Three preliminary tests, using a neural network, were carried out in an attempt to establish the need for context for accurate automatic categorisation of stressed and unstressed syllables. The results indicate that a neural network can out-perform humans in the task of identifying stressed and unstressed syllables, in conditions where no immediate context is given and where a limited amount of immediate context is provided. The overall success rates of the neural network under each condition indicate that the inclusion of context significantly improves the number of correct responses.
Keywords: Neural networks, context, stress.
Bibliographic reference. Moore, Jan / Roach, Peter (1993): "The role of context in the automatic recognition of stressed syllables", In EUROSPEECH'93, 767-770.