Keywords

1 Introduction

While it has been reported that users of mobile devices expect to use it with one-handed thumb [6], the size of mobile touchscreens is enlarging owing to the demand for increasingly larger screens. This trend inevitably enlarges the touchscreen area that users can no longer reach by using only a thumb. As a result, users are forced to use the device with two hand or change the manner in which they hold the device.

To address this problem, we designed the QAZ keyboard (Fig. 1), which is a portrait QWERTY-based keyboard. We rotated a QWERTY keyboard 90 degrees to make the keyboard cover more area where thumb can easily press keys, thus making text entry on a QWERTY keyboard with one-handed thumb input easier on a large mobile touchscreen.

Before designing the keyboard, we conducted two experiments to investigate the pointing performance with one-handed thumb input on large mobile touchscreens. Based on the results, we designed and implemented the QAZ keyboard on Android-based devices. We also conducted a long-term user study to measure the basic performance of the keyboard. This paper reports the results of those experiments and presents the QAZ keyboard.

Fig. 1.
figure 1

QAZ keyboard.

Full size image
Fig. 2.
figure 2

Normal QWERTY keyboard.

Full size image

2 Related Work

The characteristics of one-handed thumb input have been extensively researched (e.g., [5, 15–17, 21]). As a result, it has been shown that each location on a touchscreen has different accuracy and time of pointing; therefore, by using a compensation function that shifts the user’s touches, the touch performance can be improved [5, 21]. As a result, design of the user interface of a touchscreen should avoid key locations that require excessive thumb flexion or extension such as the bottom right and top left on the touchscreen.

Based on characteristics of touch behavior, text entry methods utilizing characteristics of one-handed touch input were designed. Takahama and Go [20] proposed a one-handed text entry method which provides a stable holding position, where users can input text by rubbing the screen with a thumb on a small touchscreen area. Kimioka et al. [7] proposed a text entry method by adopting two arc shaped keyboards for two-handed multi-touch gestures by using both thumbs on a tablet. While these research papers proposed novel text entry methods which were designed ergonomically, the layouts were totally different to those of existing keyboards. On the other hand, novices can use a QAZ keyboard with ease.

Some researchers have proposed keyboards that dynamically adapt the shape and position of a key to users’ hands. Sax et al. [18] and Kuno et al. [8] proposed a keyboard where a user can perform touch-typing using a soft keyboard. iGrasp [2] is a system which provides a soft keyboard to users based on how and where users grasp the devices. In contrast, we found the proper area to place a keyboard on a touchscreen based on the results of our preliminary experiments which investigated characteristics of one-handed thumb input, thereby allowing users to use the keyboard without dynamic adaptation.

To improve text entry performance, static keyboard layouts were explored. Some researchers have proposed high-performance keyboard layouts by optimizing the motion of a finger [12, 23]. Layouts that a user can easily learn have also been investigated [1]. Sipos et al. [19] presented a layout suited for thumb navigation on one-handed devices. Half-QWERTY [14] is a keyboard which only has the left-hand keys of the QWERTY layout for one-handed typing. Users can type the right-hand keys by holding the space bar to mirror the right-hand keys onto the left ones. The 1Line Keyboard [9] condenses the three rows of keys in a normal QWERTY layout into a single line with eight keys by using a language model. While these layouts show high performance, it was also reported that users rarely preferred spending their time learning a new keyboard layout [1, 13]. This result motivated us to adopt the QWERTY layout to design our soft keyboard.

3 Experiment: Pointing Performance

We conducted experiments to investigate pointing performances with one-handed thumb input using a large mobile touchscreen in order to find an optimum region to place a keyboard. 12 participants (9 males and 3 females) ranging in age from 21 to 24 years (mean \(=\) 22.8; SD \(=\) 1.14) took part in the experiments as volunteers. We used a smartphone (LG Electronics Optimus G L-01E, size: 137 mm \(\times \) 69 mm \(\times \) 9.6 mm) with a 4.7 in. touchscreen.

3.1 Procedure

We asked the participants to hold the device with their right hands. They were then asked to tap a target on the touchscreen as a trial with their right thumb (Figs. 3, 4).

Our study consisted of two experiments, Experiment A and B. In Experiment A, we motivated the participants to change the way they hold the smartphone depending on the position of a target (free hand posture) for better performance. In Experiment B, we instructed them to keep their hand posture as fixed as possible. In total, each participant performed 1440 (5 sessions \(\times \) (16 \(\times \) 9) targets \(\times \) 2 experiments) trials.

Fig. 3.
figure 3

All participants were instructed to put their elbow on a desk.

Full size image
Fig. 4.
figure 4

A target presented on the smartphone’s touchscreen.

Full size image

3.2 Results and Analysis

Figures 5 and 6 show the accuracy and time of the pointing tasks, respectively. The results show that the center of the touchscreen shows high performance, and the center of the right side shows high accuracy while the time was slow in both experiments.

Fig. 5.
figure 5

Accuracy in Experiment 1. The darker the area is, the higher the accuracy of the area.

Full size image
Fig. 6.
figure 6

Time in Experiment 1. The darker the area is, the shorter the time of the area.

Full size image
Fig. 7.
figure 7

Analyzed regions: 11 landscape regions (red; ID = \(0, \cdots , 10\)) and 32 portrait regions (blue; ID = \(11, \cdots , 42\)), which have almost the same footprint as the built-in QWERTY keyboard. (Color figure online)

Full size image

Therefore, to find an optimum region to place a keyboard, it is necessary to compare pointing performance among various regions where we can place a keyboard by taking both accuracy and time into account. To realize this, we first calculated the average accuracy and time within the regions shown in Fig. 7 and Tables 1, 2. Then, to evaluate each region, taking both accuracy and time into account, we used the following Cost function that roughly models the performance of text entry when a keyboard is placed on a certain region:

$$\begin{aligned} Cost(P,T)=NT(1+M(1-P)/P) \end{aligned}$$

Here, Cost is the estimated time to type a specific text using the keyboard placed on a certain region. Specifically, the parameter P is the success rate of the keyboard (i.e., pointing accuracy of the region). The parameter T is the average time of pointing for the region. N is a constant which is the ideal number of taps necessary to type the text (e.g., if the number of characters in a given text is 100, it is necessary for the user to tap keys n times to type one character, \(N=100n\)). Therefore, NT is the ideal total time required to type the text. Moreover, because typing usually contains errors, we model this by adopting M, which is the number of taps required to recover from an error input (usually \(M=2\) because recovering from one error input requires two taps: tapping Back Space key to delete the error input and then tapping the correct key). Because \((1-P)/P\) is the error rate, \(NTM(1-P)/P\) is the additional time required for recover. In summary, if a region has low Cost, based on the pointing performance, the region is considered suitable for text entry.

Based on Tables 1, 2, we adopted Region 23, which is portrait and the region showing the lowest Cost.

Table 1. Experimental data in Experiment A. ID is region ID. Cost 1 indicates calculated Cost as \(M = 1\), \(N = 100\). Cost 2 indicates calculated Cost as \(M = 2\), \(N = 100\). These data are sorted from the highest to the lowest performance.
Full size table
Table 2. Experimental data in Experiment B. ID is region ID. Cost 1 indicates calculated Cost as \(M = 1\), \(N = 100\). Cost 2 indicates calculated Cost as \(M = 2\), \(N = 100\). These data is sorted from the highest to the lowest performance.
Full size table

4 QAZ Keyboard

Based on the results of our experiment, we designed a portrait soft keyboard, which we designated a QAZ keyboard, and then implemented the keyboard as an Android application shown in Fig. 1. We show the design principles below.

  • QWERTY layout

    Our keyboard adopts the QWERTY layout because many people are familiar with this layout.

  • Portrait region

    Regions showing high performance might also be suitable to placing a keyboard in terms of accuracy and fast typing. Hence, our keyboard arranges the keys of a QWERTY keyboard on a portrait region that showed the highest performance in the experiment.

5 Evaluation

We conducted a long-term user study to measure basic performance of the QAZ keyboard. Four participants (P1-P4; 4 males) ranging in age from 22 to 24 years (mean \(=\) 23.25; SD \(=\) 0.83) took part in this user study. They were all right-handed. They were all Japanese. Regularly, one participant used a soft QWERTY keyboard; two participants used Grid Flick (a method for inputting Japanese text); and one participant used both methods. They were all familiar with the QWERTY keyboard because they were currently using a QWERTY keyboard to control their PCs. They were compensated with 820 JPY (approximately 8 USD) / hour for their participation. We used the same apparatus as used in Experiment A and B.

5.1 Procedure

We asked the participants to input phrases chosen at random from a set of 500 phrases [10]. The length of the phrases ranged from 16 to 43 (mean \(=\) 28.61). These phrases have only lowercase characters and no punctuation.

The participants input 5 phrases as a training task. Then, they conducted two parts of the evaluation described below:

  • Part I - Basic Performance

    Part I was a longitudinal study designed in accordance with conventional studies on text input systems [11, 22] to measure the basic performance of the QAZ keyboard. This part consisted of 10 sessions. The sessions were scheduled with one or two sessions per day. The maximum allowable interval between sessions was two days. Each session was divided into 12 blocks with 5 phrases per block. Participants could take a break freely between blocks and sessions.

    In order to normalize experimental conditions between participants, we also asked the participants to hold the smartphone without supporting it using a desk or their bodies.

  • Part II - vs. normal QWERTY keyboard

    After Part I, Part II was conducted to compare the performance of the QAZ keyboard and a QWERTY keyboard (Fig. 2). The aim of this comparison was to investigate whether the QAZ keyboard has a comparative performance improvement compared to a QWERTY keyboard. Part II consisted of two extra sessions: one session per keyboard. Specifically, Session 11 was conducted using the QAZ keyboard; Session 12 was conducted using the QWERTY keyboard. In order to normalize experimental conditions between the two keyboards, the two sessions were conducted on the same day. Furthermore, the layout, key size, and shape were equal for both keyboards; only the orientation of the keys and a keyboard were different. In Part II, we used the same set of phrases as in Part I; but they were different phrases between the two sessions to eliminate any learning effect, given that the two sessions were conducted on the same day.

Each session lasted 20–35 min. After Session 12, we asked the participants to complete a questionnaire about usability of the keyboards.

5.2 Results and Analysis

The mean text entry speed in Part I started with 11.4 wpm (SD \(=\) 1.4) in Session 1 and ended with 17.8 wpm (SD \(=\) 2.1) in Session 10 with an increase of 56 % as shown in Fig. 8. The black line in Fig. 8 is the linear regression (\(R^2\) \(=\) .9243). The fastest text entry speed was 19.9 wpm recorded by P3 in Session 7.

The mean error rate over the 10 sessions was 8.7 % (Fig. 9). Error rates slightly increased (\(R^2\) \(=\) .0005) over the sessions. The lowest error rate was 2.9 % recorded by P3 in Session 7.

Fig. 8.
figure 8

Text entry speed. (Color figure online)

Full size image
Fig. 9.
figure 9

Error rate. (Color figure online)

Full size image

P1 and P2 might have focused on speed rather than accuracy, because P1 and P2 tended to input with a higher error rate than P3 and P4, while P1 and P2 tended to input faster. Similarly, P4 might have focused on accuracy rather than speed, because P4 tended to record lower speeds. However, P3’s error rates were lower than other participants’ error rates except for Session 5 and 6. Furthermore, P3’s text entry speeds were faster than other participants throughout all sessions. These results were supported by P3’s comment. In the questionnaire, P3 commented that the size of the keys on the QAZ keyboard was larger than those of the QWERTY keyboard. We thought this subjective evaluation might be caused by the ease of moving the thumb.

The mean text entry speed from Session 11 and 12 were 18.2 wpm (SD \(=\) 0.52) and 18.4 wpm (SD \(=\) 0.87), respectively. With a paired t-test, there was no significant difference between the sessions (\(t_{11} = .728\), p \(=\) 482 > .01). The mean error rates from Session 11 and 12 were 9.09 % (SD \(=\) 0.017) and 13.1 % (SD \(=\) 0.015), respectively. With a paired t-test, the QAZ keyboard’s error rate was found to be significantly lower than QWERTY keyboard (\(t_{11} = 6.046\), p \(=\) .000 < .01). These results suggest that the QAZ keyboard will perform better than the QWERTY keyboard (Fig. 10).

Fig. 10.
figure 10

Left: the mean error rate. Right: the mean text entry speed.

Full size image

6 Discussion

6.1 Occlusion

Because a QAZ keyboard is placed near the center of the screen in our current implementation, contents on the screen are occluded by the keyboard. To address this issue, we plan to test the following two design alternatives.

  • Reduction of Keyboard Size. The first solution is to reduce the size of the keyboard. Since a QAZ keyboard is surrounded by space, flicking or dragging outward from the keyboard can also be used along with tapping. By utilizing this, keys placed outside can be shrunk, because using flick or drag can reduce the number of keys required [3, 4]. In this case, however, the optimum region may be different from the one we used in this paper, because the optimum region to tap, flick, and drag would be different. Therefore, it is necessary to investigate performance of flicking or dragging with one-handed thumb input on a large mobile touchscreen.

  • Semi-transparent Keyboard. The second solution is to make the keyboard semi-transparent. When users input a text, because they do not look at contents except for the input component, they can see only an overview of the contents. Accordingly, reducing the occlusion with an almost transparent keyboard will be feasible.

6.2 Orientation of Keys

Orientation of keys may change the learning cost of the QAZ keyboard, and therefore is to be investigated as future work. Note that this paper presents the QAZ keyboard with keys oriented to users, i.e., the letter on the top of a key is placed with the same orientation in both keyboards as shown in Figs. 1 and 2. However, this arrangement deteriorates the orientation relationships between the entire keyboard and keys of the conventional QWERTY layout. While there are some alternatives to the orientation of the keys, we believe that one such feasible alternative is to rotate the keys by 90 degrees counterclockwise so that, for example, ‘Q’ is displayed as ‘’.

6.3 Limitation

In evaluation Part II, we conducted extra two sessions to compare the performance of a QAZ keyboard and a QWERTY keyboard. However, the results of using the QWERTY keyboard could be disadvantaged due to the fact that all of participants began the evaluation using the QWERTY keyboard. To compare the performances accurately, we will evaluate the two keyboards in a counter-balanced order.

7 Conclusion

In this paper, we presented the QAZ keyboard, a QWERTY keyboard designed for one-handed thumb input on a large touchscreen. To design the keyboard, we first conducted experiments to investigate pointing performance on a large mobile touchscreen using one-hand thumb input. The results showed that vertically long areas around the center of the touchscreen would be suitable to place a keyboard in terms of accuracy and time of pointing. Based on this finding, we designed and implemented the QAZ keyboard on Android-based devices. A longitudinal study with 4 participants showed that the mean text entry speed was 18.2 wpm, and the mean error rate was 9.1 %. Moreover, a comparative study of the QAZ keyboard against a QWERTY keyboard showed that the QAZ keyboard’s error rate was significantly lower than when using a QWERTY keyboard.