Lack of regularity between letters impacts word recognition performance

Physical inter-letter dissimilarity has been suggested as a solution to increase perceptual differences between letter shapes and hence a solution to improve reading performance. However, the deleterious effects of font tuning suggest that low inter-letter regularity (due to the enhancement of specific letter features to make them more differentiable) may impair word recognition performance. The aim of the present investigation was 1) to validate our hypothesis that reducing inter-letter regularity impairs reading performance, as suggested by font tuning, and 2) to test whether some forms of non-regularities could impair visual word recognition more. To do so, we designed four new fonts. For each font we induced one type of increased perceptual difference: for the first font, the letters have longer extender length; for the second font, the letters have different slants; and for the third font, the letters have different font cases. We also designed a fourth font where letters differ on all three aspects (worst regularity across letters). Word recognition performance was measured for each of the four fonts in comparison to a traditional sans serif font (best regularity across letters) through a lexical decision task. Results showed a significant decrease in word recognition performance only for the fonts with mixed-case letters, suggesting that fonts with low regularity, such as mixed-case letters, should be avoided in the definition of new “optimal” fonts. Letter recognition performance measured for the five different fonts through a trigram recognition task showed that this effect is not consistently due to poor letter identification.


Introduction
45 The often repeated saying among typographers that "type is a beautiful group of letters, not a 46 group of beautiful letters" (1), suggests that it is only when letters work as a group that they 47 become type, a visual characteristic that we name "inter-letter regularity". To achieve this, a 48 basic principles of sign painting and font design dictates that fonts and lettering shall be based 49 on a repetition of shapes with the aim of ensuring harmony and balance between the letters 50 (2, 3) (Fig 1). This means that all lower-and uppercase letters originate in two different 51 modular systems that put together constitute the alphabet (one for lowercase letters and one 52 for uppercase letters) (4). Such an approach naturally leads to letters of relatively similar 53 shapes (and high regularity). By contrast, it has often been proposed that greater letter 54 distinctiveness, where new features are added to selected letters, could facilitate reading, as it 55 minimizes the risk of letter confusion (5-7). However, greater letter distinctiveness also 56 decreases inter-letter regularity. To investigate whether high letter differentiation could improve peripheral reading, 61 Bernard et al. (7) created a new font, referred to as Eido (Fig 2). They found that while 63 both letter and word recognition, although sentence reading speed was not significantly 64 improved. Xiong et al. (8) further found that Eido outperformed both Helvetica and Times 65 Roman for reading acuity performance, while maximum reading speed was not significantly 66 improved. Also interested in letter differentiation, Beier and Larson (9) measured letter 67 recognition of variations within the same font family and found certain letter shapes of 68 greater dissimilarity to facilitate better single letter recognition than others.

73
The absence of regularity in the Eido font (Fig 2) 163 The subject was asked to fixate on a central dot while words or pseudowords were 164 randomly presented at 10° in the lower visual field. The experimenter kept a close watch on 165 the subject to control for steady fixation on the target dot. Trials that involved eye movements 166 were discarded. When the subject was ready for a trial, he or she pressed the down arrow on 167 the keyboard, after which the exposure occurred. To carry out the task, the subject had to 168 press the left or right arrow when he or she identified a word or a pseudoword. The session 169 lasted about two hours and consisted of nine blocks of 100 trials for each font. The blocks 170 were presented in random order. A total of 450 words and 450 pseudowords were presented.

248
We ran a mixed-effect model to test whether the differences observed between the fonts 249 were significant. The dependant variable was the number of letters correctly identified, the 250 fixed variables were the font types, and the random variable was the subject identity. P-values 251 that correspond to the differences between the different fonts are shown in Table 2

General discussion
273 Our first hypothesis was that poor inter-letter regularity would impair reading performance.
274 Our results suggest that, indeed, lack of inter-letter regularity can significantly impair 275 peripheral word recognition performance. We showed this negative effect for two fonts 276 (MixedCase and Collect), both mixing lowercase and uppercase letters. These fonts with the 278 were also the fonts that significantly resulted in the poorest performances, while the fonts 279 that had a better inter-letter regularity (DejaVu and Extended) resulted in the best 280 performances. Interestingly, intermediary irregularity caused by tilted letters (Slant) did not 281 significantly affect word recognition performance.

282
The findings of the word recognition experiment cannot be explained by letter recognition 283 performances, as results were inconsistent between the two experiments. In the case of the 284 Slant font, the findings show opposite results between letter and word recognition. The Slant 285 font was the poorest-performing font with regard to letter recognition, while for word 286 recognition it showed a similar recognition rate to the two best-performing fonts and did 287 significantly better than the two mixed-case fonts. Our findings thus show an important 288 limitation of the usually accepted theory that links peripheral letter and word recognition 289 performance (22,23). It is also possible that the lack of regularity between letters causes the 290 disruption of word uniformity, and a consecutive decrease in word recognition performance 291 (24).

292
The letters in the slant conditions were either rotated to the right or to the left or had no 293 rotation. It appears that for letter recognition, this rotation is confusing, as it is difficult to 294 predict the nature of the rotation for each single letter. While for word recognition, the 295 rhythm produced by the rotations of the Slant font condition leads to greater predictability of 296 the word components and thus makes it easier for the subjects to tune into the font structure.
297 Our results differ from findings by Gauthier et al. (13), who compared recognition of letter 298 trigrams where the letters were slanted to one side to the recognition of trigrams where the 299 letters were mixed between slants to the left and right (similar to our Slant font) and found no 300 difference in performance between the two font conditions. Since our experiment did not 301 compare the Slant font with a font condition that only had a slant to one side, this may be the 302 cause of the different results.

303
The fact that the mixed-case fonts (Collect and MixedCase) are the poorest-performing in 304 the word recognition experiment confirms previous studies of the mixed-cased effect on 305 foveal recognition (17)(18)(19)(20). In the present study, we extend the findings to include peripheral 306 vision.

307
In our experiment on letter recognition, only one out of the two fonts with mixed-case 308 features was significantly outperformed by DejaVu, which indicates that the negative 309 influence of mixed-case fonts on letter recognition is less pronounced than the impact on to 310 word recognition. If letters within a word become too uncommon in relation to each other, 311 subjects may have to adopt a reading strategy based on serial processing of each single letter, 312 which is much less efficient than parallel processing drawing on orthographic lexical 313 information (25, 26).

314
For both letter and word recognition, the long extenders hold an advantage (Extended). In 315 reading situations involving smaller visual angles, a large x-height (meaning shorter 316 extenders) is known to facilitate reading (27). However, it is possible that if the x-height is 317 kept constant, longer extenders could also benefit reading at small visual angles. Our findings 318 suggest that for reading situations involving peripheral reading, long ascenders and 319 descenders may be an advantage. This is interesting, since, to our knowledge, this simple 320 change in fonts had never been directly tested, although it seems to be an easy way to modify 321 a font and improve letter recognition performance.

322
Studies into letter recognition suggest that letters are recognized by their features (6, 28-323 30). Viewing our findings in this perspective, the data on letter recognition suggests that as 325 importance. In contrast to this, the data on word recognition suggests that word processing 326 benefits from regularity. It is generally believed that for successful word processing, it is 327 highly essential to be able to recognize the letters and their features (26, 31, 32); our findings 328 add to this by demonstrating that in addition to great inter-letter dissimilarity (7), inter-letter 329 regularity within a word also contributes to successful word recognition.

331
Conclusion 332 We found evidence that a new factor, which we have labelled regularity, has a direct effect on 333 word recognition performance, as fonts of great inter-letter regularity outperformed fonts of 334 low inter-letter regularity in a peripheral word recognition task. The effect varied between 335 letter and word recognition, so that rotated familiar letter shapes had a more negative effect 336 on letter recognition than on word recognition, and mixing upper-and lowercase letters -337 which was generally detrimental -had a more negative effect on word recognition than on 338 letter recognition. Our key finding is that between letter and word recognition, great inter-339 letter regularity has the most positive effect on word recognition and less on letter 340 recognition, which shows that supplementary features can improve letter recognition, while 341 they have a negative effect on word recognition. Our findings demonstrate that the 342 typographic approach of working with inter-letter regularity is an important factor that needs 343 to be considered in the design of fonts for word processing in peripheral vision.