Collins and Loftus (1975) advanced a well known theory of semantic recognition which suggests that cognitive association involves nodes in a ``cognitive network'' which each contain semantic information. When one such node is activated (e.g., by association with an environmental stimulus) it activates other nodes to which it is connected. These nodes, in turn, activate other nodes to which they are connected. In this way, activation spreads throughout the network. Bower (1981) explained emotional associations in such a network by positing that some nodes contain exclusively affective content. Thus, association could occur as a function of both the semantic and affective content of incoming stimuli. For example, a stimulus such as a crying person might activate both person and sadness nodes in an observer's cognitive network.
The preceding explanation has been used to suggest that not only will the affective valence of stimuli influence a person's mood, but that their perception of a stimulus will be influenced by what emotion nodes are activated at a given time (i.e., their mood; Niedenthal & Setterlund, 1994). Thus, Niedenthal and Setterlund (1994) predict and find that an induced negative mood, which presumably activates a person's sadness emotion nodes, is associated with faster reaction times to words semantically associated with the ``sad'' (which presumably inherit some activation by being connected to the sadness node) than positive or neutral words on a lexical decision task. Ingram (1984) suggests that people who are depressed have strong activated connections between affective ``negativity'' nodes and multiple semantic concepts, creating feedback loops which serve to propagate or maintain depressive cognitions as portrayed in Figure 2, p. 27.
Traditionally, researchers subscribing to Bower's network theory have suggested that because depressed people begin a lexical decision task with activated nodes representing sadness, they begin the task with some residual activation of memory representations corresponding to negative stimuli, and not those corresponding to positive stimuli. Thus, these researchers predict that depressed people will be faster in responding to negative words than positive or neutral words on this task (e.g., Challis & Krane, 1988; Macleod et al., 1986; Matthews & Southall, 1991; Ruiz Caballero & Bermudez Moreno, 1992). As previously discussed, such predictions are not generally supported in the literature, and none of the theorists who made the predictions concluded that the results of their studies supported Bower's network theory. Potentially, it is erroneous to assume that lexical decisions for negative words are always facilitated by the activation of a memory structure representing negativity. Specifically, it will be suggested that the activation of a memory structure representing negativity activates multiple constructs with negative semantic content in the presence of a negative stimulus, creating interference, and thus slowing on the affective lexical decision task for negative words. The same would presumably not be true for positive words, for which less interference would occur. Bower's (1981) model might thus be appropriate even if depressed people are not facilitated with respect to negative words on the lexical decision task.