What Can Eye Movements Tell Us About Reading?

I do believe something very magical can happen when you read a good book.

-- JK Rowling

A well written story can spark vivid mental images and intense emotions in a reader. Eye movements (EMs) associated with reading behavior generate small but detectable bio-signals that suggest how the brain perceives, integrates and comprehends information acquired through words printed on a page. Although personal experience tells us our eyes scan each line of text at a steady pace, recordings of EM activity reveal a different picture.

Saccades are rapid, horizontal movements that bring each word into visual focus. These movements are followed by fixations or pauses. Individual words or phrases are recognized and given meaning during fixations according to a generally accepted theory of reading. At the end of each line of text, the eyes perform a coordinated backsweep where the reader's focus shifts to the beginning of the line below. Both eyes continue their halting movements to produce a “stair step” signature when the amplified bio-signals (in volts) are graphed with respect to time (in seconds).

Leftward EMs within the interval between sequential backsweeps are regressions. These movements enable the reader to revisit parts of a sentence and re-inspect specific words. Languages read right to left or vertically produce their own distinct EM signatures. Text comprehension requires the integration of words across sentences and subsequently paragraphs.

Description: The proposed activity, which emerged from a pilot project, introduces high school students in an Advanced Placement Psychology course to experimental reading research. Each student will generate their own electrooculograms (EOGs) or record of EM activity while reading sections from a story in two environments: a quiet setting followed by a distracting setting. The research goal is to find out whether a background distraction can influence a student's eye movement pattern during a reading task.

Key Concepts: Backsweep, EM Signature, EOG, Fixation, Regression, Saccade, Text Comprehension

ADC & Signal Display - Select an analog-to-digital converter (ADC) for the EM signals. I used a vintage Dataq digitizer (Model DI-190). The converter has a signal sampling frequency of 240 Hz/channel and a +/-5 V DC input span. I used the manufacturer's signal acquisition and display apps that require Windows 9X. The apps and user guides are available as a free download from Dataq.com, which also sells an updated version that runs on current operating systems. I addition, a post-acquisition analysis app that calculates signal derivatives is required. BTW, if you use Dataq's signal acquisition app, you must use their signal analysis app, Advanced CODAS. BTW, I tried replacing the DI-190 with a PC signal analysis app that collected signals through a sound card, but obtained disappointing results.

Alcohol Wipes (disposable)

Connectors – Crimp-on or solder pin type

Electrodes – Use disposable electrodes to pick up bio-signals associated with movements of the eyes. (Wallgreens; Item Code: 21055; ECG patches also work).

Monitor - In keeping with other reading researchers, the text should be readable on a computer monitor from a distance of about 50 cm to 1 meter.

Power Supply - Eight, AA batteries to provide 12 VDC

Signal Acquisition & Processing – An electrooculograph is the writing instrument that amplifies positional changes of the eye and filters unwanted bio-signals as well as 60 Hz interference from nearby electric lights and appliances. A record of the analog voltage output associated with EM activity is an electrooculogram. The analog voltage is approximately correlated with the number of character spaces traveled by the eyes as they read a sentence.

Research grade EOGs sell for thousands of dollars; but this project will barely make a dent in your departmental budget! I discovered a student grade, dual-channel EOG at an estate auction for $25 and then invested some time to setup and test the unit. The battery operated, analog unit is ideal for this project. However, the model is discontinued and may require some effort to locate. Try electronic auctions or schools downsizing their biology departments. If you're really jazzed and have the tech skills, "Watchmefly" has posted an instructable for a DIY EOG monitor that controls lights with your eyes.

Sound Source - Stream an audio source containing spoken words, such as a lecture, newscast or TED Talk through headphones.

Text Sample - Select two, non-sequential pages containing about 250 words per page from a grade-appropriate story. The pages should contain unrelated content so the second reading is not influenced by the first reading.

Tools & Misc Items – Contact adhesive/sealant, small screw driver from repair kit for eye glasses, solder & soldering pencil, wire snips.

USB Flash Drives - For storing each student's EOG file

Wire - Stranded, hook up wire to make jumper leads and electrode lead extensions

The battery operated, Heathkit EOG Module is compact, bio-signal conditioner featuring channel-dedicated: 50/60 Hz rejection, baseline controls (Offsets 1-4), high and low pass filtering (HPF 1&2 = 0.05 Hz; LPF 1&2 = 30 Hz) and adjustable amplifier controls (Gains 1-2). I checked both sides of the circuit board for obvious physical damage. Aside from an improperly seated integrated circuit, everything seemed satisfactory.

Rejection of 60 Hz noise is achieved with differential amplifier, A1 that increases input signals that are dissimilar in voltage, while limiting electrical noise that is common to both inputs. Voltage output from a differential amp is proportional to the difference in voltage between each input.

The specific operation of the isolation amplifier, A2 isn't clear from the graphic, but my impression is that the circuit provides a safety barrier between the subject and possible current leakage from the battery pack.

The graphic of the low frequency response curve (HPF1) gives a qualitative indication of the increase in amplifier output with respect to frequency. The graphic of the high frequency response curve (LPF1) on the right side of the board gives a qualitative indication of the drop off in amplifier output with respect to frequency. Together, the cutoff frequencies of 0.05 Hz and 30 Hz describe EOG bandpass, which refers to the expected range of EM frequencies associated with reading behavior. Because fixations have low frequencies, EOG units use minimal low frequency filtering. High frequency filtering reduces the presence of facial muscle activity as well as 60 Hz noise remaining in the signal.

Amplifier, A3 multiplies the input signal by a factor determined by Gain 1. Lastly, the (DC) Offsets 1&2 adjust the EOG baseline by adding voltage to or subtracting voltage from the signal. The 9-pin D connector on the far right edge of the board accepts 12 VDC to power the board.

I assembled five jumper cables by soldering pin connectors to the ends of stranded hook up wires. Circuit wiring took seconds thanks to the block diagram printed on the circuit board. The second channel isn't needed for this activity.

After circuit components were connected; I cut three, 30 cm lengths of hook up wire to make electrode extensions and terminated the ends with a pin connector. I plugged one end of the extension into the terminal of each electrode and soldered battery pack leads directly to the +12 V and -12 V pins of the J2 connector on the board's component side. Leads were secured with several drops of Goop sealant. Lastly, I attached the analog outputs from Channel 1 at Vo1 and the earth ground to the Channel 1 (upper left)and ground inputs (lower left) of the DI-190.

Conceptual Knowledge: This study uses an EOG to determine whether a background distraction changes a student's EM signature with respect to a baseline condition. According to reading theorists, "the eyes go where the mind goes." If a distraction competes with a reader's goal of attentively searching for the next word during a reading task, what might the eyes do?

It's possible EMs will stop when a reader's attention disengages from the task. When attention re-engages, the need to re-read poorly understood words or sentences may trigger regressions -- imagine trying to study in a library when seated near someone using a phone set to speaker.

A within-subjects design where each student serves as their own control across the collection of two data sets is used to explore these possibilities. The first data set is collected in a designated quiet location of a computer lab to establish baseline EM activity. The second data set is collected in the same location while the student is exposed to the sound source. An EM signature consistent with a distraction disengaging a reader's attention was predicted to exhibit:

• prolonged fixations with no EM activity indicating reading behavior has stalled;
• frequent regressions between backsweeps indicating re-readings of sentences.

Tech Note 1: Time constraints of a typical classroom session placed limits on the proposed experimental design. A better design would use random student assignment to one of two groups where each group is exposed to a counter-balanced presentation of the sound source across longer reading sessions.

Teacher: Obtain parental consent. Seat student in front of computer monitor and instruct student to prep left and right temples of the head as well as on the area directly behind the preferred ear with an alcohol wipe for placement of the reference or electrically neutral electrode. Have student place an electrode on each prepped location.

Tech Note 2: Brisk rubbing can reduce electrode impedance and improve signal quality.

Teacher: Insert leads into "R eye," "L eye" and "Head" inputs of circuit board. Begin data collection by launching signal acquisition app. Instruct student to perform bio-cals which simulate EMs and fixations by:

1. focusing on left margin of screen and fixating for about one second;
2. focusing on right margin of screen and fixating for about second;
3. returning focus to left margin of screen for about one second.

Conceptual Knowledge: Because the cornea and retina of the eye are oppositely charged, the left electrode picks up a positive signal when subject looks left and the right electrode picks up a negative signal. When looking right, the left electrode picks up a negative signal and the right electrode picks up a positive signal. The differential amplifier, A1 interprets the eyes looking left as a downward signal deflection and as an upward signal deflection when the eyes look right.

Teacher: Instruct student to read first page silently. Repeat instructions for second page reading. After student has positioned headphones comfortably, begin streaming and collect data. Prep, bio-cals and data collection should take less than 10 minutes. Store EOG data on the flash drive provided to each student. Instruct students to upload their data on their workstation computers.

Teacher: Make arrangements to have each workstation in your computer lab run a signal analysis app that calculates changes in eye position over time on the vertical axis (defined as velocity or reading speed) with respect to time displayed on the horizontal axis. The calculated channel should display momentary changes in reading speed as well as changes in EM direction indicating a backsweep, saccade or regression. During a fixation when the eyes pause for word recognition and meaning assignment, reading speed for practical purposes drops to zero.

Conceptual Knowledge: Here is a screen grab from a bio-cal procedure. Eye position (upper channel) and velocity, as calculated by the CODAS app, appears in the lower channel. Notice how the velocity waveform peaks and then returns to the baseline as expected when the eyes reach the fixation point. Although both channels feature clean waveforms that exhibit correct shapes, the amplitude of the position signal is too small. The voltage range of the position channel should be a full-scale deflection (FSD) that approximates the input span of your A/D converter when a student moves their eyes to opposite margins of the screen. When bio-cals do not achieve FSD, fixations recorded during the reading tasks will be difficult to identify. In contrast, the +/-5 VDC range of the bottom channel is appropriate. Dataq has a helpful tutorial on digital data conversion here.

Tech Note 3: Adjust Gain 1 with eye glass screwdriver to produce the largest deflection possible within the span of your ADC during bio-cals; but don't exceeds specified limits of your ADC!

Conceptual Knowledge: The interval defined by Backsweeps 1 and 2 in this screen shows eye speed during the inspection of a text line with a background distraction. I included the highlighted velocity band to identify fixations when the eyes were stationary because of considerable noise. Fixations, F1 - F9 are labeled above those intervals between EMs and within the velocity thresholds. Upper and lower limits of the band were eyeballed for the pilot (pardon the pun :>). The downward dip following the approximately 750 msec pause is a regression.

The second screen details a pause in EM activity followed by a regression consistent with distracted attention requiring re-inspection of a word. Digitized voltages corresponding to the regression’s start and peak amplitude appear in sidebar. Because the equipment doesn't map words on to fixations, the particular word associated with this pause can't be identified -- but overall, not too shabby!

Conceptual Knowledge: The average fixation time for silent readings of grade-relevant material is about 200 to 250 msec with a standard deviation (SD) of 100 msec. The SD is a descriptive statistic that indicates the spread or variability in a data set. A more detailed explanation is available here. For an online, statistical calculator to crunch the numbers, try this site:

Teacher: For the baseline reading task, instruct class on how to measure the duration of each fixation within the signal analysis app that you're using. Instruct students to enter the following data values on a worksheet.

• Average fixation duration
• Standard deviation
• Number of fixations
• Number of regressions

For the distracted reading task, students should fill in the same data.

Have students compare their performance during the experimental condition to the baseline condition. In keeping with predictions, an increase in the number of prolonged pauses during distracted reading should inflate the average fixation duration as well as the SD. This finding suggests that reading behavior stalled . Secondly, the number of regressions should increase if students need to re-read poorly understood sentences.

Home Assignment: Ask students to prepare talking points for in-class, group presentations describing their personal experiences involving divided attention, such as listening to a TV newscast while simultaneously trying to read and understand captions scrolling across the bottom of the screen.

Reading scientists use EOG as well as data obtained from other types of eye tracking hardware combined with sophisticated computational methods to reveal episodes of mindless reading where the eyes move across a page even though the reader is not attending to the text. Such assistive technologies have the potential to encourage mindful reading and consequently increase the impact of teaching materials through metacognition by alerting students when their attention no longer text-focused.

The proposed classroom activity developed for students in AP Psychology aimed to:

• expand their general knowledge of reading research;
• introduce them to the skill set needed for empirical research.

Specifically, students are introduced to important EM variables in reading theory, the relationships among these variables and the use of research methods to test these relationships. The preceding lab exercise offers students a unique perspective on their individual reading styles. A take home assignment aimed at reinforcing concept learning encourages students to consider and share their experiences with distracted visual attention.