Times between the lines: Embedding a temporal closure component in a mixed-initiative temporal annotation framework

High-quality temporal annotation is not possible or realistic when we rely on human annotators alone. Temporal annotation is complex, and manual annotation of temporal links is slow, produces inconsistencies, and does not provide for a complete annotation. Unsupervised automatic annotation is riot able to produce high-quality annotation either. Instead, we can combine the strengths of person and machine and let them cooperate in a mixed-initiative annotation effort. A mixed-initiative annotation framework combines high-precision preprocessing, manual annotation, temporal closure, and machine learning techniques.; This dissertation focuses on the temporal closure component and its interaction with the human annotator. The human adds temporal relations that need not be supported by textual clues; the temporal closure component adds implied temporal relations and helps the annotator add those temporal relations needed to create a complete annotation. This approach makes a complete annotation possible while making sure that the time complexity of the annotation task is linear to the size of the document.