Highsmith on Glyph Space

kentlew's picture

Cyrus Highsmith just posted some reflections on what he calls “glyph space” (. . . the Final Frontier — cue Star Trek theme) over on FB’s Type 101 blog the other day.


(I bet this will appeal to Hrant especially.)

John Hudson's picture

Brian: If we had parametrizable jamo...

I think I know what you mean by this. But consider that such parametrisable jamo don't necessarily need to exist within the font format itself: they could be a tool for creating variant glyphs which could then be applied as modular components.

Jongseong's picture

I have to admit that I do not know enough about the technical side of things. Since I want to be able to generate parametrizable jamo (like specific instances of a multiple master font, conceptually), not pick them from a pre-generated collection, I thought it would be necessary to build this into the font format. But perhaps what makes sense is a supplementary tool that generates the necessary glyphs and adds them to a font that already contains the most needed hangul syllables pre-composed.

Right now, designers of hangul fonts do use multiple masters to generate specific jamo instances and build the syllables systematically, but once they are done the font file just contains the specific glyphs and components so produced, making extension very difficult. What I would like to see is a way of preserving the generative capacity that is currently confined to the design process so that end users can easily call up additional syllables. Ideally, they would be able to do this from their word processor without needing any fancy additional software.

So this may not be a question of font format per se, but maybe something for the text engine to handle.

quadibloc's picture

I do not know enough about the required modifications of components in a Korean syllable to comment on the task itself. I would think, though, that normally it would be understood that while normal Korean text would be set in a text face, if it were desired to include syllables not in normal use, that they would be in a "typewriter" face would be accepted and understood.

Of course, that too would be within limits: while it would be acceptable in a technical discussion of archaic Korean, or when text containing foreign words transcribed in an obsolete system is quoted, I don't expect that it would be considered reasonable for everyday extensions to the basic syllable repertoire - i.e. transcriptions of foreign names by means of the system currently in use.

It might be, however, that limiting the problem in this way would be sufficient to reduce the number of variants of each element for composition to a reasonable number.

Even for the Latin alphabet, fonts allow kerning and ligatures to be specified, and so it should be technically possible for there to be several forms of each consonant and vowel to deal with changing the stroke width as the size is reduced, changing the centering depending on adjacent symbols, and so on.

The results would not be as good as individually designed syllables, however, so instead of the result being obviously due to technical limitations, as a typewriter-like syllable would be, it might be just good enough to be mistaken for poor-quality work by hand. Sort of the "uncanny valley" problem.

Jongseong's picture

John (quadibloc), something like you describe does happen, except it involves using fallback fonts that include archaic hangul syllables rather than typewriter-style face.

I listed several uses for which modern Korean syllables would be insufficient, but by far the greatest demand is for representing older orthographies of Korean. Archaic syllables turn up in Korean texts written from the creation of hangul in the 15th century until the early 20th century, and as such are indispensable for the standard Korean literature curriculum. So you can find examples in any textbook dealing in early Korean literature. (Archaic hangul is not needed for everyday use, although once they become widely available in fonts it's probably a matter of time before people decide to exploit them to transcribe foreign sounds.)

I think I've only ever seen three typefaces used for representing archaic hangul: a serif, a sans serif, and the New Gulim, the rounded sans seen in the sample in my earlier post which honestly is as suitable as Comic Sans for academic texts.

So most often, the same serif typeface that supports archaic hangul is used for the entire textbook or at least for the extracts of literature. Sometimes, the quoted texts are set in a font that doesn't support archaic hangul, so the archaic syllables from one of the fallback fonts are substituted. And I have seen examples in older books where archaic syllables were simply written by hand.

Typewriter faces in Korean are built differently from normal text faces, as they break the square paradigm, so to speak. You would never mistake one for the other. So archaic hangul syllables in typewriter faces will be too different from the surrounding text to work. In contrast, the variety in the proportions of different hangul serif designs is not that great, so if the main text uses a hangul serif font, then really the only thing to worry about in using the fall-back font is matching the weight.

blokland's picture

Earlier in this thread I mentioned the spacing mechanism I am working on, which is based on what I baptized the Rhythmic System. Basically this places a repetition of a Harmonic Model underneath the related characters to come to a spacing based on the rhythm within the letters. In essence the round parts are treated as overshoots of the straights.
The current version of LeMo contains the Harmonic System, which forms the basis for the written variants of the formal group of ‘Textletters’, basically from Textura to Humanistic Minuscule and subsequently of the typographic variant of the latter, which we call ‘roman’. This Harmonic System comprises two Harmonic Models, of which the primary one provides the construction for all letters which are not derived from the capitals.
During discussions on LeMo some colleagues noted that for instance spacing based on the Rhythmic System is restricted as such and can basically only be applied within LeMo. Actually, I am strongly convinced that the Rhythmic System was applied by for instance Jenson for the spacing of his ‘archetype’. The following illustrations show that the Rhytmic System can easily be applied on Jenson’s type and that even the word-spacing follows this structure.
The model used in these illustrations makes use of the parameters of the ‘Renaissance’ preset in LeMo. Please note that the version with the serifs places these a little bit too low for Jenson’s type; I will make a fitting preset for Jenson’s letters in the next LeMo edition. Also the photographic enlargements are a bit crude.

In most (all?) books on typography, the transition from Textura to roman type (after the invention of movable type was introduced in Italy) is just a matter of taste (Textura being considered ‘Gothic’, i.e., barbarian) and the fact that Jenson and co. were confronted with a more complex model to space than the Textura one, seems to be simply ignored.

Spacing a Textura is, because of all the straight parts, relatively easy. When based on the Rhytmic System (top row) a standardized system can be made in which the ‘i’ and the ‘l’ fit into one unit, the majority of letters, like ‘n’, ‘h’,‘u’, ‘o’,‘b’,‘d’,‘p’,‘q’, et cetera, on two units and the ‘m’ on three units. The center row shows letters made with LeMo and in the third row the ‘feet’ of these letters were centered and the arches shifted accordingly.

Actually I am quite sure that Jenson used this structure for the spacing of his type too and that he treated the curved parts simply as overshoots. In the illustration the spacing is again based on ‘Rhythmic System’ and subsequently results in a standardization of widths.

If Gutenberg and Jenson really have made use of such a structure, they were applying a sort of unit arrangement system roughly half a millennium before the manufactures of the hot metal composing machines were doing this. Later this year I will do some traveling and measuring to find out if there is proof for this theory in casted type.
One wonders if the developers of the Linotype and Monotype machines were aware of widths standardization by their precursors.
Needless to say that I am focusing on the italics now.

Christopher Adams's picture

@ Frank Blokland

In the meantime, would it be helpful to illustrate the rhythmic system of Jenson's types with a photograph from [Bad link]?

Your present illustrations are clear enough; I merely want to indicate a resource and track this thread.

Nick Shinn's picture

Frank, your theory is strongest when it applies to what Alistair Johnston refers to as the "prime rib" characters (in Alphabets to Order), that is to say, the very Latin characters that appear in the typical "Quousque tandem abutere..." specimen text of yore -- without k, v, w, x, y, and z, with their awkward diagonals.

I'm not sure how rigorously Jenson et al followed moudular spacing.
In this sample, his "h" is wider than "n", and he clearly avoids "fence posting" in the h-i-l sequence.

This topic has been discussed occasionally at Typophile; this thread provides links:

Also Type Spaces by Peter Burnhill adressses the topic.

blokland's picture

Christopher: would it be helpful to illustrate the rhythmic system of Jenson’s types with a photograph from the History of the Book (Amsterdam)?
Christopher, thank you for the link. For the illustrations I used a photograph of Jenson’s edition of Cicero’s Epistolae ad familiares from 1470. The copy is from the collection of Museum Meermanno in The Hague and I ordered this photograph more than 20 years ago when I wrote the book for the Dutch television course Kalligraferen. I have a small number of other photographs of Jenson’s work and of some publications by Aldus Manutius, of which I used a photo for the illustration at the bottom of this post. A higher resolution would be nicer for the quality of the illustrations though, so I will order new high res ones.
Nick: your theory is strongest when it applies to what Alistair Johnston refers to as the “prime rib” characters
Yes, these are what I call the primary Harmonic Models in the Harmonic Systems which we tend to call ‘roman’ and ‘italic’. The secondary Harmonic Models in both systems are derived from the capitals and have a different construction and subsequently other harmonic rules and basically are forced in the rhythm of the dominant, i.e, primary Harmonic Model. Normally this results in a tight setting of the secondary model.
When the roman lowercase letters of the primary Harmonic Model all fit in one base model (i.e., one Proportional System, such as in the ‘Renaissance’ preset in LeMo) like those of Jenson and Griffo, the Rhythmic System is clear and the subsequent ‘fence posting’ not complex. In case more than one Proportional System can be applied, as can be seen in for instance French Renaissance ‘display’ typefaces (like Hendrik van den Keere’s Canon Romain), the Rhythmic System becomes a more complex thing and seems to be dominated by the most condensed Proportional System.
Nick: I'm not sure how rigorously Jenson et al followed moudular spacing [...] [...] Type Spaces by Peter Burnhill adressses the topic.
I am not sure either when it comes to a standardized system for character widths, but most tests I did so far convince me of he fact that the Rhythmic System was used by Jenson. I admit that placing the system on your example (from which book?) is a bit trickier. Although the wider ‘h’ does seem to obstruct the ‘fence posting’, the larger counter results in a smaller distance between the ‘h’ and the ‘i’ and the total distance from the left stem of the ‘h’ to the ‘i’ seems to fit the Rhythmic System. In the second half of the word, the rhythm is clearly more obstructed and I have to do a lot of comparison to find out if this sort of deviations are common or the result of for instance some relatively poor casting.

Peter Burnhill’s book is focused on the Aldine Press and I do not recall him writing on standardized character widths. I will read my copy again, though.
Interestingly it is also no problem to put the the primary Harmonic Model with the ‘Renaissance’ preset on the letters of Griffo used in the Hypnerotomachia Poliphili (see below) to proof the Rhythmic System (although Griffo seems to condens the ‘o’ somewhat more than Jenson).


blokland's picture


The widths of many letters by Jenson and Griffo seem to fit into the Rhythmic System, but some like the ‘a’ and the ‘e’ seem to cause a shift. This can for instance be seen in the ‘Philomene’ example from Nick. These letters take obviously more space than the ‘i’ and the ‘l’, but less than the ‘n’.

If the grid based on the Rhythmic System is refined by doubling the units, the ‘a’ and the ‘e’ seem to fit in the system. The ‘i’ and ‘l’ fit on 2 units then, the ‘a’ and the ‘e’ (and ‘c’) on 3 units, the ‘b,d,h,n,o,p,q,u’ range on 4 units, and the ‘m’ on 6 units.

I mentioned Monotype before, and it is interesting to see that for instance Jan van Krimpen followed a unit arrangement which comes pretty close to the ratio 2 : 3 : 4 : 6 in his drawings for the roman of Haarlemmer. The 'i and ‘l’ were drawn on 6 units, the ‘e’ (and ‘c’) on 8 units (the ‘a’ deviates somewhat more with 9 units), the ‘b,d,h,n,o,p,q,u’ range on 10 units, and the ‘m’ on 15 units. So the ratio for the Haarlemmer widths for the letters in question is 2.4 : 3.2 : 4 : 6.

With adapted parameters in LeMo (Pen width = 10.5, Pen thickness = 1.0, X-height = 50 and Ascenders and Descenders = 30, Stretch factor = 0.92) one can reproduce the Rhythmic System for the drawing of Haarlemmer roman):

William Berkson's picture

I haven't had time to follow this interesting thread, but I would like to register my skepticism that Jenson's spacing is the best model to follow.

Having the inter-letter spacing and counter width the same in the lower case, following Jenson, is also something that Frutiger insisted on, contrary to his teachers, as he has reported. Walter Tracy, in "Letters of Credit," holds that instead the space between two n's should be less than the counter width.

Tracy's view, which reflects Linotype's practice, and I believe the practice of most other houses in hot metal days, I think is the better one. The perfect regularity to me does cause a "picket fence" effect that dazzles only slightly, but enough to detract from readability.

It is notable that none of Frutiger's seriffed faces have been as popular as his great sans faces, Univers, Frutiger, and Avenir. And the sans have lesser space between letters than within counters.

I haven't studied this, or even the photocopies above, but did Griffo and Aldus improve on Jenson's spacing, by tightening up letter spacing a bit?

Also, there is no doubt that letters which make readable words are *approximately* rhythmic. But the slight deviation from perfect regularity may be as important as the approximate regularity.

dezcom's picture

Never underestimate the power of the human eye to determine what is best for the human eye.

John Hudson's picture

The relative spacing of i and l, in a serif type, is interesting. I almost never assign these the same advance width, because the presence of the upper serif at the x-height optically closes the space between the i and a preceding letter in a way that does not happen with the l. One can see this quite clearly in the Aldine ‘nobilissimo’ image, above. I almost always end up putting less distance on the left side of the l than on the i, and still less on the left side of the b because of the absence of a foot serif and, in many styles, an under-turn.

kentlew's picture

@John Hudson: Indeed. I always assumed this was standard practice.

blokland's picture

William: I haven't had time to follow this interesting thread, but I would like to register my skepticism that Jenson's spacing is the best model to follow.
This thread is on glyph space and above I show spacing based on what I call the Rhythmic System, which will be built-in in LeMo, and I try to explain how for instance Jenson, Griffo and Van Krimpen all seem to make use of such a system. So far I did not make any quality statement nor any advocating of ‘Jenson’s spacing’.
Chris: Never underestimate the power of the human eye to determine what is best for the human eye.
What you and I see is mostly the result of conditioning based on conventions which are preserved by conditioning.
John: I almost never assign these the same advance width, because the presence of the upper serif [...]
I would not be surprised if Jenson or Griffo were aware of this fact, but that for instance the material and the restricted use on small point sizes made it practically impossible but also unnecessary to apply such details. On the other hand, there is also a good design reason for not making the spacing of the ‘i’ and ‘l’ identical and that is that making the spacing between the letters even, will give a slight disruption of ‘stem interval’. So, the rhythm of the stems versus the rhythm of the (amount) of space.

dezcom's picture

"What you and I see is mostly the result of conditioning based on conventions which are preserved by conditioning."

What I am referring to is one of your samples posted above, where the "a" does not quite fit the grid. Clearly there was an optical adjustment needed to bring it into visual balance. I think something primal and primitive to the earliest humans surviving their surroundings is operating here. Humans (and surely other animals) can spot minute things differing from the perceived norm and used that skill to know prey from predator.

quadibloc's picture

This is an interesting question, but it should be possible to settle it fairly quickly from observation. One can compare the widths of iiiiii and nnn and mm in various common unit systems to see if the "fence-post" values are the current ideal.

Thus, if we go to the absolute minimum, and ask what widths were given to letters on an IBM Executive typewriter, we find i=2 units, n=3 units, m=5 units. So we have 12, 9, and 10 units for the three cases, which differ - but not in a uniform way.

On an IBM Selectric Composer, which used a finer unit system, one that closely resembled the appearance of real type, we find i=3 units, n=6 units, m=9 units. Here, we have 18, 18, and 18 units. So this is completely consistent with fenceposting. (The internal spacing of strokes within the n and m, of course, could still be such as to avoid that.)

With Monotype Times Roman, we have i=5 units, n=10 units, and m=15 units. Again, consistent with fenceposting. But c, e, and z are 8 units, a, g, v, x, and y are 9 units - instead of all being 10 units, as would be required to maintain the rythym for all the letters.

William Berkson's picture

quadribloc, the issue is not just advance widths, but also width of counters. Currently, in the Times New Roman on my Mac, the sidebearings of the n is 315 units, and the breadth of the counter in the n is 379 units. The sidebearings are less than the counter, as Tracy recommends. Also the i is in fact more than half the width of the n.

dezcom's picture

What about the distance from stem to sidebearing to accommodate the serfs, Bill? You have to visually allow for what the serifs do for your equation. With a sans, it is easier to do the math.

William Berkson's picture

Chris, the measurements I gave were from the stem to the sidebearing. Sorry I didn't make that clear.

hrant's picture

{To Read}

enne_son's picture

{To Read}

Yes indeed! I missed this too when it happened.

Did the discussion go on elsewhere? Did Frank complete his work on this?

A theoretical basis for Frank's approach might be the spatial frequency tuning of perception in reading. It seems to me that the rhythmic system problem can be addressed with fourier transforms. These show narrow phase alignment in the stems. To keep a spacing system from implementing perfect regularity (the origin of the ‘picket fence’ effect = absolute phase alignment) there would need to be a modulatory system based on establishing the rhythmic system in the whites, or so it seems to me. I's possible that the other parts of the fourier transforms than the part that shows the phase alignment of the stems could provide a handle for that. However, building fourier transforms math into typsetting systems is probably impractical.

John Hudson's bubbles concept might be linkable to Denis Pelli's work on isolation fields, a concept which arises from work on crowding or lateral interference.

eliason's picture

As long as we're reviving the thread...
Is that Harlemmer image introducing an orange into the apple discussion? As I read it, that gridded drawing seems to represent (for the most part) widths from edge to edge of the glyph--i.e. ignoring sidebearings. What then is the relevance of overlaying the running arches over it?

John Hudson's picture

[Hrant, since you're digging up interesting old threads: I thought you'd be all over this one. My last post there opens the door to a discussion I've been meaning to have with you for a while now.]

hrant's picture

I'm on it. Like a freezer making ice-cubes. :-)


Syndicate content Syndicate content