Display of Virtual Braille Dots by Lateral Skin Deformation: Feasibility Study
table of contents4. Virtual Braille Legibility
The next step needed to evaluate the feasibility of the tangential skin deformation approach for reading Braille was to determine whether blind subjects could read the subset of characters that could be displayed by the VBD. Here, we hoped to also begin identifying the strengths and weaknesses of the concept.
4.1 Method
4.1.1 Participants
Two females and three males, experienced Braille readers, volunteered for the study. All subjects were blind from birth. Their ages varied between 22 and 55. The subjects' primary reading finger was the right-hand index. All subjects except the reference subject had never experienced the VBD or heard about our efforts.
4.1.2 Task
The reading task was designed to evaluate the legibility of sequences of first-row characters displayed on the VBD. Subjects were asked to read individual 4-character strings using their dominant reading finger. The first and last characters of each string was always 'c' (
). The two middle characters could be any of the 16 combinations of the characters 'a' (
), 'c' (), 'dot #4' (
), and ' ' (
): " ", or " " for example. The character 'c' () was chosen as the string delimiter to avoid confusion.
4.1.3 Procedure
The subjects were given written Braille instructions and had supervised practice trials until they felt comfortable with the task. They were presented with strings to read in block trials. They placed the slider to the left, waited for an audible signal, read the string, and reported verbally the two middle characters. There was no time limit but they were strongly encouraged to answer quickly. In case of doubt, they were asked to give their best guess. Subjects could stop at any time if they no longer felt comfortable (e.g. loss of tactile sensation, fatigue). They were given the choice of doing the trials with texture, without texture, or in both conditions. Some subjects were tested in both conditions while others decided to experiment with only one type. A trial block comprised 80 strings with each of the 16 possible combinations appearing 5 times in randomized order.
4.1.4 Data Collection
The experimenter logged the subject's answer for each trial. The slider trajectory was automatically recorded by the system. It was analyzed off-line to compute the duration of trials. A trial was considered to begin when the rightmost actuator first arrived at the leftmost virtual dot, and to end when it crossed this dot again for the last time in the opposite direction. In other words, the leading and trailing parts of the slider trajectory for which no actuator was affected by the Braille string were discarded.
4.2 Results and Discussion
The main results of this legibility experiment as well as those of a control experiment with conventional Braille (Section 5) are summarized in Table I.
4.2.1 Legibility
Legibility was defined by the proportion of correct identifications of 2-character strings. Results suggest that the effect of adding texture was idiosyncratic (see Table I). A dramatic improvement in performance was seen in one subject while a loss was observed in two other subjects. Retaining the best conditions for each subject, the legibility rates were between 71.3% and 98.8%.
Legibility rates were also plotted over time to assess the effect of fatigue. Without texture, no significant change with time could be noticed. However, for some subjects, performance tended to decrease noticeably after about 50 trials when texture was used (see Figure 14 for one of the worst-case examples).
Regardless of the string, individual characters having one dot, '
Table IV shows the confusion matrix for individual characters. No clear pattern emerged, except perhaps that '
Table I shows the average duration of trials. The reading speed was far from the expected Braille reading speed of 65 to 185 words per minute [Legge et al. 1999], but the conditions are so different that a direct comparison is not possible.
Correlations between reading speed and string legibility were also investigated but none could be found, even though there could be important duration variations between trials of a same subject. If it is assumed that the time taken to read a pair of characters is an indication of the confidence the subject has in her or his answer, the characters that the subjects thought were hard to read were not necessarily the ones they had difficulty reading.
The recorded trajectory of the slider was used to investigate the reading pattern used by the subjects. Three classes of patterns were identified (see Figure 15). Subjects often used one or two straight passes over the dots. On other occasions, they would explore the virtual Braille string with short back-and-forth motions. In all cases it appeared that subjects read from left to right since they moved slower in that direction.
All subjects reported that reading requires concentration, mostly because the dots were subtle (possibly due to the limited range of motion of actuators) and differed in perceived shape from physical Braille dots. Adding texture seemed to facilitate the perception of the dots for some subjects, while others found the sensation unpleasant. Some subjects also complained of loss of tactile sensation over time in both nominal and textured conditions. The occasionally-observed decrease in performance over time seems to confirm that a loss of tactile sensation was occuring with the textured representation and was likely due to the adaptation of tactile receptors.
Subjects also reported difficulties with locating the stationary Braille dots on the virtual Braille line. Most subjects mentioned that the display of characters with more than one row of points and having meaning would help reading.
Finally, contrary to our expectations, subjects reported that scanning constrained by a slider was beneficial because it guided their hand movement. They found it to be an advantage over paper Braille.
Table I: Summary of results from legibility experiments. Results from the first experiment
(vbd, Section 4) are presented in columns nominal and textured. Results from the second experiment (conventional Braille, Section 5) are presented in the column control. Trial durations are shown with their standard deviation.
4.2.2 Character Pairs Legibility
' or '', were harder to read than characters having no or two dots, '' or '' (see Table II). The legibility also varied with the 2-character string (see Table III). Except for special cases such as the pair " " which was read perfectly, no insight could be gained regarding the cause of variations in reading difficulty. It is not clear, for example, why the string " " has much lower legibility than " ".
Table II:
Average character legibility (%). The average was computed
across all subjects using individual subject means to compensate for the
unequal number of trials under the different conditions.
Table III:
Average 2-character string legibility (%). The average was computed
across all subjects using individual subject means to compensate for the
unequal number of trials under the different conditions.
' and '' were rarely mistaken for one another. Similarly, Table V shows the confusion matrix for pairs of characters. Again there was no clear pattern. However, it does seem that the most frequent errors (n=3,4,5) generally corresponded to the insertion of a single extra dot or to the incorrect localization of a dot within a character.
Table IV: Confusion matrix for
individual characters. Answers from all trials were pooled
together.
Table V: Confusion matrix for pairs of
characters. Answers from all trials were pooled together.
4.2.3 Reading Patterns
4.2.4 Verbal Reports
