A Corpus-based Chinese Syllable-to-Character System

One of the popular input systems is based on Chinese phonetic symbols. Designing such kind of a syllable-to-character (STC) input system involves two major issues, namely, fault tolerance handling and homonym resolution. In this paper, the fault tolerance mechanism is constructed on the basis of a user-defined confusing set and a modified bucket indexing scheme is incorporated so as to satisfy real-time requirement. Meanwhile the homonym resolution is handled by binding force and heuristic selection rules. Both the system performance and tolerance ability are justified with real corpus in terms of searching speed and character conversion accuracy rate. Experimental results show that the proposed scheme can achieve 93.54% accuracy for zero-error syllable inputs and 80.13% for zero-tone syllable inputs. Furthermore both robustness and tolerance of the proposed system are proved for high input error rates.