Large-Scale Legal Reasoning with Rules and Databases

tags: RuleRS Grigoris Antoniou Guido Governatori

Notes

underlying assumption of traditional legal knowledge representation and reasoning is that knowledge and data are both available in main memory.

NOTER_PAGE: (1 . 0.6985839514497639)

law has been a prime focus of attention as it is a rich domain full of explicit and implicit representation phenomena.

NOTER_PAGE: (2 . 0.20364126770060686)

the focus has always been on capturing elaborate knowledge phenomena while the data has always been small.

NOTER_PAGE: (2 . 0.35805799055967635)

Any standard reasoning system would reach its limits if data over longer periods of time need to be audited.

NOTER_PAGE: (2 . 0.6021577882670263)

building applications and property/site development are covered by a variety of local and national laws and regulations. To develop and assess relevant applications, it may be necessary to consider the legal requirements in conjunction with geodata relating to morphology of the site and its surroundings, use of space and so on.

NOTER_PAGE: (2 . 0.7909642616318273)

Stated another way, in the legal domain there is also a need for explainable artificial intelligence, as it has always been done in legal reasoning.

NOTER_PAGE: (3 . 0.35603506405933916)

A quite significant subset of legal representation and reasoning approaches relies on logic-based representation and rule-based reasoning.

NOTER_PAGE: (4 . 0.40323668240053945)

naturalness, which facilitates comprehension of the represented knowledge

NOTER_PAGE: (4 . 0.44302090357383683)

encoding of exceptions within a particular legislation, representing them explicitly by negative conditions in the rules. While this is suitable for self-contained and stable legislation, it may require some level of rewriting whenever previously unknown exceptions (or chains of exceptions) are introduced or discovered.

NOTER_PAGE: (5 . 0.3317599460552933)

deontic concepts such as permission or obligation which are a common occurrence in legislation, have to be represented explicitly within predicate names.

NOTER_PAGE: (5 . 0.40323668240053945)

well… it's not just about explicit names, that wouldn't really be much of a problem

several research efforts focused on exam- ining whether description logics and ontologies are suitable candidates for represent- ing and reasoning about legislation

NOTER_PAGE: (5 . 0.5569790964261632)

prime example of legal reasoning approaches using description logics is HARNESS [74] (also known as OWL Judge [75])

NOTER_PAGE: (5 . 0.8509777478084963)

decidable description logic reasoners such as Pellet can be exploited for legal reasoning, if, however, a significant compromise in terms of expressiveness is made. The most important issue is that relationships can only be expressed between concepts and not between individuals

NOTER_PAGE: (6 . 0.15104517869184086)

rules must be restricted to a so-called DL-safe subset

NOTER_PAGE: (6 . 0.2865812542144302)

non-monotonic logic for the purposes of legal reasoning.

NOTER_PAGE: (6 . 0.5091031692515172)

Defeasible Logic has been successfully used for legal reasoning applications [7, 33, 40, 35] and it has been proven that other formalisms used successfully for legal reasoning correspond to variants of Defeasible Logic [34]).

NOTER_PAGE: (6 . 0.5832771409305462)

philo- sophical criticism on deontic logic due to its admission of several paradoxes (e.g. the gentle murderer)

NOTER_PAGE: (6 . 0.7161159811193527)

Defeasible Deontic Logic has been successfully used for applications in legal reasoning and it is has been shown that it does not suffer from problems affect- ing other logics used for reasoning about norms and compliance [36, 35, 48]

NOTER_PAGE: (7 . 0.18543492919757248)

Defeasible Deontic Logic is a conceptually sound approach for the representation of regulations and at the same time, it offers a computationally feasible environment to reason about them [38].

NOTER_PAGE: (7 . 0.24342548887390425)

Rule-based legal reasoning approaches are more suited to legal systems that are primarily based on civil law, due to their inherent rule-based nature

NOTER_PAGE: (7 . 0.4308833445718139)

hoo boy

legal reasoning goes beyond the literal meaning of rules

NOTER_PAGE: (8 . 0.5846257585974376)

neither Carneades nor any ASPIC+ framework can be used as-is for legal reasoning

NOTER_PAGE: (8 . 0.7754551584625758)

Focusing on rule-based approaches specifically, regardless of their individual characteristics, two major issues have not yet been adequately addressed, to the best of our knowledge. These involve handling significantly large datasets and achieving efficient integration between legal rules and databases.

NOTER_PAGE: (9 . 0.24814565070802427)

The advantage of ASP is its expressiveness since it offers support for disjunction, strong negation and negation as failure and additional constructs such as aggrega- tion functions

NOTER_PAGE: (10 . 0.14632501685772084)

argumentation has significantly restricted expressiveness.

NOTER_PAGE: (10 . 0.18813216453135537)

defeasible logic seems to provide the best trade-off between expressiveness and complexity.

NOTER_PAGE: (10 . 0.2265677680377613)

The need to include rules to simulate the underlying semantics can result in a large number of rules, where the rules for the implementation of the logic exceed by far the rules cor- responding to the legal document to be encoded by the rule-base. In general, most of the additional rules are not used

NOTER_PAGE: (10 . 0.2973701955495617)

one has to perform some legal reasoning (for example to understand what are the actual requirements that apply in a given case) based on the information stored in enterprise databases.

NOTER_PAGE: (11 . 0.24814565070802427)

it could not be guaranteed that databases and rule-based systems are consistently amended.

NOTER_PAGE: (11 . 0.6581254214430209)

RuleRS [48], a possible solution where rules and databases are integrated.

NOTER_PAGE: (12 . 0.18071476736345246)

To enable their use with the rule engine used by RuleRS (SPINdle, see the next section) the rules are stored in the DFL format [51].

NOTER_PAGE: (13 . 0.4538098449089683)

since there is no direct correspondence between the literals en- coding rules and the table/attributes of the database schema, we have to establish a mapping among them to enable the integration of rules and instances in the database.

NOTER_PAGE: (13 . 0.5832771409305462)

each fact corresponds to an SQL/JSON query and a predicate is a statement that can be true or false depending on the value of its arguments/variables.

NOTER_PAGE: (13 . 0.6743088334457181)

last form may work well for a single database with small amount of rules but poses some significant challenges for large scale legal reasoning.

NOTER_PAGE: (15 . 0.7188132164531356)

the assumption of storing facts “in memory” does not hold for large scale reasoning

NOTER_PAGE: (17 . 0.1490222521915037)

the database is not capable of handling deontic concepts

NOTER_PAGE: (17 . 0.29062710721510454)

If such rules are treated as data and stored in databases, then the task of amending them if necessary becomes even harder

NOTER_PAGE: (17 . 0.4248145650708024)