Conceptual Structure and Modularity

Cognitive science makes a number of divergent claims about the structure of cognition. Some of this divergence can be traced to philosophical or "metaphysical" predilections--which at times correlate well with core modes of contrual. Some stem from different methodologies or investigative practices: for instance, a linguist looking at patterns in semantics is not usually thinking in terms of informationally encapsulated modules, domain-specific adaptations, or the way in which these patterns may be instantiated in the neural organization of the brain.

Given that the mind operates in several modalities, however, the question of how these modalities are connected remains a central problem in several approaches. How is it, for instance, that we can speak about what we see? Visual information is of a very different kind from verbal information, and a computational model of the mind is faced with the question of how these two modalities are able to interface. Is there a lingua franca of the brain in which the various modalities--such as the visual, the aural, the verbal, and the kinaesthetic--can communicate and exchange information? If so, what is the relation of this "mentalese" to language?

The lingua franca of the mind is generally thought to be inaccessible to conscious awareness; however, a case can also be made for consciousness being precisely the site at which the modalities talk to each other. In this case, the forms of consciousness would be those that are comprehensible to a wide variety of cognitive domains. (For a discussion of this proposal, see my "clearing-house" model of consciousness.)

Jackendoff (1987) reasons that mentalese is likely to consist in a set of conceptual structures into which the various modalities of information can get translated and become available to the other modalities. He explores the possibility that the use of the same words across several semantic fields is an indication of such conceptual structures, or primitives.

Three examples of semantic fields

• Position (use a positional field modifier)
• Possession (use a possessional field modifier)
• Predication/Ascription (use an identificational field modifier)

Conceptual primitives GO, BE, and STAY across semantic fields

A. Spacial position

• GO verbs (the dog ran to the gate, the stone fell on his foot)
• BE verbs (I was in town, the fly sat on the cup, the pencil lay by the pen)
• STAY verbs (I stayed in town, the dog remained by the gate)

B. Possession

• GO verbs (he gave her the book, she bought the lamp)
• BE verbs (it belonged to him, she had no pets)
• STAY verbs (she kept the book, it remained in her possession)

C. Predication / Ascription

• GO verbs (the frog turned into a prince, the snow became soft)
• BE verbs (the frog was a prince, the apple seemed tasty)
• STAY verbs (the prince remained a frog, the snow stayed soft)

Jackendoff's cross-field generalization is that "the three major concepts GO, BE, and STAY apply to three semantic fields that a priori have nothing to do with each other" (156). They are thus candidates for constituent conceptual structures of mentalese.

This move raises a number of questions. On the one hand, are cross-field generalizations a genuine indication of conceptual structures, or are they an artifact of the investigator's own creative act of categorization? On the other hand, if we grant the significans of the cross-field generalizations, how do these map onto modularity theory?

Here is one model that proposes how conceptual structures allow modular output to be shared:

• Domain-specific output is classified into conceptual primitives (conceptual primitives represent similarities across domains)

• Conceptual primitives are mapped onto semantic fields (semantic fields represent types of contextual or relational information)

• The representation retains information about the cognitive domain (cognitive domain represents information about relevant entailments)

The main linguistic cue to domain is noun category:

• the storm subsided (physical)
• the barking subsided (zoological)
• his anger subsided (emotional)

Classification of nouns into domains governs domain-specific entailments

• “This is my wife”, “This is my nephew”, and “This is my pen” have very different sets of entailments (e.g. sexual access, kin relations, utility)

• The semantic field of possession, the conceptual primitive BE, and the domain signaled by the noun category are independent dimensions of meaning

Summary

• Conceptual structure consists of conceptual primitives, semantic fields, and domain-specificity

• These are independent axes of a multi-dimensional mapping that represents patterns in modular output, contextual relations, and domain-specific entailments

In this view, conceptual structures arise from a higher-level perception of patterns (similarities and differences) in domain-specific output. Such patterns are mapped onto three independent dimensions of meaning: conceptual primitives (similarities between domains), semantic fields (contextual and relational information), and cognitive domain (the relevant inference engine).

Conceptual structure and language

• Conceptual structure is a systematization of output from domain-specific modules, and is not itself domain-specific

• It represents a level of abstraction above domain-specific processing of sensory information, and implicitly depends on this prior level

• The resultant level of meaning is of an informational complexity that is suitable for linguistic representation and communication

• Only certain aspects of the total meaning get explicitly encoded in the language

• Language calls on domain-specific modules for its meaning, and is only fully interpretable by cognitive systems with a similar set of modules

In the case of literature, which may be considered a deliberate and excessive exploration of the tacit and implicit dimensions of ordinary language (Turner 20), this final result is of general significance. It suggests that any surface meaning encodes (in the sense of containing subtle pointers to) at least three dimensions of implicit meaning: conceptual primitives that derive from similarities between domains, semantic fields that represent contextual and relational information, and the inference characteristic of the domain the phenomenon has been placed in.