Automatic recognition and transcription of Pitman's handwritten shorthand - An approach to shortforms

Abstract A verbatim written transcript of speech would have many applications in the office, in verbatim reporting and as an aid for the deaf. Unfortunately, the automatic recognition of unlimited vocabulary speech is not likely to be possible for a number of years. An alternative strategy, investigated at Southampton University, is to attempt the less complex task of automatically transcribing handwritten notes made using the Pitman shorthand notation. Pitman shorthand outlines can be split into two classes of characters, shortforms (comprising over 90 of the most frequently used words and phrases in the English language) and vocalised outlines which can represent any word pseudo-phonetically. The shortforms represent as much as 50% of normal shorthand and are recognised directly using a dynamic programming technique with typical recognition accuracy of over 90%. Vocalised outlines are recognised using a syntactic method which interacts with a knowledge source derived from analysis of a large number of shorthand outlines. This paper describes the recognition strategy for Pitman shorthand shortforms which uses the dynamic programming template matching technique.