Identification of Implicit Legal Requirements with Legal Abstract Knowledge

Transcription

1 Idenificaion of Implici Legal Requiremens wih Legal Absrac Knowledge Seiichiro SAKURAI Deparmen of Sysems Science Tokyo Insiue of Technology Yokohama, Japan Hajime YOSHINO Faculy of Law Meiji Gakuin Universiy Tokyo, Japan Absrac In order o acquire legal rules from legal exs, legal requiremens and legal effecs mus be idenified. However, some of legal requiremens are expressed implicily. Such implici legal requiremens can be found by lawyers when hey undersand legal exs. In his paper, o mechanize legal knowledge acquisiion process, a lawyer s undersanding process of legal exs is analyzed. The lawyer s undersanding process can be viewed as an abducive reasoning process, since he lawyer can inroduce implici legal requiremens which are no appeared in legal exs. This paper models such a reasoning process when lawyers undersand legal exs. Based on he analysis of lawyer s undersanding process, a knowledge acquisiion suppor sysem is proposed. 1 Inroducion In order o redlze a legal exper sysem, we mus ranslae legal exs ino machine execuable forms. If legal rules are ranslaed ino rules like Prolog program, a legal knowledge-base can be consruced. Since a legal rule consiss of legal requiremens and a legal effec, hey mus be idenified correcly. However, since some of legal requiremens are described implicily, machine ranslaion is very difficul. In spie of implici legal requiremens, lawyers can correcly idenify legal rules by clarifying implici legal requiremens. Since lawyers seem o use heir legal experise in order o fill he gap beween he real meaning of legal exs and he lieral meaning of hem, such legal experise should be clarified. This paper analyzes a lawyer s undersanding process of legal rules. In he analysis, a kind of legal knowledge seems o guide he lawyer s undersanding Permission o copy wihou fee all or pan of his maerial is graned provided ha he copies are no made or disribued for direc commercial advanage, he ACM copyrigh noice and he ile of he publicaion and is dae appear, and noice is given ha copying is by permission of he Associaion for Compuing Machinery. To copy oherwise, or o republish, requires a fee and/or specific 1993 ACM /93/0006/0298 $1.50 process. We call such legal knowledge as a legal absracion, since i is widely applicable knowledge and i is absraced from specific legal knowledge. Based on he analysis of human undersanding process, a knowledge acquisiion suppor sysem is proposed. Our sysem acquires new knowledge by using given legal absracions, which is represened by a Higher Order Horn Clause. 2 Knowledge Represenaion 2.1 Compound Predicae Formula Legal knowledge can be represened by a se of compound predicae formulas(cpf). A CPF is a rule, which can be inerpreed by a deducive reasoning engine like a Prolog inerpreer. A legal rule can be represened by a CPF as follows: legal requiremens legal effec - p + ri represens a legal requiremen and e represens a legal effec. ri and e are called compound predicae erms. A compound predicae erm can be represened as follows: predicae(id, CaseLis) predicae is a predicae name, and ID is an idenifier of predicae. The idenifier is used as a reference of predicae insance. CaseLis is a lis of pairs and each pair represens case role and filler. Each filler may be a compound predicae erm. Case and filler are separaed by :. For example, a concep, a conrac is concluded, is represened as follows: he_concluded(idl, conrac(id2, [ag :-, obj :-,...])) In he above example, idl and id2 are predicae idenifiers, and ag, which sands for agen case, and obj, which sands for objec case, are case name abbreviaions.. is an anonymous variable. Noe ha he predicae conrac appears as a suberm in he above example, and he predicae conrac should be proved 298

2 in order o verify wheher he concep, (a conrac is concluded, holds or no. referred from oher pedicae, is idenifier is omied, oo. Therefore, he above fla CPF can be represened aa follows: 2.2 Inerpreing CPF A CPF is a formula for represening a legal rule and i direcly corresponds o a legal senence. Since legal ccmceps in a CPF are represened as chunks of primiive legal conceps, i seems o be comparaively e~y for lawyers o undersand a CPF. For example, he following legal rule can be represened by a CPF in Figure 1. An offer becomes effecive when i reaches he offeree. In he above legal rule, he legal effec is an offer becomes effecive and he legal requiremen is he of~er reaches he offeree. They are represened by compound predicae erms in Figure 1. become.effecively [obj:offer(o,[ag:a, im:d]) goaofferee(b,c)]), reach(_,[obj:offer( O,[ag:A,goaofferee( B,C)]), im:d, goaofferee(b,c)]). Figure 1: Example of CPF By using compound predicae erms, he legal requiremens and he legal effec can be disinguished easily. However, since a compound predicae erm relay conain oher compound predicae erms as is suberms, compuaional inerpreaion of CPFS is no easy. In order o enable efficien inerpreaion of CPFS, a fixed legzd dicionary, including legal axonomy, is assumed. By using he legal dicionary, a CPF is convered ino a fla CPF auomaically like [5]. In he conversion, he case names are eliminaed and he ci~e names are represened by argumen places. For example, he CPF in Figure 1 is convered as follows: become-effecive(yo,d) + offer(o,a,b), offeree(b,_), reach(ao,d,b). As shown in he above example, he filler of he obj of he becomeafecive is replaced wih he idenifier. The second argumen of he become-effecive represens he obj s filler and he hird argumen represens he ire s filler. Tim is he abbreviaion of ime. Similarly, oher predicaes are convered. In a fla CPF, only idenifiers are allowed as is fillers. Since a ila CPF is a funcion free Horn Clause, i can be execued direcly by a Prolog inerpreer. If he predicae is no become_effecive( O,D) + offer(o,a,b), offeree(b,-), reach(o,d,b). In he real applicaion, we should provide oher rules, which bridge he CPFS and he given facs. For example, since he given facs may no designae he offeree, he offeree s definiion rule will be needed. Such definiion rules are assumed in he legal dicionary. 3 Human Undersa~ding Process A lawyer s undersanding process is an yzed by using Unied Naions Convenion on Conrac s for he Inernaional Sale of Goods. Since he convenion is applied o conracs of sale of goods, Aricle 23, which is applied o conclusion of conrac, is concerned, Aricle 23 is shown in Figure 2. Aricle 23 A conrac is concluded a he momen when an accepance of an offer becomes effecive in accordance wih he provisions of his Convenion. Figure 2: Unied Naions Convenion on Conracs for he Inernaional Sale of Goods Since a legal rule consiss of legal requiremens and a legal effec, he requiremens and he effec mus be idenified. In Figure 2, even a novice or a beginner of law can idenify ha he legal effec is a conrac is concluded. Since mos of aricles describe such legcd effecs explicily, he idenificaion of he legal eilec is an easy ask. However, i is no so easy o idenify legal requiremens of aricles because of implicily described legal requiremens. Of course, he idenificaion of explici legal requiremens is easy. Aricle 23 is iniially undersood as a rule shown in Figure 3. In Figure 3, unused cases are omied for simpliciy. Indeed, since some relaions beween primiive conceps are no sufficienly specified, he ideal rule should specify such relaions. A novice would undersand Aricle 23 as such a rule, since he rule seems o represen he lieral meaning of he aricle. Once such a rule is obained, even a novice ries o is validiy. The rule can be verified by amuming an ideal siuaion in which a conrac is concluded. In such a siuaion, he accepance of he offer mus

3 be_concluded(_, [ag:[a,b], obj:conrac(a[ag: im:d]) e become.effecive(_, [obj:accepance(= [ag:b, im:d). obj:offer(~ gowa]), [A, B]]), [ag:a,goa:b]), Figure 3: Iniial Undersanding of Aricle 23 become effecive. Since he following Aricle 18-2 describes when he accepance becomes effecive, he rule of Aricle 18-2 will be used for he verificaion of he CPF in Figure3. Aricle A saemen made by or oher conduc of he offeree indicaing assen o an offer is an accepance. 2. An accepance of an offer becomes effecive a he momen he indicaion of assen reaches he offeror. The CPF of Aricle 18-2 is shown in Figure 4. become.effecive(-, reach (-, [obj: accepance(ida, [ag:a, [obj:indicaion(ida, [ag:a, obj:offer(c,[ag :offeror(b,[obj: C]), goaa]), goaofferor(b,(obj: C])])]) i- obj :assen (_, [ag:a, obj:offer(c,[ag :offeror(b,[obj: C]), goaofferor(b,[obj: goaofferor(b,[obj: goa:offeror(b,[obj: C])]). goaa]), C])]), C])]), Figure 4: CPF of Aricle 18-2 If he legal requiremen of Aricle 18-2, he indicaion of assen reaches he offeror, is assumed in he ideal siuaion, a conrac is concluded can be proved by using wo legal rules, Aricle 23 and Aricle In his way, rules ranslaed from legal exs are verified. Because of such successful explanaion, he novice s undersanding process may hal. Are he above rules really correc ones? In spie of such successful explanaion, mos lawyers can provide is couner-example. For example, by means of wihdrawal of offer; even if he indicaion of assen reaches, he conrac canno be concluded. Therefore, lawyers undersanding process don hal when only such explanaion is made successfully. The lawyers massive legal experise seems o enable hem o invesigae furher. When lawyers undersand legal exs, hey also confirm wheher he legal effec can be derived from he idenified legal effecs or no. Then hey judge he validiy of he derivaion by using heir legal experise. The verificaion process can be seen as a backward reasoning process of a legal effec. Figure 5 shows such an explanaion. indicaion-o f_assen ~each I : offer_be_effecive i / accepance-b ecome-effecive / \ conrac-be_concluded Figure i / 5: Explanaion In Figure 5, each subgoal, compound predicae erm, is represened by a proposiion. The firs op goal is he conrac.be.concluded. In he explanaion, alhough offer-be_ effecive does no appear in he legal exs, i is regarded as a legal requiremen by a lawyer. The undersanding process can be seen as an abducive reasoning process since he explanaion is consruced by adding implici legal requiremens. In he search of such legal requiremens, he candidaes of he legal requiremens seem o be weakly consrained by lawyers absrac legal knowledge. In Figure 5, a general principle abou conrac, a conrac is a mach of offer and accepance weakly consrains he candidaes. Afer he verificaion, a new legal requiremen, offer is effecive, is added ino he body of he legal rule of Aricle 23. For example, he following rule can be made. conrac-be-concluded accepance-become_effecive, (1) offer-be-effecive 30 ()

4 If he legal rule whose effec is he accepance lbecomes effecive is modified so ha is requiremen par includes he offer is effecive, he same resul can be derived. For example, he following rule can be made. accepance.become.e ffecive indicaion.ofawen_reach, (2) offer_be_effecive The former legal rule (1) is preferred since he f ormer one maches he lawyer s absrac legal knowledge. Such a knowledge is called a legal absracion, since i represens absrac knowledge which is absraced from a specific legal knowledge. An example of a legal absracion is shown in Figure 6. In Figure 6, he legrd norm senence is a super concep of he conrac, and he indicaion of inenion is a super concep of boh he accepance and he offer. Since he offer is prior o he accepance, he former legal rule (1) maches he legal absracion in Figure 6. If here exiss anoher legal absracion, he laer legal rule (2) may be preferred. V Operaor C1(+)WC2(-) c W Operaor Figure 7: Inverse Resoluion as Abducion[3] I I If wo indicaions of inenion forms a legal norm senence, he legal norm senence is concluded when he prior indicaion of inenion is effecive and he poserior indicaion of inenion becomes effecive, Figure 6: Example of Legal Absracion In his way, he lawyer s undersanding process seems o be guided by legal absracions. Therefore, legal knowledge acquisiion from legal exs can be seen as abducion consrained by legal absracions. 4 Abducive Reasoning and Absrac Legal Knowledge Given he clause C. : T. + Lce and a domain heory T, proceed o seps 1,2 and 3: 1. Skolemizaion of Cc ino C.O. (one-keeps rack of 6, for skolemizaion). For each lieral Lcei in he body of C., a clause cs~g is creaed, wih C8ki : Lc=j Ela Deducive phase: we apply all possible resoluions beween he se of clauses {Ca~j } and he clauses of T. Legal rule acquisiion process can be viewed as an exracion process of a legal effec and an abducive explanaion process of legal requiremens. As an abducive explanaion engine, an inverse resoluion [3, 4, 5] mehod is used. The inverse resoluion is an inverse operaion of resoluion. Figure 7 shows ypical operaions of he inverse resoluion. The basic operaion is performed by V operaor. In he resoluion, C is derived from Cl and G 2. In he inverse resoluion, Cl is derived from C and GZ or Gz is derived from C and Cl. Rouveirol[4] has proposed Sauraion Operaor, which is an inverse resoluion operaor. The algorihm of Sauraion [4] is shown in Figure If aoms have been deduced afer sep 2, hey are ransformed ino uni clauses and added o {Cs~, }. The process hen ieraes on sep 2. If no lierals have been deduced afer sep 2, he {C8~i } are deskolemized and added o LC= o form he body of he sauraed clause, c=. Figure 8: Algorihm of sauraion[4] 301

5 Inverse resoluion can consruc rules which bridge a gap beween a newly observed fac and known facs. However, he search space of he inverse resoluion is very huge. For example, he offer-be-effecive in Figure 5 is no described in he UNCCIS. Therefore, he offer_be_effecive should be consruced aa a new predicae. While W Operaor is an operaor which has an abiliy o inven a new predicae, he conrol of applicaion of W Operaor is very difficul. Furhermore, as shown in he previous secion, he invenion of new predicae should be conrolled by legal absracions. A legal absracion is represened by Higher Order Horn Clause[l]. The legal absracion in Figure 6 can be represened by Higher Order Horn Clause. Figure 9 shows an example of a legal absracion represened by a Higher Order Horn Clause. be_concluded(=[ag: _obj:xid,im:c]) + X(Xid,[obj:_]), become-effecive( q[obj :Yld,im:C]), Y(Yld,[obj:_]), be.effecive(_jobj: Zid,im: C]), Z(Zid,[obj:-]). Consrains: X is legal norm senence. Y is an indicaion of inenion, Z is an indicaion of inenion. Z is prior o Y. Figure 9: Legal Absracion Represened by a Higher Order Horn Clause In Figure 9, X, Y and Z are predicae variables, and he consrains can be considered as predicae ypes[l]. Since he consrains are used in Higher Order Unificaion, only he same ype predicaes are allowed o bind he predicae variables. For he efficiency, all predicaes are insaniaed by predicae consans. Under he resricion, a legal absracion can be seen ss a Firs Order Horn Clause. If he legal concep hierarchy is represened by a se of Firs Order Horn Clauses, he inheriance is realized as a deducive inference. In order o disinguish he legal deducive inference wih he inheriance, a legal absracion is represened by a Higher Order Horn Clause. A legal absracion can be convered ino a fla legal absracion. For example, he legal absracion in Figure 9 can be convered as follows: be_effecive(zid), Z(Zid,a-,-). The above example, each predicae ariy is deermined by using he legal dicionary. Based on he above Iegrd absracion (3), a new rule can be obained as an insance. An insance of a legal absracion can be obained by subsiuing he predicae variables wih he same ype predicae consans. By assuming such legal absracion, he legal effec can be proved also by a heorem prover like Aprolog[l].!5 Knowledge Acquisiion Suppor Sysem KASS is a legal knowledge acquisiion suppor sysem based on he analysis of human undersanding process. Figure 10 shows he overview of KASS. KASS has wo componens for legal knowledge acquisiion. One is a bug deecion module and he oher is an abducive ressoning engine based on he inverse resoluion. The bug deecion module is based on Shapiro s algorihmic debugger[6]. A se of facs and rules are given o he module. Then he module ineracs he user o idenify bug rules. Figure 11 shows an erroneous explanaion when he offer is no effecive. If all leaves of he explanaion in Figure 11 are confirmed by he user, he following wo rules may be considered as bug rules. be.concluded(i4) + conrac(i4,a,b), offer(i5,a,b,c), (4) accepance(i6,b,a,i5), become-effecive( i6) become_effecive(i6) - offer(i5,a,b,c), accepance(i6,b, a,i5), (5) reach(i6,a) If muliple rules are considered as bug rules, legal absracions are used for selecion of a bug rule. If he legal absracion (3) is given o he KASS, he legal rule (4) is seleced as a bug rule. If anoher legal absracion is given, he legal rule (5) may be seleced. The bug rule is given o he abducive reasoning engine and i is eliminaed from he legal knowledge base. The rrmooning engine hen finds an appropriae legal absracion. If he legal absracion (3) is seleced, he following insance can be generaed. be.concluded(xid) X(Xid,a-), be_concluded(xid) conrac (Xid,a-), becomeafecive( Yld), Y(Yid,ya-), (3) become-effecive(yld), accepance(yldma _), (6) 302

6 rlegal Absracion \F I)B Engine I lhxlf!!ie1/ L Figure 10: KASS be.concluded(i4) conrac(i4,a,b) offer(i5,a,b,c) accepance(i6,b, a,i5) become-effecive(i6) offer(i5,a,b,c) accepance(i6,b, a,i5) reac(i6,a) Figure 11: Erroneous Explanaion 303

7 be_effecive(zid), offer(zid,-,_.). By referring he original rule, he above rule can be refined aa follows: be-concluded(a) + conrac(a,b,c), become-effecive(d), accepance(d,c, B, E), (7) be.effecive(e), offer(e,b,c,.). In he above legal rule (7), a new predicae be.effecive(offer(a,b,c,d)) is inroduced. This new predicaewillbe anew subgoalof he abducive reasoning engine and oher new rules may be found. If such auxiliary legal rules are found successfully, new legal rules, which saisfy legal absracions, can be learned. 6 Learning Examples Figure 12 is he iniial knowledge base provided o he KASS. Alhough Aricle 18 is described by legal senences, Aricle 18 is represened by a single legal rule. The emporal informaion is omied for simpliciy, in he real applicaion, he heory for emporal reasoning should be needed. UNCCIS goal which should be proved. conrac(il,a,b), offer(i2,a,b,c), reach(i2,b), (8) accepance(i3,b, a,i2), reach(i6,a) The above example can be paraphrased as: If he offer reaches he offerree and he accepance reaches he offeror, is a conrac concluded? While he above example is only used for he verificaion of legal knowledge-base, if he following goal is proved, hen he legal knowledge-base includes some bug rules. be-conclude(i4) ~ conrac (i4,a,b}, offer(i5,a,b,c), reach(i5,b), (9) accepance(i6,b, a,i5), rejecion(i7,i5), reach(i7,a), reach(i6,a) In he firs sep, he proof of be.concluded(i4) conains he applicaion of Aricle 18 and Aricle 23, hey may be erroneous rules. By using he legal absracion (3), Aricle 23 is idenified as an erroneous rule, and he following rule (10) is obained as an insance of he legal absrac ion. Aricle15 : become_effecive(a) + offer(a,ab,-), reach(a,b) An%de17 : be-erminaed(a) + offer(a,b,-,-), rejecion(c,a), reach(c,b) An$icle18 : become_effecive(a) + accepance(a,b, C, D), offer(d,c,b,-), reach(a,c) Aricle23 : be_concluded(a) + conrac(a,b,c), offer(d,b,c,_), accepance(e,c, B, D), become-effecive(e) be_concluded(a) i- conrac(a,b,c), offer(d,b,c,e), (lo) accepance(f,c, B, D), become_effecive( F), be-effecive(d). The above rule can be convered ino he CPF as shown in Figure 13 when he user wans o see he learned legal rule. However, he verificaion process needs anoher assumpion, be_effecive. Then, KASS S abducive reasoning engine is invoked recursively, he following legal rule (11) can be obained if appropriae legal absracions are given. auxiliary rules no-be-erminaed(x) + no(be-erminaed(x)). be-effecive(x) Figure 12: Iniial Knowledge offer(x,-,-,-), become-effecive(x), (11) no-be-erminaed(x). The following query is given o KASS. The righ hand side of ~ is assumpions and he lef hand side is he In his way, erroneous rules are removed and he correc rules can be obained from legal absracions. 304

8 bedoncluded(.,[ag: [A,B], obj:conrac im:c]) :- become-effecive(-, [obj: accepance(y [ag:b, im:c]), be-effecive(., (Y [ag:[a,b]]), obj:offer(d,[ag: A,goaB] ), goaa]), [;~of$r(d,[ag:a,goab]), :. [7] H. Yoshino e al. Foundaion of Sysemaizaion of Legal Analogy (In Japanese). Proc. of he 5h Annual Conference of JSAI, pp , Figure 13: Learning Example 7 Conclusions This paper describes a knowledge acquisiion from legal exs based on he analysis of human knowledge acquisiion process. If background knowledge is provided, we can approach he auomaic knowledge acquisiion from legal exs. By replacing he legal absracion daabase, a variey of legal knowledge can be obained. Since he acquired knowledge does no conain dynamically inerpreaion knowledge, in order o implemen a legal exper sysem, oher inference engine such as [7, 2] is required. References [1] D. Miller, e al.. Uniform proofs as a foundaion fcm logic programming. In Annals of Pure and Applied Logic, 51, pp , [2] M. Haraguchi. A form of analogy as an abducive inference. In ALT91, pp , [3] S. Muggleon and W. Bunine. Machine invenion of firs-order predicaes by invering resoluion. In Workshop on Machine Learning, pp , [4] C. Rouveirol. Semanic model for inducion of firs order heories, In 1JCA191, Morgan Kaufmann, [5] C. Rouveirol and J. F. Puge. Beyond inversion of resoluion. In Workshop on Machine Learning, pp , [6] E. Shapiro. Algorihmic program debugging. MIT Press,