Knowledge Based Transcription of Handwritten Pitman's Shorthand Using Word Frequency and Context

Abstract

The paper proposes the computer transcription of handwritten Pitman shorthand as a mean of rapid text entry to handheld devices. Handwritten outlines are bound to be variation from writers to writers and it causes pattern recognition to be prone to errors, however these imperfections can be restored by the use of heuristic approach in the interpretation stage. The transcription accuracy can be improved by the combination of three factors: firstly, incorporating contextual knowledge as used by human readers; secondly, applying knowledge of the most frequent words of Pitman shorthand; and finally, adding knowledge of collocation. Statistical analysis of a Shorthand lexicon is presented and distribution of transcription accuracy based on accuracy of segmentation is discussed in the paper. Experiments using a phonetic Lexicon with 5000 entries show that the approach is efficient and produces a satisfactory transcription accuracy of 94%.