Neural precursors of the spread of ideas
In the present study, we investigated the neural precursors of spreading ideas with enthusiasm from the perspective of the message communicator
. Although a growing body of work has examined the neurocognitive underpinnings of attitude change and behavior change from the perspective of the message recipient (Klucharev et al., 2008
; Mason et al., 2009
; Berns et al., 2010
; Campbell-Meiklejohn et al., 2010
; Zaki et al., 2011
), we know substantially less about what prompts people to share ideas enthusiastically. This type of investigation is especially relevant in the context of the new media environment, which facilitates word-of-mouth transmission of ideas to much wider networks.
In the current investigation, we examined the neural precursors of spreading ideas with enthusiasm as one way of beginning to understand the underlying processes that may lead to the successful spread of ideas. In particular, we dissect enthusiasm into component processes that can be uniquely identified through automated linguistic SA, through gestalt human ratings of combined linguistic and non-verbal cues, and points of convergence/divergence between the two. Given the growing desire and ability to leverage linguistic data to predict relevant outcomes (e.g., virality) in the context of the new media environment, understanding the overlap and divergence between mental processes captured by NLP and by gestalt human impressions is also of importance.
We hypothesized that the process of encoding information in a way that later results in enthusiastic dissemination might share common neural underpinnings with successful speaker-listener communication in general, and that ratings captured by an automated SA would be related to subsets of neural activity associated with gestalt human-coded ratings of speaker enthusiasm.
More specifically, we hypothesized that activity in regions that have been implicated in self-related processing (including MPFC, PC/PCC), reward (including VS, VMPFC), mentalizing (DMPFC, TPJ), and memory (including the MTL) during initial idea encoding may be associated with later enthusiasm expressed for those ideas.
More specifically, we hypothesized that ideas that resonate with a listener might also be more likely to be propagated by that individual. MPFC in BA10 and PC/PCC has been associated with self-related processing (Lieberman, 2010
), as well as subsequent behavior change following exposure to persuasive messages (Falk et al., 2010
; Chua et al., 2011
). We hypothesized that activity in these systems would be associated with later enthusiasm in communicating ideas. Indeed, both human coded enthusiasm and SA ratings of positivity were associated with activity in these regions during initial encoding.
However, we also hypothesized that personal connection to an idea should not be sufficient to prompt enthusiastic message propagation. Instead, outward expressions of enthusiasm for ideas also require an understanding of what others are likely to value (Krauss and Fussell, 1991
; Higgins, 1992
), and may involve consideration of what others are likely to think of us if we share (Engel et al., 1993
; Sundaram et al., 1998
; Hennig-Thurau et al., 2004
). Neural activity in DMPFC and TPJ are commonly associated with social cognition, perspective taking and mentalizing about the views of others (Lieberman, 2010
; Saxe, 2010
). We hypothesized that activity in these regions during initial encoding would be associated with participants' evaluations of ideas with respect to the value others would place on those ideas. In addition, in orthogonal analyses that we performed with this dataset, individual differences in participants' abilities to persuade others of the value of their preferred ideas was associated exclusively with activity in TPJ (Falk et al., 2012b
Indeed, human-coded enthusiasm scores were associated with activity in both of these regions, and more positive, evaluative sentiments (as coded by SA) was associated exclusively with increased activity in TPJ. Consistent with our initial hypotheses, it is possible that during initial idea exposure, increased perspective taking could have positioned participants to later argue more enthusiastically for the merit of the ideas in describing them to others, preparation that was evident both through automated linguistic analysis and human-coding. In other words, these data are consistent with the idea that preferences and recommendations may involve a contextualization of one's own thoughts with respect to those of the group. In reflecting on social influence, Allport (1954
) observed that not only are we swayed by those with whom we have direct interactions, but that we also behave in accordance with others who are “imagined or implied.” In the current investigation, we suggest the complement of this idea: that in taking in information, we may process both the value of the idea to ourselves, but also the value it is likely to have to others. To the extent that we deem the idea valuable to others, we may be more prepared to make a stronger recommendation and to argue in an evaluative fashion when describing the idea to others.
We also hypothesized that lower-level reward mechanisms could facilitate the spread of ideas from one person to the next; in this context, imagining that one will be able to tell another about a cool new television show might involve anticipation of a positive response from the other person (Engel et al., 1993
; Sundaram et al., 1998
; Hennig-Thurau et al., 2004
). The VMPFC and VS are regions commonly associated with encoding reward and value signals (Knutson et al., 2001
; McClure et al., 2004
; Knutson and Cooper, 2005
; Haber and Knutson, 2010
). Activity in VS and VMPFC are also associated with exposure to stimuli that are popular or valued by others (Plassmann et al., 2008
; Zaki et al., 2011
), and with conforming to the opinion of others (Campbell-Meiklejohn et al., 2010
Consistent with this hypothesis, we found that human-coded enthusiasm was associated with activity in VS and VMPFC during initial idea encoding. Automated SA ratings were not. One possibility is that activity in these regions tracks enthusiasm for ideas, which is subsequently encoded by non-verbal signals, or by language cues not identified with the current classification dimensions. On a broader scale, neural activity in VS and VMPFC has been shown to predict the cultural popularity of songs (Berns and Moore, 2012
). In other words, beyond encoding one's own evaluation of the incoming ideas, the VS and VMPFC might also encode value with respect to ideas in the social context. Thus, although our data also cannot speak to the flow of ideas across populations, they do hint at the possibility of common neural mechanisms supporting the spread of ideas from person to person, and the ultimate popularity of those ideas in larger groups of people.
Finally, we hypothesized that memory encoding processes, which are commonly associated with neural activity in the MTL including the hippocampus, (Cabeza and Nyberg, 2000
), as well as PC/PCC, implicated in retrieval of autobiographical memories, might be associated with enthusiastic message propagation. We found that both human coded enthusiasm and positivity as captured through automated SA were associated with neural activity in MTL.
In sum, we found that increased neural activity in several hypothesized networks previously associated with better speaker-listener communication (Stephens et al., 2010
) are also associated with encoding of ideas that are subsequently described with enthusiasm by a message communicator
. These regions include the MPFC, PC/PCC, VMPFC, VS, DMPFC, TPJ, and MTL. All of these regions were associated with gestalt ratings of enthusiasm as coded by trained human coders. One subset of these regions (MPFC, PC/PCC, DMPFC, and MTL) was associated with positive valence as classified through linguistic SA, whereas a different region TPJ was associated with more evaluative, positive descriptions, as coded by the SA.
In addition to activity in our hypothesized regions that have previously been implicated in social and affective processing, human-coded enthusiasm ratings were associated with neural activity in regions that are associated with shared sensorimotor-representations within the mirror-neuron system including the IPL and dPMC (Spunt and Lieberman, 2012
). Indeed, these regions were the primary regions associated with human coded enthusiasm scores when controlling for automated SA ratings. Thus, although prior work in the mirror neuron system literature has primarily focused on mirroring others who are seen, this type of mental simulation may also prepare individuals to share
ideas enthusiastically with others at later points in time.
Connections to the default mode network
It is also of interest that the regions observed to predict transmission of ideas with enthusiasm (VMPFC, MPFC, DMPFC, TPJ, and MTL) are often characterized as the default mode network (DMN)
. The DMN has been implicated in studies of mind wandering and other forms of stimulus independent thought; furthermore, increased DMN activity is often associated with performance decrements on tasks requiring attention or other forms of executive control (Schooler et al., 2011
). The ubiquity of mind wandering has led researchers to investigate the function and benefits of this process, however, most studies highlight increased error rates and decreased task performance with increased DMN activity; few studies have explicitly demonstrated increased task performance associated with DMN activity [c.f. recent work (Meyer et al., 2012
) demonstrating that increased activity in regions of the DMN are associated with better social working memory]. The current data suggest that activity in several regions of the DMN during the encoding of ideas is associated with more effective performance later on when pitching ideas to others. In other words, DMN activity is not always indicative of poor subsequent outcomes, and in fact, may be predictive of better task outcomes, when the task in question involves self-reference and/or social judgment. Regions of the DMN are also predictive of better outcomes when the “task” involves exposure to messages designed to facilitate positive behavior change (Falk et al., 2010
; Chua et al., 2011
Combination of neuroimaging with tools from natural language processing
Methodologically, we view this as a demonstration of the synergy of automated language analysis with neuroimaging data from fMRI studies. Although humans do not typically report specific linguistic features as contributing to their impressions of ideas, these features nonetheless are detectable by automated SA, and are associated with many of the same neural precursors as human gestalt ratings. In addition, some neural precursors are associated with SA ratings, but not human's explicit coding of communications. As such, SA, and other tools from NLP can facilitate more sophisticated understanding of the brain bases of social interaction and social cognition more broadly. For example, these tools provide a framework for analyzing data in which subjects engage in tasks that involve exposure to ideas, objects, or other socially relevant stimuli, and then provide free-form post-scan language samples expressing preferences or opinions (as opposed to relying exclusively on closed-ended reports). These methods would also allow new ways of integrating fMRI data with language recorded during other experimentally relevant social interactions (alone or in more complex groups) before, during or after the scan. The natural language data in question could include videotaped interviews (as in the present study), or other data relevant to social interaction [e.g., through the Electronically Activated Recorder; EAR; (Mehl et al., 2001
), or sharing of content online (Berger and Milkman, 2012
)]. The resulting language corpus can then be analyzed using NLP tools to provide metrics for sentiment, use of descriptive or interactive language features, and so on, that can be applied as parameters in the analysis of the fMRI data.
New, mobile media expand the circle of friends and acquaintances with whom individuals are in perpetual contact, and looking to for advice; these new media also create an unprecedented written record of the ways in which social influence unfolds. Combination of neuroimaging and NLP methods may also help to prospectively predict who is likely to share what, and in what manner (Eisenstein et al., 2010
; Tumasjan et al., 2010
; Falk et al., 2011
), as well as population level behaviors (Berns and Moore, 2012
; Falk et al., 2012a
) which lead messages to go viral. Future work in which participants type or otherwise communicate between scanners may also be of interest.
Given the multiple functions of each of the regions observed, further study will be required to test the psychological relationships posited; our discussion of possible psychological interpretations of these activations should be understood as one of many possible explanations. At a broader level, however, our results suggest that there are neural signals present during the initial encoding of an idea that are associated with the subsequent way in which the idea is conveyed to others. In prior work (Falk et al., 2012b
), many of the regions associated with our trained coders' enthusiasm ratings were also associated with ideas that spread successfully, though there were also points of divergence. In the present investigation, we begin to delve more deeply into the component processes that are associated with the intermediate step between wanting to share a great idea and the idea spreading through a culture. In particular, the results of our automated SA help dissect and contextualize the ratings made by human coders who assessed the enthusiasm expressed by each of the participants about each show using a gestalt heuristic. In particular, human coders' enthusiasm ratings captured an integrated picture of several systems at work during initial idea encoding.
Future research that takes the current findings as starting points for constructing a priori defined regions of interest (ROIs) could interrogate the psychology of the mental processes captured within these neural regions (e.g., by using activity within these regions to predict memory for specific ideas, perceived self-relevance of the ideas, associations with implicit attitudes, etc.). Within such a brain-as-predictor framework (Berkman and Falk, in press
), activity from a priori defined ROIs could also be leveraged to forecast the likely enthusiasm of communicators in spreading ideas, or even the ultimate virality of those ideas across populations.