Does sentence context help us read small print?Eliza Chong, Kelsey Hanrahan & Steve Mans!eld, Department of Psychology, SUNY College at Plattsburgh

RationaleIt is well established that sentence context helps us read faster. Reading speeds are 1.4 to 2.5 times faster for sentences than for shuf"ed sentences or unrelated words. However, it is unclear whether sentence context helps us read smaller print. Does context help us identify words that would otherwise be too small to read?

MethodsParticipants• 20 students enrolled in PSY101 General Psychology, with normal

or corrected-to-normal vision.Normal and shuf!ed sentences• Sentences were extracted from novels downloaded from Pro-

ject Gutenberg.• Each sentence had six words, and contained no extra punctua-

tion or repeated words (see Figure 1). • The initial capital letter in each sentence was changed to low-

ercase. • Shuf"ed sentences were created by randomizing word order.Print size• We tested at !ve very small print sizes: from -0.3 to +0.1 logMAR

(i.e., x-height subtending 0.042° to 0.105°.)• These sizes span the typical acuity limit of healthy college-

aged students with normal vision.• For comparison: 20/20 text subtends 0.083°; Newsprint, under

typical reading conditions, subtends 0.23°.Procedure• We measured reading accuracy as a function of print size.• The sentences were displayed using rapid serial visual presen-

tation — words were presented one at a time at the same loca-tion on the screen.

• We tested at rates of: 240, 120, and 60#words per minute.• Participants read the sentences aloud, and the experimenter

counted how many words were read accurately.

DemonstrationDoes sentence context allow reading at smaller print sizes?

Results

Discussion• At 240 wpm, normal sentences could be read at sizes 9%

smaller than shuf"ed sentences (p < 0.01).• Slower presentation rates allow reading at smaller print sizes.

Threshold print size at 60wpm is 30% smaller than at 240wpm. • Context provides no advantage when reading very small print

at slow speeds. • Perhaps the processing advantage due to context is offset by

the slow presentation rate. • Or, possibly, it is hard to establish context under these con-

ditions because the text is hard to read.

ReferencesKnoblauch, K. & Maloney, L. T. (2012), Modeling Psychophysical Data in R. New York: Springer.

AcknowledgementsSupported by a SUNY Plattsburgh Redcay award for student-faculty collaboration in the behavioral sciences.

Plattsburgh State University of New York Logo Treatment 3

The Plattsburgh State logo has been created in a variety of colors for use on different color backgrounds. The logo can also be used in white if placed on any of the darker colors from the color palette.

Standard Logo

One-Color UsagePMS 186 RedFile name: PSU_PMS186.eps

Pantone Cool Grey 8File name: PSU_CoolGrey8.eps

Black-Only UsageColor: BlackFile name: PSU_Blk.eps

Color: 45% BlackFile name: PSU_Blk_45percent.eps

Reverse UsageColor: WhiteFile name: PSU_Wht.eps

Normal sentences Shuf"ed sentences

Figure 1. We !nd that the normal sentences support reading at faster speeds, but not at smaller sizes

grieved is sadly mother our poor

all affection own her was his

our watches at both looked we

fact the girl the understand let

like late not you will staying

point gained main the was but

had promise her kept bravely she

him his this sister brought upon

gushed into tears eyes the her

of them the round rest crowded

never chair but do the would

golden not the there were waters

manfully I future the turned to

was darling it now her object

negative she the in was answered

hurt they on your account me

checked in she just time herself

London bring would soon cure its

creature a is quite little she

thinking had hours he for been

but they were very decently kept

we had some accidents with it

he made only one thing clear

we all went into the bedroom

we were all agreed upon that

she could not but be thankful

the little man smiled and nodded

this appears to be your affair

I would rather not be tempted

but the secret soon came out

and so the amiable couple parted

they both looked at each other

she would do anything for me

the conference was a long one

I remember he was in mourning

then she began to blame herself

we have got our first principle

and now my written story ends

she turned and left the house

I know the way out perfectly

PRINT SIZE (logMAR)

PR

OP

OR

TIO

N C

OR

RE

CT

0.00.20.40.60.81.0

-0.3 -0.1 0.1

160

260

-0.3 -0.1 0.1

460

560

-0.3 -0.1 0.1

660

760

-0.3 -0.1 0.1

960

1060

-0.3 -0.1 0.1

1260

1360

-0.3 -0.1 0.1

1460

1560

-0.3 -0.1 0.1

1660

1760

-0.3 -0.1 0.1

1860

1960

-0.3 -0.1 0.1

2160

2260

-0.3 -0.1 0.1

2460

2560

1120

2120

4120

5120

6120

7120

9120

10120

12120

13120

14120

15120

16120

17120

18120

19120

21120

22120

24120

0.00.20.40.60.81.0

25120

0.00.20.40.60.81.0

1240

-0.3 -0.1 0.1

2240

4240

-0.3 -0.1 0.1

5240

6240

-0.3 -0.1 0.1

7240

9240

-0.3 -0.1 0.1

10240

12240

-0.3 -0.1 0.1

13240

14240

-0.3 -0.1 0.1

15240

16240

-0.3 -0.1 0.1

17240

18240

-0.3 -0.1 0.1

19240

21240

-0.3 -0.1 0.1

22240

24240

-0.3 -0.1 0.1

25240

Presentation rate = 240 wpm

Presentation rate = 120 wpm

Presentation rate = 60 wpm60 wpm

0.0

0.2

0.4

0.6

0.8

1.0

-0.3 -0.2 -0.1 0.0 0.1PRINT SIZE (logMAR)

PRO

PORT

ION

CORR

ECT

120 wpm

0.0

0.2

0.4

0.6

0.8

1.0

-0.3 -0.2 -0.1 0.0 0.1PRINT SIZE (logMAR)

PRO

PORT

ION

CORR

ECT

240 wpm

0.0

0.2

0.4

0.6

0.8

1.0

-0.3 -0.2 -0.1 0.0 0.1PRINT SIZE (logMAR)

PRO

PORT

ION

CORR

ECT

Normal sentencesShuf!ed sentences a) individual data b) pooled data

Figure 2. a) Reading-accuracy versus print-size curves from 20 participants. Each column shows data from one participant at three presentation rates (240 wpm top; 120 wpm, middle; and 60 wpm bottom). Gray: normal sentences. Red: shuf"ed sen-tences.

b) Pooled data. These curves show the !xed effects from a mixed-effect model used to !t cumulative-Gaussian curves to the accuracy-versus-size data [as outlined by Knoblauch and Maloney (2012).] The model has threshold, slope, and sentence con-text as !xed effects:; and threshold and participant as random effects.