Greg Krynicki's MA


Abstract

My MA reports on the influence that prosody of selected Polish ambiguous utterances may have on their interpretation and translation into English. A method for parametric description of the pitch curves coextensive with these utterances is discussed. On the basis of the pitch parameters obtained by the above method, the classification of the utterances is performed with respect to their interpretation and translation into English. Three approaches to the problem of classification are presented:
  • Pattern Matching approach in the form of the Dynamic Time Warping algorithm and Nearest Neighbourhood decision rule,
  • the statistical approach based on statistical Discriminant Functions
  • neural approach in the form of two types of artificial Neural Network: a single-cell perceptron and a feed-forward neural network with one hidden layer.
The Nearest Neighbour classifier provides 66.7 - 79.5% correct classification rate depending on the disambiguated word, Discriminant Analysis performance ranged from 82.5 to 95% of correct classifications, and the neural approaches produced correct classifications in 71.2 - 84.9% of cases. The applications of the above findings in Polish-English Spoken Language Translation, Polish-English lexicography and intonation teaching are discussed and presented.

Download

Classification of Isolated Pitch Patterns by Means of Nearest Neighbour Classifier, Neural Networks and Discriminant Function, unpublished.
zipped .pdf file

17 Jan 2001