BACKGROUND Towards Integrating the FLG Framework with the NLP Combinatory
Framework 1 School of Computing and Information Science, University of Maine, Orono, ME, USA
{amin.rabinia, sepideh.ghanavati}@maine.edu 2 Researcher at Fondazione Bruno Kessler, Trento, Italy
dragoni@fbk.eu Automatic modeling of privacy regulations is a highly demanded goal in requirements engineering. The FOL-based Legal-GRL (FLG) is a semi-automated goal-oriented modeling framework for extracting and representing the legal requirements of IT systems. One limitation of the FLG framework, however, is its manual requirements extraction process. Manual extraction of legal requirements is cumbersome, error-prone, and time-consuming. To overcome this shortcoming, we integrate this requirements modeling framework with another framework that combines several natural language processing (NLP) approaches. This Combinatory framework specifically exploits NLP techniques, such as Part-Of-Speech tagging and syntactic parsing, along with NLP tools, such as C&C and Boxer, to propose an automated approach for extraction of rules from legal texts. This approach enables us to fully automate the FLG framework in order to propose a comprehensive framework for modeling privacy regulations and other legal requirements. This paper outlines the two frameworks and their integration process.
Privacy regulations enforce a set of legal requirements for
any IT system. Developers need to capture and satisfy such
requirements in their products. Legal-GRL
The FLG framework consists of three phases: A. manual
requirements extraction; B. storing the requirements in a
database; and, C. requirements models generation. The
extraction phase (A) starts with collecting and analyzing natural
language legal statements. This is a manual task that legal
experts have to accomplish using a legal ontology. The
result of this process is a set of atomized and structured
statements called Restricted Natural Language Statements
(RNLSs). In the second phase (B), the RNLSs, which are in
FOL notation format, would be stored in an SQL database.1
This database is capable of exporting the data (or the
formalized legal requirements) as a formatted XML file. In the
third phase (C), the XML file is imported into the modeling
tool support, jUCMNav
To automate the first phase of the FLG framework, we integrate it with the NLP Combinatory framework. This framework combines two sets of NLP techniques: first, Stanford Parser and C&C/Boxer, for grammatical analysis of the sentences, and second, WordNet, for handling the deontic language of the legal texts. The input data, consisting of natural language statements from a legal text, pass through two separate pipelines that perform the NLP processes, mentioned above, to extract the rules from the original statements. The result from both branches of the pipeline will be combined to confirm each other.
Fig.1. shows a sketch of the integration of the two frameworks. The integrated framework would be organized in three phases: A. Legal requirements extraction, where the legal documents are imported into the NLP Combinatory pipeline and the RNLSs would be exported. The process inside the pipeline entails application of several ontologies along with the NLP techniques upon the input statements. B. Storing the requirements, where the RNLSs would be stored and finally retrieved as an XML/GRL file. C. Goal model generation, where the XML file is imported into the modeling tool support, jUCMNav, to generate the legal requirements goal models. Within the integrated framework, the entire process of modeling would be automated. This possibility, however, demands further modifications/improvements of the two frameworks, as they are still in progress. The integration process takes place in three layers: 1) A unified format of output/input needs to be set for a proper dataflow between the two frameworks. Since the FLG framework has an SQL Server database, it can import data from a variety of file formats including a plain text file. The RNLSs, then, can be stored in the NLP Combinatory framework as a text file and imported in the FLG database.
2) The FLG framework requires a finer-grained analysis of the rules than what the NLP Combinatory framework provides as its output. The general format of such output is: Term1 => [O] Term2, where terms are parts of a sentence or statement. However, the final goal models of the FLG need further details of the requirements, e.g. actors. These elements of the statements are accessible within the NLP Combinatory framework, as it exploits the NLP techniques..
3) The FLG framework has also a legal ontology that helps simplifying the legal requirements models. During the integration process, the ontologies of the two frameworks should be integrated in order to handle syntactical modifications of the statements.
We perform the integration procedure in an iterative, nonsequential fashion. The integration of ontologies (3) should be the first priority, since it contributes to the second layer. Running the dataflow (1) between the two frameworks also can be done afterwards, when the data format is matched during 2 and 3.
The outcome of this integration is a fully automated framework for modeling of privacy requirements.