This paper demonstrates the use of Logical English as a logic programming language that can be interpreted by the s(CASP) reasoner. It shows how legal knowledge and unknown information can be expressed in a form of English that can be easily understood by users without any technical training and with only a basic knowledge of English. This research has been undertaken in the context of the CrossJustice Project.
The goal of this paper is to explore the use of Logical English (LE) and s(CASP) in the CrossJustice system, and to assess their use both for drafting legal norms in logical form and for making the system more accessible for users who lack a technical background in computing or symbolic logic.
The focus of the CrossJustice Project[
Logical English (LE)[
An LE document consists of a knowledge base of facts and rules, scenarios, queries and templates. The templates are declarations of the predicates contained in the knowledge base and scenarios, such as
*a person* has *a right* according to *an article*.
A template identifies a predicate and its variable argument places. The argument places are identified by a simple noun phrase surrounded by asterisks and starting with an indefinite article "a" or "an". The predicate itself is represented by the rest of the template. The templates are used to identify instances of the predicates in the knowledge base and elsewhere. For example, the template above can be used to identify a sentence such as:
Mario has the right to interpretation according to article 3. as an instance of a predicate, which is translated into Prolog or s(CASP) resulting in the symbolic representation:
has_according_to(‘Mario’, ‘right_to_interpretation’, ‘Article_3’).
The translation can be processed by any Prolog or s(CASP) implementation. We use a SWI-Prolog SWISH Web Platform with support for LE and s(CASP) for this processing.
Logical English can be used to model legal norms in logical form. As argued, for example, by
[
In this paper we focus on the integration of LE and s(CASP) [
The combination of LE and s(CASP) seems to be mutually beneficial. For example, the common nouns used in the names of variables in LE can be used to assign types to arguments of predicates in s(CASP) explanations. In the above example this would result in s(CASP) explanations containing an explicit reference to: the person Mario.
Recent developments have also introduced initial support for multiple languages in LE. Work on this multi-language support can be a common efort useful to both LE and s(CASP).
The CrossJustice Project is an expert system containing a knowledge base about European and
national laws relating to the rights granted to suspect or accused persons. The system is now
available for public use [
The running example in this paper focuses on Articles 3(1), 3(2) and 3(3) of Directive 2010/64.
Article 3(1) states that suspected or accused persons are to be provided with a written translation
of all essential documents in a legal case, if they do not speak the language of the proceedings.
Articles 3(2) and 3(3) give examples of what constitutes an essential document, such as a decision
depriving a person of their liberty, a charge or indictment, or a judgement. In addition, Article
3(3) states that the court has the power to decree that any document may be considered essential
for the defence of the person. This open concept[
In our example, we have used the English language to translate the above-mentioned articles from the European directive into Logical English. However, LE potentially supports the drafting of programs in multiple natural languages. The goal of the following LE rule is to verify whether a person, such as Mario, has the right to translation of essential documents, according to Article 3.
1 a p e r s o n has the r i g h t t o t r a n s l a t i o n o f an e s s e n t i a l document a c c o r d i n g t o A r t i c l e 3 _1 2 i f the language o f the p r o c e e d i n g s o f the p e r s o n i s a g i v e n language 3 and a document needed by the p e r s o n i s r e c o g n i s e d as an e s s e n t i a l document 4 and i t i s not the c a s e t h a t 5 the p e r s o n u n d e r s t a n d s the g i v e n language .
The LE knowledge base also includes rules representing the the open list of essential document types, which are stated explicitly in the legal text.
In the interest of facilitating the translation from LE into s(CASP), we have expressed each essential document type by means of a separate rule, although in LE it is also possible to express the same information with a single rule, with each condition listed as an alternative using disjunction.
We have also implemented a rule, representing Article 1 of the Directive, which states that all rights referred to in the Directive shall apply to persons who are made aware by the competent authorities of a Member State, that they are suspected or accused of having committed a criminal ofence, from the start of the proceedings until its conclusion.
In order to reason with the times referred to in this rule, we have added an additional rule representing the event calculus for reasoning about the relationship between the time of an event and the times of the fluents (i.e. time-varying facts) that start or end upon the occurrence of the event.
It is clear from these examples of rules written in LE that, although still verbose, the use of natural language connectors (if, and, or) and the use of the same words in LE templates as in the legal source can help a person with no background in computer programming to understand the LE rules and thus to understand the legal conditions for the applicability of the rules. This understanding is closely related to what we referred to previously as an isomorphic representation of knowledge.
Finally, an LE document contains one or more scenarios and one or more queries. For example: Listing 2: A case scenario
The facts on lines 5 and 9 are used together with the event calculus to derive that the proceedings take place at every time between the start at 2000 − 05 − 13 and the end at 2001 − 01 − 08.
In an LE document the goal or goals are defined in a specific query section, and in this case the goal is to verify whether the user has the right to translation on a date, 2000 − 07 − 01, which comes after the start of the proceedings and before the proceedings have ended. In the query the variables are identified by keywords such as which (as well as by a and an), and they are instantiated in the answer. 3.1. s(CASP) translation In this section we remark upon some of the issues that arise concerning the translation of rules from LE into s(CASP). For example, the s(CASP) rules generated by this translation can be seen in listing 3.
1 h a s _ a c c o r d i n g _ t o ( A , t h e _ r i g h t _ t o _ t r a n s l a t i o n _ o f _ a n _ e s s e n t i a l _ d o c u m e n t , ’ A r t i c l e _ 3 _ 1 ’ ) : − 2 t h e _ l a n g u a g e _ o f _ t h e _ p r o c e e d i n g s _ o f _ i s ( A , B ) , 3 n e e d e d _ b y _ i s _ r e c o g n i s e d _ a s _ a n _ e s s e n t i a l _ d o c u m e n t ( _ , A ) , 4 not u n d e r s t a n d s ( A , B ) .
Here negation in LE is translated into negation by failure in s(CASP), which derives that
a condition does not hold if it cannot be shown that the condition does hold. In addition
to negation by failure, s(CASP) also supports reasoning with an explicit form of negation
which expresses that the opposite (or contrary) of a predicate does hold (i.e., so-called strong
negation)[
s(CASP) generates explanation trees in English using a directive #pred in the s(CASP) program to map predicates into readable English sentences. The translation of LE into s(CASP) translates the LE templates into these #pred directives.
One of the most interesting features of s(CASP) for legal reasoning is the possibility of
defining predicates as unknown (or "abducible"), which allows the user to state that a certain
fact is neither certainly true nor certainly false[
1 i t i s unknown whether the a u t h o r i t y has d e c i d e d t h a t 2 the document D1 needed by Mario c o n c e r n s e s s e n t i a l document .
The LE to s(CASP) translator produces the abducible statement of listing 5.
1 # a b d u c i b l e t h e _ a u t h o r i t y _ h a s _ d e c i d e d _ t h a t ( n e e d e d _ b y _ i s _ o f _ t y p e ( ’D1 ’ , ’ Mario ’ , e s s e n t i a l _ d o c u m e n t ) ) .
During the process of query evaluation, when the s(CASP) reasoner encounters an abducible condition as a subgoal, it checks whether the condition is explicitly negated by a constraint in the knowledge base. If it is not explicitly negated, then the reasoning continues, both under the assumption that the abducible subgoal is true, and under the alternative assumption that the abducible subgoal is not true. 3.1.1. Possible issues in the s(CASP) translation In defining the translation from LE to s(CASP), we have encountered a few limitations, which can be overcome by extending the s(CASP) parser and reasoner.
The first limitation of s(CASP) is that it admits only one negated literal in a negative condition. In contrast, in Prolog it is possible to negate a conjunction (or disjunction) of literals. This limitation is most likely derived from the necessity to easily generate duals for all the predicates defined in the knowledge base, which is the way s(CASP) implements constructive negation.
The second related limitation is that s(CASP) does not support disjunction in the body of rules. The use of disjunction in the body of rules can be avoided in general, by writing instead multiple rules with the same head, as we do in the definition of "essential document" in the running example. It would be useful and closer (more isomorphic) to the original legal text to allow disjunctions in the body of rules. But this could complicate the generation of dual predicates.
The third limitation is that explanations in s(CASP) do not translate meta-predicates, which are predicates that have other predicates as arguments, into natural language. For example, the event calculus employs a meta-predicate to express that a fluent holds at a time, where the lfuent itself is another predicate (in this case a predicate expressing that a person has the right to translation), and this fluent is not translated into English in s(CASP) explanations.
Finally, we compare explanations in LE with explanations in s(CASP). In both cases, as in other rule-based systems, explanations display the rules used in solving a goal during the course of a problem-solving session, and they are one of the most important requirements for expert AI systems in the legal domain. without abduction, and uses the same facts given in the scenario. The other explanation, in 2, uses abduction to assume that the judge exercises the discretionary power, expressed in Article 3(3), to deem that document D1 is essential to the defence of Mario. The use of abduction is highlighted in green.
Because meta-predicates are not translated into natural language in s(CASP), the resulting s(CASP) explanation is not easy for an untrained user to read.
The open source nature of both LE and s(CASP) facilitates a cross-fertilization of ideas and approaches of the kind presented in this paper.
On the one hand, certain features of Prolog (and thus LE) are not yet included in s(CASP), such as the inclusion of disjunction in the body of the rules. Although it is possible to represent complex rules without the use of disjunction, the inclusion of disjunction would enhance the ability of s(CASP) to represent legal texts isomorphically.
On the other hand, the ability of s(CASP) to reason abductively with predicates whose truth values are unknown, is very useful in legal reasoning, especially when considering the multiple sources of law and their validity relative to one another. Future work extending the translation from LE to Prolog, to include abduction, may prove fruitful, especially when combined with extensions to include reasoning by means of argumentation.
We are grateful to Jan Wielemaker for adapting the original s(CASP) system to run on the SWISH platform2, enabling us to do the work presented in this paper.
This work has been supported by the H2020 ERC Project “CompuLaw” (G.A. 833647), and the European Union’s Justice Programme (G.A. 847346) for the project “Knowledge, Advisory and Capacity Building Information Tool for Criminal Procedural Rights in Judicial Cooperation”.