Relevance processes in multiple document comprehension

We introduce the MD-TRACE model (for Multiple-Document Task-based Relevance Assessment and Content Extraction), a descriptive model of the resources and the processes brought to bear in document-based activities. The MD-TRACE model considers the external resources that take part in functional reading activities (beyond text passages), as well as the cognitive resources that these activities call upon. The MD-TRACE model defines 5 core processes: the construction of a task model, the assessment of one's information needs, the selection, processing and integration of document information, the construction of a task product, and the assessment of product quality. These processes are helpful in identifying the many ways in which relevance processes take place during functional reading. We discuss several research questions that are derived from the revised model. Rouet and Britt Relevance in multiple document comprehension 3 In today's information age, readers are faced with textual information that comes from very diverse sources, both in print and online. Printed and electronic texts also tend to play a role in a wide range of everyday situations. For instance, shopping online requires the consumer to read, comprehend, compare and contrast various written product descriptions. Similarly, people use more textual sources when informing their decisions regarding education, jobs opportunities, investment or health, to take just a few examples. These dramatic changes in the forms and uses of text challenge traditional models of reading comprehension and call for a generalized theory of multiple text processing (Britt, Perfetti, Sandak, & Rouet, 1999; Perfetti, Rouet & Britt, 1999; Rouet, 2006). At the heart of this new perspective on text processing rests the notion of relevance, which refers to the extent to which information is consistent with the readers' needs and capacities (McCrudden & Schraw, 2007). The goal of this chapter is to examine the processes related to readers’ assessment of information’s relevance in situations in which readers make use of multiple texts to fulfill complex information needs. After a brief review of the contextual dimensions of text processing, we propose a new framework to describe the resources and processes involved in functional document-based activities. We highlight those phases and steps in the comprehension process in which relevance plays a critical part. Reading as a contextalized activity Research on text processing has initially focused on the impact of text structure on reading and comprehension processes (Cirilo & Foss, 1980; Kintsch, Kozminsky, Streby, McKoon & Keenan, 1975; Kintsch & van Dijk, 1978). Reading time and text recall were the two main indicators of processing and comprehension. These two measures were found to be strongly dependent on the contents and organization of the text. For instance, reading time increased with the number of semantic propositions in a passage (Kintsch et al., 1975) and also as a function of the status of propositions in the propositional hierarchy that makes up the literal meaning of the text (Cirilo & Foss, 1980; Guindon & Kintsch, 1984). Other textual features include the content of titles and initial mentions (Kozminsky, 1977; Kieras, 1980) or whether the order of information is consistent with prereaders' existing schemata (Kintsch & Yarbrough, 1982). Kintsch's Construction-Integration model (Kintsch, 1988; 1998) provided a comprehensive account of the detailed memory mechanisms that account for those effects. Those well-established findings were obtained using tasks in which participants were asked to read for general comprehension purposes (e.g. to be prepared for a memory test, without further instruction as what they would have to do with the information). In authentic reading situations, however, readers often approach the text with more specific needs or purposes in mind, such as answering a question, making a decision, or writing an essay on a topic. Hence, there is a need for a new perspective on reading processes that accounts for different uses of text as a function of contextual variables. Current discussions about the nature of reading emphasize the fact that contextual dimensions play a critical role in shaping the reader's strategies (Cerdan et al., this volume; McCrudden & Schraw, 2007; Rouet & Vidal-Abarca, 2002; Snow and the RAND reading study group, 2002; Vidal-Abarca, this volume). Snow et al. (2002) provided a detailed account of how individual, text and task dimensions interact to shape the act of reading. They noted that "Reading does not take place in a vacuum. It is done for a purpose, to achieve some end." (p. 15). They labeled "Activity" this pragmatic dimension of reading. Snow et al. (2002) further decomposed the Activity dimension into the purposes, processes, and consequences associated with the act of reading. They suggested that the broader sociocultural context in which reading takes places contributes to shaping each of the dimensions. Even though they focused on the classroom as a context, they acknowledged that context also plays a part in out-of-school reading experiences. The main point of their discussion is that, given a particular reader and a particular text, many different reading behaviors may be observed as a function of when, where and why the reading episode is taking place. Developing a theory of reading as a contextualized activity requires a new approach to the role and importance of information in shaping the reader's understanding of the text. Rouet and Britt Relevance in multiple document comprehension 4 McCrudden and Schraw (2007) established a distinction between text-based importance, on the one hand, and relevance (or task-based importance) on the other hand. They defined text-based importance as the degree to which a text segment includes information needed to understand the text (or structural importance), whereas relevance is the extent to which a segment contains information that is necessary to perform a certain task (e.g., to answer a question). They pointed out that in most situations, readers approach the text with more or less specific goals or objectives that may affect the relevance of textual information irrespective of its structural importance. When readers do possess those goals, they may switch from a "default" processing strategy based on textual importance to a relevancebased strategy in which they focus on text segments that are relevant to the task. Indeed, Schraw, Wade and Kardash (1993) demonstrated the impact of relevance instructions on memory for text. In a pilot study, they asked a group of participants to provide importance ratings for each segment of a narrative text describing a house. Two additional groups rated the relevance (or task-based importance) of each text segment from either a homebuyer or burglar perspective. In the main experiment, college students read the text from the perspective of a burglar or homebuyer. The main finding was that readers recalled perspective-relevant segments equally well, irrespective of their textual importance. The memory-based view of reading can account for this finding. However, some task contexts lead readers to skim, skip, or ignore textual information. Rouet, Vidal-Abarca, Bert-Erboul and Millogo (2001) found that the reading time of paragraphs in a lengthy science text presented on a computer screen was deeply affected by the type of pre-questions college students were assigned prior to reading. Low-level questions prompted a "locate-and-memorize" strategy, where only segments that directly matched the question were considered, whereas higher-level questions prompted a "reviewand-integrate" strategy where readers focused more broadly on sections of the text that contained information useful to reflect on the question. What kind of cognitive processes can those shifts in strategies be attributed to? A memory-based view of text comprehension can account for variations in the saliency of information in a readers' representation of the text (see Lassonde et al., this volume). However, we believe that there is more to goal-based reading than memory activation. The reader's controlled allocation of attention and strategic decisions also play a critical role in shaping patterns of reading time and memory traces. Contextual constraints do not just act as a bias on an otherwise context-free set of procedures and cognitive processes that would be necessary and sufficient to define reading. Instead, expert readers possess a wide repertoire of flexible and differentiated reading procedures that they use depending on both the text they have at hand, and what they want to do with the text. In doing so, readers must be able to gauge the relevance of the information, a complex and multifaceted process that we discuss further in the following sections. We focus on the use of complex and multiple textual documents, which is both one of the most challenging forms of reading and offers a rich context for a close examination of relevance-based mechanisms. We first propose a general framework to represent the relevant resources and processes that take part in multiple document comprehension. We use this framework to reflect relevance-based mechanisms at various stages of the reading process, as well as on individual, text and task characteristics that may affect those processes. Information resources and cognitive resources in multiple document use In order to fully understand how relevance assessment mechanisms shape the reading experience, one needs a general framework to represent the key steps and processed involved in the contextual use of document information. Early studies of document search (e.g., Guthrie, 1988; Guthrie & Kirsch, 1987; Mosenthal & Kirsch, 1991) emphasized that reading an entire text for the general purpose of understanding its contents differs from other situations of document use. Guthrie (1988)