There have been several theories proposed to account for the organization of the mental lexicon, from early psychological studies using word association to the idea of ‘spreading activation’.
Originally, behaviourists believed that a word derived its meaning from the sum of its associations. Therefore, the word ‘dog’ would be made up of its features, ‘bark’, ‘tail’, etc. and could be located in the lexicon with these attributes. This approach does not take into account a hierarchical structure and would not be an economical system to use to organize the mental lexicon.
Collins and Quillian (1969) refined and expanded on the use of word association. They proposed a hierarchical structure in which each word or concept would be represented by a node and each node would be linked higher up to a superordinate or hyperonym, such as bald eagle -> eagle -> bird of prey -> bird -> animal. This model is a more efficient method of storage, the idea being that by adding the feature of ‘wings’ to the bird node it is not necessary to add it to all instances of particular birds.
To test this theory they used word association experiments. They would ask people questions, such as “Is a robin a bird” and “Is a robin an animal”, and timed how long it took to respond. Collins and Quillian believed that it should take longer to respond to the second question because the node for ‘animal’ is higher.
However, there are a number of problems with Collins and Quillian’s theory. Firstly, not every concept can be represented in a hierarchical structure. Concepts such as ‘democracy’, ‘justice’, and ‘truth’ do not fit easily into such a system.
Secondly, the questions used by Collins and Quillian have been shown to lead to skewed results. Wilkins (1971) has shown that due to ‘conjoint frequency’ – how often words appear in a sentence together – the time of response can be altered. The words ‘robin’ and ‘bird’ appear more frequently than ‘robin’ and ‘animal’ and would be more associated together, therefore eliciting a faster response. When conjoint frequency is taken into account the time difference between responses becomes negligible.
Thirdly, responses to certain questions are not always as would be expected under the hierarchical model. For example, Rips, et al (1973) have found that the statement “A cow is an animal” receives a response more quickly than “A cow is a mammal” despite ‘animal’ being a higher node than ‘mammal’.
Schaeffer and Wallace (1970) have also shown that a statement such as “A pine is a church” will not be said to be false as quickly as “A pine is a flower” despite both being equally untrue. This is due to the ‘relatedness effect’. Simply, that if two things are related, despite not being from the same class, it is harder to separate them from each other.
Rosch (1973) has pointed out how the ‘prototypicality effect’ can also be a factor on speed of response. A ‘robin’ is considered to be a more typical bird than a ‘penguin’, so it will receive a more prompt response verifying it as a bird.
Collins and Loftus (1975) refined the network model already put forth by adding what is called ‘spreading activation’. In this revised model the network has become more complex and nodes are linked depending on strength and distance, rather than a purely hierarchical structure. The new model does retain certain parts of the hierarchical structure with the superordinate’s still in place, but the spreading activation element means that concepts such as ‘robin’ will activate other similar concepts, such as ‘bird’, because there is a strong link between them. There will be more distance between ‘penguin’ and ‘bird’, but ‘penguin’ will activate other strong concepts.
Although spreading activation does not describe any physiological process, it could be said to be analogous to the way that nerve cells operate in the brain. The nodes of the semantic network are connected and activating one will spread activation to the other areas they are connected to, similar to the way nerve cells can activate other areas of the brain.
With the advances made in artificial intelligence by the 1980’s, attempts were made at designing a lexical database to represent the semantic network that humans have in their mind. A group from the Cognitive Science Laboratory from Princeton University, headed by George A. Miller, created WordNet. Originally restricted to the English language, WordNet has been such a success that other researchers have now reproduced it for different languages. In WordNet “nouns, adjectives and verbs each have their own semantic relations and their own organization determined by the role they must play in the construction of linguistic messages” (Miller and Fellbaum, 1991).
From WordNet we can discover more about the conceptual-semantic relations between parts of language. We can see that nouns are connected to each other through hyponymy, such as the ‘robin’ – ‘bird’ relationship. They are also connected through meronymy, or a part-whole relation, for example ‘blade’ as part of ‘knife’. The construction of other non-English language WordNet’s have led researchers to look for lexical gaps through nouns and “might yield some generalizations pertaining to the lexicalization patterns in different languages and allow for a new perspective” (Fellbaum; 1998).
Verbs are connected through troponymy in WordNet. This means they are an action carried out in a particular manner, like ‘swipe’, ‘sock’, ‘smack’, and ‘tap’ for ‘hit’. Verbs are also related by entailment. When people eat, they have to swallow, so eating entails swallowing. Backward presupposition is also a type of entailment as in ‘untie’ and ‘tie’. To untie something, there has to be something tied in the first place. Lastly, entailment can have temporal relations, as in ‘snore’ and ‘sleep’, because to snore you must be asleep. Verbs may also have an antonymic relation, or opposite pairing, such as ‘buy’ – ‘sell’.
Adjectives can be either predicative (“The blue house”) or non-predicative (“The former employee”) and these cannot be conjoined. Unlike nouns and verbs, adjectives do not come in a hierarchical structure. Instead, the basic relation between them is an antonymic one.
From WordNet we can see that words arrange themselves through their syntactic class and are not arranged semantically. The lexicon would, therefore, be organized first syntactically and then conceptually in the mind. From aphasics we can see that some have lost certain words for types of fruit (Aitchison, 1994), but these are all nouns grouped together under one hyperonym. It is a small group of nouns around a certain concept that have been lost, not any verbs nor adjectives, because in the mind this is the organization of the lexicon.
Bibliography
Aitchison, Jean. 1994. Words in the mind: An introduction to the mental lexicon. Oxford: Blackwell.
Collins, A.M. & Quillian, M. R.. 1969. Retrieval time from semantic memory. Journal of Verbal Learning and Verbal Behaviour, 8, 240-247.
Collins, A. M. & Loftus, E. F.. 1975. A spreading activation theory of semantic processing. Psychological Review, 82, 407-428.
Fellbaum, C.. 1998. A semantic network of English: The mother of all WordNets. Computers and the Humanities, 32, 209-220.
Miller, G. A. & Fellbaum, C.. 1991. Semantic networks of English. Cognition, 41, 197-229.
Rips, L. J., Shoben, E. J., & Smith, E. E.. 1973. Semantic distance and the verification of semantic relations. Journal of Verbal Learning and Verbal Behaviour, 12, 1-20.
Rosch, E.. 1973. Natural categories. Cognitive Psychology, 4, 328-350.
Schaeffer, B., & Wallace, R.. 1970. The comparison of word meanings. Journal of Experimental Psychology, 86, 144-152
Wilkins, A. J.. 1971. Conjoint frequency, category size, and categorization time. Journal of Verbal Learning and Verbal Behaviour, 10, 382-385
Washing My Brain 
Leave a comment