Simplifying reading: Implications for
instruction
Janet Vousden
University of Warwick
Michelle Ellefson, Nick Chater, Jonathan Solity
Overview

English spelling-to-sound inconsistency and reading

rational analysis of English reading

applying the simplicity principle

analysis of some common reading programmes
Spelling-to-sound mappings

spelling-to-sound mappings in English are not
transparent at sub-lexical level

some spellings are consistent:
“ck”: duck - /dʌk/, mock - /mok/, etc
and a simple grapheme-phoneme rule will suffice;
ck - /k/

others are not:
“ea”: beach - /biːtʃ/, real - /rɪəl/, great - /ɡreɪt/, or
head - /hɛd/
grapheme
should, four,
level
country,
- “ou”tenuous,
grapheme
soul,
is
credited
journal, with
cough,
having
pompous
10 different pronunciations (Gontijo,
Gontijo, & Shillcock, 2003)
 most
e.g.,obvious
round, group,
at the

overall measure of (in)consistency in a language is its
orthographic depth: average number of pronunciations
per grapheme

for English, orthographic depth estimates
 2.1 - 2.4 (Berndt, Reggia, & Mitchum, 1987; Gontijo,
Gontijo, & Shillcock, 2003) polysyllabic text
 1.7 (Vousden, 2008) monosyllabic text

compare e.g. Serbo-Croat which has OD of 1



how do literacy levels in
English compare with
other languages?
can differences in
consistency account for
the difficulty in learning to
read English?
yes - inconsistency
clearly increases difficulty
of learning to read
compared with more
consistent languages
(Frith, Wimmer & Landerl,
1998)
Language
Real-words Nonwords
Greek
98
92
Finnish
98
95
German
98
94
Italian
95
89
Spanish
95
89
Swedish
95
88
Dutch
95
82
Icelandic
94
86
Norwegian
92
91
French
79
85
Portuguese
73
77
Danish
71
54
Scottish English
34
29
Data: % correct reading scores (adapted from Seymour, Aro,
& Erskine, 2003).
lag in performance persists through school years
% c o rre ct

Data: non-word reading
accuracy (reproduced from
Frith, Wimmer, & Landerl,
1998)
100
90
80
70
60
50
40
30
20
10
0
German
English
6
7
8
9
10
Age
11
12
13

Most often, vowel graphemes are inconsistent, but can
use immediate context to resolve ambiguity

C V C - C V or V C

ambiguity can be resolved by considering the following
consonant (a rime unit) rather than the previous
consonant (Treiman et al., 1995)

ea
pronounced to rhyme with breath when followed by
‘d’ ~80%
 pronounced to rhyme with meat when followed by ‘p’
100%
also, rime units are more consistent than graphemes
 23% graphemes inconsistent
 15% rimes inconsistent


Choosing spelling-to-sound mappings

influences from developmental literature (do rimes or
gpcs predict reading ability?)

variety of approaches from reading schemes
(Rhymeworld, THRASS, etc)

so many to choose from,
 ~2000 rime mappings
 ~300 grapheme mappings

and many are inconsistent
 15% rimes, 23% graphemes
Rational analysis

Attempt to explain behaviour in terms of adaptation to
environment, independent of details of cognitive
architecture

Solution adopted by cognitive architecture should reflect
structure of environment

e.g., Anderson & Schooler (1991) showed that the
probability that a memory will be needed over time
matches the availability of human memories
 same factors that predict memory performance also
predict the odds that an item will be needed
 i.e. reliable effects of recency and frequency

factors that affect performance of skilled readers should
be reflected in the statistical structure of the language,
e.g. frequency and consistency
effects of word frequency in naming and lexical
decision
effects of rime frequency on word-likeness judgements
and pronunciation
effects of grapheme frequency in letter search and
word priming experiments

by examining linguistic factors that skilled readers have
adapted to, could the input be more optimally structured
for learners?
Analyses of spelling-to-sound mappings

rational analysis predicts the most frequent and
consistent mappings best predict pronunciation

interested in the frequency & consistency of mappings at
level of words, rimes, and graphemes, and their ability to
predict correct pronunciation

CELEX database: 7,297 different monosyllabic words,
10,924,491 words in total
Words
1
P ro p o rtio n o f Te x t R e a d
F re qu e n cy in 1 0 0,0 0 0 's
14
12
10
8
6
4
2
0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
50
100
150
Rank order of Frequency
200
0
100
200
300
Number of Words
400
500
Onsets and rimes

Exclude 100 most frequent words:
 7,197 diffrent words, total of 2,263,264 words

Create table of onset and rime mapping frequencies,
remove all but most frequent of inconsistent mappings
Rime
Frequency
oul - /əʊl/
731
oul - /aʊl/
175
oul - /uːl/
12
ove - /ʌv/
4779
ove - /uːv/
1852
ove - /əʊv/
838
Rimes
500
50
450
45
400
40
350
35
Fre q u en c y in 1 0 0 0 's
Fre q ue n cy in 1 0 00 's
Onsets
300
250
200
150
30
25
20
15
100
10
50
5
0
0
0
10
20
30
Rank Order of Frequency
40
50
0
20
40
60
Rank Order of Frequency
80
100
0 Onsets
10 Onsets
20 Onsets
30 Onsets
40 Onsets
80 Onsets
Pro p o rtio n of T ex t R ea d (% )
60
50
40
30
20
10
0
0
20
40
60
Number of Rimes
80
100
GPCs

exclude 100 most frequent words:
 7197 diffrent words, total of 2,263,264 words

create table of GPC mapping frequencies, remove all but
most frequent of inconsistent mappings
Rime
Frequency
g - /g/
133930
g - /dʒ/
33342
g - /ʒ/
98
i - /ɪ/
153606
i - /aɪ/
46628
i - /iː/
455
GPCs
10
9
60
7
Pe rce n ta ge o f T e x t R e ad
Fre q ue n c y in 1 0 0,0 0 0 's
8
6
5
4
3
2
1
50
40
30
20
10
0
0
0
20
40
Rank Order of Frequency
60
80
0
20
40
60
Number of GPCs
80
100
P e rc en ta g e o f T ex t R ea d
100
95
90
Grapheme Phoneme
85
Onset/Rime
80
75
70
0
100
200
Number of Mappings
300
Summary

some words much more frequent than others, therefore
sight vocabulary very effective for small number of
words, up to ~100

sub-lexical units also have skewed frequency
distribution, and learning the most frequent mappings
predicts high potential outcome

high initial gains with GPCs, greater overall gain with
rimes in the long run

What is the optimal size unit to learn?

Potential benefits for reading outcome are larger for
onset/rimes, but is this out-weighed by the cost of
remembering many more mappings?

Can we measure the potential benefit from, and cost of,
remembering mappings for
 GPCs
 onset/rimes
 A combination of both ?
The Simplicity Principle

reading, like much high-level cognition, involves finding
patterns in data, but many patterns are compatible with
any finite set of data - so how does the cognitive system
choose from the possibilities?

Using the simplicity principle, choose the simplest
explanation of the data - intuitively, has long history
(Occam’s razor)

can quantify simplicity by measuring (shortest)
description from which data can be reconstructed - trade
off brevity against goodness of fit
 cognition as compression

implement with minimum description length (MDL)
 more regularity = more compression
 no regularity = no compression, just reproduce data

can measure compression with Shannon’s (1948) coding
theorem - more probable events are assigned shorter
code lengths:
length/bits = log2(1/p)

measure code length to specify:
 hypothesis about data (mappings)
 data, given hypothesis (decoding accuracy, given
mappings)
Method


determine mappings & frequencies from monosyllabic
corpus of children’s reading materials (Stuart et al.,
2003), for mapping sizes:
 words
 CV/C (head/coda)
 C/VC (onset/rime)
 GPCs
determine code length to describe
mappings
decoding accuracy, given mappings
for each mapping size
Table 1. A list of reading schemes/series used by over a third of schools in the survey
Name of scheme
% using scheme
Ginn 360
74%
Storychest
58%
Magic Circle
58%
1 2 3 and Away
50%
Griffin Pirates
43%
Breakthrough to Literacy
41%
Bangers and Mash
40%
Wide range readers
38%
Dragon Pirates
37%
Through the rainbow
34%
Ladybird read-it-yourself
33%
Humming birds
32%
Thunder the dinosaur
29%
Link Up
29%
Gay Way
27%
Monster
27%
Oxford Reading Tree
27%
Once Upon a Time
26%
Trog
26%
Included in database?
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Code length for mappings
words
CV/C
C/VC
GPCs
letter
sound
freq
letter
sound
freq
letter
sound
freq
wiː
bi
bI
1
d
d
7
t
t
17
kæn
ca
kæ
1
g
g
4
i
I
11
bʌt
do
dɒ
1
f
f
1
a
æ
11
ɒn
fro
frɒ
2
wh
w
1
g
g
7
miː
ra
ræ
1
k
k
4
c
k
4
bæk
n
n
9
an
æn
4
ay
eɪ
2
kʌm
k
k
1
og
ɒg
2
a*e
eɪ
1
frɒg
lp
lp
1
elp
elp
1
(1/p(w)) ++ log
log22(1/p(i))
(1/p(iː))+ +log
log
length = log2(1/p(b))
+
2(1/p(newline))
2(1/p(space))
log2(1/p(b)) + log2(1/p(I)) + log2(1/p(newline))
Code length for decoding accuracy
apply letter-to-sound rules to
produce a list of pronunciations
arrange in rank order of most probable
(computed from letter-to-sound
frequencies) & note rank of correct
pronunciation
bread
breId
bri:d
brɛd
bread
bri:d
brɛd
breId
code length for data, given hypothesis = log2(1/p(rank=2))
14
12
log 2(1/p )
10
8
6
4
2
0
0
0.2
0.4
0.6
p
0.8
1
Simulations

overall comparison between different unit sizes for whole
vocabulary

how does code length vary as a function of size of
vocabulary for each unit size?

optimize number of mappings by removing those that
reduce total code length

compare different reading schemes
Comparing different unit sizes for whole vocabulary
90000
80000
rules
acc
tot
70000
C ost/bits
60000
50000
40000
30000
20000
10000
0
Words
CV/C
C/VC
Rule size
GPCs
Code length as a function of vocabulary size
90000
4000
words
words
CV/C
CV/C
80000
3500
70000
3000
C/VC
C/VC
GPC
GPC
C ost/bits
C ost/bits
60000
2500
50000
2000
40000
1500
30000
1000
20000
500
10000
0
0
0
10
500
30
1000
501500
Vocabulary
Vocabularysize
size
2000
70
2500
90
3000
Optimizing number of mappings
All mappings
Mapping Size
Mappings
remaining
N
Total
code
length
N
Total
code
length
Words
3000
77,825
-
-
Onset/rimes
1141
48,612
404
29,420
GPCs
240
10,845
114
8,536
GPCs: Description length reduced by removing mainly
inconsistent, low frequency mappings
Comparing different reading schemes
Scheme
Jolly Phonics
N
GPC
rules
43
Hutzler et al. (2004)
67
ERR (Solity & Vousden, 2008)
77
Letters & Sounds
94
THRASS
106
30000
rules
acc
25000
total
C o st / b it s
20000
15000
10000
5000
0
Jolly N=43
Hutzler
N=71
ERR N=77
LettSou
N=94
THRASS
N=106
simplicity
N=111
all N=240
30000
rules
25000
acc
total
C o s t/b its
20000
15000
10000
5000
0
43
71
77
94
106
N rules
111
240
Decoding accuracy by scheme
80
70
60
% c o rrec t
50
Schemes
Simplicity
40
30
20
10
0
Jolly
N=43
Hutzler
N=71
ERR
N=77
LettSou
N=94
THRASS
N=106
all N=240
simplicity
N=111

ERR implemented as a reading intervention in 12 Essex schools:
Data: from Shapiro & Solity (2008)
60
B A S R a w S c o re
50
Comparison
ERR
40
reading
difficulty
30
20
serious
reading
difficulty
10
0
base
YR
Y1
Compariso
n
ERR
20%
5%
5%
1%
Y2
School Age
increase in reading scores significantly greater for ERR
schools
Some conclusions

small amount of sight vocabulary accounts for large
proportion of text, but only small vocabularies most
simply described by whole words
 Complements recent work by Treiman and colleagues
that shows children learn better when association
between sound and print is non-arbitrary

As a homogenous set, GPCs provide a simpler
explanation of the data

choosing the best set could be important
Descargar

Units of English spelling-to-sound mapping: A rational