Toward a stochastic speech act model of email behavior

Human interpersonal face-to-face interaction can be considered in terms of successions of speech acts. These are utterances which contain an intention, and the act of creating an utterance causes the intention to be directed toward the recipient. In this study the language contained in emails is considered in a similar manner. Due to the relatively asynchronous nature of email communication, communicators tend to generate packets of speech acts longer than those used in face-to-face communication in order to communicate efficiently. In this paper, the structure of these packets is examined by comparing probability transition matrices of speech act categories using multidimensional scaling and hierarchical cluster analysis. Results indicate that a ten-cluster system of email classification may represent a possible taxonomy of email intentions. In addition, speech act content in email communication can significantly predict membership of external categories, such as whether the email has a business or a personal nature. The use of an email intention classification system may enable future higher-level analysis of email relationships in terms of language intentions.