Micro-relation units handle relations.

• In order to represent the patterns of similarity and difference across sequences of words in the Marcus et al. task ("rules"), we need to represent relations between the words.

• In a network with a binding dimension, the relation of sameness can be represented with an excitatory (and attractive) connection, but the relation of difference requires a special mechanism.

• One way to represent difference is with explicit micro-relation units. Each MRU represents an association between two features or groups of features which belong to different (sets of) objects.

• Each MRU has two micro-roles, each with its own value along the binding dimension. When the MRU is highly activated, its roles' values on the binding dimension match those of the two (sets of) features the MRU is associated with.

• To represent a pattern containing three "slots", such as the grammatical patterns in the Marcus et al. experiment, the network could associate two binary MRUs with one another, mapping the micro-roles in the proper way.