plan.txt

0 objects in common
1 objects in common
< k objects in common

  DEV
  draw samples
  test various weightings
  human acc judgments ("left", "right", "neither", "both")
  human quality judgments

build fixed sampler w best weighing
train compiled model
train opt model?

  TEST
  get mode
  human acc judgments
  human quality judgments
  

#no objects in common
#  speaker weight vs listener acc
#  speaker weight vs human acc
#  listener acc vs human acc
#  human quality
#
#one object in common
#  speaker weight vs listener acc
#  speaker weight vs human acc
#  listener acc vs human acc
#  human quality
#
#by similarity
#  similarity vs listener acc
#  similarity vs human acc
#  listener acc vs human acc
#  human quality
#
#birds
#  speaker weight vs listener acc
#  speaker weight vs human acc
#  human quality
#
#
#model name
#speaker acc
#listener acc
#target id
#distractor id
#sentence