Time and norms: a formalisation in the event-calculus RAFAÉL HERNÁNDEZ MARÍN - GIOVANNI SARTOR Introduction Temporal concerns are ubiquitous in the law, and have been addressed in various ways by legal doctrine and legal theory (e.g., when dealing with periods, deadlines and conditions, or with prescription and usucapion or with the creation and derogation of legal norms). However, very few researches has expressly and directly addressed the link between law and time in a formal framework (cf. for all, Bulygin 1982). This gap has recently been addressed by AI & law research (cf. for all Mackaay et al. 1990), which has been employing various AI formalisms for addressing the temporal aspects involved in representing and processing legal knowledge. Among those formalisms a special place must be recognised to the event calculus both for its representative power (which makes it intuitively suitable for many legal contexts) and for its simplicity (which makes it easily accessible also to an audience having a limited formal training). The event calculus has promoted many studies in AI, where the seminal paper by Kowalski and Sergot (1986) has been followed by a large number of contributions, providing various axiomatisations, extensions and applications. There have been a few event-calculus applications in the legal domain (cf. Provetti 1992, Sergot 1995), but we think that the resources provided by the event-calculus could provide further insights into legal issues, and possibly support some useful applications. Our paper will not move into applications (although it includes a Prolog program) but it will link between an analysis of certain temporal aspects of the law, and the definition of a computable representation. 1. External and internal time of legal norms In our analysis we will use the language of first order logic, including the usual propositional connectives and quantifiers (this approach was advocated, among others, by Quine 1960). We are aware that philosophical logic and artificial intelligence (cf. for all Thomason 1984) has developed many elegant and sophisticated formalisms for dealing with time. However, we prefer to limit ourselves to first order logic since this representational device is sufficient for our concerns, and it allows as to express most clearly the temporal structures in which we are interested. In addition to the resources of first order logic, we use a special connective, , to express normative implication (the relation between the antecedent and the consequent in conditional norms)1. Sometimes we will write predicates as sequences of words and will put individual terms (variable or constants) within the corresponding predicate, in order to obtain a natural language-like formulation. We assume a discrete representation of time, that is we represent time as an infinite succession of discrete instants ... Tj ... Tk ... We denote as Ti+1 (Ti-1) the instant immediately following (preceding) Ti. We only consider conditional norms, that is those having the form Condition Effect, where the effect is assumed to be produced or brought about by the realisation of the condition. In particular, we focus on event-state norms, that is norms which establish that a certain legal state (or property) starts to hold when a certain event takes place. To indicate that event E happens at time T we write at T, E and similarly, to indicate that situation S holds at T, we write at T, S. So, for example, the norm XT ( at T, X is born in Italy at T, X is an Italian citizen ) establishes a new situation (qualification), the Italian citizenship of X starting at time T, as a consequence of the event of X's birth in Italy exactly at that time (rules on Italian citizenship are in fact more complex than this suggests, but the example suffices for present purposes). Deontic norms, conferring duties or permissions, can also be represented as event-state norms. For example the norm XT (at T, X damages Y => at T, X has the duty to repay Y ) establishes new instances of a normative state (e.g., Philip being obliged to repay Peter) as a consequence of an instance of an event-condition (e.g., Philip damaging Peter). We do not address the issue of the logic of deontic modalities. We do not consider the possibility of providing an event-calculus account of norms with are not reducible to the event-state form, and in particular of norms having an eventevent structure. While leaving this to future research, we prefer to focus here on the eventstate structure, which is sufficient for dealing with the examples here considered. Our analysis will be centred upon the distinction of two temporal profiles in legal norms. 1We do not address the difficult problem of the meaning of normative implication. For our purposes it is sufficient to assume that behaves as a conditional connective in modus ponens inferences (from p q and p, we derive q). 2 One is the so-called external time) of a legal norm, i.e., the answer to the question: during what time-interval(s) has the norm been included in the legal system? This question concerns, in other terms, the interval during which the legal norm has been valid (by validity of a norm we mean here its partaking to the legal system). The other time is the so-called internal time of a legal norm, which includes all temporal issues which are involved in the application of a valid norm. In particular it deals with the answer to the question: what time-interval(s) was the norm intended to cover? This question concerns the interval in which the conditions contemplated by the norm should happen in order that they will produce the norm's effect, i.e., the subsumption interval or the norm (cf. Hernández Marín 1996, 29-30). Interestingly, those legal notions present a similarity (which we cannot examine here) with certain notions which have been developed within research on databases, and in particular with the notions of 'valid time' (the time at which something occurs) and 'transaction time' (the time at which this is recorded in the database or belief system). 2. The internal time of legal norms (subsumption time) In the following paragraphs we will try to make the best use of our rather scanty formalism, by applying it to the different aspects involved in what we called the internal time of legal norms. 2.1. Universal temporal quantification covering both condition and effect The temporal structure more frequent in legislation is characterised by a universal time quantifier covering the whole norm (both the condition and the effect). Those norms are temporally abstract: whatever instance of the condition produces a corresponding instance of the effect. They do not include any specific limitation of their subsumption interval. This does not mean that their subsumption interval necessarily covers all the past and the future, since it may be limited according to general rules or principles, as we shall see in the following. Let us consider first a personally individualised norm, which is temporally abstract. Thus, for example, the father of John, in a morally questionable, but probably effective, attempt to stimulate his child's school performance, can issue the following promise: whenever you get a good mark, you are entitled to receive £ 1. This will amount to the formalisation T( at T, John gets a good mark at T, John is entitled to receive £ 1 ) 3 Temporal abstractness is the most sensible interpretation for all those cases in which no temporal dimension is expressly specified in legislative language. So, for example, the norm which says “he who is born from Italian parents is an Italian citizen”, would stay for T X( at T, X is born of Italian parents at T, X is an Italian citizen ) Such a general-abstract norm is doubly quantified, over instants (abstractness) and over persons (generality). The two quantifiers cover both the condition and the effect. 2.2. Partial abstractness The temporal domain of a norm may explicitly be restricted to a specified interval. Events falling outside the specified time interval do not satisfy the norm condition, and therefore do not trigger the norm effect. Let us consider for example a rule establishing a £ 10 fine for anyone who parks in front of the station from 10/6/1997 to 16/6/1997 (perhaps, for example, a sporting event is taking place over those days). This could be formalised as: T {10/6/1997 T 16/6/1997} X ( at T, X parks in front of the station at T, X is liable to a £ 10 fine ). 2.3. Existential temporal quantification of the condition or the effect Some prescriptions contemplate events which must happen within a certain interval of time. The temporal structure of the condition of those norms can be captured by means of an existential quantifier covering only the antecedent (which corresponds to a universal quantifier covering the whole norm, when the quantified variable does not occur in the consequent). Consider, for example, an insurance rule saying that if one causes any accident during 1997, one's insurance fee is doubled from the beginning of 1998: X [ T(1/1/1997 T 31/12/1997 at T, X causes an accident) at 1/1/1998, X’s insurance fee is double what it was] In order to satisfy the condition of such a norm, it is sufficient that a person causes just one accident in one instant falling within the given time interval (he/she does not need to continue causing accidents during the whole year). It may be doubted whether norms exist having an effect which is existentially quantified in its temporal variable. Such norms would establish that a certain legal effect is to happen within a certain interval, without establishing the exact moment in which the effect is to take place. Consider for example the absurdity of a norm stating that if one marries a citizen, then one shall acquire citizenship at some moment included in the following year (such a norm does not tell when this is going to happen, within the considered year): 4 X Y T1 [ ( at T1, X marries Y at T1, Y is an Italian citizen ) T2 ( T1 T2 T1 + 1 year at T2, X is an Italian citizen ) ]. Norms prescribing that a certain performance is to be accomplished within a certain time length do not fall within the (empty) category of the norms having a temporally undetermined effect. On the contrary, those norms establish their effect, i.e., the beginning of the obligation, for a determined moment, although the obligation concerns an action to be performed in any moment within a certain interval. Therefore, in such norms the time of the effect is determined, although the time of the performance is partially undetermined. Consider for example the following clause, agreed by John to Tom: if John gives Tom £ 100 today, then Tom is obliged to give John back the money tomorrow. In our model, such a clause would be represented as follows (assuming that today is the 14/4/1997). T1 {0am on 14/4/1997 T1 < 0am on 15/4/1997} ( at T1, Tom gives John £ 100 at T1, it_is_obligatory_that [ T2 (0am 15/4/1997 T2 < 0am 16/4/1997 at T2, John gives Tom back £ 100)] Note that in such a case the effect, i.e., the birth of the obligation, is determinate and indeed immediate (it takes place at T1), though the time of performance is only fixed within a subsequent interval. It would have been wrong to formalise the consequent of this sentence as T2 [ 0am on 15/4/1997 T2 < 0am on 16/4/1997 it_is_obligatory_that ( at T2, John gives Tom back £ 100 ) ] putting the whole effect (the deontic operator included) within the scope of the temporal quantifier. This would mean that there exists a precise instant T2 (although the sentence does not tell the time of T2) in which the performance should take place (performance in any other instant would be irrelevant). On the contrary, the clause is indifferent to the specific instant of Tom’s performance, provided that his performance occurs within the indicated interval. 2.4. Instantaneous conditions or effects Norms that specifically indicate the unique precise time when their condition is supposed to happen are very rare, if they exist at all. Let us thus consider a quite artificial example: a 5 contract states that if the buyer, Mr Smith, pays the price today, at 3 p.m. on 15/7/1997, he becomes the owner of the house of the seller (Mrs Jones). Such a sentence could be represented as: at 3 p.m. on 15/7/1997, Mr Smith pays the price at 3 p.m. on 15/7/1997, Mr Smith is the owner of the house of Mrs Jones Note that we have specified both the time of the condition and the time of the effect (even if the latter is not explicitly indicated in the corresponding natural language expression). Note that even instantaneous norms may include persistency assumptions (it is very hard to express any legal content without those assumptions). The norm just says that Mr Smith becomes the owner of the house at 3 p.m. on 15/7/1977, but we also assume that he remains its owner subsequently. Only if an event terminating Smith’s ownership happens, will this assumption be abandoned. 2.5. Displaced effect Certain norms may establish their effect for a time which is earlier or later than their condition. In this regard let us consider first norms establishing a non deontic qualification, i.e., the rule as according to which norms become applicable (come into force) 15 days after their publication. The effect established by this rule (the becoming applicable of another norm) is established to happen at time which is 15 days after the time when the condition contemplated by this rule (the publication of that norm) took place: TN ( at T, N is published at T + 15 days, N is applicable) Similarly, the effect may precede the condition, as in the norm stating that the recognised son is such from his birth: T1T2 { XY[ ( at T1, X is born at T2, Y recognises Y as his son ) at T1, X is son of Y ] }. In deontic norms, temporal displacement may just concern the performance of the prescribed action. Consider for example the rule that Paul must wash his hands after dirtying them: T1 ( at T1, Paul dirties his hands at T1, it_is_obligatory_that [ Te ( at T2, Paul washes his hands T2 > T1 ) ]. 6 In this formalisation the effect (the obligation) is contemporary with the realisation of the norm condition (at T2), although the prescribed action was to be performed before. 2.6. Analytical and synthetical representations of the internal time Legal language may adopt two quite different approaches in the representation of internal time, and in particular, for delimiting the subsumption interval. 1. a synthetical approach, in which all temporal and substantial elements of the norm are represented within the same sentence; 2. an analytical approach in which one sentence represents the substantive content of the norm, and other sentences specify its temporal features. In particular the analytical approach is usually adopted for specifying what we have so far called the subsumption interval of the norm. In the above example, instead of having just one sentence: 1. During the period from 10/6/1997 to 16/6/1997, anyone who parks in front of the station is liable to a fine, we could have had two sentences: 1. Anyone who parks in front of the station is liable to a fine. 2. Norm 1 is applicable2 (applies, has effect ...) from 10/6/1997 to 16/6/1997. or even three sentences: 1. Anyone who parks in front of the station is liable to a fine. 2. Norm 1 starts to be applicable at 10/6/1997. 3. Norm 1 ceases to be applicable at 16/6/1997. In analytical representations, norms must be able to speak about other norms. This means that we must have a naming convention. We shall simply put in front of each rule a univocal label to be used as its name. Let us adopt for the norm above the name pfs 2Throughout the rest of the paper we will say that a norm is ‘applicable' at a moment T to mean that T falls within its subsumption interval (i.e., that the norm is concerned with events happening at time T). Therefore, norm 2 states that the subsumption interval of norm 1 starts at T1, by prescribing that norm 1 starts to be applicable at T1. Similarly, norm 3 states that this interval ends at T2 by prescribing that norm 1 ceases to be applicable at T2. By applicable we just mean "temporally applicable" in the sense of "capable of subsuming". We do not address here other aspects of the "application" of legal norms, and are using the term applicable in this sense just because we could not find any appropriate (and not too awkward) more specific term. 7 (parking in front of the station). So, we translate the three sentences above into the following: pfs: XT1( at T1, X parks in front of the station at T1, X is liable to a fine ). pfsA1: at 10/6/1997, pfs becomes applicable pfsA2: at 16/6/1997, pfs ceases to be applicable The analytical approach has a number of advantages. Firstly, it is modular and clear, since different items of information are expressed in different sentences. This advantage is particularly significant when using formal languages: we can avoid writing awkward complex formulae. Secondly, and most importantly, temporal elements do not need to be always directly specified, but they can be made dependant upon future and possibly not yet known facts. So, a norm N1 may be accompanied by a norm N2 stating that N1 is going to become applicable when a certain authority adopts a certain implementation measure or when a certain event takes place (e.g., a situation of emergency). In particular, general rules on temporal features can be issued, such as those stating that a period after its enactment must elapse, for any norm to become applicable. For example, the general rule establishing that statutory norms become applicable 15 days after their being published may be represented as follows publication: NT( at T, N is published at T+15 days, N is applicable ). Analytical representations of time have some drawbacks. In particular, more than one sentence needs to be considered in order to determine both the substantial content of the norm and its subsumption interval. Moreover, inferences from analytical representations must take into account the interaction between substantive norms and norms which regulate subsumption intervals. However, we stick here to the analytical mode since our purpose is to model the temporal structures usually adopted in legal language and in legal reasoning. 3. Time and validity The notion of applicability of a norm, as defined above (let us recall that a norm is applicable during its subsumption interval) must be distinguished by the notion of the validity of a norm, by which we mean its inclusion in the considered legal system. A norm may be valid, but not yet be applicable, as when a norm is intended to cover only events happening some time after it is issued, or a norm may be applicable to events which preceded it validity, as when it is given retroactive effect. 8 All legal norms, or at least all positive legal norms, have a temporally restricted validity: the validity of a norm N starts at a certain time T1 and may terminate at a subsequent time T2 (for reasons of simplicity we do not consider the possibility that a norm lasts longer that one validity interval). By saying that <T1, T2> is the (maximal) validity interval of N we mean that: 1. for all T such that T1 T T2, N is valid at T, 2. N is not valid in T1-1, 3. N is not valid in T2+1. The beginning of the validity interval of any legal norm is determined by other rules (which may be explicit or implicit, and may be legal or conventional, as we shall see in the following section). For example, in some legal systems, such as Italy or Spain, a general rule establishes that the validity of legal norms begins at the time of the publication. Such a rule can be obtained by a (very liberal) interpretation of the Constitutional provisions ensuring the publication of norms. Alternatively, it can be considered a generally accepted basic convention, implicit in legal practice (corresponding to the common sense convention or moral requirement that every prescription only holds from the moment in which it is made accessible to its addressees). The end of the validity interval is established by derogation rules (which also may be explicit or implicit and may be legal or conventional). Specific legislative sentences are usually required for explicit derogation of determined norms. Among general derogation rules we can just mention those establishing that more recent norms tacitly derogate previously valid ones in the case of any contradiction between them. The usual treatment of validity in legal reasoning is based upon a persistency assumption. The fact the N is valid throughout the interval <T1, T2> cannot be derived just from the fact that N starts to be valid at T1 and ceases to be valid at T2. Those sentences just tell us that N is valid at T1 (and was not valid at T1-1) and that N is not valid at T2+1 (and was valid at T2). They do not tell us anything about all instants T such that T1<T<T2. To establish the validity of N in all those instants, we need the assumption that, once N becomes valid, it continues to be valid until it is deprived of its validity. Such an assumption is part of “legal common sense” (including various tacit conventions which underlie legal practice and legal science). 3. Rules attributing applicability and validity The representation we have adopted seems to be undermined by an infinite regress in search for both applicability and validity. As it is well known, the recursive circularity of validity has been frequently considered by legal theorists (and especially by Hans Kelsen). A norm N1 is valid only if it is qualified as valid by a norm N2 which is valid (and applicable); N2 9 is valid only if it is qualified as valid by a norm N3 which is valid (and applicable); N3 is valid ... In our framework exactly the same problem emerges as far as applicability is concerned. N1 is applicable only if it is qualified as applicable by a norm N2 which is (valid and) applicable; N2 is applicable only if it is qualified as such by a norm N3 which is (valid and) applicable; N3 is applicable... One solution to both infinite regresses consists in observing that the legal system is underpinned by social-linguistic-moral rules (conventions) which do not partake in the legal system itself, and therefore do not require the qualifications of legal validity and legal applicability. Those rules allow us to stop the recursion: N1 is valid (applicable) because it is so qualified by the non legal rule N2, for which it makes no sense to consider the requirement of legal validity and applicability. Those underpinning rules can obviously be assimilated to Hart’s recognition rule (Hart 1961) or to Kelsen’s basic norm of (Kelsen 1960). However, we do not insist that there be just one such rules, and deny that those norms are legal. As an example of such rules, let us consider the following NT ( at T, N is published at T, N is valid ) which makes validity dependant upon publication according to the view of Hernandez Marín (1966, 36ff.). 4. A formalisation in the event calculus In the following paragraphs, we shall consider the possibility of translating our temporal analysis of legislation into a formalism which is computable, i.e., into a computer program which can automatically perform temporal reasoning. Basic event calculus The axioms of the event calculus are logic programming clauses, and can be executed by a Prolog interpreter. Those axioms are therefore rules of this form (we use the syntax usually adopted for Prolog): c :- p1, , pn. where :- represent the reversed conditional, “,” is the conjunction, c is an atomic formula and the p1 pn are literals (atomic formulae or negations of atomic formulae). Negation is to be understood as negation by failure (which we denote as not), rather then as classical logical negation. This means that the formula not p does not assert that p is false, 10 but rather that there is no evidence that p, i.e., that p is not derivable from the knowledge represented in the program. The basic concept of the event calculus it that of event. Events initiate or terminate the periods in which certain states of affairs (properties, relations, situations, etc.) hold. Therefore two type of domain knowledge must be provided in the event calculus: 1. specific sentences that certain event-instances have happened, 2. general rules indicating what event-types generate what states of affairs. In our formalisation the difference between event-instances and event-types is simply indicated by the fact that descriptions of event-types contain free variables and descriptions of event-instances contain only constants. For example X is_born_in_Italy refers to an event type while giovanni is_born_in_Italy refers to an event-instance (from now on we will follow the usual Prolog convention of writing individual variable as sequences of characters starting with upper case letters and both individual and predicate constants as sequences starting with lowercase letters). General rules express the capability of any instance of a certain event-type to initiate or terminate the corresponding instance of a certain state-type: X is_born_in_Italy initiates X is_Italian. X acquires_non_italian_citizenship terminates X is_Italian. The event calculus has an in-built capacity to deal with persistence. The following law of inertia is in fact assumed: states of affairs which were started by one event are assumed to persist in the future indefinitely, unless they are interrupted by a terminating event. For our purposes, an elementary treatment of persistency is sufficient, since for us: 1. Only persistency in the future is relevant (we are not interested in deriving past states of affairs from present ones). 2. Each event is expressly located in a precise moment. Let us now introduce a formalisation of the event calculus. The first rule establishes that the state of affairs S holds at time T2 if S was initiated at a time T1, antecedent to T2 and there is no evidence (negation by failure) that the persistency of S between T1 and T2 was broken (interrupted)3. holds_at(S, T2) :initiated_at(S, T1), T1 =< T24, 3What happens to S before T1 or after T2 is irrelevant for its persistence between T1 and T2. 4We write =< instead as , as is usual in programming languages. 11 not broken(S, between(T1, T2)). The second rule establishes that state S was broken between T1 and T2, if S was terminated between T1 and T2: broken(S, between(T1, T2)):terminated_at(S, T), T1 =<T, T < T2. The third rule concerns initiation. It established that a state S initiates at time T if an event able to initiate S happens at T: initiated_at(S, T) :E initiates S, happens_at(E, T). The fourth rule has a similar function as far as termination is concerned: terminated_at(S, T) :E terminates S, happens_at(E, T). Limits of the event calculus in the legal domain Let us apply our event calculus formalisation to the following premises: happens_at(giovanni is_born_in_italy, 25/02/59). X is_born_in_italy initiates X is_italian. X acquires_non_italian_citizenship terminates X is_italian. From such premises (and the above axioms of event calculus) we can derive that Giovanni is Italian in any moment subsequent to the 25/02/59. Let us now add the information: happens_at(giovanni acquires_non_italian_citizenship, 15/6/1985). Given this additional premise we are no longer able to derive holds_at(giovanni is_italian, 15/9/1997), since the state giovanni is_italian was interrupted between 25/02/1959 and 15/9/1997 (precisely at 15/6/1985). On the other hand it is still possible to derive that Giovanni is Italian at any moment between 25/02/1959 and 15/6/1985. As this example makes clear, the formalisation of legal norms as initiation or termination rules captures the assumption already mentioned when event-state norms were introduced: from the lawyers’ perspective, legal states of affairs (for example, an 12 obligation) come into existence as a consequence of certain facts (e.g., any fact usually listed among the sources of obligations) and cease to exist as a consequence of other facts (e.g., any fact usually listed among the causes extinguishing obligations). 5. Some extensions of the event calculus The representation of legal norms in the event calculus suffers from some limitations. Let us address three major restrictions 1. Event calculus assumes that all norms are temporally universal, whereas legal rules may have a (partially of totally) limited subsumption interval. 2. Event calculus initiate and terminate rules are assumed to be always valid, whereas legal rules are limited in their temporal validity. 3. Event calculus makes a state start as soon as the conditioning event happens, whereas a legal effect may precede or follow the corresponding condition. In the following paragraphs we will overcome all those limitations, by introducing a representation formalism and an inference mechanism especially devised for legal norms. These will integrate the basic representation and inference mechanism of event calculus. Temporally limited applicability According to the analytical approach, we will preserve the basic structure of event calculus for the substantial content of legal norms, and will regulate separately their subsumption interval (applicability). For this purpose first need to provide the syntax of legal norms with a naming method. Let each legal rule be preceded by his name, in the form: N: Condition l_initiates Effect. where N is the name of the rule and l_initiates (legally initiates) expresses the relation between legal conditions and legal effect. Note that the only syntactical difference between legal rules and common event calculus rules is just the fact that the legal rules are given names. The subsuming capacity of a legal norm N1 needs to be conferred by a (legal or non legal) rule N2, which fixes the beginning of the subsumption interval. However, the syntax we have so far provided for event-state norms, does not provide any intuitive way for dealing for unconditioned norms. In fact unconditioned norms provide no event, and therefore cannot rely on any event-time for determining the time of the effect. Let us consider, for example, the norm stating: The exception to the Data Protection Act (named edpa) is applicable from the 1/1/1988”. We express this norm in a state-only form (the event is missing): 13 edpaA1: is_applicable(edpa) l_initiated_at 1/1/1988. Similarly, for indicating that the applicability of a norm N1 terminates at a certain time, for example that the application of the exception to the Data protection act ends on the 10/5/1990, we introduce an additional norm stating that: edpaA1: is_applicable(edpa) l_terminated_at 1/1/1988. Let us consider finally the norm concerning the fine for parking in front of the station from 10/6/1997 to 20/6/1997, that is: T{10/6/1997 T 20/6/1997} [ X ( at T, X parks in front of the station at T, X is liable to a fine ) ]. In our formalism, this norm (which we call pfs) is represented as the combination of three sentences. pfs: X parks_in_front_of_the_station l_initiates X is_liable_to_a_fine. pfsA1: is_applicable(pfs) l_initiated_at (10/6/1997). pfsA2: is_applicable(pfs) l_terminated_at (20/6/1997). Now we extend our reasoning mechanism in order to take into account this extension of our knowledge representation. Besides the initiate_at rule for general event calculus, we introduce an additional initiated_at rule specifically intended for legal rules (we do the same for the terminated_at rule): initiated_at(S, T):N: E l_initiates S, happens_at(E, T), holds_at(is_applicable(N), T). Let us analyse this clause. For a legal state S to start, it is not only required that there is a legal norm N according to which event E is able to produce state S, and that an instance of event E takes place. In addition, it is also required that N is applicable at T (i.e., that T is included in the subsumption interval of N). Similarly, we have to introduce a clause for unconditioned norms. 14 initiated_at(S, T):N: S l_initated_at T, holds_at(is_applicable(N), T). In order to avoid an infinite regress we need some statements on applicability that are not analytically represented legal norm. For example, a statement initiated_at(is_applicable(n0), 10/6/1900) gives applicability (from outside the legal system) to a legal norm n0, which might possibly legally initiate the applicability of other legal norms. Validity in time With validity we proceed as with subsumption capacity. Rules conferring validity are initiation sentences: pfsV1: valid(pfs) l_initiated_at 1/1/1997. Similarly, derogation norms are termination sentences, specific or general: pfsV2: valid(pfs) l_terminated_at 10/5/1998 derogation_by_conflict: N1 conflicts_with_subsequent_norm N2 l_terminates valid(N1). Obviously, we need some basic validity rules (at least one) which are not required to be valid, in order to produce their effect (the qualification of some other norm as valid). For example, we could just assume that the fundamental rule for validity is the following, which makes validity start at publication (note that this is not a legal norm, and the production of its effect is not conditioned on its being valid and applicable)5: published(N) initiates valid(N). To take into account validity in our reasoning mechanism we need a modification of the legal version of the initiated_at (and of the corresponding terminated_at clause). 5This representation could be easily extended to enable the representation of both the event and the concomitant conditions necessary for it to produce its effect (cf. Sergot 1995). This would allow us to formulate validity norms according to which the event of publication produces validity only if certain requirements are satisfied, which can be different in different legal systems and according to different legal theories (for example, being issued by a legally empowered authority, respecting certain procedural or substantial rules ...). 15 initiated_at(S, T1):N: E l_initiates S, happens_at(E, T1), holds_at(is_applicable(N), T1), holds_at_or_started_after(valid(N), T1). Let us analyse this clause. As in the formalisation introduced above, it requires that: 1. there is norm N which states that event E is able to produce state S, 2. an instance of event E takes place, and 3. N has subsumption capacity at the time when the event took place. Moreover it also requires that: 4. N is valid at the time of the event or starts to be valid after that time. This last requirement corresponds to the fact that invalidity behaves asymmetrically in regard to the past and to the future. A norm may produce its effect in a time antecedent to the beginning of its validity (as is the case for retroactive norms), while it cannot apply to situations subsequent to the end of its validity. However, effects already produced in the past are not eliminated by the fact that the norm ceases to be valid: abrogation usually is not retroactive. In other words, for a norm to produce certain effects in a certain time (included within its subsumption interval), it is sufficient that the norm has been valid in any one interval after that time, even if that interval is later terminated. We define similarly the notion of termination: terminated_at(U, T) :N: E l_terminates U, happens_at(E, T1), holds_at(is_applicable(N), T1), holds_at_or_started_after(valid(N), T1). Similar rules for initiation and termination are also introduced for unconditioned norms. Finally, the predicate holds_at_or_started_after(U,T) is satisfied in two cases: a) if the predicate U holds at time T, or b) if the predicate U started at a subsequent time (we use here the symbol “;” which means “or”). holds_at_or_started_after(U, T) :holds_at(U, T); started_after(U, T). 16 Finally, U started after T if was initiated in a moment subsequent to T. started_after(U, T) :initated_at(U, T1), T < T1. Non simultaneous norms The formalisation so far produced still assumes that every legal condition immediately produces its effect. Instead, as we have considered above (for example, when observing the rule prescribing that published norms become applicable 15 days later), legal effects may be antecedent or subsequent to their conditions. To cope with this aspect, we need to specify when a norm's effect is displaced (when no special displacement is specified, the norm's displacement will be 0 by default). To do this, we introduce a further infix operator, in , which we use for stating displacements. So, for example let us consider general rule establishes that laws become applicable 15 days after their publication. We will represent it as follows: vacatio: published(N) l_initiates is_applicable(N) in 15 Since vacatio is an analytically represented legal norm, in order to produce its effect, it needs to be supplemented by a non-legal rule stating the immediate beginning of applicability of vacatio. published(vacatio) initiates applicable(vacatio). Finally, the initiated_at (terminated_at) and the clause for legal effects need to be reformulated so as to account for a possible displacement of the effect. In the following formulation the time of the effect indeed is T0, which is obtained by adding the displacement D to the event time T1: initiated_at(U, T0):N : E l_initiates U in D, happens_at(E, T1), T0 = T1 + D, holds_at(is_applicable(N), T1), holds_at_or_started_after(valid(N), T1). Let us assume that a norm, let us call it n, is published on the 1/1/1997), which we express as the following fact: happens_at(published(n3, 1/1/1997). 17 By applying vacatio to this fact, we obtain the result that n became applicable exactly on 16 January 1997. Let us finally consider a full example in order to summarise the normative structures so far introduced: vacatio: published(N) l_initiates is_applicable(N) in 15. moral_damage: X causes_moral_damage_to Y l_initiates X must_repay Y in 0. published(N) initiates valid(N). published(vacatio) initiates is_applicable(vacatio). happens_at(published(vacatio),1/1/1945). happens_at(published(moral_damage), 1/1/1980). happens_at( john causes_moral_damage_to mary, 1/1/1990). This information allows us to infer that the moral_damage norm is both valid (according to the unnamed rule on validity) and applicable from the 16/2/1980 (according to vacatio, which turns out to be both applicable and valid). This finally allows us also to infer that John, having caused moral damage to Mary on the 1/1/1990, has to repay her. 6. Conclusion In this paper we have tried to develop first an analysis of the temporal structure of legal norms, and then a corresponding computable representation. We hope to have obtained both a satisfying conceptualisation of some significant features of legal language, and a correct translation of this conceptualisation into the framework of the event calculus. We also hope that our analysis provides some notions which may be useful for investigating the relation between time and law, and in particular for tackling the problem of legal duration. This may contribute to convince the reader having a legal background that some computational formalism can provide an insightful approach to certain legal issues. Sometimes (hopefully also in our case) the famous observation of Bertrand Russell that "a good notation has a subtlety and suggestiveness which at times make it seem almost a live teacher" may prove to be true. Finally, we hope that the reader may forgive the preliminary and provisional nature of our work, considering that we have dealt with features of temporal reasoning to which legal theory has so far dedicated a very limited attention. In fact, most studies of time 18 in law have so far mainly addressed validity without considering the "internal" temporal aspects of legal norms. Much work remains to be accomplished in order to complete the framework here sketched, in the spirit of event-calculus: 1. an analysis of the temporal dimension of different types of norms, which are not reducible to the event-state model here considered; 2. a more refined treatment of events, which also considers their length; 3. a treatment of the temporal aspects involved in action, especially those included within deontic operators; 4. a computable formalisation of those features of our formalisation that are not amenable to the extension of the event calculus here provided. Bibliography Bulygin E. 1982. Time and Validity. In Deontic Logic, Computational Linguistics and Legal Information Systems. Ed. A.A. Martino, 65-81. Amsterdam: North Holland. Hart, H.L.A. 1961. The Concept of Law. London: Oxford University Press. Hernandez Marín, R. 1996. Dos lecciones de filosofía del derecho. Murcia: DM. Kelsen, H. 1960. Reine Rechtslehre. Wien: Franz Deuticke. Kowalski, R.A., & M.J. Sergot. 1986. A Logic-Based Calculus of Events. New Generation Computing 4: 67-95. Mackaay, E., D. Poulin, J. Frémont, C. Deniger, & P. Bratley. 1990. The Logic of Time in Law and Legal Expert Systems. Ratio juris 3: 254-271. Quine, W, van O. 1960. World and Object. Cambridge (Mass): MIT. Provetti, A. 1992. The Law of Contracts in the Event Calculus. In GULP'92 - Proceedings of the Ninth Italian Conference on Logic Programming. Milano: Clup. Sergot, M.J. 1995. Unpublished slides. Thomason R.H. 1984. Combinations of Tense and Modality. In Handbook of Philosophical Logic. Volume II. Extensions of Classical Logic. Ed. D.M. Gabbay, & F. Guenthner, 135-165. Dordrecht: Reidel. Appendix. The Prolog program Let us now summarise the results of our work in a Prolog program, which we build by combining the clauses previously introduced. Such a program should run on any Prolog interpreter, after executing the following operator definitions (or similar ones): operators: 19 op(100,xfx, initiates), op(100,xfx, terminates), op(100,xfx, l_initiates), op(100,xfx, l_initiated_at), op(100,xfx, l_terminates), op(100,xfx, l_terminated_at), op(90, xfx, in), op(110,xfx, ':'). To keep the program simple, we avoid any further "syntactic sugar". Therefore, instead of writing dates in the usual format 22/10/1990, we write them 19901022. Similarly, we stick to the functional notation for predicates expressing legal knowledge. So instead of X causes_moral_damage_to Y we write causes_moral_damage_to(X, Y). /* PROGRAM NORMS AND TIME */ /* General Event Calculus */ holds_at(S, T2) :initiated_at(S, T1), T1 =< T2, !, not broken(S, between(T1, T2)). broken(S, between(T1, T2)):terminated_at(S, T), T1 =<T, T < T2. initiated_at(S, T):E initiates S, happens_at(E,T). terminated_at(S, T) :E terminates S, happens_at(E,T). /* Legal Event Calculus */ initiated_at(S, T):N : E l_initiates S in D, happens_at(E, T1), T = T1 + D, 20 holds_at(is_applicable(N), T1), holds_at_or_started_after(valid(N), T1). initiated_at(S, T):N : S l_initiated_at T, holds_at(is_applicable(N), T), holds_at_or_started_after(valid(N), T). terminated_at(S, T) :N : E l_terminates S in D, happens_at(E, T1), T = T1 + D, holds_at(is_applicable(N), T), holds_at_or_started_after(valid(N), T). terminated_at(S, T) :N : S l_terminated_at T, holds_at(is_applicable(N), T), holds_at_or_started_after(valid(N), T). holds_at_or_started_after(U, T) :holds_at(U, T); started_after(U, T). started_after(U, T) :initiated_at(U, T1), T < T1. /*Example */ published(N) initiates valid(N). published(vacatio) initiates is_applicable(vacatio). vacatio : published(N) l_initiates is_applicable(N) in 15. moral_damage : causes_moral_damage_to(X, Y) l_initiates must_compensate(X,Y) in 0. happens_at(published(vacatio),19450101). happens_at(published(moral_damage), 19800101). happens_at(causes_moral_damage_to(john, mary),19900101). 21