Algorithm for determining the feasibility of activizing Japanese passives

The authors are developing a Japanese revision support system with a rewriting capability. One of the revision support functions in the system is for activizing passive sentences. The activization of a passive sentence within a Japanese text unit is dependent on certain conditions, and it is not every passive sentence that can be activized. Thus, when attempting activization, it is necessary to assess whether it is actually possible. This paper proposes a new algorithm, covering typical sentences from technical manuals, for determining the feasibility of activization. In the algorithm, unactivizable passive sentences are assessed from contextual constraints and intra-sentential syntactic constraints. It is necessary to identify the omitted subject and the agent noun phrase to assess contextually unactivized passive sentences. This paper points out that conventional methods of identifying omitted elements are insufficient for technical manuals, because 'users,' 'the writer,' and 'subject matter for explanation' are all assumed as widely accepted and shared knowledge in manuals and are omitted from the very head of the text. Our algorithm includes a new method of identifying omitted elements by using the characteristics of manuals. According to evaluation experiments, our algorithm is capable of identifying 76.0% of activizable passives and 87.2% of unactivizable passives.