Uncertainty Detection in Natural Language Texts

Uncertainty is an important linguistic phenomenon that is relevant in many fields of language processing. In its most general sense, it can be interpreted as lack of information: the hearer or the reader cannot be certain about some pieces of information. Thus, uncertain propositions are those whose truth value or reliability cannot be determined due to lack of information. Distinguishing between factual (i.e. true or false) and uncertain propositions is of primary importance both in linguistics and natural language processing applications. For instance, in information extraction an uncertain piece of information might be of some interest for an end-user as well, but such information must not be confused with factual textual evidence (reliable information) and the two should be kept separated. The main objective of this thesis is to detect uncertainty in English and Hungarian natural language texts. As opposed to earlier studies that focused on specific domains and were English-oriented, we will offer here a comprehensive approach to uncertainty detection, which can be easily adapted to the specific needs of many domains and languages. In our investigations, we will pay attention to create linguistically plausible models of uncertainty that will be exploited in creating manually annotated corpora that will serve as the base for the implementation of our uncertainty detectors for several domains, with the help of supervised machine learning techniques. Furthermore, we will also demonstrate that uncertainty detection can be fruitfully applied in a real-world application, namely, information extraction from clinical discharge summaries.