Early Reading Fluency:
Rationale and Insights from
cognitive neuroscience
The view from the
Independent Evaluation Group
Helen Abadzi
Independent Evaluation Group
World Bank
March 2008
Habadzi@worldbank.org
Independent Evaluation Group (IEG)
Project Performance Assessment Reviews
in primary education









Comoros 1997
Burkina Faso 2000
Chile 2000
Madagascar 2001
Senegal 2001
Maldives 2002
Brazil 2002
Romania 2002
Guinea 2003
IEG - formerly OED

Honduras 2003

[Panama 2003 – observations]
Niger 2004
Yemen 2004
Uruguay 2005
Argentina 2006
Mozambique 2007
Egypt 2007






Broad findings of PPARs:
Enrollments increased everywhere
but….
(a) The schools of the poor may teach very little
Some schools closed




about 30% of the time in Mali -1999
Teachers often absent
Few if any books in class
Class time spent in little activity




copying
engaged in incomprehensible material
playing outside
Many principals, ministry officials, supervisors complacent
(b) Poorer students cannot read well until the end of primary (if
then)
The neuroscience of
reading
Memory principles
Letters = Object recognition
Visual complexity in languages and
scripts
The fluency paradox:
Minimum reading speed
needed for comprehension
If you don’t read fast enough,
by the end of a sentence you forget the
beginning!
Short-term memory
Crucial for reading comprehension
About 7 items
4 pictures
12 seconds at most
Long-term memory
To read an average sentence in an
‘average’ language roughly..
7 items in 12 seconds…

students must read at least a word per 1-1.5
second
with 95% accuracy (correlates .87 with speed).
45-60 words per minute minimum

7 words in 12 seconds equals 45-60 words per minute!




To get through the narrow opening
the mind creates chunks of
information
Letters and other small items become larger pieces that pass as
one through working memory
small items must be chunked and practiced to the point of fluent
performance


We must act in milliseconds!
Vast implications for reading, math calculations, motor skills

This is how children decode ever larger units – from syllables to
words

How do students get to fluency?
Sophisticated skills, higher-level knowledge
are created from ever-larger chunks
practiced to the point of automatic recall
Brain imaging changed the
landscape of reading assessment
Brain imaging techniques
(since about 1995 )
Example:
Brain activation patterns of literates and
illiterates
A special brain area gets activated
(fusiform gyrus, occipitotemporal
lobe)
3 primary reading areas in the brain:
2 for single letters, slow reading
1 for automatic reading
All 3 are used simultaneously
The brain becomes “programmed”
for automatic reading

Nerve “wiring” develops in children’s brains
(“White matter” needed for reading and larger working memory-Nagy et al. 2005)







The occipito-temporal pathway gets activated
The brain identifies entire words rather than single
letters
Long and short words are read equally fast
(silently)
each word or phrase becomes an item
Speed rises to 250+ words per minute
People can’t help but read
Pay attention to message rather than the print
Reading automaticity:
(the literacy vaccine!)
Almost an ‘on-off” switch
Only through consistent pairing of sounds
and letters
• Textbooks
• Feedback
• Homework
• Class time
• With practice children may pass from the off
state to on in 6 weeks
Word superiority effect:
we read faster the letters of words
we know
but…
People become a fluent readers
without knowing the language
Reading sacred texts
To the brain,
letters are just objects
Brain has rules for processing
shapes


•
•
•



Some characters are more visually complex than
others
Some languages have more letters than others
Some letter combinations are more complex
(psycholinguistic grain size)
Various languages-scripts have different
implications for reaching automaticity
Automatic readers use rules that to
beginners are incomprehensible
And instructional time must be used
We always need letters
We read by recognizing letters in a row,
with known words and context hints handled
separately



Letters account for 62% of the adult reading rate
Words 16%, context 22%, individual variance 6%
The processes are not redundant, they work on different words.
•
Like computer technicians fixing problems – letters are generalists

Words are not usually recognized as wholes, the visual system
must isolate and recognize the individual letters to get the word

Implications:
Methods that get children to read whole words are not efficient
If the children read in one language, they can read in another (in
same script).



Pelli, Dennis and Katharine Tillman. Parts, wholes, and context in reading: A triple
dissociation. PLoS ONE, August 2007, e 680.
Visual complexity of characters
Number of characters to
automatize…
probably affects the time
needed to acquire
automaticity
Dhivehi – possibly the simplest functioning
alphabet in the world
Amharic
Exact
spelling but
much larger
matrix with
some pattern
irregularities
More time
needed to
acquire
automaticity,
fluency
Kannada – high error rates
complex forms, multiple visual patterns
Half-consonant
combinations
and vowel
combinations
result in a
matrix of about
300 characters
that must be
automatized
Some are
unpredicatable
Hebrew voweled (small grains)
and unvoweled (large grains)
but the letters are separate

Kol benei ha'adam noldu benei xorin veshavim be'erkam uvizxuyoteihem. Kulam xonenu batevuna uvematspun, lefixax
xova 'aleihem linhog ish bere'ehu beruax shel axava.
All human beings are born free and equal in dignity and rights. They are endowed with reason and conscience and
should act towards one another in a spirit of brotherhood.
‫”‪Voweled Arabic – small “grains‬‬
‫”‪Unvoweled Arabic – large “grains‬‬
‫‪connected letters with rules about‬‬
‫)‪connections (and lack thereof‬‬
‫ار فَا ْست َ ْو َ‬
‫طنُوا‬
‫َو ِإ ِذ ْ‬
‫س ْهالً ِفي أ َ ْر ِ‬
‫ارت َ َحلُوا ش َْرقا ً َو َجدُوا َ‬
‫ض ِش ْنعَ َ‬
‫َاك‬
‫‪ُ .‬هن َ‬
‫صنَ ُع ُ‬
‫ن ‪‬‬
‫س َ‬
‫فَقَا َل بَ ْع ُ‬
‫ض ُه ْم ِلبَ ْعض‪َ « :‬هيَّا نَ ْ‬
‫طوبا ً َم ْش ِويّا ً أ َ ْح َ‬
‫ب‪َ ،‬و ِ ّ‬
‫ارة َ ِب ُّ‬
‫ين ِب ِ ّ‬
‫ت‬
‫الز ْف ِ‬
‫الط َ‬
‫الطو ِ‬
‫َي»‪ .‬فَا ْست َ ْبدَلُوا ْال ِح َج َ‬
‫ش ّ‬
Egypt: Grade 1 whole-word reading
“active learning” class
(without
shapes)
vowels children may identify entire words as particular
Urdu
– multiple
Urdu – multiple
issuesdifficulties
Few vowel signs
even in grade 1
Vowels are not
predictable as in
Arabic
Dots separated
from the main
body of letters
Topological
imprinting
Need to learn the
visual pattern of
each word
separately
Urdu, Farsi, Dari, Pashto
Devanāgarī alphabet for Hindi
African languages have regular spelling
can be automatized in a few months
Fluency to other languages transfers within the same script
Complex visual patterns in
various languages and scripts
(Psycholinguistic grains)
English
through, caught, bake, often, saw, sew
French
Ils etaient, oiseau, mois, etant
Bengali
jomi – earth
boithak – meeting
koThin - difficult
The brain‘s rules for recognizing
object similarity

Mozambique and Angola teach calligraphy early on.

Calligraphic and printed letters cannot be seen as
equivalent initially, only after extensive practice
e=
E=
Result: Students just “sketch”
letters (Mozambique)
The child tried to draw an O: “O sapo”
Some grade 1 textbooks don’t teach reading!
Misguided belief that text interpretation will lead
to recognition of letters (Mozambique)
Mozambique
Grade 1 outside Nampula
When should students be able to
read?
OECD reading study of 16 countries
Latin and Greek scripts
Seymour et al. 2003
Middle-class students
Best-case scenario
But..the more complex the visual
recognition or spelling
the longer it takes to automatize

Automatizing large visual patterns (psycholinguistic
grains) – takes longer,
•



may “trick” some brains
students depend more on language knowledge
And if they don’t know the language? (English,
French, Portuguese, Urdu)
Learning to read in a complex system without
knowing the language is a job for geniuses!
•
•
And if school time is also wasted?
See test results all over Africa
Reading level after 1 year of instruction
% correct
Reading lists of words
100
90
80
70
60
50
40
30
20
10
0
Seymour et al. (2003), British Journal of Psychology`
Items per minute when reading word lists
60
50
40
30
20
10
0
r accuracy/speed = .87
Seymour et al. (2003), British Journal of Psychology
U.S. Oral Reading Fluency Norms
connected text - Spring
Hasbrouck and Tindal (2006)
Grade
1
2
3
4
5
6
7
8
50th %ile
53
89
107
123
139
150
150
151
25th %ile 10th %ile
28
15
61
31
78
48
98
72
109
83
122
93
123
98
124
97
“Oral Reading Fluency Norms: A Valuable Assessment Tool for Reading Teachers.” The Reading Teacher,
59, 2006
Without fluency and accuracy there is no
comprehension (Peruvian data)
Relación observada entre fluidez y comprensión
Comprensisón (respuestas correctas de 3
preguntas)
3.0
2.5
2.0
1.5
Correlación 82%
1.0
0.5
0.0
0.0
10.0
20.0
30.0
40.0
50.0
Fluidez (palabras por minuto)
60.0
70.0
80.0
2002
NAEP
fluency
study,
p. 30
Correct Words per Minute on Grade Level Text
160
Text difficulty
increases
Text difficulty
increases
Correct Words per Minute
150
140
18 WPM
difference
23 WPM
130
22 WPM
difference
difference
120
Tindal, Hasbrouck, &
Jones, 2005
110
US students
100
Fall, winter, spring
T 2006
Torgesen,
F
W
6th Grade
S
F
W
7th Grade
S
F
W
S
8th Grade
Correct Words per Minute on Grade Level Text
120
110
Correct Words per Minute
100
33 WPM
90
difference
80
70
45 WPM
60
difference
50
40
27 WPM
30
20
Torgesen, 2006
winter, spring, fall
Good, Wallin, Simmons,
Kame’enui, & Kaminski, 2002
difference
W
S
1st Grade
F
US students
W
2nd Grade
S
F
W
3rd Grade
S
New monitoring indicator
from neurocognitive research




60 words per minute for all
•
In just about every language and script:
By the end of grade 1 students should “crack
the code”
By the end of grade 2 at the latest students
should read common words fluently
Students in grade 7 (1st secondary year)
should read about 120-150 words per minute
and give a summary of what they read
When visual patterns are complex,
basic reading needs more time and
money






(psycholinguistic grain size theory)
English reading takes 2.5 times longer to acquire over
German, Turkish, Spanish (4 months or so needed)
French and Portuguese also take longer.
Indian languages – regular spelling, complex scripts
• High error rates in south Indian scripts
Arabic – easy with vowels, difficult otherwise
• Perennially slower reading, high error rates
Urdu, Persian problematic for the poor
35 words per minute
How children should read – Cuba grade 2
Reading fluently enough
to understand?
60 wpm correctly?


Rural Indonesia grade 2
•
simple spelling rules, good class time use
Rural Niger – best 6th grader, graduating in 3 weeks
•
Study only in French, time use uncertain
Oral reading samples
Mozambique
A familia do meu melhor amigo
Overage student end of grade 3
One of 2 students who could
read fluently in the class
70 words per minute
Periurban 5th graders - Maputo
Students becoming fluent late
may always read slowly
have limited comprehension
• If they finally learn reading in grade 6
– They may read 70 wpm in grade 8
– They get no more books by grade 8
• They will read little secondary school
– university or teacher training colleges
– They cannot read fast enough to consult
sources
– Or read volumes of text
– At 110 words per minute, it takes 5 minutes per page
Can this 3rd grader in Panama solve
the problem he is reading?
How to teach complex scripts
in 1-2 years to the poor?


Lots of practice, build up letter recognition speed - phonics
Automatize the common patterns that enable students to read
most material
•



Students in English learn the most common words
Teach the more rare signs later
•
•
•
E.g. postpone certain half-consonants of Indian languages (or
use “halant”?)
Keep Arabic, Urdu, Farsi voweled throughout primary school
Keep the Arabic letters on a straight line
Get children practice with feedback!
•
E.g. read 1 minute a day with every child
Textbooks to take home with much text for all
•
Not just a few ‘generative words’ to bring from home
Teaching fluent reading to the
poor takes more effort

child and family characteristics predict initial
reading skills
•


Though not reading development over time.
High percentage of poor students in a class is
often associated with lower student reading
performance after accounting for
socioeconomic and instructional variables.
How does this happen day to day in class?
Kainz, Kirsten and Lynne Vernon-Feagans.2007. The ecology of early reading
development for children in poverty. The elemenary school journal, 107, 407-427
Teach decoding strategy
Analogies to automatize small units
Brain takes this up well
a
e
i
o
u
Letter Fatha Qasra Dhamma
‫ض‬
‫ض‬
َ
‫ِض‬
ُ‫ض‬
‫ص‬
َ‫ص‬
‫ِص‬
‫ُِ ص‬
‫ث‬
‫َِ ث‬
‫ث‬
‫ُِ ث‬
‫ق‬
‫َِ ق‬
‫ِق‬
ُ‫ق‬
‫د‬
‫َِ د‬
‫د‬
‫ُِ د‬
B
ba
be
bi
bo
bu
C
ca
ce
ci
co
cu
D
da
de
de
do
du
F
fa
fe
fi
fo
fu
G
ga
ge
gi
go
gu
‫ش‬
‫َِ ش‬
‫ِش‬
ُ‫ش‬
H
ha
he
hi
ho
hu
‫س‬
‫َِ س‬
‫ِس‬
‫ُِ س‬
Etc
cte
Mozambican NGO Progresso
“Eu leio”: Literacy through phonics
How not to teach?
Is it possible to learn reading or
other skills as time is used in many
countries?
Only a fraction of the countries’ investment
is actually converted into learning time
Class time as allotted by a government (e.g., 200 days, 1000 teaching hours)
Remaining after school closures (strikes, weather, teacher training, extra holidays)
Remaining after teacher absenteeism and tardiness
Remaining after student absenteeism
Class time devoted to
any learning task
Learning
time
relevant to
curriculum
Instructional time loss in a sample of countries
Pernambuco
(Brazil)
200
4.79
Ghana
Morocco
Tunisia
197
3.17
204
1.38
190
5.15
Days after Closures
195.21
193.83
202.62
184.85
Teacher Absence (Days)
12.76
43.01
13.36
11.55
Teacher Delays
Early Class Dismissals
5.50
2.31
39.75
2.43
6.94
6.68
1.27
1.22
No. days schools operated
% Year available for teaching
174.65
87.3%
108.6
55.1%
175.6
86.1%
170.8
89.9%
Engagement Rate in Interactive
or Passive Classroom Tasks
72.1%
70.2%
82.6%
86.7%
125.9
76.3
145.1
148.1
63.0%
38.7%
71.1%
77.9%
19.3%
21.1%
9.2%
9.9%
50.8%
30.5%
64.6%
70.2%
Student Absence (Days)
7.82
9.04
4.30
3.35
Student Delays (Number of Times)
5.64
10.61
5.187
2.63
School Year (Days)
School Closures (Days)
School Days Devoted to Learning
School Year % Spent Engaged
in Learning Tasks
Student off Task Rate
Instructional Efficiency Given Off-Task Rate
Grade-wise time use



Teachers of lower grades spent more time on
organization
younger children - higher off-task rates, given a
briefer attention span.
Older grades - less reading aloud and drill,
increase in lecturing and discussion
•
•
Morocco: reading activities 17% of the time in grades
1 and 4, 2.5% in grade 5 (and 15.1% in grade 6,
possibly to prepare for exams).
Brazil: in grade 8 copying time increased because
secondary students lack textbooks.
Percentage of Time Spent in Instructional Tasks
Instructional Strategy
% of time
U.S Classroom
Criteria
Brazil
Ghana
Morocco
Tunisia
Interactive instruction
50% or more
52.4
59.9
62.8
61.2
6.7
8.7
15.7
15.3
32.8
19.9
26.7
27.9
Discussion
6.3
24.1
6.6
6.2
Practice Drill
1.4
6.5
12.3
11.3
19.6
10.3
19.9
25.6
16.3
7.4
14.8
22.9
3.0
2.9
5.0
2.7
72.1
70.2
82.6
86.7
15% or less
27.9
28.0
17.8
13.3
6% or less
19.3
21.1
9.2
9.9
Oral Reading
Teaching, Explanation
Passive instruction
35% or less
Seatwork
Copy
Total Instructional Time
Organizing/Management
Student Off Task Rate
Examples of time loss

Class starts late (Honduras, Brazil)

Book distribution (Argentina, Honduras)

Slow motion, limited activity (Niger-Tilabery)

One student on the board – (Indonesia, Yemen)

Copying – (Guinee)

Secondary activities (Brazil, Argentina)

Group interactions (Argentina, Brazil)

Teachers doing other things – everyone off task (Brazil)
Conclusions
How to educate the poor most
efficiently?
Get results efficiently given time and money

Explicit, systematic instruction for the poor and
those who know less
• Use software for multiple repetitions that tire teachers

practice to automatize “low-level” processes that
the middle class takes for granted
•
•
•
•
much feedback, extensive practice with books - phonics
Use easily spelled African languages before reading
English
Speed and fluency in basic calculations
Focus investments in grades 1-2 – before the students fall
behind
We need much more and more
specific research on various scripts




Some syllabic scripts may not be viable
in the long run as currently used and
taught
Arabic printing and vowel inclusion
– for use by all
Some scripts need to be reformed
• Which bodies are responsible in each case?
• Eg. South Indian scripts?
Controversies and concerns






Educational philosophy vs. neurocognitive research
outcomes
How to disseminate the lines of reasoning?
•
Philosophical debates – needed but they delay decisions
Reading automaticity sounds too mechanistic to some
How to monitor these ‘concrete’ benchmarks and learning
outcomes early on
Should the poor be taught differently from the middle class?
How to undertake necessary research on memory effects in
classroom?
Alternative measurements

Wordchains
• Piloting needed for this paper-and-pencil
predictor of reading fluency
We’d better hurry
EFA is ending in 7 years
Thank you for listening!
Extra slides