Sequence Time Expression Recognition in the Spanish Clinical Narrative

Abstract

Time expression recognition is one of the open issues in Natural Language Processing. These expressions are relevant to determine temporal aspects of the text as well as to establish relationships among facts described in said text. In the clinical domain, the temporal aspects are relevant to determine, for example, a sequence of facts in a clinical history. This paper presents research on the recognition of time expressions in Spanish according to the TIMEX3 standard. First, we establish HeidelTime, a well-known state of the art rule-based system, as a reference. Next, a hybrid model (a combination of bidirectional LSTM, CNN and CRF) is introduced to try to improve the results for the Spanish language. Both architectures have been tested with a Timex3 annotated Spanish corpus (TIMEBANK 1.0) to compare them. First, the results obtained show that the neural architecture obtains better results in Spanish. Finally, the neural architecture has been tested on a corpus of Clinical Notes (English and Spanish) in order to determine the results on this domain.