SPEECH REPAIR CLUES USED FOR DETECTING SPEECH REPAIRS IN MICASE

This paper analyses speech repair clues in spontaneous speech in the MICASE corpus. An algorithm for detecting speech repairs without using prosodic information or a syntactic parser is proposed. Implementation of this algorithm into spontaneous speech is presented. Two types of speech repairs were analysed: modification and abridged repairs. The analysis shows the frequency of use of the individual speech repair clues such as editing term, word fragment and word correspondence. Modification speech repairs are shown to be the most common speech repair used in the analysed transcripts. Out of three presented speech repair clues the word correspondence is the most prevailing one used by speakers. A test for statistical significance is used to show the significance level in using speech repair clues by male and female speakers