This article appeared in a journal published by Elsevier. The

This article appeared in a journal published by Elsevier. The attached

copy is furnished to the author for internal non-commercial research

and education use, including for instruction at the authors institution

and sharing with colleagues.

Other uses, including reproduction and distribution, or selling or

licensing copies, or posting to personal, institutional or third party

websites are prohibited.

In most cases authors are permitted to post their version of the

article (e.g. in Word or Tex form) to their personal website or

institutional repository. Authors requiring further information

regarding Elsevier’s archiving and manuscript policies are

encouraged to visit:

http://www.elsevier.com/copyright

Author's personal copy

Infant

Behavior

&

Development

34 (2011) 590–

601

Contents

lists

available

at

ScienceDirect

Infant

Behavior

and

Development

Do

production

patterns

inﬂuence

the

processing

of

speech

in

prelinguistic

infants?

Rory

A.

DePaolis

a,∗

,

Marilyn

M.

Vihman

b,1

,

Tamar

Keren-Portnoy

b,1

a

Communication

Sciences

and

Disorders,

James

Madison

University,

MSC

#4304,

Harrisonburg,

VA

22807,

United

States

b

Language

and

Linguistic

Science,

University

of

York,

V/C/210,

2nd

Floor,

Block

C,

Vanbrugh

College,

University

of

York,

Heslington,

York

YO10

5DD,

United

Kingdom

a

r

t

i

c

l

e

i

n

f

o

Article

history:

Received

26

October

2010

Received

in

revised

form

12

February

2011

Accepted

23

June

2011

Keywords:

Infant

speech

perception

Infant

speech

production

Headturn

preference

paradigm

The

articulatory

ﬁlter

Babbling

Vocal

motor

schemes

a

b

s

t

r

a

c

t

The

headturn

preference

procedure

was

used

to

test

18

infants

on

their

response

to

three

different

passages

chosen

to

reﬂect

their

individual

production

patterns.

The

passages

con-

tained

nonwords

with

consonants

in

one

of

three

categories:

(a)

often

produced

by

that

infant

(‘own’),

(b)

rarely

produced

by

that

infant

but

common

at

that

age

(‘other’),

and

(c)

not

generally

produced

by

infants.

Infants

who

had

a

single

‘own’

consonant

showed

no

sig-

niﬁcant

preference

for

either

‘own’

(a)

or

‘other’

(b)

passages.

In

contrast,

infants’

with

two

‘own’

consonants

exhibited

greater

attention

to

‘other’

passages

(b).

Both

groups

attended

equally

to

the

passage

featuring

consonants

rarely

produced

by

infants

of

that

age

(c).

An

analysis

of

a

sample

of

the

infant-directed

speech

ruled

out

the

mothers’

speech

as

a

source

of

the

infant

preferences.

The

production-based

shift

to

a

focus

on

the

‘other’

passage

sug-

gests

that

nascent

production

abilities

combine

with

emergent

perceptual

experience

to

facilitate

word

learning.

1.

Introduction

One

factor

often

overlooked

in

the

research

into

infant

speech

perception

is

the

effect

that

early

pre-lexical

production

or

babble

may

have

on

perception

of

and/or

attention

to

incoming

speech,

even

though

infants

typically

begin

rhythmic

production

of

adult-like

syllables

–

‘canonical

babbling’

–

between

six

and

eight

months

(Oller,

2000),

the

period

of

the

ﬁrst

major

advances

in

speech

perception

(cf.,

e.g.,

Jusczyk

&

Aslin,

1995;

Jusczyk,

Houston,

&

Newsome,

1999;

Nazzi,

Jusczyk,

&

Johnson,

2000;

Shi

&

Werker,

2001;

Soderstrom,

Kemler-Nelson,

&

Jusczyk,

2005;

Soderstrom,

Seidl,

Kemler-Nelson,

&

Jusczyk,

2003;

Tincoff

&

Jusczyk,

1999).

This

study

tests

the

idea

that

increasing

use

of

consonants

in

production

will

directly

affect

the

processing

of

running

speech.

From

a

Dynamic

Systems

perspective,

language

development

can

be

viewed

as

a

process

in

which

relatively

simple

skills

interact

to

create

more

complex

ones

(Thelen,

1991).

Babble

is

one

such

simple

skill.

Its

effect

on

speech

perception

has

not

yet

been

seriously

investigated

despite

the

fact

that

there

is

ample

evidence

that

motoric

experiences

affect

perception,

with

wide-ranging

effects

on

social

as

well

as

cognitive

development.

For

example,

Piaget

(1952)

emphasized

the

importance

of

the

child

experiencing

and

acting

on

the

world,

suggesting

that

intelligence

is

derived

from

sensorimotor

activity.

Recent

work

with

locomotion

has

also

highlighted

the

way

in

which

secondary

effects

of

self-produced

locomotion

can

initiate

more

complex

cognitive

advances

(see

Campos

et

al.,

2000

for

a

review).

In

brief,

the

onset

of

self-produced

locomotion

leads

to

∗

Corresponding

author.

Tel.:

+1

540

568

3869.

E-mail

addresses:

[email protected]

(R.A.

DePaolis),

[email protected]

(M.M.

Vihman),

[email protected]

(T.

Keren-Portnoy).

1

Tel.:

+44

1904

433612.

0163-6383/$

–

see

front

matter ©

doi:10.1016/j.infbeh.2011.06.005

Author's personal copy

R.A.

DePaolis

et

al.

/

Infant

Behavior

&

Development

34 (2011) 590–

601 591

improvements

in

communicative

gesturing,

attention,

spatial

search,

visual-vestibular

coupling

and

depth

perception.

Thus,

a

notable

consequence

of

this

motoric

advance

is

its

ripple

effect

across

multiple

cognitive

domains.

There

is

evidence

that

babble

produces

a

similar

ripple

effect.

McCune

and

Vihman

(2001)

tracked

the

emergence

of

frequently

used

consonants

(‘vocal

motor

schemes’:

VMS),

on

the

assumption

that

repeated

practice

with

a

particular

pho-

netic

form

might

lessen

the

processing

load

of

recognizing

or

categorizing

sound

sequences

(or

word

forms)

that

contain

such

a

consonant,

freeing

up

processing

resources

for

the

pairing

of

form

and

meaning.

This

would

facilitate

the

learning

of

referential

words.

McCune

and

Vihman

deﬁned

a

VMS

as

a

supraglottal

consonant

(stop,

nasal,

fricative,

excluding

/h/,

or

affricate)

that

the

infant

produced

consistently

and

stably

over

several

observational

sessions.

They

found

that

the

use

of

at

least

two

VMS

was

a

prerequisite

for

the

transition

into

referential

word

use

in

the

20

children

they

observed.

In

the

same

way

that

self-produced

locomotion

facilitates

the

development

of

spatial

awareness,

the

ability

to

produce

consistent

patterns

in

babble

can

be

taken

to

support

memory

for

word

forms

(Keren-Portnoy,

Vihman,

DePaolis,

Whitaker,

&

Williams,

2010),

which

in

turn

facilitates

the

recognition

that

a

word

form

can

symbolize

events

or

entities

in

the

world.

This

relationship

between

VMS

and

referential

word

use

raises

the

possibility

that

babble

might

be

an

early

stepping

stone

that

has

effects

on

other

aspects

of

cognitive

development.

For

example,

as

babbling

begins

to

systematically

and

consistently

incorporate

one

or

more

adult-like

consonants,

it

could

potentially

speed

the

processing

of

these

practiced

sounds

when

heard

in

input

speech

as

well

as

when

self-produced.

Logically,

a

child’s

knowledge

of

the

speech

sounds

that

he

or

she

has

produced

might

well

be

stronger

and

richer

than

the

knowledge

of

sounds

not

yet

produced.

Babbling

necessarily

involves

hearing

one’s

own

vocal

output

and

integrating

auditory

and

proprioceptive

percepts,

although

the

process

need

not

be

considered

conscious

or

explicit.

Thus,

sounds

produced

by

the

child

provide

the

double

information

afforded

by

both

auditory

and

articulatory

experience.

This

suggests

that

babbling

that

leads

to

regularly

produced

consonants

should

boost

the

perceptual

salience

of

practiced

sounds

in

adult

(input)

speech

as

well

as

in

self-produced

vocalizations

(see

Elbers,

1997,

who

emphasizes

the

importance

of

‘output

as

input’).

It

has

been

suggested

that

an

infant

develops

an

individually

fashioned

‘articulatory

ﬁlter’,

based

on

the

particular

sound

patterns

that

the

infant

has

mastered

motorically

(Vihman,

1993,

1996).

According

to

the

articulatory

ﬁlter

hypothesis,

familiarity

with

speech

sounds

from

an

infant’s

own

production

implicitly

enhances

the

salience

of

those

same

sound

patterns

when

they

occur

in

surrounding

speech

(see

also

Locke,

1993,

p.

204).

The

articulatory

ﬁlter

was

proposed

to

account

for

the

ﬁnding

that

infants’

ﬁrst

words

tend

to

be

produced

accurately

as

well

as

being

comprised

of

consonants

that

the

infant

is

experienced

in

producing.

The

purpose

of

this

study

was

to

test

the

hypothesis

that

there

is

a

direct

inﬂuence

of

production

on

what

is

salient

in

input

speech.

We

do

not

mean

to

suggest

that

this

is

the

only

direction

of

causality

between

production

and

perception.

On

the

contrary,

the

effect

of

perception

on

early

infant

production

has

already

been

demonstrated

in

several

studies.

Speciﬁcally,

acoustic

analysis

has

shown

early

ambient

language

effects

on

babbling

(at

6–12

months

for

prosody:

Whalen,

Levitt,

&

Wang,

1991

and

at

10

months

for

vowel

and

consonant

production:

Boysson-Bardies,

Halle,

Sagart,

&

Durand,

1989,

Boysson-Bardies

&

Vihman,

1991).

In

each

case

the

ﬁndings

reﬂect

a

biasing

of

the

child’s

output

in

the

direction

of

the

ambient

language.

However,

this

study

was

designed

to

explore

the

possibility

of

the

reverse

effect,

that

is,

the

idea

that

a

child’s

production

experience

might

also

affect

the

way

he

or

she

listens

to

speech.

It

is

important

to

recognize

that

we

do

not

know

how

an

infant

settles

upon

his

or

her

own

well-practiced

(or

favorite)

consonants.

This

process

must

be,

at

the

very

least,

a

combination

of

biological

predisposition

(Davis

&

MacNeilage,

1995),

perceptual

salience

(Lindblom,

1992),

input

frequencies

and

their

effect

on

the

development

of

speech

category

was

a

function

of

the

normal

developmental

process.

The

median

age

at

testing,

for

single-

and

multiple-VMS

groups,

respectively,

was

0;10.24

and

0;10.10.

2.2.2.

Stimuli

A

native

speaker

of

British

English,

using

a

lively

speech

style,

recorded

passages

consisting

of

ﬁve

sentences,

with

nonwords

near

both

the

beginning

and

the

end

of

each

of

four

of

the

sentences

and

a

ﬁfth

sentence

with

a

single

medial

nonword

receiving

phrasal

accent.

The

duration

of

the

passages

was

between

13

and

14

s.

The

maximum

length

for

both

the

pretest

and

test

trials

was

13.8

s.

All

items

were

recorded

in

a

sound-treated

room

using

a

Sennheiser

ME

66

microphone

(with

K6

power

module)

connected

to

a

Tascam

DA-P1

digital

recorder

sampling

at

44.1

kHz.

The

speech

peaks

of

all

stimuli

were

within

±1.5

dB

as

measured

on

a

mechanical

VU

meter

(Marantz

PMD

222).

The

stimuli

were

transferred

digitally

onto

a

PC

hard

drive

for

eventual

output.

Acoustic

analysis

across

the

three

passages

revealed

no

difference

in

amplitude

(rms

and

peak),

F0

(mean,

minimum,

maximum,

and

range)

or

duration

(p

>

.05).

The

nonwords

used

as

stimuli

are

listed

in

Table

1.

Each

nonword

consisted

of

a

disyllable

with

a

CVCVC

structure.

The

consonant

was

a

stop

or

fricative

with

a

consistent

place

of

articulation

that

alternated

in

voicing

(i.e.

the

/t/

and

/d/

in

the

nonword

/tId␧t/),

consistent

with

the

deﬁnition

of

a

VMS

(given

above).

The

consonant-vowel

co-occurrences

used

were

consistent

with

those

predicted

by

the

frame

dominance

model

of

early

vocalizations

(Davis

&

MacNeilage,

1995),

which

successfully

predicts

that

infants’

early

babble

will

follow

patterns

dictated

by

the

physiology

of

mandibular

oscillations

(or

frames)

–

i.e.,

of

whole-syllable

production,

with

alveolars

followed

by

front

vowels,

velars

by

back

vowels

and

labials

by

central

vowels.

The

three

carrier

passages

are

listed

in

Table

2.

Each

carrier

passage

contained

24

syllables

in

addition

to

nine

Table

2

Carrier

passages

with

open

slots

for

the

nonwords

in

Table

1.

Passage

1

was

recorded

with

/t,d/,

passage

2

with

/p,b/,

passage

3

with

/k,g/;

all

three

passages

were

recorded

with

/v,f/

so

that

each

infant

would

hear

a

different

passage

for

each

consonant.

Passage

1

So,

who

should

the

away?

I

can

the

now.

But

go

for

a

while.

This

does

a

for

you.

Will

you

to

me?

Passage

2

Wow,

my

is

a

one.

Did

the

go

below?

We

call

a

lot.

Are

your

too

over

there?

I

see

the

here.

Passage

3

The

are

by

a

there.

I

may

the

along.

Oh,

the

is

a

now.

So

they

a

away.

Can

you

play

too?

Author's personal copy

594 R.A.

DePaolis

et

al.

/

Infant

Behavior

&

Development

34 (2011) 590–

601

disyllabic

nonwords

including

the

targeted

VMS

consonants

only.

Each

passage

highlighted

one

pattern:

Own-VMS,

Other-

VMS

or

Non-VMS.

In

order

to

avoid

changes

of

consonant

manner

between

Own-

and

Other-VMS

passages,

only

bilabial,

alveolar

and

velar

stops

were

used

as

Own-

or

Other-VMS.

The

Non-VMS

passage

contained

nonsense

words

including

as

consonants

only

/f/

and

/v/,

which

are

not

typically

mastered

by

infants

until

the

second

or

third

year

of

life

(Ferguson,

1975).

No

infant

in

this

study

produced

/f/

or

/v/

to

VMS

criterion.

We

decided

not

to

use

nasals

(which

are

also

early

favored

consonants)

as

VMS

in

the

Head

Turn

task

but

to

limit

ourselves

to

stop

consonant

VMS.

Nasals

have

distinctive

acoustic

signatures

–

such

as

shifts

and/or

reductions

in

energy

of

the

formants

of

the

surrounding

vowels

(Kent

&

Read,

2002)

and

high

average

amplitude

during

their

occluded

portion

– that

make

them

perceptually

distinct

from

stops.

Even

though

nasals

were

not

included

in

the

HPP

task,

it

was

possible

for

an

infant

to

have

multiple

VMS

with

a

stop

consonant

and

a

nasal

rather

than

two

stop

consonants.

Note

that

any

supraglottal

consonant

that

the

infant

produced

to

VMS

criterion

would

have

counted

as

a

VMS.

McCune

and

Vihman

(2001)

found

that

a

small

number

of

infants

produced

/s/

or

/l/

to

VMS

criterion.

In

this

study

only

alveolar,

bilabial

and

velar

stops

and

velar

and

bilabial

nasals

were

produced

to

VMS

criterion.

Only

infants

with

at

least

one

stop

VMS

could

be

included

in

the

sample

since

we

chose

not

to

use

nasals

as

test

stimuli.

We

did

not

actually

have

to

exclude

any

infants

since

a

stop

was

either

the

ﬁrst

or

the

second

VMS

for

every

participant.

Passage

1

was

recorded

with

/t,d/,

passage

2

with

/p,b/,

passage

3

with

/k,g/;

the

passages

containing

/t,d/,

/p,b/,

or

/k,g/

were

used

as

Own-

or

Other-VMS

passages,

depending

in

each

case

on

which

VMS

a

particular

infant

had

already

acquired

(Own)

or

had

not

yet

acquired

(Other).

All

three

carrier

passages

were

recorded

with

/v,f/

so

that

each

infant

would

hear

three

different

carrier

passages,

one

for

each

type

of

consonant.

The

passages

with

/f,v/

were

used

as

Non-VMS

passages.

2.2.3.

Procedures

The

HT

procedure

used

was

similar

to

that

described

in

Kemler-Nelson

et

al.

(1995).

Seated

on

the

caregiver’s

lap

in

a

quiet

darkened

room,

the

infants

faced

the

central

panel

of

a

three-sided

test

booth

where

a

camera

and

red

light

were

mounted.

A

blue

light

and

speaker

were

mounted

on

each

side

panel.

A

PC

and

video

monitor

were

located

in

the

adjoining

room

where

the

experimenter

controlled

stimulus

presentation

and

recorded

infant

looking

times

by

pressing

the

left

and

right

mouse

buttons.

The

computer

initiated

and

terminated

trials

in

response

to

signals

from

the

experimenter.

In

each

trial,

the

infant’s

gaze

was

centered

by

the

blinking

red

light.

The

experimenter

then

initiated

the

computer

run

trial

involving

a

blinking

blue

light

to

the

left

or

to

the

right

of

the

infant.

When

the

infant

was

judged

to

orient

to

the

blue

light,

a

trial

was

presented

from

that

speaker.

If

the

infant

looked

away

from

the

speaker

for

more

than

two

seconds,

the

trial

was

terminated

and

another

begun.

Multi-talker

babble

created

from

the

same

speaker

of

the

stimuli

used

in

the

experiment

was

delivered

to

the

headphones

worn

by

the

experimenter

and

caregiver

to

mask

the

actual

test

stimuli.

The

caregiver

also

wore

foam-insert

hearing

protection.

All

stimuli

were

presented

at

an

average

level

of

65

dB

(Tenma

72-6635

sound

level

meter).

Each

experimental

session

consisted

of

a

pretest

and

test

phase.

In

the

pretest

phase

the

infant

was

presented

with

one

passage

of

each

of

the

three

test

conditions,

Own-VMS,

Other-VMS,

and

Non-VMS,

counterbalanced

for

order

and

randomized

for

side.

This

condition

was

intended

to

expose

the

infant

to

the

test

procedures

since

our

using

the

headturn

paradigm

have

indicated

that

the

initial

trials

lead

to

overly

long

looking

times

that

do

not

seem

to

be

indexed

to

the

type

of

stimuli

presented.

The

test

phase

of

the

experiment

consisted

of

15

trials,

ﬁve

each

of

the

three

test

conditions.

Each

trial,

pretest

and

test,

consisted

of

a

randomized

presentation

of

the

ﬁve

sentences

of

each

test

passage.

The

order

of

presentation

in

the

test

phase

was

such

that

the

ﬁrst

three

trials

were

counterbalanced

across

test

conditions.

The

order

of

the

ﬁnal

three

trials

was

a

reverse

of

the

ﬁrst

three.

The

counterbalancing

at

both

the

beginning

and

the

end

was

designed

to

control

for

an

anticipated

decrease

in

looking

times,

independent

of

the

stimuli,

over

the

course

of

the

test

trials

(see

Vihman,

Nakai,

DePaolis,

&

Hallé,

2004

for

an

analysis

of

looking

time

by

trial).

The

middle

nine

trials

were

pseudo-randomized

such

that

no

more

than

two

identical

test

trials

occurred

together.

In

both

phases,

the

side

of

presentation

was

pseudo-randomized

such

that

no

more

than

three

successive

presentations

from

one

side

were

allowed.

Reliability

was

assessed

by

ofﬂine

coding,

by

a

separate

researcher,

of

two

of

the

infants’

HPP

video

recordings

(r

=

.927,

p

<

.01).

3.

Results

3.1.

Production

data

The

results

presented

below

are

drawn

from

the

18

infants

who

successfully

participated

in

the

headturn

experiment.

Table

3

documents

the

age

at

test

and

the

particular

Own-

and

Other-VMS

used

in

the

perception

test

for

each

infant

in

the

two

groups;

additionally,

the

VMS

not

tested

is

given

for

the

multiple-VMS

infants

only

(‘Non-Test

VMS’).

Interestingly,

all

but

two

infants

acquired

/t,d/

as

one

of

their

ﬁrst

VMS,

consistent

with

the

McCune

and

Vihman

(2001)

study,

in

which

13

of

the

20

infants

followed

acquired

a

coronal

ﬁrst;

we

return

to

this

point

in

the

discussion.

The

second

VMS

acquired

by

the

nine

infants

in

the

multiple-VMS

group

was

variable,

four

producing

/p,b/,

three

producing

/n/,

two

producing

/m/

and

one

producing

/k,g/.

Fig.

1

presents

a

count

of

consonants

produced

in

the

session

prior

to

the

HPP.

For

each

infant

we

present

a

token

count

of

all

productions

of

that

speciﬁc

infant’s

Own-

and

Other-VMS

consonants

(the

same

two

consonants

to

be

tested).

The

ﬁgure

also

presents

the

total

token

frequency

in

that

same

session

of

the

two

most

often

used

consonants

for

each

infant,

as

well

as

Author's personal copy

R.A.

DePaolis

et

al.

/

Infant

Behavior

&

Development

34 (2011) 590–

601 595

Table

3

Individual

infants’

Own-

or

Other-VMS

consonants

used

in

the

headturn

experiment.

The

preference

ratio

is

for

Own-

versus

Other-VMS

(preference

ratio

=

Own/(Own

+

Other).

The

number

of

sessions

indicates

the

total

number

of

30-min

observational

sessions

required

to

reach

VMS

for

each

infant.

Age

at

test

Own

Other

Non-test

VMS

preference

ratio

#

sessions

Single

0;10.1

t,d

k,g

.56

3

0;10.3

t,d

k,g

.47

4

0;10.7

t,d

p,b

.52

2

0;10.21 t,d k,g .64 4

0;10.24 t,d k,g .47 7

1;.25

t,d

k,g

.68

2

1;3.17

t,d

k,g

.44

4

1;4.0

p,b

k,g

.64

5

1;4.0

t,d

p,b

.54

5

Multiple

0;9.8

p,b

k,g

t,d

.36

1

0;9.29

t,d

k,g

n

.48

2

0;10.6

t,d

p,b

n

.47

2

0;10.8 p,b k,g t,d .46 4

0;10.10

t,d

p,b

m

.37

2

0;10.20 t,d

k,g

p,b

.47

2

0;11.0

k,g

p,b

t,d

.44

5

0;11.5

t,d

k,g

n

.44

3

0;11.25

p,b

t,d

m

.58

10

Fig.

1.

Production

data

for

each

participant

from

the

session

the

HPP

test.

Own-

and

Other-VMS

consonant

counts

refer

to

the

contrasts

used

in

the

HPP

test.

Two

Highest

refers

to

the

summed

frequency

of

the

two

most-practiced

consonants

for

each

infant

(for

2-VMS

infants

this

is

the

summed

frequency

of

the

ﬁrst

and

second

VMS).

The

preference

ratios

for

the

HPP

test

are

also

plotted

in

the

dotted

line

against

the

scale

on

the

right.

The

data

are

arranged

in

ascending

order

of

Two

Highest

consonant

counts.

the

results

of

the

perception

test

(see

below

for

a

discussion

of

the

preference

ratio).

2

A

consonant

was

included

in

the

count

if

it

was

transcribed

within

a

syllable

consistent

with

English

phonotactics.

The

single-VMS

group

produced

a

mean

of

65.7

consonants

(SD

=

15.3),

as

against

122.1

(SD

=

20.2)

for

the

multiple-VMS

group.

This

difference

is

signiﬁcant,

t(16)

=

2.228,

p

=

.041,

suggesting

that

the

division

of

children

into

single

and

multiple

VMS

groups

is

based

not

only

upon

the

repeated

and

consistent

phonetic

form

of

at

least

two

consonants,

but

also

upon

the

children’s

overall

frequency

of

consonant

production.

3.2.

Perception

Test

The

results

are

presented

in

Fig.

2.

A

mixed

two-factor

analysis

of

variance

performed

on

the

test

trials,

with

VMS

Type

(Own,

Other,

and

Non)

and

VMS

Number

(Single

versus

Multiple)

as

independent

variables

and

the

looking

time

to

each

passage

as

the

dependent

variable,

revealed

no

main

effects

for

either

VMS

Type

(F[2,16]

=

.492,

p

=

.616,



2

=

.051)

or

VMS

Number

(F[1,16]

=

.349,

p

=

.563,



2

=

.021).

However,

the

interaction

between

VMS

Type

and

VMS

Number

was

signiﬁcant

(F[2,16]

=

3.933,

p

=

.024,



2

=

.391),

indicating

that

the

preference

for

passages

was

dependent

upon

the

number

of

VMS

the

infants

had

acquired.

Contrasts

of

the

interaction

between

Non

and

Own

versus

VMS

Number

(and

ignoring

Other-VMS)

were

not

signiﬁcant

(p

=

.424),

while

the

contrast

of

the

interaction

between

Non

and

Other

versus

VMS

Number

(and

ignoring

Own-VMS)

approached

signiﬁcance

(p

=

.062).

These

results,

when

taken

together

with

the

lack

of

a

main

effect

of

VMS

type,

suggest

that

looking

times

to

the

Non-VMS

passages

are

independent

of

the

infants’

production

patterns.

The

contrast

of

2

Note,

that

infants

with

a

single

VMS

are

not

infants

who

never

produce

any

other

consonants,

but

those

who

never

produce

any

other

consonant

to

VMS

criterion.

Author's personal copy

596 R.A.

DePaolis

et

al.

/

Infant

Behavior

&

Development

34 (2011) 590–

601

Fig.

2.

Average

looking

times

per

test

trial

in

the

headturn

experiment.

the

interaction

between

Own-

and

Other-VMS

versus

VMS

Number

was

signiﬁcant

(p

=

.01),

indicating

that

the

signiﬁcant

interaction

in

the

ANOVA

is

mainly

due

to

the

differences

in

looking

times

of

the

single-

and

multiple-VMS

infants

to

the

Own-

and

Other-VMS

passages.

Paired

t-tests

of

the

main

contrast

of

Own-

versus

Other-VMS

show

that

the

multiple-VMS

infants

looked

signiﬁcantly

longer

in

response

to

the

Other

consonant

(M

=

6.61,

SD

=

1.52

versus

M

=

5.48,

SD

=

1.49,

respectively,

t(8)

=

−2.519,

p

=

.036),

while

the

single-VMS

infants

looked

longer

in

response

to

their

own

consonant

(M

=

5.98,

SD

=

2.05

versus

M

=

4.99,

SD

=

2.20,

respectively),

but

this

preference

is

not

signiﬁcant

(t(8)

=

1.735,

p

=

.121).

The

lack

of

an

effect

in

the

single-VMS

group

was

likely

due

to

the

variability

of

overall

looking

times

in

this

group

(range

2.15–8.6

s).

Often,

in

a

headturn

experiment,

infants

with

very

low

looking

times

or

atypical

values

are

excluded

from

the

ﬁnal

analysis

(see,

for

example,

Nazzi,

Iakimova,

Bertoncini,

Frédonie,

&

Alcantara,

2006).

In

this

study,

we

chose

not

to

exclude

infants

for

these

reasons

since

the

investment

in

time

for

each

infant

was

so

large.

One

way

to

factor

out

differences

in

overall

looking

time

across

infants

and

compare

the

infants’

preference

for

practiced

over

unpracticed

consonants

(disregarding

the

Non-VMS

consonants,

since

they

are

not

signiﬁcantly

affected

by

production)

is

to

make

use

of

the

preference

ratios

between

looking

times

(LT)

for

well-practiced

over

unpracticed

consonants

(preference

ratio

=

LT(Own)/[LT(Own)

+

LT(Other)]).

With

this

metric

a

value

of

.5

would

indicate

equal

looking

time

to

each

passage.

The

preference

ratios

are

plotted

against

each

infant’s

practiced

consonants

in

Fig.

1.

The

preference

ratio

for

Own-VMS

passages

for

the

single-VMS

infants

(M

=

.55,

SD

=

.086)

was

signiﬁcantly

larger

(t[16]

=

2.817,

p

=

.012)

than

for

multiple-VMS

infants

(M

=

.45,

SD

=

.063).

Six

of

the

nine

single-VMS

infants

show

a

preference

for

Own-VMS

passages,

while

eight

of

nine

multiple-VMS

infants

exhibit

a

preference

for

Other-VMS

passages.

The

categorical

nature

of

VMS

allows

the

separation

of

infants

into

two

groups

that

differ

signiﬁcantly

in

the

number

of

consonants

produced

and

in

the

frequency

and

consistency

of

use

of

these

consonants

over

time.

In

order

to

explore

the

relationship

of

the

infants’

practiced

consonants

and

the

HPP

results

independent

of

the

development

of

production

patterns

over

time,

we

ran

a

simple

linear

regression

using

consonant

counts

in

the

session

preceding

the

HPP

test.

Since

the

design

of

the

HPP

test

involved

identifying

each

infant’s

practiced

consonants

we

used

the

number

of

each

infant’s

Own-VMS

consonant

plus

the

second

most

practiced

consonant

(for

the

2

VMS

infants

this

was

the

second

VMS)

as

the

predictor

and

the

preference

ratio

as

the

dependent

variable

(see

Table

4).

There

was

a

signiﬁcant

linear

relationship

between

the

total

production

frequency

of

the

two

most

practiced

consonants

and

the

infants’

preference

for

Own-

or

Other-VMS

(R

=

.605,

p

<

.01).

Table

4

Summary

of

simple

regression

with

the

total

number

of

the

ﬁrst

two

favored

consonants

produced

predicting

the

results

on

the

HPP

test.

Variable

B

SE

B

ˇ

Constant

.583

.032

Number

of

ﬁrst

two

favored

consonants

produced −.001

.000

−.605

**

R

2

=

.366.

**

p

<

.01.

Author's personal copy

R.A.

DePaolis

et

al.

/

Infant

Behavior

&

Development

34 (2011) 590–

601 597

Fig.

3.

Infant-directed

speech

(IDS)

for

three

infants’

mothers.

The

solid

line

represents

the

pre-VMS

session

while

the

dashed

line

represents

the

session

in

which

two

VMS

were

credited

to

the

infant.

The

asterisks

indicate

the

VMS

for

the

child

whose

mother’s

IDS

is

plotted

in

the

ﬁgure.

Part

(a)

is

a

plot

for

an

infant

with

/d,t/

and

/n/,

part

(b)

for

an

infant

with

/b,p/

and

/m/,

and

part

(c)

for

an

infant

with

/g,k/

and

/d,t/.

3.3.

Analysis

of

the

infant

directed

speech

As

mentioned

above,

what

leads

an

infant

to

settle

upon

a

practiced

or

‘favorite’

consonant

or

VMS

is

unclear.

It

is

possible

that

the

speech

directed

to

the

infant

has

an

effect

on

the

consonants

produced

and

thus

plays

a

role

in

the

perceptual

salience

of

those

consonants.

To

investigate

this

possibility

we

transcribed

the

infant-directed

speech

(IDS)

of

three

mothers.

We

chose

mothers

of

infants

who

had

multiple

VMS

since

in

this

group

at

least

one

infant

had

acquired

each

of

the

stop

consonants

(alveolar,

velar,

and

bilabial).

We

transcribed

both

the

session

in

which

the

infant

was

credited

with

two

VMS

and

the

session

preceding

it.

If

there

was

a

problem

identifying

the

consonant,

the

main

transcriber

consulted

with

a

second

transcriber

to

arrive

at

a

consensual

decision.

To

assess

reliability

two

independent

researchers

(not

including

the

second

transcriber

noted

above)

transcribed

two

of

the

30-min

sessions,

yielding

agreement

on

90%

of

the

VMS

consonants.

Agreement

rose

to

99%

when

ambiguous

consonants

were

retranscribed

by

consensus.

Fig.

3

is

a

plot

of

the

proportion

of

each

VMS

consonant

in

the

IDS

directed

to

the

infant.

For

ease

of

comparison,

an

asterisk

indicates

the

VMS

for

each

infant.

It

is

clear

from

this

ﬁgure

that

although

VMS

consonant

frequency

from

each

mother’s

IDS

was

similar,

the

VMS

consonants

developed

by

the

infants

are

quite

diverse.

In

particular,

the

infant

in

panel

b

produced

the

two

consonants

with

the

least

frequency

in

the

input,

while

the

infant

in

panel

c

produced

the

two

consonants

that

were

most

frequent

in

the

IDS.

A

Chi-square

test

comparing

the

relative

frequency

in

the

infants’

production

of

the

ﬁve

VMS

consonants

(stops:

alveolar,

velar,

and

bilabial,

nasal:

bilabial

and

alveolar)

to

the

relative

frequency

in

the

mothers’

IDS

revealed

a

signiﬁcant

difference

for

each

mother–infant

pair

(

2

(4,

n

=

186,

82,

41)

=

86.0,

483.7,

84.2,

p

<

.005,

for

the

three

infants

in

Fig.

2a,

b,

and

c,

respectively).

Author's personal copy

598 R.A.

DePaolis

et

al.

/

Infant

Behavior

&

Development

34 (2011) 590–

601

4.

Discussion

The

proﬁle

of

consonant

use

for

the

infants

who

completed

the

study

was

consistent

with

that

found

by

McCune

and

Vihman

(2001).

The

most

common

initial

VMS

was

identiﬁed

as

/t,d/

in

both

studies,

although

the

18

British-English

infants

were

even

more

likely

than

the

20

American-English

infants

to

start

with

this

VMS

(88%

vs.

72%).

After

/t,d/,

the

remaining

stops

and

nasals

were

the

most

likely

VMS

in

both

studies.

Of

the

stops,

/k,g/

were

the

most

likely

to

be

acquired

last.

These

ﬁndings

are

consistent

both

with

predictions

of

early

consonant

development

based

on

diary

reports

and

markedness

theory

(Jakobson,

1941/68)

and

with

later

accounts

of

perceptual

salience

and/or

bio-mechanical

constraints

(Lindblom,

1992;

Davis

&

MacNeilage,

1995);

they

also

agree

with

empirical

ﬁndings

reported

for

a

wide

range

of

languages

(Locke,

1983).

The

results

of

the

headturn

experiments

show

that

the

infants’

looking

times

were

inﬂuenced

both

by

their

own

consonant

production

experience

(as

indexed

by

the

number

of

VMS

that

they

had

acquired)

and

by

whether

or

not

they

produced

the

particular

consonant

featured

in

the

nonwords

embedded

in

the

passages.

This

ﬁnding

conﬁrms

our

hypothesis

that

looking

times

would

be

inﬂuenced

by

the

infants’

production

patterns

and

accordingly

supports

the

postulation

of

an

articulatory

ﬁlter.

The

lack

of

any

relationship

between

three

individual

infant’s

production

patterns

and

the

IDS

they

were

exposed

to

strongly

suggests

that

this

ﬁnding

is

not

due

to

the

frequency

of

consonants

used

in

the

input.

The

difference

in

looking

times

between

the

single-

and

multiple-VMS

infants

was

unexpected.

The

multiple-VMS

infants

displayed

longer

looking

times

in

response

to

the

passage

that

featured

the

consonant

that

they

were

not

yet

consistently

producing,

while

infants

with

a

single

VMS

showed

no

signiﬁcant

preference

for

either

passage.

The

difference

in

preference

between

the

groups

was

revealed

in

the

signiﬁcant

interaction

between

Own-

and

Other-VMS

and

VMS

number

as

well

as

in

a

t-test

comparing

the

preference

ratio

between

the

single-

and

the

multiple-VMS

groups

(using

the

preference

ratio

effectively

reduced

the

‘noise’

in

the

paradigm).

The

same

type

of

relationship

was

also

evident

when

production

experience

was

measured

continuously,

in

the

correlation

and

regression

that

showed

that

the

amount

of

production

experience

an

infant

has

with

one

or

two

consonants

can

predict

their

pattern

of

preference

for

the

test

passages.

These

results

suggest

that

production

advances

may

incrementally

alter

the

manner

in

which

infants

attend

to

aspects

of

the

speech

signal.

It

is

worth

emphasizing

that

the

presence

of

a

novelty

effect

in

the

multiple

VMS

group

can

occur

only

if

the

Other

passage

stands

out

as

different

from

the

Own

passage

(novel).

In

this

case

it

is

likely

that

the

Own

passage

introduces

overly

familiar

sounds,

making

the

Other

passage,

with

its

novel

sounds,

more

interesting.

This

is

consistent

with

Hunter

and

Ames

(1988),

who

showed

that

greater

attention

to

a

familiar

stimulus

will

eventually

be

replaced

by

greater

attention

to

the

novel

stimulus.

3

It

is

likely

that

the

infants

who

are

more

practiced

at

consonant

production

(as

indexed

by

VMS

in

this

study)

will

be

the

ones

tending

to

exhibit

a

novelty

(versus

familiarity)

effect

to

a

comparison

of

produced

versus

unproduced

consonants.

The

discovery

of

this

novelty

effect

was

serendipitous,

as

had

the

study

been

carried

out

with

single-VMS

infants

only,

as

originally

planned,

we

would

have

been

unaware

that

the

transition

from

one

to

two

VMS

indexes

a

change

in

looking

times.

This

raises

the

question,

why

should

the

move

from

producing

one

to

two

stable

consonants

change

the

infants’

response

to

the

Other

passages?

One

possibility

is

that

acquiring

a

second

stable

consonant

(or

VMS)

indexes

cognitive

advance,

as

suggested

by

McCune

and

Vihman,

2001

(see

introduction

for

a

more

detailed

description).

In

addition,

DePaolis,

Keren-

Portnoy,

and

Vihman

(2007)

found

that

10-month-old

infants

who

had

acquired

two

VMS

were

signiﬁcantly

more

variable

in

their

looking

times

to

familiar

words

on

a

headturn

task

than

infants

who

had

not

acquired

two

VMS.

Both

of

these

results

support

the

notion

that

reliably

producing

two

stable

consonants

affects

the

way

that

infants

process

consonants.

Another

interpretation

is

that

the

change

in

looking

times

is

not

necessarily

dependent

upon

the

infants

producing

two

consonants

to

VMS

criterion,

but

simply

reﬂects

the

fact

that

they

are

producing

large

numbers

of

consonants.

For

example,

from

Fig.

1

it

is

clear

that

the

infants

with

the

largest

consonant

counts

are

the

ones

most

likely

to

show

a

preference

for

the

‘Other’

passages.

This

interpretation

is

also

supported

by

the

simple

regression,

in

which

the

count

of

the

two

highest

consonants

produced

is

a

signiﬁcant

predictor

of

the

preference

ratio.

Although

this

interpretation

is

plausible,

since

8

of

the

11

infants

who

show

a

preference

for

the

‘Other’

passage

are

producing

two

consonants

to

VMS

criterion,

it

is

likely

that

the

two

ways

of

operationalising

practice

in

consonant

production

are

complementary:

Infants

who

produce

many

consonants

seem

to

be

producing

many

tokens

of

perhaps

only

a

few

consonant

types.

Therefore,

it

is

hard

to

disentangle

the

role

of

the

speciﬁc

consonants

produced

from

that

of

sheer

amounts

of

consonant

production.

It

is

probable

that

these

things

combine

to

inﬂuence

infants’

preference

for

what

is

familiar

vs.

what

is

novel.

We

included

the

Non-VMS

condition

to

test

the

contrasting

hypothesis

that

production

has

no

effect

on

looking

times.

Looking

times

to

the

Non-VMS

passage

would

have

been

revealing

only

if

no

difference

had

been

found

between

the

Own-

and

the

Other-VMS

passages.

However,

this

study

did

reveal

a

signiﬁcant

difference

between

looking

times

to

Own-VMS

and

Other-VMS.

Given

that

ﬁnding,

along

with

the

fact

that

both

groups

looked

equally

to

the

Non-VMS

consonant

passages

(see

Fig.

2)

and

showed

no

signiﬁcant

difference

in

looking

times

to

Non-VMS

vs.

the

other

types

of

passages,

the

results

from

the

Non-VMS

passage

can

be

taken

to

simply

reﬂect

the

spectral

salience

of

fricatives

versus

stops.

3

Interestingly,

a

change

from

engagement

with

familiar

to

engagement

with

unfamiliar

stimuli

can

sometimes

be

seen

to

occur

even

for

individual

infants

across

experiments

(Roder,

Bushnell,

&

Sasseville,

2000)

as

well

as

in

the

course

of

a

single

experiment

(Vihman

et

al.,

2004).

Author's personal copy

R.A.

DePaolis

et

al.

/

Infant

Behavior

&

Development

34 (2011) 590–

601 599

Two

methodological

concerns

arose

in

the

course

of

the

study.

The

ﬁrst

was

the

balance

of

consonants

used

in

the

headturn

test.

Since

each

infant’s

production

pattern

necessarily

dictated

the

consonants

to

be

presented,

the

overall

proﬁle

of

consonants

could

not

be

planned

in

advance.

The

most

common

Own-VMS

consonant

was

/t,d/;

it

was

presented

as

Own

consonant

in

13

of

the

18

infant

headturn

tests

(for

eight

of

the

single-

and

ﬁve

of

the

multiple-VMS

infants),

which

raises

the

possibility

that

infants’

looking

patterns

could

be

due

to

their

having

a

preference

or

dispreference

for

the

speciﬁc

consonant

/t,d/

rather

than

to

their

individual

production

history.

However,

of

these

13

infants

for

whom

/t,d/

was

Own-VMS,

eight

preferred

the

/t,d/

passage

and

ﬁve

did

not,

and

this

difference,

using

a

binomial

test,

is

not

signiﬁcant.

This

effectively

rules

out

the

possibility

that

a

general

preference

(or

dispreference)

for

/t,d/

among

those

infants

who

produce

that

segment

might

explain

our

ﬁndings.

In

addition,

considering

the

14,

9,

and

13

times

that

passages

featuring

/t,d/,

/p,b/,

and

/k,g/

were

presented

in

all

18

of

the

headturn

tests

as

either

Own-

or

Other-VMS

(see

Table

3),

the

preference

that

the

infants

showed

for

each

of

these

passages

was

no

greater

than

chance

(deﬁned

as

a

preference

for

that

passage

half

of

the

time

it

occurred,



2

(2,

n

=

36)

=

.97313,

p

>

.05),

suggesting

that

properties

of

the

passages

are

insufﬁcient

in

themselves

to

explain

the

pattern

of

preferential

looking

times.

The

single-VMS

group

could

also

conceivably

have

a

different

pattern

of

preference

or

dispreference

for

alveolars

from

the

multiple-VMS

group,

and

these

group-dependent

preferences

could

be

the

primary

factor

responsible

for

the

pattern

of

results

that

we

found.

For

example,

the

prevalence

of

alveolars

in

the

input,

or

some

signal-based

aspect

of

alveolars

(the

high

frequency

emphasis

of

the

burst,

for

example),

could

be

driving

the

results,

with

different

effects

in

the

two

groups.

However,

the

data

fail

to

support

the

idea

that

alveolars

were

systematically

either

preferred

or

dispreferred

in

either

group

(see

Table

3

for

the

preference

for

Own-

versus

Other-VMS).

In

the

single-VMS

group,

ﬁve

infants

showed

a

preference

for

alveolars

over

velars

(3)

or

bilabials

(2).

On

the

other

hand,

three

showed

a

preference

for

the

velar

that

was

contrasted

with

the

alveolar

stop.

In

the

multiple-VMS

group

dispreference

for

Own-VMS

is

limited

to

no

one

place

of

articulation:

We

see

a

dispreference

for

alveolars

in

ﬁve

infants

(of

whom

three

showed

a

preference

for

velars

and

two

for

bilabials),

for

bilabials

in

two

infants

(both

showing

a

preference

for

velars

instead),

and

for

velars

(preferring

bilabials)

in

one

infant.

The

pattern

of

preference

appears

to

be

random

and

convincingly

rules

out

input

frequency

or

signal-based

attributes

of

alveolars

as

a

possible

explanation.

The

second

concern

was

the

age

of

the

participants

at

the

time

of

the

HPP

test.

Three

infants

in

the

single-VMS

group

were

considerably

older

than

the

rest

of

the

infants

in

the

study.

The

age

discrepancy

was

an

unintended

consequence

of

our

testing

infants

as

soon

as

we

could

identify

either

one

or

more

VMS.

The

inclusion

of

these

infants

is

developmentally

sound

since

longitudinal

samples

of

infant

consonant

production

typically

include

small

numbers

of

later

developing

infants

(for

example,

see

McCune

&

Vihman,

2001).

It

could

be

argued

that

these

infants

have

considerably

more

experience

in

processing

speech,

so

their

results

deserve

closer

examination.

Two

of

these

three

infants

followed

the

trend

of

preferring

their

own

production

patterns

while

one

did

not

(preference

ratios

of

.54,

.63

and

.44).

Removing

these

three

infants

from

the

ANOVA

reduces

the

power

from

.709

to

.563

and

changes

the

interaction

from

signiﬁcant

(p

=

.024)

to

nearly

signiﬁcant

(p

=

.056).

The

signiﬁcance

of

the

t-test

comparing

the

single-VMS

to

the

multiple-VMS

group

on

their

preference

ratio

for

Own-

versus

Other-VMS

does

not

change

when

these

three

infants

are

removed

(p

=

.012

for

all

infants

and

p

=

.018

with

three

infants

removed).

Thus,

the

behavior

of

these

infants

does

not

change

the

basic

pattern

of

results.

The

fact

that

this

interaction

of

production

and

perception

is

signiﬁcant

with

18

infants

suggests

that

it

is

highly

robust.

A

preference

for

own

sounds

was

found

in

six

out

of

nine

infants

in

the

single-VMS

group,

and

a

preference

for

others’

VMS

was

found

in

eight

of

nine

infants

in

the

multiple-VMS

group.

Note

that,

of

the

four

infants

who

did

not

show

this

pattern,

three

were

in

the

single-VMS

group.

Due

to

the

fact

that

our

samples

were

limited

to

a

30-min

recording

once

a

week

at

most,

it

is

likely

that

some

of

these

infants

were

producing

additional

consonants

that

failed

to

be

recorded.

To

guard

against

this

problem

we

provided

caregivers

with

a

detailed

questionnaire

to

be

completed

after

each

session,

but

it

was

apparent

that

parents

often

misidentiﬁed

consonants.

Thus,

while

the

multiple-VMS

infants

can

be

conﬁdently

said

to

have

had

at

least

two

VMS,

it

is

possible

that

some

infants

identiﬁed

as

having

only

a

single

VMS

were

actually

producing

a

second

VMS,

or

were

in

transition.

Our

results

suggest

the

possibility

that

production

initiates

shifts

in

the

way

that

the

infant

processes

input

speech.

Since

the

transition

from

one

to

two

VMS

typically

occurs

in

a

matter

of

weeks

at

most,

the

infants’

preferential

shift

to

a

passage

featuring

a

consonant

that

they

are

about

to

produce

is

noteworthy

and

suggests

that

the

interplay

of

familiarity

and

novelty

in

early

babble

provides

the

infant

with

attentional

pointers

toward

phonetic

advance.

It

is

worth

noting

that

this

shift,

as

indexed

by

production

operationalized

as

the

transition

from

one

to

two

VMS,

could

be

a

by-product

of

an

underlying

cause

not

identiﬁed

in

this

study.

For

example,

it

is

possible

that

the

perceptual

salience

of

stops

in

speech

directed

to

the

infant

is

the

underlying

cause

of

the

results.

According

to

this

rationale,

the

salience

of

a

stop

is

both

the

reason

for

an

infant

developing

that

stop

as

a

VMS

and

the

cause

of

the

difference

in

preferential

looking

times.

This

explanation

is

consistent

with

the

data

since

the

single

VMS

infants

looked

longer

at

their

Own-VMS

(although

the

difference

was

not

signiﬁcant).

In

addition,

even

though

the

multiple

VMS

infants

did

show

a

preference

for

the

Other-VMS,

this

could

be

due

to

their

decreased

interest

in

the

stop

consonant

which

had

commanded

their

attention

earlier,

leading

to

an

increased

salience

of

the

stop

consonant

not

yet

being

produced.

However,

the

comparison

of

the

frequency

of

consonants

in

individual

infants’

input

to

their

own

produced

consonants

does

not

favor

such

an

interpretation.

There

was

no

clear

relationship

between

input

frequency

and

choice

of

VMS

for

production.

In

addition,

the

similarity

among

mothers

in

the

frequency

of

consonants

in

their

speech

makes

this

an

unlikely

source

for

the

differences

among

infants

in

the

identity

of

the

ﬁrst

consonants

used

stably

in

babble.

Indeed,

the

lack

of

an

Author's personal copy

600 R.A.

DePaolis

et

al.

/

Infant

Behavior

&

Development

34 (2011) 590–

601

effect

of

the

mothers’

speech

on

the

phonetic

output

of

the

infant

has

been

reported

previously

in

a

larger

sample

of

French,

Swedish

and

American

infant-mother

dyads

(Vihman,

Kay,

de

Boysson-Bardies,

Durand,

&

Sundberg,

1994).

The

data

in

this

experiment

do

not

provide

a

deﬁnitive

answer

to

the

‘chicken

and

egg’

dilemma

of

how

a

favorite

consonant

emerges,

but

they

do

suggest

that

a

parsimonious

explanation

of

the

results

is

that

the

pattern

of

results

reﬂects

a

bi-directional

inﬂuence

of

production

on

perception.

Since

infants

appear

to

‘notice’

often

produced

phonetic

segments

in

continuous

speech,

these

results

also

suggest

a

mechanism

for

facilitating

the

production

of

ﬁrst

words.

Over

35

years

ago

Ferguson

and

Farwell

(1975,

p.

433f.)

noted

that

‘at

an

early

stage

in

which

a

contrast

is

absent.

.

.the

adult

words

chosen

by

the

child

will

be

highly

discriminatory’

–

i.e.,

‘selected’

for

their

sound

patterns.

Following

up

on

this

insight,

Vihman

(1993)

suggested

that

rather

than

avoiding

words

or

sounds

that

they

cannot

produce

(as

suggested

in

Menn,

1983,

for

example),

infants

‘select’

their

ﬁrst

words

based

on

implicit

matching

of

their

existing

babble

to

words

consisting

largely

of

sounds

that

they

can

produce

(cf.

also

Vihman

&

Croft,

2007).

If

infants

could

count

on

hearing

words

in

isolation,

this

fact

would

be

unremarkable,

but

at

least

one

study

has

found

that

words

are

not

consistently

produced

in

isolation

(Aslin,

Woodward,

LaMendola,

&

Bever,

1996

–

but

see

Brent

&

Siskind,

2001).

In

the

absence

of

exposure

to

isolated

words,

how

does

an

infant

extract

the

word

from

continuous

speech

in

order

to

pattern

the

output

after

the

input?

Based

upon

the

results

of

this

study,

infants

should

be

predisposed

to

notice

in

the

ongoing

speech

stream

words

that

contain

sounds

that

they

are

just

starting

to

produce.

This

is

consistent

with

the

notion

of

an

articulatory

ﬁlter

(Vihman,

1993,

1996).

It

could

be

that,

in

typically

developing

children,

the

ability

to

effectively

process

continuous

speech

in

the

prelinguistic

period

is

augmented

by

the

formation

of

a

consistently

reproducible

phonetic

pattern

in

babble.

This

is

the

ﬁrst

study

to

report

evidence

for

a

link

between

what

an

infant

produces

in

babble

and

how

that

infant

processes

speech

(although

see

Vihman

&

Nakai,

2003).

The

impact

of

this

link

beyond

the

preference

or

dispreference

for

favorite

consonants

is

an

open

question

and

should

be

explored

further.

For

example,

future

work

could

examine

how

the

relationship

between

production

and

perception

affects

the

ability

to

attach

meaning

to

word

forms

that

either

do

or

do

not

contain

preferred

production

patterns

(see

Keren-Portnoy

et

al.,

2010,

for

a

study

investigating

phonological

memory

and

preferred

production

patterns).

Models

could

also

incorporate

proprioceptive

feedback

into

the

developmental

trajectory

of

consonant

and

vowel

categorization

(for

examples

see

models

by

Kent,

1981,

and

Westermann

&

Miranda,

2004).

Most

importantly,

the

ﬁndings

of

this

study

suggest

the

possibility

that

babble

channels

infants’

sensitivity

to

phonetic

aspects

of

the

speech

stream

that

are

important

for

early

language

acquisition.

Acknowledgements

We

thank

Nicola

Williams

for

help

with

transcription

of

the

infants,

Naomi

Craig

for

running

the

headturn

experiments,

and

both

Sarah

McKain

and

Caitlin

Baird

for

transcription

of

the

IDS.

This

work

was

supported

by

a

Marie-Curie

Incoming

International

Fellowship

from

the

European

Commission

and

an

Educational

Leave

from

the

College

of

Integrated

Science

and

Technology

at

James

Madison

University.

References

Aslin,

R.

N.,

Woodward,

J.

Z.,

LaMendola,

N.

P.,

&

Bever,

T.

G.

(1996).

Models

of

word

segmentation

in

ﬂuent

maternal

speech

to

infants.

In

J.

L.

Morgan,

&

K.

Demuth

(Eds.),

Signal

to

syntax:

Bootsrapping

from

speech

to

grammar

in

early

acquisition.

Mahwah,

NJ:

Lawrence

Erlbaum

Associates.

Boysson-Bardies,

B.

de,

Halle,

P.,

Sagart,

L.,

&

Durand,

C.

(1989).

A

crosslinguistic

investigation

of

vowel

formants

in

babbling.

Journal

of

Child

Language,

16,

1–18.

Boysson-Bardies,

B.

de,

&

Vihman,

M.

(1991).

Adaptation

to

language:

Evidence

from

babbling

and

ﬁrst

words

in

four

languages.

Language,

67,

297–319.

Brent,

M.

R.,

&

Siskind,

J.

M.

(2001).

The

role

of

exposure

to

isolated

words

in

early

vocabulary

development.

Cognition,

81,

B33–B44.

Campos,

J.

J.,

Anderson,

D.

I.,

Barbu-Roth,

M.

A.,

Hubbard,

E.

M.,

Hertenstein,

M.

J.,

&

Witherington,

D.

(2000).

Travel

broadens

the

mind.

Infancy,

1,

149–219.

Davis,

B.

L.,

&

MacNeilage,

P.

F.

(1995).

The

articulatory

basis

of

babbling.

Journal

of

Speech

and

Hearing

Research,

38,

1199–1211.

DePaolis,

R.

A.,

Keren-Portnoy,

T.,

&

Vihman,

M.

(2007).

Dynamic

interactions

between

perception

and

production.

Talk

presented

at

the

Biennial

Meeting

of

the

Society

for

Research

in

Child

Development,

Boston.

Elbers,

L.

(1997).

Output

as

input:

A

constructivist

hypothesis

in

language

acquisition.