(2011).

 Paper  presented  at  Internet  Research  12.0,  Seattle,  WA,  October  10-­‐13.  

Mode  Choice  in  Multimodal  Comment  Threads:  Effects  on  Participation,  

Sociability,  and  Attitude  
 
Susan  C.  Herring  
Bradford  Demarest  
Indiana  University,  Bloomington  
 
Abstract  
 
This   study   investigates   how   participation,   sociability,   and   attitude   differ   according   to  
commenting   mode   and   commenter   gender   in   Voicethread.com,   an   interactive   Internet  
platform   that   allows  commenting   on   multimodal   slideshows   via   text,   audio,   or   video.  In  
such   platforms,   the   question   arises   as   to   whether   “richer”   modes   such   as   video   and   audio  
are  preferred  for  social  communication  and  “leaner”  text  for  more  impersonal,  contentious  
communication,   as   previous   research   has   claimed   (e.g.,   Daft   &   Lengel,   1984;   Sproull   &  
Kiesler,   1991).  Comments   in   all   three   modes   were   analyzed   for   the   frequency   of  
metadiscourse   terms   that   express   social   awareness   and   for   the   frequency   and   polarity   of  
attitude   categories   in   three   extended   public   VoiceThreads.   Video   comments   were   made  
mostly   by   males,   and   male   comments   were   more   negative,   consistent   with   previous  
research  on  gender  and  computer-­‐mediated  communication  (CMC).  Mode  differences  were  
also   found   that   provide   support   for   the   claims   of   early   CMC   theorists   associating   richer  
media   with   more   social,   interpersonal   communication   and   text   with   critical   judgment   and  
negativity.   These   findings   lead   us   to   revisit   claims   in   the   literature   that   have   previously  
been   largely   discredited,   including   technological   determinism   itself,   and   to   suggest   an  
alternative   account   that   reconciles   the   VoiceThread   findings   with   those   of   previous  
research  (e.g.,  Walther,  1996)  regarding  the  positive  sociability  of  textual  CMC.  

Introduction  
Interactive   multimodal   platforms   that   combine   video   and   user-­‐generated   textual  
comments   represent   a   growing   trend   in   convergent   media   computer-­‐mediated  
communication.   Typically   in   such   environments,   visual   content   serves   as   a   prompt   to  
which   text   comments   respond.   Examples   range   from   comments   posted   below   YouTube  
videos   to   text   annotations   inserted   directly   into   videos   and,   perhaps   less   familiarly,   into  
dynamic  visual  displays  of  audio  files,  as  on  SoundCloud.com.  The  platform  that  is  the  focus  
of  this  study,  VoiceThread.com,  takes  mixed-­‐modality  computer-­‐mediated  communication  
(CMC)  to  a  different  level,  however,  by  allowing  commenting  on  multimodal  slideshows  in  
multiple  modes1—text,  audio,  or  video—and  in  displaying  all  comments  together  in  a  single  
format.    
  VoiceThread.com  raises  questions  about  why,  how,  and  to  what  effect  users  choose  
to   participate   in   CMC   in   a   given   mode.   In   particular,   the   question   arises   as   to   whether  
“richer”   modes   such   as   video   and   audio   are   preferred   for   social   communication   and  
“leaner”   text   for   more   impersonal,   contentious   communication,   as   previous   research   has  
claimed   (e.g.,   Daft   &   Lengel,   1984;   Sproull   &   Kiesler,   1992).   Most   previous   research   in  
support   of   such   claims   has   been   experimental;   an   advantage   of   studying   VoiceThread  
comments   is   that   they   are   generated   in   authentic   contexts   of   use.   Moreover,   in   most  
experimental   studies   that   compare   communication   modes,   subjects   are   assigned   to   one  
   2  

mode   and   do   not   have   the   option   to   select   another,   whereas   VoiceThread   offers   users   a  
choice  of  commenting  modes  within  a  single  platform.  All  that  is  required  to  post  audio  and  
video   comments   is   a   microphone   and   web   cam,   which   most   laptop   computers   come  
equipped   with   these   days;   audio   comments   can   also   be   made   through   a   telephone   or  
uploaded  as  sound  files.  
  This   study   investigates   how   style   and   content   of   participation   differ   according   to  
commenting   mode   in   three   public   VoiceThreads,   employing   language-­‐focused   content  
analysis.   To   determine   if   audio   and   video   are   associated   with   greater   sociability,   all  
comments  were  analyzed  for  the  frequency  of  metadiscourse—hedges,  boosters,  personal  
pronouns,   and   other   language   forms   that   indicate   a   degree   of   social   awareness   and   an  
orientation  on  the  part  of  the  commenter  to  the  addressees  (Hyland,  2005).  To  assess  the  
effect   of   comment   mode   on   evaluative   stance,   categories   from   Martin   and   White   (2005)  
associated  with  attitude  were  adapted  and  coded  for  the  occurrence  and  polarity  of  each.  
Participation  metrics  were  also  calculated,  and  the  results  for  all  measures  are  compared  
by  thread,  gender,  and  mode.  
  The   findings   reveal   that   video   comments   were   made   mostly   by   males,   and   male  
comments   are   more   negative,   consistent   with   previous   research   on   technology   adoption  
(e.g.,  Gefen  &  Straub,  1997)  and  gender  and  politeness  in  CMC  (e.g.,  Herring,  1994).  Mode  
differences   are   also   evident   throughout   the   data,   and   provide   support   for   the   claims   of  
early   CMC   theorists   associating   richer   media   with   more   social,   interpersonal  
communication  and  text  with  more  judgment  and  negativity.  The  latter  findings  lead  us  to  
revisit   claims   in   the   literature   that   have   previously   been   largely   discredited,   including  
technological  determinism  itself,  and  to  suggest  an  alternative  account  that  reconciles  the  
VoiceThread   findings   with   those   of   previous   research   (e.g.,   Walther,   1996)   regarding   the  
positive  sociability  of  textual  CMC.    

Background  
Voicethread.com   was   founded   in   Boca   Raton,   Florida   by   Benji   Papell   and   Steve   Muth   and  
launched   in   March   2007.2   Thus   far,   the   platform   has   mainly   attracted   the   attention   of  
educators,   who   view   it   as   an   engaging   learning   and   discussion   tool   (e.g.,   Weir,   2008).  
However,   Millard   (2010)   analyzed   user   comments   from   50   randomly-­‐selected   public  
VoiceThreads   and   found   that   the   comments   showed   little   evidence   of   interaction   or  
collaboration;  rather,  they  responded  to  the  content  featured  in  the  multimedia  slideshow.  
Other  than  this,  little  scholarship  has  been  conducted  on  VoiceThreads  that  we  were  able  to  
locate.  
  VoiceThread   is   of   broader   theoretical   and   practical   interest,   nevertheless,   in   that   it  
offers   users   a   choice   of   communication   modes   on   a   single   platform.   This   has   not   been  
possible   before   on   the   same   scale   outside   experimental   contexts,   although   mode   choice  
across  separate  communication  platforms  has  been  theorized  and  researched.  Much  of  the  
research  is  based  on  early  theoretical  claims  involving  media  “richness,”  or  the  number  and  
nature   of   communication   channels   a   medium   provides.   Social   presence   theory   (Short,  
Williams,  &  Christie,  1976)  posits  that  participants  will  perceive  greater  social  presence  via  
richer   media   such   as   video   than   via   telephone   or   written   communication.   The   degree   of  
social   presence   is   equated   to   the   degree   of   awareness   of   the   other   person   in   a  
communication   interaction.   Information   richness   theory   posits   that   richer   media   such   as  
   3  

face-­‐to-­‐face   (FTF)   communication   are   better   suited   for   tasks   involving   nuanced   social  
communication,   while   leaner   media   such   as   text   are   best   suited   for   simple,   routine   tasks  
such   as   arranging   meetings   (Daft   &   Lengel,   1984).   Similar   to   social   presence   theory,   Daft  
and   Lengel   posit   a   one-­‐dimensional   continuum,   with   FTF   as   the   “richest”   and   numerical  
data  via  computer  as  the  “leanest”  medium.  Both  of  these  theories  predated  CMC,  but  they  
have  been  influential  in  CMC  research.  CMC  modes  can  be  situated  along  the  continuum  as  
in  Figure  1.    

 
Figure  1.  CMC  modes  in  the  continuum  of  media  richness  

  In   an   empirical   study   of   mode   choice,   Murray   (1988)   conducted   an   ethnographic  

study   over   a   nine-­‐month   period   of   mode   and   medium   choice   by   IBM   employees.   She   found  
that  FTF,  telephone,  paper,  and  computer  (including  both  synchronous  and  asynchronous  
CMC)   media   were   used   for   different   purposes,   as   determined   by   the   availability   of  
prospective   interactants,   time   management   (which   favored   the   use   of   CMC),   and   the  
sensitivity   of   the   message   (which   tended   to   favor   FTF   conversation).   Baym   et   al.   (2004)  
had   undergraduate   students   keep   a   diary   and   also   conducted   a   survey   regarding   their  
medium   use.   One   of   the   study’s   findings   was   that   “[p]eople   were   more   likely   to   use   FTF  
conversations   and   telephone   calls   in   more   intimate   relationships”   (p.   314),   as   predicted   by  
media   richness   theory.   Kim   et   al.   (2007)   conducted   a   social   network   analysis,   via   a   web  
survey,   of   communication   relationships   in   Korea   through   FTF,   email,   instant   messaging,  
mobile   phone,   and   SMS   text   messaging.   They   found   that,   whereas   “face-­‐to-­‐face   (FtF)   seems  
to   be   a   universal   medium   without   significant   differences   across   respondents,   “mobile  
phones  tend  to  be  used  in  reinforcing  strong  social  ties,  and  text-­‐based  CMC  media  tend  to  
be  used  in  expanding  relationships  with  weak  ties”  (n.p.).    
  The   research   described   above   mainly   focused   on   mode   choice   in   relation   to  
communicative  purpose  and/or  nature  of  the  interactants’  relationship.  Other  research  has  
studied   the   social   and   psychological   effects   on   behavior   of   different   modes.   An   early   and  
influential   study   by   Kiesler   et   al.   (1984)   found   mode   differences   in   disinhibition   (as  
measured   by   swearing,   insults,   name-­‐calling,   and   hostile   behavior),   decision   shifts,   and  
equality   of   participation,   all   of   which   occurred   more   often   in   textual   CMC   (both  
synchronous   and   asynchronous)   than   in   FTF   conditions.   The   first   two   findings   led   the  
authors   to   predict   that   textual   CMC   would   be   ill   suited   for   social   interaction,   consistent  
with   the   predictions   of   the   social   presence   and   media   richness   theories.   Kiesler   et   al.  
explain  their  findings  in  terms  of  the  paucity  of  social  cues,  such  as  facial  expressions  and  
prosody,   in   text-­‐only   CMC,   which   results   in   a   process   of   ‘depersonalization’   whereby  
   4  

messages   senders   feel   detached   from   (experience   less   social   presence   with)   their  
addressees.  
  The  explosion  of  naturally  occurring  CMC  on  the  Internet  soon  thereafter  gave  the  
lie   to   these   predictions   as   regards   the   limitations   of   textual   CMC,   however—or   so   it  
seemed.   As   early   as   1987,   Rice   and   Love   found   a   high   incidence   of   socio-­‐emotional   content  
in   the   messages   exchanged   among   medical   professionals   on   a   CompuServe   discussion  
forum.   Online   communities   arose   in   which   participants   reported   feeling   a   sense   of  
belonging  and  social  presence  (Rheingold,  1993);  reports  of  online  romances  soon  followed  
(e.g.,  Cooper  &  Sportolari,  1997).  At  the  same  time,  consistent  with  Kiesler  et  al.’s  findings  
of   disinhibition   in   CMC,   a   high   incidence   of   hostile   verbal   behavior,   or   ‘flaming’,   was  
observed  in  some  online  forums  (e.g.,  Kim  &  Raja,  1990).  To  account  for  these  behaviors,  
Walther   (1996)   theorized   that   textual,   asynchronous   CMC   was   ‘hyperpersonal’   in   that  
message   recipients   tended   to   overgeneralize   from   the   limited   social   cues   in   computer-­‐
mediated   messages,   with   the   result   that   both   positive   and   negative   perceptions   of   the  
sender   were   exaggerated.   Such   perceptions   fuel   romantic   feelings   as   well   as   feelings   of  
hostility.  
  An   alternative   approach   that   accounts   for   the   inconsistencies   between   studies  
showing  that  text-­‐only  CMC  can  be  socially  oriented  and  findings  of  studies  such  as  that  of  
Kiesler   et   al.   (1984)   is   the   Social   Information   Processing   theory   (Walther,   2002),   which  
holds   that   close   social   relationships   may   be   formed   via   leaner   media,   but   that   more   time   is  
required.  Most  previous  experimental  studies  provided  subjects  with  the  same  amount  of  
time  in  each  mode;  to  compensate  for  this,  Walther  and  his  colleagues  allowed  subjects  in  
text-­‐only   CMC   conditions   up   to   four   times   as   long   to   complete   tasks   (Tidwell   &   Walther,  
2002).   Under   such   conditions,   few   differences   in   social   perceptions   or   socioemotional   tone  
arose  across  modes,  although  when  some  communicators  used  one  mode  and  others  used  a  
richer   mode,   the   leaner   mode   was   rated   less   satisfying   (e.g.,   Walther   &   Bazarova,   2008).  
These  results  are  consistent  with  the  Electronic  Propinquity  Theory  proposed  by  Korzenny  
(1978),  which  posits  that  “users  will  adopt  the  widest  bandwidth  communication  medium  
available   to   them,”   but   that   “the   fewer   one's   choices   of   media,   the   more   closeness   one   may  
experience  even  through  the  lowest  of  bandwidths”  (Walther,  in  press).    
  In  contrast  to  explanations  that  invoke  the  influence  of  communication  technology  on  
mode  choice  and  user  behavior,  Herring  (1993,  1994,  1995,  2003)  invoked  a  social  factor,  
gender  socialization,  to  explain  why  men  and  women  participated  in  CMC  in  different  ways.  
Herring   noticed   that   ‘flaming’   was   much   more   common   in   messages   posted   to   online  
forums  by  males  than  by  females,  which  should  not  have  been  the  case  if  technology  alone  
predisposed   users   towards   expressing   hostility.   Men   also   participated   more   in   most   public  
forums   and   tended   to   post   longer   messages.   Similarly,   studies   of   technology   adoption   in  
organizations   have   argued   for   the   necessity   of   including   gender   as   a   factor   in   models   of  
technology  acceptance.  These  studies  found  that  women  perceive  email  as  higher  in  social  
presence   than   men   do   (Gefen   &   Straub,   1997),  adopt   new   technologies  for   social   reasons  
(such   as   usage   by   their   peers)   more,   and   are   more   concerned   with   ease   of   use,   whereas  
men   are   motivated   to   adopt   mainly   in   terms   of   the   affordances   of   the   technology   (its  
perceived   usefulness;   Venkatesh   &   Morris,   2000).   Despite   being   motivated   to   adopt,  
women   tend   to   be   slower   to   do   so,   consistent   with   previous   observations   that   men   tend   to  
feel  more  at  ease  with  computers  (e.g.,  Frankel,  1990).  
   5  

Research  Questions  and  Hypotheses  

The  general  research  question  that  informs  this  study  is:  How  do  users  comment  in  a  given  
mode  when  multiple  modes  are  available  on  a  single  Internet  platform?  Specifically:  
RQ1.   What   differences,   if   any,   are   there   in   sociability   and   negativity   across   the   three  
commenting  modes  available  on  VoiceThread.com?    
RQ2.   Are   there   gender   differences   in   sociability   and   negativity,   independent   of   mode  
choice?  
RQ3.   Are  there  gender  differences  in  mode  choice  and  amount  of  participation?  
Based  on  the  research  surveyed  above,  we  posited  the  following  hypotheses:  
H1:     “Richer”  modes  such  as  audio  and  video  will  convey  more  social  communication  
and  “leaner”  text  more  impersonal,  contentious  communication.  
H2.     Females   will   be   more   social   and   less   contentious   than   males,   independent   of  
mode  choice.  
H3.   Males  will  participate  more  and  comment  more  often  in  video  and  audio,  which  
are  more  novel  modes  than  text,  than  females  will.  

Methodology  
Data  
The  data  for  this  study  are  all  comments  posted  in  
three  public  VoiceThreads.  The  threads  are  prompt-­‐
triggered   discussions   set   in   academic   contexts,  
consistent   with   the   most   common   use   of   the  
VoiceThread   platform.   Because   we   had   little  
indication   of   what   to   expect   given   the   paucity   of  
previous  research  on  VoiceThread,  we  sampled  for  
diversity.   The   three   threads   vary   in   topic   and   the  
age   of   the   participants:   The   first   was   produced   by  
an   elementary   school   class   evaluating   art   in   science  
fiction,   the   second   was   produced   by   a   high   school  
driver’s   education   class   on   the   topic   of   the  
deleterious   effects   of   speeding,   and   the   third   is   a  
discussion   among   professional   educators   in  
  response   to   the   question:   What   does   the   network  
mean  to  you?  The  Sci  Fi  thread  had  155  comments,  
the   Speeding   thread   had   97   comments,   and   the  
Network   thread   had   111   comments   at   the   time   of  
our  data  collection  in  March  2011.  The  total  corpus  
consists  of  363  comments  and  22,069  words.    
  Figures  2  shows  how  an  audio,  a  video,  and  a  
text  comment,  respectively,  appear  during  playback  
of   the   Speeding   thread.3   Comments   can   be   accessed  
Figure  2.  An  audio  comment  (top  left),  a   either  by  clicking  on  the  user  profile  pictures  on  the  
video  comment  (bottom  left),  and  a  text   right   and   left   sides   of   the   central   slide,   or   by  
comment  (right)  in  the  Speeding  thread  
   6  

clicking  on  segments  of  the  playback  bar  at  the  bottom  of  the  interface.  The  playback  bar  
replays  the  comments  sequentially  in  the  order  in  which  they  were  posted.  

Analytical  Methods  
For   the   purposes   of   analysis,   all   audio   and   video   comments   were   transcribed   by   the  
authors   and   entered   into   an   Excel   spreadsheet   along   with   the   text   comments.   Each  
comment   was   first   coded   for   the   independent   variables   thread,   mode,   and   gender   of  
participant.   Gender   was   included   as   an   independent   variable   in   this   study,   because  
previous   research   has   found   gender   differences   in   participation   in   CMC   (e.g.,   Herring,  
2003),   including   in   online   learning   environments   (e.g.,   Barrett   &   Lally,   1999).   The  
comments  were  then  coded  for  the  three  dependent  variables  of  participation,  attitude,  and  
metadiscourse,  as  described  below.  
 
Participation  
The   participation   analysis   consisted   of   a   straightforward   counting   of   the   number   of  
messages   and   the   number   of   words,   as   well   as   a   calculation   of   the   number   of   words   per  
message.  The  results  of  these  measures  are  reported  using  descriptive  statistics.  
 
Attitude  
To  assess  the  effect  of  comment  mode  on  judgment  and  negativity,  we  adapted  categories  
from  Martin  and  White  (2005)  associated  with  attitude.  These  are  listed  in  Table  1  below  
(definitions  are  from  Martin  &  White;  examples  are  from  our  data).  
 
Category   Definition  and  Examples  
Affect   “Emotions;  reacting  to  behavior.”  (e.g.,  happy,  sad)  
Judgment   “Ethics;  evaluating  behavior.”  (e.g.,  dangerous,  wrong,  
should/shouldn’t)  
Appreciation   “Aesthetics;  evaluating  text/process,  natural  
  phenomena.”  
       Appreciation:reaction   •  to  things:  “do  they  please  us,  do  they  catch  our  
  attention?”  (e.g.,  interesting,  like,  dislike)  
       Appreciation:composition   •  the  thing’s  “balance  and  complexity.”  Did  it  hang  
together?  Is  it  well  crafted?  
  •  “How  innovative,  timely,  authentic,  etc.”  Was  it  
       Appreciation:valuation   worthwhile?  (e.g.,  useful,  cool,  great,  weird)  
Table  1.  Martin  and  White’s  attitude  categories  
   
We  also  found  instances  in  our  data  that  seemed  to  express  attitude,  but  in  a  formulaic  way,  
and   that   did   not   seem   to   fit   into   Martin   and   White’s   categories.   Thus   we   modified   the  
coding  rubric  to  include  the  three  additional  categories  listed  in  Table  2  below,  for  a  total  of  
eight  attitude  coding  categories.  
 
   7  

Category   Examples  
Formulaic  affect   ‘I’d  love  to  hear  your  thoughts,’  ‘I  hate  to  break  it  to  you’  
Formulaic  judgment   ‘You  have  to  wonder,’  ‘Speed  kills’  
Formulaic  appreciation   ‘Thanks  for  starting  this  thread,’  ‘take  care’  
Table  2.  Additional  attitude  categories  
 
In  addition,  we  coded  each  expression  of  attitude  for  polarity:  positive,  negative,  or  
neutral.   The  unit  of  analysis  could  be  a  word,  phrase,  utterance,  or  chunk  of  a  message  that  
expressed   a   particular   attitude,   and   more   than   one   attitude   could   be   coded   per   message,  
including  the  same  one  multiple  times.    
  Both  authors  participated  in  the  coding.  Attitude  and  polarity  codes  were  assigned  
independently  by  each  author  for  a  portion  of  the  data,  code  assignments  were  compared,  
and   disagreements   were   resolved   through   discussion.   This   process   was   iterated   until   all  
messages   had   been   jointly   coded   with   100%   agreement   (for   Speeding   and   SciFi)   or   until  
better   than   80%   agreement   was   reached   for   both   attitude   and   polarity   assignment   (for  
Network);  in  the  latter  case,  the  second  author  coded  the  remaining  messages.    
 
Metadiscourse  
All   comments   were   analyzed   for   the   frequency   of   metadiscourse—language   forms   that  
indicate  a  degree  of  social  awareness  and  an  orientation  on  the  part  of  the  commenter  to  
the  addressees.  Hyland  (2005,  p.  37)  defines  metadiscourse  as  “self-­‐reflective  expressions  
used   to   negotiate   interactional   meanings,   assisting   the   writer   (or   speaker)   to   …   engage  
with  readers.”  This  analysis  was  conducted  to  measure  the  amount  of  social  presence  in  the  
comments  in  each  mode.  
  Hyland   groups   metadiscourse   into   two   broad   types:   interactive   and   interactional,  
each  of  which  has  multiple  categories,  as  summarized  in  Table  3.  
 
  Interactive  Metadiscourse   Interactional  Metadiscourse  
  Code  glosses   Attitude  markers  
  Endophoric  markers   Boosters  
  Evidentials   Self-­‐mention  
  Frame  markers:   Engagement  markers  
  • Sequencing   Hedges  
• Label  stages    
• Announce  goals  
Shift  topic  
Transition  markers  
Table  3.  Hyland’s  metadiscourse  categories  
 
Interactive  metadiscourse  mainly  involves  reference  to  other  parts  of  the  discourse,  while  
interactional   metadiscourse   involves   the   interaction   of   the   writer   or   speaker   with   the  
audience.    
  In  an  appendix  to  his  book,  Hyland  (2005)  provides  an  extensive  list  of  metadiscourse  
   8  

terms   in   English,   based   on   his   studies   of   academic   writing.   We   modified   this   list   to   exclude  
punctuation   and   non-­‐alphabetic   symbols   (on   the   grounds   that   these   are   not   possible   in  
speech);  formal  terms  and  conventions  mainly  found  in  written  text  (such  as  ‘in  chapter  X’);  
terms   with   variable   elements   (such   as   date,   sequences   of   numbers)   that   are   difficult   to  
search   for;   and   common   terms   that   do   not   mainly   function   as   metadiscourse   (such   as   ‘and’  
and   ‘go’).   We   also   manually   reviewed   the   messages   in   our   sample   and   added   terms   that  
occurred   there   but   were   not   already   in   Hyland’s   list;   these   included   references   to   the  
multimedia   slide   show   at   the   center   of   the   VoiceThread   (e.g.,   ‘(in)   the/this  
video/chart/diagram/picture/image/slide’)   and   spoken   discourse   phenomena   such   as  
contracted   hedges   (e.g.,   ‘kinda,’   ‘sorta’)   and   discourse   markers   (e.g.,   ‘oh’,   ‘yeah’,   ‘y’know’,  
‘like’).   Finally,   we   lemmatized   terms   with   a   common   root   (such   as   ‘seem/seems/seemed’  
into   seem*)   to   facilitate   their   retrieval   from   the   corpus.   The   end   result   was   a   list   of   390  
lemmatized   terms   that   we   imported   into   a   freely   available   concordancing   program,  
CasualConc,  which  was  used  to  sort  and  count  the  frequencies  of  each.  The  results  returned  
by   CasualConc   for   each   term   were   manually   filtered   by   the   authors   to   exclude   instances  
that  did  not  function  as  metadiscourse  in  the  contexts  in  which  they  occurred.  
  Figure  3  shows  part  of  the  spreadsheet  in  which  term  frequencies  were  recorded.  
   

 
Figure  3.  Spreadsheet  in  which  metadiscourse  frequencies  were  recorded  

Statistical  Analysis  
After   the   participation,   attitude,   and   metadiscourse   data   were   collected,   counts   for   each  
category  within  each  variable  were  aggregated  to  reflect  counts  in  various  subcorpora,  first  
for   each   combination   of   thread,   gender,   and   mode,   and   then   at   higher   levels   of   aggregation  
(thread,   gender,   mode,   and   secondary   combinations   of   thread-­‐gender,   thread-­‐mode,   and  
   9  

mode-­‐gender   combinations).     In   order   to   allow   for   direct   comparisons   of   values   from  

subcorpora   of   different   sizes,   results   were   normalized:   attitude   counts,   by   the   number   of  
messages   within   the   subcorpus,   metadiscourse   counts   by   total   word   count   in   the  
subcorpus,  scaled  to  counts  per  1000  words.  
  After   compiling   descriptive   statistics,   we   further   assessed   attitude   and  
metadiscourse  data  differences  through  the  use  of  Chi  square  tests  to  determine  statistical  
significance.     Chi   square   tests   were   employed   due   to   the   nature   of   the   data   collected   as  
unranked  counts  of  exclusive  categories,  rather  than  scalar  measurements.  

Results  
Participation  
Table  4  summarizes  the  participation  results  for  the  corpus  overall.  Percentages  add  up  to  
100%  vertically  for  each  independent  variable.  
 
      Words  (%)     Valid  msgs  (%)   Avg  words/msg  
  All     22,069         3204       69.0  
Gender  
  Male     12,702    (57.6%)     177    (55.3%)       71.8    
  Female   8824    (40.0%)       108    (33.75%)     81.7    
  Unknown   543    (2.5%)         35    (10.9%)       15.5  
Mode  
  Video     2972    (13.4%)           15    (4.7%)     198.1    
  Audio     14,801  (67.1%)     102    (31.9%)     145.1    
  Text     4296    (19.5%)         203    (63.4%)     21.2    
Thread  
  Speeding   2865    (13.0%)   87    (27.2%)     32.9    
  Network   16,853    (76.4%)   109    (34.1%)     154.6    
  SciFi     2351    (10.6%)   124    (38.7%)     19.0  

Table  4.  Participation  results,  overall  corpus  

Overall,  males  contributed  more  messages  (comments)  and  more  words  than  females.  The  
majority   of   comments   were   via   text,   followed   by   audio;   relatively   few   video   comments  
were   posted   (the   SciFi   thread   contained   none   at   all).   Conversely,   video   comments  
contained   the   most   words,   followed   by   audio   comments;   text   comments   were   much  
shorter.   There   was   also   considerable   variability   in   participation   across   threads:   SciFi,   the  
thread   with   elementary   school   children,   had   the   most   messages   but   the   fewest   words  
(most   comments   were   via   text),   while   Network,   the   thread   with   academic   professionals,  
had  the  most  words  and  longest  messages  (most  comments  were  via  audio).  
  Table  4  represents  fairly  well  the  patterns  within  each  thread,  with  one  exception:  
Females  did  not  post  longer  messages  than  males  in  any  of  the  threads.  The  overall  result  
that   female   messages   were   longer   is   caused   by   the   fact   that   a   majority   of   females   in   the  
total   sample   are   in   the   Network   thread,   which   also   has   the   longest   messages.   The  
distribution  of  participation  by  gender  is  shown  in  Table  5.  
   10  

      Words  (%)     Valid  msgs  (%)   Avg  words/msg  

Speeding  
  Male     1538  (54%)       42    (48.3%)       36.6    
  Female   1154    (40.0%)       34    (39.0%)       33.9    
  Unknown   173    (6.0%)         11    (12.6%)       15.7  
Network  
  Male     9464    (56.2%)     53    (48.6%)       178.6    
  Female   7273    (43.2%)       48    (44.0%)       151.5    
  Unknown   116    (0.7%)         8    (7.3%)       14.5  
SciFi  
  Male     1700    (72.3%)     82    (66.1%)       20.7    
  Female   397    (16.9%)       26    (21.0%)       15.3    
  Unknown   254    (10.8%)       16    (12.9%)       15.9  

Table  5.  Participation  by  gender  

  Because   comments   posted   by   participants   of   unknown   gender   are   a   minority   in  

each   thread,   and   because   they   tend   to   be   shorter   than   comments   by   gender-­‐identifiable  
participants   and   contain   less   information,   making   them   harder   to   interpret,   they   were  
excluded  from  further  analysis.  

Attitude  
Figure   4   presents   the   attitude   results   for   the   corpus   overall.     Positive   expressions   of  
attitude  are  the  most  frequent  throughout  the  corpus  (totaling  488,  or  70%  of  all  attitude  
expressions),  with  negative  being  second-­‐most  frequent  at  146,  or  21%.  Neutral  comments  
constitute  the  remaining  10%,  with  68  occurrences.  Apprec:valuation  acts  were  used  most  
often,   followed   by   apprec:reaction   acts,   then   judgment,   then   affect.   Appreciation:  
composition  and  the  formulaic  acts  were  used  least  often.  
 
200  

150  

100  

50   positive  

0   negative  
neutral  

 
Figure  4.  Attitude  results,  overall  corpus  
   11  

  The  profiles  for  attitude  results  vary  considerably  by  thread,  however,  as  shown  in  
Figure   5.   The   Network   thread,   in   which   participants   mostly   praise   the   virtues   of   the  
Internet,  is  overwhelmingly  positive  and  valuative,  while  comments  in  the  Speeding  thread,  
which  concerns  the  dangers  of  driving  over  the  speed  limit,  have  mostly  negative  judgment.  
The  SciFi  thread,  in  which  participants  commented  on  art  that  was  sometimes  violent  and  
sometimes   beautiful,   has   both   positive   and   negative   comments   and   contains   more  
reactions.  
 
 
 
 
  Network  
 
 
 
 
 
 
 
Speeding  
 
 
 
 
 
 
  SciFi  
 
 
 
 
 
 
   
Figure  5.  Attitude  results  for  (from  top  to  bottom)  Network,  Speeding,  and  SciFi  thread  

Figure   6   shows   that   there   are   more   similarities   than   differences   in   the   overall  
attitude   profiles   of   males   and   females.   However,   females   are   more   positive,   especially   as  
regards  judgment  acts,  which  they  appear  to  use  more  than  males  do.  Females  also  express  
more   formulaic   appreciation   (which   is   all   positive)   and   males   express   more   formulaic  
judgment  (which  is  mostly  negative).  
 
   12  

 
 
 
  Males  
 
 
 
 
 
 
Females  
 
 
 
 
 
 
 
 
   
 
Figure  6.  Attitude  results  for  (from  top  to  bottom)  males  and  females  
  In  contrast,  the  mode  results  in  Figure  7  reveal  a  striking  pattern:  Video  and  audio  
pattern   together   in   favoring   positivity   and   valuation   acts,   in   contrast   to   text,   which   is   more  
negative   and   has   proportionately   more   judgment,   formulaic   judgment,   and  
appreciation:reaction   acts,   all   of   which   tend   to   be   more   negative   than   the   corresponding  
acts  in  the  other  two  modes.    
 
   13  

 
  Video  
 
 
 
 
 
 
 
  Audio  
 
 
 
 
 
 
 
Text  
 
   
 
 
 
 
 
 
 
 
 
 
 
Figure  7.  Attitude  results  for  (from  top  to  bottom)  video,  audio,  and  text  

Chi  square  tests  

Chi   square   statistical   tests   were   conducted   for   the   trends   and   patterns   suggested   by   the  
descriptive  statistics  presented  above.  Table  6  presents  the  results  for  those  patterns  that  
were  found  to  be  significant.  Chi  square  tests  were  run  in  relation  to  proportions  derived  
from  higher  categories  than  those  being  tested.    Specific  polarities  within  a  category  were  
tested   against   overall   proportions   of   a   category.     Overall   counts   of   a   specific   category   were  
tested   against   proportions   of   total   attitude   counts.     Total   attitude   counts   were   tested  
against  proportions  of  total  message  counts.  
 
   
   14  

  Trend           χ²  value              significancea  

Overall                
  Positive  >  Negative         184.4858     ****  
  Positive  >  Neutral         317.2662     ****  
  Network  >  Speeding  (Overall  attitude)     101.7041     ****  
  Network  >  Sci  Fi  (Overall  attitude)     185.2269     ****  
  Speeding  >  Sci  Fi  (Overall  attitude)     4.8334       ****  
  Females  >  Males  (Overall  attitude)     5.331       *  
  Video  >  Audio  (Overall  attitude)     13.929       ***  
  Audio  >  Text  (Overall  Attitude)     175.851     ****  
  Video  >  Text  (Overall  Attitude)     165.713     ****  
Thread  
  Network  >  Speeding  (Positive)       52.2693     ****  
  Network  >  Sci  Fi  (Positive)       17.6825     ****  
  Speeding  >  Network  (Negative)     52.2693     ****  
  Speeding  >  Sci  Fi  (Negative)       11.518       ***    
  Network  >  Sci  Fi  (Appreciation:valuation)   18.1266     ****  
  Speeding  >  Network  (Judgment)     17.3875     ****  
  Speeding  >  Sci  Fi  (Judgment)       14.4801     ****  
Gender    
  Male  >  Female  (Negative)b       11.11       ***  
  Female  >  Male  (Judgment,  Positive)     5.2505       *  
  Female  >  Male  (Formulaic:appreciation)   4.4041       *  
  Male  >  Female  (Formulaic:judgment)     5.9287       *  
Mode  
  Video  >  Text  (Positive)c       12.4295     ***  
  Audio  >  Text  (Positive)c       26.6178     ****  
  Text  >  Video  (Negative)   c     8.3384       **  
  Text  >  Audio  (Negative)c       30.8495     ****  
  Audio  >  Text  (Appreciation:valuation)     3.9999       *  
  Text  >  Video  (Judgment)c       5.3915       *  
  Text  >  Audio  (Judgment)   c     6.0062       *  
  Text  >  Audio  (Formulaic:judgment)d     12.8627     ***  
  Text  >  Audio  (Judgment,  Negative)d     7.6414       **    
  Text  >  Audio  (Apprec:reaction,  Negative)d   3.9615       *  
Gender-­‐Mode  
  Text:Male  >  Text:Female  (Negative)     3.9361       *  
  Text:Female  >  Text:Male  (Positive)     3.8598       *  
  Text:Male  >  Video:Male  (Negative)c     8.8826       **  
  Text:Male  >  Audio:Male  (Negative)c     22.4989     ****  
  Video:Male  >  Text:Male  (Positive)c     14.0265     ***  
  Audio:Male>Text:Male  (Positive)c     22.4102     ****  
a  *p<0.05,  **p<0.01,  ***p<0.001,  ****p<0.0001    
b  Female  >  Male  (Positive)  also  nearly  achieved  significance  (p  =  .06).  
c  No  significant  differences  were  found  between  video  and  audio  on  these  measures.  
d  Insufficient  video  tokens  were  available  to  allow  Chi  square  tests  to  be  done  for  Text  >  Video.  
Table  6.  Chi  square  results  for  attitude  analysis  
   15  

Most  of  the  trends  suggested  by  the  descriptive  statistics  turn  out  to  be  statistically  
significant.     To   summarize   the   main   trends,   expressions   of   attitude   are   mostly   positive  
overall,   but   there   is   significant   variation   by   thread:   Network   is   overwhelmingly   positive,  
while   Speeding   is   mostly   negative,   and   SciFi   is   mixed.   Males   are   more   negative   than  
females,   especially   in   judgment   acts.   Text   is   more   negative   than   video   and   audio   messages,  
and  text  also  contains  more  judgment.  

Metadiscourse  
The  results  for  the  metadiscourse  analysis  for  the  overall  sample  are  presented  according  
to   Hyland’s   categories   in   Figure   8.   Self-­‐mentions   were   used   most   frequently   (1,423  
occurrences,   or   39%),   followed   by   boosters   (645,   18%),   hedges   (452,   12.5%),   and  
engagement  markers  (438,  12%),  especially  2nd  person  pronouns.    
 
1600  
1400  
1200  
1000  
800  
600  
400  
200  
0  

 
Figure  8.  Metadiscourse  results,  overall  corpus  

Interactional   metadiscourse   was   used   much   more   than   interactive   metadiscourse.   Because  
of  the  small  numbers  for  the  interactive  categories,  and  because  they  are  less  ‘social’  than  
the   interactional   categories   and   therefore   less   relevant   to   the   present   study,   we   do   not  
analyze  them  further  here.  
Figure  9  shows  the  distribution  of  the  interactional  metadiscourse  terms  by  thread  
normalized  per  1000  words.  The  Network  thread,  in  which  participants  responded  to  the  
question   “What   does   the   network   mean   to   me?”,   overwhelmingly   favored   self-­‐mentions  
(e.g.,   ‘Hi,   this   is   [name]   and   I   just   wanted   to   say   what   the   network   means   to   me   is...’),  
whereas   the   Speeding   thread   favored   engagement   markers   (especially   non-­‐referential  
‘you’,   as   in   ‘you   shouldn’t   speed’),   boosters   (e.g.,   ‘speeding   is   definitely   not   worth   the  
consequences’),   and   attitude   markers   (‘its   amazing   how   5km   can   make   such   a   big  
difference’).   The   SciFi   thread,   in   which   participants   evaluated   and   interpreted   art   works,  
slightly   favored   the   use   of   hedges   (e.g.,   ‘possibly   the   person   that   made   the   drawing   is  
   16  

probably   an   atheist’)   and   discourse   markers   (e.g.,   ‘Ok,   so,   I   kinda   feel   like   there's   um..   a  
more  deeper  aspect  to  this  picture’).  
 
80  

60  

40  

20   Network  

0   SciFi  
Speeding  

 
Figure  9.  Interactional  metadiscourse  results  by  thread  

  In   Figure   10,   the   interactional   metadiscourse   results   are   broken   down   by   gender.  
Females  use  more  boosters,  and  males  use  more  hedges,  2nd  person  pronoun  engagement  
markers,   and   more   discourse   markers;   otherwise   the   genders   use   similar   proportions   of  
interactional  metadiscourse.  
 
80  

60  

40  

20  
Female  
0  
Male  

 
Figure  10.  Interactional  metadiscourse  results  by  gender  

  In   contrast,   numerous   differences   emerge   in   interactional   metadiscourse   use   by  

mode.  Video  and  audio  comments  contain  many  more  self-­‐mentions,  while  text  comments  
contain   more   engagement   markers   (including,   but   not   limited   to,   2nd   person   pronouns),  
hedges,  and  attitude  markers.  See  Figure  11.  
 
   17  

80  

60  

40  

20   Audio  
Text  
0  
Video  

 
Figure  11.  Interactional  metadiscourse  results  by  mode  
Chi  square  tests  
Chi   square   statistical   tests   were   conducted   for   the   metadiscourse   trends   and   patterns  
suggested   by   the   descriptive   statistics   presented   above.  The   Chi   square   tests   were   run   in  
relation   to   proportions   derived   from   total   counts   of   words   in   the   tested   subsets.   Table   7  
shows  the  results  for  those  patterns  that  were  found  to  be  significant.  
 

  Trend           χ²  value              significancea  

Overall                
  Interactional  >  Interactiveb       2000.546     ****  
  Speeding  >  Network  (Overall  metadiscourse)   8.0095       **  
  Video  >  Text  (Overall  metadiscourse)c     15.8028     ****  
  Audio  >  Text  (Overall  metadiscourse)c   16.5534     ****  
Thread  
  Network  >  Speeding  (Self-­‐mentions)     72.0725     ****  
  Network  >  Sci  Fi  (Self-­‐mentions)     45.1856     ****  
  Sci  Fi  >  Network  (Hedges)       46.8777     ****  
  Speeding  >  Network  (Hedges)       21.5244     ****  
  Speeding  >  Network  (Boosters)     22.8357     ****  
  Sci  Fi  >  Network  (Boosters)       4.6058       *  
  Speeding  >  Network  (Engagement)     130.2452     ****  
  Sci  Fi  >  Network  (Engagement)     10.8868     ***  
  Speeding  >  Sci  Fi  (Engagement)     18.6528     ****  
  Speeding  >  Network  (Attitude  Markers)   19.6604     ****  
  Sci  Fi  >  Network  (Discourse  Markers)     27.7765     ****  
Gender    
  Female  >  Male  (Boosters)       5.1122       *  
  Male  >  Female  (Hedges)       6.6942       **  
  Male  >  Female  (Engagement)       5.6535       *  
  Male  >  Female  (Discourse  Markers)     4.2774       *  
Mode  
  Video  >  Text  (Self-­‐mentions)c       96.7819     ****  
   18  

  Audio  >  Text  (Self-­‐mentions)c       100.5152     ****  

  Audio  >  Text  (Discourse  Markers)     4.9392       *  
  Text  >  Audio  (Engagement)c       21.1007     ****  
  Text  >  Video  (Engagement)c       10.9684     ***  
  Text  >  Audio  (Hedges)d       9.0758       **  
  Text  >  Video  (Attitude  Markers)c     5.8628       *  
  Text  >  Audio  (Attitude  Markers)c     9.2009       **  
Gender-­‐Mode  
  Video:M  >  Video:F  (Hedges)       4.8521       *  
  Audio:M  >  Audio:F  (Engagement)     8.6958       **  
  Audio:F  >  Audio:M  (Boosters)       8.8934       **  
a  *p<0.05,  **p<0.01,  ***p<0.001,  ****p<0.0001    
b   There   is   no   significant   difference   in   the   ratio   of   interactional:interactive   for   any   of   the   variables  

below.  Results  for  ‘overall’  apply  equally  to  ‘interactional’.  

c  No  significant  differences  were  found  between  video  and  audio  on  these  measures.  
d  Text  >  Video  (Hedges)  is  not  significant,  nor  is  Video  >  Audio  (Hedges).  

Table  7.  Chi  square  results  for  metadiscourse  analysis  

To   summarize   the   main   findings,   the   corpus   contains   more   interactional   than  
interactive  metadiscourse  overall.  Speeding  contains  the  most  metadiscourse,  followed  by  
SciFi,   then   Network,   and   each   thread   favors   different   terms,   reflecting   the   topic   of   the  
thread.   Males   use   significantly   more   hedges   (especially   in   video),   engagement   markers  
(especially  in  audio),5   and  discourse  markers,  while  females  use  more  boosters  (especially  
in   audio).   Video   and   audio   comments   contain   significantly   more   metadiscourse   overall  
than  text,  especially  self-­‐mentions,  and  audio  contains  more  discourse  markers.  In  contrast,  
text   contains   significantly   more   engagement   markers,   hedges,   and   attitude   markers   than  
video  and  audio.    

Discussion  
Research  Questions  and  Hypotheses  Revisited  
Our   first   research   question   asked:   What   differences,   if   any,   are   there   in   sociability   and  
negativity   across   the   three   commenting   modes   available   on   VoiceThread.com?   We  
hypothesized   that   “richer”   modes   such   as   audio   and   video   would   convey   more   social  
communication   and   “leaner”   text   more   impersonal,   contentious   communication.   The   first  
part   of   this   hypothesis   was   partially   supported.   Significantly   more   metadiscourse   terms  
were   found   in   the   audio   and   video   comments   than   in   the   text   comments   overall.   The  
difference   was   mainly   accounted   for   by   the   much   greater   frequency   of   self-­‐mentions   in  
video   and   audio,   however.   Text   comments   actually   had   more   engagement   markers,  
including  ‘you’,  more  hedges,  and  more  attitude  markers.  Thus  text  does  not  appear  to  be  
impersonal.   If   anything,   these   results   make   text   appear   more   other-­‐aware,   in   contrast   to  
audio  and  video,  which  seem  to  make  commenters  more  self-­‐conscious.  
  The  second  part  of  Hypothesis  1  was  also  partially  supported.  The  greater  frequency  
of   attitude   metadiscourse   in   text   is   consistent   with   the   results   of   the   attitude   content  
analysis,   which   found   more   expressions   of   judgment   in   text.   As   hypothesized,   text  
comments   were   also   significantly   more   negative   than   audio   and   video   comments   overall,  
   19  

although  male  commenters  alone  were  responsible  for  this  result.  This  gender  difference  is  
consistent   with   previous   findings   that   flaming   and   hostility   are   more   frequent   in   text  
comments   by   males   than   by   females   in   discussion   forums,   newsgroups,   and   chat   rooms  
(e.g.,  Herring,  1994,  1995,  2003).    
  Our   second   research   question   asked   whether   there   were   gender   differences,  
independent  of  mode  choice.  We  hypothesized  that  females  would  be  more  social  and  less  
contentious   than   males.   The   results   showed   that   females   were   significantly   less   negative  
than   males   in   all   modes,   in   support   of   the   second   part   of   this   hypothesis.   However,   few  
gender   differences   were   found   for   sociability,   as   measured   by   the   use   of   metadiscourse  
terms.   If   anything,   males   used   slightly   more   metadiscourse   terms   overall,   relative   to   the  
number  of  words  in  their  comments.  Moreover,  females’  greater  use  of  boosters  and  males’  
greater   use   of   hedges   is   the   opposite   of   what   has   been   observed   in   spoken   discourse   (cf.  
Coates,   1993)   and   Internet   discussion   forums   (cf.   Herring,   1995;   Herring,   Johnson,   &  
DiBenedetto,   1998).   Thus   the   first   part   of   the   hypothesis   was   not   supported.   A   possible  
explanation  for  the  lack  of  predicted  gender  differences  in  sociability  is  that  the  task/topic  
was  the  same  for  both  genders  within  each  thread,  and  almost  all  comments  were  on-­‐task.  
Herring  and  Paolillo  (2006)  found  that  gender  differences  in  word  frequencies  disappeared  
in   blog   posts   when   the   general   topic   of   the   blog   (personal   reflections   vs.   commentary   on  
events   external   to   the   writer)   was   held   constant.   Similarly,   topic   may   exercise   a   greater  
influence  than  gender  in  determining  metadiscourse  term  frequency.  
  Finally,   the   third   research   question   asked   whether   there   were   gender   differences   in  
mode  choice  and  amount  of  participation,  hypothesizing  that  males  would  participate  more  
and  comment  more  often  in  video  and  audio  than  females  would.  Males  posted  more  and  
longer   messages   in   each   thread,   consistent   with   the   first   part   of   this   hypothesis.   Males   also  
chose  to  comment  using  video  much  more  often  than  females  did,  although  there  was  no  
gender  difference  in  choice  of  audio.    
  The  hypotheses  and  results  of  the  study  are  summarized  in  table  8.  
 
Hypothesis   Supported?  
Video  and  audio  more  social   YES  overall  and  for  self-­‐mentions;    
NO  for  engagement,  hedges,  attitude  markers  
Text  more  judgmental,  negative   YES  
Females  more  social   NO  
Females  less  contentious   YES  
Males  participate  more   YES  
Males  use  audio  and  video  more   YES  for  video;  NO  for  audio  
Table  8.  Summary  of  results  

Implications  for  CMC  Theory  

Overall,   our   hypotheses   predicting   modal   differences   in   discourse   are   supported   by   the  
VoiceThread   data.   These   hypotheses   were   based   on   early   CMC   theories—the   social  
presence   theory   (e.g.,   Short,   Williams,   &   Christie,   1976),   the   media   richness   theory   (e.g.,  
Daft   &   Lengel,   1984),   and   lack   of   social   cues   theory   (e.g.,   Kiesler   et   al.,   1984;   Sproull   &  
Kiesler,   1992)—which   in   turn   were   grounded   in   the   notion   that   technical   properties   of  
communication  media  determine  the  nature  of  the  communication  that  takes  place  through  
   20  

those   media.   The   notion   of   technological   determinism   has   been   much   criticized   in  
subsequent   literature,   especially   given   the   evidence   that   naturally   occurring   text  
communication   on   the   Internet,   in   email,   discussion   forums,   chatrooms,   instant   messaging,  
blogs,   etc.,   can   be   highly   social,   even   “hyperpersonal”   (Walther,   1996).   How   can   we  
reconcile   the   findings   of   the   present   study,   which   are   also   based   on   naturally   occurring  
communication  on  the  Internet,  with  these  findings  of  sociable  text-­‐only  CMC?  
  Walther   and   Bazarova   (2008)   invoke   electronic   propinquity   theory   (Korzenny,  
1978)   to   explain   why,   under   experimental   conditions,   subjects   express   greater   satisfaction  
and   experience   more   “propinquity”   (perceived   closeness)   with   their   interlocutor(s)   in   a  
given   mode   of   communication   when   no   other   modes   are   present   in   the   environment   to  
compare  it  with.  The  tenet  that  “the  fewer  one's  choices  of  media,  the  more  closeness  one  
may  experience  even  through  the  lowest  of  bandwidths”  (Walther,  in  press)  is  not  directly  
relevant   to   the   present   study,   since   more   than   one   mode   was   available   to   VoiceThread  
users,   and   perceptions   of   closeness   were   not   addressed   in   this   study.   Nonetheless,  
electronic   propinquity   theory   helps   explain   the   apparent   inconsistency   between   our  
results  and  those  of  other  studies  of  natural  computer-­‐mediated  discourse,  if  one  considers  
that  Internet  users  for  the  most  part  previously  had  only  textual  CMC  available.  With  only  
text   to   work   with,   users   adapted   their   practices   to   enable   rich   communication.  
VoiceThread,   in   contrast,   offers   a   choice   of   multiple   modes,   and   in   that   context   of   choice,  
differentiation   of   behavior   occurs   across   modes,   consistent,   for   the   most   part,   with  
technologically  deterministic  theories  and  the  findings  of  multi-­‐mode  experimental  studies.  
Thus  the  VoiceThread  results  provide  hitherto  unavailable  evidence  that  the  technological  
properties   of   a   mode   shape   the   nature   of   the   communication   when   multiple   modes   are  
available  to  choose  from  in  a  natural  (i.e.,  non-­‐experimental)  context.    
  The   alternative   explanation—that   the   VoiceThread   participants   selected   a   given  
mode   because   they   wished   to   communicate   in   a   certain   way;   that   purpose   shaped   their  
choice,  rather  than  the  choice  shaping  their  communication—seems  less  likely,  for  several  
reasons.   First,   topic   was   a   constant   within   each   VoiceThread;   thus   we   may   assume   that   the  
purpose  for  communicating  was  similar  for  each  participant.  The  on-­‐topicness  of  all  but  a  
few   random   comments   supports   this   assumption.   Second,   almost   all   the   video   comments  
included  some  reflection  on  the  fact  that  the  commenter  was  ‘trying  out’  the  video  mode  for  
the  sake  of  experimentation,  because  it  was  novel.  That  almost  all  video  commenters  were  
male  is  consistent  with  the  findings  of  Gefen  and  Straub  (1997)  and  Venkatesh  and  Morris  
(2000)   that   males   tend   to   adopt   new   technology   earlier   than   females   do.   Finally,   the  
majority   of   commenters   used   text,   but   not   because   they   wanted   to   express   negative  
judgment;   many   text   comments   were   positive.   Participants   chose   text   presumably   because  
text  commenting  was  most  familiar  to  them;  it  was  the  default  choice.  Video  commenting  
was  selected  least  often,  most  likely  because  it  was  unfamiliar  and  perhaps  intimidating  to  
users   of   the   platform.   This   finding   violates   Korzenny’s   (1978)   tenet   that   users   will   adopt  
the  widest  bandwidth  communication  medium  available  to  them.  
  Thus  the  VoiceThread  data  partially  support  several  previous  theories,  but  they  do  
not   fully   support   any   one   of   them.   They   do   not   support   the   prediction   of   the   social  
presence,   media   richness,   and   electronic   propinquity   theories   that   richer   media   will   be  
preferred  for  social  communication.  Although  the  comments  in  video  and  audio  contained  
more   linguistic   expressions   of   sociability—interactional   metadiscourse—than   the   text  
comments,   considerations   of   novelty   and   familiarity   appear   to   have   influenced   mode  
   21  

choice  to  a  greater  extent  than  the  affordances  of  the  modes  themselves.  The  VoiceThread  
results  do  provide  compelling  evidence,  however,  that  once  a  mode  is  selected,  its  technical  
properties   influence   the   communicative   behavior   that   takes   place   in   that   mode.   In  
particular,   as   Kiesler   et   al.   (1984)   reported   long   ago   based   on   experimental   results,   text  
communication   tends   to   be   more   contentious.   This   has   also   often   been   observed   in  
discussion  forums  and  chatrooms:  At  the  same  time  that  textual  communicators  are  more  
socially  “hyperpersonal,”  they  tend  to  be  anti-­‐socially  disinhibited  (Walther,  1996).    
  However,   we   do   not   assume   that   this   is   because   text   is   depersonalized   or   that   the  
message   sender   lacks   awareness   of   the   addressee;   in   our   data,   text   messages   contained  
more  other-­‐awareness  markers  than  audio  or  video.  Rather,  we  subscribe  to  the  view  that  
asynchronous  textual  CMC  provides  a  buffer  of  temporal  and  spatial  distance  between  the  
sender  and  the  receiver  that  allows  senders  to  express  themselves  freely  without  having  to  
deal  with  the  receiver’s  immediate  responses.  In  this  respect,  textual  CMC  can  be  superior  
to  richer  media,  including  FTF  communication,  for  certain  sensitive  tasks,  such  as  firing  an  
employee   or   professing   one’s   love   (e.g.,   Sproull   &   Kiesler,   1992;   Walther,   1996).   At   the  
same   time   as   it   retains   these   medium   benefits,   we   propose   that   text   commenting   has  
become  “enriched”  through  decades  of  use  as  the  primary  mode  of  CMC,  such  that  it  is  now  
an   unmarked   mode   capable   of   fulfilling   a   wide   range   of   functions.   This   proposition   is  
supported   by   Sindoni   (in   press),   who   recently   surveyed   users   about   when   and   why   they  
choose   to   use   written   text   during   synchronous   videochats.     The   respondents   indicated   that  
they   used   text   for   precision,   secrecy,   and,   most   tellingly,   intimacy   or   to   express   face-­‐
threatening   feelings—a   varied   set   of   uses,   including   those   considered   by   the   media  
richness  theory  to  occupy  both  ends  of  the  “richness”  continuum.  
  Thus  rather  than  positing  FTF  communication  as  the  “universal  medium”  (Kim  et  al.,  
2007),   this   account   casts   text   as   the   default   mode   in   CMC   contexts.   It   is   in   the   marked  
modes,   asynchronous   audio   and   video   commenting,   that   the   technical   properties   of   the  
modes  appear  to  exercise  greater  influence,  presumably  because  their  novelty  leads  users  
to   notice   them   more.   On   VoiceThread.com,   the   immediate   reflection   audio   and   video  
provide   of   the   speakers   as   they   record   their   messages   appears   to   increase   the   speakers’  
self-­‐consciousness.  Similarly,  Sindoni  (2011)  observes  that  the  videochat  users  in  her  study  
displayed  self-­‐consciousness,  including  arranging  their  hair  as  if  their  video  image  were  a  
mirror,  when  communicating  via  video.  As  Sindoni  (in  press)  writes:    
Observing   oneself   during   a   conversation   produces   a   series   of   psychological   effects  
influencing   the   verbal   and   non   verbal   characteristics   of   the   online   exchange.   […  
C]asting   a   sidelong   glance   at   oneself   during   a   conversation   may   change,   if   not  
determine,  the  way  one  speaks,  gesticulates,  smiles  and  so  on.  (p.  10)  
While   this   is   especially   true   of   video,   audio   commenting   also   has   a   defamiliarizing   effect   in  
that  it  requires  the  commenter  to  speak  out  loud  to  a  computer,  which  can  feel  strange,  as  
anyone  who  has  tried  to  rehearse  a  presentation  without  an  immediate  audience  can  attest.  
The  resulting  self-­‐consciousness  can  affect  behavior  and  language  use.  

Conclusions  
This  study  asked  how  different  commenting  modes  are  used  on  a  multimodal  commenting  
platform,  VoiceThread.com.  The  results  suggest  that  text  commenting  is  the  default  on  the  
   22  

platform,  as  it  is  for  CMC  on  the  Internet  in  general.  However,  independent  of  the  reasons  
for   which   users   choose   to   comment   in   one   mode   or   another,   the   selection   of   a   mode  
predisposes   them   to   communicate   in   certain   ways.   In   particular,   multiple   modes   being  
available  and  all  other  factors  being  equal,  the  technical  properties  of  textual  CMC  appear  
to   encourage   openness,   brevity,   and   precision,   whereas   video   and   audio   increase   users’  
self-­‐consciousness  and  self-­‐focus.6  
  Other   factors   are   not   always   equal,   of   course.   Previous   experience   and   degree   of  
comfort  with  a  technology  may  affect  mode  choice.  Social  factors  such  as  gender  may  also  
condition  choice;  for  example,  peer  influence  appears  to  influence  women  more  than  men,  
and   men   tend   to   be   early   adopters   of   new   technology.   Moreover,   the   greater   social  
presence  of  audio  and  video  may  be  disadvantageous  for  users  in  contexts  where  they  do  
not  wish  their  gender  to  be  revealed.    
  Situational   factors   may   condition   mode   effects,   as   well.   The   lack   of   availability   of  
other  modes  to  choose  from  can  lead  an  otherwise  lean  channel  of  information  to  become  
enriched,  as  proposed  by  the  electronic  propinquity  theory.  Over  time,  users’  adaptations  
of  the  mode  to  serve  a  range  of  purposes  can  effectively  widen  the  scope  of  the  mode  itself,  
as  we  suggest  has  taken  place  for  textual  CMC.  Conversely,  some  of  the  self-­‐consciousness  
observed   in   the   present   study   in   video   and   audio   comments   may   be   attributable   to   the  
novelty   of   the   voice-­‐based   asynchronous   commenting   modes,   which   renders   their  
technological  properties  salient  to  users.    

Limitations  of  the  Study  and  Directions  for  Future  Research  

The   small,   heterogeneous   sample   analyzed   is   a   limitation   of   the   present   study.   The   three  
sample   VoiceThreads   proved   to   be   significantly   different   in   their   expressions   of   attitude,  
polarity,   and   metadiscourse,   as   well   as   in   message   length.   Future   study   of   VoiceThread  
could   scale   up   while   focusing   on   a   single   demographic   and   discussion   type,   such   as  
pedagogical   discussion   among   adults,   to   obtain   more   robust   and   consistent   findings.  
Further,  the  commenting  modes  on  VoiceThread  proved  not  to  be  equally  easy  to  use.  More  
invalid   messages   were   posted   in   video   and   audio   than   in   text;   this   could   have   biased   mode  
use   away   from   video   and   towards   text.   This   asymmetry   should   be   taken   into   account   in  
interpreting  the  results  of  this  study.  Finally,  our  analysis  did  not  directly  address  reasons  
for  mode  choice,  satisfaction  with  modes,  or  perceived  social  presence  in  different  modes,  
as   did   previous   experimental   research;   rather   we   focused   on   discourse   behavior   in   each  
mode.   As   a   consequence,   our   findings   are   not   directly   comparable   with   those   of   most  
previous   studies   of   cross-­‐mode   differences,   although   we   believe   that   legitimate   parallels  
can  be  inferred,  as  we  have  attempted  to  do  here.  
  Finally,   the   fact   that   VoiceThread   is   an   emergent   technology   and   most   users   are  
inexperienced  with  the  platform  means  that  the  snapshot  we  have  presented  here  is  likely  
ephemeral.   Our   analysis   leads   us   to   predict   that   communication   especially   in   the   novel  
commenting   modes,   asynchronous   audio   and   video,   will   evolve   as   users   become   more  
familiar  with  their  use.  Such  familiarity  will  likely  lead  to  more  strategic  usage  that  takes  
advantage  of  the  inherent  strengths  of  each  mode  but  that  is  also  susceptible  to  influence  
from  social  and  situational  factors  present  in  different  contexts  of  use.  It  will  be  interesting  
to  track  this  evolution  in  future  research.    
   23  

  The   results   of   this   study   have   theoretical   and   methodological   implications   for  
research   on   other   interactive   multimodal   platforms   that   offer   users   a   choice   of  
communication   modes,   such   as   YouTube   and   World   of   Warcraft,   as   well   as   sites   that  
combine   text   and   audio/video   more   generally.   Some   theoretical   implications   have   been  
explored   in   this   study   as   regards   media   richness.   Multimodal   platforms,   especially   those  
that   combine   entertainment   content   with   CMC,   also   raise   methodological   challenges:   the  
need  to  devise  parallel  transcription  and  visualization  displays  for  textual  and  non-­‐textual  
communication;  the  need  to  devise  new  analytical  methods  drawn  from  disciplines  outside  
communication   and   linguistics   (e.g.,   visual   semiotics);   and,   finally,   the   need   to  
conceptualize   approaches   that   capture   the   interaction   among   multiple   semiotic   systems.  
This   study   has   demonstrated   one   bottom-­‐up   approach   to   addressing   this   last   challenge,   by  
treating   modes   of   communication   as   building   blocks   of   a   larger   whole   and   seeking   to  
identify  the  effects  each  contributes.    

Notes  
1     The   term   ‘mode’   is   used   here   to   refer   to   a   CMC   option   within   a   single   platform,   as  
distinct   from   the   term   ‘medium,’   which   refers   to,   e.g.,   CMC   in   contrast   to   face-­‐to-­‐face  
communication  or  mediated  communication  hosted  on  a  non-­‐Internet  platform  (see  also  
Murray,  1988).  
2     Retrieved  August  18,  2011  from  http://www.appappeal.com/app/voicethread/  
3     The  contents  of  the  three  comments  are  as  follows.  Audio:  “Hey  ya’  guys,  welcome  to  ..  
VoiceThread.   I   want   you   to   watch   this   video   clip,   and   give   your   feedback,   either   through  
a  video  recording,  or  a  voice  recording.  It’s  a  short  clip  on  speeding,  and  I’m  interested  to  
hear   your   thoughts.”   Video:   “Hey   hey!   It’s   not   actually   Steve   Perry,   it’s   just   me   …  
checking  this  thing  out.  Uh,  this  video  clip  …  is  pretty  neat,  I  think  …  I  especially  enjoy  his  
Australian  accent,  and  I  know  it’s  Australian  bec-­‐  for  a  few  reasons,  uh  ..  the  biggest  one  
of  them  obviously  being  that  the  car’s  steering  wheel  …  is  on  the  right  hand  side  of  the  
car!     So.   Another   cool   thing   that   I   …   liked   about   this   video,   other   than   the   information   in  
it,   was   the   special   effects   they   used,   he-­‐   he’s   sorta   walked   out   in   front   of   the   car,   uh  
which   ..   I   wouldn’t   think   anybody   would   normally   do,   but   he   did   it!   And,   uh,   Steve   Perry  
–  come  back  for  us  all,  please!”  Text:  “People  may  think  that  they  have  a  quick  reaction  
time  but  in  fact  it  doesn't  matter  whether  you  can  break  fast  or  you  have  good  breaks  on  
the  car  its  what  speed  you  are  going  before  you  notice  something  that  either  has  run  out  
on  the  road   or   that   you   are   about   to   hit   something...be   aware   on   the   suroundings   [sic]  
and  do  the  speed  limit  or  this  could  be  you!!!”  
4     The   full   corpus   contains   363   comments,   of   which   43   were   invalid   because   they   were  
empty   or   contained   nothing   but   noise.   Invalid   comments   appeared   to   be   the   result   of  
problems   using   the   commenting   technology.   The   breakdown   by   mode   of   invalid  
comments   is:   video   (29%),   audio   (17%),   text   (7%).   The   thread   with   the   youngest  
participants,   SciFi,   had   the   greatest   proportion   of   invalid   comments   (20%),   compared  
with  the  teen  Speeding  thread  (8.4%)  and  the  adult  Network  thread  (1.8%).  
5     Contributing  to  this  result  is  the  fact  that  11  out  of  29  of  the  female  audio  comments  to  
the  Network  thread,  which  contained  the  most  and  longest  audio  comments,  appeared  to  
   24  

be   scripted   in   advance,   resulting   in   speech   that   sounded   more   written-­‐like   than   spoken-­‐
like.  Engagement  markers  are  common  in  unplanned  speech.  
6     Differences   can   also   be   observed   across   modes   that   are   due   to   differences   between  
speaking  and  writing.  For  example,  audio  and  (especially)  video  comments,  because  they  
typically   are   not   planned   in   advance,   contain   more   false   starts,   pause   fillers   (um,   hm,  
etc.),   and   phatic   communication   (such   as   greetings   and   closings)   than   does   text,   and  
voice   comments   tend   to   be   more   rambling   and   less   concise,   as   can   be   observed   in   the  
examples  in  note  3.  (But  cf.  note  5.)  

References  
Barrett,   E.,   &   Lally,   V.   (1999).   Gender   differences   in   an   on-­‐line   learning   environment.  
Journal  of  Computer  Assisted  Learning,  15,  48-­‐60.  
Baym,   N.,   Zhang,   Y.,   &   Lin,   M.   (2004).   Social   interactions   across   media:   Interpersonal  
communication  on  the  Internet,  telephone  and  face-­‐to-­‐face.  New  Media  &  Society,  6(3),  
299–318.  
Coates,  J.  (2003).  Women,  men,  and  language,  2nd  ed.  London:  Longman.  
Cooper,   A.,   &   Sportolari,   L.   (1997).   Romance   in   cyberspace:   Understanding   online  
attraction.  Journal  of  Sex  Education  and  Therapy,  22,  7-­‐14.    
Daft,   R.,   &   Lengel,   R.   H.   (1984).   Information   richness:   A   new   approach   to   managerial  
behavior  and  organizational  design.  In  B.  M.  Staw  &  L.  L.  Cummings  (Eds.),  Research  in  
organizational  behavior  (pp.  191-­‐233).  Greenwich,  CT:  JAI  Press.    
Frankel,  K.  A.  (1990).  Women  and  computing.  Communications  of  the  ACM,  33(11),  34-­‐45.    
Gefen,  D.,  &  Straub,  D.  W.  (1997).  Gender  differences  in  the  perception  and  use  of  e-­‐mail:  An  
extension  to  the  Technology  Acceptance  Model.  MIS  Quarterly,  21(4),  389-­‐400.  
Herring,   S.   C.   (1993).   Gender   and   democracy   in   computer-­‐mediated   communication.  
Electronic   Journal   of   Communication,   3   (2).   Retrieved   August   30,   2011   from  
http://ella.slis.indiana.edu/~herring/ejc.txt  
Herring,  S.  C.  (1994).  Politeness  in  computer  culture:  Why  women  thank  and  men  flame.  In:  
Cultural   Performances:   Proceedings   of   the   Third   Berkeley   Women   and   Language  
Conference  (pp.  278-­‐294).  Berkeley  Women  and  Language  Group.    
Herring,  S.  C.  (1995).  Men’s  language  on  the  Internet.  Nordlyd,  23,  1-­‐20.    
Herring,   S.   C.   (2003).   Gender   and   power   in   online   communication.   In   J.   Holmes   &   M.  
Meyerhoff   (Eds.),   The   handbook   of   language   and   gender   (pp.   202-­‐228).   Oxford:  
Blackwell.    
Herring,  S.  C.,  Johnson,  D.,  &  DiBenedetto,  T.  (1998).  Participation  in  electronic  discourse  in  
a   "feminist"   field.   In   J.   Coates   (Ed.),   Language   and   gender:   A   reader   (pp.   197-­‐210).  
Oxford:  Blackwell.  
Herring,   S.   C.,   &   Paolillo,   J.   C.   (2006).   Gender   and   genre   variation   in   weblogs.   Journal   of  
Sociolinguistics,  10(4),  439-­‐459.    
   25  

Hyland,  K.  (2005).  Metadiscourse.  Exploring  interaction  in  writing.  Oxford:  Continuum.    
Kiesler,   S.,   Siegel,   J.   &   McGuire,   T.   W.   (1984)   Social   psychological   aspects   of   computer-­‐
mediated  communication.  American  Psychologist,  39,  1123-­‐1134.  
Kim,   H.,   Kim,   G.   J.,   Park,   H.   W.,   &   Rice,   R.   E.   (2007).   Configurations   of   relationships   in  
different   media:   FtF,   email,   instant   messenger,   mobile   phone,   and   SMS.   Journal   of  
Computer-­‐Mediated   Communication,   12(4),   article   3.  
http://jcmc.indiana.edu/vol12/issue4/kim.html  
Kim,   M-­‐S.,   &   Raja,   N.   S.   (1990).   Verbal   aggression   and   self-­‐disclosure   on   computer   bulletin  
boards.  ERIC  document  (ED334620).  
Korzenny,   F.   (1978).   A   theory   of   electronic   propinquity:   Mediated   communications   in  
organizations.  Communication  Research,  5,  3-­‐24.  
Martin,  J.  R.,  &  White,  P.  R.  R.  (2005).  The  language  of  evaluation:  Appraisal  in  English.  New  
York:  Palgrave  Macmillan.    
Millard,   M.   (2010).   Analysis   of   interaction   in   an   asynchronous   CMC   environment.   Web  
Science   Conf.   2010,   April   26-­‐27,   Raleigh,   NC.   Retrieved   August   18,   2011   from  
http://journal.webscience.org/391/2/websci10_submission_106.pdf  
Murray,   D.   E.   (1988).   The   context   of   oral   and   written   language:   A   framework   for   mode   and  
media  switching.  Language  in  Society,  17,  351-­‐373.  
Rheingold,   H.   (1993).   The   virtual   community:   Homesteading   on   the   electronic   frontier.  
Reading,  MA:  Addison  Wesley.  
Rice,   R.   E.,   &   Love,   G.   (1987).   Electronic   emotion:   Socioemotional   content   in   a   computer-­‐
mediated  communication  network.  Communication  Research,  14(1),  85-­‐108.  
Short,   J.,   Williams,   E.,   &   Christie,   B.   (1976).   The   social   psychology   of   telecommunications.  
London:  Wiley.  
Sindoni,  M.  G.  (2011).  "Mode-­‐switching":  Speech  and  writing  in  videochats.  Paper  presented  
at  GURT  2011:  Discourse  2.0:  Language  and  New  Media.  Washington,  DC,  March  11.    
Sindoni,   M.   G.   (in   press,   2012).   Through   the   looking   glass:   A   social   semiotic   introduction   to  
the  study  of  videochats.  In  G.   Kress,  E.   Adami,   &   R.   Facchinetti   (Eds.),  The   challenge   of  
multimodal  analysis  to  notions  of  text  [special  issue  of  Text  &  Talk].  
Sproull,   L.,   &   Kiesler,   S.   (1992).   Connections:   New   ways   of   working   in   the   networked  
organization.  Cambridge,  MA:  MIT  Press.  
Venkatesh,   V.,   &   Morris,   M.   G.   (2000).   Why   do   not   men   ever   stop   to   ask   for   directions?  
Gender,  social  influence,  and  their  role  in  technology  acceptance  and  usage  behavior,  
MIS  Quarterly,  24(1),  115–139.  
Walther,   J.   B.   (1992).   Interpersonal   effects   in   computer-­‐mediated   interaction:   A   relational  
perspective.    Communication  Research,  19,  52-­‐90.  
Walther,  J.  B.  (1996).  Computer-­‐mediated  communication:  Impersonal,  interpersonal,  and  
hyperpersonal  interaction.  Communication  Research,  23,  1-­‐43.    
   26  

Walther,  J.  B.  (in  press).  Visual  cues  in  computer-­‐mediated  communication:  Sometimes  less  
is   more.   In   A.   Kappas   &   N.   Krämer   (Eds.),   Face-­‐to-­‐face   communication   over   the  
Internet:  Issues,  research,  challenges.  Cambridge,  UK:  Cambridge  University  Press.  
Walther,  J.  B.,  &  Bazarova,  N.  N.  (2008).  Validation  and  application  of  electronic  propinquity  
theory   to   computer-­‐mediated   communication   in   groups.   Communication   Research,  
35(5),  622-­‐645.  
Weir,   L.   (2008,   April   16).   Edutopia:   Voicethread   Extends   the   Classroom   with   Interactive  
Multimedia   Albums.   Retrieved   November   10,   2009   from:  
http://www.edutopia.org/Voicethread-­‐interactive-­‐  multimedia-­‐albums