To attain goals, people must proactively prevent interferences and react to interferences once they occur. Whereas most research focuses on how people deal with external interferences, here we investigate the use of proactive and reactive control in dealing with unwanted thoughts. To examine this question, we asked people to generate an association to each of several repeating cue words, while forbidding the repetition of associations. Reactively rejecting and replacing unwanted repeated associations after they occur entails slower response times. Conversely, proactive control entails constricting the search space and thus faster response times. To gain further insight into different potential proactive thought control mechanisms, we augmented the analysis of raw response times with a novel, hypothesis-based, tractable computational model describing how people serially sample associations. Our results indicate that people primarily react to unwanted thoughts after they occur. Yet, we found evidence for two latent proactive control mechanisms: one that allows people to mitigate the episodic strengthening of repeated thoughts, and another that helps avoid looping in a repetitive thought. Exploratory analysis showed a relationship between model parameters and self-reported individual differences in the control over unwanted thoughts in daily life. The findings indicate the novel task and model can advance our understanding of how people can and cannot control their thoughts and memories, and benefit future research on the mechanisms responsible for unwanted thought in different psychiatric conditions. Finally, we discuss implications concerning the involvement of associative thinking and various control processes in semantic fluency, decision-making and creativity.
Trying to stop thinking unwanted, often repetitive thoughts is a familiar experience. However, being aware of such attempts implies that the thought has already reached consciousness. Can we preempt an unwanted thought from coming to mind, similarly to how we can avoid taking an unwanted action? We examined this question using a free association task where people were instructed to avoid repeating associations. We show that to meet the task, people principally reject and replace unwanted associations after they have already reached consciousness. This rejection, however, is effective at ensuring that the same unwanted associations do not continue to come to mind endlessly. Furthermore, we found that people are able to control the extent to which associations become temporarily stronger when they repeat. Individual differences in these abilities could explain the difficulties that some individuals experience with unwanted and intrusive thoughts.
Introduction
Often, a particular cue can repeatedly evoke unwanted thoughts or memories. For example, a particular song or object can remind us of a painful, past romantic relationship, which we may not want to think of. People report using different strategies to control such unwanted thoughts or memories (e.g., trying to suppress the thought) [1–3]. Such conscious attempts for thought control reflect, by definition, a reactive process, occurring after an unwanted thought has already disturbed us (henceforth: reactive thought control). The current study investigates our ability for proactive thought control. Returning to the example above, can one preempt break-up-related unwanted thoughts or memories from coming to mind when encountering cues associated with the lost relationship? Furthermore, even when trying to reactively distract oneself from painful memories evoked, for instance, by a particular song, what mechanism can ensure that the same unwanted memory will not continue to come up over and over again? More generally, studying proactive thought control mechanisms has important implications for understanding how we avoid constant interference by unrelated thoughts, and why some people experience more intrusive, repetitive, or simply task-irrelevant thoughts than others [4–8].
Whereas proactive control has been investigated for decades [9,10], the recent dual mechanisms of control framework [11,12] offers a comprehensive account of the distinction between proactive and reactive cognitive control. Proactive control includes sustained anticipatory or very early selection mechanisms, biasing attention, and perception towards current goals. Conversely, reactive control corresponds with a transient, ’late correction’ mechanism, kicking in only after the unwanted event. These mechanisms are assumed to involve partially dissociated neural mechanisms [11,13] and different behavioral signatures. Specifically, whereas reactively inhibiting interfering stimuli slows down responses, proactive control can help avoid increasing response times (RTs) [13,14]. Proactive control thus guarantees smoother performance with less interruptions, but it is resource-demanding, and depends on the predictability of possible interferences [14,15].
We are not aware of any investigations directly contrasting proactive and reactive thought control. Interestingly, however, this distinction is implied by the two main accounts of the classical white bear phenomenon, wherein asking people to suppress a specific thought often has the paradoxical effect of increasing its frequency [16,17]. Reactive thought control is implicated in this phenomenon by the suggestion that people comply with the request to suppress the thought by trying to distract themselves when the thought comes to mind [18]. However, it has also been suggested that the to-be-suppressed thought is monitored by a continuous, unconscious process [19], potentially preempting the thought from reaching consciousness altogether. It thus remains unclear whether thought control in this task, or in any other setting, is primarily reactive or proactive, or involves an interaction between the two. Indeed, empirically investigating the generation and control of naturally occurring, daily thought is a challenging endeavor.
To simplify the problem, we focus here on thought control in the specific case of associative memory retrieval. Tasks probing associative memory are a useful tool for investigating thought processes, as evidenced by the relationship between people’s responses in a free association task and their everyday thoughts [20]. Numerous studies have demonstrated people’s ability to intentionally forget memories learned in the lab. A classic example is offered by directed forgetting experiments, showing that people are able to voluntarily forget specific words or entire lists of words upon instruction [21,22]. Efficient forgetting is also observed in the think-no-think paradigm. In this paradigm, participants are trained to produce a specific word upon presentation of a specific cue [23]. Then, participants are shown each cue and are instructed to either think or suppress thoughts of its associated word. The typical finding is reduced memory of suppressed words on a later recall test. Although it is still debated whether this reduced memory reflects interference, weakening of associative binding, or weakening of the memory itself [24–27], these findings imply that people can intentionally weaken the strength of associations formed in the lab. Critically, evidence for intentional forgetting of long-term (e.g., autobiographical) memories has been either absent [28] or limited to paradigms wherein such memories were linked to unrelated cues in the lab [29–31]. Thus, whether people can prevent predominant, long-term memories from coming to mind is unclear. Extant intentional forgetting paradigms are inherently unsuitable for answering this question because they measure recall accuracy and long-term memories are unlikely to be completely forgotten even after intensive suppression attempts.
The idea that semantic cognition involves a variety of specialized control mechanisms has been extensively studied with respect to the processing of externally presented stimuli [32–36]. To date, most such studies have focused on reactive control mechanisms. For example, delayed responses to a word (e.g., fork) that is weakly related to a prime preceding it (e.g., table) engage reactive control mechanisms required for inhibiting the dominant meaning of the prime (e.g., table-chair) [37]. However, a recent study has shown that when people know in advance what aspects of a semantic stimulus they ought to focus on (e.g., the use of an object vs. its physical features) they can proactively gate predominant associations from interfering with the processing of task-relevant stimuli [38]. Although this preliminary finding resonates with the idea that proactive control depends on the extent to which the stimuli to-be-ignored are predictable [12,14,39], these and other studies of semantic control examined the processing of externally-presented stimuli rather than the generation of thoughts or associations.
Thus, despite the great interest in control mechanisms in episodic and semantic memory, whether people can preempt specific consolidated associations from coming to mind remains unknown. For this purpose, we introduce a novel paradigm examining the frequency and speed with which people spontaneously recall unwanted associations. In this modified free association task, participants are presented with verbal cues, each repeating several times, and are instructed to not repeat associations they already had in response to previous presentations of a cue (suppress group). To achieve this goal, participants may either proactively preempt the repeated associations from coming to mind in the first place, or reactively seek an alternative association each time a repeated association comes to mind. The degree to which participants use either method can be determined by comparing the produced associations and the speed with which they are produced to those of a control group to whom repeated associations are not forbidden. By focusing on association frequency and speed, as opposed to the binary measure of accuracy employed in instructed recall paradigms, this task offers a much more sensitive test of proactive control over long-term memories.
Reactive thought control entails one central prediction–that the need to reject a repeated association (`Chair`in Fig 1A) and come up with a new association (`Desk`) delays the response. Thus, new associations to repeated cues (i.e., new/repeated trials) will be slower than associations to cues presented for the first time (i.e., new/new trials). One caveat here is that new/repeated trials tend to involve weaker associations (since the associations that have already been reported tend to be strong), and weaker associations are expected to be slower regardless of whether repeated associations are forbidden [40]. Thus, a more specific prediction entailed by reactive thought control is of a more pronounced slowdown in the suppress group compared to the control group.
PPT PowerPoint slide
PNG larger image
TIFF original image Download: Fig 1. Possible mechanisms for avoiding associations generated in previous trials. The figure illustrates pure reactive (A), and pure proactive thought control (B), as well three mechanisms of latent proactive control (C-E). Each row displays in the rightmost trial how one (`Chair`in panels A-C) or two (`Chair`and `Eat`in panels D-E) prior associations are avoided in favor of a new association (‘Desk`). Partial latent proactive control may be achieved by mitigating the natural, episodic enhancement of repeated associations (compare panel C with panel A), or by triggering proactive control after a first candidate association is rejected (panels D and E). The latter, post-rejection mechanism may be able to prevent any repeated association (D) or only the association that has just been rejected (E). Line thickness corresponds to associative strength. Dashed lines represent possible associations whereas solid lines represent the actual enacted policy. Reported associations are highlighted in blue. The unique characteristics of each solution are highlighted in yellow, shown only for the rightmost trials, together with green circles denoting acceptance and reporting of an association and red circles denoting rejection of a generated association, leading to an attempt to think of an alternative association. https://doi.org/10.1371/journal.pcbi.1010285.g001
Conversely, proactive thought control in our task may be implemented by means of several different mechanisms. One possible mechanism involves the reduction of the associative strength of an association once it is recalled for the first time (`Chair`in Fig 1B). This entails a facilitation of new associations to repeated cues (new/repeated RT < new/new RT; compare the thickness of the arrow pointing at `Desk`in the rightmost trial to that of the leftmost trial in Fig 1B), because preempting repeated associations implies a restriction of the search space, and thus less competition [40–42]. However, even if people are unable to weaken associations’ associative strength, they might be able to enact more subtle proactive control by mitigating a potential temporary strengthening of an association due to the fact that it just recently came to mind (compare the thickness of the arrow pointing at `Chair`in the rightmost trial in Fig 1C to the one in Fig 1A). The plausibility of this latter mechanism is supported by the above discussed findings of intentional forgetting of recently formed links.
Another, latent, proactive mechanism may rely on the predictability of specific interferences, which seems a key condition for proactive control over the processing of external stimuli [11,12,14,38]. Whereas it is impossible to keep in mind all of the associations previously generated for any cue [43], focusing on associations previously given to a specific, predictable cue could allow for the proactive gating of these associations. Say, for example, that a participant reported the associations `Chair`and `Eat`to previous presentations of the cue `Table`, and after several additional cues were presented, the cue ‘Table’ is presented again. The random order of the cues makes each presentation of ‘Table’ unpredictable, and this could make it difficult to proactively avoid the associations `Chair`and `Desk`. However, once the participant has reactively rejected one repeated association (e.g., `Chair`), they are now faced with the task of generating an alternative association to the same, this time predictable, cue. Consequently, the participant may now be able to proactively inhibit repeated associations (`Chair`and `Eat`; see Fig 1D), or at least those associations that have already been rejected in the current trial (`Chair`). A yet more limited possibility is that a participant is at least able to proactively avoid re-sampling an association immediately after its rejection, in a manner that may resemble classical predictability-based sensory gating mechanisms [44,45] (Fig 1E). This latter process could be relevant for understanding the difficulty that some people have in stopping a particular thought, and thus enter an endless, ruminative loop [46].
Examining raw RTs alone is insufficient for determining to what degree participants used each of the above mechanisms. Indeed, even if proactive control could be made possible after an initial rejection of an association (Fig 1D and 1E), this initial rejection is still expected to prolong RTs, even though preventing the need for further rejections will lead to a lesser delay than what might be predicted under pure reactive control. Similarly, even if participants might be able to fully avoid the natural increase in the associative strength of previously reported associations (Fig 1C), if these associations are naturally strong, they will still be likely to emerge in repeated presentation of a cue and require time-consuming reactive control. Thus, to investigate these latent modes of proactive control we developed a novel computational framework formalizing the process by which an association is sampled and then either accepted or rejected in favor of sampling an alternative association.
Several previous models have formulized memory recall as a sequential, Markovian process in which some (e.g., repeated) memories remain unreported (i.e., rejected) [47–49]. These models have been primarily applied to tasks asking participants to generate multiple responses to a cue (e.g., fluency task). Here we extend this approach to model the latent control processes preceding the report of a single association. Furthermore, whereas previous models have overlooked the fact that some associations are stronger than others, and that stronger associations take less time to recall, here we use a Semi-Markov process (SMP) model that allows states (i.e., associations) to vary in the time it takes to generate them, while retaining tractability. This allows distinguishing two plausible causes for increases in RT to repeated cues: repeated associations coming to mind and being rejected, and weaker association being generated because the strongest and fastest ones have already been generated to previous presentation of the cue. Dissociating these processes is necessary for our main goal: examining the degree to which avoiding repeated associations involves a) changes in the associative strength of repeated associations, b) rejection of repeated associations after they are generated, and c) changes in the associative strength of repeated associations immediately after the rejection of the first repeated association.
Importantly, whereas here we apply the SMP to the specific case of thought control in free association, this model may be useful for elucidating the mechanisms involved in any task in which proactive and reactive control interact, as well as other tasks involving sequential memory retrieval. For example, decision-makers faced with a realistic open-ended decision problem (e.g., how to spend a weekend) first need to generate their options, a process that has been formulized as a random-walk over semantic memory [50]. Furthermore, the interaction between semantic retrieval and control has been implicated in creativity research [51,52], yet a formal model is missing. We discuss these and other implications in the discussion.