The analysis of time helps in extracting knowledge from web contents. In order to analyze massive amounts of data, current natural language processing systems rely mainly on supervised approaches: machine learning algorithms that learn how to classify data based on corpora, annotated with the TimeML mark-up language or one of its derivatives. The quality of annotation data, thus, affects the performances of such systems. Quality can be improved by reasoning on temporal annotations. This article takes such a view. After a review of the strictly necessary background, it focuses on and discusses open issues in the area of quality of temporal annotations: inconsistency and incompleteness of annotations. Then it proposes a semantic reasoning approach as solution for improving on their quality, viz., the SOA-based Qualitative Temporal Reasoner for reasoning about temporal annotations, which leverages on existing theories and tools for qualitative reasoning. The article presents the design of the reasoner and its two main reasoning services: consistency checking for tackling inconsistency, and deduction for addressing incompleteness on demand. Afterward, the paper presents an experimental evaluation of the reasoner, sustaining why the chosen semantic reasoning approach can help improve on quality of annotations. The experiment assesses the reasoner’s performances on two different corpora and from several perspectives, e.g., the effectiveness of consistency checking in terms of the number of inconsistent documents found, and of deduction in terms of the number of annotations added. It concludes with discussion of the results of the evaluation and possible routes for future work.
|Titolo:||Qualitative Temporal Reasoning Can Improve on Temporal Annotation Quality: How and Why|
|Data di pubblicazione:||2016|
|Appare nelle tipologie:||1.1 Articolo in rivista|