Skip to Main Content
One of the problems of connected speech recognition concerns the transcription of a sequence of phonemes (or any other lower level linguistic units) into a sequence of words of a sentence. Although this problem appears to be unsolvable in its full generality, several subproblems can be solved by imposing certain specific constraints. The paper describes possible solutions to some of these subproblems and presents some results obtained using a PDP-6 computer. Using a phoneme-to-word dictionary and the structure of the English language, a computer program transforms the phoneme string of a simple English sentence into a word string. The program uses a tree-structure mechanism to keep track of the possibilities at various stages. Since the dictionary has to be finite, the program asks for human assistance when it is unable to find a suitable match. If the input phoneme string is likely to have errors, the matching routine can be required to check for possible confusions among phonemes.