David Březina, Mary Dyson

Can we selectively attend to the top halves of letters and ignore the bottom halves?

Background to the study

We presented this work at the ICTVC 8 conference in Thessaloniki in July 2022. A primary aim of the online study was to find out if letters are processed holistically. This would mean that we attend to all parts of a letter at the same time, we cannot selectively attend, and we are sensitive to spatial relations among parts. Research on face perception has demonstrated that faces are perceived holistically (e.g. Farah, Wilson, Drain, & Tanaka, 1998).

A second motivation for the study was to see whether designers differ from non-designers in how they process letters, holistically or as separate features. Research with art students has found less holistic processing of faces than ordinary observers (Zhou, Cheng, Zang & Wong, 2012). This is explained by the art students’ additional experience in drawing faces and attending to parts of a face. In a similar manner, letter designers (e.g. typeface designers, calligraphers, lettering artists) attend to parts of letters. We therefore predicted that designers, and in particular letter designers, might not process letters holistically, whereas non-designers would process letters holistically.

What we did


We asked participants in the study to categorise themselves as a non-designer or a graphic designer, typographer, letter designer, or other designer.

Participants were asked to look at a series of two letters, presented one after the other, with a pattern mask in between. The task was to say whether the top halves of the two letters were identical. The response options were ‘Sure same’, ‘Probably same’, ‘Probably different’, ‘Sure different’. To avoid any ambiguity as to what was meant by the top half, there was a horizontal line through the middle of the letters. See Figure 1.

Two letters where the top halves are different, and the bottom halves are also different.

Figure 1: Two letters where the top halves are different, and the bottom halves are also different.

The letters

A variable font was designed that varied contrast (low to high) and terminals (simple cut-off terminals to flared to serifs). Letters were composed of two halves and designed so the tops and the bottoms of all combinations connect smoothly in the middle. They were designed as monospaced letters for consistency of presentation in the task. See Figure 2.

Overview of the letter variants

Figure 2: Reading across the rows, the top halves of the letters remain the same and the bottom changes. Reading down the columns, the top halves of the letters change, and the bottom halves remain the same. The diagonal, from top left to bottom right, contains letters where the variation is applied to both the top and bottom of the letter. These are labelled ‘normal’ letters in contrast to the ‘composite’ letters where the top and bottom come from different versions of the normal letters. Highlighted in grey are letters that were used to create the pairs described in figures 3a–d below.

To test for holistic processing, a comparison is made between two conditions:

  1. pairs of letters where the relationship between the two bottom halves of the letters is inconsistent with the relationship between the top halves (incongruent letter pairs)

  2. pairs of letters where there is a consistent relationship (congruent letter pairs)

The four categories of letter pairs are shown in Figures 3a–3d:

Two ‘same’ congruent pairs (top halves identical and bottom
halves identical)

Figure 3a: Two ‘same’ congruent pairs (top halves identical and bottom halves identical)

Two ‘different’ congruent pairs (top halves and bottom halves

Figure 3b: Two ‘different’ congruent pairs (top halves and bottom halves different)

One ‘same’ incongruent pair (top halves identical but bottom
halves different)

Figure 3c: One ‘same’ incongruent pair (top halves identical but bottom halves different)

One ‘different’ incongruent pair (top halves different but
bottom halves the same)

Figure 3d: One ‘different’ incongruent pair (top halves different but bottom halves the same)


The study included a practice set of trials followed by the main session. The practice was included to familiarise participants with the procedure and to set the pace of presenting the letters (presentation rate). It was important to ensure that the task was neither too easy nor too difficult. By piloting the study, we realised that there might be quite big differences in the level of difficulty between designers and non-designers; the latter are less used to discriminating between fonts.

We included 16 practice trials using letters “d”, “u”, and “i” where we varied the rate at which we presented the letters starting at 800 msecs, then increasing the rate in steps to 200 msecs and then back to 800 msecs. The proportion of correct responses was calculated, and this was used to set a consistent rate for the main part of the study. The main part used a different set of letters: “a, b, c, e, h, j, l, n, p, q, r, t, z”.

Figure 4: An example trial from the study. Click on arrows to proceed. Participants were asked to look at a series of two letters, presented one after the other, with a pattern mask in between. The transition was automatic after a defined period of time for each letter and mask. The response options were shown, and the letters hidden, after both letters had been shown. The task was to say whether the top halves of the two letters were identical.
Figure 5: All letters and their variants used in the study.


A total of 229 people completed the online study and were broken down into:

Non-designers 118
Graphic designers 47
Letter designers 36
Typographers 7
Other designers 21


We have made the study website and data available on GitHub and licensed them under a Creative Commons Attribution-ShareAlike 4.0 International License (CC BY-SA 4.0).

What we found

Holistic processing

The ability to discriminate between same and different pairs is measured using AUC which is free of response bias, such as responding ‘same’ most of the time. Overall, the congruent pairs were easier to discriminate than the incongruent pairs. This means that the letters were processed holistically.

The results broken down by participant group are shown in Figure 6. Although the size of the difference between congruent and incongruent pairs varies across groups, the only statistically significant differences are between:

In both cases, there is a larger difference between congruent and incongruent pairs in other designers, i.e. stronger evidence of holistic processing in other designers. There is larger variation among other designers with the incongruent pairs which suggests that some participants may have found this condition particularly challenging.

Although typographers appear not to process letters holistically (similar discrimination of congruent and incongruent pairs), this result is not reliably different from the other groups because there are so few participants who categorised themselves as typographers (N=7).

The difference between congruent and incongruent pairs for all of the designer groups is not significantly smaller than the difference for non-designers. Designers generally process letters in a similar manner to non-designers, holistically, rather than attending to parts of the letter.

Figure 6: Mean AUC scores for each group of participants, comparing discrimination between same and different letters in congruent and incongruent pairs. A higher score for congruent pairs compared with incongruent indicates holistic processing. The standard error bars represent the amount of variation among participants.

Further comparisons across groups of participants

Although we aimed to achieve a similar level of performance in designers and non-designers, by adjusting the presentation rate of the letters, we did find that graphic designers, letter designers, and typographers were better at detecting the differences than non-designers. Letter designers were better than graphic designers and other designers. See Figure 7.

Figure 7: Overall AUC scores for each group of participants, combining congruent and incongruent letter pairs.

The presentation rate of the letters can also be compared across groups (Figure 8). The non-designers received longer presentation times than graphic designers and letter designers. This is expected as they would be likely to get fewer correct responses during the practice trials. Other designers also received longer presentation times than letter designers.

Figure 8: Presentation rates in miliseconds (based on practice trials) for each group of participants.

Possible explanations

We found evidence for holistic processing of letters in non-designers and designers. The responses of other designers appear to be rather different from the rest of the designer groups. But as we do not have any further information on the characteristics of this group, it is difficult to explain the results.

We may conclude that designers and non-designers process letters in the same way. However, an alternative interpretation is that, although the outcomes are the same, they are achieved by using different strategies. As experts in the analysis of letters, letter/typographic/graphic designers may have focused on the spatial relationships between letter features, e.g. the relative position of thick and thin strokes. As experts in letter perception, non-designers may have viewed the whole letter, as in reading, because it is a familiar shape.

Future work may explore whether changes to the nature of the task reveal differences in strategy.

What did you think?

What did you think of the article? We would sincerely appreciate your feedback.

Send a comment

Enjoyed the article?

Sign up for our newsletter and get notified when we publish the next one.


Farah, M. J., Wilson, K. D., Drain, M., & Tanaka, J. N. (1998). What is “special” about face perception? Psychological Review, 105(3), 482–498.

Zhou, G. M., Cheng, Z. J., Zhang, X. D., & Wong, A. C. N. (2012). Smaller holistic processing of faces associated with face drawing experience. Psychonomic Bulletin & Review, 19(2), 157–62.

More reading

See all articles