Eye tracking research applications: Three case studies
Contents
1. O the grass has thickened on the verges Q2 According to the writer things used to be O worse abroad the same abroad better abroad worse but now things are better abroad 36 Q3 For the writer the problem is that C C C C rubbish is not cleared up rubbish last longer than it used to our society is increasingly mobile Britain is a tatty country Q4 Michael Meacher Cc C C C followed the Irish example with a tax on plastic bags tried to follow the Irish example with a tax on plastic bags made no attempt to follow the Irish example with a tax on plastic bags had problems with the plastics industry who weren t bothered about the tax Q5 The writer thinks C C C C it is too late to do anything we are at the tipping point there is no alternative we need to work together to solve the problem Q6 The writer thinks that E C C C people are squalid people behave according to what they see around them people are clean and tidy people are like a vast municipal rubbish tip 37 8 BIBLIOGRAPHY Aziz W Sousa S C M and Specia L 2012 PET a tool for post editing and assessing machine translation In The Eighth International Conference on Language Resources and Evaluation LREC 12 Istanbul Turkey online Available at lt http pers www wlv ac uk in1676 publications 2012 AZIZ LREC2012 pdf gt Accessed 10 July 2012 Bohme M Meyer A2. and two P_2 path just below the words for the same sentence part in the same test is in the archetypal situation of the Romantic Wanderer or Solitary who sometimes was a poet can win no release from his own story Figure 12 P_1 Test 3 is in the archetypal situation of the Romantic Wanderer or Solitary Ones a poet can win no release from his own story Figure 13 P_2 Test 3 The determining factor to define eye tracker accuracy in this short set of experiments was how far the recorded scan path was from each line being read If the tracks were in the middle or at the edges of the lines touching the top or bottom 12 of a line they were considered accurate In both figure 12 and 13 the accuracy is acceptable whereas the scan path in figure 9 is considered inaccurate as it is unclear which line the participant is actually looking at The text was analysed both line by line and as a whole If the majority of the lines in a test were recorded accurately the tracker performance in the test was considered accurate overall Generally speaking the results were better when using shorter texts but not necessarily larger fonts Out of the four sizes considered 18 20 22 24 overall performance was better with size 22 points Test 3 where less scan paths fell in between the print lines Quite surprisingly however Test 1 Arial 18 points the smallest font adopted was fairly accurate too seeming to sugges
3. the key number will already appear in the dialogue window Just click on Activate Now and the programme will open If the last user has not deactivated the licence on their profile it will not be possible to run Tobii License Data TES The suppied license key fECTFB 5FBCE 07077 3CDA 5 is for an old version of Toby Studio Please suppi a new jicense key License key ee Organization A E mail optional v Activate Later Cancel Figure 4 Tobii Studio licence key activation Creation of new project and choice of stimuli Once the programme is open a new project and subsequent test s can be created A project in Tobii can contain several tests made of different stimuli please refer to Tobii Studio 1 X User Manual pp 9 Configuration Tool During the setup information on the exact placement of the tracker in relation to the screen must be provided The different available parameters are listed in figure 5 below For more detailed information please see the Tobii X120 Eye Tracker User Manual pp 8 12 A digital angle ruler has been provided to measure the angle width where necessary Please note that for computer monitors and other types of supported displays the height and width parameters should refer to the visible active area on the screen where one can see the actual image rather than the whole screen The standard eye tracker angle is usually around 30 degrees Due to a lack of physical
4. the length of which is exactly the same as the length of the first and second subtitle respectively Scenes are portions of a recording that need to be associated with a still image This frame will be the background on which the AOIs will be selected As the AOI in the first scene will be the misspelled word poser the static image for Scene 1 will need to contain the subtitle where the word appears The same thing goes for the background image for Scene 2 which will have to show the second subtitle containing the word power Once the correct Scene Images were selected the Visualisation view allowed the creation of the AOIs Figure 24 These can be 25 copied pasted renamed and be assigned differet colours for ease of identification s Trail 26 07 video Tobii Studio 2 22 0 aigi x Fiber by No Group Selected Narre Coit Sargies Participant F ORe a x vse Every step that takes Franco closer to in Spain Areas of Injorest Rora Basnsye HE Show Narre Labels Racargie Foe gon Aa Badwi j a Change Fort I ees reer Sie Darcorainge Q Areng of ueerent Tools A start BUO Birras 3 Material SS tyetrec W urtid Cp docune S Tobe stu i Tras 20 Screens ZShes JoeB uy Figure 24 Creation of the first AOI on the misspelled word poser Statistics view In this view it is possible to select which scenes and related AOIs to import for analysis and which metrics to run In this exper
5. the tracker is recording the user s gaze accurately If the user is focussing on a red dot but the circle does not contain the dot the tracker is having difficulties in detecting the real position of the eye When testees were looking at the three bottom dots the tracker was consistently less accurate with the red dots rarely being in the centre of the circle 13 glasses and is short sighted his right eye being considerably worse than the left This difference was detected by the eye tracker as confirmed by the track status box in the calibration process where the device at times struggled to detect his right eye as well as in the calibration results where the right eye showed long error vectors green lines The participant s eyelids were not droopy nor were the spectacles bifocal dirty or scratched The person was sitting correctly in front of the screen and the room lighting was the same as in all the other tests Several tests were carried out on two separate occasions and the results were the same This could suggest that the Tobii tracking performance with people wearing glasses could be lower when common vision problems are very pronounced Despite the claim that Tobii eye trackers have a very good tracking ability on people with glasses Tobii T X Series Eye Trackers Product Description pp 13 the device did perform considerably less well with two out of three testees wearing spectacles This could be due to a number of r
6. Questionnaire media types with the scan paths falling mostly in the middle of each line at times touching the bottom or top of the print In the PDF gaze paths tended to be above the print lines making the tracking of ocular movement less accurate Mouse clicks were recorded precisely not only in the Questionnaire which is designed to contain MCQs answered through mouse clicks but also in the other two stimuli The font used was Arial 16 points line spacing 1 5pt for the text and 14 points for the questions Considerations about the Questionnaire stimulus It is worth commenting on this stimulus as the Tobii Studio User Manual 1 X does not mention it and only a very brief and basic description can be found on pp 37 39 of the newer release Tobii Studio User Manual 2 X This stimulus is less flexible as the following aspects of text appearance cannot be changed or customised by the user 17 Formatting Font size type and other formatting options are pre set When typing or copying text in the question box the formatting is automatically removed meaning this stimulus cannot be used when comparing fonts for example in legibility and readability studies Text length At first there did not seem to be a set length limit as it was possible to type or paste large amount of text in the in the question set up window However if adding too much text one will end up with the following situation during recording Dirty Britai
7. accurately and give an good representation of gaze movement Two examples of tracking for the same scene from two different participants are shown below Both the trackings of Figure 19 and 20 were considered acceptable although in the second recording the scan path is slightly above the subtitle line 21 That s why there aren t any more youth movements Figure 19 Eye movement in Viewer_1 That s why there aren t any more youth movements Figure 20 Eye movement in Viewer_2 It must duly be noted that in most cases a participant s recording was a mix of accurate and less accurate tracking Subtitle size was suitable yet perhaps the position was too low on screen This is not particularly noticeable with the two liners above but became evident with one liners such as the one displayed in Figure 21 as these appeared on the bottom line rather than the top one thus sometimes causing some of the fixations to be only partially displayed on screen What was considered subversive then Figure 21 Here the subtitle position is too low and tracking is inaccurate hence the scan path ends up falling outside the active screen 4 b 2 Land and Freedom A practical application The second clip was taken from the 1995 film Land and Freedom directed by Ken Loach In this case a specific research question was conceived and a short experiment was designed in order to be able to give at least one practical example of how an e
8. be the easiest way namely downloading and installing the right codec as most users will not have enough administration rights to perform this action This practical issue can be resolved by liaising with IT to be granted the necessary rights choosing a different clip if possible or converting the chosen file using a different encoding compatible with Windows Media Player and Tobii 4 Subtitles need to be burnt onto the video track In the present study Any Video Converter has been used The merging of subtitles with the video can also be done with other video editing programmes such as Video Pad Power Director or Virtual Dub Once the subtitles have been burnt in it is recommended to do a trial test in Tobii to make sure their size and position is suitable 5 Tobii Studio does not resize the video which is played in actual size but limited to the eye tracker s screen resolution With the Tobii X120 Eye Tracker the maximum video size is 1280x1024 pixels therefore files with larger resolutions will be cropped during media presentation To avoid this the video needs to be resized using a video editor such as the ones mentioned above The screen resolution used in all the clips analysed here is 1024x768 The resolution must also be the same as that of the PC monitor so before making a choice it is advisable to check what sizes the monitor supports 6 The two media options of choice for working with audiovisual material were Screen Re
9. the undergrowth for years a semi permanent reminder of what a tatty little country we have now Firstly it is estimated that 10 billion plastic bags have been given to shoppers These will take anything from 100 to 1 000 years to rot However it is not as if there is no solution to this A few years ago the Irish government introduced a tax on non recyclable carrier bags and in three months reduced their use by 90 When he was a minister Michael Meacher attempted to introduce a similar arrangement in Britain The plastics industry protested of course However they need not have bothered the idea was killed before it could draw breath leaving supermarkets free to give away plastic bags What is clearly necessary right now is some sort of combined initiative both individual and collective before it is too late The alternative is to continue sliding downhill until we have a country that looks like a vast municipal rubbish tip We may well be at the tipping point Yet we know that people respond to their environment If things around them are clean and tidy people behave cleanly and tidily If they are surrounded by squalor they behave squalidly Now much of Britain looks pretty squalid What will it look like in five years Questions Q1 The writer says that it is a good time to see Britain before the trees have leaves because Britain looks perfect you can see Britain at its dirtiest you can see how dirty Britain is now
10. to make sure it worked fine For the reasons described above the Video media type was chosen for all the recordings A project was created and tests were performed with a total of 7 people watching the subtitled clip The number of participants is obviously too small to consider any results usable however it has to be borne in mind that the research question was chosen purely for the sake of demonstrating how the software works rather than to prove a specific point None of the conclusions reached is claimed to be valid or usable for anything other than providing practical evidence as to how Tobii can be used in AVT studies Replay and Visualization views Once a recording has been made it can be played back in the Replay view where the user can see the tracking overlaid onto the video The time bar can be rescaled to provide greater detail and show shorter time intervals Consider figure 22 and 23 below KEN LOACH Land and Freedom_1024x768 MPEG4 avy Fal New Scene_2 Ei New Scene_1 Scenes Opacity Trail duration Overlay color Replay UserCamere Snap to time markers IV Draw Gaze Data 0 500 Seconds m M on IV on IV Draw Mouse Clicks Transparent Dana cial Bias ET E Segments Scenes Settings Recording Rec 01 length 00 03 23 783 participant Vale Figure 22 Timeline before zooming in 24 F 00 02 26 057 EE Le gt p 000242000 000251000 00
11. 0300000 0003090 yT Ca New Scene_2 g New Scene_1 o oj on i Opacity Trail duration Overlay color Replay User Opmere T Shap fo Eve mai V Draw Gaze Data z z 3 l t l 0 500 Seconds ii On On V Draw Mouse Clicks Transparent pane MONETE E Segments Scenes Settings Recording Rec 01 length 00 03 23 783 participant Vale Figure 23 Timeline after zooming in In figure 23 the intervals between the clip length indicators the numbers in the blue bar are 9 seconds whilst in fig 22 they were 30 seconds Note how the highlighted grey segment indicating New Scene_1 is now wider compared to fig 22 while the horizontal scrolling bar also highlighted is smaller Zooming in to the timeline is useful when it is important to be able to advance the replay slowly for example when having to select an in and out time for the creation of a scene Morever by zooming in it is easier to see where a scene starts and ends once it has been created The clip can also be navigated frame by frame using the arrows keys on the keyboard which proves very useful to identify the exact frames on which a subititle appears and disappears The playback speed can also be changed to play the recording a slower or faster pace The next step involves creating scenes i e slecting a part of the recording This is a necessary operation in order to identify AOIs and be able to run statistics In this test two scenes were created
12. 2 33 7 APPENDIX Case Study 1 a A small excerpt from from a study of Mary Shelley s Frankenstein or The Modern Prometheus written by Harold Bloom Available from http legacy owensboro kctcs edu crunyon CE Frankenstein Bloom 38 39 Bloom htm A Romantic poet fought against self consciousness through the strength of what he called imagination a more than rational energy by which thought could seek to heal itself But Frankenstein s daemon though he is in the archetypal situation of the Romantic Wanderer or Solitary who sometimes was a poet can win no release from his own story by telling it His desperate desire for a mate is clearly an attempt to find a Shelleyan Epipsyche or Blakean Emanation for himself a self within the self But as he is the nightmare actualization of Frankenstein s desire he is himself an emanation of Promethean yearnings and his only double is his creator and denier 34 b Excerpts from a text on the Chinese New Year taken from the Web http wenku baidu com view c9386127192e45361066f520 html HEF 58 a A EIEN 2010 F ARITA KHAA be Aya ek aS AR ED A BE ok FTE SANTEE RT ABH A Se Ae OE Het Ae Fe We aie BK IERTE A 4 BIDAR I FFER FEA ADS HI ME FE Te TRORA JHAR IAT MLE OR te Fr ae He EPS CPP A A RA PL SCE FEF EERIE ESS FE IR SS SATS MPAA iB 2 2 EE 1A Bal A Bry Sie 5 20 ab Oe C
13. AS EAE TR SINE BA AEP SCY IE TEM NAGE DAKE SA Abo ee HA SEENA EW AE YR ITEP A Pe Se A HAMM h HAA Ait ADI S HSER SUERTE RE BR OA AYE SPN AE PLACE BALE ABE BI LAST Be tA E APE PEEN RKD EMEWA JE EL WY Die JLE WAY RW BLOC EAT RAC ARR EL HAE SP RYT WWE RMEAPRIOE LS TEER WORE SII EATERS FHE Se FU R A FY p gt iy 2002 Rik ART PAE A AZ as A EEEE E Er AB AL iho ER Fel A wi Ze AB BOR BU HEH Se PP 2a AI EAA M AOA ATR ET A MP Eit BREE AS TT AY ARIE oh th E AR AH S REAA Be at Se FEL AO REA fA HE WS Hi FTTH 35 c Reading comprehension Dirty Britain available from http www usingenglish com comprehension 29 html Dirty Britain Before the grass has thickened on the roadside verges and leaves have started growing on the trees is a perfect time to look around and see just how dirty Britain has become The pavements are stained with chewing gum that has been spat out and the gutters are full of discarded fast food cartons Years ago I remember travelling abroad and being saddened by the plastic bags discarded bottles and soiled nappies at the edge of every road Nowadays Britain seems to look at least as bad What has gone wrong The problem is that the rubbish created by our increasingly mobile lives lasts a lot longer than before If it is not cleared up and properly thrown away it stays in
14. ERRERAS A EPERERA STH PAARS RAR AB CTE TRONE gt ti SR IO MME RAKRSiwEP THEA AAA EMRAT PEARLA SRHRSSS RABRAMN DA MPRA STK Kt AE Figure 15 Characters displayed correctly during preview in Arial Unicode MS Smaller sizes were also tried with this font 20 18 14 points at 1 5 line spacing and the characters were displayed correctly However at 14 point size the characters started to be slightly blurred Several other font types were tried at 20 points some of which are listed in the table below Some fonts displayed the text correctly however most of the fonts tested in MS Word caused character corruption 15 Font type Character Corruption Characters displayed correctly Arial x Arial Unicode MS X MingLiu X Verdana x Lucida Sans Unicode x SimHei x DotumChe x Table 2 Some of the fonts tested with Chinese characters in Tobii Studio Once a usable text formatting was identified a reading test was performed Tobii shows eye movement on a sentence level but does not allow character detection nor does it distinguish which strokes or parts of the character the reader is looking at When the reader was spending more time on one or more adjacent characters looking at the component parts Tobii simply registered a very long fixation a larger circle overlaid on the text The scan path resulted in a series of long fixations which didn t reflect accurately what the
15. Eye Tracker Studies Author Valentina Ragni MA PI Serge Sharoff Centre for Translation Studies Table of Contents 1 Introduction and objectives sessessseseeesssseeeeesssseressssseeessreessssseeresssso 3 2 Tobii eye tracker and software essessseesesssseeesssseeesssseeessssreeesssereesss 4 3 Preliminary set Up sessssssseseseeressseeeesssseeeessssreesssseresssseeessssreeesssereessse 5 Ai CASE SIM GICS A E A A AEEA E E NEA 9 4 a Case study 1 Reading a text ccccccssssecccesssrecccssssscccessssseees 9 4 a 1 Reading a text in English sessoeseeessseeessssseeesssseeessssreees 9 4 a 2 Reading a text in Chinese seeeessesseeeeessseeesssseeesssseresss 14 4 a 3 Reading comprehension with MCQs in English 17 4 b Case Study 2 Watching subtitled video material 19 4 b 1 The Edukators General considerations 000000000000 19 4 b 2 Land and Freedom A practical application 23 4 c Case study 3 Translating a text in a separate application 28 5 Limitations and ODStacles ciscssssesicasivegeadiavccse cownwaedoedies eassaddoscacetaeoevecess 31 6 Conclusive TEMIAT KS vs cisiscecans seers E E Ea 32 7 RTD OC INI IE ireann eE EEEE Shas ate ond odaon EEEE EE ERa 34 SUBIDMOSLA PHY sssr a a a a EE EEE aa SEE O ens epee 38 1 INTRODUCTION AND OBJECTIVES In recent years remarkable technology advancements have be
16. Martinetz T and Barth E 2006 Remote Eye Tracking State of the Art and Directions for Future Development In The 2nd Conference on Communication by Gaze Interaction COGAIN 2006 Gazing into the Future online Available at lt http www inb uni luebeck de publikationen pdfs BoMeMaBao6 pdf gt Accessed 20 June 2012 Bloom H 1965 An excerpt from a study of Frankenstein or The New Prometheus In Partisan Review 32 4 pp 611 618 online Available at lt http legacy owensboro kctcs edu crunyon CE Frankenstein Bloom 38 39_Bloom htm gt Accessed 25 June 2012 Carl M 2012 Translog II a Program for Recording User Activity Data for Empirical Reading and Writing Research In Proceedings of the Eight International Conference on Language Resources and Evaluation LREC 12 online Available at lt http www lrec conf org proceedings lrec2012 summaries 614 html gt Accessed 25 July 2012 Corley M 2010 Making predictions from speech with repairs Evidence from eye movements In Language and Cognitive Processes 25 5 pp 706 727 online Available at lt http www tandfonline com doi pdf 10 1080 01690960903512489 gt Accessed 25 June 2012 Duchowski A T 2002 A breadth first survey of eye tracking applications In Behavior Research Methods Instruments amp Computers 2002 34 4 pp 455 470 Dyson M 2004 How physical text layout affects reading from screen In Behaviour and I
17. Tobii Studio 2 X User Manual pp 9 15 In this study the default Tobii fixation filter has been used Users should be aware of the meaning and the different effects that choosing different parameters will have on their research before setting up an experiment 6 CONCLUSIVE REMARKS The case studies presented above were not intended to provide a comprehensive description of the several alternatives available in Tobii as far as experiment design set up implementation and statistics are concerned Points of interest in each of these stages were identified and tested using some of the options offered by the software Initially the third case study was to include testing with two other separate applications namely PET and Translog II PET is a tool for the post editing of translations and Translog is a piece of software for recording and studying reading 32 and writing processes on a computer which allows to monitor various aspects of the translation progresss The idea was to study eye tracker performance when used in conjunction with these external applications relevant to linguists and translators in particular However due to lack of administrator rights it was not possible to install and run the two applications and therefore only testing of translation in MS Word was carried out Moreover a fourth case study was envisaged where eye tracking was to be used as a support in the analysis of human response to the hearing of a voice stimu
18. al confusion as to which line the participant s is actually looking at Figure 10 shows the gaze plot on a participant s regression at a defined moment A few frames later Figure 11 the same fixations appear to be related to the line below that of figure 10 This is because the test was taken using the Screen Recording stimulus and the participant had to scroll down the page to keep reading discarded fast food cartons Years ago remember travelling abroad and being saddened by the plastic ta discarded bottles and soiled nappies a f edge of every road Nowadays Britain D look at least as bad What has gone wrong i Questions Figure 10 discarded fast food cartons Years ago remember travelling abroad and being saddened by the plastic baas discarded bottles and soiled nappies at 3 edge Ox road Nowadays Britain seems at t least 9 as bad What has gone wrong Questions Figure 11 11 In fact in the PDF view scrolling is not necessary even if the text is longer than a page as the content of each page is displayed all at once so participant can read without scrolling and just need to click on the spacebar to move to the next page However as the initial idea was also to compare tracker performance for the same text in the four media types mentioned above which would have involved scrolling in the case of screen recording the excerpt was shortened considerably so that it could fit on one page and allow f
19. arato al comitato per l an cambsamento climatico una bufala che continua ad essere raccontat Fotografia Bili Clark Getty images Tmiglion mancano di convinzione mentre gion si distir passione Queste parole prese dal roena in B Yeats l inseme dee coscienze indiiduali Da il nome alla prema puntata della terza stagione di Mero Collegamenti esterni p s Figure 28 Accurate tracking while using websites for reference during translation In Tobii Studio the desired folder path location of the files to be opened at the beginning of the test must be indicated in the Screen Recording dialogue window Only one path can be entered meaning only one of the two windows will be displayed and the user will have to align the other during the recording before starting the actual translation To save time this can be avoided by aligning the two documents before commencing the test If only those two MS Word windows are open and arranged in the desired way they will appear side by side when starting the recording 30 In the horizontal alignment the visible page space is much smaller meaning not as much text is displayed on screen However this view entails a lesser degree of page movement than its vertical counterpart As you can see in fig 25 the vertical view does not allow the sentences to be displayed in their entirety which involves a considerable amount of left to right movement every time a new line is being read One can
20. be presented in the form of a chart or table This tab allows for several views and options for a full description see Tobii Studio 2 X User Manual pp 113 125 3 PRELIMINARY SET UP Prior to the implementation of any specific recording made with Tobii there are a number of points that users should be aware of Licence key activation amp deactivation Firstly the University of Leeds currently owns only one licence key meaning that only one person at the time can use the software It is therefore recommended that before exiting the programme at the end of each session the licence be deactivated in order to allow the next user to be able to open and run the programme To do so go on the toolbar at the top left of the window and select Help gt Deactivate licence gt Press OK see Figure 3 Trial 17 07 ideo Tobii Studio 2 3 2 0 File View Tools ForSight Help f Design and Fi j Replay Collect Logfiles to Desktop ihe Deactivate License X ee z User Manual different screenres Setup Help contents Fi Check for Updates About Tobii Studio Figure 3 Tobii Studio licence deactivation When the programme will be next opened by a new user Tobii will require a new licence key to be entered see Figure 4 below This is because a newer version of Tobii Studio was installed thus requiring the old key to be substituted by a new one If the licence has been deactivated by the last user
21. cording and Video Currently the Screen Recording view does not work in the replay mode i e it works while taking a test but not when playing it back The software allows a recording to be made but in the replay view where the scan paths should be overlaid on the moving images the latter simply do not show i e all that can be seen on screen is the gaze paths on a black background Given the short timeframe allowed it was not possible to investigate the root cause of this issue therefore the Screen Recording option was left out and all the recordings were made using the Video stimulus E According to the Tobii Studio 2 X User Manual pp 27 28 the ASF format is supported too however this file type has not been considered in the present case study 20 7 Using this stimulus the clip has to be uploaded to Tobii first which can take a long time especially if the video s length is that of a whole film It is therefore advisable to choose an excerpt and cut it with a video editing application to save time The clip chosen and cut from the film The Edukators for eye tracker testing was O3min 02sec long Five people watched this subtitled clip Generally speaking when calibration was good so was eye tracker performance However good calibration results were more difficult to achieve as the bottom of the screen seemed to have the weakest calibration points i e it was more challenging for the device to detect gaze position
22. duration is the sum of the duration of all the fixations made by the participant within that AOI in that timeframe The table above has been created separately however Tobii allows the export of both the chosen metrics onto an Excel worksheet and the raw gaze data to other application for statistics calculation The experiment was purely illustrative therefore the viewers were not selected according to any specific requisite Findings The third participant V_3 fixated twice on AOIL_1 poser but did not fixate at all on AOI 2 power hence the blank cells The same thing goes for V_7 who fixated three times on the word poser but none on power In AOI_2 subtitle containing the word power for V_1 and V_5 the value for first fixation duration is the same as fixation duration 0 33 and 0 54 respectively because the first fixation was in fact the only participant s fixation on the word power Comparison of the AOI data for the two selected scenes for all participants showed that 1 The first fixation duration overall across participants was longer for poser AOIL_1 than power AOI_2 by 1 2 seconds 2 The number of fixations on poser was higher than those on power in each individual participant apart from V_4 who fixated three times on both poser and power Across participants there were 22 total fixations on the word poser and only 7 on the word p
23. easons According to Bohme et al 2006 3 Systems that use infrared illumination often do not work well for users who wear glasses because of reflections on the surface of the glasses The existing systems can usually be made to work with glass wearers to a certain extent but only for some head orientations where no interfering reflections occur For other head orientations the reflections can obscure the user s eyes completely making eye tracking impossible Given that one user taking the experiments in the same conditions wore glasses and this didn t seem to affect tracker performance head orientation and the angle at which the eye reflections hit the tracker device could be a reasonable explanation for the differences registered as these are highly dependent on the individual participants physical features However as only 9 people took these reading tests and only 3 wore glasses there is not enough evidence to make a definitive statement in this sense Only further testing involving larger groups of test subjects might provide a clearer idea of how the Tobii eye tracker actually performs on people with low vision 4 a 2 Reading a text in Chinese In order to assess general eye tracker behaviour with non Latin scripts a Chinese text was analysed with the following objectives e Identify potential issues with recording character based texts e Verify if individual characters are displayed correctly during tests and
24. ed by the eye tracker and the actual dot position during calibration Calibration hans Accept Calibration Right Eye x 0 Points Marked for Recalibration Recalibrate Figure 6 Correct calibration In case of a test with several participants individual calibration data needs to be gathered for each test subject prior to recording For more detailed information see both the Tobii T X Series Eye Tracker Product Description pp 11 and Tobii Studio 1 X User Manual v 1 pp 24 28 Participant set up Each recording made within a test is associated with a participant whose data can be created and modified before or after the recording The default data consists of just the participants name but there is an option to add user generated independent variables such as gender or age which can be used to create participant groups The eye tracking data gathered during the recording can then be filtered by participant group while generating the descriptive statistics Test locked for editing It is worth mentioning that after a recording has been made the test will become locked to prevent accidental modifications during the sessions In order to edit the test this needs to be unlocked first To do so one will need to click on the lock icon located in the upper right part of the media element toolbar 4 CASE STUDIES 4 a Case study 1 Reading a text 4 a 1 Reading a text in English Media types Several m
25. edia options can be used to carry out reading tests The ones considered here were the following Web Questionnaire Screen Recording PDF Although the text chosen was available online the Web media type had to be discarded as there was an issue with recording websites This function seems to be enabled for some users but not for others due to lack of administration rights In some cases Tobii does not allow the user to upload URLs and consequently make recordings If the Web option does not work both the preview and record buttons will be greyed out Recording disabled since the timeline contains one or more web element s which require administrator rights Figure 7 Recording issues using the Web stimulus The Questionnaire stimulus is designed to have a question and at least one answer however it can be used in a reading experiment without MCQs by inserting the text in the question box and a blank answer in the answer box By doing so the software allows to preview and save the item see figure 8 where the first sentence of the reading comprehension has been used as an example Before the grass has thickened on the roadside verges and leaves have started growing on the trees is a perfect time to look around and see just how dirty Britain has become Finished Figure 8 The space available in the question box is limited hence paragraphs and long sentences would have had to be
26. een mode during a test An initial test was taken with a 12 point font but the letters appeared quite cloudy during the reading exercise figure 9 The screenshot below has not been resized and shows quite accurately the level of blur during the actual test in full screen view For a study on the effects of text layout on reading see Dyson 2004 For an exhaustive and well documented review of the serif sans serif debate see Poole 2008 10 called imagination a more than rational energy by which thought could seek to heal itself But Frankenstein s daemon am is a situation of the Romantic Wanderer or Solitary who s time can win no release from his own story by telling it His desperate desire for a mate is clearly an attempt to find Figure 9 P_1 Extra test performed on a smaller font Arial 12 points line spacing 1 5 Given the decision to compare the same text same number of sentences and same content and to have it always displayed on a single page to avoid scrolling between different pages the maximum font size was dictated by the page size The maximum font size used was 24 points as the whole text could still fit on one page with line spacing 1 5 Increasing the font size to more than 24 points would cause the text to go over two pages and was therefore avoided After a few tests it was decided to avoid scrolling because it makes it somewhat more difficult to assess gaze behaviour and it creates potenti
27. en recorded in the field of eye tracking especially regarding non intrusive remote systems B hme et al 2006 1 such as the likes of Tobii eye tracker In line with these fast technological developments eye tracking has proven a promising tool to study and assess human attention distribution on screen Schiess et al 2003 2 Despite its vast potential however according to Jacon and Karn 2003 1 there seem to be a slow progress in the use of eye tracking technology for Human Computer Interaction HCI research purposes Nevertheless the authors themselves acknowledged in that same study that the field is indeed beginning to flourish ibid This seems to be substantiated by the numerous and diverse applications the technology has today which makes eye tracking systems an attractive tool to invest in for a number of industries as well as a useful device in different scientific research fields such as physiology psychology engineering linguistics and cognitive science to name a few This study is concerned with eye tracking technology and linguistics research and in particular with providing an insight on the use of a specific system Tobii X120 Eye tracker and Tobii Studio in different linguistic experiments In particular this paper will discuss eye tracking technology in an attempt to e Raise interest in the subject and generally promote the use of this medium by researchers within the University of Leeds e Give exampl
28. es of possible applications and uses of the device in linguistics experiments e Investigate the practicalities of using the eye tracking system within the specific hardware and software arrangements currently available at the University of Leeds e Discuss and identify the most suitable set up for the case studies presented e Highlight the practical issues that might arise during experiment set up and its implementation e Touch on related problems that require to be investigated more thoroughly through experimental studies e Provide a starting point for future more in depth analysis and research projects from which other researchers in the Department can benefit We will start by describing the main features of both the eye tracker and the software it is connected to as well as highlighting preliminary actions that need to be taken when setting up an experiment We will then move on to a series of case studies aimed at describing the software and hardware performance when it comes to the following actions e Reading a text in English e Reading a text in Chinese e Carrying out a reading comprehension with multiple choice questions For a review of some of the possible applications both for diagnostic and interactive purposes see Duchowski 2002 e Watching subtitled video material e Translating a text in a separate application A brief outline of general limitations and obstacles will be then presented before drawing a conclus
29. ested Formatting options Partcipants P who took Font type Font size Line spacing Bold the test Test 1 Arial 18 1pt 9 E 5 Test 2 Arial 20 1 5pt 9 Test 3 Arial 22 1 5pt 9 Test 4 Arial 24 1 5pt 9 Test5 Courier New 20 1 5pt 2 z Test 6 Arial 12 1 5pt 1 S 5 Test 7 Arial 36 1 5pt 1 os 8 Test8 Arial 20 1pt 2 a Test 9 Arial 20 1 5pt x 4 Test 10 Arial 22 1 15pt 3 Table 1 Reading tests details Tests 1 4 were taken by all 9 participants and are discussed hereafter Additional tests were carried out but as the number of participants taking each test varied and less people in total took the additional tests the results were not analysed The font Arial was chosen based on the assumption that sans serif fonts are easier to read on screen However this assumption is not universally accepted as proven by the rich debate on the subject Despite the wealth of empirical studies on readability and legibility carried out over a period of over 70 years there still seem to be a divide within the scientific and typographic community regarding serif and sans serif fonts The font adopted by the Tobii Studio interface is a sans serif so is the default font used in the Questionnaire stimulus a sans serif type was selected for all the experiments to maintain consistency The minimum font size threshold was decided based on the clarity and neatness of the text when read in full scr
30. get text was overlaid onto a web page used for reference The user was compairing two similar Italian words to decide which one to use namely the term poesia piece of poetry poem found in the Wikipedia page or their initial translation poema poem The eye tracker picked up both fixations whose centres fall on or very close to said words on screen proving tracking was accurate The only problem registered on websites was related to scrolling up and down a page which seemed to cause the scan paths to be out of sync with the gaze is in italian Translate it using Google Toolbar Da Wikipedia lenadopedia libera Questa voce o sezione sulla i iki standard WIKI INDL a migliorarta second Traduzione ifenments Questa voce o sezione sulla sono insufficienti Puoi mighorare questa e ago tolle fone Segui i suggerimenti La seconda venuta una pper da William Buite imciusa nella raccolta Mochae artes and the Dancer venuta come alegoria per descrivere fatmosfera nelEurogi Breve Spiegazione mosis Nei versi traspare la conceziona cichca dela storia propr una fase violenta segnata dal sowerimento delle forme std vesione della Bestia rappresentata nell Apocaiisse di San Q Finiamola con questa malsana ak pagare i politici affinche neghina cambiamento climatico Proteggere l ambiente richiede un anforma radicale dei fina allora le maggior industrie cesseranno di ncopnre di soldi il senatore repubblicano James Inhofe ha dichi
31. iable data therefore the use of this tool with the Chinese text considered has not been explored 4 a 3 Reading comprehension with MCQ in English The document chosen for this experiment was a reading comprehension in English about littering and recycling in Britain see Appendix The media types considered were Screen Recording PDF and Questionnaire Initially some tests were taken using the whole text four paragraphs six questions However for ease of comparison and analysis as well as in view of previous considerations regarding text length and scrolling the final exercise was made shorter two paragraphs and two related questions and was structured in two pages Each page contained just one question preceded by the relevant paragraph of text By using this exercise in the Screen Recording stimulus it was still necessary to move from the first to the second page but scrolling down while reading would be avoided Moreover having a shorter text meant it was possible to keep it at the centre of the page where tracking performance is generally more precise The aim of the experiments was to establish which stimulus is the most effective in this type of test when an action is performed answering MCQs through mouse clicks Findings Bearing in mind that this test was taken by only one person and therefore more recordings need to be made to substantiate the findings tracking appeared quite accurate in the Screen Recording and
32. identify possible difficulties throughout this process record all the video specifications required by the software and measure accuracy in detecting eye movement when reading subtitles The following general considerations stemmed from the analysis of this first clip 19 1 Tobii only supports AVI files4 therefore the user has to make sure they either choose a clip already in this format or they convert their chosen clip to AVI using external video editing software as this operation cannot be performed in Tobii 2 Not all AVI files necessarily work Several were tried and in certain cases the software would play the audio but not the video or viceversa This depends on the file codec i e a piece of software or a device enabling video compression and decompression for digital video The standard codecs in Windows Media Player are Microsoft Video 1 and MP3 installed on all Microsoft computers The Techsmith codec is also available as it is included in the installation of Tobii Studio Tobii Studio 2 X User manual p 28 On the machine where the software is currently installed the MP4 codec also works as that is the encoding used for all the clips in this case study If one wishes to use other codecs these should be installed first Nevertheless this might present a problem as explained in the following point 3 If the audio of a clip works but the video doesn t or vice versa one cannot necessarily fix the problem in what would
33. iment the aim was to see if viewers are somewhat distracted by the spelling mistake and spend more time on it or go back to re read the word The idea was to compare number of fixations and fixation length between the two AOIs The following metrics were selected Fixation count Fixation Duration First Fixation Duration For a detailed explanation on fixations what they are how they are calculated and what types of filters are available in Studio 2 3 2 0 see Tobii Studio 2 X User manual pp 76 79 Statistics were generated for all the 7 viewers and the following results emerged 26 Test Fixation First Fixation Fixation Count Duration Duration Subjects sec sum in sec AOI AOI AOI AOI AOI AOI 1 2 1 2 1 2 Vii 2 1 0 34 0 33 0 82 0 33 V2 6 2 0 16 0 04 1 17 0 66 V_3 2 0 12 0 95 l V4 3 3 0 19 0 22 1 68 0 64 V5 2 1 1 19 0 54 2 02 0 54 V_6 3 2 0 17 0 21 0 63 0 36 V_7 3 0 37 1 22 All subjects 22 7 2 54 1 34 8 49 2 53 Table 3 Values calculated on the three selected metrics for both AOI_1 word poser and AOI_2 word power In Tobii fixation count tells the total number of fixations occurring within an AOI over the selected timeframe in this case the duration of the two subtitles respectively first fixation duration is the length of only the first fixation made on an AOI by a participant in that timeframe while fixation
34. ing to the Tobii Studio 2 X User manual p 57 this percentage shows the quality of the recording The quality of the recording is calculated by dividing the number of eye tracking samples that were correctly identified by the number of attempts 100 means that both eyes were found throughout the recording 50 means that one eye was found for the full recording or both eyes during half the time Note that the eyes cannot be found when a person is looking away from the screen this will result in a lower percentage If translating a text looking away from the screen is inevitable as the user will have to look down to the keyboard in order to type the translation meaning the device will lose track of their eyes In spite of this the accuracy of both recordings was good overall Mouse clicks were recorded correctly the scan paths were aligned precisely with the print lines in both the source and the target text and gaze movement was correctly followed as the user s eyes were switching between the two windows see figure 25 and 26 28 bo eT Stop this culture of payi for denying climate cha Protecting the environment requires a sweeping re corporations will stop throwing teg money at senato Republican senator Janes ihote told the envircement and p hoax perpetrated on the American people Photograph Bac q The best lack all convichon whe the worstAre full words Yeats s poem The Second Coming test sday s Senate hea
35. ion 2 TOBii EYE TRACKER amp SOFTWARE Experiment set ups Tobii allows for various experiment set up options The basic set up involves eye tracker connection to a primary computer screen where the stimuli are presented to the test subjects Other arrangements include connection to the primary as well as to a secondary PC screen remote Live viewer a projector TV monitor or a scene camera to track real 3 D objects For all the cases analysed in this report the basic set up was used where the X120 eye tracker is connected to the machine where Tobii Studio is installed and only one screen is used Media element types Experiment design can be customised through the use of different media often referred to as stimuli Figure 1 below shows all the media types available In this paper the following media were considered Movie Web Screen Recording Questionnaire PDF element Figure 1 Media types available in Tobii Studio Menus and Tabs When organising and carrying out an experiment in Tobii the user will navigate between the four main tabs Design and Record Replay Visualisations and Statistics File View Tools ForSight Help Design and Record Replay Visualizations Statistics de Project 2 Setup Figure 2 Tabs in Tobii Studio The Design and Record tab also referred to as view allows the user to create and modify tests manage participants calibrate and configure
36. lletin 124 pp 372 422 Schiessl M Duda S Tholke A And Fischer R 2003 Eye tracking and its application in usability and media research In MMIInteraktiv 6 6 pp 1 10 Boston William F Connell School of Nursing online Available at lt http dee square com mediapool 120 1207353 data Eye_Tracking_Application_Duda pdf gt Accessed 03 July 2012 Tobii Eye Tracking An introduction to eye tracking and Tobii Eye Trackers Tobii Technology White Paper online Available at lt http www tobii com Global Analysis Training WhitePapers Tobii_EyeTrackin g_Introduction_WhitePaper pdf gt Accessed 20 July 2012 Tobii Studio 2 X User manual release 1 0 Software Release 2 2 2010 Copyright Tobii Technology AB Tobii Studio 1 X User manual v 1 0 2008 Copyright Tobii Technology Tobii X120 Eye Tracker User Manual Revision 2 0 2008 Copyright Tobii Technology AB Tobii T X Series Eye Tracker Product Description 2007 Copyright Tobii Technology AB 39
37. lus or more precisely to investigate the immediate effects of repairs on listeners linguistic predictions Corley 2010 706 Obstacles to the implementation of this case study where identified in the limits of what media types can be utilised by the software As audio files alone are not supported in Tobii the source file would have had to be made into a flash animation However as the desired source file was not provided it was not possible to implement this particular study Future eye tracker based research in this direction is needed to test and assess participants response to speech repairs In terms of general information provided on Tobii the software and hardware features presented at the beginning of this document include what was relevant to the case studies considered rather than giving a complete overview on the Tobii technology Ergo studies of a different nature or carried out for different purposes might have slightly different needs and requirements implying the adoption of different parameters in the setup data collection and analysis Nevertheless it is hoped that the various points analysed here and the considerations stemmed from their analysis will be a useful starting point for other users who wish to customise themselves with the device and eventually use it confidently in their own experiments For more information on PET see Aziz et al 2012 For an article on Translog II please refer to Carl 201
38. n Before the grass has thickened on the roadside verges and leaves have started growing on the trees is a perfect time to look around and see just how dirty Britain has become The pavements are stained with chewing gum that has been spat out and the gutters are full of discarded fast food cartons Years ago I remember travelling abroad and being saddened by the plastic bags discarded bottles and soiled nappies at the edge of every road Nowadays Britain seems to look at least as bad What has gone wrong The problem is that the rubbish created by our increasingly mobile lives lasts a lot longer than before If it is not cleared up and nraneriv thrown awav if lt ctave in the Figure 17 This image represents the whole screen as it appears in the questionnaire stimulus during recording This initial paragraph was excessively long resulting in the text being cut A length limit does indeed exist and is ultimately dictated by the space occupied by the text on screen In the test performed with this reading comprehension the first paragraph initially 91 words had to be considerably shortened 70 words in order to fit the page Figure 18 18 Before the grass has thickened on the roadside verges and leaves have started growing on the trees is a perfect time to look around and see just how dirty Britain has become The pavements are stained with chewing gum and the gutters are full of discarded food cartons Q1 The write
39. nformation Technology 23 6 pp 377 393 online Available at lt http dx doi org 10 1080 01449290410001715714 gt Accessed 28 July 2012 Jacob R J K and Karn K S 2003 Commentary on Section 4 Eye tracking in human computer interaction and usability research Ready to deliver the promises In The Minds Eye Elsevier North Holland online Available at lt http www google co uk url sa t amp rct j amp q amp esrc s amp frm 1 amp source web amp cd 1 amp v ed 0CEoQFjAA amp url http 3A 2F 2Fciteseerx ist psu edu 2Fviewdoc 2Fdownl 0ad 3Fd0i 3D10 1 1 100 445 26rep 3 Drep1 26type 3Dpdf amp ei gg4dUPLhF4G aoQWyw4GgCw amp usg AFQ JCNG5dzyKgmZUbwnEPNaE1uAYEpgFNA gt Accessed 19 July 2012 Monbiot G 2012 Stop this culture of paying politicians for denying climate change Guardian co uk Environment blog blog 02 August Available at 38 lt http www guardian co uk environment georgemonbiot 2012 aug 02 climate change political funding us gt Accessed 28 July 2012 Noe A 2006 AVI File Format online Available at lt www alexander noe com video documentation avi pdf gt Accessed 15 June 2012 Poole A 2008 Which Are More Legible Serif or Sans Serif Typefaces online Available at lt http alexpoole info blog which are more legible serif or sans serif typefaces gt Accessed 15 July 2012 Rayner K 1998 Eye movements in reading and information processing 20 years of research In Psychological Bu
40. nngs on ch Traduzione Finiamola con questa malsa pagare i politici affinche neg cambiamento climatico Proteggere l ambiente nchiede un anforma radicale allora le maggon indusine cesseranno di ncopnre d sensore pubblicano James Inhofe ha dichiarato al comstasd cambiamento ciimatico una butaa che continua ad essere Foografia Bi Clark Getty images imighon mancano di convinzione mentre i peggiori passione Qu ste parole fratle dalla poesi La seco Yeats mi sono venute in mente mentre escola tal liticians Stop this culture of payin for denying climate chan Protecting the environment requires a sweeping reform of political funding only then Corporations will stop throwing big money at senators BoCe AaBbCe AaBb AaBbC TNoma TT No Spia Hesdngi Mesdingt Finiamoia con questa cultu PE 26 Scan paths during wanslanoni in the horizontal window alignment in MS Word 29 Tracking was also accurate on online pages fig 27 and 28 where the font had variable sizes and was at times considerably smaller than in the two translation windows uardthequardians Yeats s The Second Coming in a new verse guardthequardians blogspot com tss second coming n newe 11 Nov 2008 Yeats s The secdi oming in a verse ore OR by Sheridan N Dupre the first stanza of a work in progress UPDATED TRANSLATI Figure 27 Accurate eye tracking during online browsing In the image below fig 28 the tar
41. not see in full the source sentence nor their translation as it is being typed Moreover the tracking performance in the vertical alignment was slightly sketchier with less scan paths and more individual fixations being recorded In general the user found the horizontal alignment better as there were fewer sentences visible on screen without scrolling but they were fully displayed and therefore did not require moving the horizontal scroll bar at the bottom of the window or enlarging the page to its full size to see the text as a whole A difference was noted between top to bottom scrolling and left to right movment the latter being more easily and accurately traceable by the eye tracker perhaps becasue the linear progression of the gaze mirrored the left to right motion of the horizontal bar 5 LIMITATIONS AND OBSTACLES Jacob and Karn 2003 4 identify three main groups of limitations related to the use of eye tracking devices for research purposes technical problems labour intensive data extraction and data interpretation issues Some of the technical problems involved with the experiments have been the subject of discussion here Two technical issues that do not relate to any experiment in particular and yet might affect the implementation of any test are software crashes and eye tracker disappearance In the two month period during which all the experiments were carried out the software crashed seven times whilst different ac
42. ns if happening during an experiment with an external participant This issue could be related to the quality of the LAN cable which had to be changed once during the two months of testing Since LAN cables are not very enduring it could be advisable to have a spare one stored with the tracker 31 Another major problem encountered although not strictly a technical one was defining eye tracker accuracy For the sake of this study a very basic and intuitive definition has been outlined rather than a scientific one It was sometimes quite difficult to assess whether a scan path was not overlaid exactly with the print line because of actual tracking inaccuracy or because that was simply a feature of that participant s reading behaviour A more comprehensive and precise definition of tracker accuracy should be required as an essential pre requisite to conduct meaningful eye tracker based research in the future The labour intensive data extraction issue is partially resolved in Tobii by automated software calculations On one hand data collection seems to be relatively easy on the other however retrospective data analysis still presents researchers with the challenge to find appropriate ways to use and interpret data Jacob and Karn 2003 1 Choosing the most suitable metrics for an experiment for example is not always a straightforward task Moreover fully understanding the gaze movement parameters available and defining which one
43. or increases in font size and line spacing However media type cross comparison was not carried out due to the lack of resources and issues with one of the media types Web element therefore this could be an interesting starting point for further research or case studies Findings It must be pointed out that not enough participants took the tests for the conclusions drawn to be considered fully reliable More thorough testing involving more participants is required to investigate eye tracker accuracy with different text formatting However the findings can be useful if taken as general considerations or as the proem to further more exhaustive studies Tobii does not allow letter by letter analysis nor does it give the possibility to have a closer look at gaze movement on individual words during the replay of a test The replay view just plays back the recording which can be watched in full screen same actual size the participant was looking at during the test paused and advanced frame by frame but not zoomed in It is possible to zoom in only on a static image in the Visualisation view At individual word level Tobii cannot tell if during the reading activity a participant is looking at the top or bottom of each letter At sentence level however it is possible to detect readers tendency to look at the top or bottom of consecutive words whilst reading a whole sentence Compare gaze position in participant one P_1 scan path in the middle
44. ower 3 The values representing the total fixation duration on AOI_1 and AOI_2 last two columns of the table indicate that in each and every individual participant more time was spent during fixations on AOI_1 than on AOI 2 27 Points 2 and 3 in particular show the strongest evidence for the initial assumption that more processing time is spent on the misspelled word as this has been fixated both more times and for longer However as previously mentioned the results are not intended to prove or disprove a point but to provide an example of what can be achieved through the Visualisation and Statistics tabs in Tobii 4 c Case study 3 Translation in a separate application For the third case study a short excerpt 92 words of an English article on climate change taken from the Guardian was translated into Italian using MS Word The source text was opened in one window and the translation was typed in another Two window arrangements were tested namely the vertical and horizontal alignments The Screen Recording mode was the only viable choice for this type of experiment as both MS Word windows need to be open and navigated freely and typically some online resources will be used during translation thesaurus reference websites dictionaries and the like Gaze movement between all these applications and windows can be tracked only by recording the whole screen In both test the sample rate of the tracker was between 50 60 Accord
45. r says that it is a good time to see Britain before the trees have leaves because you Can see Britain at its dirtiest you can see how dirty Britain is now the grass has thickened on the verges Figure 18 The same question of figure 17 had to be adapted to fit on screen Paragraph parting Paragraphs cannot be separated by pressing the Enter key In the question text box one can press Enter and start a new line however when recording or previewing the item this does not show i e all the words are displayed seamlessly one after the other See figure 17 where the first two words are in fact the title Dirty Britain which should appear in bold but formatting is lost and be followed by the rest of the text on the line below In view of what highlighted above if longer or more complex reading comprehension tests are to be performed where the text cannot or is not supposed to be shortened or adapted using this media type might not be the optimal solution 4 b Case study 2 Watching subtitled video material 4 b 1 The Edukators General considerations In the second case study two clips were anlaysed with different purposes The first excerpt was taken from the German film The Edukators original title Die fetten Jahre sind vorbei by Austrian director Hans Weingartne The main objective here was to test the software in order to report the steps required to create record and replay a clip correctly in Tobii
46. reader was doing fig 16 AT RAX RAGMNTA MPRA KAKA Figure 16 The participant was an English native speaker studying Chinese Based on the assumption that mother tongue speakers of Chinese would spend less time than a foreigner on each character it would be interesting to repeat the experiment with a native to see if it yields improved or similar results in terms of scan path accuracy However it could be expected that even Chinese natives would be unlikely to have as many short fixations as the ones on figure 19 or 20 below page 21 and it could 16 be presumed that taking the same line length from an English and a Chinese text the Chinese reader would spend more time reading the Chinese than an English reader reading the English given information density and linkage of the characters in the former Moreover an obstacle to detecting gaze paths could be the lack of word spacing in Chinese which makes it more difficult to distinguish word unit boundaries and to structure the visual field As for scan path analysis and statistics since the metrics in Tobii are based on AOIs created on still images there should not be a difference between analysis of Latin and non Latin scripts as the software analyses eye behaviour fixation count length and the like in relation to a selected area of the screen regardless of its content However running statistics on a recording such as the one above would not produce any accurate and rel
47. replays e Test whether individual character detection is possible The sentence chosen was saved as PDF with different MS Word formatting options and previewed in Tobii Studio It was found that despite the PDF had been created correctly and was clearly legible when the document was previewed some of the characters didn t show correctly namely some of the strokes were blurred or not clearly detectable making it very difficult to read the sentence The first PDF had the following formatting Font type Arial Font size 14 points Line spacing 1pt 14 It was initially thought that unclear character display was due to the font size being too small This was therefore increased to 18 20 and 26 points only to discover that even with the largest font tested the software still did not display the text correctly Take for example the following characters gt and 4L During the Tobii preview some of the characters strokes disappear altogether or are barely visible figure 14 Of Pee BEACH ET H FEM LIGA MERY PEARSE FF OH 408 S NEY IADR ME AR TEREG CAP Ac A MAT HEM SULIESH BURR MPAA Fn Figure 14 Character corruption with Arial font at 14 points However when font type was changed the sentence was displayed correctly Characters were not corrupted in the Tobii preview mode with the following text formatting Font type Arial Unicode MS Font size 26 points Line spacing 1pt H
48. s are the most suitable for a specific experiment can present the user with a great deal of difficulty In Tobii Studio statistics are based on fixations and even the software creators admit that there is a degree of subjectivity to their identification Fixation identification is mostly a mathematical and statistical process that attempts to describe how the eye moves Despite the accuracy and flexibility of some of the algorithms the limiting factor is that there is no precise definition for the start and end of a fixation and thus the identification process still remains partly subjective Therefore the best identification validation process is still to perform a test where you compare the fixation filter data with an observer s subjective impression on the position and duration of the fixations Tobii Studio 2 X User manual p 76 Fixation filters are used to present the raw gaze data in a more meaningful way through classifying it as part of or not part of a fixation To do so a series of data processing functions and different algorithms are used The filters available in the 2 3 2 0 version of the software currently available within CTS are the following Tobii fixation filter Raw Data fixation filter ClearView fixation filter I VT fixation filter All but the last filter are addressed briefly in the 1 X User Manual The I VT filter has been introduced later so a definition and an explanation can be found only in the
49. space in the set up chosen for the experiments analysed here the tracker was placed directly below the monitor rather than in front of it Therefore as you can see in Figure 5 the distance from the screen is set to a negative value 0 0 deg Units Active Display Size em E Width 34 0 Tem finch Height 27 0 icm Figure 5 Configuration Tool showing the parameters used in all the experiments described in this paper Eye tracker calibration After configuration the tracker needs to be calibrated to the user s eyes Open the Calibration Dialog box in the Design and Record tab under Set up gt Calibrate Firstly a black window with a status bar at the bottom will appear This box represents the field of view of the camera inside the tracking device The user s eyes are represented by two white circles It is recommended that the user s eyes be in the middle of the track status metre and that the distance indicator be between 60 70 cm The default calibration consists of a red point moving around the screen which should be followed by the users eyes Calibration stops automatically when complete and the results are displayed in two vector plots in the Calibration Result dialog box see Figure 6 If the calibration plot shows error vectors i e green lines instead of green dots the tracker will need recalibration The size of each green line indicates the difference between the user s gaze point calculat
50. split into several parts making it unnecessarily cumbersome for both tester and testees However this option could be ideal for individual sentence analysis in other types of experiments Another reason why this stimulus was not used in this reading test is that it doesn t allow for different font types and sizes the software removes all the formatting and enables the text to be presented only as shown in figure 8 The Screen Recording option was also considered however following a comparison test the more direct PDF stimulus was preferred In the Screen Recording mode the software opens the text in the chosen application in this case 9 MS Word or Acrobat Reader whilst the eye tracker records the whole screen meaning that the application menus tabs and toolbars are also displayed potentially causing distraction Moreover this mode involves more adjustments the text might have to be zoomed in and out scrolling might be required to be made by either the testee or tester Using a PDF was much simpler as the only action requested from the participants was to press Esc when they finished reading The objectives of the experiments were the following e Test general tracker performance during a reading activity e Test accuracy in detecting gaze behaviour e Gather empirical data regarding different text formatting options The same text excerpt was presented as a PDF each time with a different formatting The following options were t
51. t that as long as the text was comfortably readable not blurred even if the font was small the tracking performance would be acceptable However one might wonder if tracking just seemed better because there was less space between the lines line spacing was 1 pt instead of 1 5 pt like in the other three tests as having more available space between print lines makes any scan path deviation more noticeable Across participants and recordings tracking was most accurate in the centre of the active screen while the lowest performance was recorded at the bottom of the screen where scan paths were often less well aligned with the text This was also substantiated across all tests by the real time checks in the calibration grid3 that all participants were asked to perform before accepting calibration and proceeding to the reading exercise In most participants there was a high degree of variability within the same recording e g the beginning of a reading exercise was tracked accurately but the scan paths towards the end bottom of the page started to deviate from the lines of text Moreover it was interesting to record the differences in reading style whereby some people seemed to read in a more linear way the paths being almost straight while moving left to right whereas other people s gaze seemed to jump up and down whilst moving left to right Another striking aspect of the reading activity is that gaze tracking was more accurate
52. the tracker In the Replay view the recording can be replayed with the gaze path overlaid on the stimuli This is also the tab where scenes need to be created Scenes are portions of the recording selected by the user They must be created to be able to use the visualisation tool and select Areas of Interest AOIs when working with certain media types namely those presenting moving images typically Movie and Screen Recording The still background image of a scene Scene Image is also chosen here The timeline detail for the recording can be increased to show shorter time 4 intervals and the playback speed can be adjusted so that the recording can be replayed at a faster or slower pace see Tobii Studio 2 X User manual pp 57 58 The Visualisations tab contains a series of tools for different visualisations The tools available in the current Tobii Studio licence are Gaze Plot Heat Map and AOI Tool In the case studies presented hereafter the AOI Tool has been used An AOI is a portion of the active screen selected by the user on which the metrics will be calculated AOIs need to be created on a static image therefore if the stimulus chosen is a video or screen recording a scene needs to be created first and a suitable scene image has to be assigned to it The Statistics view provides descriptive statistics based on the AOIs selected by the user The data are based on fixations and mouse clicks and can
53. tions were being performed e g while previewing a video before recording while replaying a clip during a screen recording test or while moving between the Statistics and Replay tabs While there did not seem to be a particular action triggering the crash the application was more likely to stop working after having been open for several hours Software hardware communication also presented technical problems In the set of experiments carried out the basic set up has been adopted where the tracker needs to be connected to a standard power socket through a power cable and to the computer through a LAN cable to USB adapter Usually a minute or so after pressing the power switch the computer connects to the LAN line and Tobii finds the tracker If the device is correctly plugged in the connection status bar at the bottom of the screen will say Eye tracker TX120 203 73000025 Ok At times however even if all these steps have been correctly followed and the tracker is switched on two blue lights flash at the sides of the device and two circular infra red lights come on the connection is not established While there did not seem to be a clear explanation for this it was found that by switching the power cable off and on again closing and re opening the programme or doing both the connection was normally re established and Tobii Studio correctly recognised the tracker However this can take some time and can therefore cause disruptio
54. when the person was reading more quickly i e reading as fast as they could rather than reading to understand as much as they could Reading pace was faster fixations much shorter and scan paths straighter It would be interesting to develop this point further by designing an experiment around a complex text and present it to two groups where group A would be specifically told to read the document in order to fully understand its content and perform a comprehension test afterwards whereas group B would be specifically told to just read through the text quickly without putting a great deal of processing effort into comprehension and then compare the resulting scan paths Three participants wore glasses during the experiment For one of them this did not seem to affect the performance as eye detection calibration results and the further real time post calibration checks were all positive However for the other two participants eye tracking was not very accurate in any test For one of them in particular the recordings were almost unusable This test subject wears prescription gt The post calibration real time checking consists of a plain screen with 9 red fixed dots calibration grid three at the top three in the middle three at the bottom on which the user has to focus their eyes being tracked and represented live on the screen by a circle If the red dot the user is focussing on is in the centre of or at least inside the circle
55. xperiment can be implemented from start to end in Tobii Studio The aims of this exercise were the following e Outline some of the interface functions when working with video material e Demonstrate the use of the visualization and statistic tabs e Introduce some of the metrics available in Tobii One subtitle in the clip presents a spelling mistake namely the word power has been misspelled poser In the following subtitle the word appears again this time spelled correctly The aim was to compare the two adjacent subtitles analysing eye behaviour to see if more and or longer fixations and therefore more processing time occur in the subtitle containing the misspelled word In this particular case the spelling mistake did not result in a random combination of letters but in an existing word with a meaning in the English language which could potentially be 23 misleading and cause misunderstanding The assumption was therefore that the word poser would draw the viewers attention to itself and overall be fixated either more times or for longer intervals or both Experiment set up The set up stage consisted of choice and preparation the audiovisual material Preparation involved selecting the excerpt from the film burning the subtitles onto the video track cutting the clip save it in a suitable screen resolution 1024x768 and with the right codec in this case MPEG4 importing the clip in Tobii and previewing it
Download Pdf Manuals
Related Search