Playpen learns rules.
- Layer of Word units (MOUs).
Hard-wired connections implement similarity relations necessary
for similarity-based segmentation.
When a sequence of words is presented, the units take on phase angles
representing the sameness or differences among the words in the
sentence.
- Layer of Sequence units (MOUs).
In the full model (not yet implemented), Sequence units associated
with specific Words and with clusters of Words.
In the simplified model, a single Sequence unit for each delay
associated with all Words.
- Layer of Grammar units (MRUs), one for each pair of Sequence
units.
- Network trained on 4 different instances of a "rule" (AAB,
ABB, or ABA).
- Following training, the Grammar layer is more highly activated
following a pattern that obeys the rule than one which
does not.
Michael Gasser