{To Follow}

Don’t say {To Follow} but give me your suggestions instead! ;)

Give me a bit of time. I just didn't want to miss following this thread!

hhp

I don’t think this is the kind of thing you can get an ad hoc answer to.
If I were to address this question seriously, it would take quite a bit of research.
Certainly I could trot out my favorites, but that would exclude faces I don’t like which I might nonetheless consider well spaced, if I were to consider them at all.

• FF Tundra
• Vesper

According to the making of for both, the designer went to great pains to space them well. There is also an article on the Diacritics Project that goes into detail about common spacing/kerning ailments. I also think Garamond Premier might have very good spacing, too. I seem to remember someone on Typophile mentioning this.

There must be at least a few excellently spaced and kerned types. I think depending on your era that meant a different thing, though.

By the way, I am curious as to what these alternative methods are. :)

My personal opinion: PMNCaecilia is a very well behaved font family. I also am a fan of Typotheque fonts, eg Brioni, Greta, etc, that look well spaced and well kerned to me (but I am not a font designer, just a user…).
Thinking about it, National is great too. : )

Hi Adam. I think there is a problem is talking only about spacing and kerning, and not about character design. Also there are issues on what kind of compromises you are going to call 'good.'

In my own experience, spacing the lower case of my Williams Caslon Text was not that tough, because I think I got the character widths and stroke weights (these interact) right. Of course I had a lot of help from Caslon and the models he was influenced by. When it came to the italic, I had a lot more problems. I widened and straightened up Caslon's italic, so it really became a different animal. Initially I had the hook terminals at the baseline kicking out beyond the shoulders. And I just couldn't get the design to space and look good. The rhythm was never right. When I pulled the hooks back under the shoulder, then things started to work. But even in the end the italic seemed to need more kerning than the roman. This may have to do with the varying slopes of characters in a true old style italic.

I say all this because I think the key issue is what rhythm is readable, and works. Then good character design and spacing and kerning all follow from this.

Also you have compromises and design philosophy issues for which there is not necessarily one right answer. A prime consideration is always to get the white space between letters visually even. A sometimes competing consideration is not to have parts of letters touching, or close to touching, as that can be distracting. For example, you might want to overlap KA, and RA, letting them touch as the lesser of evils. But you might want to modify the design of the K and R and A also, so that there is a small gap, without compromising too much on visual evenness.

You also have the need to have the letter spacing in proportion to the counters. Here there are problems involving the caps that again don't necessarily have one best solution. When you kern caps to lower case, you have the question of what the space between the cap and the lower case letters should relate to, the counters of the caps and or of the lower case. If they relate to the caps, then you have the caps standing off a bit from the rest of the word. If the space relates to the lower case, then you have a more abrupt change in color between the cap and lower case parts of the word. You can also make the caps smaller to ameliorate this problem, but then you face the question of whether caps should be clearly differentiated in color to serve as their signaling function as caps. So I think there is a fair amount of latitude involved in these decisions involving caps and lower case.

On kerning the caps to one another, I think you have the problem that the inherent shape of caps mean that you get a wide gap in combinations like KA RA LA, etc. The traditional way to deal with this is to space caps out more widely, and that can work. But now people usually want tighter spacing in caps, so it fits into running text, and you then are forced to make more compromises about overlaps and touching. And that means when you have a title, especially one pretty big, all digital fonts that I have worked with need to re-kerned to work really well. That's because when it's bigger you may want to change the spacing, and then either because of that or the original compromises the kerning doesn't work so well any more.

All that being said, I personally looked to Slimbach as my model of excellent craftsmanship on spacing and kerning. I set the same words in my font and his Minion and Adobe Garamond, and if mine didn't work as well as his, it was a sign I'd screwed up. Also, alongside general admiration for his craftsmanship, I had a couple disagreements with his spacing and kerning philosophy. I felt his spacing overall tended to be a hair too tight for text sizes, and his kerning of quotation marks, especially in Minion, was so tight as to be distracting, and so a blunder.

Doing the exercise you are talking about, I would test different rules for kerning against a number of faces of the best craftsmen working today. That would include in addition to Slimbach, certainly 'senior masters' such as Matthew Carter, Sumner Stone, Gerard Unger, Bram de Does, Adrian Frutiger (for sans only) and Hermann Zapf. There are many younger than these who are also excellent craftsmen, including Berlow, Schwartz, Hoefler, Frere-Jones, De Groot, and quite a few more. I would also include classic text faces such as Times New Roman and Sabon.

Finally, I hope Charles Ellertson chimes in. His work as a compositor has involved re-kerning existing typefaces so that they set better, so his critique of some of the best work would be interesting.

I would also include classic text faces such as Times New Roman and Sabon.

I think Sabon falls outside Adam’s original criteria. The original design, as you know, was constrained by the goal to set exactly the same on both Monotype and Linotype, and so it was unitized and had little or no kerning.

Sabon was an excellent solution to a knotty challenge and has always set surprisingly well given the circumstances. But I suspect it would not be an apt exemplar for the work Adam & team are doing.

Oh, I didn't mention Kent's Whitman, which I also used as a model for comparison, and has a lovely evenness.

Kent, I should have explained that I know that Sabon was unitized. But I think that because it was a success in those constraints, it shows how rhythm is the main thing, and the way character widths (of the black) and spacing works in Sabon is worth looking at.

Heh, I was referencing Miller Text and Minion* as my exemplars while fitting Whitman.

I don’t disagree about the success of Sabon. I’m just taking Adam at his word that unitized designs are not useful for what he’s trying to do. My understanding is that he’s intending to run some designs through his team’s new system and compare how closely the results might replicate the originals, and that, for whatever reason, unitized designs would not provide a helpful comparison.

* (the original Postscript Type 1 version)

Perhaps Sabon Next is spaced well?

hhp

Compare Goudy’s neo-Renaissance faces with those of Slimbach.
All are well-spaced, although to quite different principles.
I doubt Goudy Old Style (above) and Minion Pro (below) would look as good with each other’s spacing method. Even if they did, they wouldn’t look like themselves.

Goudy's "oo" is too tight; Minion's "vv" might need a positive kern, or a contextual alt (like Nina did in Ernestine). Speaking of which: Adam, how does the use of alts to help spacing fit in to your project?

BTW Goudy wasn't good at spacing; I was once commissioned to respace Deepdene (although I guess it's possible the bad spacing in that case was the fault of the person who digitized it).

hhp

Two different cuts of Futura:

The sidebearings of Goudy’s oldstyles are taken from Jenson; he revived Jenson’s “lost” method in toto—serif construction and sidebearings—which was predicated on solid fitting.

Yes, I came across that ignorant digitization of Deepdene too.

I suspect that Goudy's spacing was influenced by Goudy's medium: cast metal type. Yes, the “oo” is spaced too tightly—or, rather not spaced at all—yet the o fits nicely with both the l and the v without kerning.

To me as a rule round-rounds need a positive kern, otherwise everything goes out of whack.

hhp

The Goudy spacing shown in Nick’s image is typical of Monotype fitting, and of the era in general. There has been a shift in taste/philosophy/understanding — whatever you want to call it — between then and now.

Recently, I have been studying a few Monotype and Linotype settings in books I admire, rather closely, and I’m beginning to wonder if part of the je ne sais quoi of that period, in addition to the typically cited letterpress vagaries, is also the matter of a different approach to fitting.

Odd... "To me as a rule round-rounds need a positive kern, otherwise everything goes out of whack."

523 fonts and counting, not one has an oo kern pair, and no one has gone out of whack.

IIRC (which I admit happens less and less these days) I got the idea from Matthew Carter.

BTW Adam, although this is not a bad place to pursue this, I think you will get better results on a forum with a greater number of type users as opposed to designers... This isn't simply because many type designers don't actually use fonts enough, but also because designers necessarily form their own ideas of how it should be done (ideas that often arise out of expediency as much as a desire for quality) and this tends to obscure objective evaluation.

hhp

523 fonts and counting, not one has an oo kern pair, and no one has gone out of whack.

On the other hand, I'm willing to wager a beer or two that a fair number of them have lo and ov kerning pairs...

BTW, "out of whack" was not strictly accurate. The reason it makes sense (to me) is that it reduces the need to fix a bigger mess on the negative end.

hhp

Thank you for your responses so far. Certainly, I did not expect an answer "out of a hat", although I believe that some people already have their favorites that they "carry around with them", and could perhaps share the insight.

I think Typophile is the one place on the net where the most educated people working with type meet (type designers, but also critics, reviewers, users etc.). Many younger type designers have looked at existing typefaces and studied them, so perhaps they might tune in.

Indeed, I'd prefer to see newer typefaces in the list rather than the ones which have been "compromised" by the unit system. They may be very good designs, but with a coarse unit system it's hard to retroactively guess whether some decision was an "ideal" choice of a "hard-worked compromise". This is why I'd prefer to see designs that have been made without those constraints. (This is nicely illustrated on http://www.fontbureau.com/nhg/features/ where the advance widths of the old digital Helvetica and of Neue Haas Grotesk are compared).

In certainly do intend to take an "auteur sample", by picking some faces designed by type designers I know and trust (both older and younger generation) and see whether I can find some patterns there that could be applied to the methods we're researching. I do have some selection of candidates of my own, so this thread is intended mostly to broaden my own perspective rather than needing to serve as the only source for information.

Many thanks, and I appreciate your opinions coming!

Best,
Adam

IIRC (which I admit happens less and less these days) I got the idea from Matthew Carter.

Hmm. I’ve spent a bit of time inside several of Carter’s fonts, and not one of them, that I can recall, has any positive round-round kerns.

I believe what you’re describing was a Berthold approach.

Adam — I think I misunderstood your request then as paraphrased above. So, you’re examining exemplars to deduce potential patterns?

To echo someone (I think it was Bill), I find both Slimbach’s and Carter’s work, in general, to be very well spaced and kerned. What’s even more interesting, in this context, is that they both have very different approaches to achieving their results. So, you might find a comparison instructive.

But I suspect these are both already on your list.

Kent, could you characterize generally the differences you have found between Slimbach's and Carter's approaches?

Carter’s approach to fitting is more or less what I would consider traditional, the sort of approach that underlies Tracy’s elementary account in Letters of Credit. And Carter’s approach to kerning is fairly light-handed.

It’s been a while since I delved into Slimbach’s work, but his approach seemed to rely more on micro-kerning. In the original Minion there were lots of single-digit kerns — lots of ±2 or 3 units, even. And there were lots of unexpected pairs.

The re-release of Minion Pro was refitted from the original; but looking just now at the Regular, for instance — if you take /o/ as the first member and pair it with each lowercase letter, you’ll find that there are kern pairs for 24 out of the 26, both positive and negative. That’s a very different approach.

So, perhaps we have to paraphrase Duke Ellington: “If it looks good, it is good.”

Speaking of a massively-kerned typeface, it seems that Silica, by Sumner Stone, went through "six thousand different character combinations of the font before releasing it to the public" (info taken from a mid-Nineties source). I have the typeface and surely it is nicely kerned.

Thanks, that's interesting. I just did the same exercise on my Caslon and found 7 kerns following 'o' out of the 26, mainly the diagonals, with none for the oo. I did do some kerns of ±2 or 3 also, I guess influenced by Slimbach, even though I was going for a more open look than his. I guess it would be interesting to see how much rounding to the nearest 5 changes the look...

@HP "BTW, "out of whack" was not strictly accurate. The reason it makes sense (to me) is that it reduces the need to fix a bigger mess on the negative end."

That's good! It's evolved from a rule, to just making sense in less than a topic. What's best about this change is that you were condemning the spacing of nearly all per-digital types to out of whackness, which is a pretty big condemnation, even for you.

Kent, what are Slimbach's positive kerns on the "o"?

David, don't you think metal types would have been better spaced if they had kerning (I mean in the sense we use it now)? This is way before any round-round discussion.

hhp

You don’t have a copy of Minion Pro, Hrant? Comes bundled with most Adobe software. I know you’re not a “user,” but don’t you have software for testing?

If not, I’ll go look it up later. Or someone else can. Just set the string in InDesign and read out the values in parentheses.

Hrant, I just checked Slimbach's Adobe Garamond on the kerns between 'o' and a following lower case letter. Unlike Minion (the positive oo kern is 9 units), he kerns very lightly. The only kerns following 'o' are with the diagonal characters—vwxy. No kerning on oo.

I think this shows something that is important for Adam's investigation. Slimbach evidently wanted to do something different with Minion: to produce using maximum kerning a very economical typeface. The design with the narrow characters and large x-height also has the same goal.

So here we have the same designer following a different philosophy, and doing a different kind of kerning. To me Minion and Adobe Garamond succeed in different ways, with different goals. So any 'auto' kerning is going to have to have latitude to be most useful for designers.

By the way, whether Minion's solutions are the best way to be economical is another issue. I am biased, but to me a smaller size of Williams Caslon Text is a more comfortable read in the same space, partly because the tight, darker and large x-height Minion calls for more leading. Being a bit wider and looser and lighter, in other words, helps reading comfort more than size.

Kent, I should have checked myself, sorry. I do have InDesign (won it in Thomas's contest :-).

David, would you say Slimbach is wrong in positively kerning "oo"?

BTW, Adobe Garamond is older than Adobe Minion. So I would see the addition of an "oo" kern as part of a gradual improvement in Slimbach's craft.

hhp

The only problem I have with looking at Robert Slimbach’s more recent work is that, with such massive character sets, I know there are some bugs and omissions in the font, and also some compromises made because of the technical limitations of the OpenType font format (the way classes are set up etc.).

I’ve discovered some problems myself there, and even though I know how much effort goes into the design and production process at Adobe, with such ambitious efforts, I think some mistakes are inevitable.

But still, the amount of the bugs is generally a tiny fraction of the entire number of combinations, so still, I do include some of his fonts among my favorites for the purpose of my evaluation.

So I would see the addition of an "oo" kern as part of a gradual improvement in Slimbach's craft.

I would see it also as changing taste.
Whether the direction taken is the result of the person maturing, or the tenor of the times, is hard to say.

I’ve tried “…various alternatives to the current spacing and kerning model…” in several of my types, most of which could be considered Critical Design.

I put positive kerns on the rounds of Figgins Sans, partly so that it would set tight in tracked-in headlines, without collisions. The normal (“text”) fit is derived from doing the spacing with a tight fit, for headlines, as it were, then just opening up the metrics by adding space consistently to all sidebearings.

In Richler I put massively wide sidebearings on the rounds, on the principle that the face is square and high contrast, and the vertical rounds are almost like straight stems. This looks a bit silly-travesty at large sizes, but works well in body.

In Preface, I employed space-hogging glyph shapes from outside the subtly spaced foundry type tradition, to open up the colour of text.

In Goodchild and Nicholas, like Goudy I went back to try and emulate Jenson’s clockwork fitting, in the interplay of glyph form, detail and spacing.

Nick, can you explain what you mean by "Jenson's clockwork Fitting"?

If anyone has Garamond Premier, I'd be interested whether Slimbach changed his kerning philosophy for that, compared to Garamond.

Adam, I think if you restrict the character set you're observing in fonts that you suspect might suffer from such "overload" (like looking only at the basic alphabetics) you could feel much more confident; otherwise you'd be excluding some of the best stuff out there.

hhp

I believe Kris Sowersby's work is very nicely spaced. The spacing in Newzald and National is a point of reference for me, as is Commercial Type's Guardian Text.

I'm often tempted to kern oo, and other round-round pairs, but have avoided the temptation so far, instead aiming to get nice spacing with as little kerning as possible. We'll see how that works out.

Bill, clockwork:

William, I just checked Garamond Premier: "oo" (as well as "oc", "od", "oe" and "oq") is at +9. Same as in Minion... And "OO" is at +16.

When it comes to spacing, I don't mind agreeing with Slimbach one bit.

hhp

Whenever i go off to rethink spacing I look at a small number of well-spaced families in the class I'm interested in, containing as wide a variety of widths and weights as I can find.

@HP "David, don't you think metal types would have been better spaced if they had kerning (I mean in the sense we use it now)?"

Sure! Not having to saw off serifs is one kerning that helped, and the Latin uppercase AFKLPTVWXY were not designed to fit well mechanically to its entire lowercase. But that's far from saying all metal fonts' spacing was out-a-whack for lack of an oo kerning pair.

@HP "David, would you say Slimbach is wrong in positively kerning "oo"?"

Nope! But 9/1000 ain't so far from zero that I'd hold it up for the class. 10 is as small a kern as I'd use, 1/100th of an em, and that's mildly ridiculous. Carrying around 100s of +-2-9 unit kern pairs is... I'm not sure what the word is in English, in Berlowese I just bow and smile these days.

I think, that there are two general ways to do this; one is to space for a smaller number of kerning pairs even if it looks not as well in non-kerning apps, and the other is the opposite, there are more kerning pairs but it works better without kerning.

Far from some mystical development of some type designer's view, a change from one method to the other could simply signal a belief in the fonts being used primarily in environments with kerning enabled. As the web has proven, kerning is not yet a safe bet to me, so I do it the way I always have, which means o, n and p do not kern to each other.

The spacing of metal fonts should indeed not be seen as "out-of-whack", because of their inescapable physical limitations. But if you digitize a metal font and pretend it's not worth improving the spacing (assuming you're not vying for total historical accuracy) to me that's out-of-whack.

> 10 is as small a kern as I'd use

I never go below 5, so Slimbach's 2s do indeed seem like overkill to me as well. But what about sidebearing determination? I myself stop at 2 (and that's for my best effort, not just anything).

Like when making the new NHG, was the 18-unit system replaced with a 100-unit one? :-)

> a change from one method to the other could simply signal a believe in the fonts being used primarily in environments with kerning enabled.

That makes sense on the surface, but I would counter with the obvious: having kerns in place doesn't hurt when there's no kerning.

hhp

I don’t know that it makes sense to cite a positive o-o kern out of context. How large are the sidebearings to begin with? How tight are the straights & arches? And what is being gained or lost in other areas?

The reason it makes sense (to me) is that it reduces the need to fix a bigger mess on the negative end.

That may be all well and good; but in the case of Minion Pro, for example, the o is being kerned closer to most straights (on both sides, in many cases) and away from rounds. So, I don’t think Slimbach is employing this method in order to fix any “bigger messes” on the negative end. There are plenty of negative kerns.

Instead, he seems to “split the difference” and then push things back and forth.

Is this a better strategy? I don’t know.

My point is mainly that you can’t really cite the presence of specific kern pairs out of context. You have to examine the entire strategy and the results.

David, don't you think metal types would have been better spaced if they had kerning (I mean in the sense we use it now)?

You may have meant this rhetorically, but I’ve begun to wonder if the answer to this is so obvious. Yes, it may have been able to be more even (as we now define that quality), but would that make it ipso facto better? Are today’s books easier to read than yesteryear’s? I don’t have a pat answer.

Hi Adam
I’ve just been sent a booklet entitled Framing Fonts by Niko Spelbrink (Lecturis Publishing) , which is the output from his typography elective teaching at RMIT University, Melbourne, where many younger type designers have looked at existing typefaces and studied them. While more concerned with the observation and measurement of character outlines and ink coverage, the last page gives the scope of the project as a list of 41 typefaces (complete with foundry, format and version info), several of which have already been commented on in the thread so far;
Baskerville, Century Gothic, Didot, Garamond Premier Pro, Adobe Garamond Pro, Georgia, Gill Sans, Lucida Grande, Lucida Sans, Minion, Myriad, Quadraat, Scala Sans, Times, Times New Roman, Trebuchet, Arno, Avenir, Bodoni, Bodoni Std, Chaparral Pro, Cochin, Corbel, DIN, Hoefler Text, Kievit, Letter Gothic, Meta, Meta Serif Pro, Mrs Eaves, Museo Sans, Palatino, Proforma, Quadraat Sans, Sabon, Simple, Soho Std, Spectrum, Syntax, Triplex, Verdana.
Perhaps this is based on Niko’s own list of favourites – several of them are very badly spaced indeed, while others were designed on unit-based systems and/or badly spaced when they were digitised. The booklet suggests a broader range and further results in due course. I don’t know how many students are involved.
Seeing as Jos Buivenga’s Museo Sans is mentioned here, would it be inappropriate to mention the spacing & kerning service he uses? Igino Marini’s iKern (which I'm sure you know of already).

I think I did a good job on Charis roman. It is not a pretty font, but the spacing and character fit are now good enough that several different leadings work well with one setting size, to me, the sine qua non of a well-spaced font.

As it is open source, I could email it to you if you're interested.

edit:

Since I don't use class kerning, the accented letters, of which there are many in Charis, are not kerned. This may mean it is not of interest to you.

Charles, what is your view of the extensive kerning style that Slimbach has developed? Is it an advantage? Or do you go with David Berlow's view that less than 10 units isn't worth the effort?

I agree with David in principle -- then find myself using 5 or 8 units for a lot of pairs...

One oddity -- it is rare for me to change a kern value in Matthew Carter's fonts. I add pairs, but rarely change anything he's already addressed. Quite different for Adobe fonts, though I wonder if Slimbach does the kerning himself?

I’d be very surprised if Slimbach did not do the base kerning himself, since it is such an integral part of the fitting of his designs, as opposed to secondary adjustments of intransigent pairs.

> was the 18-unit system replaced with a 100-unit one?

Sorry, this isn't necessarily correct, since even if you limit your sidebearing determination to multiples of ten* if the outlines themselves don't snap to multiples of ten the advance widths will have no such coarseness.

* Actually another flaw in my question, since that's not the same as greater-than-10.

BTW, I assume that anybody who uses no kerns less than a given value should have all the kerns in a multiple of that value*. For example in Vem (Ernestine's Armenian component) all my kerns are multiples of 5.

* Except perhaps very small collision-avoidance kerns.

hhp