In our NIPS 2008 paper, A Bayesian framework for cross-situational word learning, we annotated two files from the CHILDES corpus (following Yu & Ballard, 2007) to create a corpus for use with our model.  

Source files of objects and words (aligned by line number)


Pseudo-CHAT annotated version of the files
The %ref tier refers to objects that are present at the time of the utterance.  The %soc tier contains seven tags and the object in the %ref tier that each one refers to: ih (infant's hands), im (infant's mouth), ie (infant's eyes), it (infant touching otherwise), ch (caregiver's hands), ce (caregiver's eyes), and ct (caregiver touching otherwise).


Gold standard (the human created lexicon used in our NIPS paper)

baby = baby
bear = bear
bigbird = bird
bigbirds = bird
bird = bird
book = book
books = book
bunny = bunny
bunnyrabbit = bunny
cow = cow
cows = cow
moocow = cow
moocows = cow
duck = duck
duckie = duck
eyes = eyes
hand = hand
hat = hat
kitty = kitty
kittycat = kitty
kittycats = kitty
lamb = lamb
lambie = lamb
mirror = mirror
pig = pig
piggie = pig
piggies = pig
rattle = rattle
ring = ring
rings = ring
sheep = sheep
oink = pig
bunnies = bunny
meow = kitty
birdie = duck
bird = duck
hiphop = bunny