The spread of fear: Non-perceptual generalization of conditioned fear responses in humans

The Pavlovian conditioning model suggests that an originally neutral stimulus (CS) can elicit conditioned responses (CR) after repeated pairings with a meaningful stimulus (US). A common observation is that other stimuli (generalization stimuli, GS) can also elicit these responses, even though these stimuli were never paired with the US. Such generalizing can be adaptive, as it allows to apply earlier acquired information on novel situations. However, emotional responses like fear can also spread to innocuous stimuli, which can contribute to the development of anxiety disorders. Hence, a good understanding of the mechanisms underlying generalization is required to understand fear better. So far, fear generalization research has mainly focused on the influence of perceptual similarities between stimuli. Much less is known about non-perceptual types of generalization. In this research we have focused on the influence of categorizing stimuli and functional similarities between stimuli on the generalization of conditioned fear and suppression in humans.

We borrowed techniques from the equivalence research area while we excluded potential influences from purely perceptual similarities and direct associative relations between stimuli. Two established methods were used to create de novo arbitrary categories of stimuli and functional similarity between stimuli (the matching-to-sample task and the common outcomes method, respectively). In both methods, explicitly training a limited set of stimulus relations results in the emergence of complex, indirect links between stimuli. In a matching-to-sample task, participants are taught to choose certain stimuli as belonging to a central stimulus. In a common outcomes task, two or more stimuli share an associative history with the same stimulus or event, as the latter systematically follows the presentation of both stimuli. It is assumed that only humans are capable of categorizing stimuli in the first task, while both humans and animals have succeeded in the latter task. Our first goal was to develop experimental protocols to study generalization of different conditioned responses between these stimuli. Our second goal was to investigate the effect of extinguishing these responses. More specifically, we sought to examine to what extent extinction of conditioned versus generalized responses would generalize over related stimuli as well.

In Section 1 of this thesis we focused on arbitrary categories created with a matching-to-sample task. We demonstrated partial generalization of danger expectancy and subsequently found generalization of extinction, but only if the original CS was extinguished (Chapter 2). From the results in a follow-up study we concluded that certain parameters in the matching-to-sample task may influence the amount of generalization, like the length of the training phase and whether or not the untrained stimulus relations are tested for (Chapter 3). Finally, we confirmed and extended our earlier findings described above about the generalization of acquisition and extinction, this time also in a conditioned suppression paradigm (Chapter 4).

Section 2 describes a series of experiments containing a common outcomes task. We first found partial generalization of danger expectancy and generalization of extinction, but only when the original CS+ was extinguished, which was rather consistent with our findings in section 1 (Chapter 5). A second experiment in this chapter ruled out a potential design flaw as an alternative explanation for our findings on extinction. In the next chapter, we confirmed these findings in a conditioned suppression paradigm (Chapter 6). Finally, we demonstrated that instructing participants about the associations between the involved stimuli and the outcome stimuli suffices to support generalization (Chapter 7).

In summary, we found robust, although partial generalization of multiple conditioned responses within arbitrary stimulus categories and between functionally similar stimuli in all our studies. Extinction can generalize between these stimuli as well, but only when the original CS is extinguished. These findings come from both research lines and therefore seem independent from the nature of the underlying stimulus relations.