PresentationPDF Available

The importance of nonlinguistic variability to early language learning: the case of colour

Authors:

Abstract and Figures

Variability is prevalent in early language acquisition, however, whether it supports or hinders learning is unclear: while target variability has been shown to support word learning, variability in competitor items has been shown to make the task harder. Here we tested whether background variability could boost learning in a referent selection task. Two groups of two-year-old children saw arrays of one novel and two known objects on a screen, and heard a novel or known label. Stimuli were identical across conditions, with the exception that in the constant color condition objects appeared on a uniform white background, and in the variable color condition backgrounds were different, uniform colors. At test, only children in the variable condition showed evidence of retaining label-object associations. These data support findings from the adult memory literature, which suggest that variability supports learning by decontextualizing representations. We argue that these data are consistent with dynamic systems accounts of learning in which low-level entropy adds sufficient noise to the developmental system to precipitate a change in behavior.
No caption available
… 
No caption available
… 
Content may be subject to copyright.
The importance of nonlinguistic
variability to early language learning:
the case of colour
Katie Twomey, Shirly Ma & Gert Westermann
k.twomey@lancaster.ac.uk
malizhi110@hotmail.com
g.westermann@lancaster.ac.uk
Learning words
isn’t easy
chimp!
dog
milk
foot
apple
blanket
shoe
nappy
baby
bump
arm
cup
share
hat
bye bye
tree
shoes
blue
sing
bed
park
her
why
hot
cow
bus
naughty
friend
by 36mo, they say around
about 600!
(Fenson et al. 1994)
By 6mo, infants have
begun to understand
words for the things they
see
(Bergelson & Swingley, 2012)
Nonetheless, children learn words quickly and
apparently easily
How do
they do
it?
that’s
“cow”…
…and that’s
“car”…
…so that must be
“blicket”
blicket!
From around 18mo, children rapidly map new words to
objects in (fairly) predictable ways
…attributed to innate constraints
(Markman, 1994), learned principles (Golinkoff et al.,
1994) or associative learning over time
(McMurray et al., 2012; Twomey et al. 2016)
blicket!
Word learning is a fragile,
incremental process (Smith & Yu, 2008;
Cognition) affected by task
characteristics such as repetition
(Mather & Plunkett, 2009; Infancy), pragmatics
(Grassman, Stracke & Tomasello, 2009; Cognition),
referent motion (Matatyaho-Bullaro,
Gogate,Mason, Cadavid, & Abdel-Mottaleb, 2014;
JECP) to name a few..
…?
But “fast mapping” doesn’t necessarily
lead to word learning (Horst & Samuelson, 2008; Infancy)!
Variability is important in language
acquisition, but evidence is mixed
Linguistic variability helps
(Rost & McMurray, 2009; Dev. Sci)
Nonlinguistic target variability helps
(Twomey, Ranson & Horst, 2014; Inf. Chi. Devt)
14mos learn novel word-
object associations only
when word was spoken by
multiple speakers
30mos learn words
when only when target
object is variable
blicket!blicket!
Competitor variability hinders
(Horst, Scott & Pollard, 2010; Dev. Sci; Twomey et al., 2016; Interaction Studies)
Context variability in reading hinders
(Horst, Parsons & Bryan, 2011; Front. Pscyh.)
43mos learn words
when read the same
book multiple times
24mos learn words only
when context is minimally
variable
Adding entropy to a task speeds up learning in adults
(Stephen, Dixon & Eisenhower, 2009; JEP:HPP)
What about low-level, background variability?
Context variability
(story, competitor
objects) hinders
Target variability
(words, items) helps
How does variability affect language learning?
Overall, a mixed picture
Slower learning Faster learning
On dynamic systems
approaches, behaviour
emerges in-the-
moment…
Learner (e.g., child)
Learning history (e.g., known
vocabulary, experience with objects)
Environment (e.g., visual scene)
…from a
system of multiple
interacting components
(Thelen & Smith, 1996)
Learning” is the transition from one
stable behavioural state to the next (e.g.,
being consistently able to correctly
generalize a word)
Formally, adding low-level entropy or
background noise speeds up this transition
(Stephen et al. 1999; JEP:HPP)
A “behaviour” is when these components
interact in a stable, predictable way (e.g.,
pointing to the correct referent)
Thus, if behaviour is a property
of a dynamic system,
then language
– a human behaviour –
must also be part of a
dynamic system;
so, what affects dynamic
systems in general should
also affect language
in particular
Adding background variability to a language learning task
should support learning
doff!
giraffe!
fode!
General methods
Referent selection: teach children novel words by presenting
an array of objects, all-but-one known
Five-minute break
doff!
Retention:
present just-seen objects and ask for each in turn
wheres the
cup?
wheres the
tife?
Teach toddlers 3 novel words across 15 trials (3 novel 2
known per block)
2 conditions:
half see white backgrounds (constant colour),
half see multiple different coloured backgrounds (variable
colour)
Everything else kept identical between conditions
wheres the
tife?
Run 30 x 23mo toddlers in a computerised word
learning task and record looking times using an
eyetracker
Procedure and design
Test / retention
wheres the
tife?
Six test trials, two per object
Test trials identical between conditions
Test trials always presented on grey background
Results, referent selection
Constant
Variable
Overall increase in
looking over time
(beta = 0.0022, p < .
001)
But faster on
known than novel
trials for children in
the variable
condition
(known: beta = 0.0022;
novel: beta = 0.0019;
ps < .001)
0.00
0.25
0.50
0.75
1.00
0123456
Seconds after label offset
Proportion target looking
Trial type
known novel
0.2
0.4
0.6
0 1 2 3 4 5 6
Seconds after label offset
Proportion target looking
Trial type
known novel
No consistent above-
chance target looking
0.00
0.25
0.50
0.75
1.00
0123456
Seconds after label offset
Proportion target looking
Trial type
known novel
0.00
0.25
0.50
0.75
1.00
0123456
Seconds after label offset
Proportion target looking
Trial type
known novel
0.00
0.25
0.50
0.75
1.00
0123456
Seconds after label offset
Proportion target looking
Trial type
known novel
Results, retention
Block 1
Increase in looking
over time (beta =
0.0019, p = .0012)
Block 2
Decrease in looking over
time (beta = 0.0029, p < .001)
More target looking in
variable condition (beta =
0.25, p = .0032)
0.00
0.25
0.50
0.75
1.00
0123456
Seconds after label offset
Proportion target looking
Trial type
known novel
0.00
0.25
0.50
0.75
1.00
0123456
Seconds after label offset
Proportion target looking
Trial type
known novel
Overall our hypothesis is supported: only children
who saw variable backgrounds learned novel nouns
Raises two puzzling
questions…
Theoretically, adding entropy to the
word learning system sped up a change
in behavioural state
But, only children in constant condition
showed evidence of identifying targets
during referent selection
Increasing empirical and computational evidence that
children disambiguate in word learning tasks (Bion et al., 2013;
Cognition; McMurray, Horst & Samuelson, 2012; Psych. Rev; Twomey et al., 2016; Interaction Studies)
that’s a
cow…
…and that’s a
car…
…so that must
be the
blicket
Children in the variable don’t look at the novel target at
above-chance levels during referent selection, but they do
show retention. How?
Thus, children look at competitors objects,
working out what’s not the blicket
Don’t need to learn everything
about the target to learn the
association between a target
and a word: incremental
word learning
(Yurovsky, Fricker, Yu & Smith, 2014)
Implications for our interpretation of
looking time and word learning tasks:
learning is not necessarily contingent on
robust responding during training
0
5
10
15
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Trial number
Mean switches
Condition
CC
VC
Why did only children in the variable condition learn?
contextualised
contextualised
contextualised
contextualised
contextualised
Adult learning literature:
representations are either
context-dependent (Godden &
Baddeley, 1975; BJPsych)
Recall in same context is easier for
for context-dependent
representations, but generalisation
to new contexts is easier for
decontextualised representations
or decontextualised when
leaning is repeated across
contexts decontextualised
decontextualised
decontextualised
decontextualised
decontextualised
Why did only children in the variable condition learn?
Adult learning literature:
representations are either
context-dependent (Godden &
Baddeley, 1975; BJPsych)
variable condition =
decontextualised
constant condition =
contextualised
All children had to generalise
to a context at test
Easier for children in the
variable condition to
generalize
Importantly: lack of a
behaviour != lack of learning
How do results from looking time
studies relate to behavioural
results?
Overall, background variability supports word learning!
But is this effect restricted to the visual modality?
If this is a dynamic systems phenomenon, could entropy in
other modalities have a similar effect?
Spatial location?
Sound?
Buzzzzzz!
Hummmm!
Language (acquisition) isn’t about words in isolation,
or language-specific cognitive structure….
Language is fundamentally part of a system
I said a hip hop
hippy to the hippy
to the hip hip hop
Language is fundamentally part of a system
To understand language (acquisition), we must
consider all components of that system
colourless green
ideas sleep
furiously
Huge thanks to all the
parents and toddlers who
took part, and to our funders
Twomey, Ma & Westermann (under revision).
All the right noises: background variability
facilitates early noun learning.
Cognitive Science.
This work was supported by the ESRC
International Centre for Language and
Communicative Development (LuCiD; ES/
L008955/1), a British Academy/Leverhulme
Trust Senior Research Fellowship to GW
(SF150163) and an ESRC Future Research
Leaders Fellowship to KT (ES/N01703X/1)
Attention
getter 3 n/a n/a
RS block 2
trial 3 zorch
RS block 2
trial 4 fork
Attention
getter 4 n/a n/a
RS block 2
trial 5 zorch
RS block 3
trial 1 car
Attention
getter 5 n/a n/a
RS block 3
trial 2 blick
RS block 3
trial 3 ball
RS block 3
trial 4 blick
Attention
getter 6 n/a n/a
RS block 3
trial 5 blick
Well done! n/a n/a
Five-minute break
Trial Stimuli Color Target
Engagement
1 n/a n/a
Warm-up 1 ball
Warm-up 2 fork
Warm-up 3 cup
RS block 1
trial 1 banana
Attention
getter 1 n/a n/a
RS block 1
trial 2 tife
RS block 1
trial 3 cup
RS block 1
trial 4 tife
Attention
getter 2 n/a n/a
RS block 1
trial 5 tife
RS block 2
trial 1 apple
RS block 2
trial 2 zorch
Trial Stimuli Color Target
Engagement
2 n/a n/a
Warm-up car
Ret trial 1 tife
Ret trial 2 zorch
Ret trial 3 blick
Ret trial 4 zorch
Ret trial 5 blick
Ret trial 6 tife
Well done! n/a n/a
ResearchGate has not been able to resolve any citations for this publication.
ResearchGate has not been able to resolve any references for this publication.