Identification of Speaker Origin from Transcribed Speech Text

All sentences or passages quoted in this dissertation from other people's work have been specifically acknowledged by clear cross-referencing to author, work and page(s). Any illustrations which are not the work of the author of this dissertation have been used with the explicit permission of the originator and are specifically acknowledged. I understand that failure to do this amount to plagiarism and will be considered grounds for failure in this dissertation and the degree examination as a whole. Abstract Humans have the inherent capability to recognize a speaker's native language with minimal effort from the way he speaks English and can further benefit from the physical characteristics of the speaker. Computationally, this can be an arduous task because of the spontaneous nature of speech and the unavailability to exploit the speaker attributes. This thesis looks into a new problem which we think has not yet been explored, which will deal with the detection of native language of a speaker from his/her pattern of speaking English. These speaking patterns are captured through the text transcriptions of a spoken language corpus. We will exploit the capabilities of a classifier, which will be based on a Multinomial Mathematical Distribution and will work in integration with the n-gram profiling of words, which is tried in specific with uni-grams, bi-grams and tri-grams. We present classification experiments which are based on a logical selection of 4 European and 2 Asian languages with native level American English as the base language. Fairly high degree of accuracy is achieved by the classifier in demarcating English text spoken by various non-native and native speakers of English. The uni-grams scored fairly well as compared to the bi-grams and tri-grams for our selected corpuses. However, the bi-grams show promising results for handling more complex textual variations. Apart from classifying non-native English text we also investigated the inter-language relationships that existed between various language-categories that were explored for the non-native English speakers and further found the language-group that spoke closest to native-level English for the currently selected corpuses.