This chapter addresses the question of how to translate grammatical algorithms into the language of neuronal sets.

Regular Associations, Associative Rules

There has been some discussion about the question of whether the human mind and brain use neuronal principles and connections for processing grammatically related information, or whether it uses rules and algorithms specified by grammar theories (Elman, Bates, Johnson, Karmiloff-Smith, Parisi, & Plunkett, 1996; Pinker, 1994). This proposal suggests that these positions, which are sometimes considered to exclude each other, are, in fact, both correct. This is not meant in the sense that there are two modules or systems, one for neural networks and the other one for rule algorithms (Pinker, 1997), but in the sense that rules are abstract descriptions of the neuronal machinery, as they are, without any doubt, descriptions of aspects of human behavior and action (Baker & Hacker, 1984).

If rules and algorithms are adequate descriptions of aspects of human behavior and action, they must have a basis in neuronal structure and function. As stressed in the discussion of the McCulloch–Pitts theory (Section 6.1), a neuronal network can be reformulated using calculus or by a logical formula. It is therefore reasonable to ask which putative neurobiological counterparts exist for syntactic rules and, conversely, how a neuron circuit sensitive to serial order can be adequately described algorithmically.

The cortex is an associative memory and would therefore be difficult to imagine it ignoring the correlation of words and morphemes in its input.