The effect of language complexity and group size on knowledge construction: Implications for online learning

Introduction
As  the  Internet  is  increasingly  growing,
online  education  continues  to  grow  too
(Johnson & Aragon, 2003), a phenomenon
expected  to  continue  at  a  significant  rate
(Allen  &  Seaman,  2004).  Online
discussion  forums,  or  Computer  Mediated
Discussions,  are  popular  with  educators
who  aim  at  using  IT  (Information
Technology)  to  enhance  the  quality  of
learning.  The  use  of  computer-mediated-communication  tools  can  present  new
ways  to  promote  knowledge  construction
(Schellens  &  Valcke,  2006).  Computer-mediated-communication  tools  can  help
make the construction of knowledge easier
by  working as a social  medium to support
students’  learning  by  representing
students’  ideas  and  understandings  in
concrete  forms  (e.g.,  notes)  so  that  ideas
can  be  more  developed  via  social
interactions  (e.g.,  questioning,  clarifying)
(Van  Drie,  Van  Boxtel,  Jaspers  &
Kanselaar,  2005).  One  example  of  such
tools  is  the  asynchronous  discussion
forum.  The  technology  which  is  available
in  asynchronous  online  discussions
provides  a  number  of  ways  to  foster  the
construction  of  collaborative  knowledge,
while  asynchronicity  offers  learners  the
opportunity  to  interact  at  any  time  from
any  place (Scardamalia  &  Bereiter, 1994).
The  debate  could  be  described  as  a
constructive  learning  environment  which
offers  multiple  approaches  and  actual
world examples of the topic of discussion,
that  encourages  reflection,  and  that
supports  collaborative  construction  of
knowledge  via  social  negotiation
(Jonassen, 1994).  
 
Early  analyses  of  computer-mediated
communication  using  asynchronous  tools
tended to concentrate more on quantitative
 
analysis  of  the  data,  especially  on  word
counts  and  number  of  postings.  Yet,
although  this  method  of  analysis  provides
a  survey  of  the  interactions  which  occur
online,  it  does  not  take  into  consideration
the  content  of  what  is  posted  on  the
discussion  boards.  The  analysis  of  the
content  of  the  discussion  boards,  thus,
moves  towards  a  more  semantic  labeling
of  propositions  (Donnelly  &  Fitzpatrick,
2010).  The  assessment  of  co-construction
of  knowledge  based  on  quantitative
analysis  of  postings  underestimates  the
complexity  of  the  available  issue.
Although a quantitative analysis allows the
researchers  to  understand  some  linguistic
online  behaviors,  it  does  not  allow  deep
investigation  of  the  language  complexity
in  order  to  pinpoint  the  collaborative
learning  among  learners.  Thus,  linguistic
models  for  a  qualitative  analysis  of  online
discourses have been elaborated by several
researchers;  for  example,  Interaction
Analysis  Model  by  Gunawardena,  Lowe
and Anderson (1997).  
 
More  recently,  some  researchers  have
examined if group size might influence the
levels of knowledge construction in online
discussion  forums.  Schellens  and  Valcke
(2006), for example, found that discussion
in groups of about 10 participants resulted
in larger proportions of advanced levels of
knowledge construction. Hew and Cheung
(2010)  examined  if  there  was  any
relationship  between  the  frequency  of
advanced  level  knowledge  construction
occurrences  and  group  size.  The
researchers  found  a  significant  positive
correlation  between  the  discussion  group
size  and  the  frequency  of  advanced  level
knowledge  construction  occurrences.
However,  no  indication  was  provided  by
Hew and Cheung (2010) about the optimal
group size.
 
In fact, no research study investigating the
impact  of  language  complexity  on
knowledge  construction  in  online
conversations has been reported. Language
complexity refers to the lexical variation of
a  given  text.  Consequently,  this  study
makes  an  endeavour  to  provide  some
evidence that seems to be urgently needed.
This  paper  addresses  the  effect  of  group
size  and  language  complexity  on
knowledge  construction  in  online  debates
and  tries  to  ask  these  two  research
questions:  Is  there  a  significant
relationship  between  knowledge
construction  and  group  size  in  online
debates?  And,  is  there  a  significant
relationship  between  language  complexity
and  knowledge  construction  in  online
debates?
 
The study
The  goal of this study  was to build on the
current  literature  through  exploration  of
how  group  size  impacts  participants’
construction  of  knowledge  within  a
primary asynchronous environment. It also
tries  to  investigate  the  impact  of  language
complexity  on  knowledge  construction.
This  study  is  a  longitudinal  case  study
because the data source is bounded by time
and environment (Creswell, 1998).
 
Variables of the study
Knowledge construction
Knowledge  construction  refers  to  phases
of interaction in the online debates. Phases
of  interaction  were  identified  using
Gunawardena  et  al.’s  (1997)  Interaction
Analysis Model.  
 
Group size  
Group size of an online debate refers to the
number of participants who were involved
in  the  conversations.  Two  main  forms  of
participation  are  identified  in  an  online
discussion  environment:  writing  and
reading  (Hewitt  &  Brent,  2007).  In  this
research  paper,  the  focus  is  on  the  writing
form  of  participation  because  writing  is
closely  linked  to  discussion,  and  it  is  of
greater  importance  than  reading  (e.g.,
when  the  student  is  answering  postings
from  an  existing  discussion  thread)
(Guzdial  &  Turns,  2000).  Moreover,
writing  is  a  more  observable  phenomenon
than  reading.  In  Debate  A,  group  size  is
equal  to  326  whereas  in  Debate  B,  group
size is equal to 118.
 
Language complexity
Language  complexity  (LC)  variable  is
determined  by  type  token  ratio  (TTR),
which is a measure of vocabulary variation
within a written text or a person’s speech.
The type-token ratio has been shown to be
a helpful measure of lexical variety within
a  text.  The  number  of  words  in  a  text  is
often  referred  to  as  the  number  of tokens.
However,  several  of  these  tokens  are
usually  repeated.  As  long  as  there  is  only
one type of word, the relationship between
the  number  of  types  and  the  number  of
tokens  is  known  as  the type  token  ratio
(TTR) (Williamson, 2009).  
 
A  high  TTR  indicates  a  large  amount  of
lexical  variation  and  a  low  TTR  indicates
relatively  little  lexical  variation
(Williamson,  2009).  The  following  table
features the different TTR levels:

Informants
Informants  of  the  study  are  444  online
debaters  participating  in  two  online
debates:
 
A: 326 debaters participating in the online
debate:  “Technology  in  Education”
retrieved from:  
 
http://www.economist.com/debate/days/vie
w/244,  
 
B: 118 debaters participating in the online
debate  “Internet  Democracy”  retrieved
from:  
http://www.economist.com/debate/days/vie
w/662.
 
Online debates sampling
The  first  online  debate  is  entitled
“Technology  in  education”  and  was
retrieved  from  the  website  “The
economist.com”  on  March  18th,  2011.  It
was carried over 11 days from the 15th till
the  26th  of  October  2010  and  comprised
371  comments.  It  was  coded  Debate  A.
The  second  online  debate  is  entitled
“Internet  Democracy”  and  was  also
retrieved  from  the  website  “The
economist.com” on April 13
th
, 2011. It was
carried  over  10  days  from  the  23rd
February  2010  till  the  4th  February  2010
and  comprised  128  comments.  It  was
coded Debate B.
 
Interaction Analysis Model
The  informants’  online  transcripts  were
analyzed  qualitatively  using  Gunawardena
et  al.  (1997)  Interaction  Analysis  Model
(IAM).  The  analysis  is  based  on  the  five
phases  of  knowledge  co-construction  that
usually occur during online debates.
 
Gunawardena et al. (1997)  stated
that postings coded  Level  I  and II
“represent  the lower  mental  functions”,
while postings  coded  level  III,  IV,  and  V
represent the higher mental functions:
 
 
a)  Level  I  –  making  statement  of
observation  or  opinion,  statement
of agreement among participants;  
 
b)  Level  II  -  identifying  areas  of
disagreement, asking, or answering
questions to clarify disagreement;
 
c)  Level III - negotiating the meaning
of  terms,  ideas/co-construction  of
knowledge;  

d)  Level  IV  -  testing  of  proposed
synthesis  or  construction  against
existing  literature  or  personal
understandings, experiences; and
 
e)  Level  V  -  summarizing
agreement/statements  that  show
new  knowledge  construction,
application  of  newly  constructed
ideas.  In  this  study,  we  defined
advanced  levels  of  knowledge
construction as levels II, III, IV, or
V of the model.  
 
Procedure
To apply the Interaction Analysis Model, I
read  the  postings  in  the  original  sequence
and  assigned  them  one  or  more  phases
from  the  IAM.  It  is  possible  to  code
multiple  sentences  or  a  paragraph  or  two
with a single phase; this is consistent with
the  original  application  of  the  IAM
(Gunawardena  et  al.,  1997).  I  calculated
the  frequencies  of  the  coded  phases  for
each  posting  and  for  each  informant.  Two
raters,  myself  and  an  English  assistant
colleague,  coded  the  online  transcripts.  In
order  to  conduct  inter-reliability  checks,  I
used  the  most  advanced  phase  from  each
posting  as  the  basis  for  inter-rater  checks
(Beaudrie,  2000).  Inter-rater  differences
were addressed following Chi (1997).
 
Postings were coded using the five phases
of  Gunawardena  et  al.  (1997).  For
statistical correlation, Phase I was coded 1,
phase II was coded 2, phase III was coded
3, phase IV was coded 4 and phase V was
coded 5. The ‘absence of phase’ was coded
0.  A  second  researcher  reviewed  the
coding  of  the  total  postings  in  debate  A
and  B.  The  inter-rater  was  selected  based
on  her  field  of  specialization,  applied
linguistics,  and  her  familiarity  with
discourse  analysis.  The  inter-rating
training  consisted  of  an  independent
review  of  the  Interaction  Analysis  model.
Her task was to review the coding made by
the  investigator.  It  was  easy  to  achieve  an
agreement  of  100%  because  coding
disagreement concerned only 3 postings in
Debate  B.  Total  agreement  was  achieved
after discussing discrepancies.
 
“TextMaster”  was  downloaded  from  the
Internet.  “TextMaster”  is  a  software  tool
for rapid analysis and processing of fixed-length  files.  This  software  counts  the
number  of  tokens  and  types.  Each  posting
was copied and entered in “TextMaster” to
obtain  the  number  of  tokens  and  types.
TTR  was  then  processed  for  each  posting
through  dividing  the  number  of  types  by
the  number  of  tokens.  The  value  obtained
is referred to as  language complexity. The
mean  language  complexity  was  processed
for  the  participants  who  sent  two  postings
or  more.  Numerical  data  of  TTR  was
turned  into  categorical  data  in  order  to
process  statistical  analyses.  Values
belonging to the very low TTR levels were
coded 1. Values belonging to the low TTR
levels  were  coded  2.  Values  belonging  to
the  average  TTR  levels  were  coded  3.
Values belonging to high TTR levels were
coded  4.  Values  belonging  to  very  high
TTR levels were coded 5.
 
The study investigated two online debates.
Debate  A  comprises  326  participants  and
Debate  B  comprises  118  participants.
Group  size  in  Debate  A  was  coded  1  and
group  size  in  Debate  B  was  coded  2.  The
statistical  data  analysis  was  based  on
descriptive  and  analytical  statistics.
Descriptive  statistics  were  used  to
calculate  means  and  percentages  of  the
selected  variables  of  the  study  which  are
language  complexity,  knowledge
construction,  and  group  size.  Correlation
analysis  was  used  to  describe  the
relationship  between  the  different
variables.  Spearman’s  Rho  correlations
were computed between different variables
-  language  complexity,  group  size  and
knowledge  construction  -  to  detect  any
relationship  between  them.    Simple
regression  analyses  were  computed  on
dependent  and  independent  variables  to
determine  the  significant  predictors  of

knowledge  construction.  Multiple
regressions  analyses  were  computed  on
dependent  and  independent  variables  to
confirm simple regression results. The data
were  computed  using  the  statistical
Package  for  the  Social  Sciences  (SPSS)

Findings
Table  2  reveals  that  in  Debate  A  the
relationship  between  language  complexity
and knowledge construction is positive and
highly  significant  at  the  0.01  level  of
significance. It also shows that in Debate B
the  relationship  between  language
complexity  and  knowledge  construction  is
positive and significant at the 0.05 level of
significance.  These  results  imply  that  the
higher  the  language  complexity  is,  the
higher  the  knowledge  construction  would
be.

Table  3  shows  that  the  relationship
between  knowledge  construction  and
group  size  is  negative  and  highly
significant  at  the  0.01  level  implying  that
the  less  important  group  size  is,  the  more
important  knowledge  construction  would be.
  

Table  4  shows  that  language  complexity
has  given  non-significant  results  in  the
regression  equation  for  knowledge
construction in Debate A. However, Table
5  reveals  that  language  complexity  is  the
most  consistent  predictor  of  the  variation
observed  in  knowledge  construction  in
Debate  B.  It  accounts  for  5.1  %  of  the
observed  variation.  The  regression
equation  is  significant  as  shown  by  the  t-value and the F-ratio.

Table  6  reveals  that  group  size  gives  non-significant  results  in  the  regression
equation  for  knowledge  construction.
Consequently,  group  size  is  not  a
significant  predictor  of  knowledge
construction.

Table  7  shows  that  when  group  size  is
added to language complexity in the same
regression  equation  for  knowledge
construction,  the  adjusted  R²  falls  from
4.3%  to  2.8%.  Since  the  t-value  is  not
significant  for  the  two  variables,  group
size does not help knowledge construction.
Consequently, the best regression fit is the
simple  regression  of  language  complexity
for knowledge construction in Debate B.

Discussion
In  both  debates  the  results  show  that
language  complexity  and  knowledge
construction  are  significantly  correlated.
Correlation  is  positive  and  highly
significant  in  Debate  A  and  positive  and
significant  in  Debate  B  suggesting  that  an
increase  in  language  complexity  generates
an  increase  in  knowledge  construction.  
This  finding  implies  that  using  rich  and
complex  vocabulary  results  in  consistent
conversations  which  tend  to  engender
various  ideas,  opinions  and  viewpoints.
Consequently,  this  could  promote
negotiation  and  higher  order  thinking.
Furthermore,  findings  show  that  language
complexity  is  a  significant  predictor  of
knowledge  construction.  Thus,  generating
a  high  lexical  variation  may  foster  high
levels  of  knowledge  building.  Therefore,
educators  should  mainly  focus  on
techniques  that  promote  vocabulary
richness.  
 
Besides,  students’  participation  may  vary
according  to  the  mastery  of  the  language
used.  Many  learners  may  feel  some
difficulties  when  communicating  in  their
second  or  foreign  language  which  implies
that asynchronous online environment may
be  an  effective  tool  in  evaluating  the
students’  language  proficiency.
Furthermore,  some  actions  should  be
undertaken  to  help  learners  enhance  their
language level such as undertaking reading
and  writing  sessions.  The  stress  should  be
placed  on  English,  which  is  an
international  language.  Participating  in
such  debates  using  the  second  or  foreign
language  would  be  an  efficient  practice.
Online  communication  environments  are
empowering tools for non-native speakers.
 
In  order  to  promote  rich  and  consistent
online  conversations,  students’  online
participation  should  be  fostered.  Different
roles can be attributed to students. Some of
them can play the role of moderators. They
may  be  fight-flaming  and  stop  altercation,
though.  Others  should  have  the  role  of
summarizers,  summarizing  long  and
frequent  postings  in  order  to  facilitate  the
interaction.  A  group  of  participants  may
also  find  appropriate  theories  to  back  up
informants’  statements,  thus  playing  the
role  of  theoreticians.  Giving  such
responsibilities  to  students  will  not  only
facilitate  communication  but  will  also
stimulate them to participate actively in the
discussion,  promoting,  therefore,  language
complexity and knowledge construction.            
 
The  results  also  revealed  that  the
correlation  between  group  size  and
knowledge  construction  is  negative  and
highly  significant.  This  implies  that  high
levels  of  knowledge  construction  are
achieved  by  informants  participating  in
smaller  forums.  These  findings  contradict
the  ones  reported  in  Hew  and  Cheung
(2010).  In  fact,  allowing  for  an  ongoing
increase  in  the  discussion  size  may  have
several  limitations.  First,  it  may  result  in
‘reading without writing’ on the part of the
participants.  Second,  large  groups  or
conversations  require  huge  cognitive
efforts  from  the  participants  to  react  to
others.  This  could  result  in  reading
boredom.
 
Hew  and  Cheung  (2010)  suggest  a  group
size  of  about  10  participants  in  order  to
form a  critical mass to lead the discussion
to  advanced  levels  of  knowledge  building
(P.431).  Students’  group  size  should  be
limited  in  order  to  avoid  learners’
exhaustion  and  withdrawal  from  the
debate.  In  fact,  a  big-sized  group  often
results  in  a  big-sized  conversation;  and
students  would  be  overwhelmed  by  the
number  of  postings.  Limiting  students’
number  would  therefore  help  them  go
through the five phases of interaction.  
 
Limitations and future research
The main limitation of this paper is that it
investigated  only  two  online  discussions.
To  obtain  significant  results  on  the  effect
of  group  size  on  knowledge  construction
in  online  debates,  the  number  of  forums
should  be  increased.  One  of  the  main
limitations  of  this  type  of  research  is  the
subjectivity  of  coding.  The  classification
of  messages  is  open  to  individual
interpretation.  Using  Interaction  Analysis
Model  is  based  mainly  on  personal
opinion.  The  content  might  be  understood
differently  by  coders  resulting  in  different
phases of coding.   
 
This research study could be undertaken in
other  contexts  and  by  including  other
variables.  For  instance,  it  could  be
conducted  in  another  medium  of
communication.  Other  factors  that
influence knowledge construction could be
considered,  such  as  the  amount  of
participation.  Further  research  is  also
needed  to  discover  whether  the  type  of
knowledge  or  the  amount  of  knowledge
are  significant  predictors  of  participation
level  and  knowledge  construction  that
occur  in  online  debates.  It  would  also  be
quite  interesting  to  study  knowledge
construction  in  online  conversations  from
a  sociolinguistic  perspective  and  find  out
how social variables such as age, location,
social  status,  time  or  Internet  accessibility
could  be  related  to  level  of  knowledge
constructed  but  future  data  collection  and
analysis  are  required  for  more  rigorous
findings.