Fonts for Interlude Reading:
Improving Readability in the Digital Age

We define Interlude Reading

In our age of ubiquitous digital displays, adults often read in short, opportunistic interludes. This duration continuum lies somewhere between glanceable and long form reading. We define "Interlude Reading" as the kind of reading that happens in a single brief sitting (i.e., a few paragraphs worth). In this work we consider whether reading outcomes in this unique form can be improved by tailoring typeface to the individual.

We study Interlude Reading

People are most familiar with book fonts

Content (left) participants reported reading most for leisure. Text size maps to frequency of response. The fonts commonly used for the related reading content (right). Font name is rendered in its own font type.

We consider 16 fonts

Collaborators

Shaun Wallace

PhD Candidate
Brown University

Zoya Bylinskii

Research Scientist
Adobe Research

Ben Sawyer

Assistant Professor
University of Central Florida

Rick Treitman

Entrepreneur-In-Residence
Adobe, Inc.

Nirmal Kumawat

Computer Scientist
Adobe, Inc.

Jeff Huang

Assistant Professor
Brown University

Kathleen Arpin

Learning Expert
Riverdale Country School

Towards Readability Individuation: The Right Changes to Text Format make Large Impacts on Reading Speed

Paper Abstract

In our age of ubiquitous digital displays, adults often read in short, opportunistic interludes. We consider, for the first time, whether reading outcomes in this unique "Interlude Reading" can be improved by tailoring typeface to the individual. Hundreds of participants provide a foundation for understanding which fonts people prefer and which make them more effective readers. Results reveal that while 77% believed their preferred font would be fastest to read in; this was only valid for 20%. Differences between best and worst font average 75 words per minute (WPM), with no significant changes in comprehension. High WPM variability for every font suggests that one font does not fit all. We here provide recommendations for favorable fonts related to higher reading speed without sacrificing comprehension and suggest that our methodological approach can be used to model for individuation, allowing digital devices to match their users' needs in-the-moment.

In this paper, we lay a foundation for...

understanding systematically what people like in a text format and what makes them more effective readers. We set up large-scale reading studies, covering hundreds of participants across diverse user populations, from online crowdworkers to college students and professionals. We focus our attention on paragraph reading using digital tools. Through our studies, we seek to answer the following two questions:

Universality VS Personalization:

Does one size fit all? Are some fonts generally preferred and effective across a population, or are there large individual differences?

Preference VS Effectiveness:

Do people know what's good for them? Are people most effective at reading in the fonts that they prefer most?

How do we find your most preferred font?

We designed an interface to toggle between two fonts in order to choose the preferred one, using the prompt: "What font is easier for you to read in?". Toggling between pairs of options at a time provides a simple and efficient method for assessment, motivated by other pairwise comparison tasks in the wild, such as eye exams and hearing aide adjustments.
Pairwise comparisons are a standard method across different fields to derive personal preference. However, determining a participant's preferred font among 16 fonts can be a time-consuming task if they make every possible pairwise comparison. Our system uses a double-elimination style tournament, where a font is eliminated after a participant picks against it twice. This method decreases the number of total match-ups, removes the possibility of ties, and arrives at a definitive winning font per participant.

Toggle test example

Home at Mount Vernon the candles in the windows of George Washington’s home shone brightly. This Christmas Eve, though, was different. One month earlier the United States and Great Britain had signed a peace treaty ending the Revolutionary War.


Preliminary study

Starting with the hypothesis that people’s font preferences can point to more effective fonts, we first designed and validated a method to determine a participant’s preferred font using pairwise comparisons. We recruited 60 participants (ages: 18-55, 51% female) from university mailing lists, usertesting.com, and Amazon Mechanical Turk.

Guiding questions:

  • Which are the highly-rated fonts?
  • What factors influence font preference?
  • Do people have similar preferences?

Take-aways:

  • Noto Sans is a top-rated font across participants.
  • Effective font size influences font preference.
  • Font preference is not driven by familiarity.
  • Participants vary in their preferences for fonts.

Font normalization study

Not all fonts are created equal. The preliminary study showed font size drove font preference. While prior work proposes to normalize font sizes according to a particular attribute (e.g., x-height), here we take a crowdsourced approach to finding the attribute, per font, that perceptually normalizes its size the best.
Take-aways:
  • Effective font sizes vary significantly across fonts.
  • The method to normalize a font’s size is font-specific.

How do we assess human perception for font size?

Taking Times at 16 px as our reference font, we computed three new font sizes for each font in our set, corresponding to matching the reference in each of x-height, height, or width. We recruited 60 participants ranging from professional designers to crowdworkers.

Normalized font sizes


Font study across a large population

The motivation of the present study was to determine whether people are most effective at reading in the fonts that they prefer, after controlling for font size and reading comprehension. To answer this question, we ran experiments on hundreds of participants on Amazon's Mechanical Turk.

Guiding questions:

  • Which are the highly rated fonts (controlled for font size)?
  • Is people's preferred font their most effective font?
  • What gains in reading are achievable by font choice?

Study description

  • 386 participants from Amazon’s Mechanical Turk (ages: 18-71, 46% female) out of 500 participants initially recruited.
  • Study design: 34 minutes, including (1) pre-survey, (2) preference test, (3) effectiveness test, (4) post-survey.
  • Preference test: toggle interface to compare pairs of fonts in a double-elimination tournament until a single font emerges as a winner.
  • Effectiveness test: read 10 short passages (69-93 words each) in 5 different fonts, while WPM (reading speed) was tracked along with with a comprehension test (2 multiple choice questions) after each passage.

Variables accounted for...

  • Reading speed: outliers removed; all measurements are within 100-650 WPM range [2]
  • Comprehension: outliers removed; all comprehension scores in the 71-100% range.
  • None/minor effects of content (familiarity, interestingness, fiction/non-fiction) on reading speed and comprehension.
  • Take-away: Font familiarity does not affect font preference or effectiveness.

Results

(Sortable Heatmap Table: green: top 5 and red: bottom 5 per column)

Font Most
Preferred
Win
Rate
Average
Elo Rating
Disagr-
eement
Font
Familiarity
WPM Speed
Rank
StDev
WPM
Reading
Comp.
Noto Sans 56 62% 1639 90 1.89 272 48% 108 91%
Times 56 58% 1596 115 2.50 277 50% 108 91%
Avenir Next 41 54% 1554 97 1.74 264 45% 106 93%
Helvetica 36 59% 1608 87 2.22 283 50% 102 89%
Calibri 35 55% 1573 90 2.34 276 56% 102 92%
Garamond 34 52% 1543 103 1.90 310 48% 120 91%
Arial 33 57% 1591 86 2.40 270 47% 103 93%
Open Sans 19 56% 1585 77 2.03 255 54% 91 90%
Roboto 14 53% 1556 84 1.83 268 47% 106 94%
Montserrat 13 42% 1451 90 1.77 271 57% 109 87%
Utopia 13 44% 1464 105 1.81 274 48% 116 86%
Avant Garde 11 38% 1398 104 1.83 261 29% 90 94%
Oswald 10 16% 1154 127 1.70 295 58% 99 89%
Lato 6 49% 1519 72 1.73 293 54% 99 91%
Poynter Gothic 5 44% 1473 78 1.82 265 52% 97 93%
Franklin Gothic 4 27% 1296 87 1.79 270 56% 107 89%

Preference is personal

  • Noto Sans and Times were the most preferred across all participants.
  • Noto Sans also had the highest win rate and the highest average Elo Rating.
  • Each of the 16 fonts was chosen as a favorite font by at least 4 participants.
  • High amounts of disagreement across participants in the toggle tests when picking a favorite out of two fonts.
  • Participants described their preferred font as being bold and modern, having natural character spacing, and does not cause eye strain. Qualitative preferences were not consistent across participants.

Font Preference ≠ Effectiveness

Participants read the fastest in their most preferred font 20% of the time, but they also read the slowest in their preferred font 19% of the time (out of five total fonts tested per participant), which works out to precisely chance level.
Overall participants read in their preferred font at an average WPM. Participants do no better or worse, on average, by reading in the commonly preferred fonts (Noto Sans and Times).
These findings run contrary to participants' beliefs: 73% of participants believed their most preferred font would be their most effective font to read in.

The right font can speed up reading

  • Participants read 32% faster in their fastest font compared to their slowest font.
  • Participants read 14% faster in their fastest font compared to their most preferred font.
  • The fastest font is different across individuals.

A call to action for developing reading tools and insights in the digital age:

Different fonts are effective for different people, leading us to believe that custom reading experiences can help people read more effectively. There is, given our pattern of findings, an exciting opportunity to augment reading performance for adult readers:
Our average reader could add 38 words a minute by merely adjusting their font, equivalent to an additional 3-4 pages an hour. Participants in our top quartile for the delta between best and worst font would add 93 WPM, or eight pages an hour. In both cases, average comprehension remains similar and high. In the context of interlude reading, this gain is perhaps best framed in the ability to consume more in limited windows. If a news article, journal, or forum post is roughly 700 words, requiring around two minutes for an average reader, that individual could read it 24% faster in their most effective font while still retaining normal levels of comprehension. The approximately 30 seconds saved in this case might be used to read comments or look at related posts.
The potential impacts on individual reading efficacy highlighted here point to a future in which machines help adult readers to reach for their full reading potential. We invite the present reader, and the multidisciplinary communities that will perform this work, to join us. Let us engineer better reading for everyone.

Acknowledgements

We have a lot of people to thank for giving our work the resources and context it needed to bloom: Drs. Benjamin Wolfe and Jonathan Dobres for formative discussions about reading that helped shape some of our initial ideas; Allen Ellison and Emily Seminerio of Adobe Inc. for providing us with helpful resources; Kathy Crowley and Marjorie Jordan of Readability Matters for helpful discussions of the study results; Dan Rhatigan and Tim Brown of Adobe Type for insightful discussions about typography; Tong Sun of Adobe Research for her support.

How to cite this work

@misc{readabilitymattersFormats:online,
author = {Wallace, Shaun and Treitman, Rick and Kumawat, Nirmal and Arpin, Kathleen and Huang, Jeff and Sawyer, Ben and Bylinskii, Zoya},
title = {Fonts for Interlude Reading: Improving Readability in the Digital Age},
howpublished = {\url{https://shaunwallace.org/readability/}},
month = {Oct},
year = {2019},
note = {(Accessed on 09/20/2019)}
}

Related work

[1] Ronald P Carver. 1990. Reading rate: A review of research and theory. Academic Press.

[2] Ronald P Carver. 1992. Reading rate: Theory, research, and practical implications. Journal of Reading 36, 2(1992), 84–95.

[3] Kathy Crowley and Marjorie Jordan. 2019a. Base Font Effect on Reading Performance - Readability Matters. https://readabilitymatters.org/articles/font-effect.(June 2019). (Accessed on 09/19/2019).

[4] Kathy Crowley and Marjorie Jordan. 2019b. Readability Formats Offer Instantaneous Change - ReadabilityMatters. https://readabilitymatters.org/articles/instantaneous-change. (Jan 2019). (Accessed on 09/20/2019).