The Causal Chain Model (Black & Bower, 1980; Omanson, 1982; Schank & Abelson, 1977) proposed that the causal structure of a narrative can be described as a chain of events, such as attempts to obtain goals, actions and changes of states as a consequence of these actions, that leads from the beginning to the end of the text. This model was inspired by Story Grammars (e.g., Mandler & Johnson, 1977; Stein & Glenn, 1979), which capture how different types of events tend to be ordered in narratives.

This chain begins with statements that introduce the protagonists and their goals. Once they are established, these goals lead to attempts to attain them, which lead to results. Events that have at least one cause and one consequence are part of the causal chain. Events that do not have at least one cause and one consequence are dead ends. These tend to involve actions with no consequences. The chain ends when the protagonist’s goal is attained or fails. Table 1 presents a sample story. Figure 1 presents its causal chain representation:

It can be observed that statement 4 is in the causal chain, given that it has a cause (statement 3) and a consequence (statement 5). Statement 9 represents a dead-end, given that it has a cause (statement 8) but not a consequence.

Empirical studies have found that statements that are on the causal chain tend to be recalled more frequently (Black & Bower, 1980; Trabasso, Secco, & van den Broek, 1984; Trabasso & van den Broek, 1985), more often included in summarization protocols (Trabasso & van den Broek, 1985), and judged to be more important (Trabasso & Sperry, 1985; Trabasso & van den Broek, 1985) than statements that are dead-ends by college students.

Another set of studies found that the same events are recalled more frequently by college students when they are part of the causal chain than when they are dead-ends (Omanson, 1982).

The tools of the model have also been applied to the study of the comprehension of children. Results of these studies suggest that elementary school children recall statements that are part of the causal chain more frequently than statements that are dead-ends for age-appropriate stories (Goldman & Varnhagen, 1986; Stein & Glenn, 1979; Trabasso et al., 1984; Wolman, 1991).

To summarize, this model highlights the importance of the study of the role of the establishment of causal connections in discourse comprehension. Yet, it has a number of limitations. First, it provides an intuitive definition of causality. That is, it does not provide explicit criteria to identify causal connections among statements. A second limitation is that each statement in the causal chain can have at most one cause and one consequence (van den Broek, 1990, 1994).

The Causal Network Model (Trabasso & Sperry, 1985; Trabasso & van den Broek, 1985) overcomes the limitations of the Causal Chain Model. It provides explicit criteria to establish whether a causal connection can be inferred between two statements, and highlights that statements can have multiple causes or consequences.

According to the model, there are three criteria for identifying causal connections. These are based on philosophical (Hume 1739/1964; Mackie, 1980) and legal theories of causality (Hart & Honoré, 1985). The temporal priority criterion suggests that a cause never occurs after the consequence. The operativity criterion suggests that the cause needs to be active when the consequence occurs (for example, goals are active when a protagonist attempts to attain them). According to the necessity criterion, the cause needs to be necessary for the consequence to occur. That is, one must be able to state that if the event that is considered the cause had not happened, then the event that represents the consequence would not have happened). For example, in The call of the wild (London, 1903), the author describes the protagonist’s experiences:

In this case, it can be proposed that statement 2 causes statement 1, because the protagonist’s “being flung into the heart of things primordial” occurs prior to his “every hour being filled with shock and surprise”, is in operation when it does, and is necessary for it to happen. The model also proposes that if event A is necessary for B, and event B is necessary for event C, then event A is necessary for event C, consistent with the notion of transitivity (Lewis, 1973).

Four types of causal connections are identified by the model: motivation (which connects a a goal and an attempt to attain it), psychological causation (which connects an event and the internal reaction that it generates), physical causality (which connects and event and the changes in the physical state of people or objects that it generates), and possibilitation (which connects events that are necessary but not sufficient for other events to occur).

The model also highligths that single statements can have multiple causes or consequences. As a result, it allows researchers to examine the role that the total number of causal connections a statement has, or its causal connectivity plays in comprehension.

Once the causal connections among the statements are identified, they are depicted in a causal network representation. Figure 1 presents the network for the story in Table 1, with nodes indicating the statements and lines indicating the causal relations.

Trabasso, van den Broek, and Suh (1989) examined whether the model’s criteria are consistent with readers’ notions of causality. With this aim, they asked a group of college students to read a series of stories, and to estimate the strength of the causal connections between some pairs of statements. Results indicated that the judged causal relation was high when all criteria were met, and it decreased when one of them was not (especially, the temporal priority criterion). Also, they judged two statements that were connected through a series of causal connections to be more related than two unrelated statements, supporting the notion of transitivity.

Another set of studies examined the role of causal connnectivity in the construction of the product of comprehension. Their results suggest that events with more connections are rated as more important (Trabasso & Sperry, 1985; Trabasso & van den Broek, 1985), included more often in summaries (van den Broek & Trabasso, 1986), and recalled more often and more quickly (Goldman &Vernhagen, 1986; O’Brien & Myers 1987; Radvansky, Tamplin, Armendarez, & Thompson, 2014) than are events with fewer connections.

The tools of the model have been applied to the study of the development of comprehension. Results of these investigations suggest that children establish causal connections at a young age, before formal reading education starts (Kendeou, Bohn-Gettler, White, & van den Broek, 2008; Kendeou et al., 2014; McMaster et al., 2014; van den Broek, 1989; van den Broek & Helder, 2017). For example, 4-year-old children tend to remember statements that have a high number of causal connections more often than those that have a low number of them (Kendeou, van den Broek, White, & Lynch, 2009). At the same time, older children are better able to establish causal connections among statements than younger children (Beker, Jolles, & van den Broek, 2017; Lynch et al., 2008; Pavias, van den Broek, Hickendorff, Beker, & van Leijenhorst, 2016; van den Broek, 1989, 1997; van den Broek, Helder, & van Leijenhorst, 2013; van den Broek, Lorch, & Thurlow, 1996). For example,7-year-old children establish a greater number of these connections than 4-year-old children (Thompson & Myers, 1985).

Considering these results, researchers have developed interventions that aim to focus the attention of these young comprehenders on the most causally connected events. These involve the presentation of audiovisual materials (such as drawings and televised stories), and the generation of causal questions that promote the monitoring of causal breaks (sections of the material where the comprehender needs to establish causal connections). Empirical findings suggest that these interventions facilitate comprehension (McMaster et al. 2012; Mokhtari & Reichard, 2002). Other studies have suggested that the establishment of a high number of causal connections also facilitates the comprehension of students with learning disabilities (Espin, Cevasco, van den Broek, Baker, & Gersten, 2007; Kendeou et al., 2014; McMaster et al., 2012).

Another line of research has examined the effect of the implementation of text revision procedures in the promotion of the comprehension of expository texts by highschool and college students. These procedures include the Causal-Temporal Method (which can be applied to History texts; Linderholm et al., 2000) and the Increasing Coherence Relations Procedure (which can be applied to any expository text; Vidal-Abarca et al., 2002). These procedures involve adding new statements to original texts, which increase the average number of causal connections, and allow for the integration of isolated and distant statements. Results of studies that have examined the effect of these revisions suggest that they promote question-answering and recall by highschool and college students (Barreyro, Molinari, Bechis, & Cevasco, 2012; Cevasco & de Simone, 2016; Linderholm, Virtue, Tzeng, & van den Broek, 2004; Vidal-Abarca, Martínez, & Gilabert, 2000; Vidal-Abarca et al., 2002).

A third line of research has examined the role of the causal connectivity of the statements in the comprehension of narrative and expository spoken discourse. Examining the comprehension of discourse presented in this modality is important, given that most studies in comprehension have focused on written discourse (Cevasco & van den Broek, 2013, 2017; Ferreira & Anes, 1994; Fraundorf & Watson, 2011). Results of these studies indicate that spoken statements that have a high number of causal connections facilitate recall and question-answering by highschool and college students (Cevasco, 2010; Cevasco & Muller, 2009; Cevasco, Muller, & Bermejo, 2017, 2018; Cevasco & van den Broek, 2008; Gaviria & Cevasco, 2012). These findings suggest that it could be useful for instructors to establish a high number of these connections among statements as they teach, in order to promote student learning.

One limitation of this model is that it focuses on the product of reading comprehension, but not on the processes that take place during the discourse comprehension. This is important because, even though it contributes to the study of how the outcome of comprehension is constructed, it does not explore the cognitive activities involved in the establishment of causal connections during discourse comprehension.

The Causal Inference Maker Model (van den Broek, 1990, 1994) examines the cognitive processes that take place during discourse comprehension. It integrates the central notions of the Causal Network Model with research concerning the role of the limitations in the capacity of working memory and attentional resources to model the actual coherence building processes that take place during comprehension.

According to the model, as the reader procedes through the text, he or she attempts to obtain adequate causal justification for each statement. The first statement to be considered is the immediately preceding one. If it fulfills the criteria, the reader generates a connecting inference. For example (van den Broek, 1994):

The man accidentaly dropped the crystal glass.

The glass broke.

INFERENCE: The glass broke because the man dropped it.

If the prior statement does not fulfill the criteria, then a coherence break occurs. This break results in a search for information, and there are two potential sources. One of them involves the reactivation of one or more prior statements. In this case, the comprehender generates a reinstatement inference. For example:

John’s brother beat him.

John told his parents.

The following day, John’s body was covered with bruises.

INFERENCE: John’s body was covered with bruises, because his brother beat him.

A second source of information involves the generation of elaborative inferences. These inferences involve the activation of the reader’s background knowledge. For example (Singer, Halldorson, Lear, & Andrusiak, 1992):

Dorothy poured water on the fire.

The fire went out.

INFERENCE: Water extinguishes fire.

Connecting, reinstantement and elaborative are backward inferences, given that they connect the statement that the comprehender is processing to prior statements. The model also proposes that readers generate forward inferences. These inferences create expectations about what will happen in the text. They are not necessary for comprehension, but they can facilitate the processing of upcoming events. The model distinguishes two types: predictive and anticipation of the future importance of prior events. Predictive inferences involve the activation of background knowledge in order to anticipate upcoming events. For example (McKoon & Ratcliff, 1989):

While shooting a film, the actress accidentally fell out the 14^th floor window.

INFERENCE: The actress died.

The anticipation of the future importance of prior events inferences involve the anticipation that a current statement will play a role in upcoming parts of the text, which leads the comprehender to maitain this information actíve. For example, the comprehender could maintain the goals of the protagonist active, because he or she anticipates that they will lead to attempts to attain them.

Bloom, Fletcher, van den Broek, Reitz, and Shapiro (1990) examined the generation of connecting and reinstatement inferences. They asked a group of college students to read a set of stories online. They found that reading times increased when the statement was causally connected to prior statements. The authors proposed that the time increases because the comprehender is generating the inference. Also interested in the generation of these inferences, van den Broek and Lorch (1993) observed that a statement was recognized faster by college students when it was re-presented preceded by an event to which it was causally connected, than when it was preceded by an unrelated statement. These results suggest that the generation of the causal inference facilitates recognition. Similar results were found by Trabasso and Suh (1993), and Dopkins, Klin, and Myers (1993) with a recognition task and by more recent studies (Barreyro et al., 2012; Bohn-Gettler & Rapp, 2011; Pérez, Paoleri, Macizo, & Bajo, 2014; van den Broek, Rohleder, & Narvaez, 1996).

The generation of elaborative inferences was examined by Singer et al. (1992). They asked college students to read pairs of statements, which were causally connected, and were then asked questions about background knowledge. For example:

Dorothy poured water on the fire.

The fire went out.

QUESTION: Does water extinguish fire?

Results indicated that students answered more quickly when the questions followed sentence pairs that required the activation of this information than when they read pairs did not. Similar results were found by other question-answering studies (Singer et al., 1992; Singer, Revlin, & Halldorson, 1990), and by more recent studies (Barreyro et al., 2012; Bohn-Gettler & Rapp, 2011; Pérez et al., 2014; van den Broek et al., 1996).

Other studies have examined the generation of predictive inferences. Their results suggest that the causal sufficiency of the event has a role in the likelihood that these inferences are generated. For example, if comprehenders are presented with statements such as (McKoon & Ratcliff, 1989):

While shooting a film, the actress accidentally fell out the 1^st floor window.

While shooting a film, the actress accidentally fell out the 14^th floor window.

PREDICTIVE INFERENCE: The actress died.

Results of naming tasks suggest that the inference is more available following statements that provide high causal sufficiency such as 2 than statements that provide weak sufficiency such as 1 (Murray, Klin, & Myers, 1993). Other studies have suggested that individual differences and prior context play an important role in the generation of these inferences (Barreyro et al., 2012; Calvo, 2000; Linderholm, 2002; McNamara & Magliano, 2009; Pérez et al., 2014).

The generation of anticipation of the relevance of prior events inferences was examined by van den Broek, Fletcher, and Marsolek (1989). These authors asked a group of college students to read story stems, and to provide continuation sentences. Results indicated that the continuations tended to be causally connected to the events described in the story stems. These findings suggest that the establishment of causal connections has a strong influence on the expectations of the comprehenders. Duffy (1986) found similar results.

The tools of the model have been applied to the study of the development of comprehension. Results from these studies suggest that older children are able to generate inferences that connect larger parts of the text (such as paragraphs, episodes, etc), and to infer abstract connections (such as between topics, and between story events and the emotions of the characters) to a greater extent than younger children (Diergarten & Nieding, 2015; Kendeou et al., 2009; Mouw, van Leijenhorst, Saab, Danel, & van den Broek, 2017; Rapp, van den Broek, McMAster, Kendeou, & Espin, 2007; van den Broek, 1997; van den Broek, White, Kendeou, & Carlson, 2009)². For example, 7-year-old children generate a higher number of inferences than 4-year-old children (Thompson & Myers, 1985).

The application of the tools of the model has allowed researchers to identify profiles of struggling readers. Results of think aloud and eye-movement studies suggest that they can be grouped in paraphrasers and elaborators (Karlsson et al., 2018; McMaster et al., 2012; Seipel, Carlson & Clinton, 2017). Paraphrasers are readers that tend to paraphrase or repeat the information in the text, but not to generate inferences. Elaborators are readers that tend to generate elaborative inferences, but that tend to be inaccurate or invalid according to the information in the text. The identification of these profiles has allowed researchers to develop interventions to promote the comprehension of these readers. These involve the design and generation of causal questions (such as ‘Why did Pedro decide to get a paper route?’ related to the story in Table 1, which prompt the comprehender to connect the current statement with specific prior statements) and general questions (such as ‘How does this statement relate to earlier parts of the text?’, which prompt comprehenders to make any connections between the current statement and prior statements). Empirical findings suggest that the inclusion of these questions facilitates recall by both groups of readers (McMaster et al., 2012). They also suggest that causal questions facilitate the recall of elaborators to a greater extent. It is possible that, given that these questions require the comprehender to make specific text-based connections, they facilitate that these readers focus on important information within the text. In turn, the presentation of general questions facilitates the comprehension of paraphrasers to a greater extent. It is possible that, given that these questions prompt the comprehender to make any text-based connection, they help these readers think about the text beyond the current sentence. Research has also proposed other readers’ profiles (such as literal readers, who predominantly repeat the literal text when asked to think aloud; Karlsson et al., 2018; Kraal, Koornneef, Saab, & van den Broek, 2017). Future research will allow us to discover new profiles, and finer distinctions within existing profiles.

Other studies have investigated the generation of causal inferences during the comprehension of spontaneous spoken discourse by college students. Cevasco (2009) examined the role of the presence of discourse markers in the generation of reinstatement inferences during the comprehension of narrative spoken discourse. Results indicated that the presence of the discourse marker facilitated the reactivation of prior causally connected statements. Cevasco and van den Broek (2016) observed that the presence of filled pauses (such as ‘uh’) had a negative effect on the recognition of spoken words, but not on the generation of causal inferences. These results suggest that the generation of these inferences plays a role in the processing of spoken discourse by college students.

A limitation of this model is that it does not examine the interplay between the cognitive processes and the result of comprehension. This is important, because it does not allow us to examine how passive and strategic inferential processes interplay in the construction of a coherent discourse representation.

The Landscape Model (Linderholm et al., 2004; van den Broek, Young, Tzeng, & Linderholm, 1999) is a computational model designed to capture the interplay between the processing of causal connections and the construction of a coherent representation of discourse in memory.

The model proposes that the activation of units (which can be statements, propositions, concepts, etc.) fluctuates as the reader proceeds through the text. In each reading cycle, there are four possible sources of activation: 1) the text that the reader is processing (which has the highest level of activation), 2) the preceding reading cycle, 3) the reinstatement of one or more prior cycles, and 4) the reader’s background knowledge. The reactivation of prior cycles and the activation of the reader’s prior knowledge involve the generation of inferences. The model contemplates, following the Causal Inference Maker model, the generation of connecting, reinstatement and elaborative inferences. Considering these sources of activation, in each reading cycle new statements are activated, and activation values of current statements change. In addition, the co-activation of statements leads to the establishment of connections between them. Through these fluctuating activations, a memory representation of the text gradually and dynamically emerges.

Two types of mechanisms guide access to these sources of informatiton. The first type is cohort activation. The model assumes that when a concept is activated, all other concepts that are activated become associated with it. Thus, each concept connects with other concepts to form a cohort. In turn, when one or more of the individual concepts in a cohort become active, the other concepts are also activated. The amount of activation for the secondarily retrieved propositions is a function of the strength of their relation to the primarily retrieved statement, and the amount of activation of the primarily activated statement. This mechanism is passive and similar to the activation mechanism described by memory-based views such as the Resonance Model (Myers & O’Brien, 1998; O’Brien & Myers, 1999; O’Brien, Rizzella, Albrecht, & Halleran, 1998).

The second mechanism is coherence-based retrieval. This is a strategic mechanism by which information is retrieved with the aim of meeting a reader’s standards of coherence or goals (van den Broek & Helder, 2017). These standards reflect a reader’s knowledge and beliefs about what constitutes good comprehension, as well as his or her specific goals for reading the particular text (e.g., study, entertainment). For narratives, referential and causal standards of coherence are central (Linderholm et al., 2004; van den Broek, Lorch, Linderholm, & Gustafson, 2001). Referential coherence is obtained when the reader is able to identify the reference for the objects, persons, etc, that are part of the sentence that he or she is reading. Causal coherence is obtained when the reader is able to find causal explanation for the event described in the current sentence. The model allows for the implementation of other standards of coherence.

The simulation of comprehension includes three stages. In the first stage, the activation values for each statement are calculated through an analysis of the materials, according to the theoretical model of the researcher. In the second stage, the model processes the input values cycle by cycle and produces an activation matrix that represent the state of working memory after reading each sentence. In the third stage, the model establishes and updates the connections of the emergent network memory representation. The output indicates the strength of the connections among statements at each reading cycle, and the node strength of each statement (that is, their total activation across reading cycles).