Predicting Reading Disability from Eye Movements

beadkennelAI and Robotics

Oct 15, 2013 (3 years and 5 months ago)

73 views

Predicting Reading Disability

from Eye Movements

The earlie
r
reading

disability
is detected
in school
children,
the better
the effect of professional
intervention.
We propose machine learning of eye movements during reading as an objective,
efficient and
accur
ate method for detecting children with reading problems
, relative to expectations
of what is typical for the age and grade level.

Based on
eye
-
movement recordings of 103
third
-
grade
children

with reading disabilities

(RD)
and 90 chil
d
ren

with normal r
eading capacity
(NR),

we

train
classification models that learn to

predict the status (RD or NR)
of
an
y
third
-
grade child
, given
his or
her
eye
tracking record.
We evaluate a number of
inference
models and
show that a support vector
machine

predict
s
the
status

of
a child

outside the training set
with an accuracy of 94%.

We also
investigate how classification accuracy depends on recording time and show that

nine

out of
ten
children
are correctly classified
after only
30
seconds
of
t
racking

time
.

Although
e
ye movements that

deviate from normal

are
symptom
atic

of reading problems

rather than
causing them
,

our results
indicate

that

eye mo
vement
s can be
highly useful for
early
identification of
language
-
related
deficiencies
.