Microsoft Word - Successful visual epistemic representation.docx


Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   1	
  

	
  Successful	
  Visual	
  Epistemic	
  Representation	
  

Agnes	
  Bolinska	
  

	
  

Abstract:	
  In	
  this	
  paper,	
  I	
  characterize	
  	
  visual	
  epistemic	
  representations	
  as	
  concrete	
  two-­‐	
  or	
  

three-­‐dimensional	
  tools	
  for	
  conveying	
  information	
  about	
  aspects	
  of	
  their	
  target	
  systems	
  or	
  

phenomena	
  of	
  interest.	
  I	
  outline	
  two	
  features	
  of	
  successful	
  visual	
  epistemic	
  representation:	
  

that	
  the	
  vehicle	
  of	
  representation	
  contain	
  sufficiently	
  accurate	
  information	
  about	
  the	
  

phenomenon	
  of	
  interest	
  for	
  the	
  user’s	
  purpose,	
  and	
  that	
  it	
  convey	
  this	
  information	
  to	
  the	
  

user	
  in	
  a	
  manner	
  that	
  makes	
  it	
  readily	
  available	
  to	
  her.	
  I	
  argue	
  that	
  actual	
  epistemic	
  

representation	
  may	
  involve	
  tradeoffs	
  between	
  these	
  features	
  and	
  is	
  successful	
  to	
  the	
  

extent	
  that	
  they	
  are	
  present.	
  

	
  

1.	
  Introduction	
  

How	
  do	
  scientists	
  gain	
  information	
  about	
  a	
  physical	
  system?	
  The	
  most	
  

straightforward	
  way	
  would	
  be	
  to	
  examine	
  the	
  system	
  using	
  their	
  unaided	
  senses.	
  But	
  the	
  

senses	
  often	
  do	
  not	
  suffice	
  for	
  observation	
  because	
  the	
  system	
  is	
  too	
  small	
  (like	
  a	
  

molecule),	
  too	
  distant	
  (like	
  a	
  star),	
  too	
  dispersed	
  (like	
  a	
  population),	
  or	
  otherwise	
  

imperceptible.	
  In	
  such	
  cases,	
  scientists	
  may	
  use	
  instruments	
  to	
  facilitate	
  their	
  investigation.	
  

But	
  once	
  they	
  have	
  exhausted	
  all	
  observational	
  avenues	
  –	
  once	
  they	
  have	
  done	
  everything	
  

possible	
  to	
  gain	
  information	
  from	
  the	
  system	
  itself	
  –	
  they	
  usually	
  have	
  only	
  completed	
  the	
  

first	
  phase	
  of	
  their	
  investigation.	
  For	
  scientists	
  are	
  generally	
  not	
  interested	
  in	
  particular	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   2	
  

measurements,	
  but	
  in	
  generalizations	
  or	
  patterns	
  that	
  can	
  be	
  inferred	
  from	
  them.1	
  These	
  

patterns,	
  which	
  I	
  will	
  refer	
  to	
  as	
  phenomena	
  of	
  interest,	
  are	
  rarely	
  accessible	
  through	
  direct	
  

observation.2	
  So	
  the	
  question	
  is:	
  once	
  all	
  available	
  observational	
  data	
  has	
  been	
  acquired,	
  

how	
  can	
  scientists	
  use	
  a	
  vehicle	
  of	
  representation	
  –	
  an	
  entity	
  physically	
  separate	
  from	
  the	
  

system	
  it	
  represents	
  –	
  as	
  a	
  tool	
  for	
  gaining	
  information	
  about	
  the	
  phenomenon	
  of	
  interest?	
  	
  

	
   The	
  answer,	
  I	
  will	
  argue,	
  is	
  by	
  representing	
  the	
  system	
  in	
  a	
  way	
  that	
  makes	
  these	
  

patterns	
  perspicuous	
  to	
  the	
  user.	
  I	
  call	
  vehicles	
  of	
  representation	
  that	
  are	
  used	
  as	
  tools	
  for	
  

gaining	
  information	
  about	
  phenomena	
  of	
  interest	
  epistemic	
  representations.	
  As	
  I	
  will	
  show,	
  

these	
  include	
  not	
  only	
  scientific	
  representations,	
  but	
  also	
  other	
  representations	
  that	
  are	
  

used	
  in	
  similar	
  ways	
  outside	
  of	
  scientific	
  practice.	
  By	
  bringing	
  the	
  features	
  of	
  interest	
  to	
  

the	
  fore,	
  an	
  epistemic	
  representation	
  unlocks	
  for	
  the	
  user	
  a	
  dimension	
  of	
  access	
  to	
  the	
  

phenomenon	
  of	
  interest	
  that	
  she	
  wouldn’t	
  otherwise	
  have.	
  	
  

	
   There	
  are	
  two	
  sorts	
  of	
  contexts	
  in	
  which	
  epistemic	
  representations	
  may	
  be	
  used:	
  

those	
  in	
  which	
  little	
  or	
  nothing	
  is	
  known	
  about	
  the	
  phenomenon	
  of	
  interest,	
  and	
  the	
  

representation	
  functions	
  as	
  an	
  investigative	
  tool;	
  and	
  those	
  in	
  which	
  the	
  creator	
  of	
  the	
  

representation	
  already	
  understands	
  the	
  phenomenon	
  of	
  interest	
  fairly	
  well	
  and	
  uses	
  it	
  as	
  a	
  

tool	
  for	
  conveying	
  information	
  about	
  this	
  phenomenon	
  via	
  testimony.	
  In	
  this	
  paper,	
  I	
  focus	
  

on	
  the	
  latter	
  sort	
  of	
  context	
  and	
  limit	
  my	
  attention	
  to	
  two-­‐	
  and	
  three-­‐dimensional	
  visual	
  

representations.	
  I	
  investigate	
  the	
  features	
  in	
  virtue	
  of	
  which	
  such	
  representations	
  convey	
  

	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
1	
  The	
  point	
  that	
  scientists	
  tend	
  to	
  be	
  interested	
  in	
  patterns	
  or	
  regularities	
  has	
  been	
  nicely	
  articulated	
  by	
  
Batterman	
  (2009,	
  pp.	
  429-­‐30).	
  
2	
  Why	
  this	
  is	
  so	
  will	
  become	
  clear	
  shortly	
  through	
  the	
  consideration	
  of	
  examples	
  (Section	
  2).	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   3	
  

information	
  to	
  their	
  users	
  about	
  phenomena	
  of	
  interest	
  that	
  they	
  wouldn't	
  otherwise	
  have.	
  

That	
  is,	
  I	
  determine	
  the	
  features	
  of	
  successful	
  representation	
  for	
  these	
  kinds	
  of	
  cases.3	
  	
  	
  

	
   In	
  order	
  to	
  understand	
  what	
  is	
  required	
  for	
  successful	
  visual	
  epistemic	
  

representation,	
  two	
  questions	
  must	
  be	
  addressed.	
  First,	
  what	
  kind	
  of	
  information	
  ought	
  

the	
  vehicle	
  of	
  representation	
  contain,	
  and	
  what	
  is	
  required	
  for	
  it	
  to	
  contain	
  this	
  

information?	
  Second,	
  how	
  is	
  this	
  information	
  effectively	
  conveyed	
  to	
  the	
  user?	
  I	
  will	
  argue	
  

that	
  an	
  epistemic	
  representation	
  is	
  successful	
  to	
  the	
  extent	
  that	
  it	
  contains	
  sufficiently	
  

accurate	
  information	
  about	
  the	
  phenomenon	
  of	
  interest	
  for	
  the	
  user’s	
  purpose	
  (Section	
  3)	
  

and	
  is	
  able	
  to	
  convey	
  this	
  information	
  to	
  the	
  user	
  in	
  a	
  manner	
  that	
  makes	
  it	
  readily	
  

available	
  to	
  her	
  (Section	
  4).	
  I	
  will	
  show	
  that	
  because	
  visual	
  epistemic	
  representation	
  often	
  

involves	
  tradeoffs	
  between	
  these	
  two	
  features,	
  the	
  success	
  of	
  such	
  representation	
  is	
  

determined	
  by	
  how	
  well	
  they	
  are	
  balanced	
  (Section	
  5).	
  But	
  I	
  will	
  begin	
  (Section	
  2)	
  by	
  

outlining	
  three	
  examples	
  of	
  successful	
  visual	
  epistemic	
  representation	
  and	
  showing	
  that	
  

they	
  share	
  two	
  general	
  features	
  that	
  inform	
  the	
  more	
  specific	
  features	
  considered	
  in	
  

Sections	
  3	
  and	
  4:	
  they	
  are	
  user-­‐	
  and	
  purpose-­‐specific.	
  

	
  

2.	
  Examples	
  and	
  general	
  features	
  

User-­‐	
  and	
  purpose-­‐specificity	
  are	
  widely	
  accepted	
  features	
  that	
  comprise	
  a	
  central	
  

part	
  of	
  several	
  accounts	
  of	
  scientific	
  representation	
  (cf.	
  Bailer-­‐Jones,	
  2003;	
  Giere,	
  2004,	
  

2009;	
  Mäki,	
  2009;	
  Teller,	
  2001).	
  I	
  highlight	
  these	
  features	
  here	
  to	
  provide	
  a	
  foundation	
  

	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
3	
  Many	
  authors	
  (cf.	
  Callender	
  &	
  Cohen,	
  2006;	
  Contessa,	
  2007;	
  Suárez,	
  2004)	
  focus	
  on	
  understanding	
  mere	
  –	
  
that	
  is,	
  not	
  necessarily	
  true	
  or	
  accurate	
  –	
  representation.	
  I	
  think	
  that	
  understanding	
  successful	
  epistemic	
  
representation	
  (which,	
  as	
  I	
  will	
  show,	
  isn't	
  just	
  true	
  or	
  accurate	
  representation)	
  can	
  also	
  guide	
  us	
  in	
  
understanding	
  'mere'	
  representation.	
  While	
  a	
  full	
  discussion	
  of	
  why	
  this	
  is	
  so	
  is	
  beyond	
  the	
  scope	
  of	
  this	
  
paper,	
  it	
  will	
  become	
  apparent	
  once	
  the	
  account	
  has	
  been	
  put	
  forth.	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   4	
  

from	
  which	
  the	
  remainder	
  of	
  my	
  analysis	
  may	
  be	
  developed.	
  By	
  taking	
  the	
  user-­‐	
  and	
  

purpose-­‐specificity	
  of	
  epistemic	
  representations	
  as	
  a	
  starting	
  point,	
  we	
  may	
  then	
  ask	
  

further	
  questions	
  about	
  these	
  features,	
  thereby	
  better	
  coming	
  to	
  understand	
  visual	
  

epistemic	
  representation.	
  In	
  which	
  ways,	
  precisely,	
  are	
  such	
  representations	
  user-­‐	
  and	
  

purpose-­‐specific?	
  In	
  virtue	
  of	
  what	
  are	
  they	
  so?	
  	
  

In	
  this	
  section,	
  I	
  will	
  present	
  three	
  examples	
  of	
  successful	
  visual	
  epistemic	
  

representation,	
  in	
  each	
  case	
  identifying	
  the	
  phenomenon	
  of	
  interest,	
  indicating	
  the	
  sense	
  

in	
  which	
  it	
  is	
  inaccessible	
  to	
  its	
  user(s)	
  through	
  direct	
  examination	
  of	
  the	
  physical	
  system	
  

in	
  question,	
  and	
  highlighting	
  the	
  ways	
  in	
  which	
  the	
  representation	
  is	
  successful	
  only	
  for	
  

certain	
  users	
  and	
  for	
  specific	
  purposes.	
  	
  

The	
  first	
  example	
  is	
  adapted	
  from	
  Suárez	
  (2004).	
  Suppose	
  we	
  want	
  to	
  represent	
  a	
  

system	
  consisting	
  of	
  two	
  ships	
  travelling	
  along	
  the	
  sea	
  using	
  two	
  pens	
  and	
  a	
  piece	
  of	
  paper.	
  

Let	
  us	
  further	
  assume	
  that	
  the	
  representation	
  is	
  used	
  in	
  the	
  context	
  of	
  a	
  conversation	
  

between	
  the	
  captain	
  of	
  one	
  of	
  the	
  ships	
  and	
  her	
  friend.	
  To	
  help	
  recount	
  the	
  highlights	
  of	
  

her	
  last	
  voyage,	
  the	
  captain	
  might	
  move	
  the	
  two	
  pens	
  along	
  the	
  paper	
  to	
  demonstrate,	
  for	
  

instance,	
  a	
  maneuver	
  she	
  had	
  to	
  perform	
  to	
  avoid	
  colliding	
  with	
  another	
  ship	
  that	
  had	
  

strayed	
  off	
  course.	
  In	
  this	
  example,	
  the	
  phenomenon	
  of	
  interest	
  consists	
  of	
  the	
  relative	
  

trajectories	
  of	
  the	
  ships.	
  It	
  is	
  inaccessible	
  to	
  the	
  captain’s	
  friend	
  because	
  he	
  was	
  not	
  

present	
  to	
  witness	
  her	
  collision-­‐avoiding	
  maneuver.	
  While	
  the	
  intended	
  user	
  of	
  the	
  pens-­‐

on-­‐paper	
  system	
  is	
  the	
  captain’s	
  friend,	
  it	
  would	
  be	
  suitable	
  for	
  many	
  other	
  users	
  as	
  well,	
  

since	
  little	
  background	
  knowledge	
  is	
  required	
  to	
  understand	
  the	
  relative	
  motions	
  of	
  the	
  

ships.	
  The	
  purpose	
  of	
  the	
  pens-­‐on-­‐paper	
  system	
  is	
  to	
  help	
  the	
  captain	
  relay	
  certain	
  parts	
  of	
  

her	
  voyage	
  to	
  her	
  friend.	
  	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   5	
  

A	
  second	
  example	
  of	
  a	
  successful	
  epistemic	
  representation	
  is	
  the	
  iconic	
  map	
  of	
  the	
  

London	
  Underground	
  transit	
  system.4	
  Originally	
  designed	
  by	
  draftsman	
  Harry	
  Beck	
  in	
  

1933	
  and	
  modelled	
  after	
  a	
  circuit	
  diagram,	
  the	
  network	
  of	
  railway	
  lines	
  and	
  stations	
  that	
  

comprise	
  the	
  Underground	
  is	
  depicted	
  as	
  an	
  orderly	
  array	
  of	
  intertwined	
  coloured	
  lines,	
  

along	
  which	
  lie	
  evenly-­‐spaced	
  marks	
  labelled	
  with	
  station	
  names,	
  with	
  white	
  circles	
  

replacing	
  these	
  marks	
  to	
  designate	
  interchange	
  stations.	
  Included	
  with	
  the	
  map	
  are	
  keys	
  

that	
  tell	
  users	
  how	
  to	
  interpret	
  each	
  of	
  its	
  features	
  (Figure	
  1).	
  With	
  the	
  aid	
  of	
  this	
  map,	
  

approximately	
  three	
  million	
  daily	
  users	
  of	
  the	
  Underground	
  are	
  able	
  to	
  navigate	
  this	
  

expansive	
  system.	
  The	
  phenomenon	
  of	
  interest	
  varies	
  between	
  users:	
  for	
  each,	
  it	
  is	
  the	
  set	
  

of	
  the	
  possible	
  routes	
  connecting	
  the	
  stations	
  between	
  which	
  she	
  wishes	
  to	
  travel.	
  These	
  

routes	
  are	
  inaccessible	
  to	
  her,	
  since	
  while	
  she	
  could	
  in	
  principle	
  ride	
  the	
  Underground	
  in	
  

various	
  directions	
  to	
  determine	
  which	
  one	
  connects	
  her	
  to	
  her	
  destination	
  station,	
  this	
  

would	
  be	
  extremely	
  impractical.	
  Because	
  the	
  map	
  of	
  the	
  London	
  Underground	
  is	
  intended	
  

for	
  use	
  by	
  a	
  broad	
  range	
  of	
  people	
  with	
  a	
  variety	
  of	
  backgrounds	
  and	
  cognitive	
  capacities,	
  

it	
  is	
  designed	
  to	
  cater	
  universally	
  to	
  human	
  users.5	
  The	
  purpose	
  of	
  the	
  map	
  is	
  to	
  determine	
  

the	
  most	
  efficient	
  route	
  from	
  one	
  station	
  to	
  another.	
  	
  

	
  

	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
4	
  This	
  example	
  is	
  also	
  discussed	
  by	
  Contessa	
  (2007),	
  Hoover	
  (2012)	
  and	
  Bolinska	
  (2013).	
  	
  
5	
  This	
  is	
  not	
  to	
  say	
  that	
  no	
  training	
  whatsoever	
  is	
  required	
  –	
  users	
  must	
  be	
  familiar	
  with	
  the	
  conventions	
  
involved	
  in	
  reading	
  maps	
  –	
  but	
  that	
  the	
  level	
  of	
  training	
  required	
  is	
  relatively	
  low	
  and	
  fairly	
  universally	
  held	
  
in	
  many	
  parts	
  of	
  the	
  world.	
  	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   6	
  

	
  

Figure	
  1:	
  The	
  complete	
  London	
  Underground	
  map.	
  Note	
  the	
  even	
  spacing	
  of	
  labelled	
  stations,	
  colouring	
  of	
  
railway	
  lines	
  (with	
  key	
  in	
  the	
  bottom	
  right	
  corner)	
  and	
  white	
  circles	
  indicating	
  interchange	
  stations.	
  
	
  

Finally,	
  a	
  third	
  example	
  of	
  a	
  successful	
  epistemic	
  representation	
  is	
  a	
  three-­‐

dimensional	
  model	
  of	
  a	
  macromolecule	
  like	
  DNA	
  or	
  protein.	
  The	
  phenomenon	
  of	
  interest	
  

in	
  this	
  case	
  is	
  the	
  structure	
  of	
  the	
  molecule,	
  viz.	
  its	
  three-­‐dimensional	
  shape,	
  including	
  

bond	
  types,	
  lengths	
  and	
  angles	
  between	
  constituent	
  atoms.	
  This	
  structure	
  is	
  otherwise	
  

inaccessible	
  to	
  the	
  user:	
  it	
  is	
  not	
  directly	
  observable	
  even	
  using	
  techniques	
  like	
  x-­‐ray	
  

diffraction	
  photography,	
  since	
  the	
  images	
  produced	
  using	
  such	
  techniques	
  must	
  be	
  

interpreted	
  to	
  yield	
  putative	
  structures,	
  and	
  the	
  process	
  of	
  interpretation	
  can	
  yield	
  results	
  

that	
  are	
  often	
  ambiguous	
  or	
  misleading.6	
  Unlike	
  the	
  map	
  of	
  the	
  London	
  Underground,	
  a	
  

molecular	
  model's	
  key	
  is	
  implicit,	
  so	
  users	
  must	
  be	
  told	
  which	
  features	
  of	
  the	
  model	
  

correspond	
  to	
  which	
  features	
  of	
  a	
  molecule,	
  e.g.	
  that	
  white	
  balls	
  stand	
  of	
  hydrogen	
  atoms,	
  

black	
  for	
  carbon,	
  etc.	
  While	
  any	
  user	
  who	
  understands	
  this	
  convention	
  may	
  grasp	
  the	
  
	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
6	
  The	
  problem	
  of	
  determining	
  molecular	
  structure	
  from	
  x-­‐ray	
  diffraction	
  photographs	
  in	
  the	
  mid-­‐twentieth	
  
century	
  was	
  notoriously	
  difficult.	
  See	
  Olby	
  (1974)	
  and	
  Judson	
  (1996).	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   7	
  

structure	
  as	
  a	
  whole,	
  molecular	
  models	
  are	
  most	
  useful	
  for	
  one	
  with	
  training	
  in	
  molecular	
  

biology	
  in	
  the	
  pursuit	
  of	
  further	
  aims.	
  	
  For	
  instance,	
  such	
  a	
  user	
  may	
  rely	
  on	
  the	
  knowledge	
  

of	
  the	
  structure	
  she	
  gains	
  from	
  the	
  model	
  to	
  determine	
  the	
  function	
  of	
  the	
  molecule	
  or	
  how	
  

it	
  will	
  interact	
  with	
  other	
  molecules.	
  Thus,	
  the	
  purpose	
  for	
  which	
  the	
  model	
  is	
  used	
  often	
  

extends	
  beyond	
  simply	
  learning	
  about	
  molecular	
  structure.	
  

	
  

3.	
  Containing	
  sufficiently	
  accurate	
  information	
  

	
   With	
  these	
  examples	
  in	
  hand,	
  we	
  may	
  now	
  turn	
  to	
  the	
  more	
  specific	
  features	
  of	
  

successful	
  epistemic	
  representation,	
  each	
  of	
  which	
  depends	
  on	
  the	
  phenomenon	
  of	
  

interest,	
  the	
  user,	
  and	
  the	
  purpose	
  for	
  which	
  the	
  representation	
  is	
  used.	
  The	
  aim	
  of	
  

employing	
  epistemic	
  representations	
  is	
  to	
  learn	
  about	
  an	
  aspect	
  of	
  the	
  target	
  system,	
  the	
  

phenomenon	
  of	
  interest.	
  But	
  as	
  I	
  showed	
  in	
  the	
  previous	
  section,	
  users	
  often	
  don’t	
  seek	
  to	
  

learn	
  about	
  this	
  phenomenon	
  for	
  its	
  own	
  sake,	
  but	
  rather	
  to	
  use	
  what	
  they	
  learn	
  for	
  some	
  

further	
  purpose.	
  How	
  accurate	
  the	
  information	
  they	
  gain	
  need	
  be	
  depends	
  on	
  what	
  this	
  

purpose	
  is.	
  	
   	
  

	
  

3.1	
  When	
  is	
  information	
  sufficiently	
  accurate?	
  	
  

In	
  the	
  pens-­‐on-­‐paper	
  example	
  the	
  user	
  is	
  interested	
  in	
  the	
  ships’	
  relative	
  

trajectories	
  as	
  an	
  aid	
  to	
  understanding	
  the	
  relevant	
  parts	
  of	
  the	
  captain’s	
  voyage.	
  Thus,	
  

only	
  very	
  general	
  information	
  about	
  the	
  trajectories	
  need	
  be	
  contained	
  in	
  the	
  pens-­‐on-­‐

paper	
  representation.	
  For	
  instance,	
  let	
  us	
  assume	
  that	
  one	
  ship	
  was	
  on	
  a	
  head-­‐on	
  collision	
  

course	
  with	
  the	
  other,	
  and	
  the	
  other	
  veered	
  off	
  to	
  the	
  right	
  just	
  in	
  time	
  to	
  avoid	
  the	
  

collision.	
  This	
  information	
  should	
  be	
  contained	
  in	
  the	
  vehicle	
  of	
  representation,	
  but	
  the	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   8	
  

vehicle	
  need	
  not	
  accurately	
  portray	
  the	
  precise	
  angle	
  of	
  approach	
  between	
  the	
  ships’	
  

trajectories,	
  nor	
  the	
  angle	
  at	
  which	
  the	
  second	
  ship	
  veered	
  off	
  course	
  to	
  avoid	
  the	
  collision.	
  

So	
  long	
  as	
  the	
  more	
  general	
  information	
  is	
  contained	
  in	
  the	
  vehicle	
  of	
  representation,	
  the	
  

vehicle	
  may	
  serve	
  its	
  function	
  as	
  an	
  aid	
  in	
  the	
  telling	
  of	
  the	
  story.	
  

	
   The	
  map	
  of	
  the	
  London	
  Underground	
  is	
  used	
  with	
  the	
  aim	
  of	
  determining	
  the	
  most	
  

efficient	
  journey	
  between	
  stations	
  selected	
  by	
  its	
  users;	
  thus,	
  it	
  should	
  be	
  possible	
  to	
  

determine	
  the	
  most	
  efficient	
  route	
  between	
  any	
  two	
  stations	
  in	
  the	
  network	
  using	
  the	
  map.	
  

In	
  order	
  to	
  be	
  used	
  for	
  this	
  purpose,	
  the	
  map	
  must	
  contain	
  information	
  about	
  the	
  relative	
  

locations	
  of	
  stations	
  along	
  each	
  line,	
  including	
  information	
  about	
  the	
  locations	
  of	
  

interchange	
  stations	
  at	
  which	
  users	
  may	
  switch	
  lines.	
  It	
  need	
  not	
  contain	
  information	
  

about	
  other	
  features	
  of	
  the	
  network,	
  such	
  as	
  its	
  position	
  with	
  respect	
  to	
  above-­‐ground	
  

streets	
  or	
  the	
  distances	
  between	
  stations	
  to	
  represent	
  the	
  network	
  faithfully	
  for	
  its	
  users’	
  

purposes.7	
  As	
  in	
  the	
  ships-­‐on-­‐sea	
  example,	
  being	
  more	
  accurate	
  in	
  this	
  respect	
  does	
  not	
  

contribute	
  to	
  the	
  success	
  of	
  the	
  epistemic	
  representation.	
  In	
  fact,	
  as	
  we	
  will	
  see	
  in	
  Section	
  

5,	
  it	
  could	
  even	
  detract	
  from	
  its	
  success	
  if	
  increased	
  accuracy	
  comes	
  at	
  the	
  price	
  of	
  a	
  

reduction	
  in	
  the	
  ease	
  with	
  which	
  this	
  information	
  may	
  be	
  conveyed	
  to	
  the	
  user.	
  

	
   It	
  may	
  be	
  objected	
  here	
  that	
  it's	
  not	
  the	
  case	
  that	
  the	
  map	
  of	
  the	
  London	
  

Underground	
  represents	
  features	
  such	
  as	
  distance	
  between	
  stations	
  inaccurately:	
  

distances	
  between	
  stations	
  are	
  omitted	
  from	
  the	
  map,	
  not	
  misrepresented,	
  since	
  the	
  map's	
  

content	
  is	
  determined	
  by	
  its	
  intended	
  use.	
  This	
  objection	
  raises	
  an	
  important	
  point	
  about	
  

	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
7	
  We	
  may	
  assume	
  that	
  the	
  distances	
  between	
  stations	
  are	
  sufficiently	
  similar	
  to	
  one	
  another	
  that	
  in	
  each	
  case	
  
the	
  most	
  direct	
  route	
  between	
  stations	
  –	
  that	
  requiring	
  the	
  fewest	
  number	
  of	
  stops	
  and	
  changes	
  between	
  
lines	
  –	
  will	
  be	
  the	
  quickest.	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   9	
  

the	
  relationship	
  between	
  how	
  much	
  information	
  is	
  contained	
  in	
  a	
  representation	
  and	
  how	
  

accurate	
  that	
  information	
  is.	
  	
  

	
   In	
  general,	
  the	
  fact	
  that	
  one	
  representation	
  contains	
  less	
  information	
  than	
  another	
  

does	
  not	
  render	
  it	
  less	
  accurate.	
  'Smith	
  lives	
  in	
  Canada'	
  is	
  less	
  informative	
  than	
  'Smith	
  lives	
  

in	
  Toronto',	
  yet	
  both	
  sentences	
  may	
  be	
  equally	
  accurate,	
  e.g.	
  if	
  Smith	
  lives	
  in	
  Toronto,	
  

Canada,	
  or	
  the	
  former	
  may	
  be	
  more	
  accurate	
  than	
  the	
  latter,	
  e.g.	
  if	
  Smith	
  lives	
  in	
  

Mississauga,	
  a	
  suburb	
  of	
  Toronto.	
  But	
  in	
  the	
  case	
  of	
  epistemic	
  representation,	
  we	
  are	
  

concerned	
  not	
  with	
  accuracy	
  simpliciter,	
  but	
  in	
  accuracy-­for-­a-­purpose.	
  And	
  accuracy-­‐for-­‐

a-­‐purpose	
  is	
  closely	
  related	
  to	
  the	
  amount	
  of	
  information	
  a	
  representation	
  contains.	
  Recall	
  

that	
  epistemic	
  representations	
  are	
  used	
  to	
  convey	
  information	
  not	
  about	
  the	
  target	
  system	
  

in	
  general,	
  but	
  about	
  the	
  phenomenon	
  of	
  interest;	
  moreover,	
  users	
  are	
  interested	
  in	
  gaining	
  

this	
  information	
  for	
  a	
  particular	
  purpose.	
  	
  

If	
  a	
  representation	
  is	
  not	
  sufficiently	
  information-­‐rich	
  –	
  if	
  it	
  omits	
  bits	
  of	
  

information	
  that	
  are	
  necessary	
  for	
  understanding	
  the	
  phenomenon	
  of	
  interest	
  for	
  that	
  

purpose	
  –	
  then	
  it	
  cannot	
  represent	
  the	
  phenomenon	
  of	
  interest	
  sufficiently	
  accurately	
  for	
  

that	
  purpose.	
  The	
  map	
  of	
  the	
  London	
  Underground	
  omits	
  information	
  about	
  distances	
  

between	
  stations	
  –	
  it	
  represents	
  these	
  stations	
  as	
  evenly	
  spaced,	
  but	
  users	
  are	
  not	
  meant	
  

to	
  infer	
  from	
  this	
  feature	
  of	
  the	
  map	
  that	
  distances	
  between	
  stations	
  are,	
  in	
  fact,	
  equal.	
  

Suppose	
  the	
  map	
  also	
  omitted	
  information	
  about	
  where	
  the	
  lines	
  intersect,	
  simply	
  

depicting	
  them	
  in	
  parallel	
  with	
  one	
  another.	
  The	
  omission	
  of	
  this	
  crucial	
  information	
  

would	
  render	
  this	
  map	
  not	
  only	
  less	
  informative,	
  but	
  also	
  less	
  accurate,	
  given	
  a	
  user's	
  

interest	
  in	
  getting	
  from	
  any	
  one	
  station	
  in	
  the	
  system	
  to	
  any	
  other.	
  Key	
  pieces	
  of	
  

information,	
  such	
  as	
  the	
  places	
  in	
  which	
  the	
  lines	
  intersect,	
  must	
  be	
  contained	
  in	
  the	
  map	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   10	
  

in	
  order	
  for	
  it	
  to	
  accurately	
  portray	
  routes	
  between	
  stations	
  for	
  this	
  purpose.	
  So	
  the	
  

accuracy	
  of	
  an	
  epistemic	
  representation	
  –	
  that	
  is,	
  its	
  accuracy-­‐for-­a-­purpose	
  –	
  depends	
  in	
  

part	
  on	
  it	
  containing	
  a	
  sufficient	
  amount	
  of	
  information	
  to	
  accurately	
  represent	
  the	
  

phenomenon	
  of	
  interest	
  for	
  the	
  user's	
  purpose.	
  

	
   Finally,	
  in	
  the	
  examples	
  of	
  molecular	
  modelling,	
  the	
  target	
  systems	
  of	
  the	
  

representations	
  are	
  molecules	
  like	
  protein	
  or	
  DNA,	
  while	
  the	
  phenomena	
  of	
  interest	
  are	
  

the	
  structures	
  of	
  these	
  molecules.	
  The	
  purpose	
  of	
  gaining	
  information	
  about	
  these	
  

structures	
  is	
  often	
  to	
  enable	
  further	
  research	
  into	
  molecular	
  function.	
  For	
  instance,	
  

knowing	
  the	
  structure	
  of	
  DNA	
  readily	
  suggested	
  a	
  copying	
  mechanism	
  for	
  this	
  molecule	
  of	
  

inheritance.	
  Thus,	
  unlike	
  in	
  the	
  previous	
  two	
  examples,	
  where	
  only	
  general	
  and	
  relative	
  

positions	
  of	
  the	
  relevant	
  objects	
  are	
  required	
  for	
  users'	
  purposes,	
  fairly	
  specific	
  

information	
  about	
  atomic	
  positions	
  and	
  bond	
  lengths	
  and	
  angles	
  needs	
  to	
  be	
  contained	
  in	
  

the	
  vehicle	
  of	
  representation	
  in	
  these	
  cases	
  to	
  enable	
  further	
  investigation.	
  Without	
  such	
  

specific	
  information,	
  one	
  cannot,	
  for	
  instance,	
  determine	
  what	
  accounts	
  for	
  DNA's	
  stability	
  

or	
  how	
  it	
  may	
  be	
  'unzipped'	
  and	
  replicated.	
  

	
  

3.2	
  How	
  does	
  a	
  vehicle	
  contain	
  information?	
  

	
   We’ve	
  now	
  established	
  which	
  information	
  need	
  be	
  contained	
  in	
  a	
  vehicle	
  of	
  visual	
  

representation	
  in	
  order	
  for	
  that	
  vehicle	
  to	
  epistemically	
  represent	
  its	
  target	
  system:	
  

information	
  about	
  the	
  phenomenon	
  of	
  interest	
  that	
  is	
  sufficiently	
  accurate	
  for	
  its	
  user's	
  

purpose.	
  We	
  will	
  now	
  turn	
  to	
  another	
  question:	
  how	
  does	
  a	
  vehicle	
  come	
  to	
  contain	
  that	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   11	
  

information?	
  While	
  the	
  term	
  ‘information’	
  may	
  be	
  used	
  differently	
  in	
  other	
  contexts,8	
  here	
  

we	
  are	
  concerned	
  with	
  semantic	
  information,	
  the	
  sense	
  in	
  which	
  this	
  term	
  is	
  perhaps	
  most	
  

commonly	
  used.	
  This	
  is	
  the	
  sense	
  in	
  which	
  a	
  newspaper	
  contains	
  information	
  about	
  

current	
  events,	
  a	
  map	
  contains	
  information	
  about	
  the	
  layout	
  of	
  a	
  city,	
  and	
  a	
  model	
  contains	
  

information	
  about	
  the	
  physical	
  system	
  it	
  represents.	
  	
  

	
   Semantic	
  information	
  may	
  be	
  either	
  natural	
  or	
  non-­‐natural	
  information.9	
  Natural	
  

information	
  is	
  a	
  product	
  of	
  certain	
  naturally	
  occurring	
  correlations.	
  Examples	
  include	
  

smoke	
  carrying	
  the	
  information	
  that	
  there	
  is	
  fire	
  and	
  rings	
  on	
  a	
  tree	
  carrying	
  information	
  

about	
  the	
  tree’s	
  age.	
  We	
  will	
  generally	
  not	
  be	
  concerned	
  with	
  natural	
  information,	
  since	
  

most	
  vehicles	
  of	
  epistemic	
  representation,	
  and	
  particularly	
  scientific	
  models,	
  are	
  human	
  

constructs.	
  Thus,	
  ‘information’	
  will	
  here	
  be	
  used	
  to	
  refer	
  to	
  non-­‐natural	
  information.	
  	
  	
  

	
   An	
  artefact	
  may	
  contain	
  non-­‐natural	
  information	
  by	
  convention.	
  For	
  example,	
  in	
  a	
  

weather	
  forecast,	
  a	
  picture	
  of	
  a	
  cloud	
  and	
  rain	
  labelled	
  	
  ‘80%'	
  contains	
  the	
  information	
  that	
  

there	
  is	
  an	
  80%	
  chance	
  of	
  showers,	
  and	
  a	
  sign	
  that	
  says	
  ‘Gift	
  shop	
  downstairs’	
  contains	
  the	
  

information	
  that	
  the	
  gift	
  shop	
  is	
  downstairs.	
  In	
  these	
  examples,	
  the	
  conventions	
  via	
  which	
  

the	
  vehicles	
  contain	
  information	
  are	
  known	
  to	
  most	
  people	
  who	
  encounter	
  them.	
  But	
  this	
  

need	
  not	
  be	
  the	
  case.	
  Before	
  arriving	
  at	
  a	
  party,	
  I	
  might	
  establish	
  with	
  my	
  companion	
  that	
  

either	
  of	
  us	
  uttering	
  the	
  sentence	
  ‘This	
  wine	
  is	
  fantastic’	
  indicates	
  his	
  or	
  her	
  desire	
  to	
  leave.	
  

In	
  this	
  case,	
  ‘This	
  wine	
  is	
  fantastic’	
  contains	
  the	
  information	
  that	
  its	
  utterer	
  would	
  like	
  to	
  

leave	
  the	
  party	
  (provided	
  that	
  the	
  utterer	
  is	
  either	
  my	
  companion	
  or	
  I).	
  

	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
8	
  For	
  instance,	
  Fisher	
  information,	
  Shannon	
  information,	
  Kolmogorov	
  complexity	
  and	
  quantum	
  information	
  
each	
  refer	
  to	
  different	
  notions	
  of	
  information.	
  For	
  a	
  survey	
  of	
  how	
  these	
  notions	
  are	
  used	
  and	
  related	
  to	
  one	
  
another,	
  see	
  Adriaans	
  (2013).	
  
9	
  Grice	
  (1957)	
  made	
  the	
  distinction	
  between	
  natural	
  and	
  non-­‐natural	
  meaning;	
  Piccinini	
  and	
  Scarantino	
  
(2010)	
  have	
  extended	
  this	
  distinction	
  to	
  information.	
  
	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   12	
  

	
   The	
  information	
  contained	
  in	
  a	
  vehicle	
  of	
  representation	
  need	
  not	
  be	
  true.	
  A	
  sign	
  

that	
  says	
  ‘Gift	
  shop	
  downstairs’	
  contains	
  the	
  information	
  that	
  the	
  gift	
  shop	
  is	
  downstairs	
  

even	
  if,	
  as	
  a	
  matter	
  of	
  fact,	
  it	
  has	
  been	
  relocated	
  to	
  the	
  second	
  floor;	
  in	
  this	
  case,	
  the	
  sign	
  

contains	
  false	
  information.10	
  Similarly,	
  suppose	
  I	
  really	
  like	
  the	
  wine	
  being	
  served	
  at	
  the	
  

party,	
  and	
  carelessly	
  remark,	
  ‘This	
  wine	
  is	
  fantastic’,	
  even	
  though	
  I’m	
  having	
  a	
  good	
  time	
  

and	
  don’t	
  yet	
  want	
  to	
  leave.	
  In	
  this	
  case,	
  the	
  sentence	
  nevertheless	
  contains	
  the	
  

information	
  that	
  I	
  want	
  to	
  leave	
  the	
  party,	
  but	
  this	
  information	
  is	
  false.	
  

Because	
  a	
  vehicle	
  of	
  representation	
  may	
  carry	
  information	
  by	
  convention,	
  is	
  it	
  the	
  

case	
  that	
  anything	
  may	
  carry	
  information	
  about	
  anything	
  else,	
  just	
  as	
  anything	
  may	
  denote	
  

anything	
  else?	
  As	
  noted	
  earlier,	
  information	
  is	
  a	
  multifarious	
  concept;	
  here	
  we	
  are	
  

restricting	
  our	
  attention	
  to	
  the	
  sense	
  in	
  which	
  information	
  can	
  be	
  contained	
  in	
  a	
  vehicle	
  of	
  

representation	
  about	
  an	
  aspect	
  of	
  a	
  target	
  system,	
  not,	
  for	
  instance,	
  the	
  sense	
  in	
  which	
  

information	
  is	
  carried	
  in	
  our	
  genetic	
  code,	
  or	
  that	
  in	
  which	
  it	
  is	
  stored	
  in	
  computer	
  

databases.	
  Neither	
  the	
  information	
  contained	
  in	
  our	
  genetic	
  code	
  nor	
  that	
  stored	
  in	
  

computer	
  databases	
  is	
  directly	
  accessible	
  to	
  a	
  user	
  who	
  consults	
  these	
  things	
  directly.	
  

Rather,	
  a	
  complex,	
  multi-­‐step	
  process	
  is	
  necessary	
  for	
  its	
  extraction.	
  A	
  DNA	
  molecule	
  must	
  

first	
  be	
  sequenced;	
  the	
  sequence	
  must	
  then	
  be	
  decoded	
  by	
  someone	
  with	
  the	
  requisite	
  

training	
  and	
  experience.	
  A	
  computer’s	
  hard	
  drive	
  doesn’t	
  issue	
  information	
  about	
  its	
  

contents	
  directly,	
  but	
  must	
  be	
  accessed	
  using	
  particular	
  software	
  and	
  hardware.	
  	
  

	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
10	
  Many	
  authors	
  (Barwise	
  &	
  Seligman,	
  1997;	
  Floridi,	
  2004,	
  2005;	
  Graham,	
  1999;	
  Grice,	
  1989)	
  take	
  truth	
  to	
  
be	
  a	
  requirement	
  for	
  information.	
  For	
  instance,	
  Dretske	
  contends	
  that	
  “false	
  information	
  and	
  mis-­‐
information	
  are	
  not	
  kinds	
  of	
  information	
  –	
  any	
  more	
  than	
  decoy	
  ducks	
  and	
  rubber	
  ducks	
  are	
  kinds	
  of	
  ducks”	
  
(1981,	
  p.	
  45).	
  However,	
  this	
  requirement	
  fails	
  to	
  accommodate	
  many	
  of	
  the	
  ways	
  in	
  which	
  'information'	
  is	
  
used.	
  Moreover,	
  the	
  authors	
  who	
  hold	
  this	
  position	
  have	
  failed	
  to	
  give	
  a	
  good	
  argument	
  for	
  why	
  it	
  should	
  be	
  
adopted	
  (Scarantino	
  &	
  Piccinini,	
  2010).	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   13	
  

In	
  contrast,	
  it	
  only	
  makes	
  sense	
  to	
  speak	
  of	
  a	
  vehicle	
  of	
  epistemic	
  representation	
  –	
  a	
  

tool	
  for	
  conveying	
  information	
  –	
  as	
  containing	
  information	
  if	
  that	
  information	
  is	
  available	
  

to	
  the	
  user	
  when	
  she	
  consults	
  the	
  vehicle.	
  The	
  sentence	
  'This	
  wine	
  is	
  fantastic'	
  contains	
  the	
  

information	
  that	
  its	
  utterer	
  wants	
  to	
  leave	
  the	
  party:	
  once	
  the	
  convention	
  has	
  been	
  

established,	
  the	
  person	
  who	
  hears	
  this	
  sentence	
  may	
  grasp	
  that	
  its	
  utterer	
  wants	
  to	
  leave	
  

the	
  party.	
  ‘This	
  wine	
  is	
  fantastic’	
  does	
  not	
  contain	
  the	
  information	
  that	
  its	
  utterer	
  wants	
  to	
  

leave	
  the	
  party	
  because	
  he	
  or	
  she	
  is	
  tired,	
  since	
  this	
  further	
  bit	
  of	
  information	
  was	
  not	
  

included	
  in	
  the	
  convention	
  when	
  it	
  was	
  established.	
  If	
  my	
  companion	
  utters	
  ‘This	
  wine	
  is	
  

fantastic’	
  with	
  a	
  yawn	
  and	
  vacant	
  stare,	
  I	
  might	
  infer	
  that	
  he	
  wants	
  to	
  leave	
  the	
  party	
  

because	
  he	
  is	
  tired,	
  but	
  this	
  inference	
  would	
  be	
  based	
  on	
  the	
  information	
  the	
  sentence	
  

contains	
  together	
  with	
  information	
  obtained	
  from	
  the	
  yawn	
  and	
  vacant	
  stare	
  

accompanying	
  its	
  utterance.	
  	
  

Of	
  course,	
  if	
  my	
  companion	
  and	
  I	
  were	
  to	
  establish	
  such	
  a	
  convention	
  at	
  the	
  outset,	
  

‘This	
  wine	
  is	
  fantastic’	
  would	
  contain	
  the	
  information	
  that	
  the	
  utterer	
  of	
  the	
  sentence	
  

wants	
  to	
  leave	
  the	
  party	
  because	
  he	
  or	
  she	
  is	
  tired.	
  In	
  this	
  case,	
  if	
  I	
  heard	
  my	
  companion	
  

say	
  'This	
  wine	
  is	
  fantastic',	
  I	
  could	
  immediately	
  infer	
  that	
  he	
  wants	
  to	
  leave	
  the	
  party	
  

because	
  he	
  is	
  tired,	
  even	
  absent	
  the	
  yawn	
  and	
  vacant	
  stare.	
  This	
  sentence	
  could	
  also	
  be	
  

made	
  to	
  contain	
  more	
  information	
  than	
  this,	
  such	
  as	
  that	
  its	
  utterer	
  wants	
  to	
  leave	
  the	
  

party	
  because	
  he	
  or	
  she	
  is	
  tired	
  and	
  has	
  become	
  bored	
  with	
  the	
  conversation.	
  But	
  the	
  

amount	
  of	
  information	
  that	
  ‘This	
  wine	
  is	
  fantastic’	
  can	
  contain	
  is	
  constrained	
  by	
  the	
  

limitations	
  of	
  its	
  users'	
  memories.	
  So,	
  for	
  instance,	
  it	
  cannot	
  contain	
  information	
  about	
  the	
  

names	
  and	
  birthdays	
  of	
  each	
  of	
  the	
  party’s	
  guests:	
  even	
  if	
  I	
  stipulate	
  that	
  'This	
  wine	
  is	
  

fantastic'	
  contains	
  the	
  information	
  that	
  John	
  Smith’s	
  birthday	
  is	
  on	
  February	
  24th	
  and	
  Anne	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   14	
  

MacDonald’s	
  is	
  on	
  June	
  2nd	
  and	
  so	
  on	
  for	
  each	
  of	
  the	
  party’s	
  guests,	
  the	
  only	
  people	
  familiar	
  

with	
  the	
  convention	
  –	
  my	
  companion	
  and	
  I	
  –	
  are	
  unable	
  to	
  extract	
  that	
  information	
  from	
  

the	
  sentence.11	
  Thus,	
  there	
  is	
  no	
  meaningful	
  sense	
  in	
  which	
  this	
  vehicle	
  really	
  does	
  contain	
  

that	
  information.	
  

In	
  general,	
  the	
  more	
  discernable	
  subcomponents	
  a	
  vehicle	
  has,	
  the	
  more	
  

information	
  it	
  may	
  contain.	
  A	
  discernable	
  subcomponent	
  is	
  a	
  part	
  of	
  the	
  vehicle	
  that	
  can	
  be	
  

distinguished	
  from	
  other	
  parts	
  by	
  its	
  user,	
  and	
  what	
  constitutes	
  a	
  discernable	
  

subcomponent	
  in	
  each	
  case	
  depends	
  on	
  the	
  means	
  of	
  representation	
  and	
  the	
  user.	
  For	
  

example,	
  each	
  of	
  the	
  pens	
  and	
  the	
  piece	
  of	
  paper	
  are	
  discernable	
  subcomponents	
  of	
  the	
  

pens-­‐on-­‐paper	
  system,	
  but	
  an	
  arbitrarily	
  chosen	
  point	
  on	
  the	
  blank	
  piece	
  of	
  paper	
  is	
  not,	
  

since	
  it	
  can’t	
  be	
  distinguished	
  by	
  an	
  ordinary	
  user	
  from	
  the	
  rest	
  of	
  the	
  paper.12	
  	
  

This	
  is	
  important	
  because	
  it	
  precludes	
  certain	
  vehicles	
  from	
  counting	
  as	
  successful	
  

epistemic	
  representations	
  in	
  a	
  given	
  set	
  of	
  circumstances.	
  For	
  instance,	
  consider	
  an	
  

alternative	
  to	
  the	
  pens-­‐on-­‐paper	
  representation,	
  the	
  paper-­‐on-­‐pens	
  system,	
  in	
  which	
  a	
  

blank	
  piece	
  of	
  paper	
  represents	
  the	
  ship	
  trajectories	
  and	
  two	
  pens	
  represent	
  the	
  surface	
  of	
  

the	
  sea.	
  While	
  such	
  a	
  system	
  may	
  certainly	
  denote	
  the	
  ships-­‐on-­‐sea	
  system,	
  it	
  cannot	
  be	
  an	
  

epistemic	
  representation	
  of	
  that	
  system.	
  For	
  there	
  is	
  no	
  genuine	
  sense	
  in	
  which	
  

information	
  about	
  the	
  individual	
  trajectories	
  of	
  the	
  ships	
  can	
  be	
  conveyed	
  to	
  a	
  user	
  

	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
11	
  John	
  Kulvicki	
  (2010)	
  argues	
  that	
  more	
  complex	
  or	
  detailed	
  information	
  can	
  be	
  contained	
  in	
  a	
  fairly	
  simple	
  
vehicle	
  of	
  representation,	
  but	
  may	
  not	
  be	
  ‘extractable’	
  from	
  that	
  vehicle.	
  For	
  instance,	
  the	
  information	
  that	
  its	
  
target	
  is	
  square	
  is	
  extractable	
  from	
  the	
  representation	
  ‘square’;	
  however,	
  ‘square’	
  also	
  contains	
  the	
  
information	
  that	
  its	
  target	
  has	
  four	
  sides	
  of	
  equal	
  lengths	
  and	
  right	
  internal	
  angles,	
  but	
  this	
  information	
  is	
  
available	
  only	
  via	
  an	
  inference.	
  In	
  contrast,	
  I	
  am	
  concerned	
  with	
  information	
  that	
  is	
  contained	
  in	
  a	
  vehicle	
  of	
  
representation	
  and	
  available	
  to	
  the	
  user	
  either	
  immediately	
  from	
  the	
  vehicle	
  or	
  via	
  this	
  kind	
  of	
  direct	
  
inference.	
  Thus,	
  my	
  usage	
  of	
  ‘extractable’	
  differs	
  from	
  his	
  more	
  specific	
  sense	
  of	
  the	
  term.	
  
12	
  There	
  may	
  be	
  more	
  than	
  one	
  way	
  to	
  individuate	
  subcomponents,	
  but	
  this	
  will	
  not	
  be	
  important	
  for	
  our	
  
purposes,	
  as	
  will	
  become	
  clear	
  shortly.	
  	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   15	
  

consulting	
  the	
  blank	
  piece	
  of	
  paper;	
  as	
  Suárez	
  points	
  out,	
  such	
  a	
  representation	
  “seems	
  

counterintuitive	
  and	
  unnatural”	
  (2004,	
  p.	
  772).	
  	
  

To	
  see	
  why,	
  let	
  us	
  consider	
  in	
  more	
  detail	
  how	
  the	
  captain	
  might	
  use	
  the	
  paper-­‐on-­‐

pens	
  system	
  to	
  represent	
  the	
  ships-­‐on-­‐sea	
  system	
  in	
  her	
  recounting	
  of	
  the	
  details	
  of	
  her	
  

voyage.	
  The	
  general	
  strategy	
  would	
  be	
  to	
  stipulate	
  that	
  the	
  paper	
  represents	
  the	
  

trajectories	
  of	
  the	
  ships	
  and	
  the	
  pens	
  represent	
  the	
  surface	
  of	
  the	
  sea	
  before	
  recounting	
  

these	
  events.	
  Then,	
  upon	
  reaching	
  the	
  point	
  in	
  her	
  story	
  where	
  she	
  performs	
  the	
  collision-­‐

avoiding	
  maneuver,	
  she	
  could	
  simply	
  gesture	
  towards	
  the	
  paper-­‐on-­‐pens	
  system.	
  In	
  order	
  

to	
  stipulate	
  that	
  the	
  paper	
  represents	
  the	
  trajectories	
  of	
  the	
  ships	
  and	
  the	
  pens	
  the	
  surface	
  

of	
  the	
  sea,	
  the	
  captain	
  would	
  have	
  to	
  somehow	
  get	
  her	
  friend	
  to	
  understand	
  what	
  these	
  

trajectories	
  looked	
  like,	
  either	
  by	
  describing	
  them	
  ('I	
  was	
  heading	
  due	
  north	
  when	
  I	
  

noticed	
  a	
  ship	
  coming	
  from	
  port	
  side	
  up	
  ahead…'),	
  or	
  tracing	
  the	
  trajectories	
  with	
  her	
  

hands	
  in	
  the	
  air,	
  or	
  even	
  using	
  the	
  pens	
  to	
  trace	
  them	
  on	
  the	
  paper	
  (!)	
  prior	
  to	
  reversing	
  

the	
  representational	
  arrangement.	
  Once	
  this	
  was	
  done,	
  she	
  could	
  then	
  use	
  the	
  paper	
  to	
  

represent	
  the	
  trajectories	
  of	
  the	
  ships	
  and	
  the	
  pens	
  the	
  surface	
  of	
  the	
  sea.	
  But	
  this	
  would	
  

be	
  an	
  odd	
  thing	
  to	
  do	
  at	
  this	
  stage,	
  since	
  all	
  of	
  the	
  work	
  required	
  to	
  convey	
  information	
  

about	
  the	
  trajectories	
  would	
  already	
  have	
  been	
  completed	
  in	
  the	
  earlier	
  stipulation	
  of	
  the	
  

representational	
  arrangement.	
  When	
  the	
  captain	
  reached	
  the	
  part	
  of	
  her	
  story	
  in	
  which	
  

she	
  describes	
  her	
  collision-­‐avoiding	
  maneuver,	
  gesturing	
  towards	
  the	
  paper	
  to	
  represent	
  

the	
  trajectories	
  of	
  the	
  ships	
  would	
  be	
  superfluous;	
  the	
  captain	
  could	
  just	
  as	
  easily	
  say	
  'And	
  

that’s	
  when	
  I	
  performed	
  that	
  maneuver	
  I	
  described	
  earlier'	
  instead.	
  Thus,	
  the	
  paper-­‐on-­‐

pens	
  system	
  cannot	
  be	
  used	
  as	
  an	
  epistemic	
  representation	
  of	
  the	
  ships-­‐on-­‐sea	
  system	
  

because	
  it	
  does	
  not	
  have	
  the	
  resources	
  to	
  individuate	
  the	
  relative	
  trajectories	
  of	
  two	
  ships.	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   16	
  

The	
  paper	
  is	
  a	
  single	
  uniform	
  object;	
  at	
  least	
  two	
  distinct	
  objects	
  are	
  required	
  to	
  convey	
  

information	
  about	
  these	
  trajectories.	
  	
  

	
   Each	
  of	
  our	
  examples	
  of	
  successful	
  epistemic	
  representation	
  contains	
  information	
  

about	
  the	
  phenomenon	
  of	
  interest	
  because	
  it	
  contains	
  a	
  sufficient	
  number	
  of	
  

subcomponents,	
  each	
  of	
  which	
  carries	
  more	
  specific	
  information	
  about	
  subcomponents	
  of	
  

this	
  phenomenon.	
  In	
  the	
  ships-­‐on-­‐sea	
  system,	
  these	
  are	
  the	
  individual	
  ship	
  trajectories;	
  

the	
  trajectory	
  of	
  each	
  pen	
  in	
  the	
  pens-­‐on-­‐paper	
  system	
  and	
  the	
  positions	
  of	
  these	
  

trajectories	
  with	
  respect	
  to	
  one	
  another	
  convey	
  to	
  the	
  user	
  information	
  about	
  the	
  

phenomenon	
  of	
  interest.	
  	
  

	
   Similarly,	
  in	
  order	
  for	
  the	
  map	
  of	
  the	
  London	
  Underground	
  to	
  be	
  able	
  to	
  carry	
  

information	
  about	
  the	
  most	
  efficient	
  routes	
  between	
  stations,	
  it	
  must	
  carry	
  several	
  specific	
  

bits	
  of	
  information	
  that	
  are	
  necessary	
  for	
  making	
  this	
  determination.	
  In	
  particular,	
  it	
  must	
  

carry	
  information	
  about	
  the	
  relations	
  between	
  lines	
  –	
  where	
  they	
  intersect,	
  allowing	
  users	
  

to	
  travel	
  between	
  them	
  –	
  and	
  information	
  about	
  which	
  stations	
  are	
  located	
  along	
  each	
  line	
  

and	
  in	
  which	
  order.	
  The	
  map	
  contains	
  features	
  that	
  are	
  responsible	
  for	
  carrying	
  each	
  of	
  

these	
  bits	
  of	
  information.	
  The	
  railway	
  lines	
  of	
  the	
  London	
  Underground	
  system	
  are	
  

represented	
  as	
  distinctly	
  coloured	
  lines	
  on	
  the	
  map;	
  interchange	
  stations	
  are	
  labelled	
  with	
  

a	
  white	
  circle	
  overlapping	
  the	
  lines	
  between	
  which	
  the	
  user	
  may	
  travel;	
  and	
  individual	
  

stations	
  are	
  represented	
  by	
  protrusions	
  along	
  each	
  line	
  labelled	
  with	
  station	
  names,	
  the	
  

order	
  of	
  the	
  stations	
  along	
  the	
  lines	
  in	
  the	
  map	
  reflecting	
  the	
  order	
  of	
  the	
  stations	
  in	
  the	
  

Underground	
  system.	
  	
  

	
   Finally,	
  in	
  order	
  to	
  convey	
  to	
  the	
  user	
  information	
  about	
  molecular	
  structure	
  as	
  a	
  

whole,	
  molecular	
  models	
  must	
  contain	
  specific	
  information	
  about	
  positions	
  of	
  particular	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   17	
  

atoms	
  and	
  the	
  lengths	
  and	
  angles	
  of	
  the	
  bonds	
  that	
  join	
  them.	
  This	
  information	
  is	
  

contained	
  in	
  the	
  pieces	
  representing	
  atoms	
  and	
  bonds,	
  and	
  the	
  ways	
  in	
  which	
  these	
  pieces	
  

are	
  connected	
  to	
  one	
  another.	
  

In	
  this	
  section,	
  I	
  have	
  argued	
  that	
  human	
  cognitive	
  and	
  perceptual	
  capacities	
  

impose	
  limits	
  on	
  the	
  amount	
  of	
  information	
  a	
  vehicle	
  may	
  contain	
  about	
  its	
  target	
  system:	
  

users	
  must	
  be	
  able	
  to	
  extract	
  information	
  from	
  the	
  representation,	
  and	
  they	
  must	
  be	
  able	
  

to	
  distinguish	
  between	
  subcomponents	
  of	
  the	
  vehicle	
  which	
  stand	
  for	
  subcomponents	
  of	
  

the	
  target	
  system.	
  But	
  while	
  a	
  vehicle's	
  containing	
  information	
  depends	
  in	
  part	
  on	
  the	
  

possibility	
  of	
  using	
  it	
  to	
  convey	
  that	
  information,	
  vehicles	
  of	
  representation	
  nevertheless	
  

differ	
  with	
  respect	
  to	
  how	
  well	
  they	
  convey	
  information	
  about	
  the	
  phenomenon	
  of	
  interest	
  

to	
  their	
  users.	
  The	
  extent	
  to	
  which	
  this	
  information	
  is	
  conveyed	
  is	
  a	
  second	
  important	
  

feature	
  of	
  successful	
  epistemic	
  representation.	
  In	
  the	
  next	
  section,	
  we	
  will	
  examine	
  the	
  

notions	
  of	
  syntactic	
  and	
  semantic	
  salience,	
  the	
  features	
  that	
  make	
  information	
  cognitively	
  

available	
  to	
  users,	
  and	
  thus	
  contribute	
  to	
  how	
  easily	
  information	
  is	
  conveyed.	
  

	
  

4.	
  Conveying	
  information	
  

	
   In	
  order	
  for	
  the	
  information	
  contained	
  in	
  a	
  vehicle	
  of	
  epistemic	
  representation	
  to	
  be	
  

effectively	
  conveyed	
  to	
  its	
  user,	
  it	
  must	
  be	
  readily	
  accessible	
  to	
  her.	
  In	
  this	
  section,	
  

following	
  Kulvicki	
  (2010),	
  I	
  identify	
  two	
  features	
  that	
  contribute	
  to	
  this	
  ready	
  access:	
  the	
  

syntactic	
  salience	
  of	
  subcomponents	
  and	
  the	
  semantic	
  salience	
  of	
  the	
  phenomenon	
  of	
  

interest.	
  Because	
  both	
  come	
  in	
  degrees,	
  information	
  is	
  effectively	
  conveyed	
  to	
  a	
  user	
  to	
  the	
  

extent	
  that	
  they	
  are	
  present.	
  

	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   18	
  

4.1	
  Syntactic	
  salience	
  of	
  subcomponents	
  

The	
  success	
  of	
  visual	
  epistemic	
  representation	
  depends	
  in	
  part	
  on	
  the	
  extent	
  to	
  

which	
  a	
  feature	
  of	
  the	
  vehicle	
  of	
  representation	
  that	
  carries	
  information	
  is	
  perceptually	
  

prominent	
  or	
  syntactically	
  salient	
  for	
  a	
  user	
  (Kulvicki,	
  2010).	
  When	
  the	
  trajectories	
  of	
  the	
  

ships	
  in	
  the	
  ships-­‐on-­‐sea	
  system	
  are	
  represented	
  by	
  the	
  trajectories	
  of	
  the	
  pens	
  in	
  the	
  

pens-­‐on-­‐paper	
  system,	
  they	
  very	
  clearly	
  stand	
  out	
  against	
  the	
  backdrop	
  of	
  the	
  paper.	
  

Similarly,	
  each	
  of	
  the	
  features	
  of	
  the	
  London	
  Underground	
  map	
  necessary	
  for	
  conveying	
  

information	
  about	
  the	
  best	
  possible	
  routes	
  between	
  stations	
  is	
  perceptually	
  accessible.	
  For	
  

example,	
  individual	
  railway	
  lines	
  of	
  the	
  Underground	
  system	
  are	
  syntactically	
  salient	
  

because	
  they	
  are	
  represented	
  in	
  the	
  map	
  as	
  distinctly	
  coloured	
  lines	
  set	
  against	
  a	
  

uniformly	
  white	
  background.	
  Had	
  the	
  same	
  colour	
  but	
  different	
  degrees	
  of	
  saturation	
  been	
  

used	
  for	
  each	
  of	
  the	
  lines,	
  the	
  distinction	
  between	
  these	
  lines	
  would	
  have	
  been	
  less	
  

syntactically	
  salient	
  for	
  the	
  average	
  user	
  of	
  the	
  map,	
  since	
  humans	
  generally	
  do	
  not	
  have	
  

the	
  same	
  capacity	
  to	
  recognize	
  differences	
  between	
  degrees	
  of	
  saturation	
  that	
  they	
  do	
  to	
  

distinguish	
  between	
  different	
  colours.	
  Had	
  identical	
  colours	
  and	
  degrees	
  of	
  saturation	
  

been	
  used,	
  the	
  lines	
  representing	
  railway	
  lines	
  would	
  be	
  completely	
  indistinguishable	
  for	
  

the	
  user,	
  and	
  thus	
  not	
  syntactically	
  salient	
  at	
  all.	
  The	
  pieces	
  of	
  which	
  molecular	
  models	
  are	
  

built	
  are	
  clearly	
  also	
  syntactically	
  salient:	
  they	
  are	
  very	
  easily	
  perceived	
  against	
  the	
  

backdrop	
  of	
  the	
  three-­‐dimensional	
  space	
  in	
  which	
  they	
  are	
  constructed.	
  

Note	
  that	
  whether	
  or	
  not	
  an	
  aspect	
  of	
  a	
  representation	
  is	
  syntactically	
  salient	
  

depends	
  on	
  the	
  perceptual	
  capacities	
  of	
  the	
  user(s)	
  of	
  that	
  representation.	
  If	
  someone	
  is	
  

colourblind,	
  then	
  a	
  map	
  of	
  the	
  London	
  Underground	
  in	
  which	
  the	
  same	
  colour	
  but	
  

different	
  degrees	
  of	
  saturation	
  were	
  used	
  for	
  each	
  of	
  the	
  railway	
  lines	
  might	
  be	
  more	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   19	
  

syntactically	
  salient	
  for	
  him.	
  However,	
  we	
  must	
  keep	
  in	
  mind	
  that	
  each	
  of	
  our	
  examples	
  of	
  

successful	
  epistemic	
  representation	
  is	
  designed	
  to	
  make	
  the	
  relevant	
  features	
  syntactically	
  

salient	
  for	
  most	
  of	
  its	
  intended	
  users.	
  Thus,	
  since	
  most	
  people	
  aren’t	
  colourblind,	
  the	
  

different	
  railway	
  lines	
  are	
  represented	
  with	
  different	
  colours,	
  which	
  are	
  easily	
  

distinguishable	
  from	
  one	
  another	
  for	
  the	
  average	
  person.	
  

The	
  syntactic	
  salience	
  of	
  the	
  parts	
  of	
  the	
  vehicle	
  that	
  represent	
  subcomponents	
  of	
  

the	
  phenomenon	
  of	
  interest	
  contributes	
  to	
  the	
  success	
  of	
  an	
  epistemic	
  representation.	
  But	
  

it	
  is	
  possible	
  to	
  represent	
  these	
  subcomponents	
  in	
  a	
  variety	
  of	
  equally	
  syntactically	
  salient	
  

ways,	
  not	
  all	
  of	
  which	
  are	
  equally	
  able	
  to	
  give	
  the	
  user	
  the	
  kind	
  of	
  access	
  to	
  the	
  

phenomenon	
  of	
  interest	
  she	
  seeks	
  through	
  using	
  an	
  epistemic	
  representation.	
  For	
  example,	
  

consider	
  an	
  alternative	
  representation	
  of	
  the	
  London	
  Underground	
  in	
  which	
  the	
  railway	
  

lines	
  are	
  depicted	
  in	
  parallel	
  to	
  one	
  another,	
  rather	
  than	
  as	
  a	
  network,	
  with	
  interchange	
  

stations	
  labelled	
  with	
  the	
  other	
  line(s)	
  that	
  can	
  be	
  accessed	
  at	
  them	
  in	
  parentheses	
  next	
  to	
  

their	
  names.13	
  So,	
  for	
  example,	
  along	
  the	
  Bakerloo	
  line,	
  Picadilly	
  Circus	
  Station	
  would	
  be	
  

labelled	
  ‘Picadilly	
  Circus	
  (Picadilly)’,	
  since	
  the	
  Bakerloo	
  and	
  Picadilly	
  lines	
  intersect	
  at	
  

Picadilly	
  Circus	
  Station	
  (Figure	
  2).	
  

	
  

	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
13	
  Note	
  that	
  this	
  is	
  different	
  from	
  the	
  parallel-­‐lines	
  representation	
  mentioned	
  in	
  Section	
  3,	
  in	
  which	
  
interchange	
  stations	
  were	
  not	
  labelled.	
  	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   20	
  

	
  

Such	
  a	
  representation	
  of	
  the	
  London	
  Underground	
  makes	
  the	
  same	
  bits	
  of	
  

information	
  about	
  the	
  relative	
  positions	
  of	
  stations	
  along	
  each	
  line	
  and	
  the	
  locations	
  of	
  

interchange	
  stations	
  just	
  as	
  syntactically	
  salient	
  as	
  the	
  map	
  does,	
  since	
  they	
  are	
  equally	
  

perceptually	
  available	
  to	
  the	
  user.	
  Just	
  as	
  a	
  user	
  has	
  no	
  problem	
  detecting	
  the	
  white	
  circles	
  

on	
  the	
  map,	
  she	
  also	
  has	
  no	
  problem	
  detecting	
  the	
  names	
  of	
  the	
  other	
  lines	
  that	
  can	
  be	
  

accessed	
  at	
  an	
  interchange	
  station:	
  she	
  can	
  easily	
  see	
  their	
  names	
  in	
  parentheses	
  next	
  to	
  

the	
  name	
  of	
  the	
  station	
  in	
  question.	
  However,	
  it	
  would	
  be	
  much	
  more	
  difficult	
  to	
  use	
  such	
  a	
  

representation	
  for	
  the	
  purpose	
  of	
  determining	
  the	
  optimal	
  route	
  between	
  stations	
  than	
  it	
  

is	
  to	
  use	
  the	
  map.	
  While	
  the	
  individual	
  pieces	
  of	
  information	
  that	
  are	
  necessary	
  to	
  identify	
  

possible	
  routes	
  –	
  positions	
  of	
  stations	
  along	
  each	
  line	
  and	
  locations	
  of	
  interchange	
  stations	
  

–	
  are	
  syntactically	
  salient,	
  the	
  possible	
  routes	
  themselves,	
  and	
  thus	
  information	
  about	
  

which	
  of	
  these	
  routes	
  is	
  best,	
  are	
  more	
  difficult	
  to	
  discern.	
  In	
  order	
  to	
  find	
  these	
  routes,	
  a	
  

user	
  must	
  perform	
  additional	
  inferential	
  steps	
  to	
  decode,	
  so	
  to	
  speak,	
  the	
  alternative	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   21	
  

representation.	
  But	
  this	
  kind	
  of	
  extra	
  work	
  is	
  precisely	
  what	
  a	
  user	
  seeks	
  to	
  minimize	
  

through	
  the	
  use	
  of	
  an	
  epistemic	
  representation.	
  	
  

Similarly,	
  consider	
  some	
  alternative	
  ways	
  of	
  representing	
  molecular	
  structure	
  that	
  

also	
  render	
  its	
  subcomponents	
  –	
  atomic	
  positions	
  and	
  bond	
  lengths	
  and	
  types	
  –	
  

syntactically	
  salient.	
  For	
  instance,	
  a	
  list	
  of	
  atomic	
  coordinates	
  together	
  with	
  a	
  specification	
  

of	
  the	
  lengths	
  and	
  types	
  of	
  bonds	
  between	
  each	
  pair	
  of	
  atoms	
  makes	
  all	
  the	
  relevant	
  

subcomponents	
  syntactically	
  salient;	
  nevertheless,	
  the	
  overall	
  structure	
  of	
  the	
  molecule	
  

would,	
  as	
  with	
  the	
  alternative	
  parallel-­‐lines	
  representation	
  of	
  the	
  London	
  Underground,	
  

require	
  the	
  user	
  to	
  perform	
  a	
  number	
  of	
  additional	
  inferential	
  steps	
  to	
  ‘see’	
  this	
  structure.	
  

A	
  two-­‐dimensional	
  image	
  of	
  the	
  structure	
  fares	
  better,	
  since	
  it	
  is	
  possible	
  to	
  discern	
  three-­‐

dimensional	
  structure	
  from	
  such	
  an	
  image	
  with	
  fewer	
  and	
  simpler	
  inferences.	
  But	
  the	
  

three-­‐dimensional	
  model	
  is	
  best.	
  

	
  

4.2	
  Semantic	
  salience	
  of	
  phenomenon	
  of	
  interest	
  

These	
  alternative	
  representations	
  of	
  the	
  London	
  Underground	
  and	
  molecular	
  

structure	
  lack	
  a	
  feature	
  that	
  Kulvicki	
  (2010)	
  calls	
  semantic	
  salience.	
  Semantic	
  salience	
  

refers	
  to	
  the	
  ease	
  with	
  which	
  syntactically	
  salient	
  features	
  of	
  the	
  representation	
  convey	
  

their	
  content	
  to	
  the	
  representation’s	
  user.	
  In	
  order	
  for	
  a	
  part	
  of	
  a	
  representation	
  to	
  be	
  

semantically	
  salient,	
  it	
  must	
  be	
  straightforward	
  for	
  a	
  user	
  to	
  determine	
  what	
  that	
  part	
  

represents:	
  “there	
  must	
  be	
  a	
  plan	
  of	
  correlation	
  between	
  features	
  of	
  the	
  representation	
  

and	
  features	
  of	
  the	
  data	
  that	
  is	
  easy	
  to	
  grasp”	
  (Kulvicki,	
  2010,	
  p.	
  301).	
  For	
  example,	
  

suppose	
  you	
  want	
  to	
  represent	
  a	
  temperature	
  gradient	
  in	
  which	
  the	
  temperature	
  

decreases	
  continuously	
  as	
  you	
  move	
  away	
  from	
  a	
  certain	
  point.	
  Representing	
  the	
  highest	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   22	
  

temperature	
  with	
  white,	
  the	
  lowest	
  with	
  black,	
  and	
  the	
  remainder	
  on	
  a	
  grayscale	
  with	
  

lighter	
  shades	
  corresponding	
  to	
  higher	
  temperatures	
  makes	
  this	
  gradient	
  semantically	
  

salient:	
  it	
  is	
  easy	
  to	
  interpret	
  the	
  light-­‐to-­‐dark	
  grayscale	
  as	
  corresponding	
  to	
  a	
  high-­‐to-­‐low	
  

temperature	
  scale.	
  One	
  could	
  also	
  arbitrarily	
  assign	
  shades	
  of	
  gray	
  to	
  temperatures,	
  but	
  

the	
  resultant	
  image	
  would	
  not	
  be	
  semantically	
  salient:	
  it	
  would	
  be	
  difficult	
  to	
  see	
  that	
  the	
  

temperature	
  decreases	
  continuously	
  away	
  from	
  the	
  hot	
  spot,	
  since	
  one	
  would	
  have	
  to	
  refer	
  

back	
  to	
  the	
  key	
  that	
  assigns	
  temperatures	
  to	
  colours	
  repeatedly	
  before	
  being	
  able	
  to	
  make	
  

this	
  determination.	
  

The	
  extent	
  to	
  which	
  something	
  is	
  a	
  successful	
  epistemic	
  representation	
  is	
  

determined	
  not	
  only	
  by	
  the	
  syntactic	
  salience	
  of	
  the	
  subcomponents	
  of	
  the	
  vehicle	
  

representing	
  parts	
  of	
  the	
  phenomenon	
  of	
  interest,	
  but	
  also	
  whether	
  the	
  phenomenon	
  of	
  

interest	
  is	
  semantically	
  salient.	
  In	
  the	
  case	
  of	
  the	
  London	
  Underground,	
  information	
  about	
  

the	
  phenomenon	
  of	
  interest	
  –	
  possible	
  routes	
  between	
  stations	
  –	
  is	
  semantically	
  salient	
  in	
  

the	
  map,	
  but	
  not	
  in	
  the	
  parallel-­‐lines	
  representation.	
  The	
  difference	
  between	
  these	
  

representations	
  lies	
  in	
  the	
  ways	
  in	
  which	
  relative	
  positions	
  of	
  the	
  railway	
  lines	
  vis-­‐à-­‐vis	
  

interchange	
  stations	
  are	
  represented.	
  In	
  the	
  map,	
  it	
  is	
  easy	
  to	
  see	
  that	
  the	
  intersection	
  of	
  

lines	
  with	
  white	
  circles	
  represents	
  places	
  where	
  the	
  lines	
  literally	
  cross,	
  allowing	
  users	
  to	
  

travel	
  between	
  them.	
  That	
  is,	
  the	
  locations	
  of	
  interchange	
  stations	
  are	
  semantically	
  salient	
  

and	
  this	
  is	
  necessary	
  for	
  making	
  possible	
  routes	
  between	
  stations	
  semantically	
  salient	
  as	
  

well.	
  	
  

The	
  parallel-­‐lines	
  representation,	
  on	
  the	
  other	
  hand,	
  forces	
  its	
  user	
  to	
  perform	
  

extra	
  steps	
  in	
  her	
  reasoning	
  to	
  obtain	
  the	
  same	
  information;	
  the	
  locations	
  of	
  the	
  

interchange	
  stations	
  –	
  and	
  thereby	
  also	
  possible	
  routes	
  between	
  stations	
  –	
  are	
  not	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   23	
  

semantically	
  salient	
  in	
  this	
  representation.	
  Labelling	
  interchange	
  stations	
  with	
  the	
  

name(s)	
  of	
  the	
  intersecting	
  line(s)	
  makes	
  information	
  about	
  possible	
  routes	
  between	
  

stations	
  more	
  difficult	
  to	
  grasp	
  than	
  in	
  the	
  case	
  of	
  the	
  map,	
  in	
  which	
  interchange	
  stations	
  

are	
  marked	
  with	
  a	
  white	
  circle	
  along	
  two	
  or	
  more	
  intersecting	
  or	
  adjacent	
  lines.	
  And	
  it	
  is	
  

precisely	
  this	
  information	
  that	
  is	
  required	
  to	
  determine	
  possible	
  routes	
  between	
  stations	
  

when	
  stations	
  do	
  not	
  lie	
  along	
  the	
  same	
  line.	
  

In	
  the	
  molecular	
  modelling	
  examples,	
  the	
  structures	
  of	
  the	
  molecules	
  –	
  the	
  

phenomena	
  of	
  interest	
  –	
  are	
  semantically	
  salient	
  in	
  the	
  physical	
  models.	
  It	
  is	
  

straightforward	
  to	
  interpret	
  the	
  physical	
  structure	
  of	
  the	
  model	
  as	
  the	
  structure	
  of	
  the	
  

molecule.	
  This	
  is	
  not	
  so	
  for	
  some	
  alternative	
  representations	
  that	
  also	
  make	
  the	
  

subcomponents	
  of	
  the	
  structures	
  syntactically	
  salient.	
  For	
  instance,	
  a	
  list	
  of	
  the	
  

coordinates	
  of	
  each	
  atom	
  together	
  with	
  a	
  list	
  of	
  the	
  types	
  of	
  bonds	
  that	
  join	
  them	
  makes	
  all	
  

the	
  relevant	
  subcomponents	
  of	
  the	
  structure	
  syntactically	
  salient,	
  but	
  the	
  overall	
  structure	
  

of	
  the	
  molecule	
  is	
  more	
  difficult	
  to	
  determine.	
  As	
  with	
  the	
  alternative	
  parallel-­‐lines	
  

representation	
  of	
  the	
  London	
  Underground,	
  a	
  user	
  would	
  have	
  to	
  perform	
  a	
  number	
  of	
  

additional	
  inferential	
  steps	
  to	
  ‘see’	
  this	
  structure.	
  A	
  two-­‐dimensional	
  image	
  of	
  the	
  

structure	
  fares	
  slightly	
  better,	
  but	
  since	
  the	
  three-­‐dimensional	
  structure	
  is	
  the	
  

phenomenon	
  of	
  interest,	
  the	
  models	
  make	
  this	
  structure	
  more	
  semantically	
  salient	
  than	
  

their	
  two-­‐dimensional	
  counterparts.	
  

Thus,	
  we	
  can	
  now	
  see	
  the	
  role	
  that	
  ‘mirroring’	
  may	
  play	
  in	
  successful	
  epistemic	
  

representation:	
  it	
  may	
  contribute	
  to	
  the	
  semantic	
  salience	
  of	
  the	
  phenomenon	
  of	
  interest.	
  

For	
  instance,	
  the	
  London	
  Underground	
  map	
  more	
  closely	
  mirrors	
  the	
  railway	
  network	
  

than	
  the	
  parallel	
  lines	
  representation,	
  since	
  the	
  lines	
  in	
  the	
  network	
  do,	
  in	
  fact,	
  cross	
  one	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   24	
  

another.	
  Similarly,	
  while	
  molecular	
  structure	
  could	
  be	
  represented	
  in	
  a	
  number	
  of	
  

different	
  ways,	
  not	
  all	
  of	
  them	
  make	
  the	
  structure	
  as	
  easy	
  to	
  grasp	
  as	
  molecular	
  models	
  do.	
  

Thus,	
  even	
  though	
  mirroring	
  relations	
  are	
  not,	
  strictly	
  speaking,	
  necessary	
  for	
  successful	
  

epistemic	
  representation,	
  they	
  may	
  play	
  a	
  role	
  nonetheless	
  in	
  securing	
  semantic	
  salience.	
  

	
  

5.	
  Successful	
  visual	
  epistemic	
  representation	
  

I	
  have	
  outlined	
  two	
  features	
  that	
  contribute	
  to	
  successful	
  visual	
  epistemic	
  

representation:	
  that	
  the	
  vehicle	
  of	
  representation	
  contain	
  sufficiently	
  accurate	
  information	
  

about	
  the	
  phenomenon	
  of	
  interest	
  for	
  the	
  user's	
  purpose,	
  and	
  that	
  it	
  effectively	
  convey	
  this	
  

information	
  to	
  the	
  representation	
  user.	
  Both	
  of	
  these	
  features	
  may	
  be	
  present	
  to	
  varying	
  

degrees.	
  Accordingly,	
  vehicles	
  of	
  epistemic	
  representation	
  may	
  be	
  more	
  or	
  less	
  successful,	
  

depending	
  on	
  the	
  extent	
  to	
  which	
  these	
  features	
  are	
  realized.	
  

This	
  is	
  especially	
  true	
  because	
  sometimes	
  one	
  feature	
  of	
  successful	
  epistemic	
  

representation	
  comes	
  at	
  the	
  cost	
  of	
  another.	
  Consider	
  again	
  the	
  map	
  of	
  the	
  London	
  

Underground.	
  Recall	
  that	
  users	
  of	
  this	
  map	
  are	
  interested	
  in	
  possible	
  routes	
  between	
  

stations,	
  with	
  the	
  ultimate	
  aim	
  of	
  determining	
  which	
  of	
  these	
  is	
  most	
  efficient.	
  Given	
  this	
  

aim,	
  one	
  might	
  argue	
  that	
  this	
  map	
  could	
  be	
  made	
  to	
  better	
  serve	
  its	
  users	
  if	
  it	
  depicted	
  the	
  

stations	
  and	
  railway	
  lines	
  to	
  scale,	
  thereby	
  representing	
  possible	
  routes	
  between	
  stations	
  

more	
  accurately.	
  After	
  all,	
  this	
  would	
  give	
  users	
  the	
  information	
  they	
  currently	
  are	
  able	
  to	
  

extract	
  from	
  the	
  standard	
  map,	
  as	
  well	
  as	
  other	
  potentially	
  relevant	
  information.	
  For	
  

example,	
  it	
  would	
  be	
  possible	
  to	
  better	
  estimate	
  travel	
  times	
  from	
  such	
  a	
  map14	
  or	
  to	
  

determine	
  when	
  it	
  might	
  be	
  worthwhile	
  to	
  walk	
  between	
  stations,	
  rather	
  than	
  taking	
  the	
  
	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
14	
  Of	
  course,	
  in	
  order	
  to	
  do	
  this,	
  one	
  would	
  have	
  to	
  take	
  into	
  consideration	
  things	
  like	
  congestion	
  at	
  each	
  
station,	
  which	
  would	
  determine	
  how	
  long	
  trains	
  stop	
  there.	
  	
  	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   25	
  

Underground.	
  This	
  additional	
  information	
  would	
  make	
  the	
  more	
  accurate	
  for	
  the	
  purpose	
  

of	
  determining	
  the	
  most	
  efficient	
  route	
  between	
  stations.	
  

However,	
  such	
  an	
  increase	
  in	
  accuracy	
  would	
  come	
  at	
  the	
  expense	
  of	
  the	
  syntactic	
  

salience	
  resulting	
  from	
  the	
  stations	
  being	
  positioned	
  evenly	
  along	
  the	
  lines	
  and	
  the	
  lines	
  

crossing	
  one	
  another	
  only	
  at	
  ninety	
  or	
  forty-­‐five	
  degree	
  angles.	
  For	
  in	
  this	
  case,	
  the	
  areas	
  

of	
  the	
  map	
  corresponding	
  to	
  parts	
  of	
  the	
  Underground	
  in	
  which	
  stations	
  are	
  close	
  together	
  

would	
  be	
  densely	
  populated	
  by	
  station	
  marks	
  and	
  labels,	
  making	
  it	
  difficult	
  to	
  distinguish	
  

between	
  the	
  stations.	
  And	
  the	
  ability	
  to	
  easily	
  locate	
  different	
  stations	
  on	
  the	
  map	
  is	
  

essential	
  for	
  the	
  map’s	
  performing	
  its	
  primary	
  epistemic	
  function:	
  to	
  allow	
  users	
  to	
  most	
  

efficiently	
  determine	
  the	
  best	
  routes	
  between	
  stations.	
  

Tradeoffs	
  between	
  the	
  accuracy	
  of	
  the	
  information	
  contained	
  in	
  the	
  vehicle	
  and	
  the	
  

extent	
  to	
  which	
  that	
  information	
  is	
  conveyed	
  to	
  the	
  user	
  exist	
  not	
  as	
  a	
  matter	
  of	
  law,	
  but	
  as	
  

a	
  matter	
  of	
  fact.	
  In	
  principle,	
  there	
  is	
  nothing	
  precluding	
  a	
  vehicle	
  to	
  both	
  contain	
  

information	
  that	
  is	
  maximally	
  accurate	
  for	
  the	
  user’s	
  purpose	
  and	
  to	
  effectively	
  convey	
  

that	
  information.	
  In	
  practice,	
  however,	
  it	
  is	
  difficult	
  to	
  fully	
  realize	
  both	
  of	
  these	
  features	
  at	
  

the	
  same	
  time,	
  except	
  perhaps	
  in	
  the	
  simplest	
  cases.	
  The	
  crucial	
  point	
  is	
  that	
  the	
  success	
  of	
  

visual	
  epistemic	
  representation	
  depends	
  on	
  both	
  features	
  and,	
  accordingly,	
  how	
  well	
  they	
  

are	
  balanced	
  with	
  respect	
  to	
  the	
  representation	
  user	
  and	
  the	
  purpose	
  for	
  which	
  she	
  wishes	
  

to	
  gain	
  information	
  about	
  the	
  phenomenon	
  of	
  interest.	
  	
  

	
  

6.	
  Conclusion	
  

	
   I	
  have	
  argued	
  that	
  the	
  success	
  of	
  a	
  visual	
  epistemic	
  representation	
  depends	
  on	
  the	
  

extent	
  to	
  which	
  it	
  contains	
  sufficiently	
  accurate	
  information	
  about	
  the	
  phenomenon	
  of	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   26	
  

interest	
  for	
  the	
  user’s	
  purpose	
  and	
  makes	
  this	
  information	
  readily	
  accessible	
  to	
  her.	
  In	
  so	
  

doing,	
  I	
  have	
  introduced	
  some	
  ways	
  in	
  which	
  we	
  may	
  compare	
  representations	
  to	
  one	
  

another,	
  determining	
  the	
  degree	
  to	
  which	
  they	
  are	
  successful	
  in	
  a	
  particular	
  context,	
  

defined	
  in	
  terms	
  of	
  a	
  user’s	
  background,	
  training,	
  cognitive	
  capacity	
  and	
  interests.	
  Thus,	
  

this	
  paper	
  develops	
  accounts	
  focused	
  on	
  the	
  user-­‐	
  and	
  purpose-­‐specificity	
  of	
  

representation	
  (mentioned	
  in	
  the	
  beginning	
  of	
  Section	
  2)	
  by	
  specifying	
  how	
  the	
  user	
  and	
  

purpose	
  affect	
  the	
  choice	
  of	
  representational	
  vehicle	
  as	
  a	
  tool	
  for	
  gaining	
  information.	
  The	
  

purpose	
  determines	
  how	
  accurate	
  the	
  information	
  contained	
  in	
  the	
  vehicle	
  of	
  

representation	
  need	
  be,	
  while	
  the	
  background	
  and	
  cognitive	
  capacity	
  of	
  the	
  representation	
  

user	
  determines	
  what	
  is	
  required	
  for	
  that	
  information	
  to	
  be	
  effectively	
  conveyed	
  to	
  her.	
  	
  

	
   Rather	
  than	
  giving	
  a	
  set	
  of	
  necessary	
  and/or	
  sufficient	
  conditions	
  for	
  epistemic	
  

representation,	
  this	
  account	
  instead	
  identifies	
  features	
  that	
  contribute	
  to	
  its	
  successful	
  use	
  

as	
  a	
  tool	
  for	
  gaining	
  information.	
  It	
  thus	
  both	
  departs	
  from	
  the	
  standard	
  methodology	
  of	
  

analyzing	
  what	
  is	
  required	
  for	
  a	
  scientific	
  model	
  to	
  be	
  a	
  representation	
  of	
  its	
  target	
  system	
  

at	
  all.	
  An	
  approach	
  that	
  acknowledges	
  that	
  actual	
  representations	
  vary	
  with	
  respect	
  to	
  the	
  

extent	
  to	
  which	
  they	
  give	
  their	
  users	
  epistemic	
  access	
  to	
  a	
  phenomenon	
  of	
  interest	
  

promises	
  to	
  more	
  accurately	
  reflect	
  scientific	
  practice,	
  where	
  a	
  variety	
  of	
  representational	
  

vehicles	
  are	
  used	
  for	
  myriad	
  purposes.	
  	
  

The	
  account	
  developed	
  in	
  this	
  paper	
  focuses	
  on	
  visual	
  representation.	
  A	
  question	
  

for	
  future	
  work	
  is	
  whether	
  it	
  might	
  be	
  fruitfully	
  extended	
  to	
  other	
  modes	
  of	
  epistemic	
  

representation,	
  such	
  as	
  mathematical	
  models,	
  and	
  if	
  so,	
  how?	
  Like	
  visual	
  representations,	
  

the	
  success	
  of	
  mathematical	
  models	
  also	
  depends	
  in	
  part	
  on	
  their	
  containing	
  information	
  

that	
  is	
  sufficiently	
  accurate	
  for	
  the	
  user's	
  purpose.	
  The	
  notions	
  of	
  semantic	
  and	
  syntactic	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   27	
  

salience,	
  however,	
  are	
  likely	
  to	
  have	
  limited	
  applicability	
  to	
  mathematical	
  models,	
  since	
  

visualization	
  of	
  their	
  targets	
  may	
  be	
  impossible	
  or	
  misleading.	
  For	
  instance,	
  the	
  position	
  of	
  

a	
  particle	
  may	
  be	
  represented	
  by	
  a	
  probability	
  distribution,	
  i.e.	
  as	
  a	
  ‘superposition’	
  of	
  

different	
  positions.	
  Yet	
  it	
  is	
  impossible	
  to	
  visualize	
  what	
  it	
  would	
  mean	
  for	
  the	
  particle	
  to	
  

be	
  in	
  such	
  a	
  superposition.	
  Nevertheless,	
  mathematical	
  models	
  make	
  the	
  information	
  they	
  

contain	
  readily	
  available	
  to	
  users	
  who	
  have	
  the	
  requisite	
  training.	
  Whether	
  the	
  notions	
  of	
  

semantic	
  and	
  syntactic	
  salience	
  could	
  be	
  fruitfully	
  brought	
  to	
  bear	
  in	
  the	
  realm	
  of	
  

mathematical	
  models,	
  and	
  if	
  so,	
  how	
  important	
  they	
  would	
  be	
  for	
  understanding	
  these	
  

kinds	
  of	
  models	
  as	
  tools	
  for	
  conveying	
  information,	
  are	
  questions	
  worthy	
  of	
  further	
  

attention.	
  	
  

	
  

Acknowledgments	
  

For	
  comments	
  on	
  earlier	
  drafts	
  of	
  this	
  paper	
  and	
  helpful	
  discussions,	
  I’d	
  like	
  to	
  thank	
  

Gillian	
  Barker,	
  Jossi	
  Berkovitz,	
  Anjan	
  Chakravartty,	
  Alex	
  Djedovic,	
  Eugene	
  Earnshaw	
  

Whyte,	
  Roman	
  Frigg,	
  Fermín	
  Fulda,	
  Cory	
  Lewis,	
  Greg	
  Lusk,	
  Margaret	
  Morrison,	
  Isaac	
  

Record,	
  Mauricio	
  Suárez,	
  Jim	
  Weatherall,	
  and	
  Rasmus	
  Grønfeldt	
  Winther.	
  I’m	
  grateful	
  also	
  

to	
  audiences	
  at	
  the	
  Models	
  and	
  Simulations	
  6	
  conference	
  at	
  the	
  University	
  of	
  Notre	
  Dame	
  

and	
  the	
  Center	
  for	
  Philosophy	
  of	
  Science	
  at	
  the	
  University	
  of	
  Pittsburgh,	
  where	
  I	
  presented	
  

this	
  work.	
  Finally,	
  I	
  thank	
  two	
  anonymous	
  referees,	
  whose	
  constructive	
  comments	
  

significantly	
  improved	
  this	
  paper.	
  

	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   28	
  

References	
  

Adriaans,	
  P.	
  (2013).	
  Information.	
  In	
  F.	
  N.	
  Zalta	
  (Ed.),	
  Stanford	
  Encyclopedia	
  of	
  Philosophy	
  
(Fall	
  2013.).	
  

Bailer-­‐Jones,	
  D.	
  (2003).	
  When	
  scientific	
  models	
  represent.	
  International	
  Studies	
  in	
  the	
  
Philosophy	
  of	
  Science,	
  17(1),	
  59–74.	
  doi:10.1080/02698590305238	
  

Barwise,	
  J.,	
  &	
  Seligman,	
  J.	
  (1997).	
  Information	
  Flow:	
  The	
  Logic	
  of	
  Distributed	
  Systems.	
  
Cambridge:	
  Cambridge	
  University	
  Press.	
  

Batterman,	
  R.	
  W.	
  (2009).	
  Idealization	
  and	
  modeling.	
  Synthese,	
  169(3),	
  427-­‐446.	
  
Bolinska,	
  A.	
  (2013).	
  Epistemic	
  representation,	
  informativeness	
  and	
  the	
  aim	
  of	
  faithful	
  

representation.	
  Synthese,	
  (190),	
  219–34.	
  doi:10.1007/s11229-­‐012-­‐0143-­‐6	
  
Callender,	
  C.,	
  &	
  Cohen,	
  J.	
  (2006).	
  There	
  is	
  no	
  special	
  problem	
  about	
  scientific	
  

representation.	
  Theoria,	
  55,	
  67–85.	
  
Contessa,	
  G.	
  (2007).	
  Scientific	
  Representation,	
  Interpretation,	
  and	
  Surrogative	
  Reasoning.	
  

Philosophy	
  of	
  Science,	
  74(1),	
  48–68.	
  doi:10.1086/519478	
  
Dretske,	
  F.	
  (1981).	
  Knowledge	
  and	
  the	
  Flow	
  of	
  Information.	
  Cambridge:	
  Cambridge	
  

University	
  Press.	
  
Floridi,	
  L.	
  (2004).	
  Outline	
  of	
  a	
  Theory	
  of	
  Strongly	
  Semantic	
  Information.	
  Minds	
  and	
  

Machines,	
  14(2),	
  197–221.	
  doi:10.1023/B:MIND.0000021684.50925.c9	
  
Floridi,	
  L.	
  (2005).	
  Is	
  Semantic	
  Information	
  Meaningful	
  Data?	
  Philosophy	
  and	
  

Phenomenological	
  Research,	
  70(2),	
  351–370.	
  doi:10.1111/j.1933-­‐
1592.2005.tb00531.x	
  

Giere,	
  R.	
  N.	
  (2004).	
  How	
  Models	
  Are	
  Used	
  to	
  Represent	
  Reality.	
  Philosophy	
  of	
  Science,	
  
71(5),	
  742–752.	
  doi:10.1086/425063	
  

Giere,	
  R.	
  N.	
  (2009).	
  An	
  agent-­‐based	
  conception	
  of	
  models	
  and	
  scientific	
  representation.	
  
Synthese,	
  172(2),	
  269–281.	
  doi:10.1007/s11229-­‐009-­‐9506-­‐z	
  

Graham,	
  G.	
  (1999).	
  The	
  Internet:	
  A	
  Philosophical	
  Inquiry.	
  London:	
  Routledge.	
  
Grice,	
  H.	
  P.	
  (1957).	
  Meaning.	
  The	
  Philosophical	
  Review,	
  66(3),	
  377–388.	
  
Grice,	
  H.	
  P.	
  (1989).	
  Studies	
  in	
  the	
  Way	
  of	
  Words.	
  Cambridge,	
  Mass.:	
  Harvard	
  University	
  

Press.	
  
Hoover,	
  K.	
  (2012).	
  Pragmatism,	
  Perspectival	
  Realism,	
  and	
  Econometrics.	
  In	
  A.	
  Lehtinen,	
  J.	
  

Kuorikoski,	
  &	
  P.	
  Ylikoski	
  (Eds.),	
  Economics	
  for	
  Real:	
  Uskali	
  Mäki	
  and	
  the	
  Place	
  of	
  Truth	
  
in	
  Economics	
  (pp.	
  223–240).	
  London:	
  Routledge.	
  

Judson,	
  H.	
  F.	
  (1996).	
  The	
  Eighth	
  Day	
  of	
  Creation:	
  Makers	
  of	
  the	
  Revolution	
  in	
  Biology	
  
(expanded.).	
  Plainview,	
  New	
  York:	
  Cold	
  Spring	
  Laboratory	
  Press.	
  

Kulvicki,	
  J.	
  (2010).	
  Knowing	
  with	
  Images:	
  Medium	
  and	
  Message*.	
  Philosophy	
  of	
  Science,	
  
77(2),	
  295–313.	
  doi:10.1086/651321	
  



Successful	
  Visual	
  Epistemic	
  Representation	
  

	
   29	
  

Mäki,	
  U.	
  (2009).	
  MISSing	
  the	
  World.	
  Models	
  as	
  Isolations	
  and	
  Credible	
  Surrogate	
  Systems.	
  
Erkenntnis,	
  70(1),	
  29–43.	
  doi:10.1007/s10670-­‐008-­‐9135-­‐9	
  

Olby,	
  R.	
  (1974).	
  The	
  Path	
  to	
  the	
  Double	
  Helix:	
  The	
  Discovery	
  of	
  DNA.	
  London:	
  MacMillan.	
  
Piccinini,	
  G.,	
  &	
  Scarantino,	
  A.	
  (2010).	
  Computation	
  vs.	
  information	
  processing:	
  why	
  their	
  

difference	
  matters	
  to	
  cognitive	
  science.	
  Studies	
  In	
  History	
  and	
  Philosophy	
  of	
  Science	
  
Part	
  A,	
  41(3),	
  237–246.	
  doi:10.1016/j.shpsa.2010.07.012	
  

Scarantino,	
  A.,	
  &	
  Piccinini,	
  G.	
  (2010).	
  Information	
  without	
  truth.	
  Metaphilosophy,	
  41(3),	
  
313–330.	
  

Suárez,	
  M.	
  (2004).	
  An	
  Inferential	
  Conception	
  of	
  Scientific	
  Representation.	
  Philosophy	
  of	
  
Science,	
  71(5),	
  767–779.	
  doi:10.1086/421415	
  

Teller,	
  P.	
  (2001).	
  Twilight	
  of	
  the	
  perfect	
  model	
  model.	
  Erkenntnis,	
  393–415.