Research On A Sentence-level Input System On Handheld Devices

With the rapid development of handheld devices, the Chinese input technology is becoming more and more important. The sentence-level input method can increase the input speed and have a strong practical value. Due to of the limited memory and low CPU speed of handheld devices, the word-level and the phrase-level input methods dominate and are widely used in current applications. This paper will focus on the sentence-level input method to further increase the input speed.Firstly, this paper explores various compression methods in language modeling. This is done by 1) properly integrating count cutoff and pruning; 2) applying a clustering method to further compress the model; 3) Employing the perplexity measurement to evaluate the language model. Evaluation shows that our method achieves lower perplexity and better performance.Secondly, due to the limited memory and low CPU speed of handheld devices, this paper puts forward a compact file structure and a sentence-level search algorithm. This is done by: 1) taking the CKC input method as an example and improves the input rule; 2) designing a new format to store the Chinese characters, in a reverse order; 3) introducing a sentence-level search algorithm with breadth of N. Evaluation shows that our algorithm can largely reduce the computational complexity and achieves high performance.Finally, we implement a sentence-level input system on Symbian OS. The system uses a Client-Server architecture, where the client handles input/output and the server searches the related files and calculates the output probability. In addition, the system is scalable with some interfaces for further improvement.