| Summary: | Temporal analysis is the task of determining the temporal structure of a given text. Such a structure represents the order of the events and states mentioned in the text on a time line. The main contribution of this thesis is in presenting a new processing framework for temporal analysis. The framework is a computational one and has been implemented in a system called TASTE for the temporal analysis of instructional texts. In particular, TASTE has been successfully tested on nine cookery recipes. Amongst the more important ideas explored in this thesis are the following: • We integrate qualitative information (as expressed by temporal connectives like <I>before</I> and <I>after</I>) and quantitative information (as expressed in phrases like <I>for 20 minutes</I> and <I>20 minutes before</I>) in a text into the temporal analysis framework. Previous work has only considered qualitative information but ignored the quantitative kind. • We propose a new approach to the problem of integrating the current event or state into the preceding discourse. This problem has been identified as important for solving the temporal analysis task. • We show how information from the environment surrounding a text can affect the temporal analysis of instructional texts. In particular, we show that different temporal structures for the same text can be derived in different environments. Note that the environment information is in addition to the usual information considered in temporal analysis such as information from tense and aspect, temporal connectives and real-world knowledge. An example of information from the environment for the domain of cookery recipes is the availability of resources for carrying out an action. • We incorporate techniques developed in the field of temporal reasoning into the temporal analysis task. In addition, we analyse the complexity of temporal reasoning algorithm needed in the temporal analysis of instructional texts.
|
|---|
