LawORDate: a Service for Distinguishing Legal References from Temporal Expressions

References to documents in the legal domain usually follow patterns containing temporal information in different forms (e.g. ’Directive 2001/29’). These references mislead algorithms detecting pure temporal references, and false positives occur in named entity recognition algorithms searching dates or intervals. This paper presents methods and techniques to identify these references, applied to two different domains. The first domain is that of news, where the temporal information plays a crucial role for their understanding and automatically building timelines can be hampered by the errors induced from these legal references. The second domain is that dataset descriptions. Dataset descriptions sometimes contain temporal information, not only in their dedicated metadata fields (e.g. dataset creation) but also within the text of their description. LawORDate, the system presented in this paper, is a web service able to detect legal references with temporal information in Spanish texts. The service identifies these references, avoiding their annotation by temporal taggers and enabling a further step of linking the references to the original sources and building co-reference graphs.