Misyak Et Al 2010

• Does the ability to learn statistical Non-adjacent dependencies correlate with the ability to process Non-adjacent dependencies in language?
• Can we model non-adjacent dependency learning with simple SRNs?
• allows us to see the continuous timecourse of statistical processing
• Uses both linguistic stimulus tokens and auditory cues
• on-line non-adjacency learning
• Investigation of Individual differences in language processing and statistical learning
• Participants trained in blocks of three word sequence trials.
• First and second word were random, but the third word was dependent on the first word.
  • Intervening second word creates non-adjacency
• After final block: Prediction task where participants had to say what the third word was from two word sequences
• People can learn non-adjacent sequences with only implicit exposure
• SRN can capture performance on AGL tasks
• SRNs can deal with temporal structures and associations
• Localist representations: 30 input and output units, each unique unit corresponding to each nonword
• Standard backpropagation with a learning rate of 0.1 and momentum at 0.8
• The higher the prediction task accuracy (x-axis) the shorter reading times for object relatives.
• Even the people who are bad at sequential learning are still fluent speakers and listeners
• Is it possible that sequential learning and language learning are unrelated
• Maybe children are better at sequential learning, which helps them acquire languag
• Adults then lose this ability