Typophile

First, I should have kept to the definition of legibility in "Types of Typefaces": "legibility is based on the ease with which one letter can be told from another." By keeping legibility to letters, and readability to words in extended text, the distinction is cleaner. I used a different definition: what can be read best under difficult conditions vs good conditions.

Kevin: "I don’t take this as evidence that legibility and readability are distinct, because there was no finding here that could have falsified the hypothesis that legibility is distinct from readability. If Arditi & Cho had found that lowercase performed better under stressful conditions would you say that legibility and readability are the same? Likely no."

Any claim that "you can find that x and y behave differently in some way" is not in a refutable form--you are correct. However, if you specify the ways in which they should differ, the list is finite and it becomes testable. And I did give the list: speed, comprehension, and comfort. Further the hypothesis that "x and y never behave differently" is in a refutable form, and refutation and finding of a difference is enough to establish the utility of a distinction. And in this case the hypothesis of no difference is refuted.

Arditi and Cho have refuted the claim that there is no difference. For they found that speed in reading at 50% accuracy is different between all caps and normal mixed case of the same point size under difficult reading conditions, but these differences disappear for normal readers at bigger size. That is a difference. If the same holds for identifying single letters--and in their tests ability to identify single letters was a factor--that would establish the difference under Lieberman's clearer definitions as well.

I'm sorry I missed your Sept 4 statement saying that the question remained of the reason for the differences between Arditi & Cho and past results; had I read it, I would not have been so skeptical about your willingness to experiment on this issue. But I am still curious as to what earlier studies made you think that the same design conditions are equally legible (speed + comprehension, leaving out comfort) under difficult conditions and easy ones?

I would be delighted if you are willing to do experimental tests on this issue. If you are, another test of readability under relatively good reading conditions is frequency of proof-reading mistakes. It is widely recognized that it is easier to proof extended text on paper than on the screen, so that, as well as the timed test of speed + comprehension with difficult material, seems to me a promising indicator of readability.

Yes, the theory that any typographic variable (e.g. uppercase versus lowercase) will show that one condition (e.g. uppercase) performs best in all tests has been refuted. I will have to develop a better theory.

William, I think you’re proposing that for any typographic variable that one condition will never perform best in all tests (speed, comprehension, comfort, size threshold, and time threshold). I think I can refute that with a thought experiment: I hold the firm belief that 12 point text is a better size for reading than 2 point text (I am hopeful that everyone else holds this impression too). I believe that 12 point text will perform better than 2 point text in all of these tests.

>William, I think you’re proposing that for any typographic variable that one condition will never perform best in all tests.

No, I didn't and I'm not proposing that. I don't pretend to know about all conditions in all circumstances.

What I am saying is simply what is regarded by type designers for a very long time as fundamental: different designs are most suitable for different purposes: very small type, extended text, headings, display, screen, newspaper, etc. So *some* conditions are best changed, depending on the intended purpose of the type.

And the changed conditions affect readability under different circumstances, such as size, print medium, or computer screen.

In the case of agate proportions, I would be willing to bet money that a face with such proportions will not beat out in readability (measuring speed comprehension and comfort) a good text face like Minion in print at normal sizes and decent layout.

I do think that somewhat longer extenders have a readability advantage at normal sizes, but I concede that this is going to be hard to test. I think it is most likely to show up in the tests I have mentioned--either the one using the SAT or the frequency of proofing errors.

I do think that such tests--rather than under small sizes, low light, etc.--are also most likely to enable us to identify reliably the characteristics which make screen less readable than print, the effects of spacing, etc.

paragraph

You are right, bowerbird. All yours.

Kevin, if a single powerful vendor decided 2 point was the "new world order", you know as well as I do we'd just all have to design 2 point masters and fit up the masses with thick glasses (it sounds catchy-phrase-able already), until the monitor manufacturers could fit their products with thick glasses for the poor, while the rich would 'go nano' and not need glasses, (or fossil fuels;). Then...I believe 2 point text would perform better than 12 point text in all tests. Even I could design that study.

Bill>And the changed conditions affect readability under different circumstances, such as size, print medium, or computer screen.

I agree completely. This is why I think that tests in cleartype labs and optometrically designed studies are so far away.

In print, from the type's contrast to paper to the light the user employs, there is plenty of 'positive' wiggle room and designers and readers use it... On screens, e.g. it's only a little funny that the style is selected by one user, before the size, scaling and rendering are selected by the reader. Positive wiggle room is scarce, and we see that in the number of both complaints, studies on the topic, and reader flight to better platforms.

Fortunately, identifying reliably the characters who've make the major platform's screens less readable than print, is becoming easier.

Cheers!

Thanks, Kevin, for your gracious acknowledgment that the Arditi and Cho paper refutes the idea that all characteristics which work better under difficult reading conditions also work better under normal conditions. That's the true scientific spirit!

> In the case of agate proportions, I would be willing to bet money that a face with such proportions will not beat out in readability (measuring speed comprehension and comfort) a good text face like Minion in print at normal sizes and decent layout.

Swell, but there is no theory if every set of conditions is decided on a case by case basis. What is the theory that determines that agate proportions will perform well at small sizes, but not at larger sizes? The theory that legibility is separate from readability is unscientific (according to Popper) because there is no set of conditions that would show it to be false. It is useless because it can't predict anything.

> Kevin, if a single powerful vendor decided 2 point was the “new world order”, you know as well as I do we’d just all have to design 2 point masters and fit up the masses with thick glasses

David, you caught on to my master plan. Listen up people, buy your shares now in glass manufacturing companies!

“What is the theory…” [Kevin]

I'll propose one.

At thresholds simple discrimination affordance issues become the dominant and limiting factor, and these depend on 1) the detectability of distinctive features like closure, aspect and where role architectural components meet; 2) the difference in perceptual distances between lower case glyphs as compared to the perceptual distances between the comparable glyphs in upper case.

At peaks or performance plateaus issues surrounding the the rapidity and automaticity of ‘visual word-form resolution’ or ‘gestalt processing’ become the more decisive factor. Here gestalt integrity is probably the dimension around which performance varies.

The consequence of this is that adjusting the size of counters or the proportions of the letters relative to each other at agate sizes to enhance threshold-level performance might create problems with the rhythmic tuning of counters and the phase alignment beneficial to robust, rapid and automatic visual word-form resolution on the plateau or at the peaks. As well, the phasal structure of an all caps setting and the relatively greater salience of the whites in caps that an agate setting allows (compared to all lower case or conventional upper and lower case) might be beneficial to the subject's making out what's there at thresholds, but suboptimal for gestalt integrity at peaks or plateaus.

>The theory that legibility is separate from readability is unscientific (according to Popper) because there is no set of conditions that would show it to be false. It is useless because it can’t predict anything.

I didn't explain clearly enough. The general idea that there are differences between legibility--ability to identify letters under difficult conditions--and readability under good conditions is a general research program, and not a testable theory in itself.

However, specific claims of differences are testable and scientific. For example, the theory that bigger counters are helpful to identifying letters at very small visual sizes--helpful to legibility--but those bigger counters do not help readability (speed, comprehension, comfort) at larger sizes is testable. Thus the terminology of legibility and readability can be useful, as you can couch testable theories in it.

And, given Arditi and Cho, as well as typographic practice, the theory I have stated is very likely true, as well as testable.

As Popper always said, we don't want to be stuck on terminological debates, and the theories of what features work best in what situations are what are important, and not the terminology. The point is, the idea that all features that help reading in one situation help in every other has been refuted, as you say. The question is what's next.

Peter's discussion (which I think I understand) gives some ideas on how to further identify and test such features.

Kevin: I don’t take this as evidence that legibility and readability are distinct, because there was no finding here that could have falsified the hypothesis that legibility is distinct from readability.

Bill has put forward two different definitions of legibility and readability: one his own and one from Lieberman's Types of Typefaces. The latter is closer to my use of the terms, and how I believe most typographers reckon the difference. I believe this usage and Kevin's belief that legibility and readability are not distinct can be reconciled.

I would put it this way: legibility is the property of individual recognisability of letters, and readability is a hierarchical set of properties of text including both perceptual or input properties (recognisability of words, eye comfort) and cognitive or output properties (ease of comprehension, clear use of language, etc.). I suspect Kevin would simply say that what I call legibility is simply another part of the readability set of properties and operationally indistinct from the whole ‘decoding’ process. In his favour, it must be said, we have a word ‘reading’ but no parallel gerund for recognising individual letters and, of course, the etymology of the word legible is synonymous with readable. Further, since Kevin maintains that we recognise words by recognising their constituent letters, ‘leg-ing’ is simply part of reading.

Yet it remains the case that recognisability of individual letters -- within the superset of properties that constitute readability -- is distinct from other properties, and why shouldn't it have a name. Further, because the set of properties is hierarchical, each property relying on the lower properties, one can say of legibility, as the foundational property, that it is necessary but not sufficient for readability. In the same manner, one can say that word recognition is necessary but not sufficient, or indeed that any of the perceptual properties are necessary but not sufficient: it is perfectly possible to recognise all the words in a text and still not understand what the text means. [Yes, I think reading requires comprehension, and if we have not understood something -- even imperfectly or incorrectly -- then we have not really read. I know how to pronounce Latin, such that I can speak it aloud from a text and am good enough at recognising Latin words that I can sight sing them. But I can't read Latin, and need to refer to an English translation to understand what I am singing.]

The ‘not sufficient’ clause of legibility may be expressed this way: individual letters may be recognisable while not contributing positively to readability. This is easily demonstrated. These four letters are all legible: d r w o. But would a word set in this manner be readable? Would a page of such words? Letters of different sizes may be legible, but do different size letters make readable words when mixed together? [Building on an idea I put forward in my lecture at St Bride's, one can summarise all these distinctions by saying that letters of different spatial frequency may be legible, but will they create readable words when mixed together?]

It seems to me that in order for an operational distinction between legibility and readability to be ignored, it is necessary to assume consistency of letter proportions, style and weight, i.e. to assume the characteristics of a single typeface applied to, at minimum, words. Much of the time, this is a very reasonable assumption. But only because type designers have already absorbed and applied the assumption.

John, I agree that Lieberman's definition is better.

>one can say of legibility, as the foundational property, that it is necessary but not sufficient for readability

I would add that once a threshold of legibility--of letters--is met under a given condition, it may well be that increased legibility under other conditions is irrelevant to the reading process under that condition.

As you know, at my talk at TypeCon this year I also argued that considerations of evenness of color and rhythm, aside from legibility of individual letters, also contribute to readability.

And as I said, color and rhythm are important whether or not one accepts Peter's view that gestalt processes are at work in word recognition. (And I do suspect that Peter is right.) Frankly, I think that my slides--similar to your "word" illustration here, but more extended--made the case for the importance of even color and rhythm. And of course it doesn't have to be made to type designers, who have by and large been very sensitive to these issues.

Ps. I am curious as to when the distinction was made. Lieberman has on the same page as that definition photos of Beatrice Warde of Monotype and Paul Standard of Linotype, and his book is dedicated to them. He consulted them in the writing of the book, along with many others including a young James Mosley. I suspect that the distinction was introduced by Linotype folks in the 1930s, but I am guessing.

kentlew

Griffith and Dwiggins, corresponding in 1939, did not make the distinction -- certainly not terminologically. For the most part, I've seen the terms used pretty much interchangeably. I'm sure they would recognize the broad concepts being differentiated here, but I don't think they ever articulated them quite as we do today.

Bill: And of course it doesn’t have to be made to type designers, who have by and large been very sensitive to these issues.

Although I suspect, insofar as type designers think about why they are sensitive to evenness of colour and rhythm, they probably do so in aesthetic terms or, if they relate it to reading, they do so intuitively, anecdotally, or presumptively. This is why I am excited about Denis Pelli's spatial frequency study, because it suggests to me a strong perceptual bias for evenness of colour and rhythm: we will read more faster and more comfortably if we do not have to adjust to different spatial frequencies. That strikes me as a key insight into why we're sensitive to evenness of colour and rhythm, i.e. sensitive to maintaining consistent spatial frequency, within a single channel.

Christopher Tim...

@ John Hudson: Is this the study you are referring to?

Majaj, J., Pelli, D., Kurshan1, P. and Palomares, M. (2002). The role of spatial frequency channels in letter identification. Vision Research, 42(9). 1165-1184

(doi:10.1016/S0042-6989(02)00045-7)

Yes, that's the one. It helps to understand the basic concept of spatial frequencies before tackling that study, particularly the concept of spatial frequency channels, which provides an explanation for the easily testable phenomenon of perceptual adjustment when spatial frequency changes. See: http://www.yorku.ca/eye/spafreq.htm

JH> http://www.yorku.ca/eye/spafreq.htm

No saccades, no accolades.

KL>What is the theory that determines that agate proportions will perform well at small sizes, but not at larger sizes?

Well, in theory small sizes with agate proportions bring more light in just the right places to a reader.
And in theory large sizes with agate proportions bring more cost in just the wrong places to the publisher.
So, until 5 or so years ago, it was almost never done. No one complained that it should be. So, in theory, it shouldn't be done.

Just because people are now using agates large (out of desperation from causes unrelated to 'reading'), and there is no proof that it is inferior to use of properly designed large sizes, is it a bad theory Kevin? or just a bad situation being made worse by bad developers of bad font software for bad systems? being studied, of course, by 'good' behavioral psychologists? :)

Cjeers!

Okay, another Enne-Bomb.

I have grave difficulties with the terms of reference in these debates.

On the one hand, for Kevin, decoding is decoding. On the other hand, Bill and company refuse to seriously address issues of construct validity and quickly resort to easy operationalism.

[Kevin] “We use that light to recognize features, which we build up into recognized words. We read one word, then the next; our eyes move in predictable patterns. Environmental conditions can be good or bad, but the eyes do the same thing under any condition.”

The eyes might do the same thing under widely diverse environmental conditions, but does the visual cortex follow suit? The possibility exists that under good conditions the visual cortex “build[s] up features into recognized words” one way, and that it does so another way in poor conditions. I tried to contrast these two ways in my TypeCon Atlanta presentation. Denis Pelli draws a similar contrast in his “The uncrowded window of object recognition” when he says “some objects are recognized through a single combining of features over the whole object, whereas other objects require separate combining over each of several regions of the object.” His work demonstrates that both happen.

The relevance of this is that the distinction could anchor a stronger differentiation of readability and legibility, where legibility becomes associated with performance under conditions in which the latter (slot-processing, as I like to call it) is provoked, and readability becomes associated with performance under conditions in which the former (a single combining) is enabled. This might not be entirely satisfactory however, because it doesn't tell us which functional dimensions of perceptual processing in reading legibility and readability relates to.

I tried to lay a basis for specifying which functional dimensions are in play in legibility as distinct from readability in my earlier post. This post contrasted perceptual discrimination affordance with another kind of affordance which I associate with rapid automatic visual word-form resolution. Pelli draws a similar contrast when he differentiates between feature detection and feature integration. So then legibility becomes associated with performance under conditions in which the slot-processing is provoked and perceptual discrimination affordance comes into play; and readability becomes associated with performance under conditions in which the former (a single combining) is enabled and issues surrounding and extending the rapidity and ease of visual word-form resolution ( a single “transglyphic” gathering or combining) come into play.

This brings me to my ‘refusal to address’ complaint with Bill and company. Kevin raised the ‘construct validity’ question. Ole Lund in a Reading dissertation made it the key issue in Legibility research, and complain about the dependence on easy operationalism to make up for it. Bill, leaning on Popper, doesn't want to get into endless terminological debates, but in my opinion sustained reflection on the construct validity of the terms and a social contract is the only way to resolve terminological disputes. (I almost wrote ‘therminological’, I guess because these disputes get heated.)

Neither the avuncular Ben Lieberman definition or it’s Berksonian replacement are fully serviceable. Why? Because Uncle Ben’s definition doesn’t specify what in psychophysical functional terms eyes moving along a line, or telling one letter from another, or —most flagrantly — eyes absorbing messages are. This doesn't make the definition garbage. It just makes it unserviceable for eliciting informative interpretations of empirical data, no matter how it is operationalized.

Bill’s definition leaves unexplained whether or not the contrast between deciphering and reading is intended or not, and if intended, what in function perceptual processing terms distinguishes the two. Again, this doesn't make the definition garbage. It just makes it unserviceable for eliciting informative interpretations of empirical data, no matter how it is operationalized. What I find potentially useful in Bill’s replacement is the difficult versus optimal contrast, because I can relate them to different positions on the typical reading performance curve, which, no matter what the variable, typically show a steep rise in performance, than a wide plateau followed by a gradual decline.

So while I am sympathetic to Kevin’s complaints I am frustrated by the unquestionably attractive and ostensibly benign straightforwardness of his scheme. And while I am sympathetic to Bill and company’s insistence that the broad experiential sense of distinguishable dimensions (which has crystallized into language) is real and able to be investigated, I am frustrated by his willingness to leave them be floating signifiers in terms of what in psycho-physically tractable perceptual processing or functional affordance terms the one as distinct from the other has a bearing on.

What my provocations lead to in the context of the topic for this thread is a partial explanation of why we tolerate all caps on signage but are annoyed at them in extended text unless perhaps when they are used in the Bob, Bath comment to the story at the opening link.

>[Lieberman's] definition doesn’t specify what in psychophysical functional terms eyes moving along a line, or telling one letter from another, or —most flagrantly — eyes absorbing messages are.

I strongly disagree that this makes Lieberman's distinction between legibility and readability useless. On the contrary, when you load theory into it, as you seem to want to, you muddle the concept, and make it hard to understand. His definition is so well formulated that I regret departing from it. The idea that "legibility" applies to letters and "readability" to lines of words is a clear and understandable distinction, and serviceable for discussion and theorizing about type.

Whenever people create technical terms it causes a new barrier to communication. Sometimes it's worth it, but the meanings should be a clear as possible, and not loaded with theory. The more you load it the higher the barrier to communication.

Furthermore, I agree with Popper that theories, and not terminology are what is important. Classical physics can be formulated taking force and mass as primitive terms, or work and energy as primitive terms, if I remember rightly. That shows that terminology is not decisive. It is only helpful or not.

Furthermore terminology is never, never, never completely clear and precise. Max Jammer wrote books "The Concept of Mass" "The Concept of Force", etc., and if you read them the punch line is more or less that nobody knows really what these are.

I think Lieberman's distinction is helpful because it enables you to formulate what is to me the "fundamental principle of type design": legible letters are not enough to make highly readable text. There's more to it. There's at least color and rhythm, and probably the gestalt effects you have theorized about.

Bill, I wrote: "This doesn’t make the definition garbage. It just makes it unserviceable for eliciting informative interpretations of empirical data, no matter how it is operationalized."

I could have said "fully informative."

Furthermore, I agree that Lieberman’s distinction might be helpful insofar as it associates legibility with letters and readability with words, but I need something more than 'ease' and 'identification,' or 'ease' and 'absorption' to tell me what functional factor performance measures. I want to understand why and in what sense it makes sense to say legible letters are not enough to make readable text. Clearly it is in any normal sense. Legible letters are enough: they make readily readable words. What sub-processes does the eye absorbing messages consist of? And what part of that does or should a test of readability measure? What does telling one letter from the other consist of, and what part of that does or should a test of legibility measure.

I can't believe that you have no sympathy for my hesitations.

Adding some further thoughts.

Your alternative allows one to say that what is good for deciphering or decoding or at thresholds -- a function signage must accommodate -- is not necessarily good for continuous reading under good conditions.

All I'm trying to say is that to articulate a theory of perceptual processing in reading that is able to correct or underwrite and deepen or extend our understanding of what we say or do, it is not enough to stay with what we are prone to say in our ordinary everyday engagement with these things using words and metaphors that are ready to hand and modally non-specific. Yet what we are prone to say can be an excellent place to start.

Fundamental principles don't always provide us with conjectures that support effective refutation in their native forms, as Kevin insists

Peter, the specific hypotheses I discussed with Kevin are refutable, as I said. And I explained the experiments to test readability.

I think the demarcation between identifying letters and reading words in text is useful, but I'm ready to drop it in an instant if it's not helpful. It's not important. If you want to separate different aspects of readability, such as speed, comprehension, comfort--or different aspects of comfort--that is fine. There is no need to have a precise definition of readability, or legibility. There is a need for better theories of how all these factors work in reading.

I think my disagreement with you about scientific method is over how much should be stuffed into terminology, and how much should be put in theory. I am saying that the ideas should should be put into theories. Theories are statements, and can be testable. Terminology is never testable, only useful or not.

Theories are statements connecting different terms, as in "mass = force x acceleration." The sciences that have grown most successfully focus on testing the statements, not inventing or defining new terminology. I think that if physics spent its time on first getting the definitions of mass, force, and acceleration right, there would still be no physics. Instead there would be a mass of terminological disputes, with little progress.

I know that Popper's advice in this situation would be to focus on creating theories to solve the problems of how we read. His view was that new terminology should be something that is forced on you to express your explanatory theories, not something you seek as a way to the truth. Pursued directly, they are a path to the swamp, not to the truth.

Bill, what in the process of reading as described by Kevin or Pelli or myself do your indicators of readability or legibility pick up on? Are they different for readability and legibility?

You misunderstand what impels me and how I proceed. In my forays into the literature I feel the terms that are often used to elucidate a reality I am interested in are too crude to express what I -- having read about the way the visual cortex and neural transmission works, and having been sensitized by philosophical critiques of representationalism and the rest -- see going in. So I import and adapt possibly exotic but -- to me at least -- transparent terms I have found elswhere in my hunting and gathering, and think might be applicable. Here I am doing nothing unusual in the scientific domain.

Peter, I think I do understand your frustration with existing literature on reading. Since you have interesting ideas, I can't say you have gone into a swamp, even though at times I feel your language sometimes has gone there. I don't want to belabor this, but I would just say two things: first, seeking precision in terminology can become an endless quest and a distraction from finding testable theories. Second, it is hard to communicate new ideas in any case, and new terminology can be an additional obstacle to communication. If it is possible to keep new terms to a minimum, it promotes communication. And communication is actually quite important.

Because I don't think terminology is very important to the content of a scientific theory, I don't want to put up a big defense of the distinction readability vs legibility. But to answer your question: on readability, I have said this involves speed, comprehension, and comfort. I think all of those are in the psychological literature. Legibility is different because it is, as defined by Lieberman and understood also by Hudson, a measure of ability to identify individual letters. Such tests of letter recognition are also in the literature.

In fact, Arditi & Cho tests began with legibility tests, and then switch to readability tests. To me these categories are helpful in thinking about what is going on in these experiments and others.

Also I would also add that because my introduction to type was the book Types of Typefaces I have been more ready to listen to your theories of how reading words differs from letter recognition. Perhaps you think that a benefit, and perhaps not :)

Peter: I want to understand why and in what sense it makes sense to say legible letters are not enough to make readable text. ... Legible letters are enough: they make readily readable words.

Readability is qualitative (and also measurable in terms of speed, comprehension and comfort), so saying that something is ‘readable’ isn't saying much unless you qualify the observation to indicate how readable, on a continuum from minimally readable to optimally readable (for a specific condition). Legible letters are enough to guarantee that text is minimally readable. Consider the classic ransom note made up of letters clipped from different newspaper headlines, in different types, sizes, styles and colours and pasted on a page: all the letters are legible and, of course, the message is readable (it has to be, otherwise junior's gonna get his ear cut off). But we don't typeset books like this, and not only because it takes too long and uses too much glue.

As I wrote above, all the discussion about word recognition being derived directly from letter recognition presupposes, so far as I can tell, a single typeface set in a single style and size. In other words, the model presupposes factors that Bill has identified -- colour (what I prefer to call texture) and rhythm -- and, I would argue, consistent spatial frequency, which are characteristics of good type design. It is easy to say that legible letters result in readable words -- even very readable words -- if the other factors of typeface design are a given. If you remove those other factors -- mess with spacing, mess with different typefaces, mess with different weights and styles, transcend spatial frequency channels within a word --, reducing the components of words to merely legible letters, you'll pretty quickly see 'why and in what sense it makes sense to say legible letters are not enough to make readable text'.

>Readability is qualitative (and also measurable in terms of speed, comprehension and comfort)

Show me a single study of one font vs. another, believable after even the simplest typographic vetting.

>[C]onsistent [S]patial [F]requency, which are characteristics of [g]ood [t]ype [d]esign

CSF is now more than ever, a characteristic of output that can absolutely override gtd.

And, since all this talk is aimed at screen readability...

Cheers!

Bill, my issue isn't terminology, but construct validity. John, when I say “I want to understand…” I mean I want to know what specific, single item, perceptual processing operator or primitive is at stake when questions of legibility or readability are raised.

Let me explain. Readability and legibility have a specialized meaning in typographic discourse as compared with plain language. Ben Lieberman’s definition is couched in the language of everyday ordinary lived experience, which depends on experientially tractable wholes like identification, ease, line following and message absorption. This makes the definitions palpable and the terms potentially useful as an evaluative category but not necessarily suitable as a fertile term for building greater understanding. Letter-identification, line-following and message-absorption are complex acts, and the task and beauty of science is it’s ability to break these down.

Let us say that letter identification involves feature detection, role-unit discrimination and some kind of combining or feature integration. When issues of legibility are on the table, I feel it is enriching / clarifying / disambiguating and productive to put forward that all or some of the items in this complex are under the loupe. And that a different complex of items or primitives are under the loupe when issues of readability are on the table. I feel this is a better strategy than just breaking down readability and legibility into matters of comfort, speed and comprehension, as signitors of ease.

John, following up on David's comment, it seems more informative to me with type to talk about narrow phase alignment, than having a consistent spatial frequency. You seem to associate the latter with weight. I'm not following here. I associate the former with distribution of role-unit level information. Weight I associate with salience and cue value.

Bill, to phrase my frustration another way:
I ask, what is legibility about.
You say ease, which equals speed + comprehension + comfort.
I say, ease of what?
You say, identification or deciphering.
I say, yes, but do you mean by these the gathering, streaming and assembly of the criterial visual information, which I think has to be, following Pelli and others, assessed at the role unit or distinctive feature level?
You say, maybe, but let's not get into that to deeply because it gets confusing / it's a swamp / others will recoil / 'deciphering' or 'identification' is clear enough for me.
I say, 'deciphering' and 'identification' and 'absorbing' are the swamp. Let's jump in and see what new things we can turn up.
You say, go ahead, but don't shout back to me about micro-processes and microorganisms, because I think they're microorganisms are ugly and micro-processes hard to visualize.
I say, sorry, but that's just how it is and this is why I'm here. It would be easier if you jump in which, to some extent you and others have begun to do.

I find it easier to work with the definition that legibility is the ease with which letters can be absorbed and readability is the ease with which sentences (or any larger amount of text) can be absorbed.

John proposed that there are effects of readability that are independent of legibility, such as consistency of spacing and consistency of letterform.

Therefore, it is possible to have good legibility, but poor readability by badly spacing well designed letters, or by combining well designed letters of different typefaces.

Is it possible to have the converse: poor legibility but good readability? Or is legibility a subset of readability, so letters with poor legibility will always also have poor readability?

Are lowercase letters an example of poor legibility but good readability? Ignoring the Arditi & Cho findings, uppercase letters perform better in letter recognition tests, while lowercase letters perform better in sentence or passage reading.

If lowercase letters are an example of poor legibility but good readability, is it because lowercase text tends to have better spacing or letter design consistency than uppercase?

Peter, I have no problem with discussing and testing how sub-letter features operate in reading. I'm all in favor of it. I'm afraid I haven't successfully communicated my concerns about your approach, and the role of terminology in it.

In any case, the key problem is not terminology, but lack of tests. I've suggested some tests of your theory, and that direction is the only one that will advance your theory within science, IMHO. That's the bottom line. Lay out the tests, and your theory will be clear enough for science, however lucid or awkward the terminology.

>John proposed.

As did I extensively at my presentation at your panel TypeCon, right? Here is a key slide:

The point is that all of the letters in the "legible letters" line are very readable when they are with their "mates" in style and weight, but as is the two words are barely readable. It was interesting to me that mixing both case and italic/regular, as well as mixing weights, results in such a dramatic deterioration in readability--much more than mixed case alone.

>If lowercase letters are an example of poor legibility but good readability, is it because lowercase text tends to have better spacing or letter design consistency than uppercase?

Well, it's not "poor" legibility, but sub-optimal. A threshold has to be reached, I believe, before reading can take place. I think you now understand the problem that Peter has been wrestling with, and what all we type people have been talking about.

And yes, lower case has both more even color and better spacing than upper case. If you look at lower case, most of the letters have 'four corner' adjustment of black within the x-height, so that more even color is possible. Eg compare T and t. The cap t inevitably gets more "holes" that can only partly be compensated for by spacing. In addition ascenders and descenders I think increase ease of recognition of word image--a gestalt effect.

Peter in particular has argued that letter recognition is worse in normal spacing of lower case, because of crowding. But that closeness is actually better for getting the word image, the gestalt formed by a collection of interpreted sub-letter features across the word. Thus words and letter combinations read better and single letters worse. That accounts for the Word Superiority Effect.

And the experiments I suggested to you were designed to test that in more revealing way than previous tests. The apparatus unfortunately did not allow you to get the short exposure times of the original experiments of Purcell, and so were not dispositive on the issues.

If they are repeated with the short times, I think they will be more revealing. I can discuss this at more length with you.

I have also suggested an experiment using the WSE and missing letter parts which will be even more revealing, I think. This was in part of my TypeCon talk that I didn't get to because of lack of time.

Separating the benefits for readability of only good spacing and even color from the gestalt effects is something I haven't thought about yet, but that should be doable also.

David: Show me a single study of one font vs. another, believable after even the simplest typographic vetting.

Sure: I sat down and read one book and then I sat down and read another book, and I found one easier to read than the other and got through it quicker. This is what I mean when I say that readability is qualitative and measurable in terms of speed, comprehension and comfort. This isn't a conclusion based on laboratory studies; it is a quotidian observation based on doing reading, not studying reading.

And, since all this talk is aimed at screen readability...

It is? I suppose Kevin's engagement might be, because that's his job, but I don't see anything in what Bill, Peter or I have written that indicates that this talk is aimed at screen readability or, indeed, aimed at readability at all. I believe it is aimed at understanding how reading works and being able to describe that in meaningfully precise terms. Readability is, I argue, a set of properties of text, and hence largely irrelevant to understanding reading as an activity of eyes and brains. As Peter memorably wrote on another occasion: ‘trying to construct a science of readability might be like calling physics the science of the fall-down-ability of falling objects’. Readability is only relevant to understanding reading insofar as empirical testing must try to isolate and control readability properties.

Kevin: Is it possible to have the converse: poor legibility but good readability? Or is legibility a subset of readability, so letters with poor legibility will always also have poor readability?

Are lowercase letters an example of poor legibility but good readability?

I agree with Bill: a baseline of legibility is necessary to readability. This baseline can probably be defined as the distinction between illegible and legible letters, with the caveat that if you have one or two illegible letters but put them in context of legible letters the results will be readable on the basis of context. If most or all of the letters are illegible, readability is defenestrated.

I wouldn't characterise lowercase letters as having ‘poor’ legibility. Obviously they have ‘good enough’ legibility: writing systems don't tend to develop poorly legible shapes, and if letter recognition tests indicate that particular letters are very difficult to identify that would suggest to me that identification of those isolated letters is not very important to reading that writing system.

Re. spatial frequency

My understanding is that spatial frequency involves both quantity of periodic changes across space and quality of the transitions between those changes. So yes, it relates to weight, because different weights of type have different spatial frequencies due to different arrangements of dark and light shapes within the same area. But it also relates to what I call stroke density, and the resulting quality of the transitions between dark and light shapes, and this is an area in which type on screen differs significantly from type on paper and, indeed, in which different kinds of rendering on screen differ significantly from each other, and how different monitor gammas make type in the same rendering differ, etc. However, insofar as within any one of these conditions the letters have consistent spatial frequency, spatial frequency can be isolated as a factor.

I should also clarify that when I talk about ‘consistent spatial frequency’, what I mean is restrained within a single spatial frequency channel. In other words, the reader doesn't need to change channels in order to be able to perceive different letters in a word. Within this channel, there is obviously quite a lot of room for variation of letterforms -- enough to allow of all these typefaces -- so consistency in this case does not imply uniformity.

Kevin>What is the theory that determines that agate proportions will perform well at small sizes, but not at larger sizes?

Your organization did a study where Verdana was compared to Arial. What was publicized, is that the people studied preferred Verdana. I encouraged you to investigate all the results, which showed that the Verdana preference faded as the size increased until Arial was preferred.

Finish, and perhaps you will have a theory, if you want one.

KL>Is it possible to have the converse: poor legibility but good readability?

Yes, the study that showed helvetica e more 'readable' than the garamond e 'proved' that beyond a doubt. ;)

KL>Or is legibility a subset of readability, so letters with poor legibility will always also have poor readability?

I think it depends on the poorness of the legibility and whether or not it can be hidden, (like the garamond e), in familiar ways.
But in general, poor legibility in letters will always yield poor readability because those who choose, or are condemned to fonts of poor legibility will not likely be 'super-typographers.'

DB>And, since all this talk is aimed at screen readability...

JH> It is? I suppose Kevin’s engagement might be, because that’s his job, but I don’t see anything in what Bill, Peter or I have written that indicates that this talk is aimed at screen readability...

>...and this is an area in which type on screen differs significantly from type on paper and, indeed, in which different kinds of rendering on screen differ significantly from each other...

How about now? Can I talk about crappy anti-aliasing scaling now?

>...the reader [shouldn't] need to change channels in order to be able to perceive different letters in a word.

... then why make them? ;-)

>Sure: I sat down and read one book and then I sat down and read another book, and I found one easier to read than the other and got through it quicker.

Ahh, but you read two different books. And how can you sit down twice without standing up. Show me a STUDY. I can tell you tales all day of people who can't, won't or don't read their screens John, because of spatial frequency terror. But if that's not good enough, what do we care about your book reading tales?

Cheers!

A study that shows what, David? That readability is qualitative? That we can speak meaningfully about a text being more or less readable than another text? That we reckon readability in terms of speed, comprehension (also retention) and comfort? This is all I asserted, and I challenge you to show me a study that doesn’t show this or, indeed, doesn't presuppose this, based on common sense and everyday experience. You don't need a study to prove that when we read we use our eyes and our brain. Some things are obvious to casual observation, and the fact that some things are easier or harder to read than others is one of them.

Studies may or may not have something to tell us about why some things are easier or harder to read than others -- and I agree with you about the typical faults of reading studies --, but I wasn't making any comment on that question.

David: How about now? Can I talk about crappy anti-aliasing scaling now?

Only if you take into account what else I wrote: ‘However, insofar as within any one of these conditions the letters have consistent spatial frequency, spatial frequency can be isolated as a factor.’ In other words, the fact that spatial frequency varies across reading conditions is irrelevant to the impact of change in spatial frequency so long as it is consistent within a single reading condition, whether that condition is deemed positive or negative. Crappy anti-aliasing scaling is a (presumably negative) reading condition.

Now, independent of the impact of changes in spatial frequency on reading, there is the separate question of whether specific spatial frequencies are better or worse for reading. And in looking at this question, you need to isolate different factors to determine the specific effect of changing spatial frequency within a particularly reading condition. This is where you can start talking about crappy anti-aliasing, as resulting in a specific spatial frequency. What can we say about it? Well, it is a relatively low spatial frequency, because we're losing sharp edge contrast. Is it worse for reading than a higher spatial frequency and sharp edge contrast? Is there a point at which edge contrast becomes too sharp, at which reading benefits fall away? Or is the sharpest edge contrast permitted by the medium always to be preferred, all else being equal?

And just try getting everything else equal!

> the reader doesn’t need to change channels in order to be able to perceive different letters in a word

I was reacting to this. The reader should not need to change channels in order to be able to perceive different letters in a word. It's saccadicide.

>This is where you can start talking about crappy anti-aliasing,

And I lump this together with misuse of extreme sized masters. So when one does both... and studies readability without making reference to these conditions, then what?

>And just try getting everything else equal!

That is what some of us do.

Reaching back to the comedy of definitions,

>writing systems don’t tend to develop poorly legible shapes,

The I's have it. Think about those definitions relative to sans, Ills.

Cheers!

David: I was reacting to this. The reader should not need to change channels in order to be able to perceive different letters in a word. It’s saccadicide.

Yes. We are agreeing. The point I was making is this: it is not enough for individual letters to be legible in order for text to be readable: the letters also have to work together and one aspect of this is achieving consistent spatial frequency. This is par for the course in type design.

Think about those definitions relative to sans, Ills.

I have thought about it: every style of writing the Latin script prior to the modern sans serif typographic letter made a visual distinction between I and l. The writing system didn't have a problem with this: only some type designers did.

>The writing system didn’t have a problem with this: only some type designers did.

And today?

Cheers!

Christopher Tim...

I've been off the grid in the woods for the last week or so and have fallen behind in this thread. As I continue to read read I notice references to Tinker and his studies.

Which specific studies are we referring to?

I am in the process of reviewing:

Tinker, Miles A. (1965). Bases for effective reading. Minneapolis: University of Minnesota press.

(for fun, here's what I get to see what I wake up in the morning at the camp)

Christopher Tim...

@ Moderators: Is there a formal way to re-direct conversation from a thread that goes somewhat off topic? For example:

Let's continue this conversation of in another thread: CAPITAL LETTERS versus Upper and lower case.

David: And today?

Most recent sans serif types I've looked at seem to make some effort to distinguish I and l, either through height difference, or distinct terminals, or some combination.

joeclark

> writing systems don’t tend to develop poorly legible shapes

I saw you at ATypI Vancouver tell us in the audience that you tweaked individual pixels in your Hebrew fonts to make the roughly two dozen letters that look like rectangles vaguely less illegible.

Oh, and? Chinese much?

You have used up your annual allotment of “What was I thinking?” statements.

—
Joe Clark
http://joeclark.org/

Joe, the challenges to legibility in low resolution environments, which may affect one script more than another, are generally not anticipated in the development of writing systems. Also note that in the full comment from which you quoted, I contrasted ‘poorly’ with ‘good enough’. I stand by the comment, because I do believe that writing systems are intended to be read, and are read, and this implies that the shapes used within those writing systems are indeed ‘good enough’ for reading, presuming that they are legibly made. And that's the kicker: legibility of forms within a writing system means something like ‘ability to be made in a legible way that is good enough for reading’, not inherently legible or illegible. This leaves the onus on writers and type designers to interpret the normative forms in ways that are more or less legible. And this is what we do: we don't invent new letters in order to be more legible, we interpret the existing letters. This leaves the door open to styles of writing that are less legible than others, some deliberately so.

Christopher, Typophile has followed the Talmudic model: if a more interesting topic comes up in the course of a discussion, that new topic is followed, and recorded. That's fine with me, because I love the Talmud. The main drawback might be that the topic title is misleading, but in the case of the internet, everything is indexed anyway, so the topic title actually has no special significance in search.

Also the new topic is not capital letters vs lower case, but legibility and readability. So why not leave well enough alone? What is gained by breaking the thread? I don't see it. And what is lost is something: the continuous narrative findable in one search.

At any rate, I wanted to clarify that there are several different issues here. One question is how legibility of letters in difficult conditions relates to their legibility in optimal reading conditions. It does seem that this is different. Another issue is how the identification of letters under optimal conditions relates to the identification of words. Neither Lieberman's definition nor my earlier one above--not so good--really captures that there are two issues here.

On the issue of what more there is to reading than identifying letters, I mentioned in my TypeCon talk three things: good rhythm, even color, and the formation of a word image or "gestalt" or "bouma". I do think it should be pretty easy to show that uneven spacing eg destroys the Word Superiority Effect. But this does not yet prove that "gestalt" effects are happening. This is because what may be critical is just that the letters fit into a grid. The brain may impose a grid, and look for letters within it. So the disrupted spacing may just affect our ability to identify letters.

Still, is might be possible to separate the two effects by seeing whether it is significantly more difficult to identify letters in unevenly spaced non-words vs evenly spaced non-words. When testing is done in all variations for both words and non-words, I think the result would be interesting.

I am also struck by how deadly mixing italic and roman is to readability. Testing the effects with words and non-words would also be interesting as to what it says about our mental grid or matrix.

>Most recent sans serif types [...] make some effort to distinguish I and l, either through height difference, or distinct terminals, or some combination.

Yes they do. But if I had to add an extra nose to be more easily recognized, I'd pass.

The true features of I and l that distinguish them in their character groups, are the slight exaggerations to stem thickness and spacing, relative to their nearest kin, not characters from distant groups. The writing systems can definitely handle this, (to prevent what MS calls, "the BIll Hill problem") But because of other closely related problems, they don't. So now... we've got simplified Latin, i.e. minimal information designed into the font for the reader, poor scaling and not enough pixels.

And, the challenges to legibility in low resolution environments, though not specifically anticipated in the development of writing systems, are...in the form of grids upon which most classically love fonts are plotted. But even these are defeated by the one thing no writing system could ever anticipate, a lack of competition leading to a constraint of development and trade leading to what we see today on most people's screens.

Cheers!

Russell McGorman

Most recent sans serif types [...] make some effort to distinguish I and l, either through height difference, or distinct terminals, or some combination.

funnily enough, I recently came across a sign where the font in use had "l" with a curved tail to disambiguate lc "l" from uc "I", where a double "ll" could be confused with lowercase "tt". (And was by me) Admittedly when when set too tight and viewed in poor light while in a hurry… and before a first cup of coffee.

-=®=-

> I do think it should be pretty easy to show that uneven spacing eg destroys the Word Superiority Effect.

Bill, in the study that you and Peter proposed, but decided to not discuss at TypeCon, the data showed that there was a statistically reliable Word Superiority Effect for alternating case words. That data doesn’t directly falsify your above statement, but does falsify any claim that lack of rhythm destroys the Word Superiority Effect.

If someone found that a Word Superiority Effect was still there with uneven spacing, would that falsify your above claim?

Kevin, in the experiments graciously you did, the times of exposure were too large to replicate Purcell's findings. In our communication with Purcell, he anticipated problems with the experimental set up you were using, as we related to you. In your experimental set up, with the longer stimulus times, you could not get a normal WSE except with letters as masks, rather than the overlaid Xs and Os of Purcell.

Our hypothesis concerned two levels of WSE, one at shorter times primarily due to visual effects and the second cognitive, based on an influence of word memory. The fact that the exposure times were longer, and that letter masks interfered more with stimulus identification made it plausible that memory was involved. Since no shorter times were involved, our hypothesis wasn't really tested.

Since our predictions involved longer and shorter exposure times for the exposure of stimulus, and your experimental set up did not get those, I felt it wasn't a valid test of our hypothesis. If you replicate this with a set up that can get the longer and shorter times, and careful control of the SOAs (time gap between stimulus and mask),it would be valid, and either refute or corroborate our hypothesis. Because your experimental set up did not really test what we looking for--the set-up first needed to replicate Purcell, and didn't--I felt it wasn't worth explaining the whole thing at TypeCon.

Presence of a WSE with mixed case that you found under your conditions has as you know been been found before. --In fact our hypothesis is an alternative explanation for that effect. And it does not in itself refute "any claim that lack of rhythm destroys the Word Superiority Effect." We were only using three letter stimuli, and no disrupted spacing, which still gives a pretty even look. If the s pa ci n g were like this, with longer words, I do expect there would be a destruction of the WSE, such as Purcell found a destruction of the WSE with very wide spacing of letters.

Further, with timed tests with comprehension, rather than a WSE test, I think it will be possible to show that disrupted spacing slows down reading--ie will hurt readability.

I should add that I don't think that reading test by itself it would test Peter's theory of gestalt effects. It would, though, test the idea that letters falling near a regular rhythm is important to readability.