Using the Web Ontology Language (OWL) for knowledge representation in the legal domain is very promising but has some limitations. The language is complex thus hard to comprehend, still decidability results in a limited expressiveness which may introduce serious problems in modelling. An aspect of limited expressiveness is the tree-model property of OWL, which can be overcome using rule formalisms or introducing variables, however losing decidability is not always acceptable. We propose using conjunctive queries when modelling conditions of legal norms in the HARNESS architecture. Inference services required for modelling legislation and building legal assessment applications can be feasible using grounded queries resulting in a decidable formalism. Since most of the knowledge base remains within the limits of pure OWL2, we can benefit from consistency checking services of OWL2.
Using the Web Ontology Language (OWL) for knowledge representation in the legal domain is very promising as different types of inference services are provided on top of a relatively expressive formalism, the description logic underlying OWL2.
The main benefits are strict semantics, consistency checking and feasible reasoning. In contrast to this, OWL is very complex thus hard to comprehend, resulting in a knowledge acquisition bottleneck. Remaining decidable also costs limited expressiveness which may introduce serious problems in modelling.
An aspect of limited expressiveness can be described with the tree-model property of OWL: only tree-like axioms are allowed (except for nominals, transitive properties and role inclusion axioms in OWL2). Complex structures cannot be described precisely due to the lack of cycles in axioms or predicates with arbitrary number of arguments: only unary (classes) and binary (properties) predicates are allowed.
Representing diamond-shaped structures is a frequently reoccuring problem.
Suppose a sales contract with two actors – seller and customer – where the
subject of the transaction should be joined to both the seller and the customer.
Users familiar with rule formalisms tend to use rules or other extensions
supporting variables to overcome these limitations, although, this way decidable
satisfiability checking w.r.t. the T-Box is lost. In certain cases it is possible to
represent cyclic structures using knowledge patterns and OWL2, as described
in [
An interesting approach for describing complex structures in OWL is the
representation used in the HermiT reasoner [
In this article we propose an alternative solution using conjunctive queries to solve legal assessment problems in the HARNESS1 system. The next section introduces HARNESS, a legal knowledge-based system aimed at solving legal assessment problems. In the following two sections conjunctive queries are introduced and available inference services are described, including the case, when they are mixed with class expressions. In section 4.1. we show how to use conjunctive queries in HARNESS. The last section provides an overview of possible extensions and future plans. 2
A central task in legal knowledge-based systems is legal assessment : deciding whether some case is allowed or disallowed in a certain legal environment. In everyday situations a legal expert can help individuals to answer this sort of question, but due to the increasing size and complexity of legislation this process becomes more and more difficult, although transparency of jurisdictions would demand the opposite.
During the ESTRELLA2 project an open platform was developed for legal knowledge technologies, including the Legal Knowledge Interchange Format (LKIF), a reference open source legal CMS called eXistrella, an argumentation engine Carneades and a DL-based inference system called HARNESS. The architecture of HARNESS enables solving different tasks including drafting or legal planning, although it is currently aimed solely at legal assessment.
HARNESS greatly exploits current Semantic Web technology by relying on
formal ontologies and highly optimized DL reasoners. Legal assessment requires
three distinct types of knowledge: a domain ontology, normative knowledge and
case descriptions. The domain ontology defines the concepts and constraints in
the field of interest and provides building blocks for defining individual cases.
This ontology is a specialization of the LKIF Core ontology of basic legal concepts
[
Each norm is expressed as a generic situation in which a state of action is
qualified as undesirable, permitted or prescribed [
When modelling law in the HARNESS architecture using OWL the treemodel property of OWL hinders expressing complex situations. We will present an extension that will solve some of these issues: using conjunctive queries for specifying generic cases. 3
Conjunctive queries (CQ) are well known in database systems and have been
standing in the focus of DL research for years now but not yet widely available
in OWL applications. Practical results for complex DL languages have only
appeared recently [
A conjunctive query is a conjunction of concept expressions of the form C(t)
and role expressions of the form r(t, t′) where C is a concept, r is a role and t, t′
are terms, i.e. variables or individual names [
The following example demonstrates the usage of CQs. In Section 16. paragraph (2) in the Hungarian Law on Duties3 an exemption is specified for paying duties on a land received as a gift: “In order to verify completion of the construction of the residential house [. . . ] the state tax authority shall contact the competent building authority [. . . ] If the building authority provides a certificate in proof of the occupancy permit issued to the name of the property owner, the state tax authority shall cancel the duty assessed, but suspended in respect of payment.” 3 Hungarian Law on Duties, Act XCIII of 1990, only available in Hungarian, http://net.jogtar.hu/jr/gen/hjegy_doc.cgi?docid=99000093.TV#pr123 The condition of the exemption can be formalized using conjunctive queries the following way. If a gift (?g) is a plot of land, and a building (?b) has been built on it, which is a residential house, and an occupancy permit (?p) was issued to the name of the donee (?d), the generic case is fulfilled:
GCS16 2 ≡ Donation(?t) ∧ donee(?t, ?d) ∧ subject(?t, ?g) ∧ P lotOf Land(?g) ∧ built onto(?b, ?g) ∧ ResidentialHouse(?b) ∧ permit issued(?b, ?p) ∧ OccupancyP ermit(?p) ∧ issued to(?p, ?d) (1)
A conjunctive query can be represented by a graph where each variable in the query give rise to a node in the graph. Concept names appear as node labels, role names as edges at the appropriate variables in the graph. Figure 1 represents the graph for the CQ shown above. Just like OWL classes, different inference services can be implemented to conjunctive queries: query entailment is the decision problem to answer whether a query is true in all models of a knowledge base. query answering is the problem of finding all answer tuples for a query. If the entailment is false, there will be no answers. Otherwise, there may be one or more tuples fulfilling the constraints described in a query. satisfiability is to decide if a knowledge base has at least one model in which the query is true. When building a model and respective queries, a non-satisfiable query may indicate inconsistency in the model, as the corresponding legal condition will never be fulfilled. subsumption of CQs can be defined in a similar manner as class subsumption: with respect to a knowledge base K a query Q1 subsumes the query Q2 if in all models where Q2 is true, Q1 is also true. A more specific condition may correspond to a norm with higher priority based on lex specialis.
Not all of these problems are solved for the description logic underlying
OWL2. Satisfiability can be easily answered using a DL reasoner, but it is still
an open issue whether the other problems are decidable for the DL SHOIN (D).
Latest results showed that query entailment is decidable for SHIQ [
However in the general interpretation variables in a CQ are not required to correspond to a named individual in the ABox. For so-called non-distinguished variables only the existence of a suitable element is required in the model, and answer variables are required to have a corresponding named individual. This is important as in the restricted closed-world interpretation of CQs we only use answer variables, and then all inference problems are decidable in OWL2.
A subsumption hierarchy of CQs can be derived the same way as for OWL classes. Conjunctive queries are a generalization of OWL class expressions, as all class expression can be trivially transformed to an atomic CQ with one variable:
C → Q(x) ≡ C(x) As a result subsumption can be defined across CQs and OWL named classes, and hierarchy of CQs and classes can be merged. As an example for the CQ in equation 1: GCS16 2 ⊑ Donation.
Satisfiability of conjunctive queries can be derived from subsumption the same way as for OWL classes, by defining the always unsatisfiable CQ:
Q⊥(x) ≡ ⊥(x)
Q(. . .) is unsatisfiable ⇔ Q(. . .) ⊑ Q⊥(x)
Subsumption relations across CQs and OWL classes are important for designing tool support and knowledge engineering methodology. Conjunctive queries are not yet natively supported by major ontology editors but can be integrated into e.g. class hierarchy in a relatively straightforward manner giving confidence to knowledge engineers. 4.1
Application in HARNESS
In legal inference we have two distinct modelling issues: creating a domain
ontology for enforcing valid case descriptions and formalizing normative knowledge
in legal assessment. In the assessment part we are using the HARNESS
architecture [
The domain ontology must be consistent and all kind of inferences (including full consistency) are required, OWL is an adequate formalism. With GC descriptions, however, the only inferences required are hierarchy of GCs (is one description more general than another?) and case entailment (does a case fulfill all conditions of a GC?). The problem of modelling complex structures generally occur with GCs, so we will propose to use conjunctive queries in specifying generic cases.
We can use conjunctive queries to describe generic cases while the rest of the model is still specified in OWL2. In Figure 2 parts of a general HARNESS knowledge base are shown. LKIF-core and HARNESS base are standard ontology modules: providing basic legal concepts, the base class for norms and deontic operators. Domain concepts include terminological knowledge for specifying input case descriptions. The normative knowledge is provided in part Norms, allowing conjunctive queries for modelling conditions. Consistency of an input case specification can be verified using conventional DL tools and only evaluating the deontic reasoner requires conjunctive query evaluation.
Conjunctive queries are appropriate because inference services are available to cover the tasks required in HARNESS: – query answering matches input case descriptions with relevant norms, – subsumption relations provide exceptions (lex specialis) for norms in the model and – satisfiability ensures that each norm is consistent with the domain model.
An experimental implementation for the closed-world interpretation (using only answer variables) has been provided. The reasoner can use any DL reasoner (black-box reasoning) or Pellet4 and its highly optimized algorithms. The CQ reasoner can be accessed from Prot´eg´e 45 as a plug-in and can be selected instead other DL reasoners.
When using HARNESS, legal knowledge bases with conjunctive queries are supported. The generic situations in the normative part of the knowledge base 4 Pellet: an open source OWL 2 DL reasoner developed by Clark & Parsia, LLC http://clarkparsia.com/pellet/ 5 Prot´eg´e 4 ontology editor, http://protege.stanford.edu/ can be expressed both using OWL class expressions or conjunctive queries. Generic situations can be reviewed in a combined hierarchy following lex specialis relations and also showing all cases satisfying the condition. 5
An important feature of legal expert systems is the ability to provide reliable and
comprehensible explanations for inference results. For OWL this can be achieved
using a recent feature of the Pellet reasoner: laconic justifications [
Unfortunately description logics are hard to comprehend for the casual users, so when non-expert users should be able to interpret explanations, OWL axioms have to be translated to natural language or a graphical representation. The former can be achieved with NLP tools like ROO Rabbit [13] or Ace View [14]. We already took steps on adopting these services to handle queries and translate them to our target language, Hungarian. 13. Dimitrova, V., Denaux, R., Hart, G., Dolbear, C., Holt, I., Cohn, A.G.: Involving domain experts in authoring OWL ontologies. In Sheth, A.P., Staab, S., Dean, M., Paolucci, M., Maynard, D., Finin, T.W., Thirunarayan, K., eds.: International Semantic Web Conference. Volume 5318 of Lecture Notes in Computer Science., Springer (2008) 1–16 14. Kaljurand, K.: ACE View — an ontology and rule editor based on Attempto Controlled English. In: OWLED. (2008)