Efficiency in kerning pairs

María Ramos

Does anyone know of research/statistics in kerning? How many kerning pairs are considered acceptable so it doesn’t increase too much the weight of the font files? One can refine a lot the space between pairs but I guess there is a middle point between the perfectionism of the type designer and what the reader will notice.
I know it depends much on the design of the typeface, but it would be interesting to have some numbers about the amount of kerning pairs in fonts considered to be efficient, both in their visual appearance/legibility and weight of the font files.

Andreas Stötzner

You can hardly define this issue by exact figures. Common sense is your friend.
My golden rule nº 1: Do very careful spacing of all glyphs first.
Golden rule nº 2: Restrict kerning to pairs which are absolutely neccessary.
– Makes the best of your font and keeps your file size and work load handy.

Hrant Հրանդ Փափազեան Papazian

Well one issue is technical limits. I remember during the production of Ernestine they were pretty severe for a font with many glyphs. But from what I understand things have improved since then?

The more interesting issue is what to bother kerning at all. Here I recommend kerning as much as you have time and stamina for. Even something like "¿ß" can happen in text, but it might not be worth it. This makes for very fuzzy decisions...

Probably the best way to limit things is deciding a threshold below which not to kern. Mine is 5 units, in an Em of 1000. In comparison, I use a threshold of 2 when determining sidebearings.

Andreas Stötzner said:

Do very careful spacing of all glyphs first.

But ideally while keeping in mind how kerning will eventually kick in; for example I let glyphs touch (although rarely). This is because environments that don't care enough to enable kerning probably look worse in more important ways...

James Puckett

Yesterday I added a kern pair to make ŦŦ into a ligature. It only appears in one word, ruoŧŧa. But it looks cool, so hopefully someone will come across it one day.

PS: I did more digging and ŦŦ probably exists in more than one word. Ruoŧŧa is just the only one that currently comes up on Onelook.

Thomas Phinney

The desire to limit the number of kern pairs has not usually been about file size, but about app bugs (see below).

Nowadays with class kerning, it is also more likely that there is no direct relationship between some theoretical "# of pairs" and file size, as classes work... differently. Generally if you are worried about "too many pairs" and file size, you should be using class kerning judiciously to reduce the impact of the kerning on file size.

15+ years ago, it was common for applications to misbehave when presented with more kerning data than they knew what to do with. The critical number would be as low as 3K kern pairs, but more commonly ~ 4K to 8K. There were a variety of different bad behaviors after the critical # was reached, depending on the app:

- ignore pairs after the first X pairs
- ignore pairs before the last X pairs
- read X pairs and start writing over them with additional pairs, starting at the beginning

I remember limiting fonts to 4K or 5K kern pairs because of this.

Apps have gotten better since then, of course.

Simon Cozens

On one hand I hear people talking about thousands of kern pairs. On the other I hear people like Andreas saying "space everything as well as you can, then kern only what is necessary", and Bruno Maag saying "kern straight-straight, straight-round, round-straight, round-round and be done with it". I wonder if he's oversimplifying for effect, but I take from that a pushback against over-complicated kerning and over-reliance. I wonder how to resolve this.

I have to be a bit careful saying this, but... I wonder if there is a danger in coming to think of the number of kern pairs as a proxy for font quality. And I wonder if it becomes a pride issue - a way of showing how seriously you are taking things.

I just opened up a bunch of fonts. Three were by truly world class designers, whose names you will know. The first was a display face with no kerning at all, which surprised me. It's still a great font. The second was a text face with 750 pairs (no categories) and all of the kerns were at 10 unit intervals. Applying categories and compressing, I got it down to 345. Another, which is widely regarded as an absolute classic display font, has 890 pairs.

The fourth was by a designer I had never heard of. It has 230 glyphs, and even using categories has 2557 kern pairs, and they're all over the place: six units here, eight units there; /K/V gets -32 units whereas /K/W gets -31. Does that really make a difference? I don't know. But I think the designer was able to satisfy himself that he worked very hard in a way that he would not be satisfied with if he had only 300 pairs. Is there some internal pressure on people to achieve a certain number of kern pairs?

"Space like hell, kern when you must" makes a lot of sense to me. How many "kern pairs" did some of the most famous letterpress types have? And every day I see software that still doesn't apply kerning at all. But still, it's hard for me to avoid the feeling that if I end up with less than five or six hundred pairs I haven't done a good enough job.

María Ramos

Thank you all. Of course, well-defined side bearings is a must.

Interesting to read that some type designers just use round numbers. I tend to use even numbers —both in side bearings and kerning— mainly because they work better when interpolation comes into play. 2 units can make a difference to my eye, but this will probably not make a difference to most readers.

I think some research on this specific topic will give us some clues to be more efficient. If you know of some interesting read about it, please share it here.

Khaled Hosny

All Lucida fonts has no kerning at all, and I wouldn’t have noticed if I weren’t told about it.

Kent Lew

Simon Cozens said:

How many "kern pairs" did some of the most famous letterpress types have?

Practically none. I’ll refer you back to this conversation.

Thomas Phinney

I will throw another wrench in the discussion: character set matters!

If you have an old digital font with 240 glyphs, and a newer one with CE support and small caps and oldstyle figures and 360 glyphs, and kerning is still in ye olde-fashioned pairs, then the “same” kerning will mean twice as many pairs. At least. (1.5 squared is 2.25).

So this is another reason why newer fonts have “more” kerning than older ones.

That, and the tools make it easier, of course.

James Puckett

There’s a continuum of desire to add kern pairs. On one end are fastidious type designers who try to kern every possible thing that goes wrong; they can’t bear the thought of opening a magazine and seeing gaps in W,” set with their types. The users and readers must be protected. These users have thousands of kerns in their fonts, including the numerous exceptions that come from adding the diacritical marks needed for central Europe and Vietnam.

On the other end are the realists. They know that most designers aren’t even aware that fonts have kerning. They have accepted that many designers use optical kerning by default and think kerning doesn’t exist in browsers. They don’t bother with kerning exceptions for accents because people in those countries don’t buy fonts anyway.

Christopher Bergmann

Simon Cozens said:

The fourth was by a designer I had never heard of. It has 230 glyphs, and even using categories has 2557 kern pairs, and they're all over the place: six units here, eight units there; /K/V gets -32 units whereas /K/W gets -31. Does that really make a difference? I don't know. But I think the designer was able to satisfy himself that he worked very hard in a way that he would not be satisfied with if he had only 300 pairs. Is there some internal pressure on people to achieve a certain number of kern pairs?

I may be wrong, but this sounds like the result of automated kerning, rather than of overzealous manual kerning. Depending on the kerning algorithm, it may or may not be possible to define classes or a lower threshold for kerning increments. If it is not (or the user fails to do so), some minute difference between V and W could produce the equally minute difference of one unit. I don’t expect any (type) designers to think that this one unit would make a visible difference, but any automatic process is only as good as the settings it is guided by.

Chris Lozos

A dozen years ago, I tried automated kerning.  It made such a mess that I have never tried it since. I can do kerning by hand faster than cleaning that shit mess.

I prefer to do a thorough job of kerning including diacritics.  It is not that tough.  The hardest ones are the i diacritics and I make a separate class for them.

John Hudson

Most kerning in Latin fonts involves diagonal or overhanging shapes., i.e. the shapes that are exceptional to the straight-straight, round-round, and straight-round relationships that determine the default spacing.

The only times you should ever find yourself kerning a straight to a round is an uppercase straight being followed by a lowercase round, which in some designs might benefit from being slightly tightened.

When I'm kerning, I look at everything, including straight-straight, etc. in context of words, because its a process that I find useful to confirm default spacing. But when I examine the results in the kerning, I typically see the same set of kern pairs involving diagonals and overhang shapes.

How many "kern pairs" did some of the most famous letterpress types have? 

None. And diagonal and overhang shapes produced big gaps within words. We can do better.

[Deleted User]

The user and all related content has been deleted.

Ray Larabie

The fact that fonts in previous centuries had little to no kerning seems irrelevant to how fonts are made today. Those designers were working against certain technical limitations. Technical limitations create style. We're working against a different set of limitations. Thousands of kerning pairs combined hundreds of kerning classes leaves us with almost almost unlimited possibilities.

They don’t bother with kerning exceptions for accents because people in those countries don’t buy fonts anyway.

Language support beyond your likely customer base is there to make localization easier for your customers. Multilingual apps, packaging and instruction manuals are the norm these days. Apps have fluid localization. They often start in one language, gain popularity and languages get added one-at-a-time.

Hrant Հրանդ Փափազեան Papazian

Simon Cozens said:

I have to be a bit careful saying this, but... I wonder if there is a danger in coming to think of the number of kern pairs as a proxy for font quality. And I wonder if it becomes a pride issue - a way of showing how seriously you are taking things.

Don't be too careful, it's very much a latter-day problem in the field. But equally problematic is the cavalier dismissiveness towards robust kerning. Especially when one spends so much time determining sidebearings...

Simon Cozens said:

How many "kern pairs" did some of the most famous letterpress types have?

Their kerning was limited to over-hanging bits and filed-down sides (and ligatures, in a way) but I agree with Ray: that's not very relevant.

James Puckett

FWIW, I wasn’t stating that the people on the extremely limited side of the kerning continuum are correct. Just that there are too extremes and most people fall in-between.