Keywords

1 Introduction

When using personal computers or computer terminals with keyboard, most Japanese had little problem with inputting Japanese characters through these terminals. They usually used the Roman character input method using QWERTY keyboards, just as westerners use them.

In the last decade, mobile phones with touch tone keys and displays were rapidly adopted in Japan. They are being used in addition as both telephone terminals and computer terminals, by connecting them to the Internet and accessing information they wanted to know about.

Two input methods have been developed to use touch tone keys, “one-touch-character (OTC)” and “multi-touch-per-character (MTPC)” in this time period. The seventy one Japanese hiragana characters (Japanese syllabary: 46 basic unvoiced sound and 25 voiced sound characters) were allocated to ten digit keys just as the 26 English alphabets were allocated to the digits 2 to digit 9 keys (Table 1). It is very complicated to do this, because the Japanese syllabary has almost three times as many characters as the English alphabet.

Table 1. Japanese syllabary including voiced sound characters and assignment to digits
Full size table

Basically, five characters sharing the same consonant are allocated to the same digit. Two basic ideas have been proposed to distribute voiced sound characters: one is to allocate a voiced consonant row with similar hiragana image to the same digit (ex. “は(HA)”, “ば(BA)”, “ぱ(PA)” → digit 6), the other is to request users to additionally touch the “*” key to indicate that the first touched unvoiced sound character should be changed to a voiced sound character (ex. “さ(SA)”+“*” → “ざ (ZA)”).

With the OTC method, users just touch the digit to which the desired character was allocated without indicating which one of the characters was intended (ex. Press digit “1” for “お (O)”, and press digit “3” for “す (SU)”). With the MTPC method, users need to press the same digit multiple times until they get the intended character on the display (ex. Press digit “1” 5 times to get “お (O)”, and press “3” 3 times to get “す (SU)”).

Though OTC was once used in an automated directory assistance service in the 1990’s [1], followed by research activities T9 [2] and “single-stroke-per-character” [3], MTPC has become the dominant input method because most mobile phone users preferred it for editing mail messages.

MTPC provided users with a convenient tool for communicating over the Internet even outside the home. This trend was greatly accelerated after smartphones were launched in 2008. Smartphones offer the innovative input method called “flick”, which takes full advantage of the software touch tone keys displayed on the touch screen. This method can identify an intended character when the finger flicks in the direction of the intended character; several candidates are shown around the touched key (ex. Press “3” and flick “up” to get “す (SU)” and press “3” and flick “right” to get “せ (SE)”) [4].

Though this “touch and flick” operation seemed very troublesome and time consuming to the middle and the elderly, students and young adults quickly mastered the “flick (Flick)” input method and they can now input characters much faster than with the MTPC method. This means they have an even more powerful tool for character input. Recently, scenes of young people enjoying the information they acquired by retrieval or writing/reading e-mails, can be observed everywhere in Japan in commuter trains, on sidewalks and at stations.

Thus, almost all of the young, most of the middle-aged and some of the elderly can take advantage of the search function to get target information, the holdouts became information-retrieval illiterates.

Based on the recognition that those who are not well trained in the quick input operation like “flick” should have as much opportunity to enjoy information from the Internet as the young, including dictionaries or encyclopedias, time tables, and highway traffic information, we decided to assist them and to ease their frustrations by introducing the OTC input method; it is designed to help users to effectively input Japanese characters to best utilize the smartphone environment.+

2 One-Touch-Character (OTC) Input Method for Smartphones

We had a precursory experience with the OTC input method, which was introduced in the commercial service of a fully automated directory assistance system in 1997 [1].

This service was developed assuming that touch tone telephones without displays at home represented the majority of terminals. Keywords for the directory service were address (city name, town name, etc.) and subscriber’s name. Users input these keywords as digit sequences via the touch tone keys in response to the voice prompts from the system.

If the city name entered by a user with digits created an ambiguity (an input digit sequence has multiple hiragana city name candidates), the system asked for the input of additional address section names at a lower level (ex. town name or street name). The connection between the first digits of a city name and the second digits for the additional address name input should resolve the ambiguity and identify the target address.

If the target subscriber name was ambiguous (multiple hiragana names for an input digit sequence), the system asked for address input to solve the problem. In most cases, city-town address digits and subscriber name digits could simultaneously identify the search address area and target subscriber name in hiragana.

This OTC input method and natural disambiguation process was well accepted by the users and the total number of accesses to this service reached sixteen million in nine years until the service was terminated in 2007.

After our service commenced, research started on T9 [2] and “single-stroke-per-character” [3]; both apply prediction techniques to input digit sequences to retrieve and show candidates of hiragana or kanji expressions. Similar input methods with predicative dictionary-based disambiguation were investigated in other countries [5–7]. However, these proposals achieved a very small market share compared to MTPC or Flick input method. The main reason is that these proposals target the same market, editing Japanese sentences or mail messages, and used Japanese dictionaries with huge collections of proper nouns such as place names, landmarks, and personal names. OTC-like input methods usually present many more candidates for an entered digit sequence compared to other input methods which rely on the entry of hiragana sequences.

Thus, we take two lessons from our experience [1] and, in addition, from subsequent research activities such as T9 [2] and “single-stroke-per-character” [3].

One is that the OTC approach works pretty well in search tasks if the target database is smartly organized, but not in text/messaging activities because of the difficulty of downsizing the dictionary size so as to restrict the number of candidates for each entered digit. Databases intended for search query input should respond with just one, or a few candidates, at most, for each digit.

The other is that input keywords should be short to avoid human input errors and should be fully entered by users as digit sequences. In fact, the original directory assistance service divided long addresses into short section names like city names and town names, mostly less than ten characters each, for easier input.

3 An Application Developed Around OTC Input Method

3.1 Smartphone Applications that Suit OTC Input

Taking the lessons mentioned above into consideration, we decided to develop an information retrieval application in a restricted area, such as landmarks, sightseeing points, biographical dictionaries, movie/TV talents, and athletes. One more important factor is that the data for retrieval should be freely and automatically collected by accessing the Internet and should be transposed into a defined format.

We finally selected a Japanese Who’s who as the target database. Dictionary entries were was extracted from the Japanese Wikipedia (open source) of 2013. Each entry consists of a personal name in Kanji (Chinese characters), its hiragana spelling, its ten keypad digits converted from hiragana spellings, and its URL to link to the original content.

The database contains about 130 thousand entries covering Japanese historical persons, statesmen, scholars in various areas, movie/TV talents, athletes in various sports, people in the spotlight. A check of the data showed that 65 % of personal names with different hiragana spelling were converted into different digit sequences. The remaining names collide: in one case, ten different personal names yielded the same digits. However, selecting one of those candidates is not a serious problem.

3.2 Application Development

A search application for this Who’s who was developed to determine whether the proposed OTC input method was efficient at finding a person’s name and retrieving full information from Wikipedia. A typical scene is a user watching a TV program who becomes curious about a person mentioned and wants more information about him/her. This application is named “Shirabetai (check a person in the spotlight)”.

This application was installed on the smartphone Galaxy Note (SC-01F) running Android OS.

3.3 User Interface and Operation

Four panel shots of search process execution are shown in Fig. 1. In the first panel, the user learns how to input a keyword using the OTC input method, from a simple explanation “One touch for each character!” and press the “Start” key to start a search.

Fig. 1.
figure 1

Sample application for “one touch character (OTC)” input method

Full size image

The example, a Japanese female talent in the spotlight “おおしまゆうこ(O-O-SI-MA-YU-U-KO) (大島優子),” is entered via OTC according to Table 1, see the second shot. The keyword slot shows the translation of the digits “1137812”.

The third panel displays the retrieved result. In this case, six candidates are retrieved by digits “1137812”. Two different surnames, “おおしま (O-O-SI-MA)”, “うおずみ (U-O-ZU-MI),” and three different given names, “ゆうき (YU-U-KI)”, “ようこ(YO-U-KO)”, “ゆうこ (YU-U-KO),” are retrieved. The hiragana sequences of these names clearly share the same consonants with different vowels, except the vowel row. Five of the six surnames are “おおしま (O-O-SI-MA)”, and other one is “うおずみ (U-O-ZU-MI)”. Three of six given names are “ゆうこ (YU-U-KO)”s, two are “ようこ(YO-U-KO)”s and the last one is “ゆうき (YU-U-KI)”. There are two “おおしまゆうこ (O-O-SI-MA-YU-U-KO)”s and “おおしまゆうき (O-O-SI-MA-YU-U-KI)”s with different Kanji readings. A user can use Kanji readings to select the intended one when multiple candidates appear with same hiragana sequences on the panel. As the fourth candidate “大島優子 おおしまゆうこ (O-O-SI-MA-YU-U-KO)” is the target name, the user touches it to get full information about the talent.

After this operation, the screen changes to the fourth screen to show the Wikipedia information accessed via the URL.

The above shows a typical instance of severe collision. However, as mentioned in Sect. 3.1, in most cases only a single candidate is returned. For example, the digits “163031” yields only “あべしんぞう(阿倍晋三)(A-BE-SI-N-ZO-U), prime minister of Japan” as the search result.

4 Usability Comparison Experiment

4.1 Experiment Purposes

This experiment has two purposes. One is to evaluate the popularity of this smartphone application. The other is to evaluate the usability of the proposed OTC input method, compared with MTPC and Flick input methods on smartphones.

A recent study [8] investigated the usability of text input methods for smartphone among the elderly, including MTPC (called “Mobile” in this paper) and two different Flick input methods. However they could not make it clear which input method is most suitable for the elderly, partly because they only focused to the elderly subjects who are not familiar with smartphones and the paucity of input methods at that time.

In our experiment, 30 subjects were selected with wide variety of ages, from twenties to eighties. We set three groups; Young (20–35), most are Flick experts and MTPC experienced. Middle (36–65), most are MTPC experts but novices for Flick, and Elderly (66–85) most are novices or novices for all three methods.

We intended to clarify the differences among the three age groups.

4.2 Experimental Conditions

A smartphone hosting the sample application and a name list used for character input were prepared as follows:

  1. 1.

    The application was installed on a smartphone (Galaxy Note SC-01F), and presented to each subject.

  2. 2.

    Forty personal names, famous and familiar to all subjects, were listed in hiragana characters.

  3. 3.

    Two extra input interfaces, MTPC and Flick, are prepared on the same smartphone.

  4. 4.

    Before the experiment, each subject was informed that they were to enter 40 personal names (261 hiragana characters) using the three different input methods. A detail explanation of each input method was provided if requested. Practice with unfamiliar methods was also allowed using personal names not on the test list.

  5. 5.

    The sequence of input methods was randomly determined for each subject.

Objective data (operation times and error rates) and subjective data (interviews and opinions) were collected and analyzed. Operation time to complete the character input operation was measured and the average operation time to input a hiragana character was calculated. Erroneous entries during the input operation (40 names) were counted by a watcher and averaged per personal name.

5 Results and Discussions

5.1 Application Popularity Evaluations

This application was assessed as positive by 70 % of subjects, mostly supported by Middle and Elderly, negative by 20 % and neutral by10 %. Those who answered negative were mostly Young, who can directly access Wikipedia without using an ambiguity-directed OTC input method. But they recognized that the Elderly would prefer the simple OTC input method, by acknowledging the difficulty the Elderly face in dealing with new interfaces. Middle shared the above Young opinion about the Elderly preference for OTC.

During interview collection, we had a chance to demonstrate this search application to a physically handicapped patient suffering ALS (amyotrophic lateral sclerosis). A supportive comment was obtained from him to promote this kind of application with OTC method to reduce character input labor as a welfare activity.

5.2 Operation Times and Error Rates

Remarkable trends were observed in terms of operation times and error rates as shown in Figs. 2 and 3 for OPTC and Fig. 4 for ERPW.

Fig. 2.
figure 2

Operation Time Per Character (OTPC)

Full size image
Fig. 3.
figure 3

Scattered plots of the OTPCs for 30 subjects grouped by age

Full size image
Fig. 4.
figure 4

Error Rate Per Word (personal name) (ERPW)

Full size image

  • – Operation Time Per Character (OTPC) (Figs. 2, 3 )

Comparison of average OTPCs for the three different input methods among the three age groups is shown in Fig. 2.

OPTC was shortest for Young, followed by Middle and Elderly, in every input method. OTPC ratios between Elderly and Young are 3.2 in OTC, 3.6 in Flick and 3.9 in MTPC. This suggests OTC is the most preferred method for the Elderly.

OTPC comparisons among the three methods by age groups are shown as follows.

Young: OTC ≒ Flick < MTPC

Middle: OTC < Flick≒ MTPC

Elderly: OTC < Flick< MTPC

Considering that expected average operations are 1.0 for OTC, 1.8 for Flick and 3.0 for MTPC, the Young are seen as quick learners of the Flick method as the OTC method. This also means the Young are adaptable and flexible enough to adjust to any input method.

Middle operates the three input methods pretty well. Though Middle is quickest with the OTC method, the OTPC ratios of other two methods compared to OTC are less than 2.

Elderly recorded the longest OTPC, but again OTC yields the shortest OTPC, almost half of that of others. This implies that OTC suits the Elderly, though ERPW is comparatively high.

To discover individual subject behaviors and obtain trends from different age groups, a scatter diagram was made, see Fig. 3.

Young are not concerned about the differences among the three methods. Most of them can operate even Flick input with less ERPW just fast as or faster than OTC, though they learned the new input method in this experiment. In Flick, they can operate “touch and flick” operation (two finger motions) in a moment like one touch operation for other input operations. It is rational that they need more time for MTPC input because MTPC requires exact multi touching finger motions to get the correct character.

On the other hand, the Elderly demonstrated significant differences among the three methods. Every of them achieved fastest speed (OTPC) with OTC, middle speed with Flick and slow speed with MTPC. It can be clearly said that the OTC input method best suits the Elderly. They seemed to have been annoyed for a long time by being forced to use MTPC for Japanese character input.

Nothing particular can be observed from the Middle data. Middle’s preferences for the input methods seem to mainly depend on the working environment and experience.

  • – Error Rate Per Word (ERPW) (Fig.  4 )

ERPW heavily depends on the individual’s mental attitude and willingness to experiment. Subjects who tried to break OTPC records tend to high ERPW. On the contrary, subjects who were very careful in inputting each hiragana character might have low OTPC but also low EPRW.

From average ERPW point of view, a remarkable trend can be observed in the collected data. In every input method, ERPW is the least with OTC, followed by Flick and MTPC. MTPC achieved the lowest performance regardless of age group for two reasons: average number of touches is the highest among the three input methods and entering consecutive hiragana characters in the same row (ex. “す(SU)” after “さ(SA)”) needs extra operation to set the first input character (ex. “さ(SA)” in this case). Subjects sometimes messed up this extra operation.

ERPW of Flick is comparatively low because a subject can select the intended hiragana by selecting one from five hiragana candidates displayed immediately after the first key touch.

As OTC is free from the above mentioned problems, Young and Middle recorded very low ERPW rates. Elderly made some errors due to careless operation.

6 Conclusion

Our daily life has experienced extensive change with the adoption of mobile phones with Internet access. Though mobile phones can be recognized as handheld wireless telephony devices with Internet accesses which additionally provide e-mailing functions, the smartphones that emerged in 2008 are seen more as personal computers that can perform various tasks including telephony and e-mailing.

Every year, a variety of new models with lots of new features are put on the market.

Whether the Japanese like it or not, virtually all mobile devices now have the ability to wirelessly access the Internet, either as a tablet or a smartphone with a screen panel. As these types of device normally do not have a dedicated QWERTY keyboard or touch-tone keys, users must master at least one of the character input methods provided by the vendor, either a software keyboard or a set of touch-tone keys displayed on the panel.

Aiming to send e-mail or short messages to friends or parents, most Japanese kids and students usually begin to learn Japanese character input methods using mobile phones or smartphones before using QWERTY keyboards.

Thus most of the young subjects tested herein were well accustomed to inputting Japanese characters via software touch-tone keys. On the other hand, older subjects are novices at doing so, because they generally learned to use QWERTY keyboards in the office before trying to use touch-tone-keys for character input.

In this situation, we proposed an easy-to-learn and easy-to-operate Japanese character input method named “one-touch-character (OTC)”, especially for those who have become holdouts as regards information retrieval and provided a sample retrieval application to show how this input method works.

Evaluations of this input method in comparison with MTPC and Flick method using the same touch-tone-key set on a smartphone display were executed.

As a result, the OTC input method is well accepted by Young and Middle, not to mention by the Elderly. Using OTC, the Elderly can input keywords in the shortest time with the lowest error rate. Striking to say, the Young do not mind the difference in operation of the three methods. Even with single keyword input operation, an expert can use the three methods in a mixed way, and so best utilize the advantages of each method. The results confirm that OTC method will be well received.

Regarding our sample application, the Elderly were mostly satisfied with the information provided, and appreciated its usefulness and convenience. The Young were not so impressed, because they can input keywords by hiragana characters directly into the Wikipedia site.

Further study will identify search applications that can better satisfy users’ strong curiosity via simplified keyword input interfaces. At the same time, we have to establish the OTC presence and secure market popularity in Japanese character input operations.