Comparing Word, Character, And Phoneme N-Grams For Subjective Utterance Recognition

In this paper, we compare the performance of classifiers trained using word n-grams, character n-grams, and phoneme n-grams for recognizing subjective utterances in multiparty conversation. We show that there is value in using very shallow linguistic representations, such as character n-grams, for recognizing subjective utterances, in particular, gains in the recall of subjective utterances.