast r/l in Korean• [r] and [l] cannot be used contrastively in

Korean

29

One-Year Olds Again

• One-year olds know the surface soundpatterns of the language

• One-year olds do not yet know whichsounds are used contrastively in thelanguage…

• …and which sounds simply reflectpronunciation rules

• One-year olds need to learn contrasts

Word Learning

• Stager &Werker 1997

‘bih’ vs. ‘dih’and‘lif’ vs. ‘neem’

Maybe not so bad after all...

• Children learn the feature contrasts of theirlanguage

• Children learn gradually, adding featuresover the course of development


Puzzle Solved!

• Korean speakers use the sounds [r] and [l]e.g. Korea

Seoul• Korean babies hear the difference between

[ra] and [la] … they don’t know Korean yet• Korean adults know Korean … but they

have difficulty hearing the [ra] vs. [la]contrast

30

Where’s Whorf?• We do “hear” language sounds differently

depending on what language we speak. But whenwe fail to hear a contrast that a speaker of anotherlanguage does hear, it isn’t because our physicalability to register that difference has disappeared.It’s because we have learned that that type ofcontrast is not a meaningful contrast for language.

• Our mental representations of the sounds of wordsare an abstraction from the physical signal.Allophones of /t/ are all the same sound in ourminds.

a puzzle language & mind - linguistics at...

Documents