Due to the huge and continuously increasing amount of unstructured information, i.e. text, the analysis of legal literature has become a complex and time consuming task. Texts may refer to and are referred by other texts which can lead to complex dependencies. For example, a condition specified in a text maybe excepted in another one. Furthermore, there may be exceptions of exceptions. On the other side, computer science provides means that can support information processing. Natural language processing (NLP) technologies, for instance, evolved rapidly during the last years and several architectures for NLP systems have been proposed. Most of these architectures are general purpose architectures that are not tailored to any specific domain. One exemplary adaption for a domain can be found in bioinformatics where already some NLP applications exist that are capable of recognizing proteins, for instance.
This work presents the prototypical implementation of a web based NLP application tailored to the legal domain. For this, it is analyzed what requirements such an application should fulfill. These include functional requirements that apply for an NLP application in general as well as functional requirements that are specific for the legal domain. Furthermore, nonfunctional requirements, which mainly refer to the architecture of such an application, are presented. The aforementioned existing architectures are explored and assessed against those requirements. Therefore, it is analyzed what the key concepts of the respective architectures are and what consequences they imply regarding the conformance with these requirements.
Based on this assessment UIMA has been considered as the architecture that meets the requirements best and therefore, is the fundament of the implementation which is presented in chapter 6 Implementation of the workbench. It is explicated what features the application provides and how NLP technologies have been integrated to realize these features. Moreover, it is pointed out what the main challenges have been and what the limitations of the application are.
Finally, a critical review is provided where it is discussed which of the requirements the current implementation already fulfills.