A Phonetic Rule-based Approach for Generating Casual English Sentences

抄録

We present a system for generating casual English short sentences from regular English input using a phonetic rule-based approach. This is addressed as an AI task, with the potential application of generating Twitter-style sentences for marketing or other communication purposes. Our aim was to automatically produce sentences that would appear to a human reader to be indistinguishable to sentences which are the result of human creativity. To evaluate the performance of the system, we conducted Turing-type tests with human readers, to consider firstly "human-likeness", and also "legibility" of the sentences. In this paper, we discuss the overall design of the system, the custom-made phoneme database, and the process and results of our evaluation experiment.