2016年2月16日 星期二

Research proposal v6

Do Mnemonic-devices placed on main verbs and predicates enchance people's average reading fluency?






Howard Wu

102551020@nccu.edu.tw

Advisor: Prof. J.L Tsai

Due date 2016. 1/18 5pm













View this document with latest status online

Abstract

In this research proposal I try to figure out if m-devices placed on ceartain sytactic position like main verbs and predicates will enhance reading fluency. By placing them in the correct positions, I assume they will improve readers' reading speed and decrease unnecessary regression and fixations.

I will calculate the mean (average) velovity and mean acceleration on two neighboring fixations. The data obtained by eye-tracking devices will provide accurate horizontal positional data for my research, thus answering my desired questions as the paper titled. And this will probably lead to the invention of an hint for POS(part of speech) based m-devices reading training for dyslexics who suffer from crowding effect and poor working memory.  


introduction & literature review

From the abstract of "Shorter lines facilitate reading in those who struggle." , it was written "...Given that an expected trade-off between horizontal and vertical regression was not observed when line lengths were altered, we speculate that these effects occur because sluggish attention spreads perception to the left as the gaze shifts during reading. Short lines eliminate crowded text to the left, reducing regression. " I think it might also be influential if we can utilize their limited attention to the predicates by mnemonic-devices. No such experiment has been done by eye tracking devices.

This time I need to produce the text by myself, and adjust the spacing of the mnemoinc devices so that they won't look crowded. I suspect the result that was opposite to my prediction last time was caused by crowding effect. Those m-devices were added by marking software (Skitch, made by evernote. corp) after the text sample was produced. So they look somehow crowded and awakard.



Methods: (design, material, procedure…)

From my previous experiment I found some drawbacks and I would like to avoid some of them by improving the techniques of experiment design. The rest setting and my hypothesis will remain the same. The material will be specially written with specific details. For example, a description of the objects and their orientation in the room.


I would not choose text from source such as newspaper on NYT or science textbook , even they might have more narrative materail with neutral facts. Some researchers was  claiming they randomly selcet articles from the science textbook, but it is still possible that some participants have more understanding on such a field, thus causing them to read faster and better. Because they may still have bias and hints for readers of some backgrounds. This would influence their processing speed of the text. So I'd write a novel article on my own.

For questions like how people process articles with different styles and syntatic attributes, I think it shall be postponed until I make clear of this m-devices question. But I also want to know on which material my m-devices work better. They will be arranged in the experiment next time.

  • α = velocity for each trial

  • β = acceleration for each trial

  • R = observed ROIs(Region of interest), where positive acceleration happens and ROI marks were recorded

Here are the distribution maps of the above data:

Let me list out the problems encountered so far.

  1. the Y-axis alignment is corrupted so there are dots outside the ROI. But they shall be counted into the ROI.

  2. Grouping of the test is problematic: the best grouping shall be a 2x2 matrix, or at leasat a 2x2 matrix. But here in this case, mine is a 2x1 matrix.

  3. Ways of defining efficiency. I use the looseness of the dot as an indication for reading speed. Yes we do have time stamp of each dot. But these "looseness" can not faithfully explain if their reading is efficient.

  4. the way they scan through the passage. (they just move their eyes without really reading it!)

Some methods to improve such problems

  1. fill up the whole map with ROI, so that no dots will be missed.

  2. I will make my next experiment grouping at least in 2x2 matrix.

  3. I try to make an operational definition on this: higher mean velocity throughout the trial. By mean velocity I mean the average score of each neighboring fixation. Of course I need to delete the huge regression caused by line changing.

  4. We will tell them there's a quiz after this eye-tracking experiment. I am not if I shall make one quiz paper at all. But it would be a good reference to tell their VAS (visual attentional span) in another way. Just a longer term VAS after the test.

  5. I may need a VAS for them before my test so that I could know if the participants have relatively similar working memory. I will also make sure the participants did not take pills or feel tired because they have to hurry to the test or ride fast. All these factors will cause their attention lose and dry eyes, which is very bad for reading and eye tracking calibration.  

ROI details in x-y map along with fixation map

http://fall-cicada.blogspot.tw/2015/12/dot-map-comparison.html


Mode of alpha and beta for 701 and 702 in trial 2 and trial 3

We also want to provide the δ^2(=variance) and mean of alpha/beta of all trials of both participants (alpha 2 belongs to trial 2; beta 3 belongs to trial3, etc ). Please see endnote. Though I have no idea on how to use them correctly, yet.


Now we only focus on the trial 2 & 3 of participant 701 & 702.

For Trial 2 of Participant 701: There are 37 sections (defining as neighboring fixations) with positive slope. That is, positive acceleration. There are 15 of them located within ROIs, where we can observe for the m-devices effect. I provide the observed ration to elaborate how much fidelity my experiment carries. In this case, prediction made based on 30%~40% obsered ratio is not very representative.

To address easier, I made a table below. Due to my limited time on data processing, I can only calculate so much. But I will keep on to find out if the rest answers are still so depressing.

(β > 0)

Δ x

R = observed Δ x

observed ratio

mean α throughout trial

mean β throughout trial

mean α of R

mean β of R

701 trial2

37

15

40.5%

0.329601


0.301910


702 trial2

37

12

32.4%





701 trial3

43

13

30.2%

0.312848


-0.2827


702 trial3

55

17

30.9%









Anticipating results

I expect the readers will experience better reading fluency after they get to known the tricks of predicates and main verbs. The foreground is that this would not cause extra crowding effect for them. The m-devices could be annoying for readers since they were placed and add up afterward. This would make the whole text looks crowded. I need to increase the line space greatly or clear up the layout to make the words look right.

I was expecting the group with m-devices will read far faster and has fewer dots upon the ROI of predicates.

But I figure out:
mean alpha throughout the trial is actually larger than the mean alpha of R !

This means my m-devices are only slowing them down!


To my surprise, the result seems to be the contrary to my prediction. I had deleted the huge regressions of line changing. So they shall not cause the trouble. I am not sure what happened to my data, I am not even sure if I process the data analysis correctly. (It should be correct)

Maybe I should consider my methods on calculating the "averaged" (mean) velocity is not suitable. I can not just delete the outliers caused by line changing then average them. Even though these spots(fixations) can be clearly markerd on the data. I still hold some faith to my hypothesis, but I need to renew staticstics knowledge.



Reference

  1. Olson, R. K., Kliegl, R., & Davidson, B. J. (1983). Dyslexic and normal readers' eye movements. Journal of Experimental Psychology. Human Perception and Performance, 9(5), 816–825. http://doi.org/10.1037/0096-1523.9.5.816

  2. Schneps, M. H., Thomson, J. M., Chen, C., Sonnert, G., & Pomplun, M. (2013). E-Readers Are More Effective than Paper for Some with Dyslexia. PLoS ONE, 8(9), e75634. http://doi.org/10.1371/journal.pone.0075634

  3. Stanley, G., Smith, G. a, & Howell, E. a. (1983). Eye-movements and sequential tracking in dyslexic and control children. British Journal of Psychology (London, England : 1953). http://doi.org/10.1111/j.2044-8295.1983.tb01852.x

  4. http://www.degruyter.com/view/j/cogl.2015.26.issue-2/cog-2014-0066/cog-2014-0066.xml?format=INT

Endnote:

Complete charts for both participants.


mode

701

702

α1

0~0.5

0~0.5

α2

0~0.5

0~0.65

α3

0~0.5

0~0.4

α4

0~0.5

0~0.65

α5

0~0.65

0~0.5

α6

0~0.5

0~0.7

α7

0~0.5

0~0.5

α8

0~0.5

no data

α9

0~0.5

0~0.5



mode

701

702

β1

-0.0035~0

-0.0035~0

β2

-0.004~0

-0.005~0

β3

-0.015~0

-0.004~0

β4

-0.004~0

0~0.015

β5

-0.0065~0

0~0.005

β6

-0.005~0

-0.005~0.005

β7

-0.005~0

0~0.0075

β8

0~0.005

no data

β9

-0.01~0.01

0~0.005



701

μ of α

μ of β

δ^2 of α

δ^2 of β

1

0.004146444621

-0.0000244505898

0.815874143

0.00004458453974

2

0.002332514632

-0.000008664773698

0.8333474164


3





4





5





6





7





8





9







702

μ of α

μ of β

δ^2 of α

δ^2 of β

1





2





3





4





5





6





7





8





9








沒有留言:

張貼留言