Nick, I could run that control test, yes - and when my overdraft is less booming I'd like to, but what I expect to see is that the averaging gives inkpress radiating to equal distances from the metal, except when inkpress is forced against ink coming from another direction, when the inkstreams would combine in an averaged direction and possibly travel further: ie, gumming up the crannies more than proportionally.

Outline filtering: it would have to be postscript then? Beowulf no longer works on Mac OS X, I hear.

> it is the variation itself that contributes to the look of the
> letterpress Aldine page, so averaging the variation to produce
> single letterforms seems to me not to be a very good way to
> produce a ‘faithful revival’ of the type

The problem with this is that the variation can't come from the
type anyway. It has to come from the printing process (which can
be attempted here via photopolymer letterpress, although it's no
piece of cake when you throw in issues of paper, etc.) and this
makes the arrival at the "source shape" the only way, really.

But of course there could always be more
than one source shape per character.

> what you are averaging is not letterforms per
> se but printing artifacts affecting letterforms.

Which is exactly how you get at the letterforms! :-)
That is in fact the beauty/power of averaging:
it removes some of the arbitrary interpretation.

> one knows for certain that this feature
> of the letter is Griffo’s design.

Not really - it could be an artefact too! :-)
Our minds can play tricks on us, and averaging can help tone that down.

So there are no "true" Griffo forms in the results; they are ALL
imperfect instances; and the more instances you average the
more you remove the arbitrariness.

(I remember once having a similar discussion with you concerning
what pairs to kern; somebody advocated using Google to form lists
of pairs, but you pointed out that that would include pairs from
incorrectly spelled words; at which point I put forth that there
really is no "wrong" or "right": if the pair happens, it means
it exists, and that's all a type designer needs to know.)

> the averaging methodology has merit relative to the careful selection of input.

So this becomes exactly what's anathema to the whole concept!
If you apply selectivity, you're introducing an arbitrary
judgement of what's "correct", which can only be one's own
idea of what's correct, not Griffo's. So instead of selectiveness
of input, what you really need is simply more input.

You exclude nothing, because you just don't know. You can
only feel, and that has little place in this type of effort.

> averaging gives inkpress radiating to equal distances from the metal

Note however that ink spread can be asymmetrical, and for more reasons than one. One reason is the paper: it can hold/spread ink differently in each dimension; see this example from a low-grade Armenian newspaper:

http://www.themicrofoundry.com/other/armnews2.gif

Another issue is slur: if you might end up using a cylinder press (something Aldus of course didn't use) to reproduce/test these forms, the looser your makeready the more your gain will be greater in the vertical direction than the horizontal. This is especially if you apply extra pressure (see below). In the end, every little thing affects the results; maybe even the use of ink rollers instead of what they used then: ink balls.

BTW, something I've been meaning to point out:
One serious problem with the method of using a magnifier to see the "3D"* effect (that a scanner can't show**) is that the more impression pressure a piece has the less faithful it is to the type! Which is sort of self-defeating. Quite often in fact (Fleischmann comes to mind) a designer's intentions seem to have been squashed -quite literally- by bad presswork, leaving us with an even harder job of extracting intent.

* Actually, 2.5D, in CAD-speak.

** Except of course if you use some fancy
image-processing and 3D glasses. :-)

hhp

The problem with this is that the variation can’t come from the
type anyway.

I'm not actually convinced of that. I need to go and read Peter Burnhill's book again (George, do you have this? it is invaluable for understanding the Aldine types).

But of course there could always be more than one source shape per character.

Yes. We know this was true for Gutenberg's fonts, and when I see apparent differences in the proportions of letters, e.g. one p looking narrower than the others, I really wonder how much variation is actually explainable in terms of inking and impression.

I think when looking closely at specimens of printed type, you have to make reasonable assumptions based on the competency of the punchcutter. The more a specimen looks like a complete letterform, like a well balanced and resolved design, the better a specimen it is. You have to assume this if you consider the punchcutter to be good at his job, and you can most often confirm this by looking at his larger sizes of type. If these look well designed, you can be reasonably sure that his smaller sizes were also. So if you are looking at fragmentary specimens in which the design seems weak and unbalanced, and you compare these to specimens in which the shape is clearly defined, then you have to consider that the latter show forth the design better. Yes, this requires a judgement call, but I don't think it is arbitrary: it is reasonable. Also, it is often easy to tell what are the prestige publications of a particular printer, those that have been printed with more care and attention, and these will provide a better quality of specimen than cheaper editions produced with less care.

I don't think pre-selecting specimens is anathema to the averaging concept: it is a methodological modification. In fact, a useful exercise would be to perform averaging of both pre-selected and of randomly selected specimens, and compare the results of these. You might also have different people perform pre-selections, and combine the resulting set of forms. There are lots of ways in which one might play with the methodology. I think it is too soon in the day to say that it has to be done in one way only.

I agree with you that one should not presume equal inkspread, even on a flat bed press. Letterpress impression displaces paper mass, and since paper is not of perfectly uniformly density, the displacement from one letter may be more than from an adjacent letter. Also, type was being inked using balls, not rollers, and even inking was virtually impossible. So one part of a letter might have more ink on it than another.

(I remember once having a similar discussion with you concerning
what pairs to kern; somebody advocated using Google to form lists
of pairs, but you pointed out that that would include pairs from
incorrectly spelled words; at which point I put forth that there
really is no “wrong” or “right”: if the pair happens, it means
it exists, and that’s all a type designer needs to know.)

I think that must have been someone else, because my approach to kerning involves using test words that, for Latin at least, include examples of every possible combination. Including Exxon :)

[John] "Yes, but supress them we must..."

Exactly.

... surrendering to associations of rhythm with sequence or flow is to accept a kind of phenomenological blindness.

Nice.

[George] "Are you recommending in practical terms..."

Yes, since even the most mechanical of methods interprets (contra Hrant?), you are looking for the place where the various methods converge. In the best of all possible worlds the methods will crosstalk as they proceed.

But the possibility exists that, because the glyphs are cut, the tuning of contrast is at least as much feature-awareness-based, as chiro-referential (or deliberately chiro-paraphrastic). It isn't out of the questio--probably it's likely--that Griffo engages in a more or less deliberate or overt process of 'geometrical-feature' evaluation of or play with inner and outer form. Punch cutting seems to me to encourage a more feature-analytic relation to form than the writer's learned repitition of ductal or gestural routines.

How can it be determined whether character specimens are from the same mould?

1. Statistically, by analysis of a large sample.
2. By eye

In averaging, how do you allow for the fact that fragile features are more likely to be damaged? If you want to produce a full-featured, good-looking final result, surely you would have to exclude damaged specimens from your sample selection, on a "by eye" subjective basis?

***

Should there be a constraint that specimens are always culled from the centre of a page, to minimize prining distortions?

>> The problem with this is that the variation
>> can’t come from the type anyway.

> I’m not actually convinced of that.

OK, let's think about this a little bit more deeply and with open minds.

If you want to make offset output that comes as close as possible to this particular source (ignoring for the sake of argument the factor of impression depth*) you would have to incorporate the irregularity into the letterforms/type. The ways I can think of doing this are: 1) Have a base glyph stucture and apply pseudo-random irregularity to it (render-time), in a way that mimics the original. This is not possible in current generally available font technology. 2) Choose a finite number of instances of the source glyphs and deploy them pseudo-randomly. This is doable with OpenType. 3) Create a finite number of variants pseudo-randomly (as in #1) and deploy those pseudo-randomly using OT.

* Which however you've said -rightly- is an integral part of the way Aldine typography looks/feels.

Well, #1 and #3 require averaging. And #3 has the advantage over #2 that it removes a selection process that cannot be informed by the true intentions of the original designer. Basically, #2 is the furthest from replicating the original.

The thing is, impression depth is very important here. If you want to get really close to the original, letterpress variance takes care of the irregularity inherently and "naturally". At which point the type doesn't need to exhbit irregularity (which is a good thing, because technically it can't) and that makes averaging useful here as well.

Multiple glyphs per character in the original:
That's different than irregularity. It's certainly useful to be able to group various instances of a letter as different original designs, but here again there's no reason to introduce the reviver's preconceptions (in a literal revival). What you would do instead is algorithmically decide on an instance-by-instance basis which variant of a character an instance is, and then average them separately. You might do this by determining thresholds within which an instance is a certain glyph and outside which it's another; and these thresholds might be determined by some iteration with models for which perhaps we do know the "punch history". Yes, it's error prone, but at least it's a "neutral" error, one that becomes absorbed if you process enough specimens (unlike personal interpretation, which just gets amplified).

> Yes, this requires a judgement call, but I don’t
> think it is arbitrary: it is reasonable.

Maybe arbitrary was a bad word. My point was that -no matter how
good or bad the punchcutter was- there's no safe assumption about
what he intended - not in a literal revival.

One could say that a literal revival cannot benefit
from intent; instead it spawns directly from matter.

> it is often easy to tell what are the prestige publications
> of a particular printer, those that have been printed with
> more care and attention, and these will provide a better
> quality of specimen than cheaper editions produced with
> less care.

But the punchcutter is almost never the printer,
so you can't know if he agreed with the results,
or he intended them (assuming he even cared to get
involved in the results, which is not the same thing
as not having any intent).

This business of extracting intent really seems
to make a horde of romantic assumptions that to
me are counter to the intent of design, at least
in the case of a literal revival.

> I think it is too soon in the day to say that it has to be done in one way only.

Like has been repeated, averaging is a tool. And yes, tools can be used differently depending on what the desires results are. I think in the case of a literal revival like this one, interpretation should be actively limited (not the same thing as saying it can be completely avoided). And in any case averaging shouldn't be shunned - that's what mainly motivates my own participation in this thread actually.

> type was being inked using balls, not rollers,
> and even inking was virtually impossible.

Not only that, but the necks of the sorts got more ink
than they do with rollers, and this affects ink gain.

> I think that must have been someone else

Well, Typophile's search is really cruddy now, so I can't find it.
But I wasn't thinking you'd use Google that way yourself anyway.
I just remember that as your reaction to the idea of doing so.

hhp

> In averaging, how do you allow for the fact that
> fragile features are more likely to be damaged?

You could certainly do it algorithmically.

> surely you would have to exclude damaged
> specimens from your sample selection

Why? They form part of the original!

> Should there be a constraint that specimens are always culled
> from the centre of a page, to minimize prining distortions?

Do people only see the center of the page?

Do you see what I've been getting at? In a literal revival,
intent is moot. In an interpretive revival though you would
indeed use the tool that is averaging differently, including
things like refining the source.

--

I guess the primary question really is:
What is the original? Exactly what are you reviving?
Why are you doing what you are doing? What is YOUR intent?

hhp

Do you see what I’ve been getting at?

Yes, but I thought George already decided that issue. He said that he is doing an interpretation that will be a "good" version of Griffo's type (or words to that effect), so my comments are to do with him clarifying what can be decided by eye, and what automated.

> a “good” version of Griffo’s type

That "good" contains a world though...

BTW, I think you're using "automated" too broadly.
Averaging isn't really automation, it's analysis.

hhp

it can hold/spread ink differently in each dimension
One of the few things about 15th C printing which makes my life easier is that hand-made paper will on average receive ink equally in all directions, because the fibres lie randomly.

Essentially I agree warmly with Hrant that more input is better than selected input, with the qualification that I would recognise some obvious classes of impressions as plain misprintings, like the initial on verso and final on recto letters of lines. These I have left out. One other class worth excluding is letters the variance of which from the norm built up in the first round of averaging is bizarrely high: I think this probably implies a different punch (only one case from the 2 sides and 61 examples which were picked out to produce the averaged image you've seen). Punch variation is rare and inconsistent in Aldine printing so, contra Mardersteig, I doubt that it was intentional. It'd be hard to justify on any grounds the Lascaris interlopers among the 1499 caps in the Hypnerotomachia, for instance: they're fat lumps compared with the new letters; still more obviously wrong is the appearance of Roman caps among Greek, and vice versa. Punch variation is in any case much easier to spot with software that can quantify variation than by eye; and because the distribution of extent of variation is bell-curve normal there really aren't any samples which could plausibly be considered exemplary. But paper shrink is a problem which might make taking samples only from some arbitrarily limited central region of the page, where shrinking is least, a good thing.

I have to say that I think Hrant's straw-man method #2, the choice of a finite number of instances of source glyphs, pseudo-randomly deployed, would produce (and is) just embarrassing junk.

I have Burnhill's book, John; it's main relevance to my project is in making me doubt the propriety of altering vertical proportions (for the original roman, o is 5 units tall, H 7, h 8, y 8, with 1 unit of interlinear space) through MM. Note that o is as usual slightly taller than x, such that one cannot quite accurately say that x-height is five sevenths of cap height. Can these proportions work at 10 point, I wonder? As for deciding by eye what impressions look good: the software numbers show clearly that the eye is no judge, because of the misleading attractiveness of errors in successful combination.

Peter, once one makes one's reduction of Griffo's extant work to principles of construction sophisticated enough to incorporate "feature-awareness-based" analysis, one is giving oneself too much room for interpretive manoeuvre. The method ceases to be a method at the very moment it becomes justifiable.

George, it sounds like you have thought this all through very thoroughly. I look forward to seeing the results, which is when I think I may have more to say on the method.

Thanks John - I can't wait to get started in October, assuming I get on the course. I have absolutely no training in any related field, and a year ago didn't know what an ascender was, but the pull of this sinister spiral is very strong. Hrant asks "Why are you doing what you are doing?", and the answer is that I think that there's powerful magic in this roman of which Garamond only caught a whisper in his own Griffo revivals, and which was never found again before Baskerville called the search off; I want to isolate it at original strength. The Soncino italic is just exceptionally pretty icing.

One last picture:
On the left is the output from software partially rewritten to reduce the impact of normalising colour while maintaining accuracy in superposition. On the right is an automatically produced image in which thickening of the outline signifies blunting/breakage to the type.

Edited to make this clearer: this outline is nothing to do with the eventual drawn outline for the glyph.

Your outline seems to wrap to the outer egde of the grey zone around the letter, and this seems to produce a blobbier form than I perceive when I look at the composite image.

I've been watching this thread for a while now and I can't imagine how one could expect to faithfully reproduce a typeface from a letterpress printed specimen.

There are a legion of problems. First of all it is difficult to find an Aldine specimen that hasn't been cleaned over the years. Cleaning takes out the impression. While this may resolve a certain discrepancy caused by the camera interpreting slope as 2-D it allows problems such as ink spread to be misinterpreted.

Wooden common presses were not a manufactured item and differed greatly one to another. And any publisher with the printing capacity of Aldus would likely have had many, many presses working on different sections of the same book.

There are other problems, variance in the paper, not only sheet to sheet but within the sheet causing characters to print lean or fat without reference to inking. Dampening the sheet (required with handmade papers) also introduces variance in the way the letterforms will appear due to the way the ink is accepted by any particular sheet (there is no way to make dampening standardized sheet by sheet).

There are inking problems not only as the result of ink spreading beyond the letterform (due to problems in proper visconsity and application) but in the technique the pressman used in applying the ink. As someone mentioned previously ink balls were used rather than rollers (first use of rollers was 1812). There is a significant difference (not better nor worse) in the way letterforms respond to these differing tools.

Variance in impression, due to makeready or wear on type or problems in composition or lock-up will also effect the look of the letterform.

And there is often the likelihood, in earlier printed work prior to standardization, that multiple punches were used to create the "letters."

The search for some form of standardization is wrong simply because there is no initial form of standardization. There is only variance, and to some extent, particularly with the early printers, that was the whole point of it, whether there was a choice in the matter or no.

The only true way to capture a letterform from the past would be a carbon proof from the punch itself or find an exacting way of capturing the letterform directly from the punch.

Once it goes to print, all bets are off. While I can applaud the interest and research I cannot imagine that any result could be satisfactory. If the intent is to be "faithful."

Gerald
The Bieler Press
http://BielerPress.blogspot.com

Hi John, yes, the 'outline' in the picture above is nothing to do with the outline of the glyph as it should be drawn in fontlab. It just colours in a narrow band of grey-values from the averaged master to show up the presence of large minorities of samples with unusually strong inking at various points - most notably beyond the spur and serif, but to a lesser extent above and below the left edge of the descender's bulge. The slight thickening on the exterior outline where the bowl meets the upper stem is just inkpress increasing as it is forced against itself coming from two directions.

I’ve been watching this thread for a while now
But apparently not reading it.

All the variance will be averaged, and to some extent averaged out. Quoting myself, I want to get at the forms Griffo can most reasonably be conjectured to have predicted (not necessarily to have wanted - that's unknowable). It's true that I can't be certain how much inkpress I'm incorporating, though from the smoothing in crannies I can make a pretty good guess.

The search for some form of standardization is wrong simply because there is no initial form of standardization. There is only variance, and to some extent, particularly with the early printers, that was the whole point of it, whether there was a choice in the matter or no.
Can you show me a single record of any early printer or reader admiring the variety in impressions from a single apparent punch? They boasted rather of the printing's fidelity to the coherence of the pen-made letter. Variety in printing is quite unlike pen-made variety.

The only true way to capture a letterform from the past would be a carbon proof from the punch itself or find an exacting way of capturing the letterform directly from the punch.
If you know where Griffo's punches are I'd be delighted to use them.

"I take an outline derived with consideration for inkpress in corners and blunting at spurs from an averaged image like the one two posts up, write a fixed-front ductus over it, see where the ductus doesn’t match and how the metal departs from it, assess the consistency of those modifications, and of second-order variation in manipulation (going on potentially ad infinitum but really it wouldn’t be, not with humans),"...

(this is the type speaking) Help me, Help, me. please, , , they are talking my headspace off.

My experience with this, is that the best way to revive from metal, is to do the first part of what you've done well. Examine large, a wide variety of letters. But when I do this, I've already got a list of things that go wrong, from the punch, to the paper, and from then to now (if it was a long time on the paper). So, I have a growing "typographic dictionary of errors", like "Dumb Struck" (a pack of letters with a distortion as if a punch head were too far from the user for the position of the punch face at time of strike. Or some of my other favorites like "Cat Walked", "Dropsies" or "Secret Filing" ('twixt chase and slug). My modern favorite, "Full Page Variouching", occurs, when faxing a full newspaper page to plate, burping, sometimes repeatedly, causes letters to change width and weight (variations at last!) in a typographically painful narrow strip from head to foot of an entire page, (see NY Times, (only if outside metro area???)).

Fortunately, the subset of errors in the metal of your project, limits the possibilities to the point where you can match errors with examples, and actually, (nearly factually) "find" the "right" letter but getting your eye on a particular, or a few particular letters (if you suspect intelligent design in the composition), and not use a "program" to "assume" what an "error" is. Metal type, after all, was not VooDoo. More often than not, as cited above and sited at the time of the metal men, by the metal men, it was sometime DooDoo. You just need to learn to track type scat.

"and then redraw the font appropriately"
i.e. find" the "right" letter with your mind (ooh). It is especially effective to do it this way, because, any program trying to make decisions on these kinds of letterforms, from this period, without the Error list of type, and without intelligent error thinking, like you have, on an intra-letter basis, will...I dunno, Not make an alphabet?

Thanks David; to clarify, the method you quote is not the one I have or will adopt, but my translation of Peter's thinking on revival into a practical process. As it happened, Peter rightly thought that this process didn't take into account the feature- rather than ductus-based thinking which presumably goes into punchcutting, and I responded that no method could be formulated which could do so.

But more importantly your “typographic dictionary of errors” sounds fascinating and very important to my project. Could you tell me more, do you think? I'd be perfectly happy to increase the limited pre-averaging selection I've already done to take out the scat. On the other hand, averaging of the scatless forms will stay. There is almost certainly no single “right” printed letter (and even if there were one couldn't know which it was) because, as I've said, the extent of variation is, in the technical sense, normally distributed: it follows a bell curve, such that zero or minimal variation is actually much, much rarer than substantial variation. When drawing a monochrome glyph based on the soft and lovely averaged form, my mind, for all my protests, will have plenty of use.

>avoiding the potentially banalising effects of human interpretation

I think you are coming around to see that avoiding human interpretation was a misconceived goal. Admitting that, then I think the really important questions become clear: What are you going to use the revival for? What is the revived font going to be used for by other people?

Now you are back in the boat with other revivers of Griffo, perhaps armed with some additional tools. But it is your decisions, talent and skill, and not any "objective" program that will make the difference as to what this looks like, and what it is good for.

I made, after reading Nick's and Raph's first posts, two changes to my position. First, having been undecided, I accepted that autotracing of the averaged form is unhelpful because, in preserving equally emolliating and merely emboldening inkpress, it misprepresents the perceived bite of the original. Second, I accepted that the averaged master image contains information which should be preserved and analysed in making decisions about the drawn glyph - information about likely features of the punch which are misrepresented in most impressions, because cast type degrades in predictable ways. Finally, I was culpably unclear in phrasing my initial post. Avoiding those acts of interpretation which are potentially banalising remains a goal: I dislike the fatalism surrounding interpretation (of which I don't accuse you): the repetition of truisms like "you've got to interpret", or, with more relativist conviction, that one's work is "just your interpretation", because the usually unstated assumption is that all interpretations are alike in the merit of the hypotheses they imply. But this is untrue, doubly so with newly available tools providing new kinds of evidence, but always so with the application of knowledge of the printing process (on which I'd love more detail, or recommendations of books on Aldine or incunable printing). What produces banal results is the kind of thinking that goes: Griffo was presumably a sensible man, and should have been if he wasn't; and sensible men know that typefaces are best when they're like this, or when they are consistent in this. As Justin Howes wrote, "In our own time, the result of redrawing is, unavoidably, type design which edges ever closer to this period’s dominant letter form, Times New Roman". He tried to resist that with a method which was clearly, as he knew perfectly well, unconvincing in one respect (its variation is frozen), but nonetheless he succeeded better than almost any reviver in providing as fully as possible, given current limitations, what his model had provided to an ealier generation. I'll take a different path in pursuit of the same ideal. And I shan't mind if it's never used at all; I'd be perfectly happy if it were only something for me, and I hope others, to study. As I've said, though, a secondary purpose for it is in fine book work, for which sized variants will be provided.

For now this stands out:
> I think you are coming around to see that avoiding
> human interpretation was a misconceived goal

Why so absolute? Design is pragmatic, and a pragmatic
approach is to TAME human interpretation (where it
needs that); not wallow in it, and not shun it.

Also: I don't think it's a goal, just a means.

(More on the rest soon hopefully.)

hhp

Letterpress is an inconsistent process. From a sharp original, it produces images which retain sharpness in some areas, but lose it in others.

There is some logic to this, in that the "protected" areas of the type, eg hairlines at joints, are more likely to retain fine detail than "exposed" parts of the type, such as serifs on stems, which tend to blobbyness. (There is more pressure here, as the paper is stretched over the neck of the type.)

This variety of image quality, with its slight unpredictability, is a pleasing effect.

Back in the day, type designers took this into account, which is why vertical stems on M, N, V etc. were able to be so fine, and the letters proportioned accordingly.

One way to "parse" the effect of letterpress is to mix sharp and soft detail.
In fact, Goudy did this for his pseudo-incunabula faces such as Kennerley, which were designed for letterpress anyway.

I've done this mixing on several of my faces, but here are some examples by Robert Slimbach and Carol Twombly, Adobe Caslon, Jenson, and Minion.

This is very interesting and useful: thanks. And the effect will of course be caught in the average image too.

But one of the things about most incunable printing as opposed to even, say, de Colines' stuff, is that inkpress is so substantial and erratic that even counters typically have rounded outlines, rather than just being formed by the intersection of black strokes, and I think this is a very good thing: it makes the counters as robust as rubber pillows, with their own independent shapeliness. Two printers who were too good technically for this to be so much the case are Jenson, of whom I think you have most experience, and, most obviously, Ratdolt.

How I hate the Minion e.

Yeah, you are probably right, I have not carefully read the entirety of the thread. Way beyond my comprehension.

Out of curiousity though I did go through this and try to find where you mention the source for your Aldine. Since as you say, Griffo's punches are not available. If you did mention it, I have missed it.

A reason for asking this is I had talked to Sergei maybe a year or so ago about what he was doing and I remember being a bit surprised that one of the sources for one of the Aldine italics he was working on seemingly came from a trade book on printing history. So the image he was working from was an offset halftone reproduction of a photo of one of the letterpress printed Aldine pages. Which, of course, is much further afield in terms of fidelity than working directly from an original.

Gerald
The Bieler Press
http://BielerPress.blogspot.com

Hi Gerald, sorry for my earlier testy response. The Aldine italic source is two facing sides of Aldus' 1519 Cassius Dio, scanned at 2400 dpi by Sergei. Long before 1519 Griffo had cut a replacement for this p; he constantly rethought and improved his types (such that only 40% of the 1495 De Aetna lowercase remained unaltered by the time the roman was completed with its new set of caps in the 1499 Dioscurides); it seems the typefounder simply got into a muddle, and cast this older version by accident. Next year I will probably be using the UCLA copies of the 1499 Aldine Scriptores Astronomici Veteres and the 1503 Soncino Petrarch.

Here's an additional note on my choice of italic for revival. For a long time I wanted to get at the finest prize of all: Griffo's third italic, made for his own use as a printer at Bologna. Imagine it: finally, in 1516, the greatest genius type design has ever produced* is free from all constraint. He can make the letters he wants, and print them how he wants them. He would die almost immediately, but first produced at least two editions, a Boccaccio and Bembo's Gli Asolani - the latter appropriately, since Griffo may well have known Bembo personally from the time when the future cardinal was taking a keen interest in the production of De Aetna, for which Griffo cut his first mature alphabet.

I finally tracked down that third italic. The reproduction's ghastly, but what a disappointment! It looks like a Tony Stan ITC Aldine! Having said that, it should have been very readable with that astonishing x-height to line-increment ratio, and it's unmistakably Griffo in, amongst many other things, the low-slung lowercase a, a readability master-stroke that appears in all his italics.

Nick, I think there is one further quality of the original which justifies alterations in the digital revival: I mean the characters' apparent similarity of weight, which comes from the very irregularity of weight in each printed instance. One can't get a strong sense in seeing the Aldine page that one character is heavier than another, because even if there were such a relationship in the punch, it would be lost among the inequalities of weight that come from the printing process. Griffo may even have chosen to cut variety of weight into his punches knowing that in printing these differences would have been greatly diminished. So the weight, and perhaps contrast, of characters could be tweaked upon inspection of the first draft of the digital alphabet until these seemed as harmonious as in the original. This would prevent a repeat of the major problem with Poliphilus.

As promised, here is the averaging script I mentioned above:

http://levien.com/type/avg.tar.gz

It's for Unixy systems, but it shouldn't be too hard to get running on OS X, and a sufficiently determined person can probably make it work on Windows as well.

Here's a screenshot of the script in action:

The top row is three impressions of the 'a' glyph, set in Times Roman 10.5pt, from the Letterproef 1967 specimen book of the Koninklijke Drukkerij Van de Garde, Zaltbommel. I scanned these at 2400 dpi using my Epson Perfection 4870. The next line shows the raw output of the script, a composite of 129 impressions, followed by a sharpened version (Gimp 2.0 Enhance>Sharpen, set to 90), and finally lightened through gamma adjustment (2.5) to show more clearly the contour of the metal inside the halo of ink spread.

Raph, very impressive.

Thanks Raph, this looks wonderful: I'll play with it this afternoon.

Raph,

Your images are very impressive, but can you be sure that the "inner outline" visible is not simply the effect of the sharpening? I'm not very familiar with Gimp, and even less with it's sharpening, but this appears to be exactly the effect I would expect to see when sharpening has been applied to an image.

I sincerely hope this doesn't come across as a parody of your work, but to demonstrate what I mean I've attached an image I prepared in Photoshop. It's a digital Times New Roman 'a', rasterised at 85% grey, with sharpening (Unsharp Mask) applied and then lightened. Despite the crudeness of my sample there are distinct similarities in the "inner outlines" which clearly cannot be explained, in this case, as the contour of the metal.

This may be just that I am mis-interpreting your images. My experience is in manipulating images for aesthetic purposes rather than data, so I could easily have missed something important.

Ed.

Incidentally, if anyone wants to get hold of an Aldine leaf, here is a company selling them from the same book Sergei used for $4 each. Rather more expensive is an early leaf of 1499 roman or 1503 italic: I've just bankrupted myself with one of the Scriptores Astronomici Veteres.

George, let be a little more clear on the danger of aiming at a 'faithful' or authentic revival.

First, you have yourself said you are not going to do this fully, in the sense that you are not trying to revive the look of the aldine page, with all of its variations. So what you are in fact doing, despite protestations of striving for authenticity, is trying to revive features of the typeface that you regard as good or desirable ones. And, as David Berlow pointed out, using a computer does not remove your personal choices as you are telling the computer what to do--to average in a certain way, rather than to pick out the 10 most common shapes of a given glyph, for example.

Like you, I do think that there is much worth reviving in old typefaces. But my point is that reviving the worthwhile is a process of hypothesis testing. You can identify a feature, and then draw it in various ways, and see if it is actually responsible for what you liked in the typeface. It may be, or it may not be. Once you are aware that you are engaged in a creative process of guess work and trial and error to get the effect you admired in the original, then you free your imagination and intellect to create a lot of hypotheses, and test them. If your goal is 'authenticity', you can easily limit your hypothesizing, thinking that one tool, whether it is scanning or averaging or something else give you the 'authentic' result.

You are already engaged in that creative hypothesizing, of course, so my urging is just to not let the goal of 'faithfulness' limit yourself to your first hypotheses, or one method--averaging or whatever--to such a revival. Such methods are valuable tools, but in the end your creative ideas about what is good in the typeface and your testing of your ideas will make your effort fruitful.

Incidentally, in another thread you I think rather underestimate the power of graphic design compared to type design. The look of a page is as much or more influenced by layout as by the typeface. So the way the Aldine page was laid out--the size of type, the leading, the measure etc. may well be a part of the impression you admire. Especially as Peter Brunhill found it worth the effort to figure out the system of proportions Aldus was using, it is likely this is in fact important.

Also, size matters. Some of the much admired old text samples--including yours?--are at rather large size. You cannot scale these down and get the same visual feel. So if you are aiming at a typeface family that can actually be used in a practical way--or in gaining knowledge that you or others can use for such a purpose--you will be faced with scaling issues.

Thanks for your amplifications William.
you are not trying to revive the look of the aldine page, with all of its variations. So what you are in fact doing, despite protestations of striving for authenticity, is trying to revive features of the typeface that you regard as good or desirable ones.
This doesn't follow, and as it happens it isn't the case. I'll revive features of the typeface that I regard as bad or undesirable too: I just hope I don't find too many. And I doubt I'll find any features that I can't learn to accept in reading extended text set in the revival. Choosing a body of work to revive should involve a commitment to that work, which reaches us battered by history but capable of being restored by a diet of soft gruels and attentive kindness: I have more to learn from Griffo than his work does from mine. I wouldn't have made the h in the 1503 italic so rounded, but I can see that it demonstrates a difference between humanist and mannerist ideals, and I anticipate the broadening of my own ideals to accommodate both possibilities.

I'll tell the computer what to do, and control the inputs it receives, only insofar as each choice can reasonably be expected to get me closer to the newly cast type plus average inkpress. What I hope to revive is a combination of very subtle features, varying from letter to letter, which could not but be bulldozed by any heuristic process of feature identification and redrawing. The time for heuristically developing my judgement is, in this project, afterwards, in choosing how much emolliating inkpress should consistently be removed, and, if necessary (as I hope it will not be), how much the weight of characters should be made similar, in order to get a combined effect resembling the original in bite and harmony, while preserving intact the combinatorial effects of the initial draft.

The look of a page is as much or more influenced by layout as by the typeface. So the way the Aldine page was laid out—the size of type, the leading, the measure etc. may well be a part of the impression you admire.
Sure, but satisfying and appropriate page, margin and textblock proportions aren't, with Bringhurst to hand, rocket science, and Aldine leading (1/12th of the line increment) will be incorporated into the fonts, as it was by Griffo. As for size, the horizontal and vertical proportions of the originals will be exactly replicated at ~15 and ~12 point for roman and italic respectively, and horizontal proportions and relative size of irregularities will be modified for optical similarity at other sizes using MM.

George asked a very good question: can you be sure that the “inner outline” visible is not simply the effect of the sharpening?

I wish I had some kind of definitive test to show that it's real, but right now what I mostly have to go on is a very strong hunch, developed from studying hundreds of glyph images from a half dozen or so pages. The halo I'm attributing to ink spread varies from page to page, mostly dependent on the paper - it's quite a bit stronger on coated stocks.

So the question remains open: what kind of experiment would test the null hypothesis that what I'm seeing is nothing more than a sharpening artifact? A good answer would most certainly be worth an academic paper, in my opinion.

Hi Raph, I think printing digital type onto a number of different papers from photopolymer, making averages from the results, and comparing each sharpened and unsharpened with the original glyphs, as Nick suggests, would pretty much do it. But you're too kind in attributing such a good question to me rather than to Ed! I think the first thing to find out is what these sharpening algorithms actually do.

I've just managed to convert my PNGs to PGM so I look forward to being able to try out your software. Meanwhile I've sharpened up my average - I think my two-pass blending routine is fractionally more accurate with such messy samples -

Another use for image averaging.

http://salavon.com/PlayboyDecades/PlayboyDecades.shtml

Another use for image averaging
I'm afraid they're mixing punches there.

It would have been more interesting if that guy had stuck to faces: in one study, averaging images of the faces of participants in a female beauty pageant gave an image which was found more attractive than any of the originals. Something like the same effect might be operating in my average above (depends whether you like the punch, I suppose; as I mentioned, Griffo apparently didn't).

Me: "You are in fact doing ...is trying to revive features of the typeface that you regard as good or desirable ones."

You: "This doesn’t follow." [from the fact that you are not trying to revive all features.]

Your objection to my argument is valid. My mistake.

However, the underlying issue I think remains a question you should consider, to make your projected year's work most productive: What purpose are you doing this for?

Is your decision not to revive variation but to do everything else capricious? It being not capricious, what are your guidelines? If your purpose is simply to learn, not produce a typeface to use, then what specifically are you trying to learn? What questions are you trying to answer? Producing different revival versions that revive different features might be more informative than trying to revive everything but a few features. This is just from a research point of view. The general point about research (which was made by Popper) is that the more developed and articulated your theory is, the more informative your testing and your results will be.

It is for you to say of course, but I find it a little hard to believe that you are willing to spend a year on a project reviving a typeface and are indifferent to whether anyone will use it. I say this to reinforce my question, which many of the best type designers say is always their first one: What is your brief?

I ought to say, William, that I share your doubts about my position, and that I might yet in practice change my theory.

Is your decision not to revive variation but to do everything else capricious?
I assume (do correct me) that by variation you mean differences between prints from a single punch, and that by the capricious you mean differences in treatment of similar features from punch to punch. I expect to try to revive all the capriciousness which is sufficiently marked to survive the averaging/sharpening process. I'm not sure what you mean by "It being not capricious, what are your guidelines?"

You should understand, as no one would from just seeing either Bembo or Poliphilus (which is in no sense a replica even of selected printed letters), that Griffo loved oddities, and that his work, far from being a rough predecessor to Augereau's and Garamond's, is sui generis, taking the capriciousness of Jenson and adding to it a quality which for want of a better word we might call typographic (his 1495 roman is narrower than any MS hand); his French followers would take the latter but forget the former. Thus Griffo's 1499 cap X has reversed stress, and his lc x in the Hypnerotomachia is bizarrely twisted. Take most of the capriciousness away and you get Abrams Augereau; take more away and you get Adobe Garamond. Because I find Griffo's work in its printed form uniquely beautiful, I want to learn what microproperties give it its beauty: I want to solidify these properties in a way that cannot be done by just looking at the printed page. I don't think printed variation substantially contributes to this beauty; and to me it is often ugly.

Producing different revival versions that revive different features might be more informative than trying to revive everything but a few features.
In a perfect world, I'd make three versions: one from following averaged forms in such as way as to preserve apparently damaged parts; one as I've described, taking the first version and reworking heuristically for bite and harmony; and one taking the second version and heuristically eliminating all irregularities that seem inessential to the effect.

The general point about research (which was made by Popper) is that the more developed and articulated your theory is, the more informative your testing and your results will be.
I accept the point generally, but could you clarify what theory of mine, or theoretical area, you're referring to? My theory of typographic beauty, of readability, of the nature of Griffo's work, of the morality of type revival?

It is for you to say of course, but I find it a little hard to believe that you are willing to spend a year on a project reviving a typeface and are indifferent to whether anyone will use it.
You'll have to trust me on this one. But I'll use it myself it at the fine press I expect to start within a few years. The brief is primarily to recreate a set of letterforms at original weight, and secondarily to make a book type, for digital and photopolymer printing on rag and esparto stock, in optically adjusted sizes.

>Because I find Griffo’s work in its printed form uniquely beautiful, I want to learn what microproperties give it its beauty: I want to solidify these properties in a way that cannot be done by just looking at the printed page. I don’t think printed variation substantially contributes to this beauty; and to me it is often ugly.

I see now you have a very clear concept of what you are after, which is impressive. I do think, as it seems you suspect, that you will find that when it comes to actually doing the face or faces, there are thousands of decisions to be made, and your concept will be refined and sharpened, and maybe changed by your new ideas and test designs.

My main point in posting on this thread is to make clear that the process of making a good revival is by no means an automatic process you can program. It takes many creative decisions, and in the end the result will be your product, based on Griffo, and not Griffo.

Yup, absolutely. And for the same reason - that a revival of something is not the same thing as the original - I think it's false modesty to name a revival directly after the original's punchcutter or printer, as Morison did with his infamous "recutting"s; this will either be called Horton's Griffo or something more vaguely suitable, like Astrophel - correctly reminiscent of Poliphilus, but pointing at a different book, the Scriptores Astronomici Veteres; and alluding more plainly to another Englishman's appropriation of the Italian Renaissance, Sidney's Astrophel and Stella, which was modelled on Petrach's Opere Volgari, for the printing of which Griffo made the italic I'll be using.

The idea of solidifying microproperties that give Griffo's fonts singularity strikes a chord with me as you might expect. And I'm interested in the suggestion that Griffo's featural innovations are transitional, in the sense that they form a bridge between humanist and baroque forms.

In connection with the term 'capricious', I'm wondering if you see Griffo's norm violations--vis a vis what comes before--as ad hoc letter by letter, or if you expect to find a systematic application across the glyph set(s) of a well-considered feature or contrast manipulation scheme. Are you hoping your averaging mechanism will converge on a definable or characteristic set of holotypic (Bringhurst's term) microproperties applied across the glyph set?

I didn't know you were interested in microproperties - I thought you trusted more to the role-architectural. Griffo's romans may fairly be called transitional between Jenson, whose matrices Aldus acquired by marriage, but chose not to use, and de Colines etc. Griffo's Lascaris roman, his 'prentice work, is a typical sub-Jenson Venetian. But with the 1499 caps Griffo is zipping off in a direction in which no one would follow him. Digital Poliphilus is just about accurate enough to see what he was doing, though, in approximately descending order of fakery, YRXJKZMNSQP are inaccurate (every single one of the changes seems to me a clear worsening). The widths are completely reset without much reference to either monumental or Venetian norms, and the cleverest seriffing I've ever seen is applied to bring the counters into readable relations with one another: this is the only alphabet of caps I'd want to read a book in. Such capriciousness is systematic. In the roman lowercase in its "completed", coherent state of 1499, I think serif capriciousness is more Jensonian, partly matching sidebearings (as in Goudy Lanston), and partly, along with variation of axis and vertical alignment, there for divergence's sake. I'll have a clearer idea of this when my Scriptores leaf arrives in a few days. I have no doubt at all, from the relatively low-res Hypnerotomachia scans I've studied, that Griffo was systematic about axis variation between similarly shaped letters.

But with Griffo's "norm-violations" we need to distinguish between different norms - modal Venetian forms, modal Venetian capriciousness (substantial), modal MS forms (very varied and influenced by fashions - there was no "humanist roman" style, they even had a wave of vertical-axis, geometric-bowl work which Hrant might call Modernist), and modal MS capriciousness (which could be low but which was obviously potentially very substantial). The clear departures from all the preceding Venetians and MSs in Griffo's mature work are narrowness, variety of axis, perhaps rotation of the counter, modification of serifs on open counters (with counter understood as any white space within the rectangle of the letterform), and small-eyed lowercase g. Griffo's capriciousness is not in itself norm-violation since everyone was doing it. The flat bar on e is not a violation of MS practice, though it is one of type-design practice. Making italics at all is obviously a violation of type norms, but not of MS norms. Likewise with a small but important feature of Griffo's work: stems are more equal in width from top to bottom than, following lapidary examples, type designers of sufficient skill tended to make them (Jenson, Centaur and Requiem are all waisted in this way). So:
Are you hoping your averaging mechanism will converge on a definable or characteristic set of holotypic (Bringhurst’s term) microproperties applied across the glyph set?
Yes.

"I didn’t know you were interested in microproperties [...]

See the "start" of my Typo#13 essay: "Like the structure of deep space or the oracle at Delphi, a grounded understanding of perceptual processing in reading, and a tabulation of the perceptual processing impact of manipulating typeform microvariables, are maddeningly elusive and the literatures surrounding them are, when it comes to perceptual (or cognitive) processing in reading, an ever expanding mer à boire [ sea to drink ], an in errorem variarum ambage viarum [ conflicting maze of wandering paths ], but, when it comes to tabulating impacts of manipulating of typeform microvariables, virtually non-existent.

The question I ask is: where, relative to receptive-field behaviour in the visual cortex, is the inflection-point between 'optical grammatical' and 'gestural atmospheric' matters, when it comes to tabulating the perceptual processing impact of typeform microvariables.

Emigre talked in the 1990s of type designers of that time as having a collective sense being 'primitives of the new era.' Griffo and his contemporaries and predecessors were the primitives of the metal era. I have often felt that cutting punches changes one's relation to the letterform, from a ductal routine empahasis to a situation where form-analytic featural evaluation begins to play a direct role in the very process of cutting. So there arises a tension between 'chiro-referentiality' or 'chiro-paraphraxis' and form-analytic based feature manipulation. This might be what you mean in talking about 'typographic' properties. This much I grant Hrant: feature manipulation based on form-analytic feature evaluation beyond strict chiro(para)phrastic contrast manipulation has a positive role to play.

But it seems to me that it is the relation of these two countervailing pressures that is being sorted out in this cricible where you are mucking about. At least for the metal era.

Thanks Peter. In the French old styles of the second quarter of the sixteenth century ductus is filtered through a static set of conventions, which reduced the real range of thicknesses given by the pen at different angles to a set optical thick and a set optical thin, modulated between on curves, and made serifs lachrymal or bracketted instead of the (slightly) more pen-formed straight or concave slabs of the Venetians. That is the punch-cutter's feaure-manipulation in its most normal form. Griffo's transformation of ductus is, by contrast, not regularising but actively divergent; his motivation for departure seems to have been quite opposite to his imitators'. (That said, in his italics he had to be more circumspect, because the whole point of that venture was to pretend to imitate the natural style of writing for oneself or a friend. Thus Q was given back its normal axis rather than the counter-clockwise rotation it had in 1499, though it seems that subtler things were going on in interior axes).

Hi Raph, your software works wonderfully well, but with my samples splits images from a single punch into several classes. Is there a way to increase the tolerance of difference?

I have high-resolution images now, and could send you the Soncino scans, Raph. Here's an example of that italic:

Gorgeous scans. Yes, send them to myfirstname.lastname@gmail.com if they're less than ten or so megs, otherwise we'll work out some other way to do bulk transfer.

Is there a way to increase the tolerance of difference?

Yes, it's the incredibly well-named and clearly documented tscale parameter. 4 is the default, 3 will be more tolerant of differences, 5 picker. I chose 4 to work well with my ATF scans, so it's not surprising that a different source might need different tuning.

Of course, the whole reason I made the tools work with text files is so they could be hand-edited. Here's a procedure that I find works reasonably well. After the first pass, open the "class" file and search-and-replace the "class123" names to be more descriptive. Then, keep that file open in one window while you open your favorite slideshow-style image viewer in another. Any time you see a misclassified glyph, change it in the class file. Then you can run mkblends.py again and get much cleaner composite images (with descriptive names, to boot).