Iconicity  in  the  input   1   Bibliographic Information: Perniss, P., Lu, J., Morgan, G., & Vigliocco, G. (2017). The Role of Iconicity in Linking Language to World: Evidence from Sign Language Input. Developmental Science, 1-23 Pamela  Perniss   School  of  Humanities,  University  of  Brighton   Jenny  C.  Lu   Department  of  Psychology,  University  of  Chicago   Gary  Morgan   Department  of  Language  and  Communication  Science,  City  University  London   Gabriella  Vigliocco   Division  of  Psychology  and  Language  Sciences,  University  College  London   Acknowledgements   This  work  was  supported  by  the  Economic  and  Social  Research  Council  (ESRC)  of  Great  Britain:   grant  no.  RES1 6201 281 6002  to  the  Deafness,  Cognition  and  Language  Research  Centre   (DCAL)  and  the  Susan  Rappaport  Knafel  ’52  Fellowship  to  Jenny  C.  Lu.   The research reported here was also supported by the Institute of Education Sciences, U.S. Department of Education, through Grant Number: R305B140048 at the University of Chicago. The opinions expressed are those of the authors and do not represent views of the Institute or the U.S. Department of Education. We  thank  Rachel  England  for  data  collection,  and  Sannah  Gulamani  and  Neil  Fox  for  conducting   reliability  coding  of  the  data.  We  thank   Asli   Özyürek   and   Emanuela   Campisi   for   discussion   regarding   the   methodology   an d  stim ulus  m aterials. Iconicity  in  the  input   2   Research  Highlights   • Child-­‐directed  signing  exploits  iconicity,  especially  when  referents  are  not present  (non-­‐ostensive  contexts). • Child-­‐directed  signing  uses  pointing  and  iconicity  in  a  complementary fashion. • Results  are  consistent  with  findings  that  iconicity  can  support  referential mapping:  if  iconicity  is  present  in  the  input,  it  is  available  for  use  by  the child. • Iconicity  may  be  an  important  strategy  supporting  referential  mapping, qualitatively   different   from   other   strategies   in   language   learning   and hitherto  underexplored  in  its  potential. Iconicity  in  the  input   3   Abstract   Most  research  on  the  mechanisms  underlying  referential  mapping  has  assumed   that  learning  occurs  in  ostensive  contexts,  where  label  and  referent  co-­‐occur,  and   that  form  and  meaning  are  linked  by  arbitrary  convention  alone.  In  the  present   study,  we  focus  on  iconicity  in  language,  i.e.  resemblance  relationships  between   form  and  meaning,  and  on  non-­‐ostensive  contexts,  where  label  and  referent  do   not   co-­‐occur.   We   approach   the   question   of   language   learning   from   the   perspective  of  the  language  input.  Specifically,  we  look  at  child-­‐directed  language   (CDL)  in  British  Sign  Language  (BSL),  a  language  rich  in  iconicity  due  to  the   affordances   of   the   visual   modality.   We   ask   whether   child-­‐directed   signing   exploits  iconicity  in  the  language  by  highlighting  the  similarity  mapping  between   form  and  referent.  We  find  that  CDL  modifications  occur  more  often  with  iconic   signs  than  with  non-­‐iconic  signs.  Crucially,  for  iconic  signs,  modifications  are   more   frequent   in   non-­‐ostensive   contexts   than   in   ostensive   contexts.   Furthermore,  we  find  that  pointing  dominates  in  ostensive  contexts,  and  suggest   that  caregivers  adjust  the  semiotic  resources  recruited  in  CDL  to  context.  These   findings  offer  first  evidence  for  a  role  of  iconicity  in  the  language  input  and   suggest   that   iconicity   may   be   involved   in   referential   mapping   and   language   learning,  particularly  in  non-­‐ostensive  contexts. Iconicity  in  the  input   4   Introduction   Understanding  language  development  remains  one  of  the  outstanding  challenges   of   research   in   the   language   sciences.   The   process   of   referential   mapping   –   making  correct  associations  between  form  and  meaning  –  is  a  complex  task,  yet   children   learn   form-­‐meaning   mappings   prodigiously.   An   extensive   body   of   research  has  been  dedicated  to  understanding  how  they  do  so.  Some  proposals   focus  on  child-­‐internal  mechanisms,  e.g.  innate  biases  (such  as  the  whole  object   bias;  mutual  exclusivity  bias)  that  guide  learning  (Markman  &  Wachtel,  1988;   Waxman  &  Booth,  2001)  and  powerful  capacities  for  statistically-­‐driven  cross-­‐ situational  learning  (Frank,  Goodman  &  Tenenbaum,  2009;  Smith  &  Yu,  2008).   Other  proposals  emphasise  features  of  the  communicative  context,  notably  the   role   of   joint   attention   in   establishing   common   ground   and   understanding   communicative  intentionality  (Tomasello,  1999;  Tomasello  &  Carpenter,  2007).   Child-­‐directed   language   (CDL)   –   characteristic   modifications   to   language   production   when   communicating   with   children   and   used   across   languages,   cultures,  and  language  modalities  –  has  furthermore  been  argued  to  support   referential  mapping  by  engaging  attention  and  facilitating  word  segmentation   (for   spoken   language,   Fernald   et   al.,   1989;   Thiessen   et   al.,   2005;   for   signed   language,  Masataka,  1992;  Pizer,  Meier  &  Points,  2011).  Finally,  the  coordination   of  object  naming  with  object  individuation  has  important  effects:  providing  a   label   while   pointing   to   a   referent   has   been   shown   to   be   correlated   with   children’s  vocabulary  (Iverson  et  al.,  1999;  O’Neill  et  al.,  2005);  providing  a  label   while  the  child  is  holding  and  visually  isolating  a  referent  has  also  been  shown  to   facilitate  referential  mapping  (Yu  &  Smith,  2012). Iconicity  in  the  input   5   Despite  their  diversity,  these  proposals  share  two  critical  assumptions   about  the  nature  of  the  vocabulary-­‐learning  task.  The  first  of  these  is  that  label   and   referent   are   linked   by   arbitrary   convention   alone,   reflecting   the   long-­‐ standing   tenet   of   arbitrariness   as   a   fundamental   design   feature   of   language   (Saussure,  1916;  Greenberg,  1957;  Hockett,  1960).  The  second  is  the  assumption   that  learning  occurs  in  ostensive  contexts,  where  the  co-­‐occurrence  of  label  and   referent   is   essential   to   association   mechanisms   that   link   form   and   meaning   through  temporal  binding  (Glenberg  &  Gallese,  2012).   In  this  paper,  we  explore  an  alternative  proposal:  First,  we  assume  that   language  (both  spoken  and  signed)  is  also  fundamentally  iconic,  i.e.  showing   resemblance  relationships  between  form  and  meaning,  as  exhibited  to  varying   degrees  in  the  lexicon,  as  well  as  in  co-­‐speech  gesture  and  in  prosody,  in  addition   to  arbitrary  (see  Dingemanse,  Blasi,  Lupyan,  Christiansen  &  Monaghan,  2015;   Lockwood  &  Dingemanse,  2015;  McNeill,  1992;  Perniss,  Thompson  &  Vigliocco,   2010;  Perniss  &  Vigliocco,  2014  for  overviews;  see  Liddell  2003;  Taub  2001  for   elaboration  of  the  centrality  of  iconicity  in  signed  language).  Thus,  we  propose   that  iconicity  in  the  language  input  may  provide  a  powerful  cue  to  referentiality,   allowing  the  child  to  identify  a  referent  from  aspects  of  the  communicative  form   itself  (e.g.  in  using  an  onomatopoeic  word  such  as  choo-­‐choo  to  refer  to  a  train,   the  link  between  label  and  referent  is  more  direct  and  transparent).     Second,   we   extend   questions   concerning   language   learning   to   non-­‐ ostensive  contexts,  where  label  and  referent  do  not  co-­‐occur  in  the  immediate   environment  (Jaswal  &  Markman,  2003;  Tomasello  &  Barton,  1994).  Parents   often  engage  with  their  children  in  talk  about  the  not  immediately  here-­‐and-­‐now   (e.g.   the   trip   to   the   park   yesterday),   and   such   contexts   provide   important Iconicity  in  the  input   6   opportunities  for  learning  of  words,  especially  referring  to  actions,  events,  and   properties.     A  growing  body  of  recent  research  suggests  that  iconicity  plays  a  role  in   language  development.  Language  learners  at  different  ages,  and  as  young  as  4   months  old,  have  been  shown  to  be  sensitive  to  sound-­‐symbolic  associations   (Asano  et  al.,  2015;  Ozturk  et  al.,  2013;  Maurer  et  al.,  2006;  Yoshida,  2012)  and   these  iconic  mappings  have  been  argued  to  bootstrap  children’s  word  acquisition   (Imai  et  al.,  2008;  Imai  &  Kita,  2014;  Kantartzis  et  al.,  2011;  Monaghan  et  al.,   2014;  Perry,  Perlman  &  Lupyan,  2015).  Further  evidence  that  iconicity  has  a  role   in   language   learning   comes   from   findings   that   children’s   early   vocabularies   exhibit  a  preponderance  of  iconic  forms.  For  example,  studies  looking  at  lexical   development   in   German   have   found   that   onomatopoeic   words   make   up   a   substantial  proportion  (up  to  40%,  Laing,  2014)  of  early  verbal  output,  and  then   decrease  as  the  use  of  more  conventional  word  categories  becomes  dominant   (Kauschke  &  Hofmeister,  2002;  Laing,  2014).  For  signed  language,  Thompson  et   al.  (2012)  found  that  iconicity  predicts  both  sign  production  and  comprehension   in  deaf  children  aged  11-­‐30  months  learning  BSL,  after  other  variables  (e.g.   phonological   complexity   of   the   signs)   are   taken   into   account   (contra   earlier   studies,  e.g.  Orlansky  &  Bonvillian,  1984,  which  did  not  find  learning  effects  for   iconic   signs,   but   which   were   less   well-­‐controlled   for   these   variables,   see   Thompson  et  al.,  2012  for  discussion).     For  iconicity  to  be  used  by  the  child,  it  has  to  be  present  in  the  input.   Indeed,  there  is  some  evidence  that  caregivers  make  increased  use  of  iconic   forms  in  child-­‐directed  language  (CDL).  This  has  been  found  for  Japanese,  a   language  with  a  rich  inventory  of  sound-­‐symbolic  forms  (Fernald  &  Morikawa, Iconicity  in  the  input   7   1993;  Toda,  Fogel  &  Kawai  1990;  Yoshida,  2012).  In  addition,  there  is  some   evidence  that  caregivers’  use  of  CDL  features  –  exaggerated  intonation,  slower   articulation  –  is  particularly  salient  for  onomatopoeic  words  compared  to  non-­‐ onomatopoeic   words   (Laing,   Vihman   &   Keren-­‐Portnoy,   2016;   Sundberg   &   Klintfors,   2009)   and   that   CDL   modifications   correlate   with   and   are   used   by   caregivers  to  highlight  properties  of  meaning  (e.g.  in  domains  indicating  size,   strength  and  valence;  Herold,  Nygaard  &  Namy  2011;  Nygaard,  Herold  &  Namy,   2009).       Present  study   Here  we  investigate  iconicity  in  the  input  in  British  Sign  Language  (BSL).  Sign   languages  are  notable  for  exhibiting  a  high  proportion  of  iconicity  in  the  lexicon,   compared  to  the  relative  paucity  of  iconicity  in  spoken  languages  (Taub  2001).   The  visual  nature  of  sign  languages  affords  iconic  depiction  of  a  wide  range  of   information   that   is   visually   perceived   or   motorically   experienced   (e.g.   what   things  look  like,  how  they  are  used,  where  they  are,  how  they  are  moving).   Estimates   range   between   one-­‐third   and   one-­‐half   of   signs   in   the   lexicon   of   different  sign  languages  exhibiting  some  degree  of  iconicity  (e.g.  Boyes  Braem   1986;  Zeshan  2000).  Moreover,  iconic  signs  predominate  in  reference  to  objects   and  actions  (very  common  in  child  directed  language).  Caregivers,  therefore,   have   plenty   of   opportunities   to   further   increase   the   salience   of   iconically-­‐ mapped  features,  thereby  maximising  the  imagistic  link  with  referents.  More   specifically,  we  suggest  that  caregivers  may  bootstrap  referential  mapping  by   modifying  iconic  signs  to  make  the  iconic  properties  more  salient  (similar  to   increased  CDL  modifications  in  onomatopoeia  in  spoken  language,  Laing  et  al. Iconicity  in  the  input   8   2016).   We   further   suggest   that   the   role   of   iconicity   may   be   particularly   important   in   non-­‐ostensive   contexts,   where   form-­‐meaning   resemblance   may   help  identify  a  referent  from  language  even  when  the  referent  cannot  be  directly   attended  to.  Whereas  in  ostensive  contexts,  pointing  to  the  object  can  support   referential  mapping,  in  non-­‐ostensive  contexts,  exaggeration  of  iconic  features  of   signs  may  help  bring  the  corresponding  properties  of  referents  to  the  mind’s  eye.   Thus,  in  asking  whether  iconicity  is  prominent  in  BSL  language  input,  we  are  not   looking  for  the  presence  of  iconicity  per  se.  Rather,  we  look  at  whether  and   under   what   conditions   caregivers   modify   sign   productions   in   child-­‐directed   language   to   increase   the   salience   of   those   conceptual   properties   that   are   imagistically  evoked  by  iconic  signs.   We  focus  on  modifications  typical  of  child-­‐directed  signing  (Pizer  &  Meier,   2008;  Reilly  &  Bellugi,  1996;  see  Coding  section  for  details).  CDL  modification  is   one  of  the  strategies  used  by  caregivers  to  scaffold  the  development  of  sign-­‐to-­‐ world  mappings.  Here  we  predict  that  CDL  modifications  will  be  particularly   prevalent  in  iconic  signs  (e.g.  the  BSL  sign  DRIVE  in  Figure  2A)  compared  to  non-­‐ iconic  signs  (e.g.  the  BSL  sign  for  PLAY  in  Figure  2B).  Crucially,  in  iconic  signs,   the  increased  salience  provided  by  CDL  modification  typically  emphasises  the   iconic  aspect  of  signs  (e.g.  the  shape  and  movement  of  the  steering  wheel  in  the   sign  DRIVE),   thereby  specifically  highlighting  the  element  that  describes  the   similarity   between   the   form   and   referent.   In   addition,   we   predict   that   CDL   modifications   highlighting   iconic   mappings   may   feature   particularly   in   caregivers’   language   when   referents   being   talked   about   are   absent   (non-­‐ ostensive  contexts)  compared  to  when  referents  are  present  (ostensive  contexts)   because  of  the  potential  for  iconicity  to  render  conceptual  properties  of  referents Iconicity  in  the  input   9   readily  available.  In  addition,  we  look  at  the  use  of  pointing  as  a  strategy  for   singling  out  a  referent.  We  expect  pointing  to  occur  more  in  ostensive  contexts   (when  referents  are  present,  and  the  co-­‐occurrence  of  point  and  referent  can   scaffold   referential   mapping)   compared   to   non-­‐ostensive   contexts   (when   referents  are  absent;  see  Coding  section  on  the  availability  of  pointing  in  non-­‐ ostensive  contexts).       Method   Participants   Ten  participants  were  recruited  from  the  greater  London  area  (8  female).  All   participants   were   deaf,   fluent   signers   of   BSL,   and   used   BSL   as   the   primary   language   of   communication   with   their   deaf   (three   participants)   or   hearing   (seven  participants)  children.  The  average  age  of  participants’  children  was  3;2   years  (38  months),  ranging  from  2;1  years  (25  months)  to  4;3  years  (51  months).     Materials   The  materials  used  in  the  task  consisted  of  four  toy  sets:  (1)  farm  animals;  (2)   cooking  set;  (3)  doctor’s  kit;  and  (4)  tool  bench  (see  Figure  1).  Toy  set  selection   was   based   on   the   presence   of   multiple   individual   parts   that   encouraged   manipulation  and  narrative  construction  (e.g.  visit  to  the  doctor)  and  that  had   clear  labels  (e.g.  different  animals)  exhibiting  perceptuo-­‐motor  properties  (e.g.   handling  affordances  of  tools).  The  toy  sets  were  age  appropriate  and  novel  to   participants. Iconicity  in  the  input   10   (a)  (b)  (c)  (d)     Figure  1.  Toy  sets  used  in  the  task:  (a)  Farm  animals,  (b)  Cooking,  (c)  Doctor’s  kit,  and  (d)  Tool   bench.     Procedure   Participants  were  asked  to  imagine  playing  with  their  child  in  two  conditions,   without  the  child  being  present.  This  methodology  avoids  issues  of  feedback  and   interaction  that  are  difficult  to  control,  while  still  maintaining  ecological  validity.   Our  decision  to  employ  a  methodology  where  participants  imagined  playing  with   their  child  was  further  motivated  by  the  desire  to  obtain  data  that  was  not   affected  by  local  adaptation  to  a  present  addressee  (Brennan,  Galati  &  Kuhlen   2010).  Similar  methodology  has  been  successfully  employed  in  studies  on  co-­‐ speech   gesture   use,   where   participants   were   asked   to   imagine   talking   to   different   kinds   of   addressees   (Bavelas,   Coates   &   Johnson   2002;   Campisi   &   Özyürek  2013).  Nygaard  et  al.  (2009)  offer  evidence  that  CDL  modification  of   speech   is   reliable   in   contexts   without   a   real-­‐life   addressee:   participants   instructed  to  employ  CDL  in  a  sentence  production  task  showed  remarkable   overlap  in  their  use  of  intonation.  In  addition,  Sachs  &  Devin  (1976)  found  that   children   used   CDL   when   talking   to   a   baby,   but   not   an   adult,   and   found   no   difference   in   speech   between   talking   to   a   real   baby   vs.   a   baby   doll.   In   the   Ostensive  condition,  caregivers  used  and  interacted  with  the  toy  sets  during  the   session.  In  the  Non-­‐ostensive  condition,  caregivers  imagined  playing  with  and   talking  about  the  toys  with  their  child.  Caregiver  strategy  for  addressing  the