Proposal Type:	Symposium
Domain:	Knowledge Acquisition and Expertise in Specific Domains
SIG:	Writing
Type	Invited SIG Symposium
Title	Reading during Writing
Abstract	Writing is a frequent and important activity in work life as well as in school and the importance of writing has probably increased with the use of computers. Some examples of typical writing activities in class are composing, answering questions and scoring/commenting on tests and student/peer papers. In order to improve the teaching of writing we need more knowledge about the cognitive processes that underlie text production. One important part of the writing process is reading. Students extract information from sources in order to compose a paper, they read the questions of a test and they read other students’ papers during peer review sessions. Teachers undertake similar reading activities while for example reading and scoring students’ papers and tests. However, students and teachers do not only read texts written by others. They reread their own compositions, test answers and comments in order to edit and improve them but also in order to continue their writing in a coherent way. In this symposium we will focus on the process of “reading during writing”. All papers in the symposium have used eye-tracking in combination with different methods of recording writing in order to investigate where readers look when they perform different writing tasks. Torrance and Galbraith investigate the reading behaviour of university students who are composing argumentative texts. Johansson et al compare reading and writing behaviours of keyboard gazers and monitor gazers. Alamargot et al investigate the process of writing from sources. Quinlan et al study sentence composing and error detection during proof reading. Finally, Solheim and Uppstad have investigated the process of students’ reading while answering test questions.
Equipment	Overhead projector PC and projector
Keywords	Cognitive processes/development Reading Writing

Chair list
Name	Surname	Institution	Country	E-Mail	EARLI Number
Asa	Wengelin	Lund University	Sweden	asa.wengelin@ling.lu.se

Organiser list
Name	Surname	Institution	Country	E-Mail	EARLI Number
Asa	Wengelin	Lund University	Sweden	asa.wengelin@ling.lu.se
Denis	Alamargot	University of Poitiers	France	Denis.Alamargot@univ-poitiers.fr

Discussant list
Name	Surname	Institution	Country	E-Mail	EARLI Number
Kris	Spelman Miller	University of Reading	United Kingdom	k.s.miller@reading.ac.uk
Marie	Stevenson	University of Sidney	Australia	M.Stevenson@edfac.usyd.edu.au

Paper Details

Title	EyeWrite: automatically coding fixation location during text production
Abstract	Meaningful analysis of the focus of writers' gaze within their text is problematic because the text that the writer is looking at is continually changing in unpredictable ways. Methods used in reading research that require definition of fixed areas of interest are, therefore, of no use. However, manual coding from a playback of the writing session with a gaze-location overlay is impracticably time consuming. EyeWrite editing and analysis software locates fixations in the text rather than as screen coordinates, and therefore allows measurement of the distance (in characters, words, sentences, paragraphs) between the last-inserted character and current fixation. We believe that these kinds of measurement are key to understanding reading-during-writing. In our paper we will present a summary of the functionality of EyeWrite. We will then present a preliminary description of the main features of writers' reading behaviour based on a sample of 10 undergraduate psychology students writing short argumentative essays.
Summary	We have fairly minimal understanding of the cognitive processes that underlie text production. Analysis of keystroke logs gives us some idea of where writers pause when they are composing text, and think-aloud protocols give insight into those mechanisms that are associated with conscious, high-level processing. However, we often have little idea of what writers are doing when they pause. In particular, we know very little about writers' reading behaviour. Is pausing typically associated with reading? If so, what prompts writers to look back over their text? Where do they tend to read, how much, and what function does if serve? Reading researchers have for a long time made use of eye tracking to explore where readers look in the text. Eye tracking should be similarly useful in the study of reading during writing. However, there is a problem. Readers' fixations are on static text. Writers, on the other hand, fixate on text that is unpredictably dynamic. This is particularly a problem if the writer is word processing. Not only is the text constantly expanding, but editing, line wrapping and scrolling will change the screen coordinates of part or all of the currently visible text. Standard eye tracking software will provide the screen coordinates of the focus of a writers' gaze, and head mounted eye trackers will provide sufficient accuracy to pinpoint the word that a writer is looking at on a word processor screen (with double spaced text). However, establishing what part of the text the writer was actually looking at at any particular point in time is much more problematic: Methods of analysis based on areas of interest will not work in this context because the screen location of the text is liable to regular change. One option is to manually code a playback of a screen-capture movie overlaid with fixation marker. For anything other than small scale, single-case research this is very time consuming. Answering a straightforward question that, for example, relates pause location to how many sentences back into the text the writer looks or reads during the pause requires coding each of (a) the most recently typed character, (b) the character typed next, (c) the number of sentence-boundaries that lie between the last-inserted character and the location of the fixation, and (d) the length of the fixated sentence (which needs to be controlled for). Each of these measurements must be taken for each fixation. A thirty minute writing session will involve around 2000 within-text fixations. What is required is a system that can locate a fixation not in terms of screen coordinates but at a particular character within the text. This is provided by the EyeWrite editing and analysis programs. The EyeWrite editor writes data for each keystroke and each fixation to an output file. These data include both keystroke and fixation locations measured as the distances (in characters, words, sentences, and paragraphs) from the start of the text. It therefore allows automatic output of the distance, measured in textually meaningful ways, of the fixation from the last inserted character. Further analysis of data in the output file allows a distinction to be made between, for example, fixations on the cursor (during composing), fixations associated with tracking back through the text-just-written to identify errors, scanning through larger bodies of text, and sustained reading. This opens up the potential for a sophisticated and systematic account both of writers' revision behaviour, and of the strategies that they engage in to maintain both global and local representations of the content and structure of their text. In our paper we will present a summary of the functionality of EyeWrite. We will then present a preliminary description of the main features of writers' reading behaviour based on a sample of 10 undergraduate psychology students writing short argumentative essays.
Keywords	Cognitive processes/development Reading Writing
Appendices

Authors
Name	Surname	Institution	Country	e-mail	EARLI Number	Presenting
Mark	Torrance	Staffordshire University	United Kingdom	m.torrance@staffs.ac.uk		*
David	Galbraith	Staffordshire University	United Kingdom	d.galbraith@staffs.ac.uk

Title	Gazing at the Keyboard or the Monitor: Two Different Strategies in Text Production
Abstract	If the translation processes during writing operate fluently, they draw little on the limited working memory processes, something that provides the writer with more resources for planning and revision. And when it comes to computer writing, an important aspect of translation fluency is typing skills. It has for example been shown that fast typists make fewer and shorter pauses than slow typists and have longer and more productive execution-periods. It is generally assumed that a skilled typist uses all her fingers and looks at the monitor most of the time while less skilled typists are supposed to look more at the keyboard.In this paper we compare different aspects of the writing process of "monitor-gazers" and "keyboard-gazers". One group of 15-year olds (N=20) and one group of university students (N=20) took part in the study. Equipped with an eye tracker all subjects wrote an expository text in a keystroke logging program on a computer. Based on the distribution of where the subjects looked during the text production they were categorized as either monitor-gazers or keyboard-gazers. The results indicate that the elder and thereby more advanced readers and writers tend to be "monitor-gazers", that monitor-gazers looks at the keyboard in very brief intervals, and that keyboard-gazers looks at the monitor and keyboard in a similar manner. Monitor-gazers write significantly faster and make significantly more keystrokes but do not produce longer texts. This indicates that monitor-gazers edit their texts more than the keyboard-gazers. One possible explanation for this could be that by continuously monitor their writing, the monitor gazers get a better overview of their texts and therefore better possibilities to edit their texts. Text quality, reading patterns, editing patterns and detailed analyses of fixations and saccades are currently being analysed.
Summary	As for example Bereiter and Scardamalia (1987) and Fayol (1999) pointed out, writing is a complex task that involves several costly components that affect or even steal capacity from each other. For example, McCutchen, Covill, Hoyne, and Mildes (1994) showed a clear relation between translation fluency and writing ability. If the translation processes operate fluently, they draw little on the limited working memory processes, something that provides the writer with more resources for planning and revision. Moreover, Ransdell and Levy (1996) showed that writing fluency (as measured by words per minute) is the single most important predictor of writing quality in university students. When it comes to computer writing, an important aspect of translation fluency is typing skills. For example, Alves et al. (2006) showed that fast typists made fewer and shorter pauses than slow typists and had longer and more productive execution-periods. It is generally assumed that a skilled typist uses all her fingers and looks at the monitor most of the time. Hence, she can monitor her writing continuously. Unskilled typists, on the other hand, are supposed to look at the keyboard most of the time, spending a lot of time finding the right keys. In this paper we explore this issue by comparing different aspects of the writing process of "monitor-gazers" and "keyboard-gazers". Data was collected by means of keystroke logging and eye-tracking. The keystroke logging program used was ScriptLog - which records all keystrokes and all mouse activities and their temporal distribution across a writing session. The eye tracker used was SMI I-view. One group of 15-year-olds (N=20) and one group of university students (N=20) took part in the study. Equipped with an eye tracker all subjects wrote an expository text on a computer. The subjects were asked to write a paper discussing problems that frequently occur in school settings, such as cheating, stealing and bullying. The subjects had the possibilities of looking at the keyboard, at the monitor or outside. Based on the distribution of where the subjects looked during the text production they were categorized as either monitor-gazers or keyboard-gazers. The data was then analysed for age as well as for "gazing" category. Aspects of the finally edited texts, as well as of the production process were analysed. The results indicate that the elder and thereby more advanced readers and writers tend to be "monitor-gazers", that monitor-gazers looks at the keyboard in very brief intervals, and that keyboard-gazers looks at the monitor and keyboard in a similar manner. Monitor-gazers write significantly faster and make significantly more keystrokes. Despite this no differences in text length were found between the groups. This indicates that the monitor-gazers edit their texts more than the keyboard-gazers. A possible explanation for this could be that by continuously monitor their writing, the monitor gazers get a better overview of their texts and therefore have better possibilities to edit their texts. A related explanation is that since they appear to be more fluent writers the monitor-gazers will have more capacity left for editing processes. Another possibility is that they just make more typos and therefore need to erase more. The text quality, reading patterns, editing patterns and detailed analyses of fixations and saccades are currently being analysed. Some preliminary results indicate that when the university students perform regular forward reading the monitor-gazers have significantly longer fixation durations and significantly shorter saccade lengths. This effect seems, however, absent for the 15-year-olds. Another finding is that keyboard-gazers use a significantly higher proportion of "right" cursor keys. It appears as if they use it as a visual guide for their reading. These results will be further discussed after a more thorough analysis of the reading patterns of the writers.
Keywords	Cognitive processes/development Reading Writing
Appendices

Authors
Name	Surname	Institution	Country	e-mail	EARLI Number	Presenting
Roger	Johansson	Lund University	Sweden	Roger.Johansson@ling.lu.se		*
Asa	Wengelin	Lund University	Sweden	asa.wengelin@ling.lu.se
Victoria	Johansson	Lund University	Sweden	victoria.johansson@ling.lu.se
Kenneth	Holmqvist	Lund University	Sweden	kenneth@lucs.lu.se

Title	Writers juggling error-detecting with sentence composing: Influences of cognitive load and error-type
Abstract	In the following two experiments, we examined proofreading in the context of sentence composing. We devised an experimental writing task in which participants corrected an embedded error (orthographic near-neighbors or far-neighbors) and completed a sentence (using 1 or 3 context words). Experiment 1 investigates how the cognitive demands of sentence composing influences proofreading performance. The results revealed that participants were more successful at (a) integrating 1 context word than 3 context words and (b) correcting far-neighbor errors than near-neighbor errors. In Experiment 2, we examine how sentence difficulty and error type influences the adoption of writing strategies. The results of Experiment 2 revealed that both error type and sentence complexity influenced the strategies of error-correcting. Participants most often (90% of the time) opted to complete the sentence before correcting the error. The eyetracking data revealed that error-type also influenced how participants coped with correcting errors. In the analysis of fixation transitions (into the error-zone), a significant main effect was revealed, with more transitions for near-neighbor errors than far-neighbor errors. This investigation reveals how the interplay of two factors, cognitive load and error-type, appears to influence how writers coordinate error-detecting with sentence composing.
Summary	Even highly skilled writers rarely get it right the first time. Consequently, writers endeavor to improve their texts, either through revising or editing. While considerable research has focused on the role of revising (cf. Allal, Chanquoy, & Largy, 2004), relatively little work has examined the roles of editing or the coordination of editing with other writing processes. In the following two experiments, we examined proofreading in the context of sentence composing. We devised an experimental writing task in which participants corrected an embedded error and completed a sentence. The procedure in the present experiment was designed to emulate salient aspects error-detecting and correcting in the context of sentence composing. As a writer formulates his or her thoughts into sentences, errors can occur during inscribing. The writer may or may not detect the occurrence of an error. When an error is detected, the writer may correct it immediately or later. If he or she corrects it immediately, then it will seemingly interrupt other writing processes. Because our aim was to investigate the effects of error correcting and sentence composing, we were precluded from assigning some kind of authentic writing task (e.g., writing a job letter to a potential employer). In order to simulate the conditions of error detecting/correcting, as a writer might encounter them during sentence composing, we opted for a cloze task. By varying the particular conditions of these writing tasks, i.e., error type and sentence complexity, we were able to investigate specific research questions. To vary error type, fifty percent of sentences had embedded errors, either orthographic (a) near-neighbors or (b) far-neighbors. Each experimental item consisted of four screens, presented sequentially: 1. An opening screen, which the participant clicks to begin the item 2. A blank screen, during which the first part of a complex sentence is read aloud (auditory priming), then repeated. 3. One or three context words are displayed (in a line, separated by commas) at the bottom of the screen, for 1 s and 3 s, respectively. Participants were instructed to memorize these words. 4. The partial sentence (as read aloud in screen 2) is now displayed on the screen, which may include an error. Participants were directed to use the keyboard and mouse to, in the following order: a. Correct the error, if present b. Complete the sentence, using the all the context words (either 1 or 3) displayed on screen 3. Experiment 1 investigates how the cognitive demands of sentence composing influences proofreading performance. In this experiment, subjects were obliged to first correct the error (if present), then complete the sentence. The results revealed that participants were more successful at (a) integrating 1 context word than 3 context words and (b) correcting far-neighbor errors than near-neighbor errors. The latter result suggests that orthographic near-neighbors were more likely to go undetected. Given the results of Experiment 1, we decided to focus upon the strategies employed by writers to cope with the demands of sentence composing and error correction. Thus, in Experiment 2, we examine how sentence difficulty and error type influences the adoption of writing strategies. For this experiment, we employed an eyetracking system to observe error-detecting. Given the load associated with holding context words in working memory, we hypothesize that participants will tend to complete the sentence first, then correct the error. The possibility of deferring error correction raises the question about when detection happens: Might participants detect an error but defer its correction until after completing the sentence? To address this question, we employed an eyetracking system to record the reading behavior of a subset of the participants. With the Eyelink II eyetracking system, we recorded participants’ eye fixations as they perform the experimental tasks. Analysis of reading behavior ( in terms of fixation duration, frequency, and transitions in look zones) should reveal whether errors are detected on first-pass or subsequent rereadings. The results of Experiment 2 revealed that both error type and sentence complexity influenced the strategies of error-correcting. Participants most often (90% of the time) opted to complete the sentence before correcting the error. This approach is cognitively efficient, because it allows the participant to “unload” working memory before starting to detect and correct errors. However, in some cases, participants did correct errors first, and did so significantly more often with simple sentences (1 context word) than complex sentences (3 context words). This result suggests that cognitive load influences the automatic shift between processes (error-detecting/correcting and sentence composing). The eyetracking data revealed that error-type also influenced how participants coped with correcting those errors. In the analysis of fixation transitions (into the error-zone), a significant main effect was revealed, with more transitions for near-neighbor errors and far-neighbor errors. This result suggests that detecting near-neighbor errors requires more reading analysis than far-neighbor errors. This investigation reveals how complexity manifests in even the seemingly mundane aspects of writing, coordinating error-detecting with sentence composing. The interplay of two factors appears to influence aspects of this coordination. First, the cognitive load of sentence composing appears to affect the writer’s ability to shift attention toward error-detecting. When the demands of sentence composing are high, few resources are available for analyzing the written text, in order to detect errors. However, as the demands of composing ease, and more attentional resources become available, the writer becomes more able to contemporaneously analyze the written text for errors. Another influence is error-type. It appears that detecting orthographic near-neighbors requires more reading analysis. This helps explain why this type of error frequently goes undetected.
Keywords	Cognitive processes/development Reading Writing
Appendices

Authors
Name	Surname	Institution	Country	e-mail	EARLI Number	Presenting
Thomas	Quinlan	Educational Testing Service	United States	thquinlan@gmail.com
Maaike	Loncke	University of Ghent	Belgium	maaike.loncke@ugent.be
Marielle	Leijten	University of Antwerp	Belgium	marielle.leijten@ua.ac.be
Luuk	Van Waes	University of Antwerp	Belgium	luuk.vanwaes@ua.ac.be		*

Title	K-means method to assess the visual strategy of the writer composing from sources
Abstract	Writing from sources is a frequent task at the workplace and its importance increases with the use of computers. This task requires a double competence in both reading-extracting information and inventing-composing text. The ‘Eye and Pen’ system was designed to study these skills. Based on a synchronous recording of eye movements (via an eye-tracking system) and pen movement (via a digitizing tablet), the device provides a fine-grained description of the visual strategies used by the writer while composing. To study these strategies, we recorded the graphomotor and eye activity of 25 adults while they were composing a procedural text by referring to documentary sources. From the visual patterns, some statistics are computed. One of them is the transition matrix indicating the frequency with which the eye moves from one zone to another zone. Other statistics such as the average number of different pieces of information explored when the eye leaves the writing zone are also considered. From these various statistics, a measure of dissimilarity between respondents is computed and clustering (K-means) is used to find groups of respondents with similar exploration patterns. The clustering results show that two groups of writers can be classified regarding their visual exploration activity. Further investigations show that these two groups also differ regarding their respective working memory capacity and age. We demonstrated that K-Means is relevant here to distinguish various visual strategies during writing and this K-Means categorization seems to be cognitively valid.
Summary	Since Hayes and Flower’s (1980) seminal work, writing research has aimed at solving two important issues: (i) to further describe the processes involved in writing, and (ii) to understand the dynamics of these processes (i.e., why and when a process occurs in parallel rather than to replace another process). Among the different ways of conducting real time analysis of writing processes, recording the combined measurements of ocular  and graphomotor activities of the writer can be a fruitful method. In this way, the ‘Eye and pen’ device (Chesnet & Alamargot, 2005 ; Alamargot, Chesnet, Dansac & Ros, 2006) provides a very fine-grained description of the temporal characteristics of written production and offers a new framework to understand the writing processes. From a theoretical point of view, the ‘Eye and Pen’ system represents a powerful way to infer the nature of writing processes, by bridging the gap between visual input and graphomotor output. ‘Eye and Pen’ relies on two main devices: a digitizing graphic tablet (to record spatial coordinates and pressure of the pen on the tablet surface) and an eyetracker (to record eye movements). All these observations are stamped with a common base millisecond timing. From the text written by the participant, and digitalized by the tablet, one may rebuild forward and backward on the computer screen, the trace leaved by the pen and the eye position at the same time (synchronized events). It becomes possible to improve investigations on processes engaged during the course of a pause as well as during a period of transcription. At an experimental level, this device will allow advances in the study of the visual component engaged during writing and of the functioning and dynamic of writing processes. Among the different way to compose a text, writing from sources is a frequent task at the workplace and its importance increases with the use of computers. This task necessitates a double competence in both reading-extracting information and inventing-composing text. The ‘Eye and Pen’ system was designed to study these skills. To study these strategies, we recorded the graphomotor and eye activity of 25 adults while they were composing a procedural text by referring to documentary sources (a turbine model – see Alamargot, Dansac, Chesnet & Fayol, in press). From the visual patterns, some statistics are computed. One of them is the transition matrix indicating the frequency with which the eye moves from one zone to another. Other statistics such as the average number of different pieces of information explored when the eye leaves the writing zone are also considered. From these various statistics, dissimilarity between respondents is computed and clustering (K-means) is used to find groups of respondents with similar exploration patterns. The clustering results show that two groups of writers can be classified regarding their visual exploration activity. Further investigations show that these two groups also differ regarding their respective working memory capacity and age. We demonstrated that K-Means is relevant here to distinguish various visual strategies during writing and this K-Means categorization seems to be cognitively valid. Multivariate analyses confirm the existence of contrasted visual patterns, depending on the writer characteristics. The Eye and Pen device represents a key element to study the dynamics of handwriting from source. It constitutes a powerful computer environment able to catch very fine grained phenomena related to ocular and graphomotor activities. It also provides to the researcher an easy-to-use tool helping to understand the temporal and the spatial relationships between the eye and pen indicators. Used in association with Eye and Pen, data-mining and discovery methods offer interesting perspective regarding the nature of recorded data, actually in the understanding of the complexity of the eye movements during a writing task.
Keywords	Cognitive processes/development Reading Writing
Appendices

Authors
Name	Surname	Institution	Country	e-mail	EARLI Number	Presenting
Denis	Alamargot	University of Poitiers	France	denis.alamargot@univ-poitiers.fr		*
Gilles	Caporossi	HEC Montreal	Canada	gilles.caporossi@hec.ca
David	Chesnet	University of Poitiers	France	david.chesnet@mshs.univ-poitiers.fr
Christine	Ros	University of Poitiers	France	Christine.ros@univ-poitiers.fr

Title	Students integration of text and illustration in an assessment of reading comprehension: relations between reading and reading while answering
Abstract	The present study explores the idea that eye-tracking methodology can validate the intended function of items in a reading comprehension test, as suggested by Pearson & Hamm (2005). Eye-tracking methodology has been widely used for reading (Rayner, 1998), but largely ignored in educational assessment (Tai et al., 2006) and reading assessment. In eye-tracking studies one has usually studied the first reading of the text, and this reading behaviour has sometimes been to measures of reading comprehension (Hannus & Hyönä, 1999; Hyönä et al., 2002). As a consequence, the actual process of reading when answering has not been focused. Assessments of reading comprehension typically consist of reading passages accompanied by a set of multiple-choice and constructed response questions. The present study aims at both the reading of the text (condition 1) and the reading performed while the student is answering questions related to the text (condition 2). This kind of design opens up for the study of different strategies.
Summary	Assessments of literacy have embraced the notion of functional literacy. When talking of functional literacy, one typically focuses on using printed and written information to function in society, to achieve one’s goals and to develop one’s knowledge and potential. Functional literacy is directly related to authentic reading situations and a wide range of everyday texts (including graphs, diagrams etc.) thought requisite to adequate adult functioning (Elbro et al., 1991; Kirsch, 2001; Kirsch & Guthrie, 1978, OECD, 2003). Assessments of childrens reading literacy have also changed to include both the understanding and use of authentic texts for different purposes (Campbell et al., 2001; Purves, 1993; Pearson & Hamm, 2005). For children one of the most important purposes of reading is the acquisition and use of information. At a certain age, children are expected to use their reading skills as a tool for learning, and the reading of textbook passages is a daily activity. Textbooks most often consist of both text and illustrations where illustrations play different roles. In some contexts they are peripheral to the understanding of the text while in other contexts they are crucial for learning. There is, however, little focus on the function of illustrations in these books, and the student has to discover the function of the illustration by himself. Gee (2003), and Johnson & Kress (2003) stress how meaning resources beyond language (for instance images and graphs) are combined in integral and complex ways with each other and words, and that this understanding must lead to changes in the way reading is taught, assessed and considered successfully mastered. According to their view this integration should also be included in assessments of reading. Giving tasks that require such integration in order to be correctly answered can be one way of doing this. In the present study, one of the questions in the assessment had such characteristics. When tasks that require integration of text and illustration are introduced, it becomes important to investigate whether such a task has the intended function. With a traditional approach, one gets a comprehension score at the end. A more sophisticated approach will be to relate these scores with actual behaviour. This is the aim of the present study. Eye-tracking methodology has been widely used for reading (Rayner, 1998), but largely ignored in educational assessment (Tai et al., 2006) and reading assessment. In eye-tracking studies one has usually studied the first reading of the text, and this reading behaviour has sometimes been to measures of reading comprehension (Hannus & Hyönä, 1999; Hyönä et al., 2002). As a consequence, the actual process of reading when answering has not been focused. Assessments of reading comprehension typically consist of reading passages accompanied by a set of multiple-choice and constructed response questions. The present study aims at both the reading of the text (condition 1) and the reading performed while the student is answering questions related to the text (condition 2). This kind of design opens up for the study of different strategies. The technical equipment used was the SMI iViewX Helmet with Polhemus magnetic headtracking. The sample consisted of 20 students in 7th grade (12-years olds). The students were evaluated as normal achievers by their teachers, and they were evaluated as normal achievers based on standardized scores in the Raven nonverbal intelligence test and a standardized word chain test. The texts read were two authentic pages of a science textbook from 7th grade on the subject ‘the human ear’ containing both text and illustration. The comprehension of this text is to a large extent dependent on the illustration and vice versa. The recording was undertaken in two steps. First, the students were told to study carefully all information in the textbook passages, and that they should be able to answer some questions related to the text after having read the text. In the second recording they were asked to fill out a questionaire. When answering the questionaire they had the opportunity to look at the science book. This procedure in two steps was repeated with the authentic science textbooks, giving a total of four recordings for each subject. The data will be analyzed according to two conditions. The first condition relates to the integration of text and illustration when the student reads the text for the first time. The second condition relates to how (or to what degree) the student is reading the science book passage while answering comprehension tasks. In this second condition it will be focused upon a specific question which requires integration of text and illustration. Condition 1 – reading for the first time Is there a tendency in a) how much time these two groups spend on text versus illustration? b) how many transitions between text and illustration these two groups have? c) how many transitions between text and illustration at relevant locations these two groups have? Condition 2 – reading while answering Is there a tendency in a) how much time these groups spend on text versus illustration? b) how many transitions between text and illustration these two groups have? c) how many transitions between text and illustration at relevant locations these two groups have? d) how much time these groups spend on the questionaire versus the science book? e) how many transitions between questionnaire and science book these two groups have? In the research literature on reading assessment there is little behavioural data on the working process during assessment. The present study explores the idea that eye-tracking methodology can validate the intended function of items in a reading comprehension test. In international reading tests, different types of questions are intended to seize different comprehension processes. The relation between intentions and processes remains an empirical question. The present study raises two main questions related to assessment of reading comprehension: 1) Does a particular comprehension question (item) seize the kind of reading behaviour that we assume? 2) Is this reading behaviour elicited by the question itself, or is it representative for the way students read without being guided through questions?
Keywords	Assessment Reading Writing
Appendices

Authors
Name	Surname	Institution	Country	e-mail	EARLI Number	Presenting
Oddny	Solheim	University of Stavanger	Norway	oddny.j.solheim@uis.no		*
Per Henning	Uppstad	University of Stavanger	Norway	per.h.uppstad@uis.no