Semantic memory refers to our knowledge of words, their meanings, and their relationships to each other and to the physical world. It may be thought of as a dictionary, encyclopedia, and thesaurus, all rolled into one (Tulving, 1972). A model of semantic memory refers to a description of how the semantic features of a word are represented, how these representations can be combined into larger units of meaning (such as phrases and sentences), what deductions can be made about a word based on the context in which it appears, and how word meaning is related to the perceptual systems that provide access to the world (McNamara & Holbrook, 2003).
In an object-based view of semantics, words are considered to be associated if the objects they refer to have shared properties. Closeness is a measure of the similarity between objects. The property investigated may be common features or inclusion in a common category. Category membership can be treated as a strongly weighted feature, making these two ways of classifying words essentially identical. As a quick example, consider the words cat and dog. Both have teeth, claws, fur and a tail. Additionally, both can be included in the category pets. In an object-based view of semantics, the word cat would be closely semantically related to dog.
As Stone et al (2004) points out, the cognitive science and sociocultural perspectives differ considerably in their focus to the study of language. However, despite the fact that the difference in orientation is due to the fact that the cognitive scientists have a long tradition of interest in grammar, while sociocultural group has been primarily interested in meaning and pragmatics (Stone et al, 2004, p.11), both groups share the same understanding of language productions stages, namely message, encoding of it into linguistic form, encoding of linguistic form into speech, decoding of speech into linguistic form, and decoding of linguistic form into a meaning. Accessing language appropriately depends on the organization of the knowledge base that stores words and their meanings. Meanings or concepts are stored in a semantic network, and words are stored in a lexical network.
According to Collins and Loftus (1975), concepts are stored as nodes that are organized by semantic similarity or relatedness. The greater the similarity between two concepts, the closer the concepts are stored within the network. For example, the concept of sparrow may be stored close to eagle, but hawk may be stored closer to eagle because these birds have more in common. Each node in the lexical network is linked to its corresponding concept node. When a concept is activated by a word that is heard or read, this activation spreads from this concept to the most closely related nodes. As this activation spreads outward, its strength decays over time. This spreading activation process is considered to be automatic. Automatic activation occurs subconsciously or without an individual’s intent, and it is fast acting and does not consume processing resources in working memory. After a short-lived “automatic” period of activation, word processing may continue in a strategic manner known as controlled processing. Controlled processing operates under the influence of conscious awareness, is slow acting, and consumes resources in working memory.
One procedure that allows researchers to distinguish automatic from controlled processing is a primed lexical decision task, in which participants decide whether a given letter string is a real word. In this task, words are presented in pairs, with the first word known as a prime and the second word as a target. Participants do not respond to the prime, but do make an overt response to the target. For example, if the prime-target pair is nurse-doctor, the prime nurse is presented for a set duration and requires no response. Next the target doctor is presented and the participant makes a recordable “YES or NO” response indicating whether the target is a real word.
Now, let us consider how the primed lexical decision task allows us to study processing within the semantic network. When nurse is activated in the lexical-semantic network, there is a corresponding activation of related nodes (e.g., doctor, hospital, patient, etc.). This early activation of related words, such as doctor, elicits a relatively fast lexical response. On the other hand, unrelated words, such as nurse-bread, are stored farther away from each other. Activation of nurse spreads to the activation of healthcare related nodes and, thus, is not likely to activate bread. Without prior activation, bread elicits a relatively slow lexical response.
The previous discussion of activation in semantic memory introduced the levels of automatic and controlled processing. These levels can be studied somewhat separately by manipulating aspects of the primed lexical decision task. First, let us examine how the manipulation is accomplished when the focus is on controlled processing. Controlled (strategic) processing operates under the influence of conscious awareness, is slow acting and is a mechanism of limited-capacity. In priming tasks, controlled processing may be evident before or after the target has been presented.
After a target is presented, postlexical controlled processing consists of mental operations influenced by the relationship between a prime and target. One postlexical strategy is semantic matching in which participants check the semantic relationship between a prime and a target. Viewing the prime salt followed by the target pepper activates a quicker “YES” response because these words “match” semantically.
The semantic matching strategy operates in reverse with an unrelated prime. A non-match (e.g., salt-robin) produces a bias for participants to respond “NO”, which is incorrect for the task of lexical decision. Because neurologically intact participants still make mostly accurate responses, it is possible that they suppress the urge to respond “NO.” This strategic inhibition requires effort in working memory and, therefore, is likely to add time to a lexical decision (Brown, Hagoort, Chwilla, 2000).
Automatic processing is considered to consist of the spreading activation process described earlier. Thus, spreading activation is fast acting, occurs subconsciously, and does not interfere with coexisting mental activity. Automatic spreading activation accounts for priming that occurs when the interval between the prime and target is so brief that it is nearly imperceptable. The activation spreads out following a structure that corresponds to the relatedness among concepts; and, thus, the semantic relatedness of primes is facilitative of targets at these very brief intervals.
On the other hand, spreading activation does not produce the inhibition that we might expect when a prime is not semantically related to the target. Inhibition is the result of a strategic or effortful mental activity. Therefore, more time between a prime and target is required for inhibition to develop. An unrelated prime, like a neutral prime, acts merely as an uninformative or misleading variable.