The mechanism of word crowding

Abstract

Word reading speed in peripheral vision is slower when words are in close proximity of other words (Chung, 2004). This word crowding effect could arise as a consequence of interaction of low-level letter features between words, or the interaction between high-level holistic representations of words. We evaluated these two hypotheses by examining how word crowding changes for five configurations of flanking words: the control condition - flanking words were oriented upright; scrambled - letters in each flanking word were scrambled in order; horizontal-flip - each flanking word was the left-right mirror-image of the original; letter-flip - each letter of the flanking word was the left-right mirror-image of the original; and vertical-flip - each flanking word was the up-down mirror-image of the original. The low-level letter feature interaction hypothesis predicts similar word crowding effect for all the different flanker configurations, while the high-level holistic representation hypothesis predicts less word crowding effect for all the alternative flanker conditions, compared with the control condition. We found that oral reading speed for words flanked above and below by other words, measured at 10° eccentricity in the nasal field, showed the same dependence on the vertical separation between the target and its flanking words, for the various flanker configurations. The result was also similar when we rotated the flanking words by 90° to disrupt the periodic vertical pattern, which presumably is the main structure in words. The remarkably similar word crowding effect irrespective of the flanker configurations suggests that word crowding arises as a consequence of interactions of low-level letter features.

Figure 1

Samples of the target word “home” presented in the unflanked condition and flanked conditions for the control condition where the flanking words were oriented upright. The numbers on the top row of the figure refer to the nominal word spacing between each pair of vertically adjacent words.

Figure 2

Examples of the six flanker configurations. Control, scrambled, horizontal-flip, letter-flip and vertical-flip were the flanker configurations tested in Experiment 1. The 90° rotated condition was tested in Experiment 3. See text for a detailed description of each of these flanker configurations. The vertical word spacing shown here was 1× standard word spacing.

Figure 3

Sample psychometric functions relating proportion correct of word recognition with word exposure duration (sec) are shown for two vertical spacings (0.7× and 1.5× standard spacings) and for all flanker configurations. All psychometric functions were obtained from observer S3. The gray star in each panel represents the exposure duration that yielded 80% accuracy of word reading, which was subsequently converted to reading speed in words per minute (wpm).

Figure 4

An example of the target word “rose” presented in the presence of two upright flankers (the control condition), at a 0.42× spacing, the smallest spacing used in Experiment 2.

Figure 5

Reading speed (wpm) is plotted as a function of vertical word spacing for the five flanker configurations. The unflanked condition is equivalent to having an infinite vertical word spacing (“Inf” on the x-axes). Each panel shows the data from all six observers (represented by different symbols) who participated in Experiment 1. We used bootstrapping with 5000 resampling to estimate the 95% confidence intervals. Error bars represent the averaged 95% confidence intervals for each flanker configuration and are given near the right ordinate in each panel.

Figure 6

Normalized reading speed, averaged across all six observers who participated in Experiment 1, is plotted as a function of vertical word spacing for the five flanker configurations. Normalized reading speed refers to the ratio of reading speed for the flanked to unflanked condition. Error bars represent ± 1 SEM.

Figure 7

Reading speed (wpm) is plotted as a function of vertical word spacing for the five flanker configurations. The unflanked condition is equivalent to having an infinite vertical word spacing (“Inf” on the x-axes). Each panel shows the data from all four observers (represented by different symbols) who participated in Experiment 2. Open symbols represent the unflanked reading speeds measured with the modified word list. We used bootstrapping with 5000 resampling to estimate the 95% confidence intervals. Error bars represent the averaged 95% confidence intervals for each flanker configuration and are given close to the right ordinate in each panel.

Figure 8

Normalized reading speed, averaged across the four observers who participated in Experiment 2, is plotted as a function of vertical word spacing for the five flanker configurations. Normalized reading speed refers to the ratio of reading speed between the flanked and unflanked conditions as measured using the same word list. Error bars represent ± 1 SEM. For comparison, data obtained for Experiment 1 are included in gray (replotted from Fig. 6).

Figure 9

Reading speed for individual observers (left panel) and the group-averaged normalized reading speed (right panel) is plotted as a function of vertical word spacing for the 90° rotated flanker configuration. Each symbol in the left panel represents data obtained from a single observer. Error bars in the left panel represent averaged 95% confidence intervals. Error bars in the right panel represent ± 1 SEM.