Playpen learns statistics.

• Layer of Syllable units.

• Layers of Sequence units (MOUs), one layer for each different (hard-wired) delay from the Syllable units, one unit in each layer for each syllable. Each subsequence appears as a static pattern on the Sequence layer. Sequence units connected to one another by trainable connections.

• Layer of Word units (MOUs), each connected to all Sequence units and to other Word units.

  

• Network presented with sequences of 4 3-syllable words.

• Following training, the Word layer is more highly activated following a word than a non-word (1.05 vs. 0.3323).

• Activated units in the Word layer were in phase following the presentation of a word (standard deviation = 0.0000244).

• Word layer patterns for different words were distinct (cosine between patterns for different presentations of the same word: 0.9993, between patterns for different words: 0.1196).