Bayesian network based post processing of phonetic primitives in handwritten Pitman´s Shorthand

An innovative solution to the lexical post processing of handwritten Pitman´s shorthand, denoted as the Bayesian Network (BN) based transcription is discussed along with experimental results. Unlike the conventional phonetic based transliteration approach, the paper presents a novel primitive based transcription method along with the creation of a new machine readable Pitman´s shorthand lexicon. The Bayesian Network representation is shown to be robust against stroke variation and highly effective for handling major ambiguities of handwritten Pitman´s Shorthand, including unpredictable vowel omissions and unclear thicknesses between similar consonant strokes. Pitman´s shorthand specific unigram-based rejection strategies are also introduced that are highly effective in finding the most likely candidate words for a given outline.