Content uploaded by Stephen L. Walter
Author content
All content in this area was uploaded by Stephen L. Walter on Feb 06, 2015
Content may be subject to copyright.
Page 1
Mother Tongue-based Education in Developing Countries: Some emerging insights
Stephen L. Walter
GIAL/SIL
Executive Summary
Language is an integral feature of educational practice in the classroom. Teachers communicate
content and instructions via language. Learners listen and interact via language in the process of learning.
The normal assumption is that the language of the classroom is either well-known to or quickly learned
by students so that educational outcomes meet specified expectations. When this assumption is valid,
evaluation of educational effectiveness can largely ignore the issue of language as a variable shaping
educational outcomes. But what if the assumption is NOT valid? Can we still ignore language as a
significant variable in understanding the performance of educational systems?
In many countries—both developed and developing—a significant portion of the school-aged
population enters school NOT speaking the language of the classroom. In some cases these are the
children of immigrants or other linguistic minorities. In other cases, the use of a colonial language for
educational purposes means that all or virtually all children begin school in a ―foreign‖ language.
Furthermore, such children only succeed in the classroom to the extent that they quickly master the
second language.
The last thirty years has seen slow but steady growth in doubts about the merits of the second
language model of instruction. In a number of cases, these doubts have been accompanied by or based on
experimental programs designed to test the potential of an alternative educational model in which the
mother tongues of children are used as languages of instruction in the classroom. This paper examines
some of the research which has emerged in the last ten years comparing, where possible, the educational
outcomes of similar populations (usually immigrants or linguistic minorities) participating side by side in
the two models. The intent has been to look beyond a comparison of cognitive outcomes in an effort to
identify other consequences—sometimes unanticipated—of the mother tongue model.
Specifically, the paper examines data from six different initiatives in mother tongue education from
five different countries. From the data, a number of suggestive insights about the nature, viability, and
impact of mother tongue educational models are identified and briefly explored. These deal with (1) the
optimal sociolinguistic conditions for effective mother tongue education; (2) the time taken to produce
good readers; (3) the function of teachers in mother tongue programs; (4) the impact of the mother tongue
model on developing facility in the second language or language of wider communication; and (5) an
examination of some of the possible impacts of mother tongue educational models.
Page 2
Introduction
It would seem imminently commonsensical to assert that, in the context of formal educational
settings, children are going to be most effectively taught when both children and teachers speak well the
language of instruction. That, after all, is the prevailing model in most developed countries where one
generally finds effective educational systems producing solid results.
This model, however, is NOT the prevailing model in many developing countries. Rather, it is
common to find such countries committed to the use of some exogenous or colonial language such as
English, Spanish, or French as the primary language of education from the earliest years of formal
schooling. Very often in such countries, classroom teachers have only passable proficiency in the
designated language of instruction while children speak little or none of this language upon entry into the
first year of schooling. A casual observer could be forgiven for asking whether such a model has any
chance of success. Even so, this model persists with enthusiastic support from national educational policy
makers and much of the international community. If this model were really effective, then it would seem
reasonable to ask—somewhat tongue in cheek—why we in the US have not yet decided to adopt Spanish
(or Russian or French) as our language of instruction from kindergarten on. Or why Japan has not yet
decided to use Korean as its official language of instruction in school. Or why Argentina does not use
English as its national language of instruction.
The purpose of this paper is not to retrace the history of how or why developing countries made
choices regarding language(s) of instruction in the classroom. Those choices have already been made and
―educational normality‖ in such cases is now defined in terms of the educational status quo. What we
(broadly defined) must wrestle with now are the consequences of those choices and, more specifically,
questions related to an ―alternative normality‖; i.e., is there an alternative model (of language usage in the
classroom) that might have such positive implications that it must be seriously considered as the basis for
a shift in policy?
There is already a vast literature reporting the results of research documenting the ―failings‖ of
educational programs in developing countries. In much of this literature the ―theoretical framework‖ or
evaluative lens underlying the research reported is that of classical measures of educational outcomes
(attrition rates, promotion rates, persistence rates, test scores, etc.) based on standards derived or inferred
from educational experience in developed countries. In this paper, a somewhat variant ―evaluative lens‖
Page 3
is used as a means of better understanding the nature of educational systems in developing countries.
This lens is typically referred to as mother tongue education or, more specifically, mother tongue based
multilingual education.
The present paper examines nearly a decade of research on mother tongue education experiments and
programs (mostly) in developing countries in search of insights on the interaction between linguistic
complexity and diversity on the one hand, and educational outcomes on the other. Data from Eritrea,
Cameroon, the Philippines, Guatemala, and the US will be cited. The paper is not meant to be polemical
in nature though the reader might be led to think so after reading the paper.
A Few Key Terms and Concepts
Mother tongue. In the context of formal education, the term mother tongue is normally used to refer
to the language a child learned first and usually speaks best. In a very high percentage of cases, the ―first
language a child learns‖ and the ―language a child speaks best‖ are one and the same. One can, of course,
uncover sociolinguistic settings in which these two are not the same. Such cases often involve complex
diglossia due to migration or resettlement.
Multilingual education. Prototypically, the term multilingual education refers to classroom or
school-wide instructional strategies in which multiple languages are used for instructional purposes.
There is a plethora of such strategies and continuing controversy as to which is to be preferred and under
what circumstances. When this term or model of education is used with reference to educational models
in developing countries, it almost always refers to a model in which initial instruction begins in children‘s
mother tongue with accompanying preparation to learn a second language which will serve as a language
of instruction in later years of school after sufficient mastery has been developed.
Second language medium. In this paper, the term or phrase, second language medium will be used
to refer to educational models in which children receive most or all of their instruction in a language
other than their first language or mother tongue. In the literature of bilingual education, this is referred
to as submersion (and incorrectly by many as immersion).
Data sources
The data which serve as a basis for the material presented in this paper are the following:
Page 4
1. The Eritrea National Reading Survey – 2002 (Eritrea)
At the initiative of the Ministry of Education in Eritrea, this survey was carried out in an effort to identify
areas of strength and weakness in educational delivery in the primary schools of that country. The focus was
on reading and the development of skill in English including both reading and writing. For this purpose 5
schools were chosen from each of 8 different language communities for assessment. In each school, 20
children were randomly chosen from Grades 1, 3, and 5 for assessment of their skills in reading and writing
their mother tongue. In addition 10 children were randomly chosen from Grades 2, 3, and 5 for assessment of
the development of their skill in reading and writing English, the language of higher education in the country
beginning with middle school. All testing was done by the staff of the Curriculum Unit of the Ministry of
Education. A research report (Walter and Davis, 2005) contains the results of this survey including formal
recommendations made to the Ministry of Education after the survey was completed.
2. The Kom Experimental Pilot Project in multilingual education (Cameroon)
Launched in 2007, this project was designed as a test of the mother tongue model of primary education. In
consultation with education officials of the area, 12 experimental schools were selected for implementation of
the experimental model and these were matched with 12 similar schools which continued to use the standard
educational model of the area (English as the only medium of instruction for all subjects.). Assessments of
educational outcomes are administered to all students in both programs with the assessments rendered and
administered in the respective language of instruction. In addition, assessments of oral proficiency in English
were administered to all students in Grades 1 and 2. The program is now in its fourth year although mother
tongue instruction ceases after the third year. The respective cohorts include approximately 300 children in
each model in each grade. Assessments are carried out by native speakers of the Kom language.
3. National Educational statistics from the Ministry of Education in Guatemala
The Ministry of Education in Guatemala made available the data from their national database for all primary
schools in the country for the period 1991 to 1999. The data set carries an indication of which schools are
bilingual and which are not. In addition, national census data provide a detailed breakdown of the population
by ethnicity at the level of municipios, a political administrative unit approximately equivalent to a county in
the US. The data includes information by grade and by gender on enrollment, promotions, repetitions, and
dropouts for every school. Altogether there are approximately 104,000 records in the database with a record
being the data for one school for one year.
To be forthright, it should be noted that it is very widely reported that many schools designated as being
bilingual do not actually use the mother tongue of the enrollees as a medium of instruction thus affecting the
interpretability of findings.
4. The Lubuagan First Language Component Program (the Philippines)
This program is a 15 year experiment in using the local language as a medium of instruction in the schools of
the Lubuagan District in Ifugao province of northern Luzon. Data from formal assessments is only available
for the last 4 years. The program is also small—3 experimental schools and 3 control schools. The small size
of the experimental program reflects the limited population of the area.
5. The Ixil Experiment (Guatemala)
A limited amount of data has been drawn from a very small but interesting experiment in mother tongue
education in two private schools in a highly monolingual region of Central Guatemala.
Results from the research of Thomas and Collier will also be referenced in this paper.
The binary choice myth
Both proponents and critics of the strategy of providing education via the mother tongue are tempted
to talk about (or against) the use of mother tongues in education as though these are co-identical with
Page 5
one‘s ethnic identity and are similarly immutable. For example, the existence of an ethnic enclave in a
small town might be construed as a basis for needing to offer mother tongue education in schools in that
enclave even though a sociolinguistic study of patterns of language usage might well find that most of the
children are quite or very bilingual. Conversely, the presence of a minority population living scattered
within an area where another language predominates leads some to suggest that mother tongue education
is inappropriate because such a model would be biased against the smaller minority. Consider the
following data. Figure 1 compares the efficiency rates of bilingual and non-bilingual schools in
Guatemala for the years 1991-1997. Efficiency, in this case, is defined as the ratio between the cohorts of
children who enrolled in Grade 1 in 1991 and 1992 and who graduated from Grade 6 in 1996 and 1997.
Figure 1. Comparison of efficiency rates between bilingual and non-bilingual schools across a range of
ethnic densities (prepared from national educational statistics from Guatemala, 1991-1999).
In Figure 1 we observe several discontinuities. First, efficiencies for both types of schools are
significantly higher for the category of lowest ethnic density (0 – 49.99). Second, the disparity in
efficiency between the two programs is greatest in the category of lowest ethnic density. Third, both
program types experience a significant drop in efficiency when the ethnic density exceeds 50 percent.
Fourth, as the density level gets higher, the rates of efficiency shift towards the bilingual schools actually
Page 6
showing an advantage for the bilingual schools when the schools are located in areas of very high ethnic
density (95-100 percent ethnic population).
Clearly there is a functional difference between the role of mother tongues in areas of low ethnic
density compared to areas of higher ethnic density. One interpretation of Figure 1 would suggest that
ethnic children in areas of low ethnic density are negatively affected by attending a bilingual school.1
There are any number of sociolinguistic, psychological, economic and other factors which help explain
the disadvantage observed for bilingual schools in areas of low ethnic density.
Conversely, in areas of high ethnic density, the evidence of Figure 1 suggests that bilingual schools
exhibit an advantage in efficiency over Spanish-medium schools in the same area. Again, we cannot be
certain from the data that this is due solely to the use of mother tongues as languages of instruction.
However, the data suggests rather clearly that as the density level goes up, the bilingual schools show a
gain in efficiency while the Spanish-medium schools show a steady loss of efficiency. The role of local
languages—including their potential value to the educational enterprise--seems to become more salient as
ethnic density increases (accompanied presumably by a greater likelihood of monolingualism).
Another type of evidence is found in what is commonly cited as an insurmountable obstacle to mother
tongue education—the mixing of ethnic populations (and languages). The argument can be stated rather
simply. If mother tongue education is a good educational model but 3 different ethnic groups live side by
side in a given educational region, then this model is impossible. Using three different languages as
languages of literacy in a single classroom is not feasible. If one language is chosen as THE language of
instruction and literacy, then the other two are going to be disadvantaged.
The fallacy here is an incorrectly assumed isomorphism between ethnic identity and language
proficiency. In many such settings, some language emerges as the common language and this language,
whatever it may be, becomes, functionally, the ―mother tongue‖ of the broader community. The
1 It is possible, of course, the ―bilingual‖ schools in areas of low ethnic density do not make extensive use of the
mother tongue in instruction.
Page 7
following case—with data—comes from Cameroon. In this setting, we find a relatively small population
of Fulfulde-speaking people living in the middle of a larger population group speaking a language called
Kom. The area in focus is rural and mountainous and 50+ kilometers from the nearest town of any size.
In this kind of setting, a mechanical application of a mother tongue strategy for education would state that
the Fulfulde-speaking children should go to a Fulfulde-medium school while the Kom-speaking children
went to Kom-medium schools. Cameroonian educational officials, noting the presence of Fulfulde-
speaking children in the area, have cited this as a reason why a mother tongue education model was not an
option in such a setting. In actuality, however, Fulfulde children are attending both Kom-medium as well
as English-medium schools. Tables 1 and 2 summarize the immediate, and somewhat surprising,
consequences.
Table 1. Comparison of Grade 2 Fulfulde-speaking and Kom-speaking children on a range of test measures
of academic performance under two different instructional models (data from Cameroon 2010).
Performance Under Two Different Languages of Instruction
Language Arts
Math
Oral English
Overall
English
Kom
English
Kom
English
Kom
English
Kom
Fulfulde
30.0
57.5
31.1
72.2
60.8
87.5
42.8
73.8
Non-Fulfulde
27.3
62.1
25.9
66.5
52.8
55.7
37.3
60.7
Table 2. Comparison of Grade 3 Fulfulde-speaking and Kom-speaking children on a range of test
measures of academic performance (data from Cameroon 2010)
English-medium Testing
Kom-
medium
Testing
English-
medium
Instruction
Kom-
medium
Instruction
English-
medium
Instruction
Kom
Medium
Instruction
English-
medium
Instruction
Kom-
medium
Instruction
Kom-
medium
Instruction
Language Arts
Math
Overall
Language
Arts
Fulfulde
21.0
44.0
24.1
61.0
22.1
50.1
62.9
Non-Fulfulde
19.9
40.5
21.8
49.0
20.6
43.5
65.7
Page 8
The relevant comparisons in Tables 1 and 2 are those involving Fulfulde-speaking children in
English-medium schools versus Fulfulde-speaking children in Kom-medium schools. We quickly
observe that the Fulfulde-speaking children in the Kom-medium schools substantially outperformed their
Fulfulde-speaking peers in English-medium schools by a factor of 72.4 percent in Grade 2 and by 127
percent. Interestingly, the Fulfulde-speaking children outperformed the Kom-speaking children on all
measures except the Kom-medium Language Arts measures in both grades.
How do we explain this finding? The most logical explanation for this finding is that the Fulfulde-
speaking children are sufficiently bilingual in Kom (and not English) so that they are learning much more
in Kom-medium schools than are those attending English-medium schools. Other possible explanations
include the following: (1) the greater transparency of the Kom orthography—in comparison to English—
which facilitates access to literacy in the Kom language2; (2) a greater commitment to education on the
part of the Fulfulde community; (3) selectively sending only the most able Fulfulde children to school; or
(4) some other unknown biasing factor in education in the Kom area. Further research or documentation
will be required to more precisely identify the observed effect. In the meantime, we can, at a minimum,
observe that the population of Fulfulde-speaking children living in the Kom community is NOT being
compromised educationally by their participation in Kom-medium classrooms. To the contrary, the data
suggest that this population benefits markedly from attending Kom-medium schools.
In sum, two principles seem to be suggested by these data. First, mother tongue strategies for
education are best suited to settings where there is a high degree of both linguistic homogeneity and
monolingualism. Second, children CAN be effectively educated in a second language if and only if they
speak that language well when they begin school.
2 Interestingly, a large number of settings where research has been carried out on the benefits of language of
instructions ‗happen‘ to be ones in which the mother tongue or first language has a more transparent orthography
than the language of wide communication.
Page 9
Good readers in two years
A substantial body of evidence, supported by my own research, indicates that students in second-
language instructional models in developing countries—especially in Africa—require 4-5 years to learn to
read and even after six years, read with low levels of comprehension. When asked, a Grade 6 teacher in
Cameroon noted that about a quarter of his class still could not read at all. When asked what could be
done, he merely shrugged his shoulders and changed the subject.
In contrast, data from three different countries—Eritrea, Cameroon, and the Philippines—is
consistent in demonstrating that good to average students read fluently and with good comprehension by
the end of Grade 2 and even below average students are reading well by the end of Grade 3 when being
taught to read in their mother tongue. Consider the data in Figure 2.
Figure 2. Mapping of comprehension scores by grade and program (language of instruction) from rural
Cameroon (data from Walter and Trammell, 2010).
In Figure 2, the set of red-shaded lines (those running from upper left to lower right) reflects the
comprehension scores of English-medium children on a measure of reading comprehension presented in
0
10
20
30
40
50
60
70
80
none
25 percent
50 percent
75 percent
100 percent
Percent
Level of Comprehension on Reading Assessment
Grade 1 English medium
Grade 2 English medium
Grade 3 English medium
Grade 1 Mother tongue
Grade 2 Mother tongue
Grade 3 Mother tongue
Page
10
English. The set of blue lines reflects the same performance metric for children instructed and tested in
their mother tongue (Kom). In both cases, the intensity of the hue matches progression through school.
Several interesting details stand out from this mapping. First, in the case of the English-medium scores,
more than 70 percent of Grade 1 children demonstrated zero comprehension. In fact, in reviewing the
actual tests, many of the Grade 1 children did not even attempt to answer the questions. One of the school
inspectors suggested this might have been because many of the children (at the end of Grade 1) had never
seen words put on paper before.
A second striking observation is that the curves for Grades 2 and 3 English-medium students are
nearly identical with only 15 to 20 percent demonstrating comprehension at or above the 50 percent level
in each case. Having personally observed some of the baseline testing of these children before the
experimental program began, it is very possible that even this level of ―comprehension‖ is an artifact of
random guessing as the curves for Grades 2 and 3 closely match the predictions of a random guessing
model.
The most striking feature of the curves for the measure of reading comprehension for the children in
the mother tongue program is that they reflect good levels of comprehension even in Grade 1 and change
little thereafter. Clearly, the children are able to read with comprehension after one year of school though
direct observation indicated that they do not read with much fluency or speed after just one year. Why do
the curves not show progression or improvement? Several factors come into play. First, we have found
that approximately 20 percent of children entering Grade 1 (in both programs) are under-age with some as
young as 3 years old. These are clearly not ready for school. Secondly, the tests are graded so that the
material is more demanding as the children progress through the grades. Third, children in neither
program get much experience at reading either in English or in Kom. Therefore, their reading skills fail
to progress as expected in western educational settings.
Figure 3 (next page) provides another window into the development of reading skill in mother tongue
versus second language programs. In this case, the performance of Grade 3 children in Eritrea, all of
whom are taught reading in their mother tongue, is compared with the reading performance of children in
Page
11
Cameroon being instructed in a second language (submersion) program. The four categories of
proficiency specified in Figure 3 can be defined approximately as follows:
Non-reading – no skill at all; unable to identify or read a single word; zero comprehension.
Pre-reading – some familiar words can be identified or decoded; no ability to progress through a text.
Early reading – child can slowly work his/her way through a sentence or short text with numerous
pauses, errors, and restarts. Comprehension tends to be low because of reduced
production skill.
Fluent reading – child can read through a text with little or only minimal effort. While the reading
may be slow and occasionally hesitant, there are limited errors and good
comprehension.
While these categories are not formally defined nor universally accepted, they do provide a useful
means of comparing the emergence of reading skill in the contexts being compared.
Figure 3. Comparison of skill development in reading in mother tongue versus second language
instructional models in two developing countries. (Data from Walter and Davis, 2005; Walter,
2007).
The blue (or solid) bars of various intensities characterize the level of reading skill in children being
instructed entirely in English—a second language for them. Notice especially the very low performance
after Grade 3. It is not until Grade 4 that a substantial portion of children began to show progress in
learning to read though less than 15 percent can be characterized as good readers. By the end of Grade 5,
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
Non-reading
Pre-reading
Early reading
Fluent reading
Percent of Cohort
Level of Reading Proficiency
Grade 3 (N = 270)
Grade 4 (N = 159)
Grade 5 (N = 112)
Grade 6 (N = 46)
Grade 3 Eritrea (N = 763)
Page
12
about 47 percent of children have become fluent readers. Curiously the additional experience of Grade 6
appears to provoke only the most minimal additional improvement.
By contrast, 85 percent of children from Eritrea being taught in their mother tongue are in either the
early reading or fluent reading categories at the end of Grade 3. In sum, the data in Figure 3 seem to
imply that it takes 5 or 6 years in a second language instructional model to approximate the reading skill
developed in 3 years (or less) in a mother tongue instructional model.
What about the teachers?
Obviously, when countries depend primarily on a second language instructional model, they must
field teachers who speak that language well enough to be teachers. In urban areas this is not difficult; in
rural areas, it often is. Conversely, when a mother tongue instructional model is employed, finding
proficient speakers is not an issue though teachers or teaching prospects may have minimal experience in
actually reading and writing their own language. A poorly understood (and minimally researched)
educational issue in this context is the impact upon instruction of speaking well (or not) the language of
instruction. The following evidence offers some hints.
Figure 4. Comparison of performance by schools and by program (Kom Project, Cameroon 2010).
0
10
20
30
40
50
60
70
12345678910 11 12
Schools
Mean Performance by School
Experimental
Control
Page
13
In Figure 4, the mean performance of children on a common test (rendered and taken in English, the
second language of the area) is plotted by program and by school within program from high to low. Two
features of the graph stand out. First, both programs exhibit nearly identical patterns of variation in
teacher (or school) quality. Secondly, the plots are nearly perfectly parallel being separated by a fixed
amount at every point (high performance, average performance, low performance). The underlying
statistics for the two models are the following; mother tongue model (mean = 42.49 with SD = 9.56);
second language model (English): (mean = 20.49 with SD = 8.73).
Figure 4 suggests a fixed gain in performance due to the experimental variable of approximately 100
percent across all schools. The relative advantage or gain is greater for low-performing schools
(approximately 350 percent improvement) due to the very low performance of the poorest second
language schools. The relative gain at the high end is lower being only 75 percent. The use of the mother
tongue as a medium of instruction has improved educational outcomes by a relatively constant amount at
all levels but has not converted weak teachers into good teachers.
Thomas and Collier (1997) hypothesize that children instructed entirely via a second language will, in
general, manifest learning deficits especially with respect to more challenging and difficult academic
content. What happens when those children grow up and become teachers? Do those deficits persist?
Can they be remediated? Will these deficits be passed on to the next generation of children? Over time,
will we observe a negative spiral in overall educational quality due to accumulating deficits? If so, what
steps can be taken to interrupt this negative spiral? A suggestive experiment supporting the hypothesis of
Thomas and Collier was carried out on a very small scale in Guatemala in April 2009. At the request of
the respective headmasters, all teachers in two Ixil-medium schools were tested for their knowledge and
skill in math using a standardized assessment of Grade 6 math (Spanish medium TAKS test from the state
of Texas).
Page
14
Figure 5. Performance of Mayan-speaking teachers on a Grade 6 standardized assessment of math
knowledge and skills (data from an informal experiment in Guatemala, 2009).
The performance of these teachers was cast in terms of percentiles (based on the standard used for the
TAKS test in the state of Texas). Without doubt, this is a somewhat biased scale because of the
international comparison. At the same time, the teachers taking the test had completed their studies in a
pedagogical high school in Guatemala, the normal standard for the training of primary level teachers in
Guatemala. All but one scored at or below the 14th percentile. The low-performing teachers had received
all of their education in second language (Spanish-medium) schools. An examination of group
performance on individual test items demonstrated that the test group performed reasonably well on items
involving basic computations and simple word problems. When test items required multiple
computations, modeling, or abstract thinking, performance declined quite dramatically.
Because of the very small sample size and a simple design, we cannot draw major conclusions from
this data set. Nonetheless, support is lent to the suspicions of many researchers that primary school
teachers in developing countries, themselves, have weak skill sets and are not capable of teaching that
which they do not know or understand themselves. Is there evidence that mother tongue instruction has
any potential to address such systematic deficits?
Page
15
Table 3. Comparative assessments in Grade 3 math from research sites in two countries (data from
Walter and Dekker, 2007; Walter and Trammell, 2010).
Data from Cameroon (n = 541 from 24 schools)
English-
medium instruction
Mother tongue medium
instruction
Statistics
Performance
Gain
Math
21.8
49.5
T=15.09; P < 0.001
127 percent
Data from the Philippines (n=116 from 6 schools)
Math
49.5
76.2
T = 7.25; P < 0.001
54 percent
The data in Table 3 report differential performance on math assessments in two different countries for
children in second language medium classrooms versus those in mother tongue medium classrooms. The
performance gains are substantial ranging from 54 to 127 percent in the case of the Cameroonian
program. An examination of item by item results in the case of the Cameroonian data indicates that the
differential advantage was present (though not equivalent) across all items including those involving pure
computation. Furthermore, the performance of those from English-medium schools was at or below
random guessing on all items but one—distinguishing between odd and even numbers. A possible
interpretation of these results is that virtually no effective learning is taking place because of the
communication problem in the classroom—the teachers do not speak English very well, and the children
have not learned enough English by Grade 3 to use it as a medium of learning.
Presumably, some of the children from both programs will eventually become teachers. Clearly,
those from the English-medium program are already manifesting significant conceptual deficits by the
end of Grade 3. What will such students eventually bring to their roles as teachers? If mother tongue
instruction improves learning outcomes in a content area such as math, then it is possible that it is a
strategy for interrupting the spiral of accumulating learning deficits alluded to earlier.
What about the second language?
Probably the most overriding concern expressed by those engaged in discussions over the issue of
language of instruction is whether the use of the local mother tongue as a medium of instruction is going
Page
16
to compromise progress in learning the second language which is very often the language of higher
education. Parents will say, ―My child already speaks X (the local language). He/she needs to learn Y
(the second language) to do well in school and to get a good job.‖ The well-known and oft-cited research
by Thomas and Collier (1997) was originally launched around this question, ―What is the most effective
strategy for bringing English Language Learners up to speed in English in the US?‖ During the course of
their research, the question shifted to, ―What mix of languages of instruction is going to produce the best
educational result for English Language Learners in the US educational system?‖ Contrary to
conventional wisdom, the empirical findings of Thomas and Collier indicated that children receiving as
much as 6 years of instructional support in their mother tongue not only finished their formal education at
a higher level than those submerged in English-only programs, they also achieved a greater level of
proficiency in English.
The research of Thomas and Collier was based entirely upon data drawn from US schools. Would the
same hold true in developing countries where teachers were less sophisticated and technical support was
much more limited (or non-existent). This question has received careful attention in our recent research.
The experimental project in Cameroon consists of 12 experimental schools in which the mother tongue is
the language of instruction and 12 matched control schools in which English is the language of instruction
(the standard model in northwestern Cameroon). In the experimental schools, children are taught English
as a subject for approximately 3 hours a week. In the control schools, all instruction is in English so
children are exposed to 22.5 hours of English a week (assuming full use of assigned instructional time).
At the end of the year, children in both programs take the same assessment of proficiency in oral English
which is administered one-on-one by trained test administrators. The results of this testing are provided
in Table 3 for all children in Grades 1 and 2 from the inception of the experimental program.
Page
17
Table 4. Comparison of assessments in Oral English from 2008-2010 (data from Cameroon).
Grade 1
2008
2009
2010
English
Kom
English
Kom
English
Kom
Oral English
35.4
52.5
42.8
55.0
40.8
55.5
Grade 2
Oral English
50.8
59.5
53.1
56.1
Much to the surprise of local parents and educators, the children in the mother tongue schools (Kom
language medium) outperformed their peers in the English-medium classrooms where they were listening
to English 20+ hours a week. The English ability of the mother tongue children was not compromised by
the instructional arrangement of the program. In fact, overall, they scored higher than their peers though,
after the first year, the difference is generally not statistically significant.
A comparison of results for Grades 1 and 2 in Table 4 suggests that the gap demonstrated in Grade 1
was largely erased by the end of Grade 2. An obvious question was whether the English-medium children
would surpass the mother tongue medium children in Grade 3 in the development of proficiency in
English. In Grade 3, we tested this ―hypothesis‖ by administering the assessment instrument (a test of
language arts and mathematics) entirely in English to all students in both instructional modalities. We
reasoned that this would provide a more robust test of local conventional wisdom that instruction via
English would produce superior ability in English AND superior educational outcomes. The results are
contained in Table 5.
Table 5. Comparison of results for Grade 3 children in mother tongue versus second language medium
schools (Data from Cameroon, 2010).
English-medium
instruction
Mother tongue medium
instruction
Gain in
Efficiency
Statistics
Mean
SD
Mean
SD
Language Arts
19.9
13.0
40.6
17.0
104 percent
T = 15.97; p < 0.001
Mathematics
21.8
17.5
49.5
24.3
127 percent
T = 15.09; p < 0.001
Overall
20.6
13.1
43.8
17.3
113 percent
T = 17.50; p < 0.001
Page
18
In Table 5 we find that, despite having received virtually all of their instruction in their mother
tongue, children in the experimental schools performed at a level more than twice as high as that of
children from the second language schools even when the assessment was rendered in English. Clearly,
the fact that these children were NOT being taught in English did not compromise their performance
relative to those receiving all of their instruction via English.
What accounts for the results shown in Table 5? The best explanation would appear to be that
originally advanced by Thomas and Collier. Children being instructed via their mother tongue are able to
actually learn the curricular content of instruction. Those in the second language medium classrooms
have such minimal control of English that their ability to learn is blocked by this limitation. While the
children in the mother tongue medium schools clearly were not performing at an optimal level3 when
taking an English medium test—mean scores in the range of 50 percent—the data suggests they have
learned the content plus enough English to be able to demonstrate positive academic achievement.
It is possible to suggest that the children in the English-medium program learned instructional content
in subjects like math and science but were unable to demonstrate that learning via an English-based
assessment instrument. This suggestion is countered by the observation that children in the English-
medium schools also did poorly on the math section of the assessment where nothing but computational
skill was needed to solve problems presented on the test.
Identifying and estimating impacts
Internationally, a great deal of attention has been given to achieving universal basic education. Most
agree, however, that without further education, developing countries will lack a sufficiently skilled
workforce to meet development objectives. Cost and access are two filters which often determine
whether children graduating from primary school will proceed on to secondary. But what about the
variable of language, specifically language of instruction? Walter and Morren (2002) analyzed data from
Guatemala in an effort to determine whether mother tongue medium education increased the likelihood
3 Mean scores on the Kom-medium reading test were in the range of 65 percent.
Page
19
that children would continue schooling beyond the primary level. The methodology was simple: 1,202
Mayan secondary school students were asked whether they received their primary education in second
language or in a Mayan language (mother tongue). The results of the survey are presented in Table 6.
Table 6. Data on continuing on to secondary education in Guatemala (Walter and Morren, 2002).
Type of Primary School Attended
Survey results
Expected result based on demo-
graphic and educational statistics
Spanish -medium school
620
810
Mayan -medium school
582
392
Almost 50 percent of the Mayan students surveyed indicated that they had graduated from a bilingual
school, even though only about 33 percent of all Mayan children attend such schools4. These data
provide strong statistical evidence that in such contexts, receiving L1 (bilingual) schooling increases the
likelihood of going on to secondary schooling (χ2 = 136.66; p=0.000). In this case, attendance of a
bilingual school increased the likelihood of proceeding to a higher level of education by 48 percent. It
would be fascinating to have broader and more longitudinal data to determine the reliability of this
finding.
Policy makers and other observers of influence in developing countries have frequently been heard to
observe, "There is no need for this kind of a model (mother tongue medium education). Look at me! I
came from a poor small ethnic community living in a rural area with poor and uneducated parents. I did
well in school eventually getting a Ph.D. at University X." The same argument has been voiced in
developed countries as well. These little personal case studies have often forestalled any further
discussion of the possible merits of mother tongue based education.
As the research evidence has begun to accumulate, it is now becoming clear that very bright children
can and will be successful under second language instructional models. Apparently, such children have
the intellectual capacity to quickly adjust to the language of instruction well enough to master the
4 It is widely reported that many of the schools designated as ―bilingual‖ in Guatemala‘s educational system do not
actually make use of a Mayan language as a language of instruction. We are not able to establish the extent to which
this might have affected our findings.
Page
20
curricular content being presented. This finding, however, raises a set of provocative questions, 'How
large is this population of gifted children?' and, 'What are the implications for those children who are not
so gifted?'
My initial idea for trying to answer these questions was to develop language-specific assessments of
intellectual ability—IQ tests if you will. The amount of work needed to do this well was
disproportionately large for the specific question which needed to be answered. As a result, I've been
experimenting with alternative strategies. One of these is presented below in Figure 6.
Figure 6. Gain or loss of class standing (based on percentiles) of students in the two programs (Data from
Grade 3 Cameroon 2010).
In Figure 6, students were first ranked against all peers in the same instructional mode—i.e. one
ranking for all children in the English-medium program and one ranking for all children in the mother
tongue-medium program. A necessary assumption is that the distribution of innate intellectual ability is
essentially the same in both programs AND the distribution of ability between the two programs is
essentially the same.
The second step was to combine both groups of students into a single group and then to redo the
ranking using overall test performance as the basis for the ranking. Then a scaling of these ranks was
done based on a standard percentile scale.
Gain (loss) of Percentile Standing
Grade 3 Cameroon (2010)
-40
-30
-20
-10
0
10
20
30
40
99 94 89 84 79 74 69 64 59 54 49 44 39 34 29 24 19 14 9 4
Percentile
Percentiles gained or lost
Experimental
Standard
Page
21
The third step was to compare the within-group rankings with the between-group rankings on the
basis of a percentile scale. The assumption is that any significant change in ranking of individuals going
from the within-group ranking to the between-group ranking will be, in effect, a measure of a gain (or
loss) of educational standing (achievement) attributable to the variable which defines the difference
between the two groupings—in this case, language of instruction.
In the figure, the distance of the two curves above or below the midline is a measure of the number of
percentiles gained (or lost). By dropping or raising a vertical line from any point on the midline, it is
possible to identify the impact of the experimental variable—language of instruction—for those
associated with a given position of academic achievement. So, for example, a line dropped or erected
vertically from the point marked as 54 identifies students who are in the 'average' category (the 54th
percentile). We can then look at the number of percentile positions gained or lost to estimate the extent to
which students at given percentiles benefit (or not) from being in one program versus the other. For
example, if we begin with the 54th percentile and go up vertically, we find that students in the
experimental program experienced a mean gain of 25 percentile positions relative to those in the standard
program. The curve for the Kom-medium program peaks at 28 percentile positions gained for those
between the 24th and 44th percentiles. These data support the hunch of various observers including the
author that those most benefiting from mother tongue instruction are not the bright children but the
average or even slightly below average children.
The strategy used to generate Figure 6 necessarily produces a complementary image of performance
between the two programs. It also necessarily shows less contrast at the extremes since extremely high or
low performers begin at the top or bottom and cannot climb higher (or fall lower). What this graphical
device does show, however, is more precisely where benefit occurs in the research design and something
of the extent of this benefit. The greatest benefit in the Kom-medium program was 28 percentiles. How
much of a gain is 28 percentiles? If a person is at the 50th percentile (a totally 'average' student) and
moves up 28 percentiles, that student is transformed from one who probably will not pass the Class 6 Exit
Exam to one with a strong likelihood of being able to enter and complete secondary.
Page
22
While the data in Figure 4 clearly show marked impacts on educational outcomes based on language
of instruction, an even greater impact can be derived by considering the consequences of their
performance on future work and career options. Table 7 is based on the findings of Thomas and Collier
from the US. Using their mean scores for each group—submersion, transitional mother tongue education,
developmental mother tongue education, etc.—and the statistical characteristics of the standardized test
used for their research, a map can be constructed which compares the career implications of the various
models. Table 7 contains the results of this analysis comparing only the weakest (English submersion)
and strongest (two-way developmental) models
Table 7. Educational implications of the Thomas and Collier model in terms of differential outcomes (Walter, 2003).
Standard
Deviation
NCE
equiv.
ESL Pullout (Second
language instruction
only)
Two-way
Developmental
(mother tongue
model for 6 years)
Social/Professional/economic
potential
Percent
of pop.
No. out of
10,000
Percent
of pop.
No. out of
10,000
Above 2nd
>92
.06
5.5
7.0
699
Researchers, scientists, top writers,
top intellectuals, medical doctors
1st to 2nd
71-92
1.2
120
24.7
2,471
Professors, business leaders,
professionals, journalists
Mean to 1sr
50-71
9.5
950
38,3
3,828
Teachers, mid-level managers,
engineers, programmers, bureaucrats
-1st to Mean
29-50
29.8
2,977
23,6
2,364
Skilled factory workers, equipment
operators, clerical, service workers
-2nd to –1st
8-29
37.1
3,712
5,8
580
Blue collar workers, manual laborers
Below –2nd
<8
22.4
2,236
.6
58
Hard to employ, domestics, menial
labor
The second column (NCE equivalent) divides the range of performance on the instrument of
measurement into blocks corresponding to standard deviation (21 in this case). Assuming that the
performance of the two cohorts being compared was normally distributed, the number of test-takers out of
a hypothetical block of 10,000 students is computed for each level of performance. Among students who
participated in the ESL pullout model (no instructional support in their first language), barely 10 percent
scored 50 percent or higher—a level that could be considered minimal for going on to college.
Performance below this level would be associated with blue-collar jobs, manual labor or unemployment.
Page
23
In contrast, 70 percent of students educated in two-way bilingual programs scored at or above the mean,
thus giving them potential access to university and higher level professional careers.
This analysis suggests that the workplace and economic consequences of educational language choice
are great, even in the USA where a minority of children are learners of the national language. In low-
income multilingual countries where virtually all children are learners of an official language, the
economic consequences would seem to be even greater.
Conclusion
In a nutshell, the insights gleaned from the data examined can be approximately stated as follows:
1. Patterns of language usage in any given country or region are NOT monolithic. Consequently,
some sociolinguistic settings are much better candidates for mother tongue education than others.
2. The research data is uniform in supporting the claim that mother tongue education programs are
capable of producing capable readers in 2-3 years rather than the 5 reported for many second language
medium programs.
3. The use of the mother tongue as a medium of instruction raises the level of educational outcomes
across the board but does not convert poor teachers into good teachers.
4. The research evidence consistently contradicts the claim that heavy reliance upon the mother
tongue for instructional purposes in early grades will compromise ability to learn and use the second
language. It appears that the gains in mastery of basic educational skills coming from mother tongue
instruction more than compensate for reduced exposure to the second language in a classroom context.
5. The research data suggests that the impacts of mother tongue instruction are substantial and are
measurable both in the short term—specific measures of educational outcomes—and in the long term via
more global measures such as access to higher education and to more advanced career opportunities.
Page
24
6. The research evidence suggests that the greatest beneficiaries of mother tongue education are those
of average ability and potential.
The data on the effectiveness of mother tongue instructional models in developing countries is still
relatively small but growing. This paper has sought to distill useful and applicable generalizations from
the body of extant evidence. The reader will have to come to his or her own conclusions about the
adequacy of the research data as well as the aptness of the generalizations drawn. In the view of the
author the data are sufficiently suggestive to make for some very interesting theorizing about the role of
language of instruction as a variable in formal primary education. Even further, visible and consistent
implications for policy makers are beginning to emerge.
Page
25
References
Thomas, W. and Collier, V. 1997. School effectiveness for language minority children.
http://www.ncela.gwu.edu/files/rcd/BE020890/NCBE_School_Effectiveness.pdf Last accessed
14 July 2010
Thomas, W. and Collier, V. 2002. A National Study of School Effectiveness for Language Minority
Students’ Long-Term Academic Achievement. Center for Research on Education, Diversity and
Excellence, Santa Cruz CA. http://crede.berkeley.edu/research/crede/research/llaa/1.1_final.html
Last accessed 5 August 2010
Walter, S. 2003. ‗Does language of instruction matter in education?‘, in Wise, Mary, Headland, Thomas
and Brend, Ruth (eds) Language and Life: Essays in Memory of Kenneth L. Pike,. Dallas TX:
SIL International/University of Texas 611-635
Walter, S. 2007. ‗Preliminary Report of Some Findings on Primary Education in Boyo Division‘,
Unpublished research report. Available from the authors or SIL Cameroon
Walter, S. 2008. ‗The language of instruction issue‘, in Spolsky, Bernard and Hult, Francis (eds) The
Handbook of Educational Linguistics. Malden MA: Blackwell. 129-146
Walter, S. and Trammell, K. 2008. ‗Results of the First Year of the Kom Experimental Education
Project‘, Unpublished research report. Available from the authors or SIL Cameroon.
Walter, S and Trammell, K. 2009. ‗The Kom MLE report for 2009‘, Unpublished research report.
Available from the authors or SIL Cameroon.
Walter, S. and Trammel, K. 2010. ―The Kom Experimental Mother Tongue Education Project Report for
2010.‖
Walter, S and Davis, P. M. 2005. Eritrea National Reading Survey. Dallas, SIL International
Walter, Stephen L. and Ronald Morren. (2004) ‗Twenty years of bilingual education in Guatemala‘,
Paper presented to the GIAL Academic Forum, Dallas, TX
Walter, S. and Dekker, D. 2007. ‗The Lubuagan mother tongue education experiment (FLC): a report of
comparative test results‘, Unpublished research report. Available from the authors
Walter, S. and Dekker, D. 2008. ‗Mother Tongue Instruction in Lubuagan - Results for 2008‘,
Unpublished research report. Available from the authors.
