SEMANTIC NETWORK
A representation of the way in which concepts are related to each other, showing the connections in terms of nodes and links.
The first such model, the Hierarchical Network Model (Collins and Quillian, 1969), attempted to represent not simply the links between lexical items but also the features which characterised each item. It was based upon relations of hyponymy, and consisted of a hierarchical structure with a node for ANIMAL at the top and nodes for basic level items such as BIRD and FISH below it. Below BIRD were subordinates such as CANARY or OSTRICH. Attached to each node were the attributes (properties) associated with it. These attributes were only stored at the highest possible level. Thus, features such as has skin, can move, eats, breathes were stored at the level of ANIMAL not that of BIRD– though BIRD shared them by implication since it was dominated by the ANIMAL node.
This model gave rise to a theory of semantic distance. It was hypothesised that it would take longer to confirm the truth of the ‘category statement’ A canary is an animal, than it would A canary is a bird, because the process involved crossing two nodes (CANARY ! BIRD !ANIMAL) rather than one. A similar result was expected with ‘property statements’ such as An ostrich has skin since, again, two nodes would have to be traversed in order to access a property which was not that of the ostrich alone but associated with animals as a whole. Both assumptions were supported by evidence. In addition a category size effect was recorded: the larger the size of the category, the more time was needed for the search.
The Hierarchical Network Model was criticised on various aspects of its design; and it was also found that the model’s predictions did not always hold true. The model was revised (Collins and Loftus, 1975), and based instead upon the principle of spreading activation. It now represents the way in which exposure to a lexical item enhances the recognition of other words associated with it. The effect is represented in terms of an electrical impulse running between connected items. This more recent model is no longer hierarchical; instead, there is massive interconnection. As before, concepts are represented as nodes, but now properties too can form nodes: CHERRY is linked to RED and RED to ORANGE and GREEN. The distance between concepts represents the strength of the connection between them: CANARY would be connected closely to SINGS but not closely to SKIN. The model thus accords with recent connectionist approaches to language processing.
See also: Lexical storage, Propositional network, Spreading activation
Further reading: Reeves et al. (1998: 196–202)
