656565656

Ori Ben-Dor

Considering the lowermost point at the bottom of each figure, which of the following statements describes your impression?

a) Bottom of 5 looks lower;
b) Bottom of 6 looks lower;
c) They look the same height. 

Jasper de Waard

Same height.

Wes Adams

Both glyphs lean left.

Ori Ben-Dor

Left? I've asked myself if the 6 doesn't lean right...
What about the height?

John Hudson

It depends what part of the bottom of the letters I am focused on. If I focus on the relationship of the left side of the 5 to the right side of the 6, then the five looks very slightly lower. If I focus on the relationship between the right side of the 5 and the left side of the 6, then they look to be the same height.

Thomas Phinney

(a)

Ori Ben-Dor

I'm with Thomas, 5's bottom looks lower to me too.

I guess that's because the radius of the curvature at the bottom is smaller in the 5.

I've checked some fonts, and haven't found one that compensates for differences in the curvature radius by playing with the height.

Should I?

Does that suggest a flaw in the design and I should simply make the 6 narrower or something?

Or should I simply use Jasper's impression as an excuse and leave it?

Ori Ben-Dor

To demonstrate my claim about other fonts, here's Helvetica Neue Black. Do they look the same height to you?

John Savard

To me, the 5 and 6 look very much the same at the top, and also the same at the bottom, but I'm less sure. With the 8 and the 0, at first the 8 looked bigger, and later, the 0 looked bigger.

Ori Ben-Dor

By the way, come to think of it, what I'm suggesting here kind of defies the logic of overshooting.

The bottom of the 8 is less curved than that of the 0, so it looks higher (at least to me, and probably to Thomas too), and I'm suggesting that it (that is, the bottom of the 8) should be lowered to compensate for this.

On the other hand, if the bottom of the 8 had been even less curved to the point where it had been straight, we would have said the opposite: that it's the bottom of the 0 that should be lowered (overshooting).

What's going on here? 

John Hudson

What's going on is that overshoot is intended to provide a general optical adjustment between flat and rounded forms within a narrow band of space. So typically there isn't a lot of variation in precise heights for similar shapes. The other thing to bear in mind is that the optical effect for which overshoot compensates is actually size-specific, while overshoots in typical fonts are not (variable fonts with a size axis linked to actual text size could provide better results). In your images, we're looking at the numerals quite large, at sizes for which we use edge recognition rather than shape recognition. At these sizes, we see the relationship between the bottoms of the numbers in different ways than we would at small sizes.

The other thing that is going on is that our perception of the relative heights of the bottoms of the letters is not only affected by the radius of their curvature. As you've observed, some of what you perceive seems contrary to the idea that broader flatter curves, being closer to flat, require less overshoot than tighter curves. That's because other aspects of the structure of the glyph affect how we perceive the top and bottom. A narrower shape like the zero, with tighter, more pointed vertical extremes than the eight is going to appear taller; indeed, is going to appear to thrust upward and downward, whereas the horizontally aligned counters of the eight make it look wider, thrusting outwards. This is, of course, a problem with tabular numerals, especially at heavier weights: the zero ends up too narrow, as in the Helvetica Neue Black example.

FInally, as I noted in my earlier response, at these sorts of sizes, where one can really only fixate on parts of the shapes, it is possible that one will have a different perception depending on whether one is looking at the relationship on the left or right of a numeral, at least in the case of an asymmetrical one like five.

Ori Ben-Dor

Thanks, John. That's very insightful.

Do you have any practical conclusions? Would you pull down Helvetica's 8 or my 6? Would you change anything following your observation that 56 and 65 make different impressions?

(Let me add, referring to your first point, that I first observed the problem while looking at a test printout set in 60 pt, which is realistic for my font.)

John Hudson

Does it bother you? If so, do something about it. But bear in mind that the optical effect may be due to multiple factors, so mi inclination is only to fiddle with actual overshoot amounts after all other aspects of a glyph are finalised.

Georg Seifert

Continuing on Johns remark the differences in perception might have be caused by the size (and resolution) of the screen people where using. 

Ori Ben-Dor

Academically speaking, it does bother me. In practice, if Helvetica suffers from this problem, I guess I could live with my font suffering from it too. Either way I guess I'll just follow your advice and leave it for the very end.

attar

It's easy to get obsessed over this kind of details. Usually I just set my overshoots to about the same value or according to the shape width, then after modeling all your glyphs when you print test pages you can actually look and tweak overshoots if needed.

Kent Lew

I've checked some fonts, and haven't found one that compensates for differences in the curvature radius by playing with the height.

FWIW, I have on occasion placed the overshoot of an eight or a zero differently than the general overshoot value, if I felt they looked smaller (in the case of eight sometimes) or larger (in the case of lining zero sometimes).

I’m not saying that I was right to do so, or that one should (or shouldn’t). Just that it isn’t completely unprecedented. ;-)

Hrant Հրանդ Փափազեան Papazian

The more non-flat a bound is the more overshoot it needs. To avoid the torture of having a different overshoot amount for every single different profile (not to mention due to their internal whites) we compromise by discretizing the overshoot amounts, like how most people overshoot triangle shapes more than rounds, but not different rounds. The degree of discretization depends on many things, not least intended use size; for a display font that really cares, variant overshoot amounts even within rounds is entirely rational.

BTW Frutiger even applied different "overshoot" amounts to flats of different length! I suspect diagrams like the one below ("Type Sign Symbol", p. 21) have quietly made a type designer queasy...

Nina Stössinger

In practice I think probably the most crucial thing to do is, rather than staring at large shapes on screen, to stare very closely at very good printouts at the size you are optimizing for, and adjust things that feel too bouncy or not aligned at that size and in context. And then once that feels good enough, learn to live with the fact that different shapes will behave differently to some degree depending on that size and context due to whatever else is going on in that shape.

Related thing I’ve wondered about: when people make the /ae or /oe glyphs from the constituent parts does anyone adjust the overshoots? To my eye it often seems like those wide glyphs, whose overall profile becomes flatter as they cross the baseline/x-line twice, seem to need maybe less than their constituent parts, but I also wonder if that’s overthinking it.

Ori Ben-Dor

Hrant H. Papazian said:

The more non-flat a bound is the more overshoot it needs.

Doesn't the 80808080 example convince you that's not always the case?

==

Nina Stössinger said:

stare very closely at very good printouts at the size you are optimizing for

Well, that's exactly the problem: I first noticed the illusion while looking at a printout at about the size I'm optimizing for.

On the other hand, it wasn't a very good printout. Which brings me to another issue: I really need a better printer. But that's for another thread.

Hrant Հրանդ Փափազեան Papazian

The problem with relying on specific outputs is that you don't know which one an end-user will see...

> Doesn't the 80808080 example convince you that's not always the case?

That's due to the internal white. Like how darker weights need a taller x-height to look the same height.

Ori Ben-Dor

Hrant H. Papazian said:

That's due to the internal white. Like how darker weights need a taller x-height to look the same height.

The internal white space surely plays a role here, but I truly think the curvature plays a role too. Try hiding everything except for the bottom; doesn't what's left from the 0 still look lower? To me it does.

Nina Stössinger

Hrant H. Papazian said:

The problem with relying on specific outputs is that you don't know which one an end-user will see...

Which is why I’d think it’s a good idea to work with good quality prints, so you don’t design for a specific distortion but factor as much as possible of that out. Btw as Georg pointed out above, of course screens are specific too, so arguably you’re always looking at “specific output”. But screen + print at least already make for 2 different specifics.

John Hudson

Yes. Unless your brief is to design for a specific output device or condition, it is better to proof at the high end, simply because there is too much variation in the low and medium end. Of course, one gets clients saying 'We want this to work really well on screen, but also in print, and we do a lot of print-on-demand, but also high-end offset, and...'. Oh well, at least I don't hear 'We need it to fax well' anymore.

Hrant Հրանդ Փափազեան Papazian

> it is better to proof at the high end

Yes. The screen. (Reduction lenses welcome.)