Abstract :
[en] Information systems in several regulated domains (e.g., healthcare, taxation, labor)
must comply with the applicable laws and regulations. In order to demonstrate
compliance, several techniques can be used for assessing that such systems meet their
specified legal requirements. Since requirements analysts do not have the required
legal expertise, they often rely on the advisory of legal professionals. Hence, this
paramount activity is expensive as it involves numerous professionals. Add to this,
the communication gap between all the involved stakeholders: legal professionals,
requirements analysts and software engineers. Several techniques attempt to bridge
this communication gap by streamlining this process. A promising way to do so is
through the automation of legal semantic metadata extraction and legal requirements
elicitation from legal texts. Typically, one has to search legal texts for the relevant
information for the IT system at hand, extract the legal requirements entailed by these
legal statements that are pertinent to the IT system, and validate the conclusiveness
and correctness of the finalized set of legal requirements.
Nevertheless, the automation of legal text processing raises several challenges,
especially when applied to IT systems. Existing Natural Language Processing (NLP)
techniques are not built to handle the peculiarities of legal texts. On the one hand,
NLP techniques are far from perfect in handling several linguistic phenomena such as
anaphora, word sense disambiguation and delineating the addressee of the sentence.
Add to that, the performance of these NLP techniques decreases when applied to
foreign languages (other than English). On the other hand, legal text is far from being
identical to the formal language used in journalism. We note that the most prominent
NLP techniques are developed and tested against a selection of newspapers articles.
In addition, legal text introduces cross-references and legalese that are paramount
to proper legal analysis. Besides, there is still some work to be done concerning
topicalization, which we need to consider for the relevance of legal statements.
Existing techniques for streamlining the compliance checking of IT systems often
rely on code-like artifacts with no intuitive appeal to legal professionals. Subsequently,
one has no practical way to double-check with legal professionals that the elicited
legal requirements are indeed correct and complete regarding the IT system at hand.
Further, manually eliciting the legal requirements is an expensive, tedious and error-prone activity. The challenge is to propose a suitable knowledge representation that
can be easily understood by all the involved stakeholders but at the same time
remains cohesive and conclusive enough to enable the automation of legal requirements
elicitation.
In this dissertation, we investigate to which extent one can automate legal processing
in the Requirements Engineering context. We focus exclusively on legal requirements
elicitation for IT systems that have to conform to prescriptive regulations. All our
technical solutions have been developed and empirically evaluated in close collaboration
with a government entity.
