Genre Effects on the Generalization Inference a Thesis Submitted to the Graduate School in Partial Fulfillment of the Requirements for the Degree Master of Cognitive and Social Processes by Wesley Stoller

THESIS: Genre Effects on the Generalization Inference STUDENT: Wesley A. Stoller DEGREE: Master of Cognitive and Social Processes COLLEGE: Science and Humanities DATE: July, 2010 PAGES: 50 The constructionist theory has emerged as a leading perspective in the field of reading inferences and makes the assumption that readers cannot generate inferences when text is inconsiderate or lacking coherence. The generalization inference has been documented as allowing the reader to condense multiple, consecutive propositions into a singular macroproposition. Research has shown that the genre of a text can affect the perception and the set of processes used by the reader to comprehend text. In the present study, participants read ten short narratives, eight of which contained generalization inference lexical decision tasks with genre and coherence of text manipulated. Participants were shown to be no more likely to draw the generalization inference from incoherent text when primed by genre, but were shown to be capable of drawing the generalization inference from incoherent text. These results do not support the constructionist hypothesis and suggest that further research is needed. GENRE EFFECTS ON THE GENERALIZATION INFERENCE 3 Genre Effects on the Generalization Inference In our modern society, the ability to read is an early benchmark of progress, and is often a requirement for not only academic (Nippold & Sun, 2008; Rytkönen, Aunola, & Nurmi, 2005) but social success (Tournaki, 2003). The underlying psychological mechanisms that allow the reader to integrate and comprehend text has been studied for many years (e.g., Kintsch & van Dijk, 1978; Mandler, 1980; Mani & Johnson-Laird, 1982; van den Broek, 1990). Of particular interest has been inference generation both during (Graesser, Singer, & Trabasso, 1994) and after reading (McKoon & Ratcliff, 1992). An inference occurs when the reader bridges the gap between the explicitly stated text and the underlying meaning or message of the text. Inferences have been studied as a major component of reading because often word meanings are ambiguous and sentence relationships are arbitrary without context. Despite these challenges successful readers quickly extrapolate the author‟s intended meanings through the use of inferences, often with little cognitive effort required (McKoon & Ratcliff, 1992). Inferences in Text Comprehension The context of the text can be viewed as a contributing factor to the processing of textual information. Comprehension of text has been shown to improve drastically when the reader is primed with a title that evokes relevant world knowledge (Bransford & Johnson, 1973 as referenced by Long, Oppy, & Seely, 1994). Past research has assumed that inference generation is highly contingent upon the coherence of the text (Graesser, et al., 1994). However, because readers' ability to integrate text is influenced by the context (Einstein, McDaniel, Owen, & Cote, 1990), inference generation may also be enhanced or diminished by contextual primes. If context can influence the processing of textual GENRE EFFECTS ON THE GENERALIZATION INFERENCE 4 information, then it may be possible for inference generation to occur despite textual incoherence. This study investigated the effect of context, established through text genre, on readers' inference generation. Coherence is an important element in reading comprehension and inference generation. Coherence has been considered the foundation of the text that allows the reader to make inferences (Graesser, McNamara, & Louwerse, 2003). The definition of coherence varies drastically from study to study. One common theme within most definitions of coherences is that coherent text is ordered in a way in which ideas flow fluidly between one another. Past research has shown that coherence is actively maintained through integrating information from: the current text, information earlier in the text, and background knowledge of the reader (van den Broek, Risden, Fletcher, Thurlow, 1996). Kintsch and van Dijk (1978) offered a definition of coherence for text comprehension and discourse that requires sentences and propositions to be connected at a local level (i.e. readers can identify the connections between consecutive ideas) and propositions organized at a macrostructure level, or global level, where readers identify the connections between non-consecutive ideas. Kintsch and van Dijk argued that many inferences may be required by the reader in their effort to establish and maintain local and global coherence. Before inferences can properly be examined they must first be defined. The term inference is used in at least three different contexts: cognitive science, logic, and everyday speech (Kintsch, 1993); which makes establishing an agreed upon definition a difficult task. One definition commonly used in the cognitive sciences is that inferences should be considered reading tools (i.e., cognitive processes) that are conscious and GENRE EFFECTS ON THE GENERALIZATION INFERENCE 5 controlled or automatic and uncontrolled (Kintsch, 1993). More specifically, some inferences are done within the awareness of the reader and can be modified to fit their motivation, and some are outside of the awareness of the reader. Depending upon the nature of the text and the motivation of the reader, inferences are either information generative or integrative. Generative inferences are those that produce new information for the reader. For example, a causal-antecedent inference enables the reader to better understand the text by generating new information regarding how something happened given past information (e.g. After reading, “The vase hit the wall. It cost a lot to replace,” a reader would generate the inference “broke,” which had not been explicitly stated). Integrative inferences do not generate new information; rather, they allow the reader to incorporate information provided by the author. For example, referential inferences are integrative because they allow the reader to understand which character the author is referring to by the given pronoun, but do not require the reader to infer any words that had not been previously stated (e.g. After reading, “Philip was a good looking and intelligent young man. Although he was quite shy he never had trouble making friends,” a reader would generate the inference that the pronoun “he” was in reference to “Philip”). The leading perspective on how readers generate inferences is the constructionist theory (Graesser, et al., 1994; Singer, Graesser, & Trabasso, 1994). The constructionist theory makes two defining assumptions. The first assumption is that the reader's representation of text embodies the goals of the reader. The first assumption presupposes that the motivation of the reader is an important underlying mechanism for the generation of inferences. The second assumption is that both local and global coherence are necessary for comprehension (Graesser, et al., 1994). Similarly to Kintsch and van Dijk GENRE EFFECTS ON THE GENERALIZATION INFERENCE 6 (1978), the constructionists say that local coherence occurs when causal-antecedent relationships are present within the chronology of the text. When the reader is able to integrate all of the sequential prepositions within their‟ short-term memory, the text is considered locally coherent. Global coherence, as described by Graesser, et al. (1994), is the organization of local bits of information into higher-order ideas. A globally coherent narrative would be one that allows the reader to condense multiple ideas into a conceptually broader one that expresses the underlying message behind each smaller idea. By organizing multiple ideas into a more general one, readers are making a specific type of global inference called a generalization inference (van Dijk, 1980). Generalization inferences allow readers to integrate knowledge into easier to remember chunks of information that can be better accessed at a later time. Because of the important role generalization inferences have in the global comprehension of text, they were the focus of the current study. Generalization inferences are made when multiple similar ideas are summed up with one idea that reflects every aspect of those original ideas (van Dijk, 1980). Generalization inferences have been studied as a crucial element of constructing a macrostructure of the text during reading comprehension (Kintsch & van Dijk, 1978). The macrostructure of a text is created in order to condense and organize the many individual detailed propositions of the text, but maintains both the truth and meaning of the text (Kintsch & van Dijk, 1978). Under the constructionist theory, generalization inferences are constructed online under most processing conditions (Graesser, et al., 1994). Unlike other global inferences, generalization inferences do not convey an underlying moral or theme of a narrative; rather they are the combination of multiple GENRE EFFECTS ON THE GENERALIZATION INFERENCE 7 concepts into one encompassing idea. More specifically, generalization inferences are rules for developing macropropositions that through abstraction allow the reader to utilize all propositions of the text while requiring less information. Generalization inferences can thus be seen as an important early step towards the development of a macrostructure of the text. The constructionist theory adopts the perspective that readers are subject to the 'search after meaning' (SAM) principle. The SAM principle causes the reader to make commitments as to the meanings of each concept while it is read, unless the reader is unwilling (e.g., their reading goal is inconsistent with finding meaning), incapable (e.g., lacks necessary background knowledge), or unable (e.g., the text is inconsiderate; the text lacks global coherence and a coherent message; Graesser et al., 1994). The constructionist theory presupposes that referential and causal antecedent inferences are generated to maintain local coherence. Similarly, thematic inferences, a type of global inference related to the generalization inference, are generated to achieve global coherence (Graesser et al., 1994). An alternative perspective, the minimalist theory of inference processing (McKoon, & Ratcliff, 1992), takes a different approach than the constructionist theory. This position assumes that at the most basic level readers strive only to achieve local coherence, resulting in only simple inference generation, such as pronoun resolution; which is similar to the referential inference. From this perspective, global coherence is thought to be achieved by accessing information stored in long-term memory after reading has occurred. Global coherence is achieved after reading unless the reader has a specific, cognitively-taxing goal that requires establishing global inferences during GENRE EFFECTS ON THE GENERALIZATION INFERENCE 8 reading. From the minimalist position, the only inferences generated automatically during reading are those needed to establish local coherence; all other inferences are explicitly done via the motivation of the reader (McKoon & Ratcliff, 1992). The minimalist position reminds us that perhaps not all readers strive to construct a mental model of the text during the actual reading of the text. A mental model can have three levels of representation, including: the surface structure, the textbase and the situation model of the text (Zwaan, 1994). A surface structure representation of the text is the exact structure of the text; specifically, the words and syntax. A textbase is a representation of the gist meaning, outlined by propositions explicitly stated within the text. A situation model of the text is the reality a reader creates, combining the gist of the text with the reader's own past experiences, as well as the reader's concept of their own reality. The minimalist position argues that the situation model is not automatically generated during reading unless the reader is specifically motivated to do so. Because the constructionist approach assumes that inferences are not cognitively taxing, it would seem that readers who automatically generated inferences in order to maintain local and global coherence would create the fullest understanding of a text during their first read, which would imply no need to reread a text to gain additional information. The minimalist approach, on the other hand, assumes that the reader will construct their mental model of the text after information has been integrated into long-term memory, requiring few inferences to maintain local coherence unless otherwise motivated; implying that reading is subject to the limitations of the readers‟ memory. The current study investigated the extent to which readers construct inferences while reading, and whether that construction is influenced by the genre of the text. GENRE EFFECTS ON THE GENERALIZATION INFERENCE 9 Text Genre and Comprehension The perspective of the reader has been examined as a primary factor for inference generation (Just & Carpenter, 1980; Mandler, 1984; Warren, McConnell, & Rayner, 2008; Zwaan, 1994). The question of whether readers‟ expectations regarding textual content influences reading processes has been explored by Zwaan. Specifically, Zwaan (1994) noted that textual variations between genres are vast enough that the reader may use different processes depending upon what genre is presented. For example, in his study, Zwaan presented the same texts as either being from a news story or a literary story and measured which levels of representation were developed (surface, textbase, or situation model). Zwaan found that readers would integrate textual information differently when primed with the genre of the text (news story as opposed to literary story), despite being of the same literature style (discourse narrative). Subjects who were primed to believe they read a news story developed more robust situation models and had faster overall reading times than subjects primed to believe they were reading a literary narrative, which had more accurate memories for surface information. Zwaan concluded that these results indicate that readers allocate different processing resources to text which are dependent upon their expectations from their past experience with other texts like it (i.e., genre). Zwaan also noted that authors typically leave readers in the dark as to what the theme or purpose of the narrative is. This could have resulted in the development of an expectation within the reader that the most important information in a narrative will be at the end, causing readers to integrate information without drawing direct conclusions until the end of the narrative. This would imply that the most efficacious placement of GENRE EFFECTS ON THE GENERALIZATION INFERENCE 10 generalization inferences would be at the end of the narrative. Taking this literature a step further, Zwaan (1994) hypothesized that because entire narratives often have ambiguous meanings, readers may attempt to construct a single textbase representation containing all surface information; that is, a model of all explicitly stated information of the text (similar to the macrostructure). When the text is ambiguous or indeterminate, readers will delay constructing a situation model until enough information is provided (Zwaan, 1994). In the present study, I will expand upon Zwaan‟s investigation of how differences in genre influence readers‟ processing a text, specifically when the text‟s theme is ambiguous. In addition to the reader‟s perspective, the reader‟s motivation is also thought to be an important element in inference generation during reading comprehension (McKoon & Ratcliff, 1992; Graesser, et al., 1994). The motivation of the reader influences the construction of a goal. The constructionists claim that the goal of the reader can either be underdeveloped (e.g., to read in order to be entertained or to extract general ideas), or very specific (e.g., to read in order to determine the need for a major life decision). Genres exert a direct influence upon the perceived goal of the reader (Graesser, et al., 1994). For example, readers of literature will construct the goal of being entertained; conversely, readers of expository texts will construct a goal of expanding knowledge. The constructionists recommend that any study of inference generation consider how readers‟ goals influence their discourse processing. By the constructionists‟ own recommendation, genre should be considered a primary factor influencing the processing of a text but is often ignored, particularly as an independent variable. Expectation-driven processing cues, such as the title or writing style of the text GENRE EFFECTS ON THE GENERALIZATION INFERENCE 11 has been studied as a priming mechanism for reading that causes the reader to process the text in an expectation-consistent manner (material-appropriate processing framework; Einstein, et al., 1990). This material-appropriate processing framework posits that different types of texts (genres) will prime the reader to use different types of processing. When the text‟s genre causes readers to process items more deeply (i.e. make more relational connections), they are more likely to recall textual information and that information is more likely to persist in memory long after reading (Einstein, et al., 1990). Einstein and colleagues (1990) showed that the context of the text is a powerful tool that not only affects the motivation and goal setting of the reader but also the underlying processes that go into comprehending the text. The effects of genre context on reading comprehension have been studied in an eye-tracking design by Warren and colleagues (2008). They found eye-movement disruptions when impossible literary scenarios were presented in a real-world context but not when presented in a fantasy context. The authors noted that fantasy context literature allows for the reader to eliminate a sense of 'unnaturalness' in impossible events after their first fixation. Real-world impossible events (e.g., The cat flew to the moon) lagged in processing times in gaze (of word) and go-past (particle immediately after word) segments of reading, compared to fantasy-impossible times which resulted in processing times similar to real-world and fantasy possible times. These results indicate that readers' processing times of fantasy primed impossible narratives are more similar to real-world primed possible processing times than real-world impossible narratives, suggesting that readers' expectations about the genre of the text influences their processing of the narrative. The preceding literature indicates that readers' processing of text is influenced GENRE EFFECTS ON THE GENERALIZATION INFERENCE 12 by their expectations, based on their concept of the narrative itself (i.e., the genre). The current study will examine how readers' expectations, primed by genre, affect inference generation. The folk tale genre has been studied under the premise that it has a simple structure with explicitly stated action sequences (McDaniel, Hines, Waddill, & Einstein, 1994). McDaniel et al. (1994) examined this idea under the assumption that textual consistency was necessary to construct a causal-antecedent representation of a narrative. McDaniel et al. had participants either use a word-unscrambling mnemonic technique to improve readers' recall of narratives, or provided no such intervention. Additionally, McDaniel et al. primed readers with the genre of folk-tale or with the genre of a news story. Readers who were primed by the folk-tale genre recalled more information from the text, but showed no effects of the mnemonic intervention. The results of McDaniel and colleagues' study indicate that the folk-tale schema influences readers to develop a more complete representation of the content of text based on a causal-antecedent structure, allowing for a greater degree of recall. The McDaniel et al. study provides further evidence that the genre of a text provides the reader with an expectation that cues genre-specific processes within the reader. Although folk tale was used as a priming genre in this context, there is very little literature explicitly describing what researchers believe the folk-tale genre to be. In studies such as the one conducted by McDaniel and colleagues, the definition for the genre was provided by the authors and not based on past research or an established definition. For the purpose of this research I have made the following dichotomous definition of the genres that are of primary interest: a genre in which events are factual and possible (e.g., biography), and a genre in which events can GENRE EFFECTS ON THE GENERALIZATION INFERENCE 13 be non-factual and impossible (e.g., fairy tale). The context that is established by a text‟s genre has been shown to affect the processes utilized by the reader (Zwaan, 1994) and the degree to which incoherent information will be comprehended (Warren, et al., 2008). Although inferences have been studied under the assumption that local and global coherence are prerequisites of reading comprehension (Myers, O'Brien, Albrecht, & Mason, 1994), given the research done on the effect genre has on the processes involved in reading, genre may also influence the readers' ability to draw inferences under conditions in which global coherence is not maintained. The present research investigated whether or not the genre of the text influences whether or not inferences (specifically, generalization inferences) are generated within incoherent text. Hypotheses Generalization inferences have been studied under conditions in which a series of related sentences are presented sequentially (Ritchey, 2001). According to the constructionist theory, when this sequence is broken, maintaining global coherence should no longer be possible, and therefore generalization inferences should no longer be made. When readers are primed by the context of a genre that allows for impossible events, not based on fact (e.g., fairy tales), readers should expect the information presented to be capable of being inconsistent with their concept of reality, allowing them to integrate information regardless of the sequential consistency of that information. This study was designed to manipulate sequential consistency by presenting readers with texts in which a series of sentences are presented in chronological order or non-chronological order. The present research was designed to question the constructionist theory in its GENRE EFFECTS ON THE GENERALIZATION INFERENCE 14 claim that global coherence, and therefore the construction of generalization inferences, cannot be not achieved if the text is inconsistent, choppy or pointless (Graesser et al., 1994). The present research makes the counter claim that readers will draw generalization inferences when they perceive an inconsistent text as belonging to a genre (e.g. fairy tales) that allows for events inconsistent with their concept of reality. I predicted that because readers‟ background knowledge of such a genre is that in which the impossible is possible, they will draw generalization inferences from incoherent text. Similar to their predictions for generalization inferences, the constructionist theory predicts that causal inferences will be generated under most normal processing conditions (Graesser, et al., 1994). The constructionists define causal inferences as a direct linkage between current actions or events and previous contexts. That is, the reader creates a causal chain linking the present textbase information with the immediately preceding information. Previous research has shown that causal inferences are important for comprehension (Graesser, et al., 1994) and are drawn on a regular basis (van den Broek, 1990). As is predicted by the constructionist theory, causal inferences should be generated online during the reading of non-reality based narratives as well as reality based narratives. Because causal inferences have been shown to be drawn so regularly, they were assessed in the present experiment to ensure that readers were drawing inferences and therefore making an effort to comprehend the text. For the present research I used a 2 (factual-possible genre v non-factualimpossible genre) x 2 (coherent vs incoherent) x 2 (target v non-target word) design. The constructionist perspective assumes that readers who view a text as incoherent (i.e., lacking in global coherence) will abandon inference generation (Graesser, et al., 1994). GENRE EFFECTS ON THE GENERALIZATION INFERENCE 15 Half of the experimental texts were designed to be incoherent by presenting sequential events required for readers to draw generalization inferences in a non-sequential order. The coherent condition was the condition in which the sequence of events was presented in order. The dependent measure of this study was the reaction times on lexical decision tasks, which are indicators of inferences generated (Long, et al., 1994; Zwaan, 1994). Additionally, I manipulated whether or not the word used within the lexical decision task was a target word (a word that would be inferred from the text) or a non-target word (a word that would not be inferred from the text). The target word manipulation was conducted in order to measure whether participants were drawing an inference or not. For causal inferences, I predicted an effect for the lexical decision task word condition (i.e., target vs. non-target.). Because causal inferences are an integral part of reading comprehension (McDaniel, et al., 1994), participants should be faster to verify inference-related target words than control words or nonsense words in both genre conditions. I will measure reaction times to these words for causal inferences in order to assess whether or not participants are reading the text and are drawing basic inferences while they read the text. Because the coherence manipulation is implemented only for the four sentences that instantiate the generalization inference, this manipulation should have no effect on causal inferences. For generalization inferences, I predict an interaction of genre and coherence. Participants‟ reaction times for inference-related target words should be equally fast in the non-factual/impossible coherent, non-factual/impossible incoherent and factual/possible coherent conditions, and these times will be faster than reaction times for inference-related target words in the factual/possible incoherent condition. Because the GENRE EFFECTS ON THE GENERALIZATION INFERENCE 16 non-factual/impossible genre suggests that impossible events are possible, I predict readers primed with the non-factual/impossible genre will integrate the four sentences of the generalization section regardless of their coherent (i.e., chronological) or incoherent (i.e., non-chronological) nature and will draw the intended generalization inference, resulting in faster reaction times to verify a word that represents this inference than a neutral word or non-word. Reaction times for inference-related target words in the factual/possible-coherent condition should be as fast as reading times in the other genre conditions because readers should have no difficulty constructing a generalization inference, resulting in faster reaction time to verify a word that represents this inference than a neutral word or non-word. A difference should appear for reading times of inference-related target words in the factual/possible-incoherent condition. Because the nature of a factual/possible genre is one in which readers are primed to believe that all events in the text are truthful and possible, readers encountering impossible (i.e. incoherent, non-chronological) events should abandon attempts at constructing a generalization inference. Because they will not have constructed an inference, they will not have activated the inference-related target word, and will be as slow to verify this word as the control word, which is unrelated to the story. Also, because of a lack of research on the subject, in order to determine which two genres most accurately represent this criterion and which word represents the ideas expressed in the last four sentences of each story (i.e., generalization inference target word), two pilot studies were conducted.