Typedesign and science, or: look at what I made!

Jasper de Waard

To avoid random sampling effects it's probably better to have both groups do the adjustments in both directions (so, two groups, in counterbalanced order), but now we're getting into the nitty gritty of research design

Nick Shinn

“I attended Davy's lectures to renew my stock of metaphors.”
—Samuel Taylor Coleridge (poet of “The Rime of the Ancient Mariner”) on chemist Humphry Davy.

Vasil Stanev

Mark Simonson said:

An interesting insight from the study:

We believe the most probable cause for the vertical-horizontal illusion, and by extension also the thickness illusion, lies in a probabilistic estimation of three-dimensional length, based on the visual properties of natural surroundings.

This makes sense to me.

There is generally more happening below then above in nature, that is why the top oval of 8 is smaller, and so on in stem width.
Humans evolved from apes avoiding dangers in a jungle while moving in 3D space, which is why we see and think like we do. This explains fear of snakes, aversions of creeping things, love of the color green and awe when looking in the sky. And the vacant male stare into the fire/tv after bringing home the bacon. 

Jasper de Waard

@Vasil Stanev Those are just theories. I doubt there is much evidence to support your story.

@Nick Shinn Que?

Vasil Stanev

Jasper de Waard said:

@Vasil Stanev Those are just theories. I doubt there is much evidence to support your story.

Strange, they seemed pretty logical to me so I did express them as a statement. (Of course we can never know what apemen thought, but we do know how nature looked like).
What are the alternative theories? Always willing to learn.

Jasper de Waard

Of course evolution shaped our perception and behavior to a large extent, but just because a theory makes sense doesn't make it true. Your account of the 8 is new to me, and I don't really see how it would work to be honest. I think a much more logical hypothesis would be that objects that are higher in our visual field are generally further away, but it's only a hypothesis, and a very difficult one to confirm. I guess I'm just trying to say: a little more modesty about what we know about ourselves seems appropriate.

Ray Larabie

I think we should get a group of children to walk on their hands their entire lives. Then, when they're around 40 years old, quiz them on which part of an 8 looks bigger. I think it would be a worthwhile study.

notdef

What we do “know” from the arts tradition is that we perceive the vertical center of an area to be slightly above the mathematical center. To my eyes, the top of ‘8’ appears larger than the bottom if not reduced slightly. Similar illusions apply to B, S, etc.

Congratulations, Jasper! This is inspiring work. I think it was a wise decision to measure the effect in non-letter shapes. This knowledge is valuable beyond just type design.

Craig Eliason

Frode said:

What we do “know” from the arts tradition is that we perceive the vertical center of an area to be slightly above the mathematical center. To my eyes, the top of ‘8’ appears larger than the bottom if not reduced slightly. Similar illusions apply to B, S, etc.

Is it that we see the top as larger, or is it that we feel it should be smaller? That is, do we perceive inequality but want equality, or do we perceive equality but want inequality?

John Hudson

Or do we perceive inequality and want a different inequality?

Mark Simonson

To me, it has always seemed that the top looks larger (or top-heavy) when a character is perfectly symmetrical vertically.

I noticed this years ago with Kabel Bold. The S looks upside down.



If you turn it 180°, it still looks upside down.



(There is a slight top/bottom optical correction above, but it is much smaller than usual.)

Nick Shinn

I suspect the illusion you investigated is more prevalent amongst type designers.

(This is entirely my subjective opinion, based on how, after a day working on italics, upright romans appear to backslant.)

**

To check my bias, I rotate a print-out through 45°.
Can’t do that in FontLab, though.

Mark Simonson

FWIW, I recall noticing the Kabel thing in high school, long before I was a type designer or had any formal training in graphic design. At the time, I think I thought it was intentionally top-heavy. Later on I discovered it was actually symmetrical.

Craig Eliason

You were always a type designer, you just had a latency period.

John Hudson

@Jasper de Waard

To avoid random sampling effects it's probably better to have both groups do the adjustments in both directions...

With that approach, you will get some subjects optically equalising the contrast at lighter overall weights and some at heavier weights. I think also it will be harder for subjects and will take a lot longer. My thought would be that some shapes will have the x direction locked and some the y direction locked, so the adjustment only affects the other orientation. The shapes themselves could be a variety of weights, which might indicate whether the perception of difference in weight is more pronounced at lighter or heavier weights.

Jasper de Waard

Ah, I'm sorry, that's not what I meant. In one trial they adjust ver thickness until it's equal to the static hor thickness. In the next trial they adjust hor thickness with static ver thickness. (So I meant both directions, as in, first in one, and then in the other, but I can see the confusion.) This can be repeated several times (to obtain a more reliable average), as is actually common practice in this kind of research.

Of course, testing this at different weights is easy, and could be interesting.

Eris Alar

How would you do this and avoid people measuring based on where the slider is as opposed to looking at the shape? I suspect if I was doing it, I'd assume if the slider was in the middle, or at on end, then that is zero contrast, I probably would not look at the shapes. 

Jasper de Waard

Yes, it's better if the slider is invisible, and controlled by arrow keys. That should be possible, though, right?

Jonathan Hoefler

Thomas Phinney said:

I think it would be an interesting experiment to take a variable font with a contrast slider, and let the participants adjust that slider until they perceive zero contrast.

Since letterforms are pleiotropic, I think it would be very difficult to disentangle one set of influences from the others. The perception of zero contrast in a lowercase E would presumably be shaped by the way we perceive weight accumulation around intersections; zero contrast in a capital N might have more to do with expectations about stress and hierarchy. Even reducing things down to an I and an O could contaminate the results owing to overshoot. This is why I'm especially excited about this study, since it seeks to test a single hypothesis in a clean and rigorous way.

Ray Larabie said:

I think ITC Avant Garde is a good candidate.

Perhaps another reason to stick with geometric figures is that typefaces are prone to cultural and fashionable biases. I wasn't certain whether Ray was being sarcastic here; one person's normative typeface is another person's museum piece (and another person's horror!)

Jasper, kudos to you and your colleagues, and thanks to the three of you for writing a paper that's so readable by a non-technical audience. This is really a lovely and valuable piece of research, and I'd be thrilled if these lines of inquiry continued!

Hrant Հրանդ Փափազեան Papazian

Big thumbs up to empirical research in type design! Bravo, Jasper and team.

Georg Seifert said:

Have you done the horizontal/vertical test with native Arabic speakers? As their writing system has a reversed contracts compared to Latin it might be that they trained in the opposite direction.

Extending the testing with subjects who are [more] familiar with writing systems with horizontal contrast (arguably even nativity in such) is critical in cementing, attenuating or (hopefully not) invalidating this result. We need to get at the underlying "firmware" which we can then judiciously apply to a given contrast tradition.

Adam Jagosz

@Jasper de Waard Great job! And an interesting read. There's one thing I don't quite grasp, though:

The continuous line represents the optimally fitted Gaussian curve for the group mean data.

As a layman, I don't know what group mean data is, or what the curve was fitted to.

AbiRasheed said:

I just felt the paper was overcomplicating a simple issue with all the math and phrasing. 

That's how a bachelor thesis works

Ray Larabie

@Jonathan Hoefler I was being serious about ITC Avant Garde Gothic. Avant Garde, regardless of its readability, features shapes that are supposed to look like circles but actually aren't. It would be easy to overlay a pure geometry version and interpolate. It has a minimum of tapering where bowl meets stem. I can't comment on any digital versions; I'm looking at the analog ITC Avant Garde Gothic Medium.

Jasper de Waard

Thank you everybody for the kind words, I'll be sure to pass them on to the others too.

@Jonathan Hoefler I'm very glad you enjoyed the read, and I'm fairly sure that I will do more research along these lines. I fully agree with your analysis of letters vs geometric shapes — there are simply too many confounds.

@Hrant H. Papazian I agree that non-latin research would be interesting, but there is also a problem with cross-cultural methods. If the illusion is smaller/bigger (or even reversed) in a different sample, my hypothesis would be that this has to do with their writing system, but it's difficult to prove — there will always be variability in other aspects too. In other words: you can show correlation, but not causation using cross-cultural comparisons.

@Adam Jagosz Group mean data means that for every point on the x-axis (-12, -9, -6, etc.) we took the mean % of 'same' responses in the entire sample. These averages are represented in data points on the graph. Then we created a Gaussian curve (this is just a kind of distribution) in such a way that it would come as close as possible (called 'fitting') to the data points. Thus, the Gaussian is used as a model to represent subjects' response distributions, and the top of the Gaussian represents the PSE (the point where hor and ver lines are perceived as equally thick).

I would say it's how science works, but that's a matter of debate of course

@Ray Larabie I get your argument, but you're not going to convince a scientific reviewer with an argument about what the supposed intentions of a certain typeface are. 

Adam Jagosz

I mean, I see one dashed curve and one continuous, and don't know how to read them. I did get that far that since the curve peak is not at x=0, the illusion is present, but there is something true about what Abi complained about: you really need to read the whole thing to figure it out. Yeah, science in general, but for some people the science part of life ends at bachelor's thesis

Jasper de Waard

Aha, I guess the legend and the text underneath the figure could have been a bit more clear there. The dashed line is the Gaussian fitted to the data from the 'vertical frame' condition, while the continuous line is fitted to the 'horizontal frame' condition data.

Adam Jagosz

Ah yes, it's right there. The line patterns could have been indicated in addition to point colors, and the whole could have been left-aligned instead of each label centered beneath a graph, which trips you into thinking each is related to a graph when they both describe both of them. And there I was thinking that one curve is fitted and the other "unfitted" Thanks.

Jasper de Waard

In case you missed it, an interesting take on the matter by @Jonathan Hoefler (a few days old):
https://www.typography.com/blog/turning-type-sideways

Hrant Հրանդ Փափազեան Papazian

Since we've only been making tall structures for a blip in time compared to the æons of evolution, I think it must have to do with gravity (which BTW is quite impervious to social conditioning).

Christian Thalmann

Hrant H. Papazian said:

Since we've only been making tall structures for a blip in time compared to the æons of evolution, I think it must have to do with gravity (which BTW is quite impervious to social conditioning).

Then again, trees. And people, for that matter.

Hrant Հրանդ Փափազեան Papazian

I guess if the trees are straight enough. Which means the whole test has to be redone based on forestation nativity.  ;-Þ