We see that line length diminishes at the same time as the height of the letters, but that the decrease in the width is much slower than that of the height, because the engravers have recognized, without fully realizing, that the decrease in readability is mainly attributable to the decrease in the width of the letters. That is an important fact that we have noticed for a long time.

ETA: graveur probably should be taken as 'typecutter' in this context. Whether lisibilité is translated as 'readability' or 'legibility' depends on how you're using the terminology...

Thanks Jongseong, this is much better than my previous Bablefish translation... but am trying to make it clearer...

By tweaking this a little would this still be an accurate translation.... (or have I gone too far...)

Upon a first glance, it is seen that line length decreases in direct proportion to the diminishing height of the letters. However under closer scrutiny, the line width decreases at a slower rate than to that of the height. The punch-cutters have recognised that by allowing the letters to become proportionally wider as the text becomes smaller maintains legibility. They understood that a negative impact in legibility is mainly attributable to the decrease in the width of the letters resulting from a 'natural' reduction in scale.

I have included the Fig from the Book however - it doesn't follow the rule observation above very consistently!

_________________________________
Michael Hernan

I would probably describe that as an expanded paraphrase of the original rather than a translation. Having said that, I think it's an accurate depiction of the intent of the original. The original doesn't spell out that the letters are made proportionally wider, just that the decrease is slower than the decrease in height, but that comes out to mean the same thing.

@Jongseong Thanks for that.

>expanded paraphrase

-will do.

Yes, it was the 'slower' that was throwing me.

Task done - Next....

_________________________________
Michael Hernan

I have included the Fig from the Book however - it doesn’t follow the rule observation above very consistently!

First, I don't think that "en même temps" is to be interpreted as meaning "directly proportional". I read it as meaning that if point size decreases, so does the length of the line displaying a full alphabet. I would rather translate as follows: "The line length (in fig 60) is seen to decrease with point size" (which is a weird way of saying that the line length is an increasing function of the point size). The citation also means that if character height is decreased by some percentage, then the width will be decreased by a smaller percentage. That seems to be true except for the lines at 5 and 4 points. Indeed 4 points 20% less than 5 points but the length at 4 points seems to be about 24% less than that at 5 points.

Michel, You are right, I have introduced "directly proportional" as a layman assumption, so as to make the next sentence make more sense. i.e so it follows : However under closer scrutiny, the line width decreases at a slower rate than to that of the height.

I like the way you describe the relationship by assigning percentages to each the length and height - so I can lose the "slower" description - Cool.

Its funny that at 4 and 5 points the rule breaks. I wonder if this is because such small text is not supposed for extended text and therefore can be made to become narrow again?

_________________________________
Michael Hernan

Michael, Maybe I was too categorical in my statement.

Since all the lines are equally spaced vertically with a constant increment of 1 point, the line length would in fact be directly proportional to the point size if all lines ended on the diagonal red line in the following figure (since both vertical and horizontal scales are then linear).

This would happen if the width of each character was proportional to its point size i.e. if incrementing height by a certain percentage would correspond to incrementing width by the very same percentage. Now, most characters are taller than wide. For a character that is twice taller than wide, if proportions are always kept that way, the rate of increase of the height of that character will always be twice the rate of increase of its width; that is a trivial mathematical fact, having nothing to do with the skills of punchcutters; I thus can't image this is the kind of speed Javal had in mind and that is why I took for granted that he was comparing relative changes instead of absolute changes. If my interpretation is correct, then the line ends are not on the red diagonal (I guess they should be on some convex curve) and the line at 4 points should be much longer.

After looking at other threads and links concerning that book, I now feel like having a look at it. I guess I'll do that tomorrow.

Michel

One sees that line length decreases with letter height, but the decrease in width is much slower than that of height, because engravers recognized, instinctively, that decrease in readability is mainly attributable to decrease in letter width. Therein is a major effect which we [I] have long pointed out.

In the chart, letter height decreases from 14 to 4, a factor of 3.5.
Line length decreases by a factor of 3.
All he's saying is that 3.5 is bigger than 3.
I don't think he intends there to be any exactitude in his demonstration, if one takes "sans bien s'en rendre compte" quite literally, i.e. "without properly committing themselves to counting".

[edit] removed.

[edit] "se rendre compte de" means "to realize". "Je me suis rendu compte qu'elle était ravissante dès que nos regards se sont croisés" should have nothing to do with counting!

@ Hernan: Can you cite the article you are referencing?

All he’s saying is that 3.5 is bigger than 3.

You say it in terms of factors, I said it in terms of percentages. Both approaches are equivalent.

“se rendre compte de” means “to realize”.

If that is so, what does "recognize without realizing" mean?!

The text in question was written over 100 years ago. Words and phrases have many meanings, which change according to time and context; dictionary translations also change, and writers can indulge in word play.

With that in mind, I translated "sans bien s’en rendre compte," as "instinctively", taking into account that the word "compte" may perhaps contain at least a trace of its discrete meaning.

**

Interestingly, the word "capital" (French) would have translated very nicely into "capital" (English) 100 years ago, although now that is an archaic usage (e.g. "Capital idea, Watson!" -- Sherlock Holmes). Is "capital" (French) in the Javal usage similarly archaic?

Nick, translating "sans bien s'en rendre compte" by "instinctively" seems quite fine to me. I was just astounded by your literal translation. I never thought that Baudelaire, who died in 1887, sounded oldish, but maybe I am getting old myself (I even studied Villon (who died in 1463) and texts in sixteenth century French when I was in high school).

[Added] I think I could indeed use the word "capital" in class the same way Javal used it and I am quite sure my students would understand.

[Added] I just Googled "fait capital" and found it on recent blogs. I am not getting that old after all!

@ Boyer: Is this you? You both appear to be academics from Montréal…

@ Dean: Yes this is me.

I found the first translation by Jongseong very accurate and to the point, there's no need in over analyzing the syntax.
But I think it's important to note that the punchcutters didn't work in vectors and so scaling up and down was not available. In fact they didn't even think of it that way they had no intention of scaling the shapes they were cutting, because they were not executing architectural drawings. What they were doing were size specific fonts (size specific fonts were a technical limitation here, not a readability or aesthetic issue) then it made sense that all the sizes have the same style or be on "typeface"

:D I just finished reading Counterpunch by Fred Smeijers and became an expert on punchcutting!

To clarify the source:

Compact Typographie (Section 3) (page 219) from:

Javal, Émile. Physiologie de la lecture et de l’écriture. Bibliothèque scientifique internationale, 105. Paris: F. Alcan, 1905. (p219)

Reprinted Edition Published by Les Classiques des Sciences Humaines, Paris, 1976

@Michel, "instinct"- perfect.

Of interest: Looking at how the cascade is working, I suggest it might be that three drawings were made and used on a pantograph (Assuming c.1900 at Deberny Foundry) to create multiple type size:

A. LARGE used for 14, 13, 12, 11
B. MEDIUM 10, 9, 8, 7, 6 (and 4)
C. SMALL 5pt only

by using the medium drawing for the smallest size means that a narrower set is achieved.

@Nick, @Michel, "The punch-cutters instinctivly knew"- perfect.

Of interest: Looking at how the cascade is working, I suggest it might be that three drawings were made and used on a pantograph (Assuming c.1900 at Deberny Foundry) to create multiple type size:

A. LARGE used for 14, 13, 12, 11
B. MEDIUM 10, 9, 8, 7, 6 (and 4)
C. SMALL 5pt only

by using the medium drawing for the smallest size means that a narrower set is achieved.

For me this is one of those black art - secret knowledge things, - that continue to bug me.

A properly performing optical cascade is a rare thing. It would be an interesting project to compare from different foundries who were considering such things up until the beginning of the 20th century. I think by the time we get to Univers, new rules are in place and old ideas have gone (lost). We are now grappling for this insight.

Where cascades are irregular (such as this one), it is really interesting to try and work out what was going on in-terms of intention, limitations and practical approaches to achieving intended results. Odd discrepancies such as the narrow set 4pt add added intrigue also reminding us that rules are to be broken when necessary.

Why I have been fascinated in Javal, is that he is a lay person with a keen interest in optics. My imagination has him in correspondence with M. Tegu (perhaps actual meetings) where the secrets of type manufacture are being explained to him (Javal is blind), then they are being recounted in the book but only as half facts because Javal doesn't quite have the right knowledge to know for instance, if the punch-cutters (engravers) actually knew what they were doing!

But it is precisely this outside position (mentality) which gave him permission to create with M. Dreyfuss radical designs solving optical problems.

/m

> a constant increment of 1 point

Is that correct? Or would a constant increment be a fixed percentage of size. Going from 9 point to 10 point is 111.11%, while going from 10 to 11 is 10 percent, etc. (5 to 6 is 120%).

@Don: Or would a constant increment be a fixed percentage of size.

If you do that, you get what is called a logarithmic scale. That seems to be the right scale indeed because the word "slower" used by Javal does not apply to width versus height but to the log of the width versus the log of the height (the "speeds" that are compared are the rates of change of those logarithms, i.e. logarithmic derivatives).

@Michael: I suggest it might be that three drawings were made and used on a pantograph

Correct me if I am wrong but I thought a pantograph would just scale down proportionately the given pattern like a postscript font is scaled down when we select the point size. If so, the line ends for a fixed design should be aligned like what occurred to Adobe Garamond Pro in this grab:

For Fig 60, we rather get

and the line at 13 points looks shorter than it should for the pattern to be the same one scaled proportionately.
Michel

@Michel
Essentially you are correct - a pantograph (punch-cutting machine) will make very strict reductions as shown in your top diagram.
However I am finding that in some instances and perhaps with this example (fig. 60) that artistic licence might be taken with the pantograph drawings. Perhaps tracing the drawing's outlines with adjustments being done on the fly. The operator of the machine acting with the same sensitivities as the traditional punch-cutter? I can't confirm this but open this hypothesis up...

p38 Letters of Credit, Walter Tracy
Says of Updike:
I have sometimes questioned whether a machine can be so managed that it will ever produce those fine and imperceptible qualities of design given to it by the hand of a clever type cutter... That there has been an improvement of late [he was writing in 1922] in type cut by machine is undeniable... This improvement, I learn, has come to pass through a more sympathetic and subtle manipulation of the machine itself, and by the rules by the eye of the workman who operates it.'

Tracy says no more on the matter. =(

_________________________________
Michael Hernan

There is considerable variation in the letter details for the different sizes in this sample. Take a look at the descender of the y through the different sizes, for example.

I don't think there's much to be gained by detailed analysis of the measurements of Javal's specimen.
For the purposes of his argument, he was quite satisfied to consider the "hauteur des lettres" to be the nominal point size of the type.
However, to this day there is still no standard, objective way of defining type size.
The problem is fundamental, as was discussed in a recent Typophile thread about the definition of the em.
Still more recently, another thread discussed the design of "small" type, and the adjustments that type designers make to scale a style up or down (and by "scale" I don't mean in a strictly mechanical sense).

These are some of the things that may be modified:
* x-height relative to cap height
* descender length relative to x-height
* serif width relative to stroke width
* sidebearings relative to letter width
* horizontal-to-vertical proportion of letter
* size on the body (i.e. how close extenders come to the edge of the em)
* contrast between thick and thin strokes
* weight

I don't think this is a longer list for the designer of digital type than for the punchcutter, it's just that considerations which were once absolute are now relative.

The amount of press gain in play (or anticipated) is hugely significant. A type which has an even scaling of weight, such that different sizes appear to be the same weight on coated stock, will, on uncoated stock, appear to be a heavier weight the smaller it gets.

The rule expressed by Javal is equivalent to saying that the width divided by the height of an "m" should increase as the point size decreases. Another way to say it is that if we scale all the "m" to the same height, the m width should increase as the point size decreases. I went to the library and took pictures of all the "m" from 14pt to 4 pt and scaled them proportionately so that they all be of the same height. Here is the result:

I fail to see an increase in the width as the point size decreases.

Michel

I've got Javal's original, and if you look at it, you clearly see that it's not the same typeface used for all the sizes (I would say there are at least 4 or 5 different typefaces here), so the inconsistencies in the cascade are not really surprising.

It guess I should have taken the l height to scale. Here is what that gives. One can now see how the m (I guess the x height) grows, but for me that gives no convincing evidence for the law expressed by Javal.

Michel

Added remark: if the point size is proportional to the l height, then the law expressed by Javal becomes (I guess) that the m width divided by the l height should increase as the point size decreases.

...the law expressed by Javal.

What law?
He made a general observation, with a single typeface (or several, as it transpires?!) as example.
And his definitions of distance-- "hauteur", "largeur", and "longeuer"--are non-typographic with no mention of point size, x-height, letterspace, or picas.
Furthermore, line-space (leading) is fudged by equalization.

Nonetheless, there is some truth in his observation, that display sizes tend to be more condensed than regular text, and micro-sizes expanded horizontally.
He attributes this to the demands of readability, however, my theory is that it is because in toto, there is more room for letter details to be heavied up in the vertical axis--which they must be when reduced in size, to avoid disappearing from sight-- by expanding into "ascender space", than along the x axis. Therefore in the smaller sizes, extra proportional "heft" is accomodated in the vertical space of the type body, by increasing x-height, but that produces a condensed letter shape, so to maintain letter proportions, the character width is increased.

Re rules: I transcribed (by hand) the Compacte Typographie Chapter and translated it though BabelFish.
I counted approximately 16 Rules which I agree are more like observations.
Javal didn't list these observations in an easily accessible way but weaved them into the text.
Most are to do with resemblance or quite obvious stuff as Nick listed a couple of posts up. I have diagramised the resemblance rules and complimented these with 'Grant and Legros'' and Weidmann's obsevations to be included in my essay.

Javals observations: @Nick: Javals observation is a good one, it is a shame about the "fraudulent" figure that discounts his case!

Typographic FAIL?

@phillipe_g My copy unfortunately is a stereotype. =(

However not *all* the fonts are different...
The top end of the scale sees 12 and 14pt from the same family (more Didot in appearance)
13 pt is doing its own thing! It has the open gait in-between the legs of the m and quite wide p and q making the set alot wider.

However from 11 to 4 pt looks like its the same font, and this is the focus of interest anyhow.

Even though the figure in this instance is a bit of a mash-up. I hope to see other cascades and deconstruct some more punch-cutters intentions.

/michael

I don't think you can say the figure is fraudulent : it is more than probable that the typefounder did not have all sizes in the same typeface, so Javal had to do with it. It was a common practice at the time to have to use a different typeface for each font size.

Concerning how many typefaces there really are, I'll let you judge by yourself from this 1200dpi scan of my book:


It seems quite clear to me (judging from the a, the g as well as the y) that the fonts from 14pt to 11pt are all different. For the smaller sizes, it's more difficult to judge: some of the typefaces seem alike, but between the letters a, g, y, f/j and t, there's always one which is quite different.

Here is what happens with the font Computer Modern which has opticals for 12, 10, 9, 8, 7, 6 and 5 points. The line corresponding to 4 points is obtained by scaling 12 point to 4 points. The blue line is thus where the lines would end if the the 12 pt font had been scaled instead of using the optical. I guess this is the kind of figure Javal hoped would appear in his book (if he was blind, he never saw how the figure came out).

Michel

Well, I might be completely off the trail, but how was the original figure done? Was it actually typeset or engraved? It would be very hard to achieve the desired outcome in metal ... just a thought.

The paragraph that comes before the figure is: "Note that usual typographical characters don't give specimens that would come out if we simply scaled down the letters. Here is a series of types". The figure was thus clearly typeset with available characters. I think Javal expected the available fonts would behave in such a way that when the lines are typeset with them, the right ends would make about the same curve as CM does here (of course, at 4 points, it was also an optical that was to be chosen). I don't know how many fonts were available at the time but it is my feeling the typesetter may have chosen the fonts to get a figure to his liking; he did not understand what Javal had in mind. Don't forget that Javal graduated in 1860 from Ecole des mines de Paris before studying medicine. He had an excellent background in mathematics and physics.

Note: the straight blue line corresponds to the imaginary line the reader would have traced from the end of the line at 14 points to 0 points, at the bottom of the figure.

Here is a graphical representation of what I think Javal expected:

It is the line at 4 points in the scan that upset me. It seems to have been chosen so that the 0 aligns with the end of the 14pt line and I felt it was a cheat. Remove that line and the feeling left by the rest of the figure that appeared in the book it not that bad.

It would be very hard to achieve the desired outcome in metal

Really?

Contrary to the sample in Javal's book where it seems the x height divided by the l height increased as the point size decreases, heights are kept fixed in CM. Here are F, x, f and g in 12, 10, 9, 8, 7, 6 and 5 points (just to better see what is involved in the above CM grab):

Edit: I should of course have said "Here are F, x, f and g in CMR12, CMR10, CMR9, CMR8, CMR7, CMR6 and CMR5."

...heights are kept fixed in CM.

...because two parameters, horizontal scaling and "swell" are varied with optical size in that typeface.
It is possible to also vary x-height and other parameters, using Superpolator, or by eye.

In CM, it looks like a mechanical interpolation.

@paragraph
are you asking asking if it is engraved because of the original transcript with the use of the phrase "les graveurs"?
It is likely that the Javal figure either way was made to approximate set type.
I believe the irregularities have been valuable to fuel this discussion regardless of the figures manufacture.
It is leading and exposing us to other examples (most recently with CM) and other *known* intents by designers.

I was surprised to see Computer Modern used as an example! However upon reflection I realise that it is a great model to continue the argument because it forms an ideal to which to contrast against.

Without undermining the sophisticated algorithms used by Computer Modern, it is still (to me) a very single minded and inflexible approach which is insensitive to broader or task specific usage.

The original (Deberny) Javal (Fig.60) example is a more sophisticated model. (If we can assume that the figure *was* broadly the intention of the author)... Where we see a single arch in Computer Modern, in Fig 60 three arches can be seen which correspond to broadly accepted typographic usage's; Text, Footnote and Caption.

I have an open agenda in promoting such a 'range' concept or 'size specific' design as I believe high quality typography comes out of such considered usage.

Multiple approaches (non exclusive) have been revealed so far:

A: CM - Design changes algorithymically where interpolation between a big and small size affects the with of a type to become wider the smaller it gets.

B: Size Specific (Type I) - Litterally size specific with out attributed task. Each size has a slightly different design at the discretion of the punch cutter or designer seen with the 18c Caslons (seen also in Miniscule (inspired by Javal/Dreyfuss experimental designs)) without a specific task attributed to the design

C: Size Specific (Type II) What I will call the AdobePro model (is there a more accurate term for this) which assigns a names use to a design.

D: ??

Note: I think it is worth mentioning for clarity for anyone new to this topic that what is not obvious until someone is made aware of the fact (was so in my case) that all the above conditions are in contrast to much of what exists in out digital type Library, which I can state confidently are types for the most part developed to work best at 9-12pt.

Here is a grab of a pdf produced by XeTeX using the Utopia Std opticals. The fontspec package chooses Display from 21 points above, Subhead from 20 to 14 poionts, Regular from 13 to 8 and Caption from 8. I checked in Acrobat to be sure the point sizes and the fonts are as indicated and so they are. This is a font I use a lot for mathematics (with the Fourier package) and that I like.

I just found the FontShop Optical Sizes link. It misses Kobayashi's FF Clifford (with FF Clifford Six) as well as Adobe fonts. The link nevertheless allows seeing that the widening of characters as font size decreases that Javal promotes during three pages in his book actually occurs with Eldorado (with Eldorado Micro), ITC Bodoni (whith Bodoni Six), MVB Sirenne (with Sirenne Six) and Times New Roman

where the middle font is called "Times New Roman Seven" and the top is "Medium and Small Text".

Michel

Just to add something to the discussion I have made a rough example of the waterfall with line-transport corresponding to the typesize (eg diminishing transport). It looks quite different, I think:

Less of a variance, right?

. . .
Bert Vanderveen BNO

And here is Computer Modern set solid 10/10, 9/9, 8/8, 7/7, 6/6 and 5/5.

I added cmr17 and cmr12 to the picture. The text is still set solid, as above. The yellow lines are obtained by scaling. The first three lines are cmr17, the next three cmr12, the next two cmr10 and then cmr9, cmr8, cmr7, cmr6 and cmr5, always set solid. The scaling is always proportional, as usual; cmr17 scaled at 15 points gives about the same length as cmr12 scaled at 14 points. We also see that cmr17 gives a shorter line than we would have expected from the rule of thumb we could have been tempted to induce from the above examples.

Michel

The more I look at the fit for sizes under 10 points with the CM font, the more I am surprised. If I rely on Haralambous 1993, Parametrization of PostScript fonts through METAFONT — an alternative to Adobe Multiple Master fonts (pdf, 552K), sizes under 10 points are to be treated differently from those above 10 points. This thread being concerned with small sizes, we may just forget what happens above 10 points. It is also said that, from the 62 parameters used to generate a CM font, two are relevant for this scaling, "cap-stem" and "slab", shown here.

On page 155 (Section 7, optical scaling) Haralambous goes on to say:

Just for the record, those values are (cmr17 size being actually 17.28 points)

What I find surprising is that those values supposedly obtained after actual measurements have no direct obvious relationship with the line length. On the other hand, if all the lines in the grabs above end on the blue line, then that means that the length L(s) of the line at s points (for s less than 10) satisfies the formula L(s) = ms(s-1)+b where m and b are constants to be found by measurement. I wrote a XeLaTeX program that outputs the exact value of Length for each size s and then used Excel to find m and b (by linear regression of Length vs s(s-1) .

I get that the length Length at size s is 0.5427s(s-1)+71.5982 with an error that is less than 1% in all cases but one, and less than a quarter of a percent in all cases but two.

How could such a precise relation (quadratic in the size s) hold for Monotype Modern 8A, or even for any CM family whose parameters would have been filled by inspection and measurement?

Michel

Haralambous' paper (pdf, 341K) is now to be found in EPODD, Volume 6, Issue No. 3, September 1993, pp 145-158.