Multilingual Touchscreen Keyboard Design and Optimization

Figure 7. Letter Frequency in English, French, German, Spanish and Chinese
20%
18%
16%
14%
12%
10%
8%
6%
4%
2%
0%
English French Spanish
German Chinese
a b c d e f g h i j k l m n o p q r s t u v w x y z
Table 1. Correlation coefficients of letter (bottom-left) and digraph frequency
distributions (top-right)
English French Spanish German Chinese
English 0.75 0.74 0.69 0.36
French 0.88 0.86 0.70 0.30
Spanish 0.86 0.92 0.66 0.24
German 0.88 0.90 0.81 0.33
Chinese 0.48 0.44 0.46 0.51
Table 1 shows the correlation coefficients across these five languages. As shown,
Chinese is loosely correlated with other four languages. In letter frequency distribution,
the correlation coefficients between Chinese with other languages are all below 0.51,
while those across other four languages are all above 0.8. In digraph distribution,
correlation coefficients between Chinese with other four languages are all below 0.4. In
one case, between Spanish and Chinese, the digraph correlation is only 0.24. These low
correlations impose a challenge to simultaneous optimization. If the positive result in the
last section was largely due to the strong digraph correlation between English and French,
the current goal of optimizing for five languages including Chinese would have to place
greater hope on the flexibility of the optimization space itself.
Following the proposed multilingual optimization methodology, each of the five
languages is weighted equally in the optimization process. The mean time of inputting a
character is represented as:
t�0.2tEng �0.2tFren �0.2tSpanish �0.2tGerman� 0.2tChn
(9)
t Eng , t Fren , t Spanish , t German and tChn are the mean times of typing a character in the
corresponding languages, which are predicted by the Fitts-digraph model (Equation 5).
18