Alternate numerals in script fonts

Hey all, just an observation, I was wondering if anyone had some insight.

It seems to me that fonts advertised as having lots of alternate glyphs (usually script fonts) only come with one set of numerals, so if I am setting some text with the number 484 both 4's would be identical.

Is my observation correct? If yes, is there a common reason?


  • The user and all related content has been deleted.
  • This is a moot point since I'm pretty sure it's a rule that you have to write numbers as words on wedding invitations.

    Anyway, good question. Are there any script type drawers here? Some script designs use letter (and letter pair) frequency to target the most important characters to create alternates for. In theory, number frequency is random, making it impossible to prioritize or target key characters. I'm curious what the actual frequency of numbers and number pairs is IRL though, has anyone crunched that?
  • Nick ShinnNick Shinn Posts: 1,712
    I’ve implemented alternate sets of figures in a couple of “pseudo random” script typefaces: Fontesque Pro (one alternate) and Duffy Script (three alternates).image
  • James M - I'm referring to fonts that mimic some hand-created process, like handwriting or sign writing.
  • The user and all related content has been deleted.
  • Studio Lettering Swing has alternate figures, but they are intended for different languages.
  • The only overarching issue on this is that script numbers have to stand up legibly outside of any context, while letter alternatives in scripts are contextual enhancements, supposedly to form more realistic words from a given hand. What you'd expect from most hands is consistent figures.

    That leaves random alternates, words, roman numbers, OS figs and other vertical options, and diagonals, as well as hybrids. Sign painting shows lots of this.

  • Good point. but it would be nice to have some alts, even if i have to manually select them from the glyphs panel :-)
  • Numeral, exclamation mark and question mark alternates are important. You rarely see three of the same letter in a row but with numerals, question marks and exclamation points, it happens all the time.
  • Nick ShinnNick Shinn Posts: 1,712
    Good point Ray. Continuing this line of thought, it’s unfortunate that double quotes are constructed from single quotes, when “quotes within quotes” produces such a nasty combination. Perhaps it would make more sense to make single and double quotes somewhat different, to mitigate this (admittedly rare) situation.

    And of course, as “smart quotes” have demonstrated, it’s dysfunctional to have single right quote and apostrophe sharing the same Unicode point.
  • In font where I'm going for a hand lettered look, I try to point the quotes together a bit and lower the quote on the outside. That and a little positive kerning prevents a triple quote abomination. Example: Chickweed

    I think I got the idea for positive kerning quotes from you, Nick. I've been doing it to all my fonts for years. I even went back and posikerned quotes in my old fonts. I don't add a lot of space; just enough so readers can differentiate the single from the double quote - break up the cluster.
  • Nick ShinnNick Shinn Posts: 1,712
    Yeah, I’ve always liked those symmetrical (“American”) double quotes that are thin lines, like there used to be in old movie title card and comic book display lettering. ATF Garamond had “backwards” left quotes too, and so did Verdana at first—but of course they are no good for German.

    Perhaps one could make them language-specific in the feature.

    To bring this discussion back to alternate figures; the coding I did for FF Fontesque made do with only one alternate, on the principle that there are almost no words in which a letter repeats thrice consecutively (a few oddities in German, IIRC). However, that doesn’t apply to figures, you really need at least four sets. But that means you have to have four of every character for the random effect to work…
  • But that means you have to have four of every character for the random effect to work…
    I believe it’s possible to write a carousel rotation without having equal sets of everything. I wrote a proof of concept for one of Richard Lipton’s scripts — five variants of ascenders and four variants of descenders only — but we didn’t end up pursuing it (for other reasons).

    I’d have to look back through my notes to remember exactly how I did it.
  • It's possible to do complicated rotations with different numbers of alternates for every glyph. But, obviously, it get very complicated very quickly.
  • When I make 3 variations, I can type a string of repeated glyphs and the pattern isn't obvious until I get about 7 or 8 in a row. Then you can clearly see a repeating "wave". I'm not a font scientist but I just found, from experience that 3 is the key number of variations for most situations.

    I find, with fonts that are supposed to look hand drawn, seeing a sequence of identical glyphs is visually jarring. If you just want to make hand drawn letters look natural, then 3 variations is the key.

    With 2 glyph variations, the pattern is slightly noticeable in a sequence of 3 repeating glyphs and a bit obvious in a sequence of 4. If you have 3 variations, you can go with a longer set before the pattern jumps out at the reader.

    Even if you have 5 glyphs in a row, the reader probably won't pick up on the pattern unless they're looking for it.

    In other words:

    12121=a bit obvious. An easily detectable "wave" appears.
    12312=probably good enough. The sequence doesn't repeat in it's entirety and it doesn't jump out at the reader.
    12341=slightly better than good enough.

    If someone's really looking for the pattern, they can see it but, to me, what matters is that the pattern isn't visually distracting to the reader. I can't see a justification for the 4th variation unless it's an effect when the repeats are really, really obvious - like maybe a ransom note font. I doubt that there are enough situations with more than 5 identical glyphs in a row to justify the extra work and font bloat. I'm not setting out to fool forensics - just to make a font more visually pleasing.
  • The user and all related content has been deleted.
  • James — The basic concept follows this form:
    feature calt {
    	sub @default @default' by @calt_1;
    	sub @calt_1 @default' by @calt_2;
    	sub @calt_2 @default' by @calt_3;
    } calt;
    where you have four variations distributed into four parallel classes.

    Here is a nice article about a slightly more sophisticated application of the principle:

  • The "ignore sub" lookup is invaluable here.
  • Yes, I was just giving James a basic introduction to the concept. The Font Feed article I linked goes into Martin Wenzel’s 2-part cycle, similar to that described in your Glyphs tutorial.
  • I'm impressed by Martin Wenzel's method in Duper. I took a different approach when 'randomising' the form of decorations inserted in the {ss07} and {ss08} styles of Gabriola. Instead of cycling, I use pairs of randomly classed glyphs* -- sometimes at a remove from the decoration insertion point -- to contextually trigger the form, and then a secondary cycling lookup to ensure that the same decoration doesn't occur twice in sequence (the latter made easier by the fact that the decorations are categorised as marks in the GDEF table, meaning that in the cycling lookup I can opt not to process any non-mark glyphs). I think the effect of this feels more genuinely random for the user, because the form of decoration is being affected by things that are not adjacent to it, even by things that are across word boundaries (results will vary in software, especially broswers, which may not process word spaces as glyphs).

    * Creating these classes involved actually randomising a list of context glyph names, then cutting the list into blocks and creating a number of classes with names like x.random1, x.random2, etc., which are then used in GSUB context statements.
  • Nick ShinnNick Shinn Posts: 1,712
    edited September 2013
    I think 2 sets is OK for text, using a two-part code.

    The first part creates the “121212…” toggle sequence, and the second part deals with proximity situations where both “1”s are the same character.

    With this system, (barring three-in-a-row characters such as figures) glyphs don’t repeat with fewer than three other glyphs between them.

    I calculate that puts enough distance between the glyphs—in terms of viewing arc degrees—to remove the duplication from a single fixation in the fovea, making it imperceptible to immersive reading.

    The technique is partly explained in this article, in the section about Fontesque:

  • "...making it imperceptible to immersive reading"

  • I would also suggest these techniques and ideas could apply to rough/grunge/distressed/worn styled fonts too, where 'in the real world' these effects happen to each glyph uniquely.
  • Nick ShinnNick Shinn Posts: 1,712
    David, I reason that the eye/brain only gets enough information from each “grab” (between saccades) as is necessary to identify characters/words. Subtle differences in execution of individual characters would not therefore be perceived if they occur in separate grabs.

    As I understand the workings of the brain, different cells identify different optical qualitites. The research of Hubel & Wiesel: “By measuring the electrical impulses of cells in the visual cortex, the scientists discovered that cells respond to straight lines, movement and contrast – features that delineate objects in the environment. They further found that some cells fire rapidly in response to horizontal lines, while other cells prefer vertical lines or angles. Cells with similar functions are organized into columns, they said, tiny computational machines that relay information to a higher region of the brain, where an image is formed.”

    Accordingly, I would expect that the quality of identicality would register as a particular combination of cell firings, if glyphs are in the same grab. But I doubt that small differences in character renderings, or lack therof, would make it to that “higher region of the brain, where an image is formed”, if they occur across a saccade.

    Having said that, I would like to believe that vision is a more profound and mysterious phenomenon than this mere decoding process which scientists have attached to reading, and that there is some kind of mental ’bot which searches for the similarites that create large-scale patterns, as indicators of broader categoric significance, such as cultural style.

    The fact that there is so much redundancy in the most successful “book” faces, in which not only do serifs vary from character to character, but, when printed by letterpress, individual glyphs of the same character vary too, suggests that identicality of letter features may in fact by problematic—perhaps because our “similarity-detector” processing is unnecessarily over-stimulated.

  • I reason that the eye/brain only gets enough information from each “grab” (between saccades) as is necessary to identify characters/words. Subtle differences in execution of individual characters would not therefore be perceived if they occur in separate grabs.
    Alternatively, even if such differences are perceptible in the fixation, we might filter them out in cognition, since they don't contribute to word recognition or comprehension. Of course, this presumes that the differences are indeed 'subtle'. If they cause errors in character recognition, or bump spatial frequency channels, they're going to be impediments to reading.

    I've read a fair amount of text in 18th Century and earlier editions, in which the printing has been such that no two instances of a letter on the page are visually identical, and sometimes quite radically different as a result of worn sorts, uneven inking, variable thickness paper, not to mention actual variant sorts (some clearly designed for specific contexts, but not always reliably used as intended). While reading, I am not conscious of these things, nor am I conscious of what effect, if any, such inconsistency might be having on my reading in terms of speed, number of regressions, etc.. But then when reading I'm not conscious of much except the meaning of the text. That's rather the point, I think.
  • Eris AlarEris Alar Posts: 231
    Does this relate to context? As a graphic designer I am unlikely to set a large portion of text in a script font, especially one with lots of personality. Whereas I would probably use it for short bits of text (like headings or titles or logos), and in this context the repeated letterforms become a lore more obvious.
  • Context. Yes. I have a pseudo randomization scheme for advertising faces that only subs repeated instances of individual letters up to ten letters deep. It works well for short headlines and the benefit is that it allows for a "preferred form' for each character. But it doesn't work so well for a font intended to set longer lines of text because repeating pairs of preferred letters begin to create a noticeable pattern. In those cases, I have a slightly more complicated scheme that rotates more randomly (kind of like the above examples).

    H I P P O P O T A M U S S M U T
    1 1 1 2 1 3 2 1 1 1 1 1 2 2 2 2
  • Deleted AccountDeleted Account Posts: 739
    edited October 2013
    "As a graphic designer" you should choose your Typography and then find the font(s) that make it happen on the devices you are targeting. If you choose Random-Looking Typography, and your device can handle it, you'll need to match the font's arbitraryations to your text length(s), at least, if not the arbitraryations required for the character/glyph interchange rate required per word. If you're choosing a Script Typography, the overarching issue is written in my first post above, and it wasn't an opinion.

    As a type designer, you should know that hooking up a glyph randomizer, to a universal saccade predictor, (as if that could work on everyone!?), for figures, is about as likely/useful, as this being the last post ever. In general, I think this is related to one of the basic problems of OS/UA/OT: that the user is interested in Typography and when that Typography involves non-registered OT features, who are not related to the Typography by technical or nominal hook, the work you've left the user to figure out is much more than a bad thing. Leaving Typography out of font software for software to figure out, we know how that works out.
Sign In or Register to comment.