Comprehension: from clause to conspiracy narrative

ABSTRACT

In this article, I outline the evolution of my research in the domain of narrative comprehension. I cover three phases: (1) the event-indexing model, which explores connecting events in narratives; (2) the immersed experiencer framework, which focuses on event representations and mental simulations; and (3) a preliminary exploration of conspiracy thinking, in which I reveal parallels with coherence generation in narrative comprehension. I emphasize the interconnectedness of these domains and call for further investigation into the intricacies of narrative comprehension.

“Rolf, dinner’s ready!”

I can still recall my father’s firm yet gentle voice pulling me from some enchanting fictional realm back to the reality of my everyday life as a young teenager. Briefly disappointed that the world I returned to was not populated by hobbits, elves, ents, and, yes, even orcs, I would climb down the stairs to rejoin the family for the meal. These days, I find it more difficult to become immersed in narratives than I did back then. Still, it is fair to say that throughout my life, I have had a fascination with stories: how we understand them and how we can become engrossed in them. This fascination has driven my pursuit of various avenues of inquiry over the years.

My aim in this article is to describe these pursuits, tracing my intellectual evolution from its inception to my current interests. To underpin my line of argumentation, I will draw upon a series of theoretical and review articles that I have either authored or coauthored. These articles serve as the signposts along which I will map the progression of my thoughts on comprehension, moving from narratives to clauses and sentences to conspiracy narratives (Zwaan, 2004, 2014, 2016, 2022; Zwaan & Radvansky, 1998).

It is important to acknowledge at the outset that numerous scholars have contributed and are contributing to research in the domains that I will be discussing. In no way do I want to diminish their important contributions. However, the primary objective of this article is not to provide an exhaustive review of the very expansive body of literature on comprehension; rather, it is to articulate the course of my own ideas in this area of inquiry.

In the sections that follow, I will describe the three main interests of my research over the years. A fourth interest is open science and replication (e.g., Zwaan, Etz, Lucas, & Donnellan, 2018). Although close to my heart, it is less related to the focus of this article and so I will not discuss it further here. Instead, I will transition from discussing situation models to mental simulations and then consider conspiracy thinking, with the aim of demonstrating the interconnectedness of these topics.

Situation models

Comprehending narratives is not comprehending language per se. Rather, it is comprehending situations and events by way of language. This is the basic insight I took from my early readings about concepts such as mental models (Johnson-Laird, 1983) and situation models (Van Dijk & Kintsch, 1983). (As an aside, although it was far from a captivating story, I remember an occasion in which I was so engrossed in Johnson-Laird’s book that I missed my train stop and ended up having to walk a long distance to my destination.) The concept of the textbase, which was introduced several years earlier (Kintsch & van Dijk, 1978) focused on the analysis of propositions that can be derived from the text itself. Situation models, in contrast, take event representations as their basic units and consider how they are connected in the comprehender’s mental representation of the situation described in the text.

Zwaan and Radvansky (1998) was intended in part as an effort to elevate the status of the situation model. It had previously been perceived as a kind of receptacle for causal, spatial, and instrument inferences that were not considered part of the textbase because they were not explicitly referred to in the text but rather had to be inferred by the comprehender using semantic and episodic information retrieved from long-term memory.

The event-indexing model

The precise definition of an event remains a topic of ongoing debate within the field of cognitive science (e.g., Yates et al., 2023). However, the event-indexing model, which we presented in a number of articles (Zwaan, Langston, & Graesser, 1995; Zwaan, Magliano, & Graesser, 1995; Zwaan & Radvansky, 1998), represents an attempt to specify potential connections among these events and their effects on the comprehension process. More specifically, the event-indexing model identifies five dimensions along which events can be integrated, drawing inspiration from research in linguistics, computational linguistics, and discourse psychology.

Although researchers in these different fields might each entertain somewhat different taxonomies than we did, we identified five situational dimensions: time, location, entity, motivation, and physical causation. In the event-indexing model, each clause in a unit of discourse corresponds to an event (actions are subsumed under this heading). These events are characterized by indices on each of the five dimensions. An event

1. unfolds at a specific moment in time,

2. takes place within a particular location or spatial region,

3. involves one or more entities (primarily protagonists but also objects),

4. may be part of protagonists’ plans and goals,

5. may be either the physical cause or consequence of a previously mentioned event.

We derived two general hypotheses from this framework, one concerning the online processing of discourse, the continuity hypothesis, and one concerning the resulting representation in long-term memory, the overlap hypothesis. According to the continuity hypothesis, all other things being equal, the more indices the event currently being processed shares with the previously mentioned event, the easier it should be to integrate its representation into the evolving situation model. According to the overlap hypothesis, all other things being equal, the more overlap on the five dimensions there is between two event representations in long-term memory, the stronger the link between two these nodes should be.

To elaborate this line of reasoning a bit further, the basic idea behind the event-indexing model is that a story is situated in a specific spatiotemporal framework (e.g., from Bill cooking an egg in his kitchen to the French Revolution taking place in Paris) and that this provides the setting for a narrated series of events that are interconnected on one or more of the five situational dimensions. Consider the temporal dimension. When two events take place within the same time frame they share an index on the temporal dimension. Two events are part of the same time frame when they immediately follow one another (“He entered the building. A moment later, he realized he’d forgotten to lock his car.”) or when one event is encapsulated by another one (“The hikers spotted a deer while taking a rest on the mountain ridge”).

Similarly, when two events take place within the same spatial region, for example a room, they can be considered to be connected on the spatial dimension (e.g., “The guitarist was noodling on his guitar. Five minutes later, the drummer entered the studio”).

A key dimension is entity. This rather broad concept encompasses both human protagonists and objects. Just like protagonists, objects can sometimes be the focus of events (e.g., “The vintage guitar tumbled from the stage”). Two consecutively narrated events are connected on the entity dimension when they have the same focal entity (e.g., “The vintage guitar tumbled from the stage. Its neck was cracked”).

The fourth dimension is the goal dimension. When two events are part of the same goal-plan structure, they are connected on the goal dimension. For example, the sentences “Sifan wanted to run the marathon. She went to the track to do interval training” are connected on the goal dimension, given that doing intervals is part of marathon training. On the other hand, the sentences “Sifan was running in the woods. A deer crossed the trail” are not connected on the goal dimension, as the second event is not part of Sifan’s goal of training for the marathon.

The final dimension is physical causation. The sentences “Max revved the engine. The engine started to roar” are connected on the causation dimension, as revving the engine causes the engine to roar. Incidentally, these two events might be connected on the goal dimension as well, of example if Max was trying to impress his opponent Lewis with the sound of his engine. In contrast, the sentences “Max revved the engine. Lewis put on his gloves” are not causally related given that revving the engine does not cause someone to put on gloves.

Problems with the event-indexing model

This taxonomy of five dimensions is not without its problems. For one, the causal and temporal dimensions do not seem to be entirely separate. To give a simple example, if A occurred before B, B cannot be the cause of A. Moreover, the size of the time interval between two events may make a causal connection more or less likely. For example, if the vintage guitar from the previous example tumbled from the stage and a month later there was a crack in its neck, the tumbling episode is not a likely cause. However, in practice the occurrences of discontinuities on the temporal and causal dimensions are not highly correlated, as analyses of several literary narratives show (Zwaan, Magliano, & Graesser, 1995).

Another issue with the taxonomy is whether it is complete as is or whether additional dimensions should be added. Various researchers over the years have suggested to me dimensions that they thought should be added. As far as my own thinking is concerned, I have a suggestion for one additional dimension, which I discuss later in this article.

Empirical evidence for the event-indexing model

There is empirical evidence for the event-indexing model. It can be found in a number of articles that I was a coauthor of, as well as in many articles by other researchers over the years up to the present time. For example, there are various articles examining the continuity and overlap hypotheses on single dimensions. Much of this research predated Zwaan and Radvansky (1998) and is reviewed there. In fact, the 1998 article was an attempt to integrate these somewhat disparate research efforts in the form of a single model from the perspective that model studies on individual dimensions are treated as special cases of event indexing. At the same time, there was little research on the temporal dimension. This prompted our early efforts to address this issue (e.g., Zwaan, Madden & Whitten, 2000; Radvansky et al., 1998; Zwaan, 1996). More directly pertinent to the event-indexing model were studies that investigated multiple dimensions simultaneously in an attempt to test the continuity and the overlap hypotheses (Therriault et al., 2006; Zwaan, Magliano, & Graesser, 1995; Zwaan, Langston, and Graesser, 1995). These studies yielded empirical support for both the continuity and the overlap hypothesis.

Current status of the event-indexing model

Nearly three decades after its introduction to the literature, the event-indexing model continues to be a subject of research and finds utility in research on event segmentation and cognition (e.g., Wang et al., 2023). This sustained relevance has even prompted extensions of the model, broadening its application beyond its initial focus on language processing. Researchers have extended this framework into various domains, including films (Magliano et al., 2001), real-world activities (Radvansky & Zacks, 2014), and comics (Cohn & Magliano, 2020).

Mental simulations

The next phase in my thinking about comprehension was prompted by the question of what is “in” an event node. It is all well and good to study how event nodes are connected in a network, but if the event nodes are merely abstract labels the entire network is meaningless. This pertinent issue was brought to my attention by a well-respected researcher in a review of one my manuscripts about the event-indexing model. I do not recall whether the manuscript in question was ever published or whether it was relegated to the file-drawer, although I suspect the latter was the case. I do recall, however, that the reviewer’s comment had a major impact on me. My initial response was “Why me?” as I felt the same criticism could have been leveled at pretty much any other researcher in the field of discourse processing or psycholinguistics. Also, I was still relatively junior at the time. However, when I put self-pity aside and gave the issue some more thought, I realized the criticism was justified.

The symbol-grounding problem

The issue that was pointed out to me in the review is the symbol-grounding problem (Harnad, 1990). Briefly, the symbol-grounding problem occurs for any system of meaning that relies on abstract symbols, such as words, the predicates and arguments in propositions, or the nodes in a semantic network or in the event-indexing model. How do these symbols acquire meaning if they are not connected to real-world entities or concepts? The symbols we use in comprehension and reasoning must in some way be connected with perception and action. They need to make a connection with how we perceive and interact with the world or else they are just free floating and essentially meaningless. This is the core of the symbol-grounding problem. Thus, as the reviewer was at pains to point out, the nodes in the event-indexing model are not grounded in perception and action, which renders them meaningless according to the symbol-grounding view.

To address the lack of grounding in models of comprehension, my students and I started thinking of ways to address the grounding problem in the domain of language comprehension. Very naïvely, I saw this at the time as a matter of reculer pour mieux sauter, as the French call it (taking a step back to make a longer jump). The idea was to invest time in tackling the grounding problem, which necessitated a focus on smaller linguistic units, namely, sentences, typically single-clause sentences. Clauses, being inherently tied to individual events, would enable us to build momentum for a more substantial advancement into the domain of discourse processing. The envisioned step back has taken a great deal longer than I had anticipated.

The immersed-experiencer model

The immersed-experiencer model (Zwaan, 2004) was an attempt to improve on the event-indexing model in that it addressed the symbol-grounding problem. This model was heavily influenced by research on cognitive grammar (e.g., Langacker, 1987, 2001), perceptual symbol systems (Barsalou, 1999), and cognitive neuroscience (Pulvermüller, 1999). The foundational idea, as with the event-indexing model, is that event representations are construed based on background knowledge and linguistic cues. By the time this article was written, this view was already less controversial than it had been in the decade before

Each event is captured in an attentional frame, a notion that was based on the work on cognitive grammar (Langacker, 1987, 2001). The transition between attentional frames occurs along the same situational dimensions as proposed by the event-indexing model (time, location, entity, goal, causation) but adds the dimension of perspective. I mentioned earlier that I would return to the issue of the number of dimensions. If another dimension is to be added, my proposal would be to add perspective.

Consider the sentence “The mausoleum that enshrines the tsar overlooks the square” (taken from Kintsch, 1998). Consider next the sentence “The mausoleum that enshrines the tsar has a marble floor.” There are two attentional frames in each of these two sentences. The first frame is the same across the two sentences: the museum enshrines the tsar. But the second frame is different across the two sentences. The mausoleum overlooks the square in the first sentence, while the mausoleum has a marble floor in the second sentence. This difference is relevant in terms of perspectival continuity, as I argued in my 2004 article. In the first sentence, there is a break in perspectival continuity. The perspective shifts from inside the mausoleum to outside the mausoleum, the square. In contrast, in the second sentence, the perspective remains within the mausoleum. In terms of the continuity and overlap hypotheses, the immersed experiencer model would predict that it should be easier to integrate the two construals in the second sentence than in the first. As a result, their bond in long-term memory should be stronger.

Current status of the immersed-experiencer framework

Looking back, I am forced to conclude that I have given this article short shrift over the years. One reason is that I had just started working on the article when 9/11 occurred. Of course, this was a very disturbing event for everyone, myself included. This event may have given the paper a negative connotation in my mind, being representative of a period I was reluctant to revisit. It is possible that this is part of the reason why I had receded into the background of my thinking in subsequent years. However, a more important reason is probably a more practical one. The main impetus for the development of the immersed-experience model was the need to address symbol-grounding problem. This required a focus on the contents of event representations rather than the connections among them, which had been the focus previously. Although cognitive grammar and perceptual symbol systems held promise with regard to addressing the symbol-grounding problem, there was little empirical evidence for them. In fact, I remember discussing Barsalou (1999) with my graduate students. We were wondering whether an empirical case could be made for perceptual symbols in discourse comprehension. Our initial ideas led to a commentary on the Barsalou article (Zwaan et al., 1999). Subsequent to this, we were on a path to answer our question. This required a lot of attention to experimentation, which went at the expense of the event-connection component of the immersed-experiencer model.

I reread the 2004 article during the preparation for the current article; it was the first time I had done so since I wrote it more than twenty years ago. During my reacquaintance with the article, I noticed that it contains quite a few ideas that have as yet to be tested. Most notably perhaps, the notion of perspectival continuity has not been tested (not by me, nor by anyone else to my knowledge). As it stands, I can only claim that there are arguments to be made for including perspective as the sixth situational dimension besides time, location, causation, motivation, and entity, while acknowledging that empirical evidence for this claim is lacking at the moment.

Empirical evidence

My coauthors and I have, however, collected a great deal of evidence that addresses the symbol-grounding problem. I briefly discuss three lines of research here. One line focuses on how certain linguistic constructions give rise to specific mental representations. We have specifically examined the role of grammatical aspect in the mental simulations of events (e.g., Eerland et al., 2017; Liao et al., 2023; Madden & Zwaan, 2003; Zwaan, 2021). Results indicate that conveying events as ongoing (e.g., “He was walking to the church.”) versus completed (e.g., “He walked to the church”) impacts mental simulations in a variety of ways, for example, with regard to where comprehenders mentally locate the target entity (Liao et al., 2021; Madden & Zwaan, 2003) and how many iterations of an action people perceive (compare “John was hitting Jack” to “John hit Jack”; Eerland et al., 2017). In this line of research we have not only looked at language comprehension, but also at production. Different visual stimuli give rise to different aspectual constructions in a manner that mirrors the comprehension data (Liao et al, 2021, 2023).

The second line of research has examined whether particular features of target entities are mentally simulated, such as their shape (e.g., Zwaan, Stanfield, & Yaxley, 2002; Zwaan & Pecher, 2012), orientation (Stanfield & Zwaan, 2001; Zwaan & Pecher, 2012), and color (Hoeben-Mannaert, Dijkstra, & R. A. Zwaan, 2021; Zwaan & Pecher, 2012). This work was heavily influenced by perceptual-symbol theory (Barsalou, 1999) and cognitive grammar (Langacker, 1987). The concept of mental simulation aligns with that of situation models (Zwaan, 2016). To put it simply, mental simulations offer a more intricate, perceptually vivid representation compared to the conventional conception of situation models, in which, for example, protagonists might be represented by tokens, as we had argued earlier (Zwaan & Radvansky, 1998). Consider the sentence, “The ranger saw the eagle in the sky.” According to the framework of mental simulation, the reader envisions the unmentioned detail that the eagle has its wings outstretched (Zwaan, Stanfield, & Yaxley, 2002). Similar results have emerged concerning aspects like orientation and color (see Zwaan & Pecher, 2012 for replications of the original findings). In contrast, in the event-indexing model, these perceptual aspects of the situation would not be represented.

The third line of research can be viewed as a combination of the first and second line. In this series of studies, we examined whether comprehenders represent shape changes (Kang et al., 2020). Consider these two sentences: “The woman bought the ice cream” and “The woman dropped the ice cream.” The first sentence describes the object in its original state, an ice cream as we imagine it to be. The second sentence, however, describes the ice cream cone in an altered state. For example, its orientation is changed and also its shape.

When the pictures of objects are presented in isolation, participants respond faster when the object in its original state than when it is in its altered state. But this pattern changes when a linguistic context is introduced. Participants responded faster to the modified state of the object when it matched the situation conveyed by the sentence than when it mismatched. For example, people responded faster to a picture of an ice cream in its original state than in its altered state (upside down) but this pattern reversed when the picture was presented after the sentence “The woman dropped the ice cream.”

Interestingly, this effect occurred for past-tense sentences but not future-tense sentences (e.g., “The woman will drop the ice cream”). This makes sense from the perspective that, in the current situation model, the ice cream is still in its original state. It has not yet happened in the story that the ice cream has been dropped. These findings show the importance of taking into account the dynamics of event representation in language comprehension, specifically the interaction between semantic object knowledge and the episodic knowledge introduced by the sentential context (the current situation model, as per Zwaan & Radvansky, 1998).

Despite my neglect of the 2004 article, I was still interested in how mental simulations evolve from sentence to sentence. My student Lara Hoeben-Mannaert (Hoeben Mannaert et al., 2019; Zwaan, 2022) has taken a first stab at investigating how mental simulations carry over from clause to clause. In one series of experiments (Hoeben-Mannaert, Dijkstra, & Zwaan, 2019), she presented participants with sentences such as “The boy rode on the red bicycle to the station. At the station he stepped off of his bicycle.” The question of interest here is whether the color of the bicycle is carried over from the first to the second sentence. This hypothesis was tested by using a picture-recognition task. The results indicate that the color was still activated in the second sentence. If the second sentence contained the color word, as in “The boy rode on the bicycle to the station. At the station he stepped off of his red bicycle,” responses were not faster when than the color was mentioned in the first sentence. When the narrative focus was away from the bicycle for several sentences and then returned to it, color no longer seemed to be activated, unless it was mentioned explicitly again. These initial results suggest that perceptual features are carried over from clause to clause, provided that the focus remain on the target object.

My current thinking on symbol-grounding in discourse comprehension

My current thinking on the issue of grounding in discourse comprehension is more nuanced than it was at the outset. As I see it now, the most productive way of understanding the relationship between research on event simulation and that on event connection is to consider them as operating at distinct levels of analysis. Research focused on situation models, including investigations inspired by the event-indexing model, primarily involves exploring the interconnections and relationships among event nodes. For this level of analysis, it might be practical to abstract away from the precise content of event representations and from how they flow into one another in mental simulations. This could, for example, be useful if one wanted to examine the coherence of a narrative at the situational level.

Conversely, research on mental simulation, such as that proposed by the immersed-experiencer model, predominantly concerns itself with the contents of event representations and, as such, it has to address the symbol-grounding problem. It requires unpacking what is “in” an event node. The difference between these levels of analysis can be viewed as a matter of zooming in or out. In this sense, the two views are complementary.

Embeddedness in the environment

In Zwaan (2014), I tried to articulate a more nuanced position on symbol grounding in language comprehension. I distinguish communication situations in which perceptual grounding may be more prominent and others in which it might be less so. To put it differently, communication situations differ with regard to the degree to which they are embedded in the environment and this has impact on the extent to which they are grounded in perception and action.

At the maximum level of embeddedness, the communication situation is one in which the discourse is about the very environment that the interlocutors are co-present in. Think of two people sitting on a bench in the zoo, observing an commenting on the monkeys at play. In such a case, visual representations do not have to be derived from long-term memory, given that the referenced entities (monkeys, trees, jungle gyms) are within clear sight.

At a lower level of embededness, which I call projections, a speaker is using part of the surroundings to build a referential world, and guides the listener via language and gestures to projects onto this mental simulations. Consider for example, a tour guide in, say, Pompeï. The guide uses the ruins of the city as a basis on which she invites the visitors to conjure up what the town looked like right before the eruption of Mt. Vesuvius, mentally populating it with senators, merchants, and prostitutes who are all walking its streets and interacting. Projections are a combination of visually accessible information and a verbal stream, often accompanied by deictic gestures. Another example at this level of embeddedness projects into the future rather than the past. Take a soccer coach who is, during the match, pointing out to a player where he needs to make runs to play more effectively. The spatial layout of the pitch and the players on it are perceptually available as a canvas onto which the desired runs can be projected via language and gestures.

At the next level of embeddedness we find narratives of the kind typically studied by discourse psychologists. Such narratives are divorced from the immediate environment. They refer to situations that are not within sight, a feature of language that Hockett (1960) called displacement. Displacement implies that the environment does not contain relevant information about the situation described in the narrative. Comprehenders have to rely on long-term memory representations of settings, entities, and events to populate their mental simulations.

At the lowest level of embeddedness is discourse that might be so abstract so as to render perceptual representations unnecessary or even unhelpful. This might be the case for abstract scientific or philosophical treatises, for example, although one often sees the use metaphors in such contexts.

The 2014 article describes these ideas in more detail. The point I want to take from it here is merely that the degree to which perceptual representations are involved in discourse comprehension varies as a function of the level of embeddedness.

Methodological observations

There is an inherent problem with regard to testing more subtle predictions about comprehension, such as those that can be derived from the immersed-experiencer model or from cognitive linguistics. This problem is caused by the limitations of traditional psycholinguistics-based methods of experimental design. Experiments in this tradition present participants with a large number of stimuli, sentences in psycholinguistics and short narratives in discourse studies. These stimuli appear in different conditions, which means that there are multiple versions of each sentence or narrative. The different versions of a story cannot, of course, be shown to the same participant, so counterbalancing schemes are employed, which require a large number of stimuli. Furthermore, such experiments commonly also include filler texts, which are used to obscure the manipulations used in the experiment. An experiment may contain a number of filler texts that is equal to or sometimes even larger than the number of experimental texts. Researchers are normally not interested in how these large numbers of stimuli are processed.

This approach faces a severe problem. The experimental designs that are employed create a rather unnatural reading situation. Each text is necessarily short – there are a great many to read, after all— and has to meet a number of design demands regarding length and content. This means that there is no opportunity for the authors, usually the experimenters themselves, to develop a narrative world that the reader might become engrossed in, certainly not to the extent that my teenage self was in Middle Earth (or even my 20-something self in the world of mental models, for that matter). Therefore, whatever fictional world experimenters manage to conjure up in the few sentences they have at their disposal has to be quickly discarded by the participant to make room for the next fictional world described in the next stimulus item.

It is not unusual for experiments to have upwards of 40 stimulus texts. The rapid succession of minimalistic fictional worlds is likely not conducive to the kinds of mental simulations envisioned by the immersed-experiencer framework.

To be sure, I am not just criticizing other researchers here, as I have conducted many experiments of this kind myself; I cite several of them in this article. My goal is merely to point out an inherent methodological problem for researchers wanting advance the study of narrative comprehension.

Nevertheless, to drive this methodological point home bit further, compare the predicament of an experimenter having to write dozens of stories that have to fit a specific experimental manipulation to that of, say, the author of a fantasy series. Fantasy authors often have hundreds, and in many cases even thousands, of pages at their disposal to build up their fictional worlds. In fact, the quality of their “world-building” is a dimension on which fantasy and science fiction authors are often being evaluated. A brief internet search will reveal many sites with world-building tips for budding fantasy authors.

Furthermore, fantasy authors often use external (to the written text) devices to enhance the experience of their fictional worlds. Their books commonly use maps to provide additional information on the spatial layout of the fictional cities, countries, and continents that the narration is about. Often, even a glossary is provide in support of further immersing the reader in the fictional world.

One mechanism for world building in novels is that the same entity or setting can be described multiple times from multiple perspectives at multiple times. It is at least an empirical possibility that people gradually build up a perceptual representation and that it becomes stronger each time the target entity or setting is mentioned. Take, for example, the One Ring in Lord of the Rings. Given its central importance to the story, it is described multiple times. The reader learns not just about its appearance but also about the effect it has on the characters involved. Here are some examples from the first part of the trilogy, The Fellowship of the Ring:

1. “It was round and plain, with no carving or inscription, only a tiny golden dot on its surface that could be seen when held up close to the eye. It was a plain gold band, but it seemed to have a life of its own.”

2. “There it was: round and unadorned, as it had always been, though it had been lost for many years. It was as if it had been waiting for him.”

3. “He held out the Ring. In the torchlight it gleamed, this ring that he had bought many years ago.”

Each of these descriptions is somewhat different than the others and they become progressively less detailed. Nevertheless, the Ring’s visual aspect is evoked each time. It is plausible that such descriptions have an additive effect, something which is yet to be investigated, as far as I can tell, although the study by Hoeben Mannaert et al. (2021) is a start.

Besides the fact that experimenters are more interested in empirical design than in entertaining the reader, they may also lack the skills to do so as they are not professional fiction writers. Even if they had the skills, the experimental designs prevent them from using them. It would seem, then, that several factors conspire against participants being engrossed in fictional worlds in psychology experiments on discourse.

A notion that is relevant here comes from the important work of Richard Gerrig, my predecessor as a Distinguished Scientific Career winner, namely the notion of transportation (e.g., Gerrig, 2023). Fantasy authors use everything they have in their writerly toolboxes to transport their readers to and set them on a journey in the narrative worlds of their invention. Experimenters typically are not in possession of such toolboxes and are also highly constrained by the rigor of experimental design.

As an aside, I should point out that I am focusing on fantasy novels here for two reasons. First, it provides thematic continuity to this article. Second, fantasy is probably the genre that makes the most elaborate attempts to evoke fictional worlds. Nevertheless, it is easy to find world building in other genres as well. Examples that come to mind are historical novels, biographies, police procedurals, and some literary novels. To provide another anecdote, I still vividly the feeling of immersion in Cold War Dresden I experienced when I read Uwe Tellkamp’s Der Turm some ten years ago. In fact, it is the most memorable fictional world of all the ones I have been exposed to in a long time. The general point there is the lack of world building in experimental narratives makes them stand out from narratives people read for pleasure.

I want to bring up one additional problem with using traditional psycholinguistic experimental designs. It has proven linguistically impossible to create narratives in which discontinuities on each of the five original situational dimensions are experimentally manipulated and counterbalanced across narratives. In Therriault et al. (2006), we managed to accomplish this for the three dimensions of time, location, and entity. The results provided converging evidence with our earlier results (Zwaan, Langston, & Graesser, 1995; Zwaan, Magliano, & Graesser, 1995), but designing the studies also showed us that experimentally manipulating more than three dimensions would be next to impossible. Moreover, even if it were possible to construct reasonably naturalistic narratives that systematically manipulate the presence or absence of situational continuity on six dimensions, counterbalancing and having a sufficient number observations per cell for statistical analyses would require an impossibly large number of stimulus texts, which would exacerbate the problem of rapidly changing minimalistic fictional worlds that I touched upon earlier.

My coauthors and I recognized this challenge right from the outset. To tackle it, we opted at the time for an approach involving the use of naturalistic stories and the application of multiple-regression techniques in our research, as documented in our earlier work (Zwaan, Magliano, and Graesser, 1995). With today’s collaborative and technological advancements, we have the opportunity to significantly enhance this approach by harnessing the power of team science, online participant databases, real-time measurements, and cutting-edge data science techniques. For example, a team of researchers could join efforts to test large numbers of participants, using a set of naturalistic texts. Real-time measurements, such as reading times, and even eye fixations, can be obtained in online or lab-based experiments. Large-language models could be used to code the texts into meaningful structures relevant to the hypotheses of interest. Finally, techniques such as multilevel generalized linear models can be used to analyze the data. I am optimistic that in the years to come, we will make substantial strides in unraveling the intricate processes through which language prompts comprehenders to construct fictional worlds.

Conspiracy narratives

Several years ago, I became concerned – more concerned than I already was, to be precise – about how political discourse was evolving, both in the Netherlands and abroad. The rise of certain political parties seemed to be driven by irrational thinking about secret groups controlling the lives of many. I started reading some of the literature on conspiracy thinking. There were many useful insights to be had from my readings, but it occurred to me that most of the research was conducted by social or political psychologists. I felt that cognitive psychologists and in particular discourse psychologists might be able to provide useful insights.

At first sight, the shift from discourse comprehension to conspiracy thinking might seem like a radical one, but in actuality, it is not. There is quite a bit of overlap between discourse comprehension and conspiracy thinking, as I point out in a recent article (Zwaan, 2022). Here are the main points.

First of all, conspiracy thinking can be viewed as a, highly flawed, attempt to understand the world. In fact, forming a stable coherent mental representation of the world is one of the key motivating factors for many of us but, as it turns out, particularly for conspiracy thinkers (Bowes et al., 2023; Douglas et al., 2017). Conspiracy thinkers form narratives to understand the world. A central role in these narratives is typically reserved for a cabal of secretive and malevolent individuals. Often, who these individuals exactly are is not revealed. They could be referred to as “The Elites,” “The Deep State,” “The Illuminati,” or even a race of malevolent shapeshifting reptiles. The putative malicious intentions harbored by these groups form a key mechanism in tying events together in a coherent conspiratorial narrative. Recent research suggests that this explanatory coherence is what makes these narratives convincing to conspiracy thinkers, even though the individual parts of the narrative are improbable (Fernbach & Bogard, 2023). To illustrate this point, Flat Earthers appear to be convinced by the overall coherence of their narrative. Among other things, this narrative states that:

1. the Earth is flat and situated below a dome;

2. the Arctic is a the center of our disk;

3. Antarctica is a wall of ice that surrounds the Earth’s surface;

4. Antarctica cannot be visited;

5. the sun and moon are small lights that are close to Earth;

6. the moon landings are fake;

7. there is no space program;

8. NASA and many others (scientists, governments, and journalists in many different countries) are conspiring to make us believe the earth is a sphere;

9. all scientific papers about space are fabrications.

Taken by themselves, these elements of the Flat Earth narrative are highly implausible and their combination is therefore necessarily even more so. But this is of no concern to Flat Earthers. They are convinced by the coherence of the narrative. If the Earth is covered by a dome, it makes perfect sense that you cannot travel to the moon. So this must mean that the moon landings are fake and that astronauts are merely actors playing astronauts. If people could travel to Antarctica, they would be able to see for themselves that it is a wall of ice. Then they would understand that the Earth is flat. No wonder that no one is allowed to visit Antarctica. And so on. One might speculate that creating more awareness that narrative coherence does not equal truth would be a worthwhile enterprise.

Goals and plans in conspiracy thinking

There is yet another noteworthy connection between conspiracy thinking and narrative comprehension. Conspiracy narratives are, well, narratives. They revolve around a group with malevolent intentions. These putative intentions for these putative groups can be used to “connect the dots.” In conspiracy narratives, the goal of the cabal is to exert total control over the lives of ordinary people. To achieve this, they have to conspire, manipulate, deceive, and fabricate.

Discourse researchers have long argued and shown that narratives revolve around the plans and goals of protagonists (e.g., Graesser et al., 1994; Schank & Abelson, 1977). In narratives – from the everyday stories we tell each other to literary novels of the highest caliber – protagonist intentions are often not stated explicitly but have to be inferred by comprehenders, who do this routinely (e.g., Graesser et al., 1994). Conspiracy thinkers make abundant use of this coherence-generation mechanism to produce their narratives, connecting the dots by explicitly linking events via inferred goals (Zwaan, 2021) to achieve explanatory coherence (Fernbach & Bogard, 2023). This is why I argue that it might be useful to view conspiracy thinking as a form of situation-model construction.

At this early stage of my study of conspiracy thinking I have adopted more of an event-connecting perspective, as described in the first section of this article, than an event-simulation perspective, as described in the second section. I have considered in greater detail how ideas in conspiracy narratives are connected than how they are grounded in perception and action. However, this is not to say that an event-simulation perspective might not be fruitful – on the contrary. Consider again the Flat-Earthers. It would be interesting to examine what kinds of mental simulations they perform when reading “The sun is setting” of “The ship sailed around Antarctica.” Do they simulate these events like the rest of us, despite their belief in a flat earth, or have they fully internalized the Flat-Earth doctrine of the sun being a small moving light that is very close to the Earth and Antarctica being a wall of ice? These are intriguing questions for future research.

Conclusion

In this article, I have given an account of my lifelong fascination with stories and their comprehension, tracing the evolution of my research ideas and interests. Central to my thinking over the years has been the notion that narrative comprehension is a manifestation of event comprehension, guided by the medium of language. Within this broad framework, I have outlined three key stages in my ongoing research journey.

I began by exploring the process of narrative comprehension, which involves integrating event representations into a coherent narrative structure. I paid particular attention to the event-indexing model, a comprehensive framework that identifies five essential dimensions for integrating events: time, location, entity, motivation, and physical causation. I also discussed empirical evidence supporting both the continuity hypothesis and the overlap hypothesis, both of which can be derived from the event-indexing model.

Subsequently, my focus shifted to the mental simulation of events, building upon the event-indexing model with the immersed experiencer model. This more recent framework aimed to succeed the event-indexing model and was targeted at the contents of an event node.

Lastly, I offered a brief overview my current research interests: the intersection of conspiracy thinking with situation-model construction. I outlined some of the ways in which narrative comprehension and conspiracy thinking overlap, shedding light on intriguing research avenues.

Throughout this article, I have drawn connections between these diverse areas of research and shared my current insights regarding the uncharted territories within each domain.

Although, due to its special nature, this article is predominantly retrospective, I wish to conclude on a forward-looking note. My hope is that this work will inspire upcoming generations of discourse psychologists to continue their systematic investigations of discourse comprehension and to cultivate a research interest in the study of conspiracy thinking, thus contributing to the continued advancement of our field

Acknowledgments

I am deeply honored to be the 2023 recipient of the Distinguished Scientific Contributions Award from the Society for Text and Discourse. I thank my co-authors, and in particular Art Graesser, Joe Magliano, and Gabe Radvansky, for their many contributions to the work I describe here. I am grateful to the Society and its senior members for their support throughout the years and for this recognition.

I thank Tobias Richter, Mark Stadler, and an anonymous reviewer for helpful feedback on an earlier version of this article.

Disclosure statement

No potential conflict of interest was reported by the author(s).