Performance Optimization Of Turkish Virtual Keyboards

Sinan Uşşaklı, Cevdet Aykanat, Bülent Özgüç

Text entry is one of the major problems of handheld and tablet computers. The main aim of this work is the design of first virtual keyboard for Turkish. Our contributions are fourfold: formulation of the virtual keyboard layout design problem as a combinatorial optimization problem; a Kerninghan­–Lin (KL) based formulation for the solution of virtual keyboard layout problem; creation two different digraphs for Turkish from two distinct sources representing formal and informal language usage; construction of a visualization tool to compare and contrast the features of inter- and intra-language usage habits of Turkish and English. Experiment results show that the proposed KL algorithm outperforms the existing optimization method based on Simulated Annealing (SA) in terms of the average amount of time the algorithm requires to find the best solution(s) and confirms that it is a very good candidate to obtain “a good” layout in a very short period of time. The best virtual keyboard layout designs obtained by both KL and SA algorithms achieve a performance of 40.95 word per minute, which is an improvement about 50% compared to the existing Turkish keyboards.

Digraphs

ek$i sözlük
     

Alphabetic

html
txt
xls

Full

html
txt
xls
METU Turkish Corpus
     

Alphabetic

html
txt
xls

Full

html
txt
xls

Digraph Coloring Results

ek$i sözlük
     

Black and White

png
jpg
bmp.zip

Color

png
jpg
bmp.zip
METU Turkish Corpus
     

Black and White

png
jpg
bmp.zip

Color

png
jpg
bmp.zip
Soukoreff & MacKenzie Bit95
     

Black and White

png
jpg
bmp.zip

Color

png
jpg
bmp.zip
Zhai, Hunter & Smith News Corpus
     

Black and White

png
jpg
bmp.zip

Color

png
jpg
bmp.zip
Zhai, Hunter & Smith Chat Corpus
     

Black and White

png
jpg
bmp.zip

Color

png
jpg
bmp.zip

Performance Comparison Of Proposed Result versus Turkish QWERTY and Turkish F Keyboards

Top 10.000 most used letter sequences of and performance comparisons

Length 2

txt
xls

Length 3

txt
xls

Length 4

txt
xls

Length 5

txt
xls
As Words
txt
xls

Most Used Words

ek$i sözlük
txt
xls
METU Turkish Corpus
txt
xls

 

Most Used Letters

ek$i sözlük
txt
xls
METU Turkish Corpus
txt
xls

Contact

sinan at ussakli dot net
aykanat at bilkent dot edu dot tr
ozguc at bilkent dot edu dot tr