The purpose of this study was to investigate some Tunisian geography teachers'/researchers' reading of research articles (RA) in English in their field, in particular their use of metadiscourse and the factors that might affect this use. A combination of quantitative and qualitative methods addressed the purposes of this study. Two major reasons have motivated the choice of this topic: theoretical and contextual.
Theoretical Justifications Of The Study
A good number of studies have highlighted the facilitative role that metadiscourse plays in reading comprehension (e.g. Camiciottoli, 2003). Metadiscourse is defined, in the present study, as “self-reflective linguistic material referring to the evolving text and to the writer and to the imagined reader of that text (Hyland and Tse, 2004, p. 156). In defining reading comprehension, I adopted the componential interactive approach (e.g. Grabe, 2008). According to this approach, readers are active participants who actively take knowledge, connect it to previously assimilated knowledge and make it theirs by constructing their own interpretation. They develop, modify and even reflect on all or some of the ideas displayed in the text.
Research has shown that second language (SL) and foreign language (FL) reading comprehension process is highly complex (Grabe, 2008; Koda, 2007; Sheng, 2000). Indeed, a wide range of variables intervene in the process: linguistic, metalinguistic, cognitive, metacognitive, social and psychological (Koda, 2005; Pressley, 2006; Rapp et al., 2007). Researchers reported the difficulty to clearly understand the exact degree of the contribution of each variable to the final product. All aspects of the reader variables interact with one another and interact with textual and contextual factors (e.g. Dhieb-Henia, 2003).
Different models have been proposed in the literature in order to take account of these factors (the Top-down approach, the Bottom-up approach, the Interactive approach) (Grabe, 2008). Also, there has been a debate among SL reading researchers about whether SL reading is a language problem or a reading problem (Alderson, 1984, Bernhardt and Kamil, 1995; Grabe, 1991; Khaldieh, 2001). Some researchers contended that some SL linguistic knowledge threshold was necessary in order to get first language (L1) reading knowledge to engage and first language reading strategies to transfer (Cummins' (1979) threshold level of language proficiency and Clarke's (1978) linguistic ceiling). Others argued that reading difficulties in a SL can be caused by a 'deficient' reading ability in general, or can be caused by a failing transfer of L1 reading ability to an FL (the linguistic interdependence hypothesis, or alternatively called the common underlying principle (Bernhardt & Kamil, 1995).
Reading in English for Academic purposes (EAP) is still a more complex process (Flowerdew and Peacock, 2001). EAP is concerned with the English required for specific academic purpose such as studying at universities and colleges, doing research or publishing papers. Based on the literature, the present study views EAP reading as the ability to read an EAP discourse as the product of a particular disciplinary culture (e.g., Swales, 2004). Daoud (1991, p. 6), for instance, recommended that non-native readers (of EAP) should acquire “those abilities which would allow them to recognize the existence of certain types of implicit presuppositional rhetorical information, abilities that the experienced native reader possesses". Studies in Contrastive Rhetoric (CR) have demonstrated that texts are shaped by their cultural origins even if they participate in international discourses such as those of the science disciplines (Burgess, 2002; Connor, 2004; Kaplan, 1966, 1987). Mauranen (1993) asserted that “science, or more widely, academic research, does not exist outside writing, and so we cannot represent it, or realize it, without being influenced by the variation in the writing cultures that carry it” (p. 4).
This suggests that the belief that scientific academic discourses merely transmit natural facts is debatable. Hyland (2005) argued that discourses "are never neutral but always engaged in that they realize the interests, the positions, the perspectives and the values of those who enact them" (p. 4). The RA, for instance, is a discourse written by a particular writer belonging to a particular scientific discourse community in the purpose of engaging an audience and persuading them of his/her findings (Adel, 2006; Hyland, 2005; Martin, 2004). Webber (1994) explained “scientists are writers who fight for their ideas to be accepted, recognized, and to be competitive and get their work published” (p. 257). The tendency of scientific writers to choose a particular range of rhetorical devices, might reveal their attempts to establish interpersonal relations, to interact with their audience and to express personal attitudes about the content of their texts and about their audience(s).
Metadiscourse is the linguistic system that enables writers to achieve these goals (Adel, 2006; Hyland, 2004; 2005; Martin, 2004). It represents some internal stylistic map whereby an external reality is created and conveyed. Metadiscourse also helps to perceive discourse as a social action between two parties, author(s) and audience, researchers and scientific communities; in other words, writers and readers are viewed as social agents and texts as a social enterprise in which writers do not only want their messages to be understood (an illocutionary effect), but also to be accepted (a perlocutionary effect) (Hyland, 2005). Hyland (2001, p. 550) argues “a central aspect of the writer-reader dialogue involves careful interpersonal negotiations in which writers seek to balance claims for the significance, originality, and truth of their work against the convictions of their readers”. The reader's ability to construct the writer-intended meaning, via metadiscourse, is one major key to a successful comprehension.
This idea of interaction was grounded in Halliday's (1985) metafunctional theory of language. Halliday (1985) argued that language is a 'system of meanings' and maintains that the writer needs to operate at three levels: the ideational, the textual and the interpersonal. On the ideational plane, the writer supplies information about the subject of the text and expands propositional content, on the interactive plane, he/she does not add propositional material, but helps readers organize, classify, interpret, evaluate, and react to such material.
Many researchers have attempted to investigate the contributions of metadiscourse to language teaching. 'However, most of the literature on metadiscourse has focused on the writing skill. The immense part of these studies has compared writers' use of metadiscourse across cultures and disciplines (Dahl, 2004; Hyland & TLe, 2004; Ifantidou, 2005; Lee, 2002; Perez & Macia, 2002; Steffensen & Cheng, 1996). Only a few studies have examined the role that this crucial part of discourse plays in reading comprehension skill.
There is still an opaque picture of the correlation between the recognition of metadiscourse markers and reading comprehension performance. Indeed, a good number of the studies conducted with native readers of English has yielded inconclusive results (e.g., Crismore and Vande Kopple, 1997; O'Keefe, 1988). While some researchers demonstrated the positive role of metadiscourse (e.g. Vande Kopple, 2002), other researchers concluded that metadiscourse did not have larger effects in their studies (e.g., Crismore, 1989).
Research on the interaction between SL reading and metadiscourse seems to be still in its infancy. The little research carried out has demonstrated the complexity of the entreprise (Camiciottoli, 2003; Daoud, 1991; Dhieb-Henia, 2003; Mustapha and Premalatha, 2001). In fact, research has shown that many factors could intervene in the reading process and hamper EAP readers from using metadiscourse, namely language proficiency, prior disciplinary knowledge and metacognitive strategies (e.g., Camiciottoli, 2003; Daoud, 1991; Dhieb-Henia, 2003). Language proficiency refers to the ability to understand technical and semi-technical language used by a particular academic discourse community. Metacognition is the awareness readers have of their own mental processes and the subsequent ability to monitor, regulate, and direct themselves to a desired end. Prior knowledge includes knowledge of the rhetorical conventions of the genre and the subject discipline. It has been also interesting to note that no study has investigated SL experts' use of metadiscourse when reading materials in their fields.
Local Justifications Of The Study
Reading RAs is a prerequisite for doing research, updating one's knowledge and ultimately publishing RAs. Publication is not optional or a matter of personal choice in Tunisia. The last reform of postgraduate studies has given prior importance to the number of published RAs (Labbassi, 2000). Therefore, researchers who would like to become visible in the international scientific community, have to read and write RAs in English, the international Lingua Frinca in the age of electronic communication. In fact, in Tunisia there are few specialised local journals. Labassi (2008, p. 4120 aptly put “reading and writing English have become unavoidable conditions for joining academic communities in almost all disciplines”. The Tunisian geographical society seems to be eager to integrate into the international geographic community. Indeed, the Tunisian Geographers Association, in collaboration with the International Geographical Union (IGU), managed to organise the 31st International Geographical Congress (IGC) in the country in 2008. Tunisia was the first Arab and African country to host the biggest appointment for geographers all over the world. Adding to that, there is a clear political intention to enhance English status in Tunisia. English is gaining ground over French, which has been until recently the principal language of modernity, as well as the dominant economic language. (Champagne, 2007; Labassi, 2009a, b; Tossa, 1995). The Tunisian government is aiming at creating an English-speaking workforce to enhance the prospects for successful integration into the global economy. Under the New Maitrise reform of 1998, English was introduced in the curriculum of all undergraduate students from all disciplines (Labassi, 2009 a). What is ironical, however, is that “potential researchers and professionals who have to read a literature, which is up to 90% in some disciplines in English, are not offered courses in English” (Labassi, 2009 a, p. 249).
However, while the literature abounds with arguments for and against the role that metadiscourse plays in reading comprehension, little research has been conducted to assess the reading practices of the Tunisian geography society.Therefore, the present research aims to fill in this gap and gain more insight into the reading practices of some Tunisian university teachers/researchers of geography; in particular it aims to assess the extent they use metadiscourse to facilitate their comprehension of RAs in English in their field and to find out about the variables that may hinder this use.
The main aims of the present research are three-fold: (1) to determine to what extent Tunisian geography faculty researchers use metadiscourse markers when reading research articles in English in their fields, (2) to assess whether this use facilitates their comprehension of research articles in English in their fields, (3) to find to what extent this use is related to the participants' proficiency in English, to text familiarity (defined in the present study in terms of both content and formal knowledge) and to their use of metacognitive reading strategies.
The present study addresses the following questions:
1. To what extent do Tunisian geography faculty researchers use metadiscourse markers when reading research articles in English in their fields?
2. To what extent does this use relate to their comprehension of research articles in English in their fields?
3. To what extent do their proficiency in English, discipline-related knowledge (content and genre) and metacognitive reading strategies contribute to their use of metadiscourse markers and comprehension of research articles in English in their fields?
Significance Of The Study
This study explored the use of metadiscourse by some Tunisian geography faculty when reading research articles in English in their disciplines. Theoretically, findings from the
study can help clarify the role metadiscourse plays in SL reading comprehension. The research into metadiscourse markers is useful in itself. It helps us have an insight into how writers interact with their readers to establish a reader-friendly atmosphere and to persuade them of their findings. Such information can add to our understanding of factors contributing to language pedagogy, in particular to the teaching of reading comprehension skill.
At the practical level, information from this study can be useful for making decisions about organizing training sessions to university teachers/researchers to introduce them to the rhetorical conventions of academic genres, namely the research article and to the rhetorical importance of metadiscourse. They should be made aware of the facilitative role of metadiscourse markers when reading academic materials in English. Metadiscourse markers should be taught explicitly in EAP reading comprehension classes as a means to enhance the researchers' reading comprehension ability.
This thesis consists of six chapters. The first chapter describes the background of the research and the context in which the research was conducted. The second chapter comprises two sections: the first is a review of research into the nature of foreign language reading. The second section provides a descriptive account of what metadiscourse is, then explores the studies conducted on the effects of metadiscourse, reviews a few taxonomies on metadiscourse and ends with a description of the taxonomy to be adopted in the present study. Research questions are presented after the discussions of these reviews. Chapter three concerns the methodological decisions taken for this research. It describes how the participants were selected, how the material and instruments were piloted, and which materials and instruments were finally used. It provides as well an account of how the main study was conducted and how the data were analysed. Chapter four reports the findings of the present study. Chapter five discusses the key findings from this study and highlights their implications for the area of EAP. The contributions and limitations of this research and suggestions for future research are presented in chapter six.
The previous chapter introduced the context for this research. This chapter clarifies the two central variables underlying the present study, namely EAP reading comprehension and metadiscourse use. It can be divided into three parts. In the first part, I will focus on the reading variable. Since foreign language reading modeling has been strongly influenced by first language reading theories, I will first discuss four different approaches to reading in a first language, I will then explain how they have been adopted in and adapted for a foreign language reading context. Next, I will elaborate on theoretical concepts particular to reading in English for academic purposes (being the focus of the present study) and I will draw attention to studies that investigated the interaction between reading in English for Academic purposes and language proficiency, background knowledge and reading strategies. This part aims at underscoring the complexity of the EAP reading process and the need to account for the many factors intervening in it. In the second part, I will introduce the second variable of the present study, metadiscouse. I will first try to define the concept and then provide an overview of its main assumptions and classifications to highlight the confusion surrounding the term and emphasize its benefits to EAP readers, and last I will detail Hyland's (2005) taxonomy, the theoretical framework within which the present study is conducted. In the last part I will review some empirical studies that have investigated the interaction between metadiscourse and EAP reading. This part will also draw attention to the scarcity of research in this area.
Understanding the processes involved in reading comprehension is a prerequisite to select reliable and valid research instruments. Alderson (2000) stated “if we are not able to define what we mean by the ‘ability to read', it will be difficult to devise means of assessing such abilities” (p. 49). Likewise, Hogan (2004, p. 1) maintained
The real question we are asking when we look at assessing reading is: What distinguishes a good reader from a poor reader? Implicit in this question is an even more fundamental question: What are we doing when we read?” Assessment is an attempt to answer the first question, but if we cannot at least try to answer the second, we do not know what we are assessing, and any measure or description of reading proficiency we suggest is meaningless.
Thus, in what follows I will first attempt to define the construct of reading comprehension, and then present a number of models that provide a framework for organizing and explaining the nature of reading comprehension. I will give due prominence, however, to the issues specific to reading in EAP dwelling upon the complex cognitive processes that EAP readers go through when they read. I will try to show how certain variables such as linguistic proficiency, background knowledge and reading strategies interact with reading comprehension process.
Definition Of The Construct Of Reading Comprehension
Reading comprehension has often been a subject of controversy among teachers and scholars. Kintsch (1998, p. 2) stated “the terms understanding and comprehension are not scientific terms but are commonsense expressions. As with other such expressions, their meaning is fuzzy and imprecise”. In defining the construct of reading comprehension, I adopted Kintsch's (1998) theory of human text comprehension, as it is a “widely-recognized” theory of text comprehension and as it has “consistently defined the research agenda for the field of text comprehension” (Grabe, 2008, p. 3).
The concepts of ‘Comprehension' and ‘Understanding' are used interchangeably in the present study as “a matter of linguistic variation” following Kintsch. ‘Comprehension' is to be understood in relation to ‘perception' and ‘problem solving'. Both ‘perception' and ‘understanding' involve unconscious and automatic processes. They “can each be described as a process of constraint satisfaction” (p. 3). They differ, however, in that ‘understanding' should result in an action, be it “an overt action in the environment or a mental event” (p. 3). Kintsch (1998, p. 2) stated “understand is used when the relationship between some object and its context is at issue or when action is required”. As for the ‘problem solving' process, it is more complex and involves more demand on cognitive resources. It is an action readers resort to when they fail to understand something. Kintsch (1998) maintained “perception and understanding are the processes people normally use; when an impasse develops in perception or understanding, they resort to problem solving as a repair process” (p. 3).
Reading is the process in which the reader sequentially deals with letters, words and sentences. It was defined by Sheng (2000, p. 2) as “the process of recognition, interpretation, and perception of written or printed material”. Comprehension, on the other hand, involves the ability of the reader to grasp and interpret the meaning of written material, and to reason about cognitive processes that lead to understanding. In other words, it not only covers cognitive understanding of the materials at both surface and deep structure levels, but also the reader's reactions to the content. Sheng (2000, p. 2) maintained “it is a more complex psychological process and includes in addition to linguistic factors (phonological, morphological, syntactic, and semantic elements), cognitive and emotional factors”. In sum, the process of reading deals with language forms while the process of comprehension, the end product, deals with language content.
The term ‘reading comprehension' can, in some respects, be considered a complex construct, as highlighted by numerous reading researchers (Carrell, 1988; Grabe, 1991; Kintsch, 1998; de Beaugrande, 1981). According to Brumfit (1980), reading comprehension is a complex activity covering “a combination of perceptual, linguistic and cognitive abilities” (p. 3). It is a constructive thinking process which “involves application, analysis, evaluation and imagination” (Taylor, 1984, p.391). Grabe (2008) attributed the complexity of the reading comprehension process to the multiple purposes of reading and the complex cognitive processes involved. In the present study, reading is viewed as a process of communication between a writer and a reader and this communication will be incomplete unless it is affected by the reader's evaluation and appreciation. In fact, both the reader and the writer contribute to the reading process.
Despite the uniqueness of second language reading processes (Geva and Wang, 2001; Koda, 2007), second language reading has drawn extensively on first language reading research. Different models have been adopted and adapted based on a variety of L1 reading theories. Below is a review of these theories.
Reading Theories In A First Language
A reading model provides an imagined representation of the reading process. It provides ways to represent a theory and explain what reading involves and how reading works based on available evidence. Goldman, et al., (2007) explained “the term model refers specifically to a representation of the psychological processes that comprise a component or set of components involved in human text comprehension” (p. 27). According to Samuels (1994), a good theoretical model has three characteristics: it summarizes a considerable amount of information discovered in the past; it helps explain and make more understandable what is happening in the present, and it allows one to make predictions about the future (p. 816). Researchers, however, are somehow cautious about the comprehensibility of the model because of its inability to account for all the available evidence that exists. Dhieb-Henia (2002) warned that the models are not "always backed up by sufficient empirical evidence to validate (them)" (p. 18). In the same fashion, Grabe (2008) argued “to assert that a model must be an accurate synthesis, (...) is problematic”. Thus, these researchers recommend that we consider these models as a possible representation of the reading process, rather than absolute models. They nonetheless stress the key role that these models play in “synthesizing information and establishing central claims” (Grabe, 2008, p. 84).
Reviewing the literature, four major approaches have been proposed in an attempt to understand the reading process. The major distinction between the approaches is the emphasis given to text-based variables such as vocabulary, syntax, and grammatical structure and reader-based variables such as the reader's background knowledge, cognitive development, strategy use, interest, and purpose (Lally, 1998). The following sections review these approaches and discuss them with reference to the specific context of the present study.
The Bottom-Up Approach
The concept of decoding is central to what is usually called the bottom-up
approach to reading. The term 'bottom-up' originated in perception psychology, where it is used to signify the processing of external stimuli (Mulder, 1996). In reading
research, the term is not always used in a consistent manner and has drifted away from the original meaning it had in perception psychology. Nevertheless, the term always focuses on what are called lower order processes, i.e. decoding ability and word recognition ability (Mulder, 1996). These abilities are believed to form the key to
proficient reading. In other words, the reader perceives every letter, organizes the perceived letters into words, and then organizes the words into phrases, clauses and sentences. Meaning, at any level, (e.g. word or phrase), is accessed only once processing at previous (e.g. lower) levels has been completed (Carrell, 1988). The argument is that bottom-up processing requires a literal or fundamental understanding of the language. Carrell (1993, p. 2) maintained
Reading (is) viewed primarily as a decoding process of reconstructing the author's intended meaning via recognizing the printed letters and words, and building up a meaning for a text from the smallest textual units at the bottom (letters and words) to larger and larger units at the top (phrases, clauses, intersentential linkages.
Advocates of this theory argue that successful reading comprehension of a text relies heavily on an efficient application of bottom-up processes. The best known representative of this type of approach is Gough's (1972) model. The model is summarised by Urquhart and Weir as follows:
T]he reader begins with letters, which are recognized by a SCANNER. The information thus gained is passed to a DECODER, which converts the string of letters into a string of systematic phonemes. This string is then passed to a LIBRARIAN, where with the help of the LEXICON, it is recognized as a word. The reader then fixates on the next word, and
proceeds in the same way until all the words in a sentence have been processed, at which point they proceed to a component called MERLIN, in which syntactic and semantic rules operate to assign a meaning to the sentence. ... The final stage is that of the Vocal System, where the reader utters orally what has first been accessed through print. (Urquhart &Weir, 1998: 40)
The decoding approach, however, has been viewed by many as inadequate. According to Eskey (1973), the approach underestimates the contribution of the reader. The reader, according to this approach, does not read the text through a self-determined, predefined perspective or goal, but rather lets the text itself (and therefore its author) determine the reading process (Urquhart & Weir, 1998). The approach fails to recognize that readers utilize their expectations about the text based on their knowledge of language and how it works. Similarly, Carrell (1984) stated that this view assumes a rather passive view of reading. Grabe (2008) stated “we know that such an extreme view of reading is not accurate, and no current model of reading depicts reading as a pure bottom-up process” (p. 89). Criticism of the bottom-up theory has given impetus to the Top- down theory.
The Top-Down Approach
Whereas the bottom-up approach gives incoming information a central place in the reading process, the top-down approach focuses on the knowledge a reader already possesses. It stresses what are called higher order cognitive processes. The top-down theory posits a non-linear view of the reading process, i.e. from higher levels of processing, and proceeds to use the lower levels selectively. It assumes that readers interrogate the text rather than process it completely. They get meaning by comparing their expectations to a sample of information from the text. The proponents of this theory argue that reader's experience and background knowledge is essential for understanding a text.
Grabe (2008) explained “top-down models assume that the reader actively controls the comprehension process, directed by reader goals, expectations, and strategic processing” (p. 89). Carrell (1993, p. 4) stated
In the top-down view of second language reading, not only is the reader an active participant in the reading process, but everything in the reader's prior experience or background knowledge plays a significant role in the process. In this view, not only is the reader's prior linguistic knowledge (“linguistic” schemata) and level of proficiency in the second language important, but the reader's prior background knowledge of the content area of the text (“content” schemata) as well as of the rhetorical structure of the text (“formal” schemata) are also important.
According to this view, the reader's background knowledge may compensate for certain syntactic and lexical deficiencies. Readers start with their background knowledge (whole text) and make predictions about the text, and then verify their predictions by using text data (words) in the text (Urquhart & Cyril, 1998). Clarke and Silberstein (1977, p.136-137) stated that “more information is contributed by the reader than by the print on the page, that is, readers understand what they read because they are able to take the stimulus beyond its graphic representation”. The most frequently cited representative of this approach is Goodman's (1976) top-down model. He defined reading as a process of verifying hypotheses — hypotheses that are based on knowledge which the reader possesses. His model, also called the guessing game theory, is summarised by Bossers as follows:
According to Goodman, the reading process consisted of sampling and selecting cues, on the basis of which an interpretation is predicted or guessed, which is subsequently tested against the semantic context, and then confirmed or rejected as the reader processes further language, and so on. These stages of the process were called 'features of the reading process' or 'effective strategies' or 'effective reading behaviour' alternatively. (Bossers,1992:10).
A very influential theory that is usually discussed in relation to the top-down
perspective is that of schema theory. The prior knowledge gained through experiences, stored in one's mind and activated when readers encounter new information is referred to in the literature as schemata (Carrell, 1980; Widdowson, 1983).
Schema theory, which comes from cognitive psychology, owes much to the work of Bartlett (Rumelhart, 1981) and Piaget (Orasanu and Penny, 1986). Schemata, the plural form of schema, also called ‘building blocks of cognition' (Rumelhart, 1981, p. 3), refer to “abstract knowledge structure (s) stored in memory” (Garner, 1987, p. 4). They are defined as the mental framework that helps the learner organize knowledge, direct perception and attention, and guide recall (Bruming, 1995), as cognitive constructs which allow for the organization of information in long-term memory (Widdowson,1983) and as the underlying connections that allow new experience and information to be aligned with previous knowledge ( McCarthy ,1991).
Within the framework of schema theory, reading comprehension is no longer a linear, text-driven process, but is the process of the interpretation of new information, and the assimilation and accommodation of this information into memory structures or schemata (Anderson & Pearson, 1984). Schemata are not static entities, however, but are continually constructed and reconstructed through the processes of assimilation and accommodation. In other words, a comprehension of a text involves activation of relevant schemata, which are initiated as a result of ‘bottom-up observation', and proceeds through a constant process of testing the activated schemata, evaluating their suitability, and refining or discarding them (Rumelhart, (1984, pp. 3, 6). Rumelhart (1981, p. 4) stated “according to schema theories, all knowledge is packed into units … (which) are the schemata. Embedded in these packets of knowledge, in addition to the knowledge itself, is information about how this knowledge is to be used”. The schemata or old information which “provide general “ideational scaffolding” for new information” (Garner, 1987, p. 7) are activated to make new incoming information comprehensible. According to Alderson (2000, p. 17) schemata act as ‘filters' for new information. Grabe (1991, p. 390) stated that schema theory has been an extremely useful notion for describing how prior knowledge is integrated in memory and used in higher-level memory processes”.
Three different types of schemata are distinguished in the literature: linguistic schema, formal schema and content schema Carrell (1983). Linguistic or language schema refers to readers' existing language proficiency in terms of vocabulary, grammar and idioms. According to Alderson (2000, p. 80), linguistic knowledge includes “phonological, orthographic, morphological, syntactic, and semantic information”. These schemata are needed to recognize words and how they fit together in a sentence. Readers, may via repeated exposure, be able to generalize a pattern or guess the meaning of a word, which may not have initially been part of their linguistic schema. Without linguistic schemata, it is impossible for the reader to decode and comprehend a text.
Formal schema, often known as textual schema, includes “discourse-level knowledge, including …text organization and cohesion, text types and associated conventions, as well as metalinguistic knowledge” Alderson (2000, p. 80). It can include knowledge of different text types and genres, and also of their organization, language structures, vocabulary, and level of formality/register differently. Schooling and culture play the largest role in providing a reader with a knowledge base of formal schemata (Singhal, 1998).
Content schema refers to a reader's background or world knowledge (Carrell and Eisterhold, 1983; Carrell, Pharis, and Liberto, 1989). Content schema provides information that aids in the interpretation of meaning. That is, readers attempt to match content schemata with text information. As the reader does this, he/she builds a mental concept for the meaning of the text. This concept is constructed partially out of information previously known and partially by the new information presented in the text. The processes of building and refining mental concepts of meaning allow comprehension to occur (Armbruster, 1986).
Despite its importance, the schema theory has some shortcomings. the literature does not mention where these schemata originate from and/or how they are acquired. Alderson (2000) pointed out that schema theory does not clarify how completely new information is dealt with (e.g. how similarities between old and new information are detected). The main shortcoming of a top-down approach is that it is impossible to see how a reader can begin by engaging the text as a whole, then proceed to paragraphs, then to individual sentences, ending with single letters. In other words, this process (background knowledge, then paragraphs, then sentences, then single letters) may not represent what really happens (Alderson, 2000).
Urquhart and Weir (1998, p. 70) also listed different reasons 'for believing that schemata are not very useful in reading research'. They signal the problematic relationship between the main characteristics of schemata: on the one hand schemata are necessarily pre-existing and pre-structured, but on the other hand schemata should be instantly adaptable, instantly activated and acquired for schema theory to be successful. They criticized the fact that schemata were rarely described in detail: it was not clear what exactly a schema comprised. The view of reading as an interactive process has been considered as “a typical compromise solution” (Grabe, 2008) and is explored in the following section.
The Interactive Theory
This theory considers comprehension as drawing upon both top-down and bottom-processes. The claim is that bottom-up processing influences top-down processes, and vice versa. In fact, the two approaches tend to be compensatory. If ottom-up skills are weak, there is the risk of misunderstanding the basic meaning from which top-down skills are built. On the other hand, if top-down skills are ignored, learners become passive readers, and do not develop the analytical skills important to good readers. Dubin and Bucina (1991, p. 197) posited “the two processes, bottom-up and top-down, are complementary; one is not able to function properly without the other”. Likewise, Grabe (2008) argued “readers most definitely are not either top-down or bottom-up readers. Instead, readers are, of necessity, always both bottom-up and top-down readers” (p. 56).
According to this interactive approach to reading, the balance between the complex interactions of different types of information activating various skills varies with text, reader and purpose (Alderson, 2000). In the same vein, Grabe (1991) maintained that the term ‘interactive' refers to three different conceptions: First, the interaction that occurs between the reader and the text whereby the reader constructs meaning based on the knowledge drawn from the text and the existing background knowledge the reader has; Second, the interactivity occurring simultaneously between many component skills that results in reading comprehension; Third, the textual interaction that is the interaction between various linguistic dimensions within the text that together underline a text as such rather than a series of unconnected sentences. Therefore, Grabe (1991) asserted that from an interactive approach, the reading process is seen “as a kind of dialogue between the reader and the text” (p. 350) involving “both an array of low-level rapid, automatic identification skills and an array of higher-level comprehension interpretation skills” (p. 383). In the literature on process models one approach is often cited as representative of interactive approaches: Rumelhart's (1970) interactive Model. The model is interactive in that there is no hierarchy in the activation of processes or absolute direction in the interplay of different information resources.
Limitations associated with interactive process models concern first the nature of the interaction. Existing interactive models have not been able to give a full account of how and when (particular) interactions take place exactly. They are limited to explaining the results of the interaction. In a recent publication, Grabe (2008) criticized interactive models on the ground that little empirical evidence supports their assumptions. He argued (p. 90):
Research demonstrates that (a) fluent readers do not wait for context information to support automatic word-recognition processes; (b) inferences do not impact automatic word-recognition processes; and (c) eye-movement patterns follow consistent and fairly automatic processes; they are not usually under the conscious direction of the reader during fluent reading.
Grabe (2008) supported instead restricted interactive models. The latter are hybrid models that combine two important features of the major theories of reading comprehension: (1) the interaction of information sources and (2) restrictions on these interactions. In other words, the more automatic the reading process at the bottom-up level is, the less influential the top-down level is. Grabe (2008) stated “restricted interactive models are primarily bottom-up driven with respect to automatic processing (word recognition, syntactic parsing, proposition formation). At the same time, automatic processing may require interaction among processes and resources within a given component skill level.” (p. 90).
A second, related criticism to the interactive approach is that they do not clarify the relative importance of the different sources of information involved in the interaction (Bossers, 1992). This lack of explicitness of the interaction gives way to innumerable possible variations of the interaction. Interactive views, therefore, seem to contribute “rather superficially to the unraveling” of the process of reading. There was more advocacy for componential interactive approaches (e.g. Grabe, 2008).
The Componential Theory
A componential theory of reading “attempts to identify a set of functionally defined information processing systems or components which, in interaction with one another, accomplish the more complex performance—in this case, reading with comprehension” (Frederiksen, 1981, p. 5). This definition may hint that readers could differ depending on the degree of interaction among these components. Indeed, the interaction of these components may explain individual differences and reading difficulties. Frederiksen (1981, p. 7) explained “a skilled reader possesses many, highly automated components, while a less skilled reader has a smaller number of such components, and those may be quantitatively less automated”. Likewise, Gibson and Levin (1975, p. 454) suggested “there are as many reading processes as there are people who read, things to read, and goals to be served”.
A model that would be a good example of such an approach is the Interactive-Compensatory model (ICM) presented by Stanovich (1980, 1986, 2000). The model stipulates that whenever a weakness or lack in knowledge or skills is encountered which inhibits or distorts reading comprehension, it is compensated by other knowledge resources or skills. Stanovich (1986) explained “deficiencies at any level in the processing hierarchy can be compensated for by a greater use of information from other levels and that this compensation takes place irrespective of the level of the deficient process” (p. 49). A deficit in any particular process will result in a greater reliance on other knowledge sources, regardless of their level in the processing hierarchy. In other words, the model suggests that effective reading depends on the dynamic interrelationship among existing knowledge; it is possible for most learners to compensate for weaknesses in one area using strengths in other areas. For example, a lack of vocabulary knowledge may be compensated by far-reaching knowledge of the topic of the text.
The ICM was developed primarily to explain developmental and individual differences in the use of content to facilitate word recognition during reading. It theorizes on the differences in individuals' reading abilities. That is to say, poor readers use contextual information to compensate for weak word recognition skills, a contextual facilitation of word perception strategy; good readers who read with ease and in an automatic fashion consider less often this strategy. Contextual facilitation or facilitation of word perception is useful only to poor readers to compensate for their difficulties in decoding. Good readers perceive words using data-driven strategies saving cognitive capacity for comprehension monitoring.
The ICM includes five main components: cognitive abilities, an organized knowledge base, strategies, metacognition, and motivational beliefs. Cognitive abilities refer to one's general capacity to read. The knowledge base refers to organized, domain-specific knowledge and general knowledge in long-term memory. Strategies refer to procedures that enable readers to solve specific problems. Metacognition includes knowledge about oneself as a learner, and how to regulate one's learning. Motivation refers to beliefs about one's ability to successfully perform a task,as well as one's goals for performing a task. Knowledge and regulatory skills such as strategies and metacognition are combined into one overarching module because of the close relationship among these three components.This model has been adopted by many researchers (Carrell, 1993; Rumelhart, 1980; Sanford and Garrod, 1981; Van Dijk and Kintsch, 1983).
Anderson et al. (1982) posited that “comprehension of a message entails drawing information from both the message and the internal schemata until sets are reconciled as a single schema or message" (p.187). Similarly, Dubin and Bucina (1991) stated “the two processes, bottom–up and top–down, are complementary; one is not able to function properly without the other” (p. 197).
In the first section, I gave a brief insight into the major reading theories in the first language reading process, as they have strongly informed SL reading research (the latter will be reviewed in the following section). The major conclusion to be drawn is that reading is conceived, by most current researchers, as an interactive cognitive process in which readers interact with the text, the writer and other contextual variables (e.g. Barnett, 1989; Grabe, 2008; Hudson, 1998; Koda, 2007; Paran, 1994). The level of reading comprehension of the text is determined among others by how well the reader variables (knowledge of the topic, foreign language abilities, awareness of the reading process) interact with the text variables (text type, structure, syntax, and vocabulary) (Hosenfeld, 1979). Reading is a complex activity. Indeed, Alderson and Urquhart (1984, p. xxvii) maintained
…the study of reading must be interdisciplinary. If the ability involves so many aspects of language, cognition, life and learning, then no one academic discipline can claim to have the correct view of what is crucial in reading: linguistics certainly not, probably not even applied linguistics. Cognitive and educational psychologists are clearly centrally involved; sociology and sociolinguistics, information theory, the study of communication systems and doubtless other disciplines all bear upon an adequate study of reading.
This view of reading is adopted in the present study. But as the participants in the present study are EAP readers, it is important at this stage to shed light on SL reading research and in particular on EAP reading.
SL Reading Comprehension Theories
SL/FL theories of reading were based primarily on the principles of the bottom-up, the top-down and the interactive approaches. Following the bottom-up approach, some researchers claimed that decoding was the key to FL reading and that FL readers were faced with problems similar to beginning technical reading. Indeed, the studies were unidimensional in nature. The areas of investigation were: text structure (e.g., Ross, 1994); syntax (e.g., Berkemeyer, 1994) and word knowledge (e.g., Laufer & Hadar, 1997). These studies identified an insufficient speed and accuracy in L2 decoding as the main obstacle to proficient L2 reading. This approach was questioned by many because the reading comprehension process was mainly hindered by a lack of SL vocabulary knowledge rather than by poor decoding. The graphic input is usually processed correctly while reading in an SL, but no meaning can be assigned to what is read, thus hindering the efficiency of the reading process (e.g., Bernhardt, 2005).
With regard to the top-down approach, empirical studies using Goodman's (1971) guessing game theory yielded problematic results (Bhatia, 1984; Clarke, 1979, 1980; Connor, 1981). They argued that Goodman's approach could only partially explain their results. For example, Goodman's theory did not account for the fact that the advantage good L1 readers had over poor L1 readers was not found reflected in SL reading (Clarke,1979) and it did not account for the slower SL reading speed found in skilled bilinguals (Segalowitz, 1986). Further objections against top-down approaches were based on studies that indicated that poor SL readers also do a lot of guesswork while reading (while this is characteristic of proficient reading according to Goodman's theory) and that people reading in an SL need extensive vocabulary knowledge in that language in order to read with comprehension (Bossers, 1992).
The view of SL reading as interactive was adopted by a good number of researchers(Bernhardt, 1986; Carrell, Devine & Eskey, 1988). Reading is regarded an interactive process wherein the reader variables (interest level in the text, purpose for reading the text, knowledge of the topic, foreign language abilities, awareness of the reading process, and level of willingness to take risks) interact with the text variables (text type, structure, syntax, and vocabulary) (Hosenfeld, 1979). Bernhardt (1991) developed a model in an attempt to capture in a comprehensive manner the variables in the second language reading process.
However, it was only since the mid 1980s that researchers increasingly have showed interest in adult reading and in second or foreign language reading (Alderson, 1984). They found out that there are numerous other variables (of unequal and variable importance) that can influence how second language readers go about trying to understand an academic text and how successful these efforts would be. Bernhardt (2005) maintained “it was clear that the variables involved (in FL reading) are significantly more complicated than the set involved in general L1 reading” (p. 135). When the reader has different conventions or assumptions from the writer of a given text, the contract breaks down and comprehension can suffer. Eskey (1986) further explained that the mismatch of the conventions or assumptions between the reader and writer might be because of different culture values:
The literate second language reader is a product of a culture which may have very different ideas about reading from those that the unwary teacher takes for granted. Such a student may have completely different conceptions of what reading is, how it should be done, and what it normally is used for from those of the teacher in what might be called the standard amercian academic setting (p.4)
Therefore, the researchers have denounced the application of L1 to SL reading research. Bernhardt (2005), for example, criticized “the slavish replication of studies conducted in first language” (p. 133) and “the overadoption of Schema Theory” (p. 134). Weber (1991) maintained that SL reading research has been “derivative and nonoriginal”. Bernhardt (2005) gave the example of Carrell's (1983) study which replicated the famous “washing clothes passage” and “balloon serenade” used in Bransford and Johnson (1973) with English as a second language. Bernhardt (2005) argued that first and second language reading processes have different underlying dimensions. She stated “the mere existence of a first-language renders the second language reading process considerably different from the first-language reading process because of the nature of the nature of information stored in memory” (p. 112).
A new debate emerged among SL reading researchers about whether second language reading is a language problem or a reading problem (Alderson, 1984, Bernhardt and Kamil, 1995; Grabe, 1991; Khaldieh, 2001). Coleman (1971) reported that lexical complexity of texts account for 80% of the variance in SL reading comprehension, implying that the higher one's degree of lexical knowledge, the higher the degree of comprehension. Schoonen et al. (1998, p. 87) brought evidence that “foreign language vocabulary is the best predictor of FL reading comprehension” (p. 89). One's knowledge of discourse and text structure has also been reported to affect reading comprehension. For example, understanding coherence, or superficially non-existent relations between sentences, is a determining factor in reading comprehension (Taylor, 1985, p. 2).
This debate has been translated into two hypotheses dominating the field: the linguistic threshold hypothesis and the linguistic interdependence hypothesis. They have provided the frame work for many empirical studies. The linguistic threshold hypothesis, also known as the short-circuit hypothesis, stated that: “In order to read in a second language, a level of second language linguistic ability must first be achieved” (Bernhardt & Kamil, 1995, p. 17). According to this hypothesis, problematic reading in an SL was caused by a language problem — more precisely, inadequate knowledge of that language. What exactly the term 'language knowledge' embodies differs considerably: some considered it to be inadequate knowledge of vocabulary, others stated it to be inadequate knowledge of grammar, and others also stated that a strong interaction between both vocabulary and grammatical knowledge was necessary (Ulijn & Salager-Meyer, 1998).
A concept related to this hypothesis is that of linguistic ceiling. This concept was first formulated by Clarke (1979). In a 'weaker' interpretation, it was stated that 'below a certain threshold in SL proficiency, comprehension processes which were used by readers in L1 could not be used as effectively in SL reading' (Ulijn & Salager-Meyer, 1998, p. 82). In a 'stronger' interpretation, it was stated that 'a certain threshold level of SL language knowledge is necessary before LI reading ability transfers to L2 reading' (Yamashita, 2002, p. 81). Both interpretations share the idea that a certain level of language knowledge needs to be reached before understanding can take place. Berhardt (1986) and Koda (1987) demonstrated that a given amount of SL grammatical/linguistic knowledge was necessary in order to get L1 reading knowledge to engage and L1 reading strategies to transfer. Bossers (1991) stated that the students who are strategic readers in their first languages may be inhibited to transfer L1 strategies into L2 because of L2 proficiency problems. He (1991) maintained
Poor foreign language reading is due to reading strategies in the first language not being employed in the foreign language, due to inadequate knowledge of the foreign language. Good first language-readers will read well in the foreign language once they have passed a threshold of foreign language ability. (p. 47)
Although not always considered as different from the above description and related to the 'stronger' interpretation of the language threshold , Alderson (1984) discussed what he calls a modification of the linguistic threshold hypothesis. This modification states that SL reading problems are caused by a lack of transfer of L1 reading strategies to reading in an SL due to an inadequate knowledge of the SL. SL language competence, in this view, is a prerequisite for the transfer of strategies across languages. Usually this modification is referred to by the second name under which the hypothesis is known, i.e. the short-circuit hypothesis. Bossers (1992) stated that this hypothesis is only relevant for good L1 readers, defined as those who efficiently use high level processes (strategies), since weak LI readers have nothing to transfer to the SL. Good LI readers' SL reading process can be 'short-circuited' by a lack of (or insufficient) SL language knowledge (relevant to a particular text) or by slow and inaccurate lexical processing, leaving little capacity for higher level processing (inhibiting transfer of reading strategies).
Opponents of the linguistic threshold hypothesis warn of the 'trap' of the threshold concept — i.e. an absolute linguistic level, valid for all tasks and all subjects (Urquhart & Weir, 1998). They argued that the language threshold depends on a number of factors. These include task demand, cognitive development, and level of background knowledge. Mulder (1996) called the strict adherence to the language threshold concept too narrow (reading as an exponent of language knowledge rather than a creative meaning reconstruction process), even though she recognized the importance of vocabulary knowledge to FL reading. Although many researchers did not seem to strictly adhere to this hypothesis (e.g. Alderson, 1984), they did recognize the existence of some kind of threshold: [S]ome threshold does, (...), appear to be necessary before other abilities, like one's firstlanguage reading ability, can be brought to bear upon the task of reading in a foreign language. (Alderson, 1984, p. 19)
The second hypothesis dominating the discussion on FL reading is the linguistic interdependence hypothesis, or alternatively called the common underlying principle (Bernhardt & Kamil, 1995). This hypothesis stated that LI reading abilities (strategy use) transfer to reading in an FL. Underlying is the idea that '[r]eading performance in a second language is largely shared with reading ability in a first language' (Bernhardt & Kamil, 1995, p. 1718). This hypothesis is associated with the view that failing sL reading comprehension is a reading problem. Reading difficulties in an FL can be caused by a 'deficient' reading ability in general (thus in the LI as well), or can be caused by a failing transfer of LI reading ability to an FL. This second part, poor FL reading caused by a problematic transfer of LI reading ability, has mainly been supported by studies from bilingualism. Cummins (1979), doing research on bilingual children's performance in an educational context, introduced the concept of Cognitive Academic Language Proficiency (CALP). Based on his findings, he stated that there is one single underlying dimension which directs our CALP and therefore, CALP underlies language proficiency in whatever language.
SL reading is, currently, most often seen as a balanced combination of the above mentioned approaches (e.g. Grabe, 2008). Comprehensive models that can account for all SL variables are missing in the literature though. Indeed, the research in the 1980s and 1990s tried to capture a “holistic depiction” of the interaction of variables in the second language reading process, but there has been no consensus about a ‘satisfactory' integrated model of these variables (Bernhardt, 2005, p. 136). Most of the researchers applied L1 reading models such as Stanovich's interactive compensatory model to the FL context. Bernhardt (2005) commented on Stanovich's (1980) model as one that 'captures the current knowledge base regarding literacy knowledge, language knowledge with particular emphasis on vocabulary, and dimensions under investigation, but not yet explained.' (Bernhardt, 2005, p 140). The only model reported to have been designed especially for FL reading contexts is Bernhardt's(1991 and 2000). Grabe (2008) stated “in L2 reading research, only one general descriptive model of reading has been proposed (Bernhardt, 1991, 2000)”. Bernhardt (2000) proposed a three-dimensional compensatory model, which she considered an improvement on her former model (1991) for the conceptualisation of the multi-dimensional process of reading. The model “illustrates that knowledge sources are not additive, but rather operate synchronically, interactively and synergistically” (Bernhardt, 2005, p. 141). The model was, however, criticized on the ground that it was “somewhat vague in its specification of component abilities and in its implications for reading development” (Grabe, 2008, p. 104).
In a later study Bernhardt (2005), advanced what she called the requirements for a comprehensive model of FL reading: 1) 'acknowledge the significant contribution of first language reading ability to second language comprehension', 2) 'enable a conceptualization of comprehension as consisting of different elements and influences', 3) 'concede that in the reading of cognate languages there is no such thing as "no knowledge" if the reader is already literate and, at the same time, admit that when switching to non-cognate languages, the threshold is set at a very different point', and 4) 'encompass a consideration of unexplained variance in individual performance and after considerable time in learning'(Bernhardt, 2005, pp. 138-139).
This limited development in the field of foreign language reading research may be attributed to the complexity of SL reading comprehension process. Indeed, SL readers invoke a unique set of constraints. Reading in English for Academic purposes is, however, more demanding and more challenging for those who are educated in their native languages and have different cultures and educational systems. Flowerdew and Peacock (2001) explained
EAP reading involves a number of specific difficulties. The registers/genres of different disciplines are different from those of ‘general English'. Students may do well in ‘reading lessons' in general English, but have difficulty in reading in their subject areas. Also the aims are different: reading narrative may be fro enjoyment alone, but in subject areas students often read to perform some task- to learn about something, get information, learn how to do something or draw material for argument (p. 185).
Reading In English For Academic Purposes (Eap)
I start this section by explaining what EAP is. EAP is concerned with the English required for specific academic purpose such as studying at universities and colleges, doing research or publishing papers. EAP reading is unique in several ways. First, the materials are read within a specific academic setting. The academic setting is characterized by a particular academic culture and a particular disciplinary culture and those involved are expected to be (come) academically literate. Academic culture concerns the values, roles, assumptions, attitudes, patterns of behavior, etc. in that setting. Disciplinary culture concerns the theories, concepts, norms and terms of a particular academic discipline (Flowerdew & Peacock, 2001). Operating successfully within the academic setting is usually linked to an understanding of the academic culture and discipline. Johns (1997, p. 2), in defining, academic literacy argues that literacy involves 'ways of knowing particular content, languages, and practices', that it includes 'strategies for understanding, discussing, organizing, and producing texts' and that '(i)t relates to the social context in which a discourse is produced and the roles and communities of text readers and writers' (Johns, 1997: 2). In essence, she says, '(w)hat this term does is integrate into one concept the many and varied social, historical, and cognitive influences on readers and writers as they attempt to process and produce texts' (Johns, 1997: 2). Braine (2002), calls good reading and writing skills, research skills and knowledge of one's chosen field of study only 'the foundation' for becoming academically literate.
Many researchers have noticed that difficulties SL readers face when reading academic texts are not necessarily due to insufficient SL proficiency. Rather, those difficulties relate to the specific characteristics of academic genres. Spiro and Taylor (1987) maintained that texts that are markedly different in type make distinct demands upon readers' knowledge and expectations about reading, and have important consequences for cognitive processing. Similarly, Flowerdew and Peacock (2001) observed that “each academic discipline differs in its ways of arguing for a particular point of view, interpreting data, considering different sides of an argument and drawing conclusions” (P. 187). Thus, comprehension should be seen in the large context of the discipline community (Meyers, 1991). Swales (1990) introduced the concept of genre. He considered a genre a “class of communicative events, the members of which share some set of communicative purposes” (p. 58). These shared communicative purposes are recognized by the expert members of a discourse community and turn out to be the rationale for a genre. As a result, the schematic structure of the discourse and the participating members' choice of content and style are under the influence of the rationale embedded in a genre. Bhatia (1993, p. 15) added that there are restrictions on the vocabulary, grammar and discourse structure of any particular genre, and these purposes are well understood and skilfully used by expert members of a particular discourse community to fulfil their own purposes. Dudley-Evans (1994) cited studies on the rhetoric of different disciplines and the practices of different discourse communities which found, for example, that there is variation in the patterns of argumentation favoured by different disciplines. In a study of dissertations of students in engineering and biology. In another study Dudley-Evans (1993) found that there are even differences in the organisational patterns of the discussion sections across disciplines. Dudley-Evans and St John (1998) furthermore argued that the 'adopted' set of text genres varies between discipline areas and the features of a particular genre may itself also vary between discipline areas. Clapham (2001) also contended that even within one discipline, there are variations in terms of level of subject specificity of academic reading material, 'ranging from those that are comprehensible to all educated readers to those highly technical ones which can only be read by experts in some limited sub-field' (p. 89). In an earlier version, Clapham (1996) found that even within the same text genre or between different text sections, variations in specificity occur and several discourse types may be used in one text document.
This makes it extremely important for readers to be familiar with these conventions. For example, failing to recognize the canonical order of the experimental research article (Introduction, Methods, Results and Discussion (IMRD)) and the different communicative functions of its sections may result in a reading comprehension failure. The section “Results”, for instance, indicates that what follows is offered as pure description without interpretation, whereas the section “Discussion” indicates that the authors' interpretations are to follow (Norris and Philips, 1994).
Readers of research articles should also be aware that the discourse of the research article is not merely expository. Rather, it consists of interactions among scientists in which the maintenance of face is crucial (Brown and Levinson). In fact, writers seek to convince their audience(s) of the validity and importance of the results being reported and ultimately to get them accepted and used by a consensus of the community (Latour, 1987). As claim making tends to involve tensions (stating a new claim always means denying or superseding the claims of others), discourse makers are expected to have recourse to politeness devices to stress solidarity and to mitigate both claims and denials of other claims (Brown and Gilman 1960; Meyers, 1991). One of mitigating a criticism is through the use of hedges.
A further cognitive demand upon readers of research articles in specialized fields is interpreting the pragmatic meaning of the discourse. Speech act theory (Austin, 1962; Searle, 1979) holds that the readers' success in interpreting a particular academic discourse depends on the ability to discern the illocutionary point of an utterance, i.e., the communicative function of the utterance (for instance, to assert, to suggest, to question, to promise, and so on). Readers should also be aware that these illocutionary functions vary according to the context in which they occur (e.g. suggesting in one context may be asserting in one another). Interpreting pragmatic meaning is important as it helps readers decide whether the statement being offered, for example, is in support or critical of another. Pragmatic interpretation is basic to an ultimate goal of academic literacy, the critical interpretation of academic texts (i.e., the questioning of the believability of what the text says). Olson and Bbu, (1992) even claimed that if readers fail to critically interpret the texts, they are academically illiterate. They (1992, p. 184) maintained that “the interpretation, analysis, and criticism of written texts is what critical thinking is and what it is for”. Likewise, Feynman (1969) maintained that just because something is asserted, even by an expert, it is fallible and must, therefore, be questioned.
The critical thinking of the text can be interpreted using metadiscourse. Hyland (2005) argued that metadiscourse would enable readers to think about illocutionary points that show degree of certainty and uncertainty, such as hedging (e.g. possibly, might, tend), framing, which makes a text cohesive (e.g. in the following section, the first point is) and evaluating (e.g. surprising, important). A further and more comprehensive discussion about metadiscourse will be presented in the second part of the literature review.
Reviewing the literature, we have seen how issues related to what variables interfere within the reading process and how they contribute to the ease with which an academic text is understood are essential to be able to answer the present research questions. However, the majority of studies interested in L2 literacy in academic contexts have focused on writing competence. Only a small portion of EAP research has focused mainly on reading. Thus, in the following section, I will limit my review the studies that investigated the interaction between background knowledge, language proficiency and reading strategies with EAP reading, the independent variables of the present study.
The Interaction Between Language Proficiency, Background Knowledge And Metacognitive Reading Strategies And Reading Comprehension
Before reviewing the studies into the effect of language proficiency, background knowledge and metacognitive reading strategies and reading comprehension on EAP reading performance, conceptual definitions of these variables are needed.
Language proficiency refers to the type of language used in an academic setting to convey a certain communicative purpose. It is defined, in the present study, as the ability to read academic language used by a particular academic discourse community. The term metacognitive refers to “one's knowledge concerning one's own cognitive processes and products or anything related to them” (Flavell, 1970). Examples of such processes are minoring activities for comprehension purposes, self-questioning and taking corrective actions when comprehension fails (Brown, 1980; Mokhtari and Reichard, 2002). A strategy is defined as “any overt purposeful effort used or activity used by the reader to make sense of the printed material with which he or she was interacting” (Jimenez, Garcia and Pearson, 1995, p. 76). As for the concept of background knowledge, it is referred here as knowledge of the subject-area treated in the text and the overall text organization. The latter has been defined differently in the literature (e.g. genre (Swales, 1990), text structure (Roller, 1990), text type (Meyer and Freedle, 1984) and rhetorical structure (Lee, 2006)).
Research on the relationship between these variables and reading comprehension has yielded different results. Some studies have proposed results congruent with the suggestion that partial or false background knowledge can interfere with comprehension (e.g., Carrell and Eisterhold, 1983; Gordin, 1987); in other studies, however, no correlation was found (e.g. Bernhardt, 1991a). Bernhardt (1991b) attributed this to the tendency among researchers “to attribute more substantial weight in language text processing to knowledge of topic than to knowledge of language” (p. 37) and to “the artefacts of manner in which the measurement was taken” (p. 39). Ridgway (1997) explained these contradictory results: “the field in which we are working is not that only of reading in a foreign language, but also a leading edge in research into human cognition and psycholinguistics” (p. 152). Below is a brief overview of a number of empirical investigations in the field (Block, 1986; Carrell, 1984, Garner, 1987).
Tan (1990) carried out a study to measure the role of prior knowledge and language proficiency as predictors of reading comprehension among 317 undergraduates of medicine (141), law (95) and economics (81). The participants took three tests: (1) a general English proficiency test (a short version of a British Council test (Davis and Alderson, 1977)); (2) prior knowledge tests to measure the subjects' background knowledge in the disciplines involved in the study, and (3) discipline-related English reading tests. Results gave empirical evidence to show that prior knowledge of the topic domain of the text and competence in the target language were significant predictors of comprehension. The strength of prediction from a score on a language proficiency test was, however, two times greater than that of the score on a test of prior knowledge.
Alderson and Urquhart (1983, 1984, 1985) investigated the effect of background knowledge on students attending English classes in Britain in preparation for courses at British universities. They gave groups of students in three different disciplines –business and economics, science and engineering, and liberal arts- the Social Studies and Technology Modules of the English Language Testing Service (ELTS) test. In the first study, the low sample size prevented performance of significance tests, although the effects were in the expected direction. In the second study, students in the field of development administration significantly outperformed engineering students on content related to the former area, while the two groups performed about equally well on content related to the latter area (with students of science and mathematics performing similarly to those in development administration). Among the findings of the third study, the combined group of science, mathematics and engineering students significantly outperformed development administration students on the reading module that was related to the former areas, while the two groups performed about equally well on the module related to the latter area. The researchers concluded, therefore, that although students in some disciplines achieved higher scores if they were given tests in their own subject areas, these findings were not consistent across disciplines.
The same inconclusive results were also found by Peretz (1986). The author administered a reading test with passages in science and technology, biology and humanities/social sciences to students in these three areas. She observed a significant interaction between major-field area and text content in the expected direction, but she also found that science and technology students outperformed the other groups overall. Language proficiency levels seem to play at least as important a role as background knowledge in the comprehension of reading texts. Peretz (1986) concluded “ the question whether students would perform better… if the content of the reading passage were related to their general field of study than If the reading passage were related to another subject was not answered conclusively” (pp. 11-12). She attributed this problem to the difficulty to assess background knowledge “a student who is in business studies may well have previously worked in another discipline such as science, or may well have scientific interests in his or her spare time” (Clapham, in Flowerdew and Peacock, 2001, p. 87). Clapham (1996) also explained the inconclusive results by the varying level of specificity of the subject-based texts.
Clapham (1996) investigated the effect of background knowledge on reading test performance on 842 non-native English speakers, most of whom were about to start undergraduate or post-graduate studies at English medium universities. The students were asked to complete two versions of the IELTS reading test, one in their own subject area and one outside it. This was meant to see whether the students had higher scores on the test in their own subject area. Findings revealed a significant interaction effect in each of the three groups (Business students, medicine students and engineering students).
Davis, Lange and Samuels (1988) conducted a study to assess the effects of an instruction on the organization of a journal article on the reading performance of forty intermediate-level college students of French. The participants were divided into two groups. While the first group was asked to read a scientific article in French with a canonical order; the second had to read the same article but with no canonical order. The texts were adapted from Science magazine for use in psychological experimentation. The participants were also asked to complete a free recall task in English. Results showed that non-native readers trained on the structure of scientific articles had significantly better recall than non-instructed readers. The authors, however, reported that this effect was only evident when the texts were read in normal and canonical order. Nevertheless, this study demonstrated that knowledge of structure can enhance comprehension of well-organised text for intermediate-level foreign-language readers.
Similarly, Dhieb-Henia (2003) found that metacognitive strategy training improved the 62 undergraduate biology students' familiarity with and proficiency in reading research articles. Prior to the experimental course, the students used the bottom-up approach as a strategy for reading, which made it difficult for them to cope with the reading tasks, the length of the article and the time allotted to finish the tasks. The participants mainly focused on the content of the texts and paid little attention to their metatextual features. They were blocked by the presence of difficult words; After the course, the participants adopted the top-down approach. Their reading protocols demonstrated that they used their background knowledge of the RA genre, namely the metatextual elements, to understand the RA and to overcome the difficulty of some highly technical words. There were few complaints about the length of the reading materials. She also stressed the effectiveness of retrospection as a method for evaluating the participants' reading behaviors.
While these studies have acknowledged that language proficieny, background knowledge and metacognitive strategies are important variables in reading comprehension, there is as yet little evidence to suggest the extent to which they are able to predict success. Widdowson (1979) explained this issue by the fact that scientific inquiry constitutes its own independent, cross-linguistic “secondary cultural system” (p. 51). One of the reasons for the inconclusive findings of the studies described above may have been that “for research of this sort it is difficult, if not impossible, to place students into appropriate subject areas” (Clapham, in Flowerdew and Peacock, 2001, p. 87). The students were allocated according to their academic field, but the fact that people study in one particular subject area does not, of course, mean that they are ignorant about other subjects and unfamiliar with other rhetorical styles, since they may well read books and articles in subjects outside their own academic field (Clapham, in Flowerdew and Peacock, 2001, p. 87). A further problem might have been that the reading passages varied according to their sources. For example, a passage might tend to be more specific if it came from an academic journal than if it came from a popularization such as the New Scientist, where scientific discoveries are rewritten for the educated layman. For example, in Clapham's study the business module contained the greatest variety of texts: a study document, a British government paper and a paper on career education. Attempts were made to overcome these lacunas in the present study: the participants were asked about their leisure reading in the background questionnaire and the four texts were chosen from academic professional journals. Swales (1990) stated that research articles are highly specific to their field of study. Because research articles include different sections with different communicative functions, the whole article was administered to the participants, and because even in one discipline, articles may vary according to the sub-discipline, the participants were given articles representative of each sub-discipline.
In the present chapter, I have first given an overview of different theoretical approaches to reading (the bottom-up approach, the top-down approach (including schema theory), the interactive view of reading (including the interactive-compensatory) and the componential approach).I have discussed how these approaches were initially developed in order to explain first language reading and I have then explained how they were later adopted in and adapted for foreign language reading research. I have referred to concepts and hypotheses which came into being due to the foreign language aspect: The linguistic threshold hypothesis and The linguistic interdependence hypothesis. The language-or-reading-problem question has been the subject of many investigations. Next, I have discussed reading in an academic context. I have referred to the role of academic and disciplinary culture and to definitions of academic literacy and academic language. Finally, I have described the findings of several empirical studies on FL reading and focused on studies that investigated particular components involved in reading, relevant to my study, namely language proficiency, metacognitive reading strategies and background knowledge. Although the issue of how these factors influence reading is not fully understood, there is an agreement in the aforementioned studies that they are significant contributors to the prediction of FL reading. It is worth to highlight at the end of this chapter, the small number of empirical studies conducted on research article genre EAP reading. Hence, the frequent call among science educators for more research on reading science (e.g. Shymansky et al., 1991). Most Studies examining the role of schemata in relation to academic reading performance have only examined texts considered traditional in nature, such as magazine articles, narratives or textbooks (e.g. Bea, Potter, and Clark, 1980; Carrell, 1984). They have employed similar methodologies in that participants read texts and then report aloud whatever information went through their heads while reading. The structures inherent in the texts (e.g., compare-contrast, problem-solving structures in expository text, and standard versus structurally interleaved versions of stories) were identified. Recalled information was analyzed for specific variables such as the number of propositions recalled, and the temporal sequence of story components. It is worth noting that language proficiency in some studies was not measured at all (Alderson & Urquhart, 1983, 1985). Therefore, the present study intends to fill in this gap.
Having recognized the complexity of EAP reading process, its multi-dimensional properties and its many interactive components, we have to move to a further interaction parallel to that between the reader and the text. This interaction takes place between the reader and the writer and is referred to in the literature as metadiscourse (Hyland, 2005; Crismore et al., 1993; Vande Kopple 1997). The reading activity is viewed in the present study as resembling a problem solving activity where metadiscourse is used in managing the reader's interaction with the text. Metadiscourse plays a crucial role in mediating the relationship between what writers intend to argue and their discourse communities (Camiciolotti, 2003).
It is interesting now to examine how this concept has been defined, studied and classified in the literature, how an awareness of its mechanism can be useful to readers and how this variable can contribute to reading comprehension especially in EAP settings.
Metadiscourse is a controversial term in current discourse analysis and language education "referring to an interesting, and relatively new, approach to conceptualizing interactions between text producers and their texts and between text producers and users" (Hyland, 2005: 1). It has been shown to be a typical feature of academic discourse, both written (Mauranen, 1993; Valero-Garces, 1996; Bunton, 1999) or spoken (Mauranen, 2002; Thompson, 2001).
In this section we set out to give a brief picture of metadiscourse and to describe why it has attracted the interest and attention of so many practitioners working in discourse analysis. We begin by providing a definition of the term, then review the empirical studies into metadiscourse use, look at the classification schemes analysts have proposed to identify metadiscourse, and finally outline the taxonomy to be adopted in the present study. We will show that there is also a "lack of systematicity" in classifying the features the concept is supposed to encompass; nevertheless the advantages of metadiscourse in research education are highly emphasised. In defining metadiscourse, the key notions of proposition, meaning, function and discourse community will be explored.
Metadiscourse: Towards A Definition
The concept of metadiscourse has been defined by a number of theorists. The term itself was initially introduced by Zellig Harris in 1959 who referred to ‘metadiscourse kernels', but there are earlier references to this phenomenon such as those found in the work of Volosinov (1930). Volosinov made reference to ‘metasemiotic discourse (that is, communication about another communication at another level)' (cited in Threadgold 1988: 343). Later definitions of metadiscourse reinforce these earlier distinctions between textual material which provides the reader with a primary message and material which functions to provide information about the message. Meyers (1980: 2), for example, defined metadiscourse as ‘information which does not add new content on a topic, but which gives emphasis to certain aspects of the semantic content or points out aspects of the structure of the content'. Williams (1981), although he initially defined metadiscourse in very general terms as a ‘level of structure in a sentence', proceeds to follow Meyer, specifying that metadiscourse is ‘discourse about discourse, words and phrases and clauses-even sentences- that refer not to the subject “out there” but to the act of discoursing, to the speech event that the discourse and its reader create' (p. 195). Vande Kopple (1985, p. 340), following Williams, defined metadiscourse as ‘discourse about discourse' but he provides alternatives by bringing together collocations such as ‘writing about writing' and ‘communication about the primary communication', all of which highlight the initial contrast made by Volosinov (1930). Crismore (1984) used the same definition as Williams, she defined metadicourse as ‘an author's discoursing about the discourse' (p. 279). She further qualified this definition by stating that ‘it is the author's intrusion into the discourse, either explicitly or non-explicitly, to direct rather inform the reader/listener (p. 279). In another paper, she stated that metadiscourse is ‘a rhetorical domain that regulates the communicative function of language' (1991, p. 191). Crismore et al. (1993, p. 40) restated the same definition and define metadiscourse as "linguistic material in texts, written or spoken, which does not add anything to the propositional content but that is intended to help the listener or reader organize, interpret and evaluate the information given". Vande Kopple (2002) maintained that there are two levels of meaning:
On one level we expand ideational material. On the levels of metadiscourse, we do not expand ideational material but help our readers connect, organise, interpret, evaluate, and develop attitudes towards that material (p. 93).
Generally, most of these definitions of metadiscourse seem to be based on the dichotomy between 'the propositional content' (the subject matter of a text) and the 'non-propositional content' (any linguistic material that does not add information to the propositional content of the text). There seems to be a high degree of consensus between these different theorists as to what constitutes metadiscourse. The latter is reported to refer to a range of discoursal features that go beyond the subject matter to convey the writer's conception of the target audience and the strategies he/she thinks most appropriate to convince them.
These definitions appear to be influenced by the traditional transitional-interactional approach to language (Brown and Yule, 1983; Jacobson, 1960). This approach regards the primary function of language as the transmission of information and communication of knowledge (the transactional function). The interactional function, which is used to express personal relations and attitudes, is relegated to a secondary position. Coates (1987, p. 113) warned, however, that "there has been a dangerous tendency among many linguists, philosophers and semanticists to concentrate on the referential function of language at the expense of all the others". Similarly, Hyland (2005, p. 9) asserted
What the 'informational' view of language almost completely ignores is that all discourse, no matter how explicitly 'informational', is created between participants who bring to the encounter certain affiliations, experiences, expectations and background understandings. These interpersonal dimensions influence how they will interpret and respond to the message and how they will engage in the interaction.
No separation should be therefore made between propositional and metadiscourse meanings. A statement can sometimes have a dual function. Hyland (2005) gave as an example the Socratic paradox 'I am lying', which simultaneously expresses a proposition and a commentary on it. Moreover, an omission of metadiscourse markers from a text can change its meaning (Hyland and Tse, 2004). Hyland (2005, p. 23) asserted "Vande Kopple and others are simply wrong to state that metadiscourse is a separate 'level of meaning'. Texts are communicative acts, not lists of propositions". Discourse studies, therefore, must see metadiscourse as part of a discourse community's practices and goals. A discourse community is made up of a group of individuals who have their own rules about what can be communicated and how it can be communicated (Swales,1990) .
Hyland (2005) adopted the functional approach to metadiscourse whereby an array of linguistic items can perform metadiscoursal roles depending on the rhetorical context in which they are used and the pragmatic function they fulfil. Metadiscourse markers can be of many types and adopt various forms. They can range from a simple word 'perhaps' to a full sentence (for example 'the next chapter deals with the problem of poverty'), several sentences or even a whole paragraph. Hyland (2005, p. 37) defined metadiscourse as
The cover term for the self-reflective expressions used to negotiate interactional meanings in a text, assisting the writer (or speaker) to express a viewpoint and engage with readers as members of a particular community.
As discussed above, metadiscourse has been understood in different ways. It is described as “a subject that is extremely dishevelled” (Sangster, 1987). Nash (1992, p. 100) maintained
The word 'metadiscourse' may have a reassuringly objective, 'scientific' ring, but its usage suggests boundaries of definition no more firmly drawn than those of, say, 'rhetoric' or 'style'. One reader may perceive a clear stylistic intention in something, which another reader dismisses as a commonplace, 'automatized' use of language.
This fuzziness may be attributed to the fact that metadiscourse may be realized by various linguistic forms and may fulfil a number of pragmatic functions in the text. Crismore (1989) maintained “metadiscourse is problematic because almost all properties of written discourse may implicitly or explicitly signal various dimensions of the communicative situation” (p. 49). Hyland (2005, p.18) argued "the lack of systematicity is partly a result of the heterogeneity of the features of spoken and written discourse which can signal the dimensions of context that metadiscourse refers to: the sender, the receiver or the organization of the message". Likewise, Swales (1990) maintained that ‘metadiscourse' can be realized through all kinds of linguistic elements ranging from punctuation and typographic marks (such as parentheses used for clarification or underlining for emphasis), to affixes or to whole sentences and clauses. In fact, it is more difficult to differentiate metadiscourse references from propositional discourse because distinctions should not only be made from a linguistic perspective but also from a functional one. The term functional in applied linguistics refers to how language works to achieve certain communicative purposes for users. Hyland (1999) asserted "features must be identified functionally… (as) the meaning of metadiscourse only becomes operative within a particular context” (p. 6). Likewise, Crismore (1989) stated “what is metadiscourse in one situation may be discourse in another” (p.49). Audience, purpose and situation are all variables that can affect the denotation of metadiscourse. Hyland (2005, p. 24) argued Metadiscourse is a relative concept in that text items only function as metadiscourse in relation to another part of the text. So what might be metadiscourse in one rhetorical context may be expressing propositional material in another.
Metadiscourse reflects the idea that language is essentially dynamic and that communication "is more than just the exchange of information, goods or services, but also involves the personalities, attitudes and assumptions of those who are communicating" (Hyland, 2005, p. 3). Discourse is an outcome of interaction, in which a text producer makes decisions about how to have an effect on the receiver. A writer is able not only to transform a "difficult" or "dry" text into coherent, "reader-friendly prose", but also to relate it to a given context and convey his or her personality, credibility, audience-sensitivity and relationship to the message (Hyland, 2000).
The concept of metadiscourse can vary according to what the author wants to emphasize. Some writers, for example, have restricted the term to features of rhetorical organisation by including only those text elements which refer to the text itself, such as 'we now turn to another topic' or 'this will be discussed in the next chapter', describing this as metatext or text reflexivity (Bunton, 1999; Mauranen, 1993; Valero-Garces, 1996). Enkvist (1975) pointed out that the primary function of ‘metatext' is to describe the text in which is located and to express the writer's presence. Later the same terminology is used by Mauranen (1993) as “text about the text itself” and “those elements in text which at least in their primary function go beyond the propositional context” (p.108). Bunton (1999) resorted to Nunan (1993)'s distinction between ‘discourse' and ‘text' to better elucidate his claim. He aptly argued that Just as ‘discourse' is often seen as a broader term for the written record of it (Brown and Yule, 1983; Nunan, 1993), ‘metadiscourse' should be seen as a broader term, including interpersonal and textual elements, and ‘metatext' as a narrower one with only textual elements (1999, p. 44):
Other researchers, for instance Brown and Yule (1983) deployed the term ‘metadiscourse' to refer to a set of adverbials in ‘which the speaker/ writer specifically comments on how what he is saying is to be taken' (p. 132). Similarly, Skelton (1988) employed the term ‘comments', but he included not only adverbials, but also features such as modal auxiliaries and ‘copulas other than be which indicate degrees of certainty and lexical verbs, such as those of believing, arguing, doubting' (100-101).
In the next part, we will look at different theorists' taxonomies of metadiscourse features. A taxonomy or a taxonomic scheme is defined by the Wikipedia dictionary as a particular classification arranged in a hierarchical structure. Typically, it is organised by subtype –supertype relationships. Classifying metadiscourse markers is a highly delicate and challenging task. Authors have the option to present their metadiscourse overtly through words, phrases or clauses, or covertly through subordination (Nagy, 1988). According to Crismore (1989), taxonomies “must show that all categories of metadiscourse are a matter of degree” (p. 60) and “must consider the overt and covert continuum and its planes”. Although taxonomies are open and not all inclusive, they can be models which can help researchers apply appropriate research methods to answer their questions. How successful researchers carry out their studies will partly depend on their use of the appropriate taxonomy.
Researchers seem not to agree on how to delimit metadiscourse. Two main traditions in the analysis of metadiscourse take a broad approach or a narrow approach, also known as the “integrative approach” and the “non-integrative approach” in Mauranen's (1993) terminology. In the broad approach, metadiscourse is seen as the means whereby the writer's presence in the discourse is made explicit, whether by displaying attitude towards or commenting on the text, or by showing how the text is organized. To take it from the reader's perspective, metadiscourse helps readers to organize, interpret, and evaluate the information in a text. This approach is quite popular but problematic as the concept of metadiscourse “tends to become all-inclusive” (Adel, 2006, p. 171).
The narrow (non-integrative) approach to metadiscourse, on the other hand, primarily investigates aspects of text organization, while largely excluding the interpersonal elements. There is also a terminological difference between the narrow and the broad approach; instead of “metadiscourse”, researchers in the narrow strand commonly adopt the term “metatext”, coined by the Finnish linguist Enkvist in 1975. Metatext is defined as “elements (…) whose function in the first place is to describe the text in which they are located” (Markkanen et al., 1993, p. 142). The tables below summarizes some of the research taking a broad and a narrow approach to metadiscourse, respectively.
In the present study, I adopted the broad approach to metadiscourse as it is more comprehensive and can help answer my research questions. Therefore, only the taxonomies within the broad approach of metadiscourse will be reviewed below. The breadth of meanings realized by metadisourse markers has resulted in different ways of categorizing them (Cheng and Steffensen, 1996; Crismore and Farnsworth, 1990; Crismore, Markkanen and Steffensen, 1993; Hyland and Tse, 2004; Hyland, 1998, 2005; Valero-Garces, 1996; Vande Kopple, 1985, 1997). In fact, a writer can deploy a multitude of resources to build, shape and organize his/her discourse.
Most of the classifications of metadiscourse have been inspired by the Hallidayan distinction between the textual and interpersonal macrofunctions of language (Halliday, 1973). Halliday's grammatical theory analyzes a text from three metafunctions: ideational, interpersonal and textual. The ideational or referential function represents the external world, including not only the representation of physical experiences and internal/mental processes, such as, thoughts and feelings (the experimental subfunction), but also the fundamental logical relations that exist among these experiences and processes (the logical subfunction). Logical relations are expressed linguistically through the syntactic devices of co-ordination, hypotaxis, indirect speech, etc. The interpersonal function encompasses the relations between the addressor and the addressee in a discourse situation or speech event. It includes "all that may be understood by the expression of our own personalities and personal feelings on the one hand, and forms of interaction and social interplay with other participants in the communication situation on the other hand" (Halliday, 1973, p. 66). Linguistically, the interpersonal function is realized through the use of first/second person pronouns and speech acts, such as questions and directives. The textual function is concerned with the way language establishes links with itself and the way to produce a text that is linguistically cohesive and semantically coherent. Halliday (1973) defined the textual function as "the component that enables the speaker to organise what he is saying in such a way that it makes sense in the context and fulfills its function as a message" (p. 66).
Many researchers are cautious, however, about the way Halliday (1985) separated these different functions (Crismore and Farnsworth, 1988; Schiffrin, 1980; Vande Kopple, 1985). They noted that grammatical and lexical elements can convey the three functions simultaneously. Crismore (1989) stated
Because metadiscourse may be used simultaneously for different roles or functions, it is impossible to identify types of metadiscourse that function only interpersonally or textually and to separate metadiscourse and ideational content with confidence at all times
Pioneering work on metadiscourse can be traced back to the eighties and early nineties. Vande Kopple's (1985a) and Crismore et al.'s (1993) taxonomies have been benchmarks in the field. Indeed, most subsequent classifications (Crismore and Farnsworth, 1989 a/b; Steffensen and Markannen, 1993; Hyland, 1993, 1999) have either shadowed or been inspired by these taxonomies in particular. Given their importance, I am going to review their taxonomies in detail, then move to "the most substantial revisions" of these taxonomies by Hyland (1999, 2005).
Vande Kopple's (1985, 1997) Taxonomy
Vande Kopple (1985), expanding upon Lautamatti's (1978) and Williams' (1981) classification of metadiscourse types, developed what is reported to be the first theoretically sound taxonomy. Indeed, many writers have used his taxonomy such as Crismore and Farnsworth (1989, 1990), Intaraprawat and Steffenson (1995), Cheng and Steffensen (1996) and Hyland (1999). Vande Kopple (1985) offered a classification system for metadiscourse, which he later defined as “discourse that people use not to expand referential material but to help their readers connect, organise, interpret, evaluate, and develop attitudes towards that material” (Vande Kopple, 1997, p. 1). Vande Kopple's (1985) classification of metadiscourse consists of seven kinds of metadiscourse markers divided into textual and interpersonal types (see table 1). What is common to the seven categories is that they are non-propositional. “They do not expand the propositional information of the text. They do not make claims about states of affairs in the world that can be either true or false'' (Vande Kopple, 1985, p. 85). It is noteworthy that he identifies the multifunctionality of some elements and introduces a seventh type, Commentary, which is restricted to that metadiscourse the writer uses to directly address readers and invite them to implicit dialogue.
Vande Kopple's (1985) taxonomy has been criticised and revised by many writers because of the vagueness of the categories and the functional overlaps (Crismore et al., 1993; Hyland, 1998, 1999, 2005; Nash, 1992; Vande Kopple, 1997, 2002). Vande Kopple (1985) warned against these problems when presenting his taxonomy stating “the boundaries and internal characteristics of these kinds would probably have to be more closely surveyed in future work” (p. 8”). He attributed this overlap to the multifunctionality of metadiscourse functions, “sometimes one form can fulfill a metadiscoursal function in one place and a referential function in another” (Vande Kopple, 1997, p. 1).
Hyland (2005) gave two instances of these overlaps. First, he pointed to the difficulty of distinguishing the category “narrators” and the subcategory “attributors” in academic writing. He argued that citations, in addition to the fact that “they provide propositional warrants (…) and meet conventions of precedence, (…) might also be used to offer a narrative context for the research or establish an intertextual framework to suggest a cumulative and linear progression of knowledge” (Hyland, 2005, p. 33). Second, he showed that the boundaries between the categories of commentary and of attitude markers are not obvious.
Vande Kopple (1997) tried to refine his taxonomy. In his revised classification, he “left essentially unchanged” some categories such as text connectives and code glosses, deleted the categories of validity markers and narrators and added a new category, epistemological markers. He (1997) justified this new introduction:
Since 1985 I have come to see that various kinds of metadiscourse are linked in the overarching function of indicating some stance on the part of the writer toward the epistemological status of the referential material conveyed (p. 4).
He included in the new category two subcategories: modality markers and evidentials. By modality markers he refers to the linguistic items that express how committed the author is to the truth of that material. Evidentials encompass the kinds of evidence the author has for his/her text (Anderson, 1986).
Vande Kopple (1997) also changed some of his terminology replacing the term “hedge” that he had used in his early work with the term “shields”, which he adopted from Prince's, Frader's and Bosk's (1982) terminology. He justified this change on the grounds that hedges “make things fuzzy in one of (at least two distinctly different ways” (Vande Kopple, 1997, p.5).
In the new taxonomy, Vande Kopple kept the textual/ interpersonal division. Within the interpersonal cluster, he included illocution markers, epistemology markers, attitude markers, and bits of commentary.
Despite these changes, Vande Kopple (1997) still qualified his taxonomy as tentative. He recognized that his taxonomy is far from being comprehensive. Alternative divisions of the categories could be made and further categories could be added. He stated
I have tentatively assigned each kind of metadiscourse to either the interpersonal or the textual categories…Perhaps the kind of categorization that will emerge in future research will find ways to show overlaps between textual and interpersonal functions of language (p. 9)
Crismore, Markkanen And Steffensen's (1993) Taxonomy
Crismore et al. (1993) defined metadiscourse as
The linguistic material in texts, whether spoken or written, that does not add anything to the propositional content but that is intended to help the listener or reader organize, interpret, and evaluate the information given (p. 40).
Crismore et al. (1993) used Vande Kopple's (1985) system as a basis for classifying metadiscourse, but modified it. Crismore examined social studies writing to determine the types and amounts of metadiscourse in textbooks and nonschool books. Although they kept the textual and interpersonal division, they "collapsed, separated, and reorganized his subcategories" (see table 2). For example, they "dropped the subcategories of narrators and temporal connectives" (1993, p. 46) and created two new ones: textual and interpretative markers. Text markers replace "the more limiting text connectives" and include sequencers, reminders, logical connectives, and topicalizers. As for the second new category, it helps "readers interpret and better understand the writer's meaning and writing strategies" (Crismore et al., 1993, p. 47) and encompasses code glosses, announcements, and illocution markers. For interpersonal metadiscourse, they "separated validity markers into three separate subcategories: hedges, certainty markers, and attributors and kept attitude markers and commentary as subcategories (Crismore et al., 1993, p. 46).
As can be seen in the table above, Crismore et al. (1993) followed the functional approach in their classification of metadiscourse:
Our decisions were based on what appeared to be the primary function of the item in a particular context (...) Thus our analysis is not strictly a linguistic analysis but rather a functional analysis (1993, p. 47-48)
They were cautious, however, and admitted that "metadiscourse is (a) messy" concept (p. 54). Thus, they "limited (their) study to the more explicit instances of metadiscourse, realizing that the rich complexity of language also leads to more subtle uses of metadiscourse" (p. 48). This may explain why the taxonomy has been criticised by a number of linguists such as Hyland and Tse (2004) and Hyland (2005). Hyland (2005) found, for instance, the textual/interpretive division unclear as logical connectives, sequencers, reminders and topicalizers also "assist the reader in interpreting" the text (p. 33). Furthermore, he pointed o to a "confusion within these categories". Although reminders and announcements refer to ideas mentioned earlier or later in the text, they are included in two different subcategories, textual markers and interpretive markers, respectively. Besides, he sees that the subcategory of logical connectives is not "entirely transparent" as the items it includes are identified "syntactically rather than functionally". Crismore et al. (1993) seemed to choose only the items that "are the product of choice rather than syntactic necessity" (Hyland, 2005, p. 34). Thus, they counted coordinating conjunctions (such as and and but) and conjunctive adverbs (therefore, in addition) as metadiscourse, but not subordinating conjunctions (like because and which) because the latter according to them affect grammaticality when omitted. Finally, Hyland (2005) highlights a paradox in Crismore et al.'s definition of metadiscoure. In fact, while they identified metadiscourse as the linguistic items that do not add anything to the propositional content of the text, they considered logical connectives as metadiscourse. He argued "it seems clear, however, that conjunctions operating to link elements of proposition might justifiably be seen as part of these propositions" (p. 34-35). He describes as “muddying the water” the attempt to establish clear boundaries for metadiscourse by referring to the items that go beyond the propositional matter according to syntactic criteria. However, Crismore (1990) did recognise the problem of imprecision in her definition of metadiscourse, she commented “we cannot be precise and completely systematic about metadiscourse. Because the concept of metadiscourse involves several confusing elements, it is difficult to be very precise and explicit about it theoretically' (49).
Crismore at al's taxonomy is not applicable for my dissertation since it was compiled with reference to a different genre than that used in the present study, namely the textbook genre. In textbooks, there is relatively little saliency, fewer hedges and less evaluative metadiscourse because authors appear as authorities and truth-givers (Aguilar, 2009).
Hyland (1999) created a modified version of Crismore et al.'s (1993) taxonomy in an attempt to overcome the problems reported above.
Hyland's (1999) Taxonomy
Hyland viewed metadiscourse as "a central pragmatic construct" that helps establish a social and communicative engagement between writers and readers. Furthermore, he maintained that it is a crucial rhetorical device for writers as it "allows them to engage and influence readers in ways that conform to a discipline's norms, values and ideology, expressing textual and interpersonal meanings that their audience is likely to accept as credible and convincing" (Hyland, 1999, p. 6). This socio-pragmatic view implies that in choosing a particular metadiscourse, a writer has to consider three variables: audience, purpose and situation.
Hyland (1999) developed a metadiscourse taxonomy based on Crismore et al. (1993). He distinguished between textual metadiscourse –that refers to the devices that "represent the audience's presence in the text in terms of the writer's assessment of its processing difficulties, intertextual requirements and need for interpretative guidance" (1999, p. 7) - and interpersonal metadiscourse –which “allows writers to express a perspective towards their propositional information and their readers” (1999, p. 8). The second category is considered as “essentially interactional and evaluative” (Hyland, 1998, p. 443). It "relates to the level of personality, or tenor, of the discourse and influences such matters as the author's intimacy and remoteness, expression of attitude, commitment to propositions and degree of reader involvement"(1999, p. 8).
Hyland (1999), however, reshuffled the subfunctions of Crismore et al.'s (1993) taxonomy (Table 5). In fact, within the textual metadiscourse category, he shifted the sub-classes sequencers, topicalizers and illocution markers to a new subcategory he named frame markers, "which signal boundaries in the discourse or stages in the argument" (p. 7). Furthermore, he moved reminders and announcements into another new subcategory he labelled endophoric markers "which refer to other parts of the text." Finally, he added a new category, evidential markers, which replace attributors and "indicate the source of textual material". As for interpersonal metadiscourse, while he kept the functions hedges and attitude markers, he introduced three new functions: emphatics (which "indicate(s) the degree of commitment, certainty and collegial deference a writer wishes to convey"), relational markers (which includes "devices that explicitly address readers, either to focus their attention or include them as discourse participants") and person markers (which "refers to the degree of author presence in the text").
Although Hyland (1999) described his taxonomy as a "useful means of revealing the meanings available in the text", he draws attention to the "great deal of pragmatic overlap between the categories" (p. 8). When describing his taxonomy, he admitted "attitude and relational markers are often difficult to distinguish in practice"(italisization mine) (p. 8). He also put forward some examples of the difficulties he encountered when he designed his taxonomy:
Connectives, for example, principally link textual material but can also solicit reader collusion when presenting claims (Barton 1995); hedges have both epistemic and affective roles, indicating either uncertainty or deference to disciplinary norms of appropriate interpersonal stance; code glosses both supply necessary information and imply a position of superior knowledge to the reader. (p. 8).
Ifantidou (2005) also questioned the applicability of this typology. She found that the central boundary between ‘textual' and ‘interpersonal' metadiscourse appeared to be fuzzy and the two definitions seemed to overlap. She stated
It is not clear how the textual/interpersonal (or interactive/interactional) distinction is really being drawn. When it comes to assigning individual lexical items to particular subcategories of these two basic categories, things become even more complicated. For example, in Hyland's taxonomy, ‘evidentials' (Z states, according to X), ‘emphatics'/‘boosters' (in fact, definitely), and ‘hedges' (might, perhaps) all in fact refer to or express the writer's/speaker's degree of commitment to the propositional content; so why multiply categories if just one, e.g., ‘evidentials', would do? Similarly, how do ‘person markers'/‘self-mentions' (we, they . . . – ‘‘explicit reference to authors'') and ‘evidentials' (according to X, Z states . . . – ‘‘refer to source of information from other texts'') really differ – or do they need to differ, since both indicate reference to writers or other ‘authors'? (p.1328)
Hyland (1999) attributed these problems to the fact that language is polypragmatic and that "writers frequently seek to achieve several concurrent purposes, appealing to readers on both affective and logical levels simultaneously"(p. 8). Hence his assertion that his taxonomy "can (...) only approximate the complexity and fluidity of natural language use"(p. 8). As my thesis investigates the RA genre, I decided not to adopt Hyland's (1999) taxonomy since it was the outcome of work on university introductory textbooks and the role these textbooks have in students' acquisition of a specialised disciplinary literacy. Hyland (1999) attempted to illustrate that these text types employ similar discourse strategies to create a credible and trustworthy narrative voice.
Hyland (2005) presented a new model which he thought could be a better representation of metadiscourse features. He asserted that his new framework is a comprehensive and pragmatically oriented means of studying interpersonal features in academic articles. He (p.37) asserted "I propose a more theoretically robust and analytically reliable model of metadiscourse, based on a number of core principles and offering clear criteria for identifying and coding features".
Hyland's (2005) taxonomy was adopted in the present study for data analysis as it seemed to offer a more holistic view of metadiscourse. Furthermore, it is the most recent praised taxonomy in the literature. In fact, Rodina (2007) maintained that “his model serves as a model for metadiscourse research in general” (p. 481). Likewise, Paltridge (2007) stated “it (Hyland's (2005) taxonomy) sifts out conflictions in previous work on metadiscourse and adds clarity to our understanding of how metadiscourse works in written texts” (p. 227). Similarly, Aguilar (2009) wrote “Ken Hyland is possibly the metadiscourse researcher par excellence in contemporary research” (p. 86). Hyland has analysed the pragmatics of academic metadiscourse without neglecting the significance of the wider context in which such metadiscourse features occur- the academic community. His extensive and intensive work on (mostly written) academic metadiscourse encompasses citation, authorial stance, hedging, disciplinary and genre differences and, above all, the interpersonal role of metadiscourse.
A further reason that has motivated my choice was the fact that Hyland's (2005) taxonomy was, in great part, drawn from the academic genre investigated in the present study. This is highly important because a good knowledge and awareness of the community and of the discipline and genre conventions are usually reflected in the use of metadiscourse (Aguilar, 2009). Hyland's (2005) taxonomy is displayed below.
Hyland's (2005) Taxonomy
Hyland's (2005) metadiscourse taxonomy is based on the view that metadiscourse becomes operative only within a particular social context and within the norms of a specific discourse community. The same language items can perform both metadiscoursal and non-metadiscoursal roles depending on the discourse context, the discourse community, the writer's purpose and the audience. Hyland (2005, p. 37) argued
The key assumption here is that rhetorical features can be understood and seen as meaningful only in the contexts where they occur, and as a result metadiscourse must be analysed as part of a community's practices, values and ideals.
Constructing an audience is an equally important prerequisite, according to Hyland (2005), in the choice of metadiscourse. The notion of audience is reported to be “notoriously elusive”. While some analysts consider the audience as real people whose expectations must be taken into consideration by the writer, others view them as fictious (Kirsch and Roen, 1990; Park, 1986). Grabe and Kaplan (1996, p. 207-11) pointed to five main parameters of audience that influence writing:
1. The number of readers- whether a text is written for oneself, a single person, a small group or a large heterogeneous group will have an impact on linguistic and rhetorical choices.
2. Whether readers are known or unknown- the degree of looseness of the reader is likely to influence the extent of the interactional involvement features in the text.
3. The relative status of participants- metadiscourse choices will also vary depending on whether the writer has an equal or lower status than the reader. In spoken contexts it seems that equal status creates more interactional negotiation.
4. The extent of shared background knowledge- writers are likely to be more explicit in their use of metadiscourse where they assume the reader's lack of cultural, institutional or social familiarity with issues.
5. The extent to which specific topical knowledge is shared_how far writers can rely on readers knowing about the topic will influence not only the amount of detail that can be included, but also the elaboration of ideas and assumptions through code glosses, the amount of evidential support required, he frequency of explicit transitions, and so on.
Hyland (2005) shared Grabe's and Kaplan's views and maintains that relative to its purpose and setting, each piece of writing (professional or academic) can either have a single audience or heterogeneous audiences. Thus, writers have to consider the communicative context of their texts before constructing their audience(s). This will help them choose among the various linguistic items, the appropriate metadiscourse markers that would enable them to create a “convincing reader-environment”. Indeed, the writer's awareness of variables such as the social status of both the writer and the potential reader, the number of intended readers, the degree of the readers' familiarity with the subject matter and the shared cultural or community norms, would enhance the interactivity of the text. According to Hyland, interaction
involves the writer's sense of his or her personal relationship with readers. In deciding whether to establish an equal or hierarchical affiliation, adopt an involved or remote stance, or choose a convivial or indifferent interpersonal tenor, we are at least partly constrained by the dominant ideologies of our institutions. (2005, p. 12)
This view has been confirmed by comparative studies of metadiscourse which highlight the variability of metadiscourse across cultures and genres (Clyne, 1981; Holmes, 1982; Intaraprwat, 1988).
Hyland (2005) presented three key principles of metadiscourse. These are (p. 38):
1. Metadiscourse is distinct from propositional aspects of discourse. Despite this dinstinction, however, there is no hierarchical relation between them; nor is metadiscourse secondary in importance. In fact, Metadiscourse which organizes the propositional content and conveys the writer's stance towards it and the reader and the propositional content ('the communicative content' of discourse) interact simultaneously in the text. Hyland (2005, p. 38) writes "we can see these two dimensions as two simultaneous aspects of language in use, referring to two main types of entity: things in the world and things in the discourse".
2. Metadiscourse refers to aspects of the text that embody writer-reader interactions; in other words, all metadiscourse categories are essentially interpersonal since they need to take into account "the readers' knowledge, textual experience and processing needs" (Hyland and Tse, 2004). He argues that textual markers can play both a textual and an interpersonal function depending on the writer's orientation and goal. He maintains (2005, p. 41) "like other features of textual metadiscourse, the transitions and links that conjunctions mark between clauses can be oriented towards either the experiential or the interpersonal, to either propositional or interactional meanings". Thus, instead of the traditional textual and interpersonal division, Hyland and Tse (2004) and Hyland (2005) adopted Thompson's (2001) terminology and renamed their categories interactive and interactional, respectively. Interactive markers are used by the writer to shape and organize his/her text in such a way as to become "reader-friendly". Interactional markers, on the other hand, concern the writer's comment and evaluation of his/her material. Thompson (2001, p. 61) points out that these two functions "are essentially the two sides of the same coin". He gives the example of a directive which in addition to its interactional function, serves an interactive function by signalling where the text is going next.
3. Metadiscourse refers only to relations which are internal to the discourse. Hyland (2005) highlights the dual role of references. He argues that references can perform two roles: internal and external. The former concerns internal reference and set up relations between parts of the discourse and help in the unfolding of the text. In contrast, the second concern external reference and involve reporting and signaling relations between events in the world. It is at the internal level that metadiscourse is used.
Hyland (2005) summarised and commented on his three principles
There are (…) good reasons for distinguishing metadiscourse from the propositional content of a text and for seeing it more broadly as encompassing the interactional aspects of discourse, using the criteria of external and internal relations.
The classification scheme proposed by Hyland (2005) embodies these principles (see table 6).
Hyland (2005) adopted the functional approach when designing his taxonomy. He viewed metadiscourse as the tool that allows writers to signal their presence and to express their stance towards their propositional content and their readers. The taxonomy also "acknowledges the contextual specificity of metadiscourse" (p. 48). Hyland's scheme is inspired by early work. In fact, it employs Thompson and Thetela's (1995) distinction between interactive and interactional resources, includes both the stance and engagement features introduced in Hyland (2001), and builds on Hyland's previous metadiscourse taxonomies (1998 and 2000).
The taxonomy is based on the two dimensions of interaction: the interactive and the interactional. These two functions encompass more specific functions, which will be outlined below.
Based on their awareness of their audience's expectations and knowledge, writers employ interactive resources to arrange their propositional content in ways that make it appear persuasive and coherent. This category includes five sub-categories:
a)Transition markers: These are mainly conjunctions and adverbial phrases that have internal reference and convey pragmatic connections between stages in an argument. They express addition, causation, comparison and contrast. Examples of this category are furthermore, moreover, similarly, likewise, equally, thus, therefore, consequently, etc.
b)Frame markers: They "signal text boundaries or elements of schematic text structure" (Hyland, p. 51). They serve to sequence parts of the text (first, then, at the same time), to explicitly label text stages (to sum up, by way of introduction), to announce discourse goals (my purpose is, I argue that, I hope to persuade, there are several reasons why), and to indicate topic shifts (well, right, OK, etc).
c)Endophoric markers: These guide the reader through the discussion to orientate him/her towards a particular interpretation of the discourse (refer to the next section, as noted above).
d)Evidentials: they are “metalinguistic representations of an idea from another source” (Thomas and Hhawes, 1994, p., 129). In academic writing, they draw on a community-based literature to support the writer's claims.
e)Code glosses: They are deployed to signal a restatement of an argument for the purposes of explanation, exemplification and elaboration (in other words, this can be defined as, for example).
They “alert” the readers to the writer's stance towards both the propositional content and the readers themselves and reveal the way he/she engages with the socially determined positions of others They “act to anticipate, acknowledge, challenge or suppress alternative, potentially divergent postions” (Hyland, 2005, p. 52). This category encompasses five sub-functions
a)s: these devices indicate that a statement is an opinion and not a Hedgefact, and thus leaves it open to discussion. They show the degree of precision and reliability the author wants to emphasise (possible, might, perhaps, etc).
b)Boosters: They “allow writers to close down alternatives, head off conflicting views and express their certainty in what they say” (Hyland, 2005, p. 52). Therefore, they help boost an argument by highlighting the necessity of mutual experiences to draw the same conclusions as the writer. Examples of such devices are clearly, obviously and demonstrate.
c)Attitude markers: They express the writer's emotional attitude to propositions. They reflect surprise, agreement, importance, obligation, frustration and so on. They are signaled by attitude verbs (agree, prefer, etc.), sentence adverbs (unfortunately, hopefully, etc.) and adjectives (appropriate, logical, remarkable, etc.)
d)Self-mention markers: They convey the writer's presence in a text and his position with regard the propositional content, the discourse community and the target audience. These are expressed by the frequency of first-person pronouns and possessive adjectives (I, me, mine, exclusive we, our, ours).
e)Engagement markers: These are meant to involve readers in the discussion, as discourse participants. “In addition to creating an impression of authority, integrity and credibility through choices of hedges, boosters, self-mention and attitude, writers are able to either highlight or downplay the presence of their readers in the text” (Hyland, 2005, p. 53). These devices are important because they secure social and rhetorical objectives.
In the previous sections I have examined how researchers have approached metadiscourse. I have demonstrated that there has been a certain amount of confusion surrounding the term and imprecision in defining it. Furthermore, there has been no consensus about the categorization schemes that have been proposed. While most analysts adopted a functional approach to metadiscourse, taxonomies sometimes confused functional with syntactic criteria. Hyland and Tse (2004) warned “the failure to pin the term down precisely has meant that analysts have been unable to confidently operationalize the concept in real texts, making analysis an elusive and frustrating experience” (p. 156). This may be attributed to the fact that language features have potentially multifunctional roles and that items only function as metadiscourse in relation to a particular context and to other parts of the text. Hyland and Tse (2004) stated “not only is (metadiscourse) an open category to which new items can be added to fit the writer's needs, but the same items can function as metadiscourse in some parts of the text and not in others” (p. 158).
Despite this confusion, there seems to be an agreement about certain features of metadiscourse. The areas of agreement can be summarized as follows:
1. Metadiscourse exists and its benefits are evidenced in different disciplines such as EAP. Crismore (1989) notes “metadiscourse is part of natural spoken and written language and that it has been used in different genres and disciplines throughout the centuries” (p. 237).
2. Metadiscourse requires a methodology of study appropriate to its elusive nature. Several researchers have pointed out the lack of a firm theoretical basis for metadiscourse (Van Dijk, 1984; Beauvais, 1986; Hyland, 2000, 2005).
3. Metadiscourse has many functions: it frames the propositional content, helps readers situate the text in a particular socio-cultural context, ensures the coherence and cohesiveness of discourse, signals rhetorical moves within a discourse, indicates the author's stance towards his/her text, and boosts reader/ writer interaction.
4. Metadiscourse fulfills the interactional function of language.
5. Metadiscourse dimensions are “a set of continuums”. The boundaries between its categories are fuzzy. There is not always a clear separation between metadiscourse and propositional content.
6. Metadiscourse is considered necessary and equal in importance to the “primary” discourse in communication.
In recent years, there has been a great deal of work on metadiscourse from a variety of perspectives. Metadiscourse has been investigated in students' coursework in writing and composition (Williams, 1981; Vande Kopple, 1985), mechanical engineering proposals (Longo, 1994), advertisements (Fuertes-Olivera et al., 2001), business letters (Vergaro, 2002), editorials (Le, 2004), research articles (Swales, 1990; Hyland, 1998), non-native PhD medical students' dissertations (Bunton, 1999), university textbooks (Hyland, 1999), L2 lecture comprehension (Thompson, 2003), EAP textbooks (Moreno, 2003) and informal conversations (Craig, 2000). Metadiscourse has been also investigated with respect to intercultural variation (Mauranen, 1993; Valero-Garces, 1996). However, since the present study focuses on EFL readers' use of metadiscourse when reading RAs in English in their field, only the studies relevant to my research questions will be reviewed below.This will provide me with information on which I can base my own research, or with insights regarding the appropriateness of various types of research method which I could adopt and adapt for my own study.
Studies On Metadiscourse
Studies On Metadiscourse And The Writing Skill
A good number of studies have highlighted the ways authors, via the use of metadiscourse, make their discourses more "reader-friendly" in an attempt to engage their readers . Hyland (2001) examined how disciplinary communities constructed readers in published research articles. His corpus comprised 240 published research articles, three from each of ten leading journals in eight disciplines in the fields of mechanical engineering, electrical engineering, marketing, philosophy, sociology, applied linguistics, physics, and microbiology. He found that in the “soft” fields of the humanities and social sciences, writers used more interactive features (questions, inclusive first person, indefinite, and second person pronouns and items referring to readers, directives, including imperatives, obligation modals referring to actions of the reader, and adjectival predicates controlling a complement to-clause, directing readers to a particular action, references to shared knowledge, and asides addressed to the reader, marked off from the ongoing flow of text) to bring their readers into the discourse “as a participating equal” (Hyland, 2004, p. 222).
Abdi (2002) also found the same results when analysing the discussion section of 55 academic research articles from the social sciences (SS) and natural sciences (NS) to investigate the way writers used interpersonal metadiscourse to partly reveal their identity. The analysis showed that the SS writers employed interpersonal metadiscourse (hedges', ‘emphatics' and ‘attitude markers') more frequently than the NS writers.
Likewise, Chu and Yu (2003) provided counter evidence for the myth that professional scientific writing should be impersonal. After examining sixty Editor's letters from MISQ, ISR and CACM (collected for the years 1990, 1995 and 2000), they found out that editors used interpersonal devices more frequently than textual devices to build their credibility and affective appeal.
In the same vein, Hyland (2004), after analysing 40 Masters and doctoral dissertations written by Hong Kong Chinese students from six academic disciplines (Electronic Engineering, Computer Science, Business Studies, Biology, Applied Linguistics, and Public Administration), demonstrated that the students in social science disciplines ‘employed the more metadiscourse overall (56% of the normed count). They tended to use more interpersonal metadiscourse to offer a credible representation of themselves and their work, to negotiate social relations with their audience and to gain a more sympathetic hearing for their own findings. Hyland (2004, p. 332) accounts for this finding by the fact that “these fields all deal with human participants and rely on qualitative analyses or statistical probabilities to construct and represent knowledge. For these reasons, they require elaborate exposition and considerable tentativeness in expressing claims”.
The argument from the review above is that readers should be aware of the efforts made by writers to promote understanding and acceptance and to produce a desired effect, depending on their underlying purposes and perception of readers' expectations. Readers as active participants, according to the componential interactive approach to reading, have the power to refute writers' claims. Below I will review the studies that investigated the extent readers use metadiscourse when attempting to comprehend written discourses.
Metadiscourse Studies On Reading Comprehension
Research on metadiscourse has been available for the last thirty years but research into the correlation between metadiscourse and reading comprehsion is still in its infancy. A good number of the prior studies have been conducted with native speakers.
The Case Of Native Readers Of English
One of the first studies on the interaction between metadiscourse and reading comprehension can be found in Vande Kopple's (1985b). Vande Kopple prepared two expository 4-paragraph texts, one with validity markers distributed throughout the text (beginning, middle, and end positions) and another one where metadiscourse was added to paragraphs from which the first sentence had been excised. Based on his definition of metadiscourse- discourse about discourse that does not convey propositional information- he hypothesized that nontopical participants and adjuncts that do not contribute information to the propositional content will not be remembered, and will have either negative or no effects on readers' memory for the propositional discourse. His results showed that validity marker clauses were recalled far less well than propositional material and had a detrimental effect on the recall of propositional discourse. In spite of not discarding the possibility that validity markers had an effect that was not revealed in the immediate recall tests, he suggested that a more accurate model of the discourse processes was called for. This study is important as it suggested that the order and/or place of metadiscourse may affect the recall process. It paved the way for future research.
Positive correlation was also found byHaas and Flower (1988). They looked at how readers comprehend text by focusing on what they call "rhetorical reading" and found that more accomplished readers constructed meaning from the text by representing "the author's intent, the context, and how other readers might respond" (p. 181) using "cues in the text" (p. 176). While Haas and Flower did not call these cues metadiscourse, they described the functions of metadiscourse in explaining how readers used these cues to construct representations of the author and context.
O'Keefe (1988), on the other hand, found no significant effects on comprehension for the textual, informational metadiscourse (logical connectors) or for the attitudinal metadiscourse. The results of a questionnaire designed to measure students' attitudes revealed that they were positive for informational than for the attitudinal metadiscourse. This limited finding could be attributed to the fact that the author used a small scale study and short reading passages adapted from periodicals, business case studies, or student-written essays.
Similarly, Crismore (1989) reported that metadiscourse did not have larger effects in her study. She conducted a study to see whether including informational and attitudinal metadiscourse in passages of social studies textbooks would influence reading retention (among other factors) with sixth graders. She found that there was some improvement in retention after reading passages with both types of metadiscourse, but only with certain participant subgroups. She also reported that while the informational metadiscourse did seem to have some beneficial effect, it was not great, and it was not necessarily limited to the information on which the metadiscourse focused. Furthermore, the students seemed, in general, to react neither negatively nor positively to the informational author commentary or to the added length.
A further interesting finding was reported by Crismore and Vande Kopple (1997) when investigating the effects of hedging on reading retention. They found that it was the interactional nature of the hedges that made them memorable, not the fact that the writer distanced themselves from claims, or qualified them. They conducted an experimental study and had the experimental group read passages from both social studies and science textbooks containing various types of hedging and the control group read corresponding passages in which all hedging had been removed. Even if other factors (e.g., gender, subject matter and type/position of hedging) were also found to have an influence on retention scores, the authors reported that, in general, students learned more from reading science and social studies passages with hedging included. They concluded that low-intensity hedges, with first-person pronouns used in most controversial places (e.g. where a claim is made) were positively interpreted, connecting propositional content to the writer.
These inconsistent findings showed that the interaction between the use of metadiscourse and L1 reading comprehension is not straightforward. In SL instructional contexts, it has been posited that this interaction is still more complex (e.g., Camiciottoli, 2003). However, this area of investigation is under -researched.
The Case Of Non-Native Readers Of English
Studies with non-native speakers have yielded inconclusive results. Mustapha and Premalatha (2001) investigated the effects of metadiscourse on Malaysian university students' understanding and critical reading of texts. Two expository passages written in English were manipulated so that the author's metadiscourse was deleted. Two groups of Malaysian second-year undergraduates then read the manipulated or unmanipulated versions of each passage (about 750 words) and answered comprehension questions and critical reading questions. The students' language proficiency was measured against a standard English language examination. The text was chosen as it was not biased to any cultural group taking part in the research project and the text content was also general in nature. The results showed that participants reading the non-doctored text tended to fare better than those reading the doctored text. It is interesting to note, however, that the participants said when interviewed that the author's intrusion into the reading passage sometimes irritated them more than helped them comprehend text. The authors (2001, p. 37) explained “this “irritation” could only mean that the presence of metadiscourse - or author's intrusion, impinges on their text processing capacity, creating certain memory overload which would then have a negative effect on their text comprehension”.
Similar results were found in a study of Camiciottoli (2003). The author explored the effects of metadiscourse on ESP reading comprehension with 55 Italian university students. Two groups of students read selected extracts from two versions of the same text differing according to quantity and type of metadiscourse. Each group then took a reading comprehension test and their mean scores were compared. The findings suggested that a more pronounced use of metadiscourse might be associated with improved comprehension in some cases. Yet, a post-reading questionnaire showed that the students had substantially no awareness of metadiscourse.
Parvaresh and Nemati (2008) replicated the study of Camiciottoli (2003) to shed light on the effects of metadiscourse markers on the comprehension of English texts by Iranian EFL learners and then compared the results achieved with those in their native language (Persian). They attempted to measure the participants' awareness of those markers and their interaction with those texts in both languages by using a follow-up questionnaire. The study included the proficiency level of the learners, measured by Nelson English Proficiency Test, as a moderator variable. Based on an original English text, adapted from an article written published in Language Magazine:, a set of 11 True/False questions was developed and used once with that text and once with its doctored version in which metadiscourse markers had been removed. In metadiscourse identification process, Hyland's (2005) interpersonal model of metadiscourse was utilized. The texts and questions were also translated into Persian and used for a Persian reading comprehension test. The analyses showed that the participants performed significantly better on the un-doctored texts although they had read them first, regardless of whether the texts were in their L1 or in their SL. The results revealed that for SL it was the lower proficiency learners who benefited more from the presence of metadiscourse markers. The results of the follow-up questionnaire, adapted from the one used by Camiciottoli (2003), also revealed that the difficulty of a text in its general sense had nothing to do with the presence or absence of markers.
Although the findings of the studies discussed above do not provide clear evidence for the fact that the presence of metadiscourse in a text improves comprehension, they do suggest that it has a facilitating role and is an effective technique for enhancing critical reading. There are, however, some limitations to these studies. First, these studies have overwhelmingly relied on expository adapted texts, often concerning general themes. Moreover, the extracts were short, ranging from 120 to 700 words. This length does not allow for a large use of metadiscourse. Swales and Feak (2000) maintained that longer texts and complex material have more metadiscourse than shorter texts and simpler material. Besides, most of the studies used the same methodology: doctored vs non-doctored texts, which might affect the natural process of reading. Indeed, by modifying the original text, the researchers might have changed the communicative effect and function of the discourse. In addition, this methodology falls short because the researchers assumed that the participants read better doctored texts because of the presence of metadiscourse markers without any tests to measure the presence or absence of any gain . Furthermore, none of the studies cited above actually assessed the extent the participants' prior disciplinary knowledge and metacognitive strategies contribute to their use of metadiscourse and ultimately to their understanding of the materials they read. In two studies, the participants' language proficiency level was assumed, and not measured though metadiscourse awareness was found to be associated with increases in language proflciency. There is also a problem, in my view, with the reading comprehension tests administered. Camiciottoli (2003) used a reading comprehension test, consisting of four-multiple choice questions, each with four options was administered. The first question was of a global nature referring to the content of all the three extracts. The second, third and fourth questions focused instead on the individual content of the extracts. The questions were limited to the identification of the main point of the extract (e.g. In these extracts, the author mainly... / The main point of Extract A is...). I reckon this type of question alone cannot assess the the participants' comprehension of the passages. Parvaresh and Nemati (2008) also deployed one type of question: a set of 11 True/False questions. A more varied question list, in my opinion, can provide a better assessment of the respondents' comprehension of the passages. Finally, the 5-item post-reading questionnaire designed by Camiciottoli (2003) and adapted by Parvaresh and Nemati (2008) is, in my view, too short and overoptimistic. It is not likely that three questions could really measure the students' awareness of metadiscourse. The questions asked were too general, making it difficult for respondents to supply answers. For example, question number four “does the author try to create a dialogue with the readers?” is a bit confounding for non-specialists in linguistics. The term ‘dialogue' is subject-specific. They were not trained to answer this type of questions. The present study attempted to consider these methodological shortcomings when deciding about the methodology to be used. This will be explored in detail in the following chapter.
The review of literature has emphasized the importance of metadiscourse in academic discourse. Metadiscoursal markers provide clues as to how the writer consciously or unconsciously attempts to direct the way the reader is supposed to make sense of the text and in this reveals something about the writer's own suppositions and moves. The affirmations the writer makes are never simply “matter of fact”. Implicitly and/or explicitly the writer always works within a dialogical framework (Bahktin, 1973, p. 6) to accommodate what he/she postulates with the potential or hypothetical readership. In other words,
The literature demonstrated, however, that there is still a lack of consensus on the contribution of metadiscourse to reading comprehension, in particular in SL contexts. This may be attributed to the nature of metadiscourse. In fact, the latter is difficult to delimit. Researchers have offered different explicit criteria for distinguishing metadiscourse. Crismore et al. (1993) summarised the situation
Metadiscourse is an admittedly messy but very important part of language use; thus, in studying it, we should expect a certain degree of impreciseness and subjectivity, reconsiderations and refinements of what does or does not count as an instance of metadiscourse or of a proposition (p. 54).
Second, the methodologies deployed seemed not be adequate in addressing this issue.
This section presents the research setting and methodology used to answer the research questions outlined in the first chapter. It describes the informants' profile as well as the instruments and procedures employed to investigate the extent the participants used metadiscourse when reading research articles in English in their field.
The Participants' Background And Setting
Fifty participants out of 150 were selected for this study among Tunisian university researchers in the field of geography from four faculties: the Faculty of Human and Social Sciences of Tunis, the Faculty of Arts, Letters and Humanities of Manouba, the Faculty of Letters and Human Sciences of Sfax and the Faculty of Letters and Human Sciences of Sousse. The selection of the participants has been done according to the convenience and availability of the universities whose authorities were thought to be more co-operative. It was also based on the results of a background structured interview (see appendix A). A list of eligibility criteria was compiled to evaluate the degree of fit between the study sample and the population the participants were intended to represent. Participants were only selected if they:
1. were still doing research,
2. were reading RAs in English,
3. had a Pre-intermediate level threshold in English (based on self-report).
The population formed a homogeneous group with regard to their age, linguistic, educational, and cultural background. Most of my participants ranged from 30 to 59. They were all Tunisian and spoke Arabic (Tunisian Arabic), as their first language, and Modern standard Arabic as their second language. They spoke French as their third, and English as their fourth language. Thirty had a doctorate, obtained in Tunisia or in France. The languages of instruction for all of them had been French and Arabic. 80 % of them had had four years of English (at school). Only ten of them had had private training in general English. Eleven of them had been to an English-speaking country but for a short period of time (from two weeks to a month). Characteristics of the participants are summarised in the table below.
On The Use Of A Subject Specialist Informant (Ssi)
In the present study, I had recourse to a subject specialist informant to (1) help select appropriate materials for the study and (2) get some background disciplinary knowledge. The SSI consulted was a professor assistant at the department of Geography in the Faculty of Letters and Human Sciences of Sousse.
Many applied linguists stressed the importance of consulting a specialized informant in the target field as each discipline has its own special jargon and specific rhetorical conventions which might not be familiar to language specialists (Bhatia, 1993; Douglas, 2000; Johns, 1993; Selinker and Douglas, 1989; Swales, 1990). Swales (1993, p. 287) maintained that each genre has its own “jargon, acronyms, code signals, inner circles, specialized discourse communities and the like".
Bhatia (1993) stressed the importance of the information provided by SSI in genre analysis and identified the grounds for cooperation between them and the discourse analyst. He (1993, p. 34) maintained that “the analyst double-checks his findings against reactions from a specialist informant, who is generally a member of the disciplinary culture in which the genre is routinely used. Selinker and Douglas (1989) presented a broad description of the profile of an SSI. First, the SSI must be trained and competent. He/She must have "a feel for technical language". Second, he "should be willing to give technical answers, and be willing to put up with come-back questions from the investigator" (Selinker and Douglas, 1989, p. 103).
The genre selected for this study was the published RA on the basis of many criteria: First, it has received extensive attention in genre analysis as it occupies a prominent position in research publication and is considered the main means employed by researchers for the dissemination of knowledge, the publication of claims, and their ratification (Holmes, 1997; Hopkins and Dudley-Evans, 1988; Myers, 1990; Swales, 1990). The present study adopted the approach of the New Rhetoric School which studies genres not only in terms of their textual patterns, but in terms of the social action that these patterns perform in relation to particular audiences and rhetorical situations (Atkinson, 1996; Bazerman, 1988; Freedman and Medway, 1994; Russell, 1997). This school has highlighted a major finding about the RA which is that the structure of an article, its language and style, the author's persona and use of the literature are mainly determined by the social structure of the scientific discourse community. Thus, in the present study we had to be sensitive about the differences (linguistic and rhetorical) existing between the sub-disciplines of economic geography and physical geography, though they both belong to the 'mother discipline' geography.
Second, the RA is a persuasive resource where knowledge claims are negotiated and where their veracity is assessed (Berkenkotter and Huckin, 1995; Casanave, 1995; Myers, 1990). Indeed, the authors' claims are very often produced within a context of hotly contested social and environmental values and competing economic and political interests (Helal, 2007; Ruiying and Allison, 2003). Gross et al. (2002, p. 3-4) argued it is "only through publication that new knowledge claims become fully engaged in a struggle for existence". Acceptance of claims depends, among others, on the discursive and rhetorical strategies deployed by the author. The RA, then, is rich in metadiscourse, the focus of this study. Metadiscourse constitutes the major means by which writers vary and negotiate arguability by adjusting the dialogic status of propositions. Writers can engage with alternative positions by “acknowledging”, “disclaiming” (‘denying' and ‘counter-expecting'), and “proclaiming” (‘concurring' and ‘pronouncing') propositions (White, 2003). A number of studies have emphasized the use of metadiscourse markers in RAs, such as the use of hedging, modality and reporting verbs (Dahl, 2004; Dhieb-Henia, 2003; Hyland, 1996; Salager-Meyer, 1992; Thompson and Ye, 1991).
The participants in the present study were invited to read four research articles. A preliminary list of articles from the discipline of geography was compiled. Ten RAs were presented to the SSI to rank the articles according to their close relation to the participants' disciplines. He selected four research articles (RA) from two disciplines (human geography and physical geography). The articles were piloted with four geography university teachers who specialised in the two disciplines mentioned above. Using Flesch-Kincaid readability measures, the four articles were judged equivalent (see table 8).
The first article (A1) is entitled “Visual threshold carrying capacity (VTCC) in urban management: A case study of Seoul, Korea” written by Kyushik Oh and published in the Journal of Urbanism in 1998. This article presented the Visual Threshold Carrying Capacity (VTCC) approach for maintaining and enhancing the landscape quality of the the downtown of Seoul. The author argued “the purpose of the VTCC method is not to limit development and growth, but to distinguish areas for preservation and areas for active development in advance” (Oh, 1997, p. 8). He thought that rapid urbanization and high- rise and massive development from the 1960s to the present has significantly deteriorated the landscape of Seoul and has obstructed the view of the beautiful mountains surrounding the city.
The second article (A2) is entitled “Cities' contribution to Global Warming: Notes on the Allocation of Greenhouse Gas Emissions” by David Satterthwaite published in the Journal of Biogeography in 2008. Global Warming is an issue that has been and will continue to be hotly contested in both public and private arenas. The author argued against the assumption that cities are responsible for global anthropogenic greenhouse gas emissions. He blamed agriculture, deforestation, heavy industries, fossil-fuelled power stations and high-consumption households for gas in general, since these emissions would be heavily concentrated in cities in high-income nations and they should be ascribed to the individuals and institutions whose consumption generates them” (2008, p. 1).
The third article (A3) is entitled “The Mediterranean vegetation: what if the atmospheric CO2 increased? and written by Cheddadi, Guiot and Jolly published in Landscape Ecology in 2001. Biogeography is the study of the past and present geographic distributions of plants, animals and other organisms. The biosphere includes all life that exists on earth. The authors aimed to demonstrate the ability of a model BIOME3 to simulate future changes in the Mediterranean vegetation for different climate scenarios.
The fourth article (A4) is entitled “Migration in Europe” written by Boswell for the Policy Analysis and Research Programme of the Global Commission on International Migration in 2005. Migration is at the centre of a fierce debate throughout much of Europe. The author presented the economic underpinnings of migration to Europe and shows the increasing and profound ambivalence of Western European countries about it. Issues of labour migration, irregular migration, asylum and integration have become highly politically contested. Populist mobilization on immigration themes has placed even liberal oriented governments under pressure to pursue restrictive approaches.
The participants were invited to read the whole articles at their own pace and time so as to render the reading process as natural as possible. This is a further contribution of the present study, as earlier studies on the impact of metadiscourse on SL reading used short simplified texts (e.g. Camiciottoli, 2003).
Instruments And Procedures
In order to answer the research questions, a number of methodological issues should be considered. The study attempted first to predict other factors than metadiscourse that might affect comprehension, namely the participants' level of English proficiency , their metacognitive reading strategies and their prior disciplinary knowledge (formal and content schema). Thus the study was undertaken in two phases and used both quantitative and qualitative data. In fact, it relied on a mixed- method approach. The use of different instruments, such as structured interviews, the TOEFL and oral protocols enhanced construct validity by providing multiple points of data collection (Gall et al., 2003). Table 11 summarizes the instruments that were used in the present study.
The Early Phase Of The Study
The first phase of the study gave us insight into the participants' proficiency level in English, their prior disciplinary knowledge and their metacognitive reading strategies. A test of English as a Foreign Language (TOEFL), a metacognitive reading strategy questionnaire, and five disciplinary knowledge quizzes were administered to the participants.
A Structured Interview (Appendix A)
The structured interview was administered to the target population with the following objectives in mind:
i. To obtain background information about the participants.
ii. To select the participants for the present study
The structured interview (SI) has been favoured by many researchers (for example Bailey, 1978; McDonough and McDonough, 1997; Nunan, 1992; Seliger and Sohamy, 1989). It is thought to be an 'extraordinary' method of collecting data as it allows an interpersonal connection between the interviewer and the interviewees, 'an incredibly rich interaction' (Dowsett, 1986) and 'more personalized responses' (McDonough and McDonough, 1997). Seliger and Sohamy (1989, p.166) claim that "(this instrument) permits a level of in-depth information gathering, free response and flexibility that can not be obtained by other procedures". The SI shares many similarities with quantitative written questionnaires. The typical qualitative interview is a one-to-one ‘professional conversation' (Kvale, 1996, p. 5) that has a structure and a purpose ‘to obtain descriptions of the life world of the interviewee with respect to interpreting the meaning of the described phenomena' (pp. 5-6). In this format, the researcher follows a pre-prepared, elaborate ‘interview schedule/guide', which contains a list of questions to be covered closely with every interviewee, and the elicited information shares many of the advantages (for example, limited comparability across participants) and disadvantages (for example, limited richness) of questionnaire data.
The SI offers many advantages. First, it gives the interviewer a great deal of flexibility 'in changing the order of questions and for more extensive follow-up of responses' (McDonough and McDonough, 1997, p. 183). Second, since the designer is asking the questions, the interviewer can ensure that the interviewee focuses on the target topic area and that the interview covers a well-defined domain, which makes the answers comparable across different respondents.Third, the designer can clarify any misunderstandings in the interpretation of the questions. The other side of the coin is, however, that in a structured interview there is generally little room for variation or spontaneity in the responses because the interviewer is to record the responses according to a coding scheme (the only occasion when the interviwer may not stop the interviewee being spontaneous is when responding to follow-up questions). Moreover, interviews may be disadvantageous especially if the interviewer influences and biases the interviewees' responses. Furthermore, the main weakness of the interview is that it is time-consuming to set up and conduct, and that it requires good communication skills on the part of the interviewer (Dornyei, 2007). Additionally, because the interview format does not allow for anonymity, there is a chance that the respondent will try to display him/herself in a better than real light. Finally, interviewees can also be too shy and inarticulate to produce sufficient data, or at the extreme, they can be too verbose, generating a lot of less-than-useful data (Dornyei, 2007).
The researcher was careful to design a good interview guide and to select a few relevant questions. When wording the questions, the researcher tried to avoid the following problems reported in the literature:
1- Response bias: the fear of the informant being judged negatively or exposed to criticism incites him\her to give subjective answers. Daoud (1991) warned “one should be aware of the possibility that the respondents might try to project a positive image of themselves as proficient readers in their field instead of giving more realistic answers (p. 43).
2- Lack of comprehension: when the respondent does not understand the question, he\she will give inaccurate answers. Thus, researchers should avoid using technical words (Wilson,1990)
3- Lack of objectives: the inference of the interviewer during the administration of the interview may influence the respondent and elicit, therefore, the desired answers (Hague, 1993).
4- Too few items: it is impossible to formulate basic judgements because a single question cannot detect the presence or extent of an attitude.
The main function of the interview guide, according to Dornyei (2007, p. 137) is to help the interviewer in a number of areas: (a) by ensuring that the domain is properly covered and nothing important is left out by accident; (b) by suggesting appropriate question wordings; (c) by offering a list of useful probe questions to be used if needed; and (e) by listing some comments to bear in mind. McCracken (1988) emphasizes the importance of the interview guide due to the demanding objectives interviewers want to achieve and the multiple factors they have to attend to during the interview. The guide was combined with an ‘interview log' to leave space in it for recording the details of the interview (for example, participant, setting, length) as well as for the interviewer's comments and notes.
As for the response format adopted, I relied heavily on the closed-response format. Closed-response questions are questions in which the particiapnt is offered a set of pre-determined answers to a specific question. Such specific questions channel participants' responses, even when the participants have no clear idea about the subject. They are time-saving, easy to answer, score and analyze (Hague, 1993; Oppenheim, 1992; Wray, et al. 1998). These questions are essentially used to obtain quantitative data, to allow for comparability and structuring for analysis. However, one disadvantage of closed questions is that they are so directive and patronizing. As a consequence, the respondents might provide flippant and unhelpful answers. In addition, closed questions can make respondents prejudiced against some issues and possible responses. The questions vary from checklists which present the informant with a range of alternatives (e.g., questions 8, 10, 11), to two-way questions in which the respondents are asked to choose among two answers Yes or No (e.g., questions 7 and 9).
The SI designed in the present study was built on Daoud's (1991) questionnaire for two main reasons. First, the latter meets the objectives of the present study. Second, it would help avoid the risk of unreliability and invalidity. Seliger and Shohamy (1990) recommended that
In most research studies, a researcher should first begin by searching for available data collection procedures. Using a ready-made instrument which has been developed by experts and for which information regarding reliability and validity is more advantageous than developing a new procedure, provided that it is appropriate for the given research (p. 189)
The SI in the present study included eleven questions sequenced logically from general to specific, and from easy to more challenging. This would help the interviewees feel more relaxed and more motivated to answer. The first questions helped to create trust in this initial ice-breaking period (Dornyei, 2007; Henerson et al., 1987). Thus, the first section (questions 1 to 7) included questions on the respondents' native and foreign language(s), the number of years they had studied English at second and tertiary levels, their training in English, the language of instruction at school and university, and so forth. The primary objective of this section is to assess the participants' language proficiency and to find out about the linguistic environment in which they had operated which might have affected their proficiency in English. The second section of the questionnaire (questions 8 to 11) focused on the respondents' needs for English. These questions as stated above would help me select the participants for the present study.
A Test Of English As A Foreign Language (Toefl) (Appendix B)
The participants' English proficiency is very important in answering the research questions. Determining the participants' language proficiency is a very tricky task. Indeed, Lee and Schallert (1997, p. 716) stated “the construct of language proficiency is not a simple one as it relates to language competence, metalinguistic awareness, and the ability to speak, read, and write the language in contextually appropriate ways”. Language proficiency has been defined differently by different researchers in the field. For example, Hymes (1972) distinguished linguistic competence, or knowledge of the rules and systems of a language, from communicative competence, or knowledge of the social rules of language use. Canale and Swain (1980), on the other hand, identified four subcategories of communicative competence: linguistic, discourse, sociolinguistic, and strategic competence.
Different studies used different measures to assess proficiency in English. A number of researchers have decided on their participants' SL proficiency level according to the language class level they attended (Carrell, 1988, 1991; Clarke, 1978, 1980), or reading class (Hudson, 1982) or according to the number of language courses participants had taken before the experiment ( Raymond, 1993). Other researchers, such as Sasaki, 1993) used multiple measures of L2 proficiency: a short-text multiple-choice test, a long-text multiple-choice test, a free composition and a cloze test. A third category of researchers preferred use a standardized language proficiency test (Chen and Donin, 1997; Clapham, 1996; Tan, 1990), which I believe is the most reliable method for assessing L2 proficiency.
Since, in the present study, language proficiency is used to refer to the ability to critically read and comprehend an academic discourse written by a particular academic discourse community, the reading battery of a practice version of the Test of English as a Foreign Language (TOEFL) was administered. The TOEFL is a battery of academic English language ability designed primarily for admission of non-native speakers of English to higher education institutions. It is a standard test known for its validity and reliability and has been used by a number of researchers(e.g. Perkins, Brutten, and Pohlmann, 1989). Though the reading battery scores were only part of the TOEFL test scores, they were thought to be somewhat indicative of the language proficiency of the participants.
The reading battery of the TOEFL measured the participants' ability to understand university-level academic texts (Enright et al., 2000). The tasks were expected to engage a complex of abilities, processes, and strategies that were appropriate for different purposes. It comprised three sets, each of which contained 12-14 items associated with a reading passage of approximately 700 words. The reasons given for this length, according to Schedl (2004), was that relatively long texts (older versions of the TOEFL comprised only 400-500 words) better represented the academic experiences of faculty. Likewise, Enright, Grabe, Koda, Mosenthal, Mulcahy-Ernt and Schedl (2000) believed that the manipulation of the reading task along with text length would allow for the fair, reliable, and efficient evaluation of individual academic reading proficiency.
The passages were of different types (argumentation, exposition, and historical narrative) and of different structures (classification, comparison/contrast, and cause/effect), with information presented from more than one perspective or point of view. Despite being academic and persuasive, none of the passages resembled a RA, with all the persuasive features I discussed earlier. There is no standard test, to my knowledge, that tests English reading proficiency against the RA genre. Success on the TOEFL reading section depended on how well the participants understood the text passages and the attached questions.
The Reading Battery contained different tasks assessing the four reading purposes outlined in the Reading Framework (Enright and Schedl, 2000). These were (1) reading to find information –locate discrete pieces of information in a text, (2) reading for basic information –understand the main ideas or the main points of the text or form some understanding of the main theme of the text-, (3) reading to learn –develop an organized understanding of how the main ideas, supporting information, and factual details of the text form a coherent whole-, and (4) reading to integrate-when two texts are used in the prompt- generate an organized understanding of the relationships among the kinds of information found in different sources. There were three basic question formats in the test:
a. Multiple choice questions: This question type consisted of a question and four answer choices. They tested and evaluated the participants on four main skills: their ability to scan text for key facts and important information (questions 4, 7 and 8), to understand vocabulary in context (questions 1 and 9) and references (question 5), to make inferences about what is implied in a passage (questions 2, 3, 6 and 11), and to paraphrase a sentence in the passage (question 10).
b. Insert a sentence: This question type consisted of a sentence and four places marked in the text. The participants were asked to choose where the sentence best fit into the passage (question 12). This question was meant to test the participants' ability to recognize the organisation of a passage and the relationship between ideas.
c. Summary: This question type contained six answer choices. The participants were asked to select three of the choices that expressed the most important ideas in the passage. The question aimed at assessing the participants' ability to summarize a passage.
All the information needed to answer the questions could be found within the passage. It was not necessary for the participants to have any prior knowledge about the topic in order to answer the questions. The participants had 60 minutes to complete the Reading section. I strove to make sure that the conditions were the same at every administration of the instrument. Each question in the set was worth one point, with the exception of the last question in each set, which was worth two points.
A Survey Of Reading Strategies (Sors) (Appendix C)
To assess the participants' metacognitive awareness and perceived use of reading strategies while reading for academic purposes, Sheorey's and Mokhtari's Survey of Reading Strategies (SORS) was used. Sheorey and Mokhtari (2001, P. 2) defined reading strategies as “mental plans, techniques, and actions taken while reading academic or school-related materials. The SORS was thought to be an appropriate measure for a variety of reasons: first, it was developed to be used with adolescent and adult English as a Second Language (ESL); in other words, it aimed to collect information about the various techniques an ESL learner used when he/she read academic materials in English, namely the general academic article. Second, it has been used by many studies and reported to be highly efficient (e.g. Lee, 2006; Lei, 2009; Poole, 2005). Finally, most instruments found in the literature aimed at assessing native speakers' metacognitive awareness of reading processes and did not consider some of the strategies specific to bilingual readers such as translating from English into one's native language or using both languages when reading to enhance their reading (Mokhtari and Reichard, 2002).
The instrument was an adapted version of the Metacognitive-Awareness-of-Reading-Strategies Inventory (MARSI), which was originally developed by Mokhtari (1998–2000). It was used to measure native English speaking students' awareness and use of reading strategies while reading academic materials. Sheorey and Mokhtari (2001) reported that MARSI was validated using a large native speaker population (n=825), representing students with reading abilities ranging from middle school to college. The internal consistency reliability coefficients (as determined by Cronbach's alpha) for its three subscales, which were based on the results of factor analysis, were as follows: Metacognitive (0.92), Cognitive (0.79), and Support strategies (0.87). The reliability for the overall scale was 0.93, indicating a reasonably dependable measure of metacognitive awareness of reading strategies.
To adapt the MARSI to ESL situation, Sheorey and Mokhtari (2001) made three important revisions:
First, we refined the wording of several items to make them easily comprehensible to ESL students. Second, consistent with relevant research about reading strategies used across languages (c.f., Jimenez, Garcia, and Pearson, 1996), we added two key strategies clearly not used by L1 readers but often invoked by L2 learners (‘translating from one language to another” and “thinking in the native and target language while reading”). Finally, we removed two items (namely “summarizing information read” and “discussing what one reads with others”) which do not specifically constitute reading strategies as conceived in the current research literature on metacognition and reading comprehension (p. 4)
Sheorey, Mokhtari and Reichard (1999) reported that a pilot-test of the revised instrument on a population of ESL students studying at a university in the United States (n=147) gave consistent results relative to the instrument's overall reliability (Cronbach's ALPHA=0.89), a “reasonable degree of consistency in measuring awareness or perceived use of reading strategies among native and non-native speakers of English” (Sheorey and Mokhtari, 2001, p. 8).
The SORS consisted of 28 items, each of which used a five-point Likert scale ranging from 1 (“I never or almost never do this”) to 5 (“I always or almost always do this”). The participants were asked to read each statement and circle the number that applied to them, indicating the frequency with which they use the reading strategy implied in the statement. Thus the higher the number, the more frequent the perceived use of the strategy concerned (Sheorey and Mokhtari, 2001, p. 8).
The SORS measured three broad categories of reading strategies, namely, metacognitive strategies, cognitive strategies, and support strategies. A brief description of each SORS category and the number of items within each category are given below (Sheorey and Mokhtari, 2001, p. 8-9):
1. Metacognitive strategies are those intentional, carefully planned techniques by which learners monitor or manage their reading. Such strategies include having a purpose in mind, previewing the text as to its length and organization, or using typographical aids and tables and figures (13 items: questions 1, 3, 4, 6,8,12,15,17,20,21,23,24, and 27).
2. Cognitive strategies are the actions and procedures readers use while working directly with the text. These are localized, focused techniques used when problems develop in understanding textual information. Examples of cognitive strategies include adjusting one's speed of reading when the material becomes difficult or easy, guessing the meaning of unknown words, and re-reading the text for improved comprehension (8 items: questions 7, 9, 11, 14, 16, 19, 25, and 28).
3. Support strategies are basically support mechanisms intended to aid the reader in comprehending the text such as using a dictionary, taking notes, or underlining or highlighting the text to better comprehend it (7 items: questions 2, 5, 10, 13, 18, 22, and 26).
Assessment Of The Participants Discipline-Related Background Knowledge (Discourse Type And Subject-Area)
The objective of the assessment was to evaluate the contribution of discipline-related background knowledge to the participants' reading comprehension of the RAs under study. In fact, there is much evidence in the literature to support the hypothesis that familiarity with the subject-area treated by the text enhances foreign language reading comprehension (e.g., Clapham, 1996; Tan, 1990) (see section 2.1.4. in the Literature Review).
Different methodological designs have been used by researchers. In fact, the designs varied significantly in the way the independent variables (the participants' discipline-related knowledge) and the dependent variable (the participants' reading proficiency) were assessed, as well as the statistical analyses conducted to analyze these variables. In earlier studies, in most cases, the participants' existing discipline-related knowledge was not measured, but rather it was assumed on the basis of their areas of study specialization (Alderson and Urquhart, 1983; Koh, 1985; Ridgway, 1997). In the studies that actually assessed the existing discipline-related knowledge, very different methods were used. In one study by Chen and Donin (1997), the presence of discipline-related knowledge was assessed with an oral interview about the particular topic, whereas in another study (Clapham, 1996) the participants completed a questionnaire that covered information about reading habits in the subject area as well as familiarity with the content area. In yet another study, a more direct measure of discipline-related knowledge that included multiple-choice questions about the text topic was also used (Tan, 1990). It is worth noting that being specialized in a particular discipline does not mean, that the participants do not have background knowledge of other fields or they have extensive knowledge within their field (Uso-Juan, 2006).
In the present study and in line with Chen and Donin (1997) and Tan (1990), the researcher deployed quizzes to assess the participants' familiarity with the RA genre and the topics of the RAs administered in this study. They are considered as useful ways to have a complete picture of the participants' discipline-related knowledge (genre and content). These are described in more further details below.
The Participants Familiarity With The RA Genre Quiz
To test the participants' familiarity with the structure of the RA, I used the scrambling technique. This methodology has been used successfully to measure the effects of knowledge of discourse structure in various problem-solving situations such as the experimental report (Davis, Lange and Samuels, 1988; Poulsen, Kintsch, Kintsch and Premack, 1979, Taylor and Samuels, 1979). As the RA genre per se is not a single genre, two types of tasks were administered to assess the participants' familiarity of the two assigned RA subgenres, namely the review and the experimental RAs (Myers, 1991; Swales, 1990, 2004). In the first task, the participants were given a scrambled- order of experimental RA and were asked to reorder the sections in the canonical structure found in conventional scientific journal and match them with their appropriate communicative functions (Swales's (1990) IMRD).
In the second task, the participants were asked to match four types of review articles with their definitions and functions. The Noguchi's (2001) generic description of the review article inspired the second task. Noguchi (2001) worked on twenty-five “review articles” published by the top-ranked U.S. journal Proceedings of the National Academy of Sciences in 1993 and came out with the following four categories: history (presenting a historical view of a facet of the field), status quo (describing the current situation in a field), theory/model (proposing a theory or model to resolve some issue in the field) and issue (calling attention to some issue in the field (p. 142). Meyers (1991) noted that the review article “draws the reader into the writer's view of what has happened, and by ordering the recent past, suggests what can be done next” (p. 46). Commenting on the review articles, Swales (2004, p. 211) stated
it would seem then that there are no preferred orders here. Rather, success lies in the LR writer convincing the reader that there is an organizing mind at work, which is assembling potentially disparate material in a coherent way and is doing so with suitable variation in citation patterns, a judicious selection of reporting verbs, a developed sense of appropriate paraphrase, and a feeling for what are the right moments for direct quotation, telling illustration, or critique.
The Prior Knowledge Tests
A measure of the candidates' prior knowledge in the disciplines involved was conceptualized in two ways:
a. The extent to which they were familiar with their own subject-area, namely physical geography and human geography.
b. The extent to which they were familiar with the topic under discussion in the discipline-related reading texts, namely, global warming, vegetation simulation model, urban management and immigration.
To this end, the participants' topic-domain knowledge for each assigned article was measured by means a vocabulary quiz. The quiz comprised twenty words and the respondents were asked to tick ten they expected to find in the RAs under study. Because the RAs dealt with specialized topics, the researcher enlisted the help of the subject specialist informant.
A quiz on technical academic vocabulary was considered to be an appropriate measure in assessing the readers' level of preexisting knowledge about the topic. “A technical word is one that is recognizably specific to a particular topic, field or discipline” (Coxhead and Nation, in Flowerdew and Peacock, 2001, p. 261). Technical vocabulary “for any particular subject (…) consists of probably 1,000 words or less. It could provide coverage of up to 5% of the running words in a text” (Coxhead and Nation, in Flowerdew and Peacock, 2001, p. 252). There are degrees of ‘technicalness' depending on how restricted a word is to a particular area. In the present study, the quiz included the most technical vocabulary that appear rarely if at all outside this particular field (Yang, 1986 and Becka, 1972). Studies investigating the vocabulary needed for certain academic disciplines showed how knowledge about technical vocabulary can facilitate EAP reading (e.g. Corson, 1997; Martin, 1976; Meyer, 1990).
The Second Phase Of The Study
This phase aimed to assess whether the participants perceived and used metadiscourse markers when reading a research article in their discipline especially as such elements are intended to help readers distinguish fact from opinion.
To achieve these goals, the participants were tested via a post-reading questionnaire and a recall task. These would allow me to have as reliable, valid and comprehensive data as could be obtained and to increase the probability that the results obtained are a true measure of the participants' use of metadiscourse.
A Post-Reading Questionnaire (Appendix E)
A questionnaire was administered requiring the participants to reflect on the reading activity. It was partly adapted from Crawford Camiciottoli's (2003) and from Pérez and Macia's (2002) studies. Participants were asked to
1) assess the perceived level of difficulty of the RA
2) assess the author(s)' success in facilitating the reading process.
Most questions were framed in such a way as to elicit both quantitative and qualitative data since the participants were asked to justify their answers by adding their own comments.
An Oral Recall Task (Appendix F)
To measure the participants' reading comprehension and use of metadiscourse, I had recourse to the recall protocol. The latter has been claimed to be a powerful measure for tracking a reader's psychological processes while minimizing researcher bias (Bernhardt, 1983, 1991; Hayes, 1989; Hayes & Flower, 1980; Johnston, 1983; Lee, 2007). It has also been claimed that compared to multiple-choice questions, free recall tasks produce richer evidence helpful for understanding a reader-based constructivist model of reading processes (Bernhardt, 1991; Heinz, 2004; Roebuck, 1998). Bernhardt (1991) noted, “a free recall measure provides a purer measure of comprehension, uncomplicated by linguistic performance and tester interference” (p. 200). Congruent with this claim, Dhieb-Henia (2003, p. 110) argued
It is now widely agreed that product-oriented [tests] or observational techniques are not sufficient for studying the process of reading comprehension whereby meaning is mostly gathered through silent reading and therefore is inaccessible to direct observation. This explains the direction in research for using more process-oriented techniques.
I used the immediate retrospection method, which is completed directly after the reading tasks, so as to ensure that the participants' short-term memory can be assessed and its contents reported and 'guarantees a minimum of interference with the reading process' (Dhieb-Henia, 2003, p. 110). The participants were asked to read the RA section by section and after each section, they had to recall everything they could remember about what they had just read (using words from the original passage or their own words) without looking back at the text. Berkemeyer (1989, p. 131) maintained:
The immediate recall protocol demands that the reader comprehends the text well enough to be able to recall it in a coherent and logical manner... This procedure allows misunderstanding and gaps in comprehension to surface; a feature that other methods of evaluation cannot offer
To avoid the problem of rote learning, the participants were not told beforehand that they would be asked to recall the text. Moreover, they were administered a distraction task (a questionnaire asking for comments about the text) between the initial reading of the text and the request for an oral recall. I was unable, unfortunately, to audiotape the oral protocols. The researcher was forced to take notes. The participants refused to be audiotaped.
The participants were allowed to recall the RA in any language they liked as research has shown that the participants' language choice may be important rather than the language itself. Dhieb-Henia (2003, p. 111) stated ' in EFL contexts, where command of the foreign language was only partial, there was some concern about using this language for reporting on the reading process as it might interfere with comprehension'. Doni and Silva (1993) and Chen and Doni (1997) demonstrated that the language used affected production. When reporting in their native language, people recalled more and demonstrated more use of higher-level processing, compared to reporting in their second or foreign language. Likewise, Barnett (1989) contended that 'students may not be able to show their comprehension of a text due to their inability to express themselves well in English in their recall protocols' (p. 45). In the same vein, Levine and Reves (1998) reported that when protocols were conducted in English, the statements produced by students with lower language proficiency levels were shorter. Similarly, Alderson (2000) asserted that the recall should be completed in the test taker's L1 because otherwise it becomes a test of English proficiency instead of reading. Bernhardt (2005) reinforced this recommendation until learners reach the highest SL proficiency/fluency levels.
Scoring Of The Recalls
The data from recall protocols were analyzed in such a way as to yield information about the participants' reading comprehension and use of metadiscourse. In order to avoid the problem of subjectivity of marking mentioned by Alderson (2000), the researcher performed the scoring in collaboration with four colleagues. The group consisted of two Tunisian-born university applied linguist researchers, the subject informant of the present study and one British-born university applied linguist lecturer at Coventry University, UK. Members of the group were invited to point out the sentences that they thought were basic to each reading passage in order to create the list of main ideas for each reading passage. This simple but time-consuming task was done individually. Thus, only those ideas identified by at least four of the experts were considered basic main ideas. Interrater reliability ranged from .95 to .97 (2-way mixed). The same procedure was followed to identify the metadiscourse markers deployed in the RAs under study using Hyland's (2005) taxonomy (the subject informant was not involved as he is a Geography teacher). The scoring scale for the recall test was 0 to 100 points in whole numbers (decimals were rounded off to the nearest whole number). Further details about the scoring of the recall protocol are given below.
Reading Comprehension Scores
The articles were divided into quantifiable units using a system developed by Mayer (1985) for research on scientific prose. In the Mayer system, passages were divided into idea units, with each unit communicating one event, state or action. Idea units consisted of one predicate and one or more arguments. Arguments corresponded to noun phrases. Predicates can be main verbs, expressions of location, time markers or text organizational signals (Bransford and Franks, 1971; Carrell, 1983). The assumption is that informational quantity is a good reflection of global reading comprehension. However, as different reading variables may have qualitatively varied effects on readers' comprehension (Bernhardt, 1991; Carrell, 1987, Lee, 2009), I decided in this study to consider the relative content value of each information unit recalled rather than informational quantity. This strategy “offers a more revealing window into the quality of reading comprehension” (Lee, 2009, p. 163). The final recalls were scored according to the important ideas retrieved. The tables below present the idea units for all the RAs.
To assess the extent the participants used metadiscourse when reading the assigned RAs, two steps were followed: First, the RAs were analyzed sentence by sentence and coded according to Hyland's (2005) taxonomy. Although some phrases could be concluded to belong into more than one category, only the major category was represented in order to reduce the complexity of the coding. Hyland's list is by no means exhaustive, a fact he confirmed in a personal communication to me (July, 30th, 2009). He wrote “you can always find more than I have listed which do a similar job. If you are examining texts by hand, then you will spot more candidates for inclusion, particularly in the interactional categories”. Therefore, other interactional items than those identified in Hyland's list were included. Then, the coding process was compiled by two coders independently, and then the different clarifications were discussed in order to arrive at consistent results. The reliability of the coders was assessed by measuring the inter-rater reliability. The inter-rater reliability means that the results of content classification are invariant over time. Dornyei (2007) maintained inter-rater reliability could be ascertained when the same content was coded more than once by the same coder. In my study, the inter-rater reliability of the interactive category was 89.7 percent; the interactional one was 87.5 percent. This meant that the combined reliability index of interactive and interactional categories reached 85 percent.
The next step was to analyse the participants' data in terms of metadiscourse markers or metadiscursive relations recalled and to compare them against the metadiscourse markers identified by the coders. Each metadiscourse marker or relation recalled was worth one point and the total number was turned into a percentage to be compared with the total number of metadiscourse items/ relations found in each RA.
The following passage was taken from the article on Global Warming (Satterthwaite, 2008, p. 1) to demonstrate how the analysis was conducted.
It has become common, for cities to be blamed for generating most of the world's greenhouse gas emissions. For instance, many sources claim (a) that cities are responsible for 75-80 percent of global greenhouse gas emissions, including Munich Reinsurance, the United Nations Human Settlements Programme, (…). Perhaps (b) this was an estimate for the United States, which was then assumed (c) to be valid worldwide. However (d), these figures are certainly (e) inaccurate (f) when applied to cities worldwide, although (g) it is difficult (h) to produce an accurate figure because(i) there are too few cities for which there are detailed greenhouse emission inventories.
The italicized expressions can be grouped in the following categories:
(1) Hedges: claim, perhaps, assume.
(2) Boosters: certainly.
(3) Attitude markers: inaccurate, difficult.
(4) Transition markers: although, however, because.
In the following, I will illustrate how the use of these markers can orient the reader's attention while reading.
They indicate how the writer assesses the probability or truth of the propositional content and how committed the writer is to that assessment. Hedges are used to help readers judge or respect the truth value of the propositional content the way the author wishes them to. In the above passage, the writer uses linguistic markers such as “perhaps”, “assume” and “claim” to show a lack of complete agreement with sources blaming the cities for the world's greenhouse gas emissions. These hedges highlight the author's commitment to the sources' discourse and may influence the readers' overall perception of the global warming issue. Jlang and Blan (2008, p. 4) assume that “he information signalled by these markers will be highly activated as a result of the increased attention that readers pay to textual information that marked by hedges”.
Since these markers can indicate the writer's attitude towards the content or their own readers, they can explicitly inform the reader of the author's perspective on a particular idea and important information in the text. For instance, in examples (f) and (h) the idea that it is difficult to produce accurate figures on worldwide cities' greenhouse gas emissions is more likely to be accepted by readers with the help of the adjectives “inaccurate” and “difficult” which indicate the author's opinion towards the information. Hence, in the case of these features, readers will be prompted to focus on the current signalled idea and make a connection with the subsequent cycle. This, as Jlang and Blan (2008) claim, “in turn will result in increased memory for textual content produced by those attitude markers” ( p. 4).
These markers can help the author to access the truth or probability of the text content. In example (e), the writer used the adverb “certainly” to put his positive emphasis upon the following proposition. Thus, readers are expected by the author to pay more attention to the propositions contained in that clause and are more likely to be persuaded to accept the author's opinion (the inaccuracy of the figures advanced by the sources mentioned in his article).
In the above passage, the following sentence segments contain examples of the use of transition markers.
a. However (d), these figures are certainly inaccurate when applied to cities worldwide,
b. Although (g) it is difficult to produce an accurate figure because(i) there are too few cities for which there are detailed greenhouse emission inventories.
For readers, these transition markers can be utilised as a frame or architecture for the mental representation of the passage as a whole. Meanwhile, the readers' attention will focus on the important textual information cued by these conjunctions. In examples (d) and (g), “however” and “although” mark the contrastive information which provided for the greenhouse gas emission figures. In this case, the reader constructs the meaning and structure of the passage when these two clauses have been activated and made connections with each other.
The above discussion illustrates how the use of metadiscourse can help focus readers' attention. Signalled or marked information results in stronger memory representation in readers' minds and helps them to recall better. The tables below show the distribution of metadiscourse markers in the four RAs.
Prior to administering the instruments to the participants, they were piloted. Piloting is highly recommended in the literature. Nunan (1992) maintained 'it is very important that questions are piloted with a small sample of participants before being used' (p. 151). Likewise, Weir and Roberts (1994) maintained that piloting allows the evaluator to improve the quality of the instrument by detecting any questions or instructions that may be ambiguous or confusing to the participants. In the same vein, Oppenheim (1992) argued that, though piloting is a time-consuming activity, it allows the researcher to check the wording of questions and its ordering and to reduce non-response rate. The instruments were tried on a smaller group with a similar profile: four non-native speakers of English university teachers/researchers of geography. This pilot audience did not take part in this study. The try-out phase offered several advantages: First, it allowed the researcher to estimate the time required for each instrument. Second, the researcher gained useful feedback concerning the wording of questions and the layout of the questions. For example, it was found out that specialized terminology such as 'skills' and 'genre' were not clear for most of the participants and that some questions were redundant. Finally, it ensured the reliability and the validity of the instruments. In fact, the instruments were administered more than once to the same group over a period of time and consistency in the results was obtained.
The instruments were administered in the period from November 2008 and December 2009. Data were collected at four sites: the Faculty of Social and Human Sciences of Tunis, the Faculty of Arts, Letters and Human Sciences of Manouba, the Faculty of Letters and Humanities of Sousse, and the Faculty of Letters and Humanities of Sfax. After obtaining the permission of the head of departments and the deans of the universities, I combined a list of all geography teachers/ researchers. Then I sent them solicitation letters describing the objectives of my study and requesting them to participate in the study and to suggest some time for us to meet. In the first meeting, the teachers were invited to complete a background structured interview. Findings from the interview were meant to select the participants for the study. I only selected the participants who seemed to be somewhat proficient in English (pre-intermediate level), and to regularly read RAs in English. Some participants (n=5) dropped out of the study because they disagreed about the orol protocol procedure, and their data
are not included in the analysis. Others (n=3) declined the TOEFL because they refused to be evaluated. Of 100 geography teachers/ researchers, only half were returned. The selected participants were contacted to arrange further meetings for the other instruments to be done. The site coordinator determined the interval between sessions according to the availability of the participants. The TOEFL test was administered first. The metacognitive reading strategy questionnaire was completed in another session. At the end of the second meeting, arrangements were made with the participants to meet again for the oral protocols. Four appointments were set for the four research articles. The participants decided about the timing of the meetings. Prior to reading each RA, the participants had to complete a disciplinary knowledge quiz. After the reading, the participants were asked to complete a post-reading questionnaire and to perform a free oral recall protocol. The participants dealt with the research articles at their own space and time. Indeed, researchers have emphasised the role context variables such as environmental and situational elements play in reading comprehension (Walberg, Hare and Pulliam, 1981, p. 154).
When conducting the protocols, the researcher first explained the reason for the protocol as understanding the purpose of the questions would increase the motivation of the respondents to respond openly and in detail. The participants were given a little overview of what would happen to the protocol data and were reassured on the issue of confidentiality. I strove to ensure a relaxed, non-threatening atmosphere in which the respondent could feel comfortable to express him/herself. Some small talk was helpful to build rapport with the respondents. I then kept silent and listened without interrupting the respondents. Carry-on feedback backchannelling signals were offered to appear a particularly sympathetic listener (for example, nods, ‘uh-huh' noises, one-word utterances like ‘yeah'. Miller and Crabtree (1999) highlighted the important role of small gestures such as the ‘attentive lean', the ‘eyebrow flash' and the ‘sympathetic smile'. Robson (2002) recommended that the researcher should let the respondents dictate the pace without being rushed or interrupted. They should be patient and resist stepping in too quickly with a new question. She, however, interrupted the natural flow of the account from now and then when she felt it drifted away. I made sure to use phrases for inoffensive interruption and refocusing (for example, ‘Let me stop you here for a moment and go back to what you said earlier to make sure that I understood you well'. Dornyei (2007) stated “this is an area where the skilful use of various probes can make a real difference” (p. 140). Richards (2003) pointed out that a golden rule for conducting a protocol is ‘always seek the particular' (p. 53). When the interviewee was not very forth-coming about a certain topic, I used various probes, including ‘silent probes' that involve remaining quiet to indicate that she was waiting for more, ‘echo prompts' (repeating the last word spoken by the respondent with an interrogative tone), or simply asked the participant to elaborate on what he said. When the protocol drew to a close, I used pre-closing moves such as summarizing or recapping the main points of the discussion. Dornyei (2007) recommended this strategy as it allows the participant to correct anything that the researcher may have misunderstood and to make any comments that might be relevant/important but which have not been covered in the rest of the protocol. Kvale (1996, p. 128) suggested the following end “I have no further questions. Do you have anything more you want to bng up, or sk about, before we finish the interview?). At the end I re-expressed her gratefulness and respect, and discussed the logistics of how to keep in touch in the future. Table 18 summarizes the data collection procedures.
Qualitative Data Analysis
Qualitative data included the major instrument of the present study: the free oral recall protocols. I typed the data into word-processing files. Then, they were coded using the computer program Nvivo 2(Richards, 1999). The unit of analysis for the protocols was the idea unit for the reading comprehension score and the metadiscursive relation (interactive or interactional) for the metadiscourse score (see sections 188.8.131.52.1.1. and 184.108.40.206.1.2., respectively). Appendix M provides an illustration for both the coding and scoring methods.
Quantitative Data Analysis
To answer questions about the extent the participants' use of metadiscourse related to their comprehension of research articles in English in their fields and the extent the participants' language proficiency, metacognitive reading strategies and disciplinary knowledge contributed to their use of metadiscourse and comprehension of RAs in English in their fields (i.e., Research Questions 2 and 3), data of the coded protocol, of the TOEFL, of the SORS and of the related disciplinary knowledge quizzes were tallied and scores were computed. The scores were then turned into percentages to facilitate statistical analyses (Wolfe et al., 1998).
Various statistical analyses (spearman correlations, multiple regression analyses, ANOVA and multilevel analyses) were conducted to assess the relation between (a) the use of metadiscourse and reading comprehension, and (b) ESL reading comprehension, metacognitive reading strategies and content and genre knowledge on the one hand and use of metadiscourse and RA comprehension on the other hand.The following table offers further details about the data coding and statistical analyses for this study.
Titles, figures, bibliography and appendices were not included.
The Flesch/Flesch–Kincaid readability test is designed to indicate comprehension difficulty when reading a passage of contemporary academic English. Among the core measures used by the Flesch Reading Easiness are word length and sentence length. Scores between 90 and 100 can be interpreted as easily understandable by an average 11-year-old student; 60 to 70 means that the text is easily understandable by 13- to 15-year-old students and 00 to 30 means that the text is best understood by university graduates.
I am aware, however, that reading one article (about 1000 words long) in a foreign language is already a demanding task for busy university researchers. Asking them to summarize after reading adds more difficulties. I took account of this issue when interpreting the data.
It is produced by the Educational Testing Service in Princeton New Jersey, and the International English Language Testing System (IELTS), developed by the university of Cambridge.