| Part of understanding third-person fictional narrative text is determining for each sentence whether it takes some character's psychological point of view (is subjective) and, if it does, identifying the character whose point of view is taken (the subjective character). This dissertation is a computational investigation of how a reader can perform these tasks. We identify regularities, found during extensive examinations of naturally occurring texts, in the ways that texts initiate, continue, and resume a character's psychological point of view, and present an algorithm that exploits these regularities in order to recognize subjective sentences and their subjective characters.;Linguistic elements (syntactic, lexical, and morphological) are an important source of information for recognizing subjective sentences. While there are some that always indicate that a sentence is subjective, we show that there are many that only potentially do so. In deciding if an element is subjective, we show that it is useful to consider the kinds of things that have appeared since the last subjective sentence (which determine the current text situation). Elements with weak potential to be subjective should usually be interpreted to be subjective only in the most continuous of situations, while those with strong potential to be subjective can be subjective in less continuous situations.;The subjective character may be identifiable from the sentence itself. One case we identify is a sentence that reports a character's private state, such as wanting or seeing. We specify a way to distinguish reports from other uses of private-state sentences, so that it can be determined if the subjective character is identifiable from the sentence.;We also identify characters who are likely to become subjective characters (called expected subjective characters); if the subjective character is not identifiable from the sentence, it should usually be identified to be one of these. One is the last subjective character and the other is a recent actor; we specify conditions under which these are expected subjective characters and give a heuristic for resolving competition between them.;We show that references are understood differently in conversation, subjective sentences, and non-subjective sentences; namely, they are understood with respect to different sets of beliefs. We describe an implementation of the algorithm and demonstrate it in the body of the dissertation. |
