Expertise reversal effect on reading comprehension: A case of english for specific purposes (esp)

74 Huynh Cong Minh Hung. Journal of Science Ho Chi Minh City Open University, 7(4), 74-83 
EXPERTISE REVERSAL EFFECT ON READING COMPREHENSION: 
A CASE OF ENGLISH FOR SPECIFIC PURPOSES (ESP) 
HUYNH CONG MINH HUNG 
Ho Chi Minh City Open University, Vietnam – hung.hcm@ou.edu.vn 
 (Received: June 30, 2017; Revised: November 07, 2017; Accepted: November 29, 2017) 
ABSTRACT 
Cognitive Load Theory assists researchers in designing instructional procedures that can lead to enhancement 
of reading skills. This paper aims to examine cognitive load effect as expertise reversal effect on reading 
comprehension of English for Specific Purposes (ESP). An experiment was designed to investigate whether the 
expertise reversal effect can be applied to reading comprehension of ESP. The implications of the experiment 
findings can be used in teaching and learning ESP reading comprehension. The findings will help instructors design 
more appropriate reading comprehension instructions with alternative versions to integrate different domains such as 
English for Geography and Mathematics effectively and to test the expertise reversal effect on reading 
comprehension. 
Keywords: Cognitive Load Theory; Expertise reversal effect. 
1. Introduction 
Cognitive Load Theory (CLT) has 
developed since the 1980s and attracted many 
researchers all over the world. CLT is 
concerned with the limitation of working 
memory. According to CLT, reading 
comprehension is defined as a constraint of a 
limited working memory (Eskey and Grabe, 
1988). It will be more difficult for learners if 
working memory goes beyond its limitations 
(Goldman, Varma and Cote, 1996). Another 
difficulty for reading comprehension is the 
various levels of readers. According to 
Daneman and Capenter (1983) and Perfetti 
(1985), low level readers who do not have 
enough automation of schemas in reading 
comprehension may generate increased 
cognitive load. Obviously, differences 
between high level readers (experts) and low 
level readers (novices) are explained by using 
levels of expertise (Chi, Feltovich and 
Glasser, 1981). There are several instructional 
effects generated by CLT as the expertise 
reversal effect when instructions useful for 
novices may be unhelpful for more expert 
readers (Kalyuga, Ayres, Chandler and 
Sweller, 2007). The Expertise Reversal Effect 
is examined not only in natural sciences but 
also in well-structured domains like literacy 
texts (Kalyuga and Renkl, 2010) and biology 
texts (McNamara, Kintsch, Songer, 1996). 
The results of McNamara et al.’s (1996) 
experiments showed that novices would 
benefit from information added to original 
instructional text while experts were 
beneficial from original instructional text 
(McNamara et al., ibid). Oksa, et al. (2010) 
used Shakespearean text to differentiate 
instructional effectiveness and found that it 
was difficult for novices to comprehend the 
text, which used a lot of sophisticated 
Elizabethan English language. 
McNamara et al. (ibid) investigated the 
effect of text cohesion on readers’ 
comprehension. The results demonstrated that 
low level readers benefited more from high-
cohesive texts whereas high level readers 
benefited more from low-cohesive texts. This 
is because high-cohesive texts employed 
many anaphoric referents, sentence 
connectives, background information, 
meaningful headings and paragraphs while 
low cohesive texts do not contain so much 
structuring information (Tubingen, 2011). 
 Huynh Cong Minh Hung. Journal of Science Ho Chi Minh City Open University, 7(4), 74-83 75 
McNamara et al. (ibid) clarified that low-
cohesive text required high level readers to 
engage in compensatory processing to infer 
unstated relations in the texts as germane 
possessing, while high cohesive text seduce 
high level readers to more passing processing 
instead of activating relevant prior knowledge 
of their own. In an effort to support the 
germane cognitive load explanation, O’Reilly 
and McNamara (2007) did a study about its 
effect on reading comprehension and found 
that learners with high prior knowledge and 
low reading skills did not benefit from high 
cohesive texts while skilled learners with high 
knowledge and reading skills would benefit 
from high cohesive texts. On explain their 
findings, O’Reilly and McNamara (ibid) 
considered that good reading skills assist high 
knowledge learners in involving in germane 
cognitive load processing. Kalyuga et al. 
(2007) explained that high knowledge 
learners, as skilled readers, know how to 
apply active processing strategies into well 
guided text instructions. McNamara et al. 
(ibid) stated that information added to an 
original biology instructional text for 
coherence enhancement was advantageous to 
low-knowledge readers only. However, an 
original minimally coherent format text was 
useful for high-knowledge readers more than 
an enhanced one. 
Unlike the study done by McNamara 
et al. (ibid), this experiment was conducted 
within the framework of CLT in which 
cognitive load approaches were used to 
measure effort and the efficiency. 
Accordingly, the current experiment used 
expanded and reduced versions instead of 
high-cohesive and low cohesive texts used by 
McNamara et al. (ibid) in their study. In the 
expanded and reduced versions, the sentences 
were added or removed while in the high-
cohesive texts and low cohesive texts, the 
content of the versions were modified by 
changing cohesive devices. 
Though CLT has been introduced since 
2007 (Huynh, 2007), no studies on cognitive 
load effects as expertise reversal effect have 
been carried out in the Vietnamese context. 
The paper is the first study in Vietnam to 
investigate the expertise reversal effect on 
EFL area related to reading comprehension. 
Based on a review of the study by McNamara 
et al (ibid), the experiment had the following 
aims: 
Firstly, the experiment was investigated 
within the CLT and assumed that cognitive 
processes caused expertise reversal effect 
while McNamara et al’s (ibid) study did not 
measure any cognitive load and was just 
based on learning outcomes and studying 
times. McNamara et al. (ibid) firstly used 
different cohesive versions of a biology text 
and a history text (McNamara and Kintsch, 
1996). The experiment assumed that high 
knowledge readers (or experts) do not benefit 
from expanded versions because they are 
overloaded by extraneous processing due to 
redundant information. 
Secondly, the experiment used the 
subjective ratings in the expertise reversal 
effect. The experiment assumed that how 
high level readers (experts) and low level 
readers (novices) perceived difficulty of 
comprehension of different versions 
(expanded and reduced versions). 
2. Method 
Participants 
The participants were 120 Vietnamese 
second-year students consisting of 60 second-
year students studying in the department of 
Geography and 60 second-year students 
studying in the department of Mathematics, 
Ho Chi Minh City University of Education. 
Their English proficiency was quite different 
because the students took different English for 
Specific Purposes (ESP) courses for 
Geography and for Mathematics, respectively. 
The participants were divided into an expert 
group and a novice group. The expert group 
76 Huynh Cong Minh Hung. Journal of Science Ho Chi Minh City Open University, 7(4), 74-83 
consisted of the 60 students from the 
Department of Geography because the 
material used in this experiment was a 
geographical text that required them to have 
appropriate English proficiency in Geography. 
The novice group included the 60 students 
from the Department of Mathematics. They 
were categorized as novices because they 
were not familiar with the materials used in 
the experiment. Both experts and novices 
were randomly assigned to either a reduced or 
an expanded text version group. 
 Materials 
The Geographical text entitled “What 
killed the dinosaurs?” was extracted from the 
book “Earth Science” (Feather R.M., Snyder 
S.L., 1993). The original text had 124 words. 
A reduced version included text in which 
some sentences were removed from the 
original text. The reduced version had 60 
words. 
The expanded version consisted of extra 
seven sentences added to the reduced version 
to explain more about dinosaur extinction. For 
example, sentences such as “In the search for 
answers to what killed the dinosaurs, 
scientists have looked beyond fossils. There is 
increasing evidence that the impacts of 
meteorites have had important effects on 
earth, particularly in the field of biological 
evolution” were added to the first paragraph 
to explain evidence of dinosaur extinction. 
The length of the expanded version was 237 
words. 
Procedure 
Half of the experts and novices were 
randomly allocated to either of the two 
reduced or expanded text versions. During the 
learning phase, participants were required to 
read either of the two versions and answer 6 
questions in 12 minutes (2 minutes each). 
After the learning phase, participants were 
given the test questions. They were required 
to answer the test questions without seeing the 
text. 2 out of 5 questions were identical to 2 
questions presented during the learning phase 
for the two versions. The 2 identical questions 
were “When did the last species of dinosaurs 
become extinct?” and “How long had 
dinosaurs dominated the land?” These 2 
questions were chosen because they serve as 
background for understanding both versions 
of the text. 
After the learning phase, participants 
ranked the subjective difficulty score of the 
textual materials from 1 as “extremely easy” 
to 9 as “extremely difficult”. The duration of 
the test phase was 10 minutes (2 minutes for 
each question). The appendix presents the 
questions used in both learning and test 
phases. 
Scoring 
In both phases, one mark was given for a 
correct answer and a zero mark for an 
incorrect answer. An answer was deemed 
incorrect if it had a wrong choice or lacked 
key words of the correct answer. The answers 
to the questions were explicitly stated in the 
text and only one sentence was required as an 
answer for each of them. For example, the 
correct answer to question 1 of the learning 
phase “What is one theory of dinosaur 
extinction?” was “A hypothesis of dinosaur 
extinction is that a large meteorite collided 
with earth”. The key words for the answer 
were “a large meteorite”. Similarly, the 
correct answer to question 5 in the test phase 
“How long had species of dinosaurs 
dominated the land?” was obtained from the 
sentence “Species of dinosaurs had dominated 
the land for 130 million years” with the key 
words being “for 130 million years”. 
The maximum total score for the tests 
was 6 marks in the learning phase and 5 
marks in the test phase. The total score of 
each participant in the two phases was then 
converted to a percentage for analysis. 
3. Results 
The performance scores in the learning 
phase and the test phase were analyzed by a 2 
 Huynh Cong Minh Hung. Journal of Science Ho Chi Minh City Open University, 7(4), 74-83 77 
(instructional text versions: reduced and 
expanded version) x 2 (expert and novice 
groups) ANOVA (see Table 1). The 0.05 
significance level was used throughout the 
analysis. The performance mean scores in 
Table 1 are expressed graphically in Figure 1 
and 2 for each expertise group indicating the 
mean scores of participants. 
 In the learning phase, the main effect of 
version indicated that there was no significant 
difference, F (1,116) = 2.50, MSE = 889.0, 
p = .116. The main effect of expertise group 
indicated a significant difference, F (1,116) = 
5.28, MSE= 889.0, p = .023, partial Eta 
Squared = .044. The experts (geography 
students) obtained higher scores than the 
novices (mathematics students). There was a 
significant interaction between expertise 
groups and versions, F (1,116) = 12.41, 
MSE = 889.0, p = .001, partial Eta squared 
= .097. Following the significant interaction, 
simple effects tests indicated that, for the 
expert group, in the learning phase the 
reduced version had significantly higher mean 
scores than those of the expanded version, F 
(1,116) = 13.04, MSE = 889.0, p < .001, 
partial Eta Squared = .101. For the novice 
group in the learning phase, the expanded 
version did not differ significantly from the 
reduced version F (1,116) = 1.88, MSE = 
889.03 p = .215. Figure 1 describes the 
distribution of the learning scores of novices 
and experts in two versions: reduced and 
expanded. The figure shows the lowest score 
and the highest score. 
In the test phase (see Table 1), the main 
effect of expertise groups showed a significant 
difference, F (1,116) = 5.93, MSE = 297.3, p 
= .016, partial Eta Squared = .044 and the 
main effect of versions was significantly 
different, F (1,116) = 7.00, MSE = 297.3, p = 
.009, partial Eta Squared = .057. There was 
also a significant interaction between the two 
groups and versions, F (1,116) = 84.8, MSE = 
297.3, p < .001, partial Eta squared = .422. 
Simple effect tests showed that, for the expert 
group, the reduced version had significantly 
higher mean scores than those of the 
expanded version, F (1,116) = 70.3, MSE = 
297.3, p < .001, partial Eta Squared = .377, 
while for the novice group, the expanded 
version was better than the reduced version, F 
(1,116) = 21.5, MSE = 297.3, p < .001, partial 
Eta Squared = .157 (see Figure 2). Figure 2 
revealed that higher knowledge students 
learned better from the reduced version than 
from the expanded version, while the lower 
level students learned better from the 
expanded version than from the reduced 
version. 
Table 1 
Percentage means and Standard deviations of performance scores in the Experiment 
Phase Group Version Mean Std. Deviation N 
Learning Novice Expanded 54.96 34.52 30 
Reduced 44.40 28.48 30 
Total 49.68 31.82 60 
Expert Expanded 48.29 31.36 30 
Reduced 76.09 23.85 30 
Total 62.19 30.97 60 
Total Expanded 51.62 31.87 60 
Reduced 60.25 30.55 60 
Total 55.93 31.89 120 
78 Huynh Cong Minh Hung. Journal of Science Ho Chi Minh City Open University, 7(4), 74-83 
Phase Group Version Mean Std. Deviation N 
Test Novice expanded 26.0 24.15 30 
Reduced 5.33 10.41 30 
Total 15.6 21.18 60 
Expert expanded 14.66 8.60 30 
Reduced 42.0 20.5 30 
Total 23.3 24.6 60 
Total expanded 15.33 20.94 60 
Reduced 23.66 24.56 60 
Total 19.50 23.11 120 
Figure 1. Performance scores in the learning phase 
Figure 2. Performance scores in the test phase 
3030 3030N =
GROUP
expert groupnovice group
L
E
A
R
N
IN
G
120
100
80
60
40
20
0
-20
VERSION
expanded version
reduced version
3030 3030N =
GROUP
expert groupnovice group
T
E
S
T
100
80
60
40
20
0
-20
VERSION
expanded version
reduced version
9610012
56214338
34
8520673
 Huynh Cong Minh Hung. Journal of Science Ho Chi Minh City Open University, 7(4), 74-83 79 
Mental effort ratings (Table 2) 
demonstrated that the main effect of version 
was not significant, F (1,116) = .011, 
MSE = .747, p = .916. The main effect of 
expertise group was significant, F (1,116) = 
22.5, MSE = .747, p < .001, partial Eta 
Squared = .163 (see Table 2). There was a 
significant interaction between the groups and 
versions, F (1,116) = 18.7, MSE = .747, 
p < .001, partial Eta Squared = .139. Simple 
effect tests revealed that the effort scores of 
the expanded version were higher than those 
of the reduced version for the expert group, 
F (1,116) = 9.83, MSE = .747, p = .002, 
partial Eta Squared = .078 while the effort 
scores of the reduced version were higher than 
those of the expanded version for the novice 
group, F (1,116) = 8.92, MSE =.747, p = .003, 
partial Eta Squared = .071 (Figure 3). 
According to Paas and Van Merrienboer 
(1993), an efficiency score can be generated 
by using the difference between the z score of 
performance and the z score of effort. The 
main effect of version was not significant, 
F (1,116) = 1.34, MSE = .921, p = .209. The 
main effect of expert groups was significant, 
F (1,116) = 21.4, MSE=.921 p < .001, partial 
Eta Squared = .156 (See Table 4). There was 
a significant interaction between the groups 
and versions, F (1,116) = 27.0. MSE = .921, 
p < .001, partial Eta Squared = .189. Simple 
effect tests indicated that the reduced version 
was relatively more efficient than expanded 
version for the expert group, F (1,116) = 20.2, 
MSE = .926, p < .001, partial Eta Squared 
=.148. In contrast, the expanded version was 
relatively more efficient than the reduced 
version for the novice group, F (1,116) =8.17, 
MSE = .926, p = .005, partial Eta Squared 
=.066 (Figure 4). 
Table 2 
Effort and relative instructional efficiency in the experiment 
Group Version Effort Efficiency 
Mean SD Mean SD 
Expert Expanded 5.53 0.937 .0520 1.18703 
 Reduced 4.83 0.648 .8025 .83503 
 Total 5.18 0.837 .4273 1.08561 
Novice Expanded 5.60 1.102 -.3472 .82794 
 Reduced 6.27 0.691 -.5074 .95412 
 Total 5.93 0.972 -.4273 .88933 
Total Expanded 5.57 1.015 -. 1476 1.03442 
 Reduced 5.55 .982 .1476 1.17745 
 Total 5.56 .994 .0000 1.07729 
80 Huynh Cong Minh Hung. Journal of Science Ho Chi Minh City Open University, 7(4), 74-83 
3030 3030N =
GROUP
novice groupexpert group
S
C
O
R
E
7.5
7.0
6.5
6.0
5.5
5.0
4.5
4.0
3.5
VERSION
expanded version
reduced version
Figure 3. Effort scores of the two groups 
Figure 4. Efficiency scores of the two groups 
4. Discussion 
As expected, the results showed that in 
the learning phase there was a significant 
interaction between the two groups and the 
two versions. The results demonstrated that, 
for the expert group, the reduced version 
outperformed the expanded version. The 
experts had better English proficiency in 
Geography. Thus, the experts were able to 
answer the question quickly and accurately. 
To comprehend the reduced version, the 
experts found it easy to find key words and to 
answer the questions. However, for the 
expanded version, the experts found it more 
difficult to answer the questions because the 
information provided and added to the version 
were redundant and caused an extraneous 
cognitive load. The results of the experiment 
in the learning phase for experts were 
different from previous studies (Oblinger and 
Oblinger, 2005; Chujo and Utliyama, 2005) in 
which text length had no significant effect on 
3030 3030N =
GROUP
NOVICE GROUPEXPERT GROUP
S
C
O
R
E
3
2
1
0
-1
-2
-3
VERSION
EXPANDED GROUP
REDUCED GROUP
68
 Huynh Cong Minh Hung. Journal of Science Ho Chi Minh City Open University, 7(4), 74-83 81 
reading comprehension. 
Contrary to expectation, in the learning 
phase, the results revealed that the expanded 
version did not significantly outperform the 
reduced version for the novices. The results in 
the learning phase for novices were consistent 
with some previous studies (Jalilehvand, 
2012; Strother and Ulijn, 1987; Mehrpour 
and Riazi, 2004) which showed a non-
significant effect of text length on reading 
comprehension. 
The results do not accord with McNamara 
et al’s (1996) data. Even though the expanded 
version had extra seven sentences explaining 
more about dinosaurs’ extinction, this 
addition seemed not enough to fill the gap 
between novice and experts’ background 
knowledge and the content of the text. One 
reason might be that English is the mother 
tongue of high school students in McNamar