Understanding the Brain p. 7 Josephson
main scenario assumed to be implicated in learning sign-referent connections and determining the design of the
neural systems involved.
As in previous cases, the target moves on to more advanced possibilities once a given kind of target has been
achieved. The basic scenario just discussed is one where communication involves single signs, which we
identify with the instances of a specific type of system (lexicon, say). The abstract system on which the
hardware is assumed to be patterned would be superseded by a series of alternative scenarios, based on particular
abstractions, extending the range of usability of language. One extension is straightforward, involving
performing the elementary sign generation process more than once. This leads naturally to scenarios where more
than one sign is involved, but without any specific constraints on the order. This basic form opens up certain
possibilities for more advanced variants, including one mentioned above, involving the use of symbol position
to convey basic semantic relations. The way such a capacity can evolve is for it to be naturally present in an
earlier stage in evolution in a very inefficient form, in which case circuits may evolve especially adapted to the
performance of the task concerned in an efficient manner; in other words, a new abstraction starts to have an
important influence on the design.
We move on now to consider the role played by syntax in the language system; which abstraction might be
relevant here? A general feature of language is the relationship between syntactic form and semantic form, but
we cannot simply postulate a general ability to relate the two, and neither does Arbib s (2000) complex
imitation human skill especially help mediate the syntactic aspect of language. A more appropriate mechanism
invokes the ability to handle competently two things happening at once: for example with a speaker discovering
that an existing process fails because there is no word corresponding to some construct that is being represented
verbally (e.g. an object needs to be indicated by more than a single lexical item in order to avoid ambiguity), so
starting a new activity while keeping the other on hold. The listener, correspondingly, may be able to detect
that two distinct processes are occurring and again try to keep track of what is happening.
We may tentatively hypothesise an abstraction conceptual unit , associated with a specific system that is active
when the concept concerned is active, as an explanatory mechanism, such systems being aspects of a general
information management process, the functional equivalent of a set of boxes in which information may be
stored temporarily (there being possibly a connection here with the mechanism for pre-planning a grasping
action, regarded by Arbib as a precursor of a component of the language system). Roughly in accord with
Jackendoff s parallel architecture involving three systems, phonology, syntax and semantics, such a
manipulation system may be work together both with a grammatical unit such as a phrase, and with a node in a
semantic structure. In the process of attempting to connect language and meaning, such a unit could assist by
forming provisional links with units in the systems that are being connected. Developments of the language
system could investigate and later utilise processes combining units of this kind in various ways, thereby
building up structures of organisation equivalent to syntactic structures.
Note that Arbib s hypothesised complex imitation does not feature directly in this account, which is better
considered as referring to an alliance between a system for simple imitation and a system for managing
concepts, the essential mechanism for developing complex language being manipulation or management
abilities applied to the different problem of connecting linguistic structure with meaning. We will assume,
without attempting to justify this assumption, that training mechanisms using such manipulative mechanisms
can lead to the ability to interconvert between linguistic utterances and semantic structures in the way
characteristic of language. The concept of support by an environment, referred to in earlier sections, is relevant
in connection with discovering relationships between language and meaning in a language system: just as one s
ability to create a viable plan can be assisted by testing the plan against the reality, one s ability to decode
linguistic messages can be assisted by the referents being ready to hand to act as confirmation that one is
applying the correct process.
Next we discuss the typing (e.g. noun phrase, verb phrase) characteristic of universal grammar descriptions of
language. In combination with the above discussion we identify this with processes that assign types to the