Logique A n a ly s e 143-144 (1993), 245-260 PROOF-THEORETICAL CONSIDERATIONS ABOUT THE LOGIC OF EPISTEMIC INCONSISTENCY* Ana Teresa de CASTRO MARTINSt and Tarcfsio H. C. PEQUENO Abstract The Logic o f Epistemic Inconsistency (LEI) provides an interesting example o f a paraconsistent formal system from the proof theoretical point o f view. Although cut elimination theorem fo r its sequent calculus presentation can only be proved with a restriction, this does not damage its main desirable outcomes, that are still attained. Therefore, consistency, subformula property and conservativeness o f logical constants definitions are we ll preserved. Th e motivation f o r L E I designing, it s sequent calculus presentation and mo st remarkable theorems are presented. Instances of the unremovable cut-proof pattern are analysed. I. Introduction The L o g ic o f Epistemic Inconsistency (LEI) is a paraconsistent lo g ic designed to support reasoning under conditions of incomplete or unaccurate knowledge. In these circumstances, conclusions must be grasped on the basis o f partial evidences. In the course o f reasoning, partially supported conflicting conclusions may emerge. This kind of situation asks for a criterious consideration o f the total evidence available in order to solve these conflicts but, it may eventually happen, the available knowledge not being able to do so, making some o f these pairs o f conflicting partial conclusions turn out as unremovable conflicts, being this the best that could be possibly done under these circumstances. It is our opinion that, when the term reasoning is used to express a more general thinking process than mere deduction, as in the situation ju st described, its role is not just the grasping of conclusions, as for deduction, but to perform a more comprehensive analysis of the available knowledge. * research partially supported by PICD/CAPES and CNN. t PhD student at Federal University of Pernambuco 246 M A R T I N S & PEQUENO As a consequence o f this disposition, the whole set o f emerging conclusions, which includes the unresovable conflicting pairs, should be assimilated and reasoned out by a suitable logic, able to do so without provoking a logical collapse. The kind o f inconsistency so arising should be called epistemic, by reflecting not an inconsistency in the state of affairs itself but a deficiency in our knowledge about it [Pequeno & Buchsbaum 91]. Incomplete knowledge reasoning has been the topic o f study o f the so called nonmonotonic reasoning in artificial intelligence, fo r more than a decade now. The current attitude in the field has been to avoid the hazadous o f epistemic contradiction by taking a choice between two alternatives. One, which has been called credulous, consists in accepting a ll conflicting conclusions by splitting them into self consistent subsets [Reiter 80]. The other, sometimes called skeptical, consists in simply rejecting all those conclusions [McCarthy 80]. In [Pequeno 901, it is suggested to keep the whole set of conclusions, no matter they possibly being contradictory, in a single theory, treating them paraconsistently. L EI has been designed for th is purpose and i t is intended to be used in combination wit h a nonmonotonic logic called IDL (fo r Inconsistent Default Logic [Pequeno 90]) wh ich takes care o f the dynamics o f this kin d o f reasoning (the incoming o f knowledge and the dismissing o f no lo n g e r supported conclusions). LEI logical stmcture also fits a general pattern o f situations in which a plurality o f views on a same subject is involved. This is the case, fo r instance, when a same phenomenon is reported by many observers or, in general, when it is observed in experiments under slightly varying conditions, being these variations undetected and/or unrecorded along the experiment, but anyway enough to affect it. Then, disagreement may also occur, and again, by a lack of knowledge. This fact make of these situations instances o f epistemic inconsistencies. It is curious that, in the first case, epistemic inconsistencies seem to arise fro m the underdetermination o f conclusions wh ile , in the second one, they seem to come f ro m th e ir overdetermination. The designing of LEI semantics has been based on the intuition coming from this last scenario, while its Hilbert style axiomatics is intended to match the pattern o f reasoning upon incomplete knowledge described before. LEI soundness and completeness establish an equivalence between these two perspectives [Pequeno & Buchsbaum 91]. Reasoning upon partial/multiple knowledge is a central issue in many Artificia l Intelligence applications and its automatization is certainly profitable. Tableaux systems for L EI have been developed in [Corrêa & Buchsbaum & Pequeno 93] and [Buchsbaum 91], and a sequent calculus in [Martins & Pequeno 94]. Ou r aim here, is to discuss proof-theoretical issues through L E I sequent calculus by stressing L E I metalogical prop- PROOF-THEORETICAL CONSIDERATIONS A ND EPISTEMIC INCONSISTENCY 247 erties. In section 2, LEI sequent calculus — proved correct and complete with respect to the given LEI axiomatics — and main L E I theorems are presented. Proof-theoretical considerations are detailed in section 3 and conclusions, as well as further developments, are pointed out at the end. 2. LEI sequent calculus and main theorems LEI has been conceived as a maximal approximation of classical logic able to support, without trivializing, epistemic contradictions arising from the application of IDL (default) rules, which introduce defeasible conclusions. To distinguish these conclusions fro m those irrevocable ones, a ? sign marking the first ones is adopted in LEI. The intuitive meaning o f a ? is 'there are evidences to a ' o r ' references, irrefutable, monotonic, formulas w i l l be denoted b y roman a i s capital letters, ?-formulas p l a u s i b l e ' . by greek lower letters and sequence of formulas by greek capital letters. F o r i s attained i n L E I b y re je ctin g e x fa lso se q u itu r t Paraconsistency h e quodlibet, s a a -•>k (-,ce ----> e p), to formulas suffixed b y ?. Technically, L E I differs fro m da Costa paraconsistent calculi Ci [daCosta 74] in some key o f aspects, mainly: first, L EI retains more classical theorems than Ci — a m e t remarkable example of this is the schema -, (a A -,a ) usually taken as an a inner expression of the non-contradiction law. Although rejected in Ci, it is still a theorem in LEI. Second, a recursive semantics has been provided for LEI. Whereas d e n o t e s negation in LEI, a paraconsistent negation, - a is introduced a s a handy abbreviation f o r a ---> p A -ip , b e in g ' p ' a propositional letter. - can be shown to wo rk as a classical, also called strong negation in LEI. LEI sequent rules are: Structural Rules Exchange LXI-, a , P , r I- A F, f i, a , r I- A I- A , a , 13, A ' F A , 0, a, A ' 248 MARTINS & PEQUENO Weakening LW I - A I", a , I- A I' I- A R W F I- a , A Contradiction LC I', a , a , I- A F, a , I- A I- a , a IR C I a , A I' I- a -4 1 3 , A A RC Fa? A ' Identity Rules a I- a [Identity A I- a , '?' A F ' , a I- A ' Cut F, F ' I- A , A ' A A I- a , A r , a ? F A ' ? Cu t ?F' I- A , A ' R OperationalC Rules F - 0 , Question mark a? I- A Li • • F, a ? ? I- A I- a A R , ? IF I- a ? , A F, a ? -31.3 ?I- A I- a - 0 , A ? (a?-0?)? F A a ? -0 ? , A Negation I- A , A I', - A I- A F, a , F A F 1- a , A R 2 ? PROOF-THEORETICAL CONSIDERATIONS AND EPISTEMIC INCONSISTENCY 249 Conjunction a A F, A a AP A , A I', a A [3 A A a, A F ' F /3, A R F, a A P , A , A ' A Disjunction LF, a I- A F . P I- A ' F, R, a 4 I- A , A ' F 1- a , A F F 0, A I- a y 0, A F a v p, A R 2 V Implication • a , A F L R 1 • F ' , a -4 3 A , A ' F I- a - 0 , A , p 1 a ll variables are held constants and, either a is ?-closed or R2 ? and ' R - areAnot applied after the first time a occurs justified by being a F premise. , DoubleaNegation l a I- A F F - o t A D IT, I , a , A Distribution of ' ---,' over ' y ' A F, A -43 A LD, F, - ( a v P) F A ,' • - - i o t A -0, A p LDA r , - ( a A P) F A • 1- Distribution of ' -, ' over ' A ' F, 1 A I' I- - 0 4 ) , A - D 7 y -0, A ✓ - ( a 4 ) , A RD/7 Distribution o f ' -, ' over LD - I "' a A -13 F, (a 0 ) F A F I- a A- p, A p 1- A -D-7+ 250 M A R T I N S & PEQUENO Distribution of ' —1' over '?' LDT F, ( • 1 '-, --,(a ? ) I- A a ) ? Distribution of o v e r I LD,? a -• 7 A ( a v f l ) ? I- A (-,a)?, A F ( a ?) , A ROT v Classical Negation f 3 I ? IF a )- ? ) , a , A First Order A Rules A R - Universal Quantification ' L V 1 ' F, V xo t(x) i- A 1 ' 1at is free for x in a (x ) 2 x must not occur free in 1 ( t ) ,Existential A Quantification I I- Vxot(x), A IL3 a ( x ) F F xce (x), A IA 3xce(x) 1- A I1 Ax must not occur free in F, A 2I t is free for x in a (x ) -F Quantifiers Equivalence I a 3x(--,a) F A F 3 x ( - 1 a ) , A R —1 (- t ) —,(Vxa) F A ---1(Vxa), A 0 2 4 , Vx ( F V x(--, a ), A R ) A 1", A -F - , ( 3 x a ) , A 0 R , a ) 2 3 A , A R V PROOF-THEORETICAL CONSIDERATIONS A ND EPISTEMIC INCONSISTENCY 251 Extended Rule for Classical Negation Classical Negation a, A L eF, - a A I " F R - , a I- A - a, A e LEI sequent rules are closely related t o L E I axiomatics presented in [Pequeno & Buchsbaum 91]. Paraconsistency has been captured by restricting t o classical, ?-free formulas. Double negation rule, distribution rules of o v e r A, y and a n d quantifier equivalences are derivable in classical logic but not in LEI and must be explicitly stated. Central focus has been devoted to the new and peculiar ? operator. It has been defined by operational rules but its structural aspects are also expressed by two new rules: ?-Cut and ?-RC. The former states that: ' i f a comes from 13 and y comes from a ? , then certainly y also comes from the stronger a , or from its proof using T h e latter is closely related to the operational rule R premise a deduction where R 2 2? Question ? w a smark rules reveal operational ?-features. ? states that fro m aa p pcan l iextract e d its . plausibility a ? . The opposite is obviously not valid. . you Nevertheless, a weak converse of R ? additional L2 ? expresses that external ?'s are irrelevant I—? —?'s are superfluous. ? to an implication when both antecedent and succedent are ?-closed (see — i s R definition vC a l 2.2i below). d . R2 ? provides the propagation of ? along inferences when used in conjunction with L —) to simulate a sort of ?-Modus Ponens ?r (a?,a---> 13 / ? ) as in the se t a t e proof s below: ts h a t T t T2 o 2a F r h— r e 1 a> l s I t ? The same is achieved in LEI , axiomatics using Modus Ponens and rule t 3 — a ha -4 S r a as in the>following: ? ea ? -4 H 0? 0? s F p ?a a 2 ? a? m , 0 ? e a -? c ? 4H o I 1p n 3? 0 t ? e ? , x a t ? I H 252 M A R T I N S & PEQUENO a —> f3 a? a ? - 0 ? P? Distribution o f ? over y is assured by L D . Its symmetrical RD:, is derived. R- states that it is impossible to be sure about a and, at the same time, t o have an evidence to it s classical negation ( - a )? A theory including a and ( - a)? is trivia l in LEI. Notice the double line at rules R R 2 - d o not hold constant the variables over evidences implicitly quantified by ? ?. On the other hand, the imp lica tive fo rms o f R 2(a ? >p)—>(a?—> a n d 0?)R a n -d a, —>-((i a)?) . wo e u ld . ca rry o u t a ? —> a a — which is not reasonable. n Main LEI theorems are: d R Theorem 2.1. all thesis of classical logic hold to ?-free formulas in LEI. I Corollary 2.1. the negation symbol b e h a v e s classically for ?-free formulas in LEI: t m B) —> — > —B)—> a A r Theorem k 2.2. The defined symbol - really behaves as classical negation: s (a p ) - ) ((a —>- a ) a a a r Ae form of replacement theorem can be recovered with the help of a special implication introduced by definition [Pequeno & Buchsbaum 91]. s t Definition 2.1. (strong implication) a 0 is a short form for (a —> f3) A r i -96 - › —,a a n d (strong double implication) a <=> p is a short form fo r ca1 0 / 3 ) A - - t— , a ) Theorem 2.3. (replacement) Let a ' be a formula obtained fro m a b y i substituting 13 ' fo r some occurrences (not necessarily a ll) o f 13 . Then o I -n Definition 2.2. (?-closed formula) A formula is ?-closed, i.e., it is under the Io vscope of ?, if it has one of the forms a?, p, 0 4 are e ?-closed formulas and / 1 4 , A , V1 ,0 y , w h e r e a n d r y0 'R — e > n ta p a p i ll PROOF-THEORETICAL CONSIDERATIONS A ND EPISTEMIC INCONSISTENCY 253 Theorem 2.4. I f a is a ?-closed formula, then F- a •4 a ? . A restricted version of the deduction theorem can be kept also. Theorem 2.5. I f F, a F p and if does not occur any free variable in the set of premisses, then F F a —>13 unless a is not ?-closed and the L EI rules R of 2 being a premise. ? o r 3. Proof-theoretical issues about LEI R — Cut elimination theorem was established by Gentzen f o r classical and a intuitionistic logics and assures that every derivation using sequent calculus r be transformed into another one with the same endsequent, in which no can e occur [Gentzen 35 ]. Prawitz's No rma l Fo rm theorems assert a n cuts u equivalent result but fo r natural deduction calculus [Prawitz 711. Proofs theoretical issues concerned cut-elimination and normalization theorems in LEI e will be now discussed. d Cut elimination in LEI, established in [Martins & Pequeno 94], was proved a with a restriction: some proofs where R 2f propagation of ? along ---> to simulate ?-Modus Ponens (a?, a —> p ip ? ). In Gentzen transformation o f his Natural ?t fact, i s appealing a p p lt oi history, e d aDeduction e i m calculi i n NgJ and NK , f o r intuitionistic and classical lo g ics respectively, to his correspondent Logistic Sequent Calculi L I and L K had tr h e requested the use of cut rule to effectively translate elimination rules to left tones [Ungar 92]. I t will enforce our argument below in favour o f cut h maintenance on such proofs. To illustrate this, consider, as a typical e example, the following proof: f i a Fo t 1 r a 3l - a v f i 1 3 1 a v1 3 s 'P I- a -> v a ? F o t ? ( a v f l ) ? 1 - ( a v ) ? F / 3 - - - ) ( a v i 3 ) f i ? F f l ? ( a v P ) ? i ( a v P ) ? t t a? - + ( a v p )? a ? , a ? --)(cx v1 3 ) ? i( a v1 3 ) ? 1 3 ? - ) ( a v1 3 )? 1 3 ? , 1 3 ? --(a v p ) ? F ( a v f i)? a ? Eja vf i) ? 1 3 ? ' ( a v t r ! ) i a ? v( PF (a vp )'? . ( a vt 3 ) ? m a ? v p?1- ( c t v 0 ) ? e ot?,/ )5? -* v P)? a o c c u r s i n a p 254 M A R T I N S P E Q UE NO I f we translate this example in a natural deduction style we get (vide in [Martins & Pequeno 94] the straightforward derivation o f LEI natural deduction rules from LEI sequent ones): 1. [ a ? v 13?1 2. [ a 3. [ a 4. a 35. a 6. a 7. ( 2 ? ( 5 a) 8. L / 9. L P 3 lo. a v 1 ?] ] v) - ' (a v p) I ? ( a v p)? 1 v0 )? E ] —› (3, 4 ) —› (2, 6 ) p 11. ) 12. a ? —› (a v 13)? 6 13. ( a v fi)? — › 14. ( a v 1 (15. a ? \ / 0 ? ( a v f i ) ? 3 )? a In v this translation, note that we do not have a characterization o f a redundant inference. In fact, in step 5, fo r instance, we have introduced an P implication that was subsequently eliminated, but only in step 7. We have ) an intermediate step (the application of 1 permuted in order to explicit this candidate to redundancy. This proof is 2 indeed in normal ? o r r u lform: e the R analytical part corresponds to y elimination — steps l to 14 — while the synthetical part is the -4 introduction — step 15 2 — and the minimum part ? ) w h i c is the h conclusion o f the analytical part and the premise of the synthetical one, i.e. step 14. No reduction steps (0 , n , c a n n o t can be applied in order to get rid o ff redundances. In fact, Normal Form b e theorem was established in LEI [Martins & Pequeno 941 as in classical logic just adding new ) Nevertheless, nonadditional Ç reductions were necessary. 3 a n d Backing to the sequent calculus proof, we can easily see r e d u c t i correspondent o that ru le R n s between 2 t othe elimination of —> through the cut rule and the introduction of through Ra r u l e , we have an application o f R ? d e rewriting process behind cut elimination, there is no way to be free of such rl e 2a l l intermediate rule by permuting since this rule cannot be broken i n g cut upward, t o yw ? . i A p p t e a ltheorem through the deduction represented by R ---> I n fact, R—> is also e d it s h h restricted to cases analogous to the ones related to quantifiers where either a l ul o e q the discharged premise a is ?-closed or rules R w s2 t after the f iirst toime a appears as a premise. I t is a consequence o f tn ? a n d R a r e h m at n o e ra p pk l i e d cr h au rl a ce ts e r . i z a PROOF-THEORETICAL CONSIDERATIONS A ND EPISTEMIC INCONSISTENCY 255 interpreting ? as a sort of existential quantification of hidden variables over evidences. Hence, if the discharged premise a is not under the scope of ? ( a is not ?-closed), the application o f R constant such ?-quantified hidden variables. 2 we shall show that the main desirable consequences ? Toaend n this d section, R of will be kept even though we have maintained some - cut-elimination t o a special occurrences of cuts in proofs. Such cut-proofs can be framed in the w i l l following pattern: n o t h o l d 1 F1 ta 13 I4 The desirable outcomes 1 Fof cut-elimination are: 3 A a, 1. Consistency o f thisl calculus: even i f F A ? never prove the empty sequent F-- since 1 above aI n dproof, A we would , -r e x i s t a t l e a st 7 I always w e r e A 02HmFI- y, p8 o tr y, y5 1 or e E 2 2-?a w hnruelers et o id, e n t it y a x io ms u p p e r t o F o tional ? F y, , AI h, U t 2 e A -23 8AA a a ? ,n d A3 2 2. Subformula the following example: 2 amProperty: n e observe d 3,c1F o 0 eF3 T a r? f a, r o1 3 3 2) a At irp Im( t A hU i ,- I r ( a A3 tl 3 ) - ) a ah? 1 F? A F e,a,0 A) ?/- )3a )? ?( a1 A. 0 ) ? - o a ? , ( a A 1 3 ) ? 1 - a ? A , 0 ) ? a (?a A (S 3 4 - 0 ?t3 aI ( a pA 0 )(pa?.Ap- )4l? 11- a3? ? j (a A c1 3 A) ? I 3 ? ) ? i ft? I( a- A 1 ,/3- 3 ?) ? ry c i a t (a A P)?, ( a A 13)? r a ? " 0 ? A 1 ( ) F ( 0 ( In ) ia3 o Ie as 5o? pA ?t p t f- 3 ? ae o0 A . ) c? ?A r1 ae this first ? ta I n i - it will ( a AP) , ? r a ? A f l ? ( a A/ 3 ) ? - * a ? AP? p r3 example we have two occurrences o f cut rule. Consider, fo r instance, the one where the cuted formula is ( a AO)? i such > p ?-formula . N is o the t eimplicative t h form a otf the formulas which occur in the v sequent A conclusion (i.e., ( a A13)? I - 1 e23 ? ) o f t h l, i s r uyA l e . . 3 256 M A R T I -a F13 a Af il - a a A3 1 - - A 13) -+ a N S & PEQUENO a ? l-a ? 1 3 ? 1 -p ? A1 3 )? 1 -(a At i)? (a A# )? 1 -(a AP) ? a ? , f 3 ? 1 -a ? Af i? (a A J3 ) Ap )-4 1 3 A / 3 ) ' - * a ? ? ( a A p ) ? - ) a ? , ( a AP) ? - 3 ? , A f i ) ? F - a ? A p ? Af3)?-)13?, ( a A p )? f r a ?A/3? I * a ( ? . a A ( ) ( a A I3)? a F a 13 ? 3 ) ( a AA0 ? - > a ? A p ? ? f ? l ) This second example has the ? same endsequent but the p ro o f tree is somehow different. Now, the -›same cuted formula ( a A0)?--> 0 ? is not the implicative f o rm o f ( a A P)? However, ? h athe ? consequent A 1 3 ? of , , a A p)? f i ? , i. e . 0 ?, can be obtained by ( application of R t h e a it what in the first Aproof. Hence, whenever the cuted formula is A rt c is o uv ehappens p not in the cut sequent conclusion, you can ae implicative s( the q u form e of formulas ) rewriting steps of cut-elimination [Martins & rewrite the proof applying the A n t ? Pequeno 94] in order to accomplish this constraint. , p c o) n? c l u ( The last example is the following, where we consider as hypothesis Is - o a a <,>i a and [3 n<,:, /3.?: A a ? 1 A a?1- a ? 3 p a?A/3?1- a ? ) a ? F a? ?FGe ?A /3 ?) -4 a ? ( a ?A ? ) ( a -a?A )9?)? a ? ? I- a ? b e ? 1 - - ( a ? A 0 7 ) ? - ? ? ( a ? At i ? ) ? - - ) a ? A ? , ( a ? Al 3 ? ) ? 1 - a ? f o (a?,\,(3?)?F af ? ( a ?A )3?)?1- )3? r e i (cr?AP?)?, (a ?? A/1 ? )? 1 - a ? Ali? a p (a ? Ap ? )? p l 1 (a A fl)?1- a Af i y i o c 7 A 1- (3a ?A 13)? A a A p ) n g , c u1. u sin g replacement theorem (see theorem 2.3), and the hypothesis ' a •::,> a ?' and 'f3 f i ?' t 2. t h e same proof of (a ?A p?)?1- a? . A c In this example, the cuted formula is (a?Ap?)?—> a?? and the cut setquent conclusion u is (a ? A P?)?1- a?. Although the former is not the ima l fo rm o f the latter, you can apply the replacement theorem to plicative lsatisfyythis restriction since we have a ? ? <,> a? (see theorem 2.4). , PROOF-THEORETICAL CONSIDERATIONS A ND EPISTEMIC INCONSISTENCY 257 To sum up, subformula property is somehow modified to the following: 'all formulas that appear in proofs where the only remaining possible cuts are those pointed above are either subformulas (or rewriting formulas) o f those wh ich o ccu r i n th e endsequent o r a re imp lica t ive f o rms o f subformulas (or rewriting formulas) o f them. Rewriting formulas are those obtained through distribution rules or by the replacement theorem applied to strong equivalences'. 3. Inversion Principle: R 2 not have a le ft (operational) ru le to confront with . Indeed, f ro m it does not make sense to eliminate ? in order to get a —› • ? i s?—> Pa? , n a nTherefore, o m a we l really o u do not want an elimination rule to deal with such s case. Thus, the inversion principle, as stated in classical logic, holds to r allu LEI llogical e constants but to ?. It does not carry out any dangerous results. Th e conservative property, the main related result o f the i n inversion principle, still remains by the use of ?-RC (see below) which t couldh be thought e as the 'symmetrical' rule of R s 2 e n s4. Conservative e Property: prima facie, it would seem that ? is not con?. t servative h since we have an ?-rule ( R a In t we do not have an operational ?-rule 'to destroy' the effect of 2 fact, introducing i ? ) w i t h? over o u —>, t i.e., t a n y d s y m o ma-401-.6, e t r i c a l 1. e o sn a? —on I", e -A. 3 ? In truth, this rule will allow to prove (a ? —) )6?) --)(a p ) , and it does not make sense. Nevertheless, ? is s t ill a conservative operator but characterized somehow differently fro m Gentzen operators. In Gentzen (see [Gentzen 35] and [Prawitz 71]), each logical constant is defined locally by introduction and elimination rules in his natural deduction calculi NJ and NK or by right and left operational rules in his logistic sequent calculi L i and L K. The introduction o f any lo g ica l symbol does not presume a premise with a fixed pattern as the implicative formula a —>fl in the case o f R logical symbols, with the exception of ?, obey Gentzen style. 2 ? It urges a n dthe necessity o f characterizing ? properly. We have done it not only by operational ?-rules but also by ?-RC and ?-Cut rule. ?-RC and ?a ? Cut reflect a sort of structural property, an ordering of credibility or plau— › sibility, associated to ?-formulas. Thus, a —)/3 ma y be considered as [ 3 ? stronger than a ? ---)P ? in ?-RC and a stronger than a ? in ?-Cut. Ergo, i n the definition o f the new logical constant ? is characterized not only by t h e c a s e o f L 258 M A R T I N S & PEQUENO (question mark) operational rules as in standard Gentzen logical operators, but also by the structural rule ?-RC and the identity rule ?-Cut. To see why it is still a conservative definition, we have only to analyse if the case where cut is preserved in the proof, i.e., when we use R keep property of ?, i.e., all we have introduced by means 2 ? the , sconservative t i l l of application of R 2 Consider ? c athis n rule: b e d i s I' c a h—43, a A R r g 1 e d 2 ? b - y a we n' have o already said, it has not and could not have an operational As t h1 e left rule 'to cancel' the introduction of ? on the right hand side symmetrical r F. This of cancellation ? will be done through the structural rule ?-RC. Iua r l -?. e 1"1-a?-3/3?, A ?— RIPC - a —q3, A -? Thus, a,the operator introduced b y R 2restore ? -Aon c the a nthird line b thee same formula on the first line which was object 0 of R e l i m i n a t e d , 2 b y ? AR ? -4. Conclusions C and Further . Works a W e p The p l Logic o f Epistemic Inconsistency is here presented through sequent calculus rules. I t is a remarkable example o f a formal system where a i c normal form theorem can be proved but not unrestricted cut elimination. It a hast been shown that a special cut-proof pattern cannot be eliminated in iproofs o where R n . 2 ? rewriting i the s proof in order to become free o ff cut application since R cannot a p? p be 2 l i'sp e lit' into R-3 (the deduction theorem) due to existential in terpretation o f ? over evidences. Fortunately, the preservation o f this cutd . proof does not damage the desirable outcomes o f cut-elimination: 1 ) the I n calculus consistency has still been maintained; 2 ) subformula property, s u although slig h tly modified: th e cuted remaining fo rmu la has a n imc h plicational fo rm o f endsequent subformulas. Thus, formulas in a (quasi) ccut-free i r proof c is still predictable; 3) the inversion principle has not been u m s attested to ? tsince a symmetrical left operational rule to R a c oe e s 2 ?n d n o t m a k e s , t h e r e i PROOF-THEORETICAL CONSIDERATIONS A ND EPISTEMIC INCONSISTENCY 259 sense, but 4) the conservativeness of all logical constants has been assured, even for ? where ?-RC plays the role of symmetrical rule to R 2 As ? future work, we aim to extend sequent calculus treatment to those dynamic aspects of incomplete knowledge reasoning covered by IDL. The idea is to get an uniform formalization encompassing all the aspects of this kind of reasoning. Laborat6rio de inteligencia artificial Departamento de computa0o Universidade Federal do Ceará P.O.Box 12166, Fortaleza, CE 60455-760 Brasil e-mail: ana@ lia.ufc.br REFERENCES Buchsbaum, A.R. 'L6gicas da Inconsistencia e Incompletude: Sernântica, Axio ma tiza 0 o e Automatizaçâo'. P h D Thesis. Departamento d e Informdtica. PontifIcia Universidade Catélica. Rio de Janeiro. 1995. Correa, M.S., Buchsbaum, A.R. & Pequeno, T. 'Sensible Inconsistent Reasoning: A Tableau System for LEI.' Technical Notes of AAAI Fall Symposium on Automated Deduction in Non-Standard Logics, October 1993. da Costa, N.C.A. 'On the Theory o f Inconsistent Formal Systems.' Notre Dame Journal of Formal Logic, 15. 1974. pp.497-510. Gentzen, G . 'Untersuchungen t ib e r d a s lo g isch e Schliessen', i n Mathematische Zeitschrift, B . 39, pp. 176-210, 405-431. (En g lish translation 'Investigations into Logical Deductions', in The Collected Papers o f Gerhard Gentzen, M.E.Szabo (ed.), series Studies in Logic and The Foundations o f Mathematics, North-Holland, Amsterdam, 1969, xiv-1-338pp, pp.68-131.) Martins, A.T.C. & Pequeno, T. 'Sequent Calculi for the Logic of Epistemic Inconsistency and for the Inconsistent Default Logic'. Qualification Exam. Departamento d e In fo rma tica . Universidade Federal d e Pernambuco. Recife. To appear. First Draft 1994. Martins, A.T.C. & Pequeno, T. ' A Sequent Calculus f o r the Logic o f Epistemic Inconsistency'. I n : Proceedings o f the I 1 th B ra zilia n Symposium on A rtificia l Intelligence. Fortaleza, Bra zil, Oct. 17-20, 1994. McCarthy, J. 'Circumscription - A Form o f Nonmonotonic Reasoning'. Artificial Intelligence, 13:27-39, 1980. Pequeno, T.H.C. ' A Logic fo r Inconsistent Nonmonotonic Reasoning.' Technical Report 90/6. Department of Computing, Imperial College. London. 1990. 260 M A R T I N S & PEQUENO Pequeno, T. H. C. & Buchsbaum, A . R . ' T h e L o g ic o f E p iste mic Inconsistency'. In: Proceedings of the 2nd International Conference on Principles o f Knowledge Representation and Reasoning. Boston, USA. 1991. Prawitz, D. 'Ideas and results in proof theory'. In: Proceedings of the 2nd Scandinavian Logic Symposium (ed. by I.E.Fenstad), North-Holland, Amsterdam, 1971. pp. 235-307. Reiter, R. ' A Logic o f default reasoning'. Artificia l Intelligence, 13:81132, 1980. Ungar, A .M. 'Normalization, Cut-elimination and the theory o f proofs'. CSLI lecture notes, 28, 1992.
© Copyright 2026 Paperzz