This paper introduces a methodology for the encoding of rules into a semantic logical format to facilitate the automated reasoning process. We demonstrate how to identify, capture, combine, and thus formulate all the components from rules into a computationally-oriented formalism. The need for the methodology is motivated by the desire for automated reasoning of automated vehicle information regarding tra c rules. We use Defeasible Deontic Logic as a formal foundation of our methodology. The overtaking tra c rule is our use-case to illustrate the usefulness of our methodology. Through this use-case, it is seen that the logical semantic representation of the tra c rules seems conceivable to support automated reasoning. This paper includes the source materials, the use cases, proposed methodology, and the example of encoding.
Road crash is a major concern of global public health due to the epidemic growth
of road fatalities. Every day, more than 3,7001 people die due to road crashes.
From world road crash statistics, it was found that the driver's behaviour is the
main contributing factor for 90% of these crashes [
Rule encoding is one of the important requirements for compliance checking,
automated reasoning, and legal validation [
Tra c rules are often detailed and complex and, therefore, it is a big challenge
to encode them. There have been several e orts to encode tra c rules for di erent
purposes [
Legal reasoning has some special features, which are norms and exceptions.
Usually, norms set baseline conditions but later, they are open to exceptions.
These exceptions are also expressed as norms. Norms prescribe behaviour using
permissions, obligations, and prohibitions. Therefore, we need a logical approach
to handle such exceptions in the tra c rules. In legal reasoning, DDL has been
successfully proposed [
There are several situations in con ict within the Queensland tra c rules, which may create problems for automatic tra c rule reasoning regarding AV. For example, in Tra c rules, Part 12, Division 3, Rule 170 says (2) A driver must not stop on a road within 20m from the nearest point of an intersecting road at an intersection with tra c lights, unless the driver| (a) stops at a place on a length of road, or in an area, to which a parking control sign applies; and (b) is permitted to stop at that place under this regulation.
the driver may stops at a particular place, or in a particular way because the condition of the driver, a passenger, or the driver's vehicle makes it necessary for the driver to stop in the interests of safety, and the driver stops for no longer than is necessary in the circumstances.
These two rules can contradict each other regarding stopping on a road. The
main purpose of these examples is not to show the di erence in tra c behaviour
but rather to nd the solution to follow the appropriate rule in such di erent
situations. These type of con icts often may be solved by the explicit priority
rules. However, Leens and Lucivero [
Tra
c Rules Encoding The proposed methodology consists of four modules as shown in Figure 2. Here, we introduce our methodology to translate the tra c rules into the semantic logical format. The input of the system is tra c rules (natural text) as shown in Figure 1(a). In the rst module, we de ne atoms from the rules. In the second module, we determine the norms. The if-then structure is identi ed and generated from the rules in the third module. Finally, we use Defeasible Deontic Logic (DDL) on the atoms, norms and if-then structure to make the semantic logical format. The modules are explained below. An example of making a semantic logical format of tra c rules is shown in Figure 1.
De ne Atoms This section outlines in brief that how we de ne atoms from rules. An atom is a predicate symbol including constants or variables that contains no logical connectives. Here to de ne atoms, we use terms of the rule sentence. A term is a variable or an individual constant in the sentence. This work deals with those variables and constants that refer to subject (s), predicate (p), property (pr), object (o), and quali er (q) (Figure 3) in the rule sentence.
In natural language a subject is refers to that term about which something is said in the sentence. The something which is said about something is the predicate of the sentence. The predicate of a subject-predicate sentence indicates relation or a property. The object is what that subject does something to. In another word, object are the results of action. Quali ers are that terms which usually enhance or limits the other word meaning. In one sense it can be said as an adverb of the sentence. Before generating terms, some article preprocessing are done on the text. Like, we are not considering the verbs (auxiliary, principal and modal) for this task. In logic, Subjects are variable or constant in the rule sentence that refers to the entity always. Predicate always refers to the properties or action of entities. Properties indicate the relation between subject and predicate. Object refers to the properties of the entities. Quali ers refer to the variable that enhances or limits the entities.
Atom is a combination of these terms and composed as a statement which only can be evaluated as true or false. For example \The bus breaks the tra c rule". According to the linguistic perspective, the term `bus' is the subject of the sentence as this sentence is about this. The term `tra c rule' is the object of the sentence as the subject is doing something to it. The term (verb) `breaks' is the predicate of the sentence as it expresses the relation between the subject (bus) and object (tra c rule). In logical approach, `bus' is that variable (subject) which refers the entity of the sentence. `breaks' is the predicate constant which refers the action of the entity. `tra c rule' is an individual of the sentence which refers the properties of the entity. So in a logical way, we can represent the above example as Predicate (subject, object): B(b,t): Breaks (bus, truck). Therefore the atom can be represented as Subject-Predicate-Object: bus Breaks theTra cRule.
In tra c rules, the rule structure are not equally structured. Due to this heterogeneity of the rules information, the atom structure varies. An example of identifying terms and de ning atom from the tra c rule is shown in Figure 1 (b) & (c). Throughout the empirical study of the Queensland Tra c Rules, we semantically de ne the atom in terms of ve aspects which are: Subject-PredicateObject; Subject-Predicate-Quali er-Object; Subject-Property, Subject-PredicateObject-Object; Subject-Quali er-Predicate-Object. Some examples are shown below.
Example 1 : Overtaking tra c rule section 151A: 3 (c) \the rider rides in a school zone".
Determining Norms Norms stipulate the conditions in the rule to perform
speci c actions. Every norm is represented by one or more rules, which could
either constitutive or prescriptive rules [
Prescriptive rules are determined based on conceptual semantic understanding and some special keywords, which are \must", \must not", \should", \ought", etc. For prescriptive rules, we consider only obligation, prohibition and permission norms. We identify the constitutive rules through our knowledge of understanding and descriptive notions in the sentences. Some examples of descriptive notions are \it is", \means", \does", \does not", etc. An example of determining prescriptive and constitutive norms from overtaking tra c rules, is shown Figure 4.
Generate if-then Structure Typically rules contain conditions and norms,
which control the behaviour of the subject. From a legal perspective, rules use
conditions on some actions to achieve particular behaviours. Rules can be analysed
based on the subject's behaviour and the circumstances [
if (X1; : : : ; Xn) then [norms] f Y g Here, (X1; : : : ; Xn) is the antecedent (premises), and Y is the conclusion (consequent) of the rule. In a tra c rules context, norms represent the situation of actions. Some actions are mandatory to follow; some are not. For example, \Rule 144A: Keeping a safe lateral distance when passing bicycle rider". The rule expresses an obligation for the driver while passing a bicycle rider. A simple example of generating an if-then structure is shown in Figure 5.
Rules Encoding Defeasible Deontic Logic (DDL) is the combined form of
defeasible logic and deontic logic. DDL can deal with both normative and
defeasible reasoning based on defeasible logic [
Defeasible theory consists of ve distinct knowledge foundations: strict rules,
facts, defeasible rules, superiority relations, and defeaters [
DL consists of a nite set of literals, where a literal is either an atomic proposition or its negation. Given a literal X, X denotes its complement. That is, if X = Y then X = :Y , and if X = :Y then X = Y .
Facts (F ) are conclusive and unambiguous statements. A fact is represented either in the form of state a airs (literal or modal literal) or actions that have been performed and are considered as always true. For example, \Honda is a Motorbike", is represented by: Motorbike(Honda).
A rule (r 2 R) describes the relationship between premises and conclusion and
we can specify the strength of this relationship. Based on the relationship strength
of the rules, we can di erentiate defeasible rules, strict rules and defeaters [
Strict rules are rules in the classical sense: whenever the premises are indisputable (e.g. a fact) so is the conclusion. For example, \a motorbike is vehicle," formally: Motorbike(Honda) ! Vehicle(Honda).
Defeasible rules are rules that can be defeated by contrary evidence. An example of a tra c rule is: \Part 11 Division 4 Section 151 B (3): the rider is taken to have unlawfully edge ltered along the length of road|(a) the rider does not hold an O type licence for the class of the motorbike". In summary, a motorbike can edge lter if the rider holds an O type licence, so formally we can write: Motorbike(Honda) ) EdgeFilteringVehicle(Honda). From this information, it can be concluded that a motorbike can edge lter on the road unless there is any evidence is provided that the motorbike cannot edge lter. A defeasible rule with empty premises would be considered as a presumption.
Defeaters are rules, that are used to prevent the conclusion. As an example, :(rider HoldOTypeLicence Vehicle(Honda)) ; :EdgeFilteringVehicle(Honda). From this rule we can state that Honda is an edge ltering vehicle but if the rider of a Honda does not hold a O type licence, then it cannot edge lter on the road. This statement can prevent the conclusion of edge ltering. This is not also supporting the no edge ltering.
Defeasible Logic is a non-monotonic, skeptical approach that does not support a contradictory conclusion. It aims to resolve the con icts between knowledge. For example, suppose there is information, and it has support to conclude A, but also there is information which does not support A and prevents it from concluding A. If the support for A has priority over :A , then it might be possible to conclude A. In such scenarios no conclusion can be made unless the rules are prioritized. The superiority relation ( ) used the priority set among the rules, where one rule may override the conclusion of other rules. An example of edge ltering concept using the Turnip Engine is shown in Table 1.
In Table 1, we can see that, no conclusive decision can be made until we use the superiority relation . Without using the superiority relation, we can only conclude that Honda is an Edge Filtering vehicle. After using the superiority relation like r1 r0 and r2 r1, then we can conclude that (as in the result section of Table 1) a professional rider can Edge Filter if he/she has a O Type Licence but learner or provisional rider cannot edge ltering.
In addition to defeasibility, tra c rules also engage with deontic concepts, which are obligation (O), permission (P), and prohibition (F). For example, considering the concept \overtake", we can de ne these notions as: [F]overtake [O]overtake [P]overtake [O]:overtake [F]:overtake :[O]:overtake
An example of representing tra c rules using DDL is given below. For this example, we use the Queensland Overtaking tra c rules (Part 11 Division 3 rule 140 in Tra c Rules).
; (Empty Set) ) [F] Overtake driver (HasClearViewOf approachingTra c) ^ driver (CanSafelyOvertake vehicle) ) [P] Overtake
A simple and complete example of encoding Overtaking Tra c Rules: 141 (1) using Defeasible Deontic Logic (DDL) is shown in Figure 1. 3
This work uses Defeasible Deontic Logic (DDL) to de ne a logical semantic
representation of tra c rules; however, there are some issues regarding the
accuracy and completeness of the representation. Given the sophisticated and
varied nature of tra c rules, identifying all the terms, norms, rule types and
conditions is a challenging task, as these components found in explicit or implicit
linguistic forms. As noted, the generic interpretation is not represented here. Only
ve di erent aspects of combination between subject, predicate, object, property,
and quali er are considered for de ning atoms. In terms of norms determination
from rules, only explicit types of norms (obligation, permission, and prohibition)
are considered, although there might be di erent types of permission and other
normative e ects [
Encoding the complex and varying nature of tra c rules is a challenging task. Any wrong encoding of the rule can adversely e ect on the reasoning process. Therefore, we proposed a Defeasible Deontic Logic (DDL) based encoding methodology for translating tra c rules into a semantic logical format (machine-computable), which can be used for automatic tra c rule reasoning to validate automated vehicle legal behaviour. The methodology incorporates the components and behaviour of regulations such as atoms (de ned from terms), norms, and if-then structure to analyse the rule content as well as identify the actions and activities of the rules. DDL is applied to the characteristics of the rules to resolve con icts and understand the norms and exceptions more explicitly. In the future, we plan to enhance the scale and scope of this proposed methodology. We intend to cover all possible combinations of terms and normative e ects for this task. Besides work on the overall tra c rules, we also plan to work on other domains to make this encoding methodology more e cient and standard in the eld of Law and AI research.