Natural language generation (NLG) is characterized by the automated creation of language, based on databases and linguistic models. In the field of NLG, in-depth classification of natural language is an open challenge, especially within Asian languages. One example is conversational Japanese, for which there currently exist very few speech act models. To address this deficiency, the presented study proposes a linguistic categorization of illocutionary speech acts. The model developed is based upon Brown and Levinson's politeness theory and Bach and Harnish's speech act categorization. A case study is conducted to test applicability of the proposed theoretical constructs using machine learning methods on conversational data. The results obtained can effectively contribute to the generation of a corpus of conversational Japanese.