Logic and Databases Past, Present, and Future

AI Magazine Volume 18 Number 3 (1997) (© AAAI) Articles Logic and Databases Past, Present, and Future

Jack Minker

■ At a workshop held in Toulouse, France, in 1977, a number of other individuals also had the idea Gallaire, Minker, and Nicolas stated that logic and of using logic as a mechanism to handle data- databases was a field in its own right. This was the bases and deduction, and they were invited to first time that this designation was made. The participate in the workshop. The book Logic impetus for it started approximately 20 years ago and Data Bases (1978), edited by Gallaire and in 1976 when I visited Gallaire and Nicolas in Minker, was highly influential in the develop- Toulouse, France. In this article, I provide an ment of the field, as were the two volumes of assessment about what has been achieved in the 20 years since the field started as a distinct disci- Advances in Database Theory (Gallaire, Minker, pline. I review developments in the field, assess and Nocholas 1984a, 1981) that were the result contributions, consider the status of implementa- of two subsequent workshops held in Tou- tions of deductive databases, and discuss future louse. Another influential development was work needed in deductive databases. the article by Gallaire, Minker, and Nicolas (1984b), which surveyed work in the field to he use of logic and deduction in databas- that point. es, as noted in Minker (1988b), started in The use of logic in databases was received by Tthe late 1960s. Prominent among devel- the database community with a great deal of opments was work by Levien and Maron (1965) skepticism: Was deductive databases (DDBs) a and Kuhns (1967), and by Green and Raphael field? Did DDBs contribute to database theory (1968a), who were the first to realize the impor- or practice (Harel 1980)? The accomplishments tance of the Robinson (1965) resolution princi- I cite in this article are testaments to the fact ple for databases. For early uses of logic in data- that logic has contributed significantly both to bases, see Minker (1988b), and for detailed the theory and the practice of databases. It is descriptions of many accomplishments made clear that logic has everything to do with the in the 1960s, see Minker and Sable (1970). theory of databases, and many of those who A major influence on the use of logic in data- were then critical of the field have changed bases was the development of the field of logic their position. In the remainder of this article, programming: Kowalski (1974) promulgated I describe what I believe to be the major intel- the concept of logic as a programming lan- lectual developments in the field, the status of guage, and Colmerauer and his students devel- commercial implementations, and future oped the first Prolog interpreter (Colmerauer et trends. As we see, the field of logic and data- al. 1973). I refer to logic programs that are bases has been prolific. function free as deductive databases (DDBs), or as datalog. I do so because databases are finite Intellectual Contributions of structures. Most of the results discussed can be extended to include logic programming. Deductive Databases The impetus for the use of logic in databases In 1970, Codd (1970) formalized databases in came about through meetings in 1976 in terms of the relational calculus and the rela- Toulouse, France, when I visited Herve Gallaire tional algebra. He provided a logic language and Jean-Marie Nicolas while on sabbatical. and the relational calculus and described how The idea of a workshop on logic and databases to compute answers to questions in the rela- was also conceived at this time. It is clear that tional algebra and the relational calculus. Both

the relational calculus and the relational alge- between logic-based systems and knowledge- bra provide declarative formalisms to specify based systems; (10) a formalization of how to queries. This was a significant advance over handle incomplete information in knowledge network and hierarchic systems (Ullman 1989, bases; and (11) a correspondence that relates 1988), which only provided procedural lan- alternative formalisms of nonmonotonic rea- guages for databases. The relational algebra and soning to databases and knowledge bases. the relational calculus permitted individuals I address the area of implementations of who were not computer specialists to write DDBs in Implementation Status of Deductive declarative queries and have the computer Databases, where commercial developments answer the queries. The development of syn- have not progressed as rapidly as intellectual tactic optimization techniques (Ullman 1989, developments. I then discuss some trends and 1988) permitted relational database systems to future directions in Emerging Areas and Trends. retrieve answers to queries efficiently and com- Formalizing pete with network and hierarchic implementa- Formalizing Database Theory tions. Relational systems have been enhanced Reiter (1984) was the first to formalize databas- databases to include views. A view, as used in relational es in terms of logic and noted that underlying through logic databases, is essentially a nonrecursive proce- relational databases were a number of assump- dure. There are numerous commercial imple- tions that were not made explicit. One has played a mentations of relational database systems for assumption deals with negation, that facts not significant large database manipulation and for personal known to be true in a relational database are computers. Relational databases are a forerun- assumed to be false. This assumption is the role in our ner of logic in databases. well-known closed-world assumption (CWA), understanding Although relational databases used the lan- expounded earlier by Reiter (1978). The of what guage of logic in the relational calculus, it was unique-name assumption states that any item in not formalized in terms of logic. The formal- a database has a unique name and that indi- constitutes ization of relational databases in terms of logic viduals with different names are different. The a database, and the extensions that have been developed domain-closure assumption states that there are are the focus of this article. Indeed, formaliz- no other individuals than those in the data- what is ing databases through logic has played a signif- base. Reiter then formalized relational databas- meant by icant role in our understanding of what consti- es as a set of ground assertions over a language tutes a database, what is meant by a query, ᑦ together with a set of axioms. The language a query, what is meant by an answer to a query, and ᑦ does not contain function symbols. These what is how databases can be generalized for knowl- assertions and axioms are as follows: … edge bases. It has also provided tools and Assertions: R(a1, , an), where R is an n-ary meant by an … answers to problems that would have been relational symbol in ᑦ, and a1 , an are constant answer to extremely difficult without the use of logic. symbols in ᑦ. … a query, In the remainder of the article, I focus on Unique-name axiom: If a1, , ap are all the some of the more significant aspects con- constant symbols of ᑦ, then and how tributed by logic in databases: (1) a formaliza- ≠ … ≠ ≠ … (a1 a2), , (a1 ap), (a2 a3), , tion of what constitutes a database, a query, ≠ databases (ap–1 ap) . and an answer to a query; (2) a realization that Domain-closure axiom: If a , …, a are all can be logic programming extends relational data- 1 p the constant symbols of ᑦ, then generalized bases; (3) a clear understanding of the seman- ᭙ … . ((for knowledge tics of large classes of databases that include X((X = a1) ٚ ٚ (X = ap alternative forms of negation as well as dis- Completion Axioms: For each relational bases. junction; (4) an understanding of relationships 1 … 1 … m … m symbol R, if R(a1, an), , R(a 1, , a n) denote between model theory, fixpoint theory, and all facts under R, the completion axiom for R proof procedures; (5) an understanding of the is properties that alternative semantics can have … … → ᭙X1 ᭙ Xn (R(X1, , Xn) and their complexity; (6) an understanding of 1 … 1 … = X1 = a1 ٙ ٙ Xn = an) ٚ ٚ (X1) what is meant by integrity constraints and m … m . (( a ٙ ٙ Xn = a = how they can be used to perform updates, 1 n semantic query optimization (SQO), coopera- Equality Axioms: tive answering, and database merging; (7) a ᭙X(X = X) formalization and solutions to the update and ᭙X ᭙Y ((X = Y) → (Y = X)) ((view-update problems; (8) an understanding of ᭙X ᭙Y ᭙Z ((X = Y) ٙ (Y = Z) → (X = Z … … bounded recursion and recursion and how ᭙X1 ᭙Xn (P(X1, , Xn) ٙ … (they can be implemented in a practical man- (X1 = Y1) ٙ ٙ (Xn = Yn → … ner; (9) an understanding of the relationship P(Y1, Yn )) .

22 AI MAGAZINE Articles

Example 1 illustrates the translation of a small database to logic. It is clear that handling such databases through conventional tech- Consider the family database to consist of the Father rela- niques will lead to a faster implementation. tion with schema Father(father, child) and the Mother rela- However, it serves to formalize previously tion with schema Mother(mother, child). Let the database be unformalized databases. The completion axiom was proposed by Clark FATHER father child (1978) as the basis for his negation-as-failure rule: It states that the only tuples that a relation j m can have are those that are specified in the rela- j s tional table. This statement is implicit in every relational database. The completion axiom MOTHER mother child makes this explicit. Another contribution of logic programs and databases is that the formal- r m ization of relational databases in terms of logic r s permits the definition of a query and an answer to a query to be defined precisely. A query is a The database translated to logic is given as follows; we do statement in the first-order logic language ᑦ. not include the equality axioms because they are obvious. Q(a) is an answer to a query, Q(X), over a data- Assertions: Father(j, m), Father(j, s), Mother(r, m), Mother(r, base DB if Q(a) is a logical consequence of DB. s), where Father and Mother are predicates, and j, m, s, and r are constants. Deductive Databases Unique-Name Axiom: Relational databases are a special case of DDBs. ((j ≠ m), (j ≠ s), (j ≠ r), (r ≠ m), (r ≠ s), (m ≠ s)) . A DDB can be considered as a theory, DDB, in Domain-Closure Axiom: which the database consists of a set of ground . ((᭙X)((X = j) ٚ (X = m) ٚ (X = s) ٚ (X = r) assertions, referred to as the extensional data- Completion Axioms: base (EDB), and a set of axioms, referred to as ᭙X ᭙X )(Father(X , X ) ← ((X = j) ٙ (X = m)) ٚ) the intensional database (IDB), of the form 1 2 1 2 1 2 . (((X = j) ٙ (X = s)) ← 1 2 P Q1, …, Qn , (1) ← ᭙X1᭙X2)(Mother(X1, X2) ((X1 = r) ٙ (X2 = m)) ٚ) . (((where P, Q1, …, Qn are atomic formulas in the ((X1 = r) ٙ (X2 = s language ᑦ. Databases of this form are termed datalog databases (Ullman 1989, 1988). A datalog database is a particular instance of a more general Horn logic program that permits function symbols in clauses given by formula 1. The Example 1. Translation of a Small Database to Logic. recognition that logic programs are significant for databases was understood by a number of individuals in 1976 (see Gallaire and Minker rem of the DDB. For alternative definitions of [1978] for references). The generalization per- integrity constraints, see Reiter (1990, 1988) mits views to be defined that are recursive. and Demolombe and Jones (1996). The recognition that logic programming In DDBs, the semantic aspects of a data- and databases are fundamentally related has base’s design can be captured by integrity con- led to more expressive and powerful databases straints. Information about functional depen- than is possible with relational databases dencies—that a relation’s key functionally defined in terms of the relational algebra. determines the rest of the relation’s attribute— That logic programming and DDBs are fun- can be written via integrity constraints. For damentally related is a consequence of the fact example, assume the predicate flight for an air- that databases are function-free logic pro- line database and that the attributes Airline grams. As shown in many papers and, in par- and No. are a composite key for the relation. ticular, Gottlob (1994), the expressive power of One of the functional dependencies—that the logic programming extends that of relational departure time is functionally determined by databases. airline and flight number—is represented by In addition to defining a database in terms ⇐ of an EDB and an IDB, it is necessary to formal- Dtime[1] = Dtime[2] ize what is meant by an integrity constraint. flight(Airline, No., Dtime[1],-, …,-), Kowalski (1978) suggests that an integrity con- flight(Airline, No., Dtime[2],-, …,-) , straint is a formula that is consistent with the where ⇐ is used to distinguish a rule from an DDB, but for Reiter (1984) and Lloyd and integrity constraint. Topor (1985), an integrity constraint is a theo- Likewise, inclusion dependencies, which are

FALL 1997 23 Articles

also common semantic information from a query fails, a user, in general, cannot tell why database’s design, are easily represented. For the failure occurred. There can be several rea- example, say the predicate airport records vari- sons: The database currently does not contain ous information about airports known to the information to respond to the user, or there database. We want to ensure that any airport will never be an answer to the query. The dis- that serves as a departure or an arrival of any tinction could be important to the user. Anoth- flight known to the database is also in the air- er aspect related to integrity constraints is that port relation. The first of these—that the of user constraints. A user constraint is a formula departure airport is known—could be repre- that models a user’s preferences. It can con- sented as follows: strain providing answers to queries in which airport(-, …, -, Fieldcode) ⇐ flight(-, …, -, the user might have no interest (for example, Fieldcode) . stating that in developing a route of travel, the user does not want to pass through a particular The major use made of integrity constraints city) or provide other constraints that might has been in updating, to assure that the data- restrict the search. As shown by Gaasterland, base is consistent. Nicolas (1979) used techniques from DDBs to speed database update. Godfrey, and Minker (1992b), user constraints, Blaustein (1981) has also made contributions which are identical in form to integrity con- to this problem. Reiter (1978a) showed that straints, can be used for this purpose. Although one can query a Horn database with or without integrity constraints provide the semantics of integrity constraints, and the answer to the the entire database, user constraints provide query is the same. However, integrity con- the semantics of the user. User constraints can straints can be used to advantage in the query be inconsistent with the database; hence, these A cooperative process. Although integrity constraints do not two semantics are maintained separately. To answering affect the result of a query, they can affect the maintain the consistency of the database, only efficiency with which answers can be comput- integrity constraints are relevant. A query can system ed. Integrity constraints provide semantic be thought of as the conjunction of the query provides information about data in the database. If a itself and the user constraints. A query can be information query requests a join for which there will never optimized semantically based on both integrity be an answer because of system constraints, constraints and user constraints. to users about then an unnecessary join on two potentially As noted previously, integrity constraints are why a large relational tables in a relational database versatile; they do more than just represent system or performing a long deduction in a dependencies. General semantic information particular DDB is not needed when integrity constraints can be captured as well. Assume that at the query imply the answer is empty. The process of using national airport in Washington, D.C. (DCA), integrity constraints to constrain a search is that no flights are allowed (departures or succeeded called semantic query optimization (SQO) arrivals) after 10:00 PM or before 8:00 AM because or failed. (Chakravarthy, Grant, and Minker 1990). the airport is downtown, and night flights McSkimin and Minker (1977) were the first to would disturb city residents. This information use integrity constraints for SQO in DDBs. can be captured as an integrity constraint. Hammer and Zdonik (1980) and King (1981) Such knowledge, captured and recorded as were the first to apply SQO to relational data- integrity constraints, can be used to answer bases. Chakravarthy, Grant, and Minker (1990) queries to the database more intelligently and formalized SQO and developed the partial sub- more informatively. If someone asks for flights sumption algorithm and method of residues. out of DCA to, say, Los Angeles International The partial subsumption algorithm and Airport leaving between 10:30 PM and 12:00 method of residues provides a general tech- AM, the database could simply return the emp- nique applicable to any relational database or ty answer set. (There will be no such flights if DDB that is able to perform SQO. The general the database is consistent with its constraints.) approach to SQO described in Chakravarthy, It would be better, however, for the database Grant, and Minker (1990) has been extended to system to inform the querier that there can be perform bottom-up evaluation (Godfrey, Gryz, no such flights because of the Washington, and Minker 1996); to include databases with D.C., flight regulations. negation in the body of clauses (Gaasterland Using user constraints and integrity con- and Lobo 1993); and to handle recursive IDB straints, one can develop a system that informs rules (Levy and Sagiv 1995). users why a query succeeds or fails (Gaasterland A topic related to SQO is that of cooperative et al. 1992). Other features can be incorporated, answering systems. A cooperative answering sys- such as the ability to relax a query, termed tem provides information to users about why a query relaxation, given that it fails, so that an particular query succeeded or failed. When a answer to a related request can be found. See

24 AI MAGAZINE Articles

Gaasterland, Godfrey, and Minker (1992a) for a model. The unique minimal model is the same survey of cooperative answering systems. as all the atoms in the fixpoint and are the SQO, user constraints, and cooperative only atoms provable from the theory. answering systems are important contributions To find if the negation of a ground atom is both for relational database and DDB systems. true, one can subtract, from the Herbrand base They will eventually be incorporated into com- (the set of all atoms that can be constructed mercial relational database and DDB systems. from the constants and the predicates in the Indeed, I cannot imagine a DDB developed database), the minimal Herbrand model. If the for commercial systems to be successful if it atom is contained in this set, then it is does not contain both SQO and cooperative assumed to be false, and its negation is true. answering capabilities. How can one expect Alternatively, answering queries that consist of users to understand why deductions succeed or negated atoms that are ground can be achieved fail if such information is not provided? How using negation-as-finite failure as described by can queries doomed to fail because they violate Reiter (1978b) and Clark (1978). user constraints or integrity constraints be The first approaches to answering queries in allowed to take up a significant amount of DDBs did not handle recursion and were pri- search time if the query cannot possibly suc- marily top-down (or backward reasoning) ceed? I also believe that these techniques must (Gallaire and Minker 1978). Answering queries be incorporated into relational technology. As in relational database systems was a bottom-up discussed in Implementation Status of Deduc- (or forward-reasoning) approach because all tive Databases, systems are beginning to incor- answers are usually required, and it is more I cannot porate SQO techniques. Practical considera- efficient to do so in a bottom-up approach. imagine tions of performing in a bottom-up approach The major approaches to handling recursion have been addressed by Godfrey, Gryz, and are based on the renaming of the Alexander a DDB Minker (1996). (Rohmer, Lescoeur, and Kerisit 1986) and mag- developed for ic set (Bancilhon et al. 1986) methods, which Extended Deductive make use of constants that appear in a query commercial Database Semantics and perform search by bottom-up reasoning. systems to The first generalization of relational databases Bry (1990) reconciles the bottom-up and top- was to permit function-free recursive Horn down methods to compute recursive queries. be successful rules in a database, that is, rules in which the He shows that the Alexander and magic set if it does not head of a rule is an atom, and the body of a methods based on rewriting and the methods contain both rule is a conjunction of atoms. These databases based on resolution implement the same top- are DDBs, or datalog databases. Subsequently, down evaluation of the original database rules SQO and other DDBs that might contain negated atoms by means of auxiliary rules processed bottom- cooperative in the body of rules were permitted. These up. For pioneering work on recursion and alternative extensions and their significance alternative methods, see Minker (1996). Min- answering are described in the following subsections. ker and Nicolas (1982) were the first to show capabilities. Horn Semantics and Datalog One of the that there are forms of rules that lead to bound- early developments was by van Emden and ed recursion, in which the deduction process Kowalski (1976), who wrote a seminal paper must terminate in a finite number of steps. on the semantics of Horn theories. Van Emden This work has been extended by Naughton and Kowalski made a significant contribution and Sagiv (1987). Example 2 illustrates a rule to logic and databases by recognizing that the that terminates finitely regardless of the state semantics of Horn databases can be character- of the database. ized in three distinct ways: (1) model theory, The efficient handling of recursion and the (2) fixpoint theory, and (3) proof theory. These recognition that some recursive cases might three characterizations lead to the same inherently be bounded contributes to the prac- semantics. tical implementation of DDBs. An understand- Model theory deals with a collection of mod- ing of the relationship between resolution- els that capture the intended meaning of the based (top-down) and fixpoint–based database. Fixpoint theory deals with a fixpoint (bottom-up) techniques and how the search operator that constructs the collection of all space of the latter can be made identical to atoms that can be inferred to be true from the top-down resolution with program transfor- database. Proof theory deals with a procedure mation is another contribution of DDBs. that finds answers to queries with respect to Extended Deductive Databases and the database. van Emden and Kowalski (1976) Knowledge Bases The ability to develop a showed that the intersection of all Herbrand semantics for theories in which there are rules models of a Horn DDB is a unique minimal with a literal (that is, an atomic formula or the

FALL 1997 25 Articles

negation of an atomic formula) in the head and literals with possibly negated-by-default literals If a rule satisfies the condition that it is singular, then it is in the body of a clause has significantly expand- bound to terminate in a finite number of steps independent ed the ability to write and understand the of the state of the database. A recursive rule is singular if it semantics of complex applications. Such claus- is of the form es, referred to as extended clauses, are given by ← L ← M , …, M , not M , … not M , (2) R F ٙ R1 ٙ … ٙ Rn , 1 n n+1 n+k … where F is a conjunction of possibly empty base (that is, where L and the Mj, j = 1, , (n + k) are literals. Such databases combine both classical nega- EDB) relations and R, R1, R2, …, Rn are atoms that have the same relation name iff (1) each variable that occurs in an tion and default negation (represented by not immediately preceding a literal) and are atom Ri and does not occur in R only occurs in Ri and (2) each variable in R occurs in the same argument position in referred to as extended DDBs. The combining

any atom Ri where it appears, except perhaps in at most one of classical and default negation provides users atom R1 that contains all the variables of R. with greater expressive power. Thus, the rule Logic programs where default negation can R(X, Y, Z) ← R(X, Y’, Z), R(X, Y, Z’) appear in the body of a clause first appeared in the Workshop on Foundations of Deductive is singular because (1) Y’ and Z’ appear, respectively, in the Databases and Logic Programming in August first and second atoms in the head of the rule (condition 1) 1986. Selected papers from the workshop were and (2) the variables X, Y, and Z always appear in the same published in Minker (1988a). The concept of argument position (condition 2). stratification was introduced to logic programs by Apt, Blair, and Walker (1988) and Van Gelder (1988), who considered stratified theo-

ries in which L and the Mj in formula 2 are atomic formulas, and there is no recursion through negation. Apt, Blair, and Walker, and Example 2. Bounded Recursion. Van Gelder, show that there is a unique preferred minimal model, computed from strata to strata. Przymusinski (1988) termed this minimal model the perfect model. When one has a The rules theory that is stratiﬁed, one can place clauses in different strata, where predicates in the head of ← r1: p q, not r a rule are in a higher stratum than predicates ← r2: q p that are negated in the body of the clause, as ← r3: q s explained in example 3. Thus, one can com- r4: s pute the positive predicates in a lower stratum, ← r5: r t and the negated predicate’s complement is true make up a stratiﬁed theory. Rule r5 is in the lowest stratum, in the body of the clause if the positive atom but the other rules are in a higher stratum. The predicate p has not been computed in the lower stratum. is in a higher stratum than the stratum for r because it The theory of stratified databases was fol- depends negatively on r. q is in the same stratum as p lowed by permitting recursion through nega-

because it depends on p. s is also in the same stratum as q. tion in formula 2, where the L and Mj are The meaning of the stratified program is that s, q, and p are atomic formulas. Example 4 illustrates a data- true, but t and r are false. t is false because there is no defin- base that cannot be stratified. In the context of ing rule for t. Because t is false, r is false. s is given as true; DDBs, they are called normal DDBs. Many hence, q is true. Because q is true, and r is false, from rule r1, papers have been devoted to defining the p is true. semantics of these databases. A summary of these semantics is given in Minker and Ruiz (1994). The most prominent of this work for the Horn case are the well-founded semantics (WFS) of Van Gelder, Ross, and Schlipf (1988) Example 3. Stratified Program. and the stable semantics of Gelfond and Lif- schitz (1988). The WFS leads to a unique three- valued model. Stable semantics can lead to a collection of minimal models. For some DDBs, this collection can be empty. Fitting (1985) also defined a three-valued model to capture the semantics of normal logic programs. For additional work, see Minker (1996).

26 AI MAGAZINE Articles

There have been several implementations of the WFS. Chen and Warren (1993) developed a top-down approach to answer queries in this Consider the database given by semantics, while Leone and Rullo (1992) devel- ← r1: p(X) not q(X) oped a bottom-up method for datalog databas- ← r2: q(X) not p(X) es. Several methods have been developed for ← r3: r(a) p(a) computing answers to queries in stable model r : r(a) ← q(a) . semantics. Fernández et al. (1993) developed a 4 bottom-up approach based on the concept of Notice that clauses r1 and r2 are recursive through negation. model trees. Every branch of a model tree is a Hence, the database is not stratifiable. According to the model of the database, where a node in a tree is WFS, {p(a), q(a), r(a)} are assigned unknown. However, for an atom that is shared by each branch below the stable model semantics, there are two minimal models: {{p(a), r(a)}, {q(a), r(a)}}. Hence, one can conclude that r(a) is the node. (See example 6 for an illustration of true, but the disjunct, p(a) q(a) is true in the stable model a model tree.) Bell et al. (1993) developed a ٚ semantics. method based on linear programming. See Minker (1996) for additional methods to compute the well-founded, the stable model, and other related semantics. A further extension of normal DDBs, pro- Example 4. Nonstratifiable Database. posed by Gelfond and Lifschitz (1990) and Pearce and Wagner (1989), permits clauses in formula 2, where L and Mj are literals, and, therefore, combines classical and default nega- DDBs, together with integrity constraints, per- tion in one database. The semantics for normal mit a wide range of knowledge bases to be DDBs was described by Minker and Ruiz (1994). implemented. Many papers devoted to knowl- These notions of default negation have been edge bases consider them to consist of facts used as separate ways to interpret and deduce and rules, which is one aspect of a knowledge default information. That is, each application base, as is the ability to extract proofs. Howev- has chosen one notion of negation and applied er, integrity constraints supply another aspect it to every piece of data in the domain of the of knowledge and differentiate knowledge application. Minker and Ruiz (1996) defined a bases, which can have the same rules but dif- new class of more expressive DDBs that allow ferent integrity constraints. One should define several forms of default negation in the same a knowledge base as consisting of an extended database. In this way, different pieces of infor- DDB plus integrity constraints. mation in the domain can be treated appropri- Since alternative extended deductive seman- ately. They introduced a new semantics called tics have been implemented, the knowledge the well-founded stable semantics that character- base expert can now focus on the problem to izes the meaning of DDBs that combine the be implemented, that is, on writing rules and well-founded and the stable semantics. Schlipf integrity constraints that characterize the (1995) has written a comprehensive survey arti- knowledge bases, selecting the particular se- cle on complexity results for DDBs. mantics that meets the needs of the problem, The development of the semantics of ex- and employing a DDB system that uses the tended DDBs that permit a combination of required semantics. The field of DDBs has con- classical negation and multiple default nega- tributed to providing an understanding of tions in the same DDB are important contribu- knowledge bases and their implementation. tions. The study and development of results in Extended Disjunctive the computational complexity of these databases are important contributions to database Deductive Database Semantics theory. They permit wider classes of applica- In the databases discussed previously, information to be developed. tion is definite. However, there are many situ- Knowledge bases are important for AI and ations where our knowledge of the world is expert system developments. A general way to incomplete. For example, when a null value represent knowledge bases is through logic. appears as an argument of an attribute of a Work developed for extended DDBs concern- relation, the value of the attribute is unknown. ing semantics and complexity applies directly Uncertainty in databases can be represented by to knowledge bases. Baral and Gelfond (1994) probabilistic information (Ng and Subrahman- describe how extended DDBs can be used to ian 1993). Another area of incompleteness aris- represent knowledge bases. For an example of es when it is unknown which among several a knowledge base, see example 5. Extended facts are true, but it is known that one or more

FALL 1997 27 Articles

clauses are given by formula 3, where the literals are restricted to atoms, and there is no Consider the following database, where the predicate default negation in the body of a clause. Next, p(X,Y) denotes that X is a professor in department Y, a(X,Y) I discuss the semantics of EDDDBs, where there denotes that individual X has an account on machine Y, are no restrictions on clauses in formula 3. and ab(W, Z ) denotes that it is abnormal in rule W to be individual Z. Disjunctive Deductive Databases We want to represent the following information, where As noted in Minker (1989), work in disjunctive mike and john are professors in the Computer Science theories was pursued seriously after a work- Department: shop organized in 1986 (Minker 1986). The First, as a rule, professors in the Computer Science field of DDDBs started approximately in 1982 Department have Vax accounts. This rule is not applicable with the appearance of a paper I wrote (Minker to mike. He might or might not have an account on this 1982), in which I described how one can machine. answer both positive and negated queries in Second, every computer science professor has a Vax or an such databases. For a historical perspective of IBM account but not both. These rules can be captured in disjunctive logic programming and DDDBs, the following DDB: see Minker (1989). There is a major difference p(mike, cs) ← between the semantics of DDBs and those for p(john, cs) ← DDDBs. Whereas DDBs typically have a ¬p(X, Y) ← not p(X, Y) unique minimal model that describes the a(X, vax) ← p(X, cs), not ab(r4, X), not ¬a(X, vax) meaning of the database, DDDBs generally ab(r4, mike) ← have multiple minimal models. a(X, vax) ٚ a(X, ibm) ← p(X, cs), ab(r4, X) As shown in Minker (1982), it is sufficient to ¬ a(X, ibm) ← p(X, cs), a(X, vax) answer positive queries over DDDBs by show- ¬a(X, vax) ← p(X, cs), a(X, ibm) ing that the query is satisfied in every minimal a(X, ibm) ← a(X, vax), p(X, cs) . model of the database. Thus, in the DDDB a ٚ The third rule states that if by default negation, predicate b, there are two minimal models: (1) {a} and (2) p(X, Y) fails, then p(X,Y) is logically false. The other rules {b}. The query, a?, is not satisfied in the model encode the statements listed previously. b; hence, it cannot be concluded that a is true. From this formalization, one can deduce that john has a However, the query (a ٚ b) is satisfied in both Vax account, but mike has either a Vax or an IBM account minimal models; hence, the answer to the ,but not both. query a ٚ b is yes. To answer negated queries it is not sufficient to use Reiter’s (1978) CWA because as he noted, from the theory DB = a ٚ b, it is not possible to prove a, and it is not possible to prove b. Hence, by the CWA, not a and Example 5. Knowledge Base (Baral and Gelfond 1994). not b follow. However, {a ٚ b, not a, not b} is not consistent. The generalized closed-world assumption (GCWA) (Minker 1982) resolves this is true. Therefore, it is necessary to be able to problem by specifying that a negated atom be represent and understand the semantics of considered true if the atom does not appear in theories that include incomplete data. A natur- any minimal model of the database. The al way to extend databases to include incom- GCWA provides a model-theoretic definition plete data is to permit disjunctive statements of negation. An equivalent proof-theoretic def- as part of the language, where clauses can have inition, also presented in Minker (1982), is disjunctions in their heads. These clauses are that an atom a can be considered false if when- represented as ,ever a ٚ C can be proven from the database ← ,L1 ٚ L2 ٚ … ٚ Lm M1, …, Mn, not Mn+1, then C can also be proven from the database … not Mn+k (3) where C is an arbitrary positive clause. For and are referred to as extended disjunctive claus- related work on negation in disjunctive theo- es. Such databases are referred to as extended ries, see Minker (1996). For surveys on nega- disjunctive deduct ive databases (EDDDBs). Foun- tion in DDBs and DDDBs, see Shepherdson dations of Disjunctive Logic Programming by (1987), Apt and Bol (1994), and Minker (1993). Lobo, Minker, and Rajasekar (1992) describes In DDBs, it is natural for the fixpoint opera- the theory of disjunctive logic programs and tor to map atoms to atoms. However, for includes several chapters on disjunctive deduc- DDDBs, it is natural to map positive disjunctive databases (DDDBs). Example 5 illustrates tions to positive disjunctions. A set of positive the use of such a theory of databases. disjunctions is referred to as a state. A model I first discuss the semantics of DDDBs, where state is a state whose minimal models all satisfy

28 AI MAGAZINE Articles the DDDB. The concept of a state was defined by Minker and Rajasekar (1990) as the domain of a fixpoint operator T whose least fixpoint p ∈* characterizes the semantics of a disjunctive logic program P. The operator is shown to be monotonic and continuous; hence, it con- verges in a countably infinite number (ω) of a(1) iterations. The fixpoint computation operates bottom-up and yields a minimal model state that is logically equivalent to the set of mini- a(2) b(2) mal models of the program. The Minker- Rajasekar fixpoint operator is an extension of the van Emden–Kowalski fixpoint operator. If one considers all model states of a DDDB and b(1) intersects them, then the resultant is a model state, and among all model states, it is minimal. Consider the following example given by the database: Hence, one obtains a unique minimal model in -a(1); a(2) ٚ b(2); b(1) ٚ b(2)}. There are two minimal mod} a Horn database, but one obtains a unique els for this database {{a(1), a(2), b(1)}, {a(1), b(2)}}. These model state in a DDDB. See Decker (1991) for a models can be written as a tree (see figure above). related fixpoint operator for DDDBs. Answering queries in DDDBs has been studied by a number of individuals, as described in Minker (1996). I focus on the work of Fernán- dez and Minker (1991), who developed the Example 6. Model Tree. concept of a model tree. They show how one can incrementally compute sound and complete answers to queries in hierarchical DDDBs. is available on the World Wide Web. A model tree is shown in example 6. A DDDB See Minker (1996) for references to work on is hierarchical if it contains no recursion. Fer- the complexity of answering queries in dis- nández et al. (1993) show how one can develop junctive logic programs and Eiter and Gottlob a fixpoint operator over trees to capture the (1995) for complexity results for propositional meaning of a DDDB that includes recursion. logic programs. The tree representation of the fixpoint is equiv- The development of model-theoretic, alent to the Minker-Rajasekar fixpoint (Minker fixpoint, and proof procedures placed the and Rajasekar 1990). Fernández and Minker semantics of DDDBs on a firm foundation. compute the model tree of the extensional Methods to handle DDDBs are being devel- DDDB once. To answer queries, intensional oped and should eventually enhance imple- database rules can be invoked. However, the mentations. The GCWA and alternative theo- models of the extensional disjunctive part of ries of negation have enhanced our the database do not have to be generated for understanding of default negation in DDDBs. each query. Their approach to computing Complexity results provide an understanding answers generalizes both to stratified and nor- of the difficulties in finding answers to queries mal DDDBs. in such systems. Loveland and his students (Loveland, Reed, and Wilson 1993) have developed a top-down Extended Disjunctive approach when the database is near Horn. Deductive Databases They have developed a case-based reasoner Fernández and Minker (1995) present a new that uses Prolog to perform the reasoning. fixpoint characterization of the minimal mod- This effort is one of the few that have imple- els of disjunctive and stratified DDDBs. They mented DDDBs. Loveland, Reed, and Wilson prove that by applying the operator iteratively, (1993) introduced a relevancy-detection algo- in the limit, it constructs the perfect model rithm to be used with SATCHMO, developed by semantics (Przymusinski 1988) of stratified Manthey and Bry (1988), for automated theo- DDDBs. Given the equivalence between the rem proving. Their system, termed SATCHMORE perfect model semantics of stratified programs (SATCHMO with RElevancy), improves on SATCH- and prioritized circumscription (Przymusinski MO by limiting uncontrolled use of forward 1988), their fixpoint characterization captures chaining. Seipel (1995) has developed a sys- the meaning of the corresponding circum- tem, DISLOG, that incorporates many different scribed theory. Based on these results, they pre- disjunctive theories and strategies. The system sent a bottom-up evaluation algorithm for

FALL 1997 29 Articles

stratiﬁed DDDBs. This algorithm uses the mod- Second, there is a positive (negative) arc el-tree data structure to represent the informa- from a predicate node p to a rule node ζ iff p tion contained in the database and to compute appears positive (negative) in the body of ζ answers to queries. Fernández and Minker and an arc from ζ to p (resp., and also an arc (1992) develop the theory of DDDBs using the from p to ζ) if p appears in the head of ζ. concept of model trees. Work on updates in The positive dependency graph of P is a sub-

DDDBs is described in Fernández, Grant, and graph of Gp containing only positive arcs. A Minker (1996). directed cycle in Gp is called negative if it con- Four alternative semantics were developed tains at least one negative arc. A DDB P is HCF for nonstratiﬁable normal DDDBs at approxi- if for every two predicate names p and q, if p mately the same time: (1) Ross (1989), the and q are on a positive directed cycle in the

strong WFS; (2) Baral, Lobo, and Minker dependency graph Gp, then there is no rule in (1990), the generalized disjunctive WFS P in which both p and q appear in the head. It (GDWFS); and (3, 4) two semantics by Przy- is shown in Ben-Elyahu, Palopoli, and Zem- musinski, an extension of the stable model lyanker (1996) that answers to queries semantics (Przymusinski 1990) for normal dis- expressed in this language can be computed in junctive databases and the stationary seman- polynomial time. Furthermore, the language is tics (Przymusinski 1990). A number of other sufficiently powerful to express all polynomial important semantics have been developed. time queries. It is further shown in Ben-Elyahu Przymusinski (1995) describes a new semantic and Palopoli (1994) that there is an algorithm framework for disjunctive logic programs and that performs, in polynomial time, minimal introduces the static expansions of disjunctive model finding and minimal model checking if programs. The class of static expansions the theory is HCF. An efficient algorithm for extends both the classes of stable, well-found- solving the (co–NP-hard decision) problem of ed, and stationary models of normal programs checking if a model is stable in function-free and the class of minimal models of disjunctive disjunctive logic programs is developed in programs. Any static expansion of a program P Leone, Rullo, and Scarcello (1996). They show provides the corresponding semantics for P that the algorithm runs in polynomial time on consisting of the set of all sentences logically the class of HCF programs, and in the case of implied by the expansion. The stable model general disjunctive logic programs, it limits the semantics has also been extended to disjunc- inefficient part of the computation only to the tive programs (Gelfond and Lifschitz 1991; components of the program that are not HCF. Przymusinski 1991). The disjunctive WFS In addition to work on tractable databases, (DWFS) of Brass and Dix (1995) is also of con- consideration has been given to approximate siderable interest because it permits a general reasoning. In such reasoning, one can give up approach to bottom-up computation in dis- soundness or completeness of answers. Efforts junctive programs. have been developed both for DDBs and DDDBs As noted previously, a large number of dif- by Kautz and Selman (1992) and Selman and ferent semantics exist for both EDDBs and Kautz (1996), who developed lower and upper EDDDBs. A user who wants to use such a sys- bounds for Horn (datalog) databases and compi- tem is faced with the problem of selecting the lation methods; Cadoli (1993), who developed appropriate semantics for his/her needs. No computational and semantic approximations; guidelines have been developed. However, and del Val (1995), who developed techniques many complexity results have been obtained for approximating and compiling databases. See for these semantics. Schlipf (1995) and Eiter also Cadoli (1996) for additional references con- and Gottlob (1995) have written comprehen- cerning compilation, approximation, and sive survey articles that summarize the com- tractability of knowledge bases. plexity results that are known for alternative A second way to determine the semantics to semantics. be used is through their properties. Dix (1992) In addition to the results for extended dis- proposed a large number of criteria that are junctive theories, there is work in investigating useful in determining the appropriate seman- tractable cases of disjunctive theories. Ben- tics to be used. Properties deemed useful are (1) Eliyahu and Dechter (1994) introduced the elimination of tautologies, where one wants the concept of a head-cycle–free (HCF) clause. Let semantics to remain the same if a tautology is a clause consist of a disjunction of literals. A eliminated; (2) generalized principle of partial

dependency graph Gp is associated with each evaluation, where if a rule is replaced by a one- program P as follows: step deduction, the semantics is unchanged; First, each clause of the form, formula 2, and (3) positive-negative reduction; (4) elimination each predicate in P is a node. of nonminimal rules, where a subsumed rule is

30 AI MAGAZINE Articles eliminated, the semantics remains the same; the user to develop knowledge base systems. (5) consistency, where the semantics is not Second, alternative concepts of negation empty for all disjunctive databases; and (6) have been developed as evidenced by the dif- independence, where if a literal l is true in a pro- ferent semantics for logic programs (for exam- gram P, and P’ is a program whose language is ple, WFS and stable semantics for extended independent of the language of P, then l logic programs and alternative semantics for remains true in the program consisting of the disjunctive logic programs). union of the two languages. Third, complexity results have been found A semantics can have all the properties that for alternative semantics of DDBs, including one might desire and be computationally alternative theories of negation. tractable and yet not provide answers that a Fourth, methods have been developed to user expected. If, for example, the user expect- permit prototype systems to be implemented. ed an answer r(a) in response to a query r(X), Fifth, DDBs can be used as the computation- and the semantics were, for example, the WFS, al vehicle for a wide class of nonmonotonic- the user would receive the answer that r(a) is reasoning theories. unknown. However, if the stable model seman- In this section, I showed how relational The field tics had been used, the answer returned would databases can be formalized in terms of logic, of DDBs be r(a). Perhaps, the best that can be expected permitting databases to be extended beyond is to provide users with complexity results and what is possible with relational databases. Var- has made criteria by which they can decide which ious extensions were discussed, such as DDBs. significant semantics meets the needs of their problems. EDDBs, DDDBs, and EDDDBs. Alternative the- Understanding the semantics of disjunctive ories of negation were discussed, and the intellectual theories is related to nonmonotonic reasoning. semantics of the alternative databases, includ- contributions The field of nonmonotonic reasoning has ing negation, were described. These extensions over the past resulted in several alternative approaches to the were shown to be useful for developing com- performance of default reasoning (Moore 1985; plex knowledge bases. The role of integrity 20 years. McCarthy 1980; McDermott and Doyle 1980; constraints and other constraints for such sys- However, Reiter 1980). The survey article by Minker tems was described. (1993) and papers by Eiter and Gottlob (1995) these and Cadoli and Schaerf (1993) cite results Implementation Status of contributions where alternative theories of nonmonotonic reasoning can be mapped into extended dis- Deductive Databases have not junctive logic programs and databases. Hence, The field of DDBs has made significant intel- been DDDBs can be used to compute answers to lectual contributions over the past 20 years. matched by queries in such theories. See Cadoli and Lenz- However, these contributions have not been erini (1994) for complexity results concerning matched by implementations that are avail- implementa- circumscription and closed-world reasoning. able in the commercial market. In the early tions that See also Yuan and You (1993) for a description 1970s, when Codd (1970) introduced the rela- of the relationships between autoepistemic cir- tional model, there were numerous debates in are cumscription and logic programming. They use the database community about the efficacy of available two different belief constraints to define two such systems relative to network and hierarchi- in the semantics: (1) the stable circumscriptive cal systems (Date 1995). These debates ended semantics and (2) the well-founded circum- when an effective relational system was imple- commercial scriptive semantics for autoepistemic theories. mented and shown to be comparable to these market. The work in Yuan and You (1993) and that on systems. Now, some of those individuals who static semantics developed by Przymusinski are prominent in relational databases claim (1995) appear to be related. that DDBs are not effective and are not need- Another area to which DDDBs have con- ed. Although I believe otherwise, these com- tributed is the null-value problem. If an attribute ments can be addressed better either when a of a relation can have a null value, where this full commercial implementation of a DDB is value is part of a known set, then one can repre- available or when many of the techniques sent this information as a disjunction of rela- introduced in DDBs find their way into relations, where in each disjunction, a different val- tional databases. I believe that both of these ue is given to the argument. For papers on the are beginning to happen. null-value problem both in relational and In the following subsection, I discuss the deductive databases, see Minker (1996). stages through which implementations of There are several significant contributions DDBs have progressed and some contributions of DDDBs: made during each stage. Following this, I dis- First, greater expressive power is provided to cuss the reasons why I believe that no current

FALL 1997 31 Articles

systems are commercially marketed and spec- (1978) on the Deductively Augmented Data ulate on how this situation might change. Management (DADM) system, and Minker (1978) on the Maryland Refutation Proof Pro- Deductive Database Systems cedure System (MRPPS 3.0) represent work dur- There have been three stages of implementa- ing the second stage of development of DDBs. tions of DDBs: (1) pre-1970, (2) 1970 to 1980, These papers appear in Gallaire and Minker and (3) 1980 to the present. Each stage has (1978). Table 2 provides a brief summary of contributed toward understanding the prob- some of the features of these systems. lems inherent in developing DDB systems. DADM precomputed unifications among First Stage: Pre-1970s Two efforts stand premises so they did not have to be recomput- out during this period: the first by Levien, and ed during deduction. Variables were typed. Maron (1965) and Kuhns (1967), who devel- Inference plans and database-access strategies oped a prototype system that demonstrated were created from a premise file without the feasibility of performing deduction in data- requiring access to database values. bases and the second by Green and Raphael MRPPS 3.0 performed top-down searches for (1968a, 1968b), who recognized that the reso- large databases. It permitted arguments of lution method of Robinson (1965) was a uni- predicates to contain function symbols and form procedure based on a single rule of infer- had a knowledge base index to access the data. ence that could be used for DDBs. This was the The deductive system used a typed unification first general approach to DDBs. The work by algorithm and a semantic network. The SQO Levien and Maron (1965) and Kuhns (1967) on method described in McSkimin and Minker Relational Data File (RDF) started in 1963. A (1977) was incorporated into the system. procedural language, INFEREX, executed infer- Answer extraction, natural language process- ence routines. Plausible and formal inferenc- ing, and voice output were part of the system. ing were both possible in RDF, as was temporal The DEDUCE 2 system performed deduction reasoning. The system was implemented on a over databases. Nonrecursive Horn rules were file consisting of some 55,000 statements. The used and were compiled in terms of base rela- work by Green and Raphael (1968a, 1968b) tions. Integrity constraints were used to per- resulted in a system termed the question- form SQO on queries. Problems with respect to answering system (QA-3.5). It was an outgrowth recursive rules and termination were also dis- of Raphael’s thesis on semantic information cussed (Chang 1981). retrieval (SIR) (Raphael 1968) that performed Third Stage: 1980 to Present A large deduction. QA-3.5 included a natural language number of prototype DDBs were developed, component. Another deductive system, rela- and most are described in Ramakrishnan and tional store structure (RSS), started in 1966 was Ullman (1995). I briefly discuss several major developed by Marrill (Computer Corporation efforts: work at the European Computer 1967). The system had 12 deductive rules built Research Consortium (ECRC) led by Nicolas, at into the program and was able to incorporate the University of Wisconsin led by Ramakrish- other deductive rules. The association store nan, and at Stanford led by Ullman. They processor (ASP), developed by Savitt, Love, and attempted to develop operational and possibly Troop (1967), also performed deduction over commercial DDBs. They contributed signifi- binary relational data. The inference rules, cantly to both the theory and the implemen- specified as relational statements, were han- tation of DDBs. Detailed descriptions of contri- dled by breadth-first, followed by depth-first, butions made by these systems and others can search. These efforts, as well as those cited in be found in Ramakrishnan and Ullman (1995). Minker and Sable (1970), were important pre- In table 3, I list some of the capabilities of sys- cursors to DDBs. In table 1, adapted from tems developed in this stage, adapted from Minker and Sable (1970), I list some capabili- Ramakrishnan and Ullman (1995). ties of systems developed during this stage. Implementation efforts at ECRC on DDBs Second Stage: 1970 to 1980 Whereas the started in 1984 and led to the study of algo- first stage could be characterized as using ad rithms and prototypes: deductive query-evalu- hoc techniques for deduction (except for the ation methods (QSQ-SLD and others) (Vieille work by Green and Raphael), the second-stage 1986); integrity checking (SOUNDCHECK) (Deck- systems were based on the Robinson resolu- er 1986); the DDB EKS(-V1) by Vieille and his tion principle, as first recognized by Green and team (1990); hypothetical reasoning and Raphael. The SYNTEX system built by Nicolas checking (Vieille, Bayer, and Kuechenhoff and Syre (1974) used logic as the basis for 1996); and aggregation through recursion deduction. The work by Chang (1978) on the (Lefebvre 1994). The EKS system used a top- DEDUCE 2 system, Kellogg, Klahr, and Travis down evaluation method and was released to

32 AI MAGAZINE Articles

Name QA-3.5 (Green and ASP (Savitt, Love, RDF (Levien and RSS (Computer Raphael 1968a, and Troop 1967) Maron 1965) Corporation 1967) 1968b) (question- (association storing (relational data file) (relational answering system) processor) structures system) Organization Stanford Research Hughes Aircraft RAND Corp. Computer Institute Corp. Corporation of America Designers Raphael, Green, Savitt, Love, and Levien and Maron Marill and Coles Troop Computer PDP 10 IBM 360/65 IBM 7044 360/65 IBM 360/75 Programming Lisp 1.5 Assembly language Assembly language Assembly language language Input language Near–natural Stylized input form Stylized input forms Near–natural model language model and a procedural analyzed by language model based on simple language FOREMAN language based on matching transformations and sentence templates context-free grammar Syntactic analysis Earley algorithm N/A N/A Match of sentence technique for context-free against stored

grammar templates Semantic analysis Semantics stack N/A N/A Pattern ⇐ action technique built during syntax operation invoked analysis phase as a result of template match Intermediate First-order Binary relations Binary relations n-ary relations (n ≤ language predicate calculus 7 as implemented) Data structures Lisp-chained list Relational Files quadruplicated Statement elements structures statement elements and ordered by are hash coded and randomized (coded) statement number open statements and replicated and three elements linked to statements stored corresponding under each element closed statements Inference Formal theorem Inference rules Plausible inference Twelve general procedures proving by specified as rules specified in a rules of deductive Robinson resolution relational procedural language logic used procedure statements handled called INFEREX by breadth ﬁrst followed by depth Output language Near–natural Relational Relational Near–natural language generated statements statements language generated in a synthesis phase from n-ary relational statements

Table 1. First-Stage Deductive Database Implementations (adapted from Minker and Sable [1970]).

ECRC shareholder companies in 1990. 1986), semantics for stratified negation and set Implementation efforts at MCC Corpora- grouping (Beeri et al. 1991), investigation of tion on a DDB started in 1984 and emphasized safety, the finiteness of answer sets, and join- bottom-up evaluation methods (Tsur and Zan- order optimization. The LDL system was imple- iolo 1986) and query evaluation using such mented in 1988 and released in the period methods as seminaive evaluation, magic sets 1989 to 1991. It was among the first widely and counting (Beeri and Ramakrishnan 1991; available DDBs and was distributed to univer- Bancilhon et al. 1986; Sacca and Zaniolo sities and shareholder companies of MCC.

FALL 1997 33 Articles

Name MRPPS 3.0 (Minker 1978) DADM (Kellogg, Klahr, DEDUCE 2 (Chang 1978) (Maryland refutation and Travis 1978) proof procedure system) (Deductively augmented data management) Organization University of Maryland System Development IBM San Jose Corp. Designers Minker, McSkimin, Kellogg, Klahr, and Travis Chang Wilson, and Aronson Computer UNIVAC 1108 N/A N/A Programming language SIMPL N/A N/A Input language model Multisorted, well-formed Primitive conditional DEDUCE (Chang 1976) formulas statements and natural (based on symbolic logic) language Intermediate language Clausal form Primitive conditional DEDUCE statements Data structures Semantic networks, Predicate array, premise Connection graph knowledge base index array, semantic network, predicate connection graph (Sickel 1976; Kowalski 1975) Inference procedures SL resolution (Kowalski Connection graph Connection graph and Kuehner 1971) and Lush resolution (Hill 1974) Output language Natural language voice Primitive condition DEDUCE and English (Powell 1977) statements Features Semantic query Semantic query Semantic query optimization, multisorted optimization, multisorted optimization variables, no recursion, variables, no recursion non-Horn clauses, clauses not necessarily function free, relations not in first normal form

Table 2. Second-Stage Deductive Database Implementations. SL = Linear resolution with Selection function. LUSH = Linear resolution with Unrestricted Selection function for Horn clauses.

Implementation efforts at the University of optimization by selecting from among alterna- Wisconsin on the CORAL DDBs started in the tive control choices. CORAL provides imperative 1980s. Bottom-up and magic set methods were constructs such as update, insert, and delete implemented. The system, written in C and rules. Disk-resident data are supported using C++, is extensible and provides aggregation for the EXODUS storage manager, which also pro- modularly stratified databases. CORAL supports vides transaction management in a client-serv- a declarative language and an interface to C++ er environment. that allows for a combination of declarative Implementation at Stanford started in 1985 and imperative programming. The declarative on NAIL! (Not Another Implementation of query language supports general Horn clauses Logic!). The effort led to the first paper on augmented with complex terms, set grouping, recursion using the magic set method (Bancil- aggregation, negation, and relations with hon et al. 1986). Other contributions were tuples that contain universally quantified vari- aggregation in logical rules and theoretical ables. CORAL supports many evaluation strate- contributions to negation: stratified negation gies and automatically chooses an efficient by Van Gelder (1988); well-founded negation evaluation strategy. Users can guide query by Van Gelder, Ross, and Schlipf (1991); and

34 AI MAGAZINE Articles

Name Developed Recursion Negation Aggregation Update Integrity Optimization Storage Interfaces constraints ADITI U. General Stratified Stratified No No Magic sets, Extensio Prolog (Vaghani Melbourne seminaive nal et al. database, 1991) intensional database COL INRIA ? Stratified Stratified No No None Main Machine (Abitebo memory learning ul and Grumbac k 1991) CONCEPT U. Aachen General Locally No Yes No Magic sets, EDB only C, Prolog BASE Stratified seminaive (Jeusfeld and Staudt 1993) CORAL U. General Modular Modular No No Magic sets, Extensio C, C++, (Ramakri Wisconsin Stratified Stratified seminaive, nal extensible shnan, context database, Srivastav factoring, inten- a, and projection sional Sudarsha pushing database n 1992) EKS-V1 ECRC General Stratified General Yes Yes Query- Exten- Persistent (Vieille subquery, sional Prolog et al. left-right database, 1992) linear intensional database DECLARE MAD General Locally General No No Magic sets, EDB only C, (Kiesslin Intelligent Stratified seminaive, Common g and Systems projection Lisp Schmidt pushing 1994) LDL, MCC General Stratified Stratified Yes No Magic sets, LDL++, seminaive, SALAD left-right (Chimen linear, ti et al. projection 1990) pushing Extensio C, C++, SQL nal database only LOGRES Polytech. of Linear Inflationa Stratified Yes Yes Seminaive, Exten- INFORMIX (Cacace Milan ry algebraic X sional et al. semantics forms database, 1990) intensional database NAIL- Stanford U. General Well Glue only Glue No Magic sets, EDB only None GLUE founded only seminaive, (Morishit right linear a, Derr, and Phipps 1993) STARBURS IBM General Stratified Stratified No No Magic sets, Exten- Extensible T Almaden seminaive sional (Mumick (variant) database, et al. inten- 1990) sional database

Table 3. Existing Implementations of Deductive Databases (adapted from Ramakrishnan and Ullman [1995]).

FALL 1997 35 Articles

modularly stratified negation by Ross (1990). A VALIDITY is now being further developed language called GLUE (Morishita, Derr, and and marketed by Next Century Media, Phipps 1993), developed for logical rules, has Inc., a California corporation in which the power of SQL statements as well as a con- Groupe Bull has some equity interests. Its ventional language for the construction of principal office [is] in the San Francisco loops, procedures, and modules. area. Implementations of DDBs in the first, sec- The VALIDITY DOOD software platform is ond, and third stages of their development currently mainly used to develop NCM’s have demonstrated the feasibility and practi- products in electronic media for interac- cality of the technology. Tools and techniques tive media applications. Two of these have been developed to produce efficient products enable marketers to target their DDBs. advertising messages to household clus- ters, to individual households, and to spe- Prospects for Commercial Implemen- cific consumers, based on the user’s tation of Disjunctive Databases1 expressed and implied interests and pref- One might address why after 20 years of theo- erences, and to convert the data coming retical research in DDBs, no commercial sys- from the user into a database of ongoing tems exist. To place this statement in perspec- and useful information about these cus- … we should tive, it is well to recall that approximately 12 tomers. A third product enables marketers years passed before relational systems were to measure the effectiveness of their not abandon available commercially. As Ullman has stated media plan and expenditures in a timely all research on a number of occasions, DDB theory is more manner, based on a full census of the subtle than relational database theory. Howev- entire audience, rather than on samples on theories of er, many prototypes have been developed start- which are fraught with inherent biases negation and ing from the 1960s, as described in the previous and errors. alternative subsection. However, none of the systems in No commercial systems exist for several rea- Ramakrishnan and Ullman (1995) are likely to sons. First, most prototypes were developed at semantics, become commercial products, with, possibly, universities. Without commercial backing for but we must two exceptions: (1) ADITI (Ramamohanarao the venture, universities are not positioned to 1993) and (2) VALIDITY (Friesen et al. 1995; Ling, either develop or support maintenance take stock of Mendelzon, and Vieille 1995), developed at the required for large system developments. Sys- what we have Bull Corporation. According to a personal com- tems developed in research organizations con- munication with Ramamohanarao, leader of trolled by consortia (ECRC and MCC) have not accomplished the ADITI effort, that I had in May 1996, they are had full backing of consortia members. Sec- and make perhaps one year from having a commercial ond, implementation efforts to develop a com- it more product. In a communication that I received mercial product were vastly underestimated. A from him on 20 February 1997, he stated: “We large investment must be made to develop a accessible have now completed most of the difficult parts DDB that both competes and extends relation- for users. of the low-level implementation of ADITI. I am al technology. According to industry stan- very hopeful of getting the beta release of the dards, an investment on the order of $30 to system by December 1997. The task was much $50 million is required to develop and place a harder and time consuming than I have ever database system in the market, no matter what anticipated.” technology it relies on. Furthermore, research- Whether it becomes a product remains to be ers tend to change their interests rather than seen. I believe it will depend on moving the consolidate their work and invest in technolo- system from a university setting to industry. gy transfer toward industry. Third, until Implementers and application specialists, recently, no convincing demonstration has rather than university researchers, are required. been made of a large commercial problem that At the Bull Corporation, Nicolas and Vieille requires a DDB, which might be why the MCC have headed an effort to develop the VALIDITY and ECRC developments were terminated. DDB system that integrates object-oriented fea- However, now, a large number of applications tures. In Minker (1996), I reported that the could take advantage of this technology, as evi- VALIDITY DDB effort had been ongoing for about denced by the book by Ramakrishnan (1995) four years. It appeared to be entering the mar- and the applications being performed by the ketplace and was being moved from the Bull VALIDITY deductive object-oriented database Corporation to a new company that will be (DOOD) system. In addition, Levy et al. (1995) responsible for its maintenance, marketing, studied the problem of computing answers to and improvements. In a personal communica- queries using materialized views and note that tion with Nicolas on 7 March 1997, he stated: this work is related to applications such as

36 AI MAGAZINE Articles global information systems, mobile comput- nated. In other systems, equalities and other ing, view adaptation, and the maintenance of arithmetic constraints are being added to opti- physical data independence. Levy et al. (1996) mize search. I believe it will not be long before describe how DDBs can be used to provide uni- join elimination is introduced into relational form access to a heterogeneous collection of technology. One can now estimate when it will more than 100 information sources on the be useful to eliminate a join (Godfrey, Gryz, World Wide Web. Fourth, apparently no uni- and Minker 1996). The tools and techniques versity researchers have tried to obtain venture already exist, and it is merely a matter of time capital to build a product outside the universi- before users and system implementers have ty. Efforts by some from MCC to obtain ven- them as part of their database systems. ture capital did not succeed. Another technology available for commer- Does lack of a commercial system at this cial use is cooperative databases. The tools and date forebode the end of logic and databases? techniques exist, as evidenced by COBASE (Chu, I believe that such a view is naive. First, there Chen, and Merzbacher 1994) and CARMIN still is a strong prospect, as noted previously, of (Gaasterland et al. 1992). With the introduc- commercial DDBs. Second, considering that it tion of recursion and SQO techniques into took 12 years before relational database tech- relational database technology, it will be neces- nology entered the marketplace, there is no sary to provide users with cooperative respon- need for alarm. Third, as the following devel- ses so they understand why certain queries fail opments portend, relational databases are or succeed. It will also permit queries to be starting to incorporate techniques stemming relaxed when the original query fails, permit- from research in DDBs. ting reasonable, if not logically correct, Indeed, many of the techniques introduced answers to be provided to users. Because user within DDBs are finding their way into rela- constraints can be handled in the same way tional technology. The new SQL standards for that integrity constraints are handled, we will relational databases are beginning to adopt see relational systems that incorporate the many of the powerful features of DDBs. In the needs of individual users into a query, as rep- SQL-2 standards (also known as SQL-92) (Melton resented by their constraints. and Simon 1993), a general class of integrity Two significant developments have taken constraints called asserts allow for arbitrary place in the implementation of commercial relationships between tables and views to be DDBs. First is the incorporation of techniques declared. These constraints exist as separate developed in DDDBs into relational technolo- statements in the database and are not gy. Recursive views that use the magic set tech- attached to a particular table or view. This nique for implementation are being permitted, extension is powerful enough to express the and methods developed for SQO are being types of integrity constraint generally associat- applied. Second is the development of a ed with DDBs. However, only the full SQL-2 DOOD, VALIDITY, that is in commercial use as standard includes assert specifications. The well as the development of the ADITI DDB that intermediate SQL-2 standard, the basis for most is scheduled to undergo beta testing in Decem- current commercial implementations, does ber 1997. It remains to be seen how long one not include asserts. The relational language for can make patches to relational technology to the next-generation SQL, SQL3, currently pro- simulate the capabilities of DDB systems. vides an operation called recursive union that supports recursive processing of tables (Melton Emerging Areas and Trends 1996). As noted in Melton (1996): “The use of the recursive union operator allows both linear In the previous sections, we discussed many (single-parent, or tree) recursion and nonlinear theories and semantics for negation in both (multiparent, or general directed graph) recur- extended DDBs and DDBs. We understand a sion. This solution will allow easy solutions to great deal about negation, except for how and bill-of-material and similar applications.” when to use a given theory, which will be an Linear recursion is currently a part of the area of confusion when users want to apply client server of IBM’s DB2 system. IBM is using the work. Much more work is needed if the the magic set method to perform linear recur- areas of implementation and application are to sion. Also, indications are that the ORACLE data- catch up with the intellectual developments base system will support some form of recursion. achieved over the past 20 years. The ﬁeld is sat- A further development is that SQO is begin- urated with alternative theories of semantics, ning to be incorporated into relational databas- and work is needed on more fertile topics. es. In DB2, cases are recognized when only one Unless we do so, funding for logic and data- answer is to be found, and the search is termi- bases will wane, as I believe it has in the United

FALL 1997 37 Articles

States. However, we should not abandon all (1995) noted the need for declarative seman- research on theories of negation and alterna- tics of triggers. He has developed a unified tive semantics, but we must take stock of what semantics for active DDBs and has shown how we have accomplished and make it more acces- active database rules relate to transaction-con- sible for users. scious stable model semantics. Baral and Lobo The role of logic will be of increasing impor- (1996) proposed a first step toward characteriz- tance because of the need to handle highly ing active databases. A clear semantics, sound complex data (partly as a result of the advances implementations, and a better understanding in networking technology and the reduction of complexity issues are required in active in the cost of both processing time and prima- databases. Work in the situation calculus and ry, secondary, and tertiary memory). These datalog extensions apply here. data will require more complex models of data Data mining and inductive inference deal access and representation. Advances will with finding generalizations extracted from a require formal models of logic rather than ad database or a logic program. Generalizations hoc solutions. Below, I briefly mention some can be integrity constraints that must be true fertile areas for further exploration. This listing with respect to the database or generalizations of important areas to investigate is not intend- that might be true of the current state but ed to be exhaustive. might change if there are updates. Database Temporal databases, which deal with time, administrators will need to determine which are important for historical, real-time data- generalizations are integrity constraints and The bases and other aspects of databases. Work in which apply only to the current state. SQO can role of this area has been done by Snodgrass (1987), handle either case and inform the user which Chomicki (1995), and Sistla and Wolfson constraint might be applicable to a query. As logic (1995). A paper on applying transition rules to demonstrated in Muggleton and De Raedt will such databases for integrity constraint check- (1994), logic programming can be used to form ing appears in Martin and Sistac (1996). inductive inferences, and Knowledge Discovery be of Transactions and updates need further in Databases (Piatetsky-Shapiro and Frawley increasing attention. Semantic models of updates exist 1991) covers work on knowledge data mining. importance (Fernández, Grant, and Minker 1996) that Laurent and Vrain (1996) discuss how to cou- assure views and data are updated correctly. ple DDBs and inductive logic programming to because Transactions that require sequences of learn query rules for optimizing databases with of the updates, long-duration transaction models, update rules. and work-flow management are areas that Integrating production systems with DDBs need to require work. In emerging applications of data- is needed to develop a formal approach to inte- handle base systems, transactions are viewed as grate and develop the semantics of rule-based sequences of nested, and most probably inter- systems. See Minker (1996) for references. highly active, subtransactions that can sparsely occur Logical foundations of DOODs is needed. complex over long periods of time. In this scenario, new No formal definition exists that covers all data. complex transaction systems must be aspects of object-oriented databases. Efforts designed. Logic-based transaction systems will have been undertaken by Kifer and his cowork- be essential to assure that an appropriate and ers (Kifer, Lausen, and Wu 1995) to develop a correct transaction is achieved. See Minker formal foundation for object-oriented data- (1996) for references. bases. Work is required to develop techniques, Active databases consist of data that protect a formal theory of updating, and all tools and their own integrity and describe the database techniques for DDBs. semantics. They are represented by the formal- Description logics restrict knowledge repre- ism event-condition-action (ECA) (Xerox sentation so that deduction is tractable but suf- Technologies 1989) and denote that whenever ficiently powerful to represent knowledge nat- an event E occurs, if condition C holds, then urally. See Minker (1996) for references to trigger action A. It has a behavior of its own systems that incorporate description logics. In beyond passively accepting statements from DDBs, representational power is also limited to users or applications. On recognition of certain allow for more tractable deduction. Some of events, it invokes commands and monitors these limits are a restriction to Horn clauses, and manages itself. It can invoke external no logical terms, no existential quantification, actions that interact with systems outside the and so forth. Research in DDBs has sought to database and can activate a potentially infinite extend the representational power yet preserve set of triggers. Although declarative con- tractability to the greatest extent possible: For straints are provided, the ECA formalism is example, DDDBs allow for general clauses, and essentially procedural in nature. Zaniolo the addition of null values allows for a type of

38 AI MAGAZINE Articles existential quantification. DDBs and descrip- and constraint-intensive queries are required tion logics have remained distinct, but their in many advanced applications. Constraints goals are similar. can capture spatial and temporal behavior that Heterogeneous databases integrate multiple is not possible in existing databases. Relation- databases into one system that do not neces- ships between these areas need to be explored sarily share the same data models. There is the further and applied to DDBs. Spatial databases need for a common logic-based language for defined in terms of polynomial inequalities are mediation and a formal semantics of such investigated by Kuipers et al. (1996), who con- databases. Work on HERMES by Subrahmanian sider termination properties of datalog pro- et al. (1994), on TSIMMIS by Chawathe et al. grams. (1994), and by Ramakrishnan (Miller, Ioanni- Abductive reasoning is the process of find- dis, and Ramakrishnan 1994) and his col- ing explanations for observations in a given leagues illustrate the efforts in this area. Het- theory. Given a set of sentences T (a theory) erogeneous databases are also needed to and a sentence G (an observation), the abduc- handle textual data. Kero and Tsur (1996) tive task can be characterized as the problem of describe the ᑣᑫ system that uses a DDB finding a set of sentences (abductive explana- ᑦᑞᑦ++ to reason about textual information. tion for G) such that T ʜ ∆ |= G, T ʜ ∆ is con- Language extensions for the semantic integra- sistent, and ∆ is minimal with respect to set tion of DDBs is proposed by Asirelli, Renso, inclusion (Kakas, Kowalski, and Toni 1993). and Turini (1996). The language allows media- A comprehensive survey and critical tors to be constructed, using a set of operators overview of the extension of logic program- for composing programs and message-passing ming to the performance of abductive reason- features. ing (abductive logic programming) is given in Multimedia databases (Subrahmanian and Kakas, Kowalski, and Toni (1993). They outline Abductive Jajodia 1995) is an emerging area for which the framework of abduction and its applica- reasoning new data models are needed. These databases tions to default reasoning and introduce an have special problems, such as the manipula- augmentation-theoretic approach to the use of is the tion of geographic databases; picture retrieval abduction as an interpretation for negation as process of where a concept orthogonal to time can failure. They show that abduction has strong finding appear in the database: space; and video data- links to extended disjunctive logic program- bases, where space and time are combined. ming. Abduction is shown to generalize nega- explanations Temporal and spatial reasoning are needed. tion as failure to include not only negative but for Logic will play a major role in the develop- also positive hypotheses and to include gener- ment of query languages for these new data al integrity constraints. They show that abduc- observations models and will permit deductive reasoning, tive logic programming is related to the justifi- in a given and a formal semantics will provide a firm the- cation-based truth maintenance system of oretical basis for them. Doyle (1979). Inoue and Sakama (1996) devel- theory. Combining databases relates both to hetero- oped a fixpoint semantics for abductive logic geneous and multimedia systems. Here, one is programs in which the belief models are char- trying to combine databases that share the acterized as the fixpoint of a disjunctive pro- same integrity constraints and schema. Such gram obtained by a suitable program transfor- databases arise in distributed system work and mation. For a summary of complexity results the combining of knowledge bases. In addition on abductive reasoning and nonmonotonic to handling problems that arise because the reasoning, see Cadoli and Schaerf (1993). combined databases might be inconsistent, one High-level robotics is an area of active has to handle priorities that can exist among research in which logic plays a significant role. individual facts. A formal treatment and refer- Knowledge bases are used to solve problems in ences appear in Pradhan and Minker (1995). cognition required to plan actions for robots Integrity constraints, SQO, and constraint and deal with multiple agents in complicated logic programming (CLP) are related topics. environments. Work in deductive and disjunc- SQO uses constraints in the form of integrity tive databases relates to this problem. In some constraints to prune the search space. These instances, a robot can have several options integrity constraints introduce equalities, that can be represented by disjunctions. Addi- inequalities, and relations into a query to help tional information derived from alternative optimize search (Chakravarthy, Grant, and information sources such as sensors can serve Minker 1990). CLP introduces domain con- to disambiguate the possibilities. Universities straints. These constraints might be equalities engaged in this research are the University of or inequalities and might even be relations Toronto (Lesperance et al. 1994), the Universi- (Jaffar and Maher 1994). Constraint databases ty of Texas at El Paso (Baral, Gelfond, and

FALL 1997 39 Articles

Provetti 1996), the University of Texas at many new areas that will be important in the Austin (Gelfond and Lifschitz 1993), and the near- and long-term future. It is clear that the University of Linkoping (Sandewall 1994). field of logic and databases had a significant Kowalski and Sadri (1996) discuss a unified prehistory before 1970 and a well-defined area agent architecture that combines rationality of research, complete with past achievements with reactivity that relates to this topic. and continued future areas of work. Applications of DDB techniques will be In the past 20 years, we have seen logic and important. Greater emphasis is required to databases progress from a fledgling field to a apply DDB technology to realistic problems. fully developed, mature field. The new areas DDBs have been shown to be important for that I cited that need further investigation both relational and deductive systems on such show that we have not nearly exhausted the Logic and topics as SQO (Chakravarthy, Grant, and work in this field. I envision that many more databases Minker 1990), cooperative answering (Gaaster- workshops will be held on this topic. Logic and have helped land, Godfrey, and Minker 1992), and global databases have helped the field of databases be information systems and mobile computing a scientific endeavor rather than an ad hoc col- the field (Levy et al. 1995). lection of techniques. We understand what of databases Commercial implementations of DDBs and constitutes a database, a query, and an answer DOODs are needed. The deductive model of to a query and where knowledge has its place. be a databases is beginning to take hold, evidenced I look forward to the next 20 years of this field. scientific by the textbook by Abiteboul, Hull, and Vianu I hope that I will have an opportunity, then, to (1995). The prospect that the ADITI (Ramamo- look back and see a field that has accomplished endeavor hanarao 1993) system might be available in a much and is still vibrant. To remain vibrant, rather than year and the applications for which the DDB we will have to take on some of the new chal- an ad hoc VALIDITY are being used indicate that progress is lenges rather than be mired in the semantics of being made in developing a commercial DDB. more exotic databases. We will have to address collection The merger of object-oriented and deductive implementation issues, and we will have to be of formalisms is taking place, as illustrated by the able to provide guidance to practitioners who proceedings of the DOOD conference series will need to use the significant developments techniques. (Ling, Mendelzon, and Vieille 1995; Ceri, in logic and databases. Tanaka, and Tsur 1993; Delobel, Kifer, and Masunaga 1991; Kim, Nicolas, and Nishio Acknowledgments We 1989). That the VALIDITY (Friesen, Lefebvre, and This article is a condensation, and in some cas- understand Vielle 1996) system is currently in use by cus- es an expansion, of an invited keynote address what tomers indicates that the object-oriented and presented at the Workshop on Logic in Data- deductive formalisms are soon to be available bases in San Miniato, Italy, in 1996. Those constitutes a commercially. Additional features will be interested in the longer version of the article database, required for commercial systems such as coop- should refer to Minker (1996). A number of my erative answering (Gaasterland, Godfrey, and colleagues contributed their thoughts on what a query, Minker 1992a) and arithmetic and aggregate they considered to be the significant develop- and an operators (Dobbie and Topor 1996). ments in the field, including Robert Demo- It is clear from this discussion that logic and lombe, Hervé Gallaire, Georg Gottlob, John answer databases can contribute significantly to a Grant, Larry Henschen, Bob Kowalski, Jean- to a query large number of exciting new topics. Hence, Marie Nicolas, Raghu Ramakrishnan, Kotagiri and where the field of logic and databases will continue to Ramamohanarao, Ray Reiter, and Carlo Zanio- be a productive area of research and imple- lo. Many of my former and current students knowledge mentation. also contributed thoughts, including Sergio has its Alvarez, Chitta Baral, Jose Alberto Fernández, Summary Terry Gaasterland, Parke Godfrey, Jarek Gryz, place. Jorge Lobo, Sean Luke, and Carolina Ruiz. I discussed major accomplishments that have Although many of the views reflected in the taken place in logic and databases during the article might be shared by those who made 20 years since 1976. Among these accomplish- suggestions, I take full responsibility for them. ments are the extension of relational data- The full paper is dedicated in honor of Hervé bases, the development of the semantics and Gallaire and Jean-Marie Nicolas with whom I complexity of these alternative databases, the have worked as co-author, co-editor, colleague, ability to permit knowledge base systems to be and friend and who have helped to make the represented and developed, and the use of log- field of deductive databases a reality. Support ic programming and DDBs to implement non- for this article was received from the National monotonic reasoning systems. I discussed Science Foundation under grant IRI 9300691.

40 AI MAGAZINE Articles

Note Beeri, C.; Naqvi, S.; Shmueli, O.; and Tsur, S. 1991. Set Constructors in a Logic Database Language. Jour- 1. This subsection reflects comments made at the nal of Logic Programming 10(3–4): 181–253. Workshop on Logic in Databases in San Miniato, Italy, 1 to 2 July, 1996, in the panel session, Deduc- Bell, C.; Nerode, A.; Ng, R.; and Subrahmanian, V. 1993. Implementing Stable Model Semantics by Lin- tive Databases: Challenges, Opportunities, and ear Programming. In Proceedings of the 1993 Inter- Future Directions, by Arno Siebes, Shalom Tsur, Jeff national Workshop on Logic Programming and Ullman, Laurent Vieille, and Carlo Zaniolo, and in a Non-Monotonic Reasoning, June, Lisbon, Portugal. personal communication by Jean-Marie Nicolas. I Ben-Eliyahu, R., and Dichter, R. 1994. Propositional am wholly responsible for the views expressed in Semantics for Disjunctive Logic Programs. Annals of this subsection. Mathematics and Artiﬁcial Intelligence 12:53–87. References Ben-Eliyahu, R., and Palopoli, L. 1994. Reasoning with Minimal Models: Efficient Algorithms and Abiteboul, S., and Grumback, S. 1991. A Rule-Based Applications. In Proceedings of the Fourth International Language with Functions and Sets. ACM Transactions Conference on the Principles of Knowledge Representa- on Database Systems 16(1): 1–30. tion and Reasoning, 39–50. San Francisco, Calif.: Mor- Abiteboul, S.; Hull, R.; and Vianu, V. 1995. Founda- gan Kaufmann. tions of Databases. Reading, Mass.: Addison-Wesley. Ben-Eliyahu, R.; Palopoli, L.; and Zemlyanker, V. Apt, K., and Bol, R. 1994. Logic Programming and 1996. The Expressive Power of Tractable Disjunction. Negation: A Survey. Journal of Logic Programming Paper presented at the Twelfth European Conference 19–20: 9–71. on Artificial Intelligence (ECAI96), 12–16 August, Budapest, Hungary. Apt, K.; Blair, H.; and Walker, A. 1988. Towards a Theory of Declarative Knowledge. In Foundations of Blaustein, B. 1981. Enforcing Database Assertions: Deductive Databases and Logic Programming, ed. J. Techniques and Applications. Ph.D. thesis, Comput- er Science Department, Harvard University. Minker, 89–148. San Francisco, Calif.: Morgan Kauf- mann. Brass, S., and Dix, J. 1995. A General Approach to Bottom-Up Computation of Disjunctive Semantics. Asirelli, P.; Renso, C.; and Turini, F. 1996. Language In Nonmonotonic Extensions of Logic Programming, eds. Extensions for Semantic Integration of Deductive J. Dix, L. Pereira, and T. Przymusinski, 127–155. Lec- Databases. In Logic in Databases (LID’96), eds. D. ture Notes in Computer Science 927. New York: Pedreschi and C. Zaniolo, 415–434. New York: Springer-Verlag. Springer-Verlag. Bry, F. 1990. Query Evaluation in Recursive Databas- Bancilhon, F.; Maier, D.; Sagiv, Y.; and Ullman, J. es: Bottom-Up and Top-Down Reconciled. Data and 1986. Magic Sets and Other Strange Ways to Imple- Knowledge Engineering 5:289–312. ment Logic Programs. In Proceedings of the ACM Cacace, F.; Ceri, S.; Crespi-Reghizzi, S.; and Zicari, R. Symposium on Principles of Database Systems, 1–15. 1990. Integrating Object-Oriented Data Modeling New York: Association of Computing Machinery. with a Rule-Based Programming Paradigm. In Pro- Baral, C., and Gelfond, M. 1994. Logic Programming ceedings of the ACM SIGMOD Conference on Man- and Knowledge Representation. Journal of Logic Pro- agement of Data, 225–236. New York: Association of gramming 19–20: 73–148. Computing Machinery. Baral, C., and Lobo, J. 1996. Formal Characterization Cadoli, M. 1996. Panel on “Knowledge Compilation of Active Databases. In Logic in Databases (LID’96), and Approximation”: Terminology, Questions, Ref- eds. D. Pedreschi and C. Zaniolo, 195–215. New erences. In Proceedings of the Fourth International York: Springer-Verlag. Symposium on Artiﬁcial Intelligence and Mathemat- ics (AI/Math-96), 183–186. Available from Martin Baral, C.; Gelfond, M.; and Provetti, A. 1997. Repre- Columbic, Bar-Ilan University, Department of Com- senting Actions: Laws, Observations, and Hypothe- puter Science, Tel Aviv, Israel. sis. Journal of Logic Programming 31(1–3): 201–243. Cadoli, M. 1993. Semantical and Computational Baral, C.; Lobo, J.; and Minker, J. 1990a. Generalized Aspects of Horn Approximations. In Proceedings of Disjunctive Well-Founded Semantics for Logic Pro- the Thirteenth International Joint Conference on grams: Declarative Semantics. In Proceedings of the Artificial Intelligence, 39–44. Menlo Park, Calif.: Fifth International Symposium on Methodologies for International Joint Conferences on Artificial Intelligent Systems, 465–473. New York: North Hol- Intelligence. land. Cadoli, M., and Lenzerini, M. 1994. The Complexity Baral, C.; Lobo, J.; and Minker, J. 1990b. Generalized of Closed-World Reasoning and Circumscription. Disjunctive Well-Founded Semantics for Logic Pro- Journal of Computer and Systems Science 43:165–211. grams: Procedural Semantics. In Proceedings of the Cadoli, M., and Schaerf, M. 1993. A Survey of Com- Fifth International Symposium on Methodologies for plexity Results for Non-Monotonic Logics. Journal of Intelligent Systems, 456–464. New York: North Hol- Logic Programming 13:127–160. land. Ceri, S.; Tanaka, K.; and Tsur, S., eds. 1993. Proceed- Beeri, C., and Ramakrishnan, R. 1991. On the Power ings of the Third International Conference on Deductive of Magic. Journal of Logic Programming 10(3–4): and Object-Oriented Databases—DOOD’93. Heidel- 255–300. berg: Springer-Verlag.

FALL 1997 41 Articles

Chakravarthy, U. S.; Grant, J.; and Minker, J. 1990. Decker, H. 1986. Integrity Enforcement on Deduc- Logic-Based Approach to Semantic Query Optimiza- tive Databases. In Proceedings of the First International tion. ACM Transactions on Database Systems 15(2): Conference on Expert Database Systems, 381–395. Men- 162–207. lo Park, Calif.: Benjamin Cummings. Chang, C. 1981. On Evaluation of Queries Contain- Delobel, C.; Kifer, M.; and Masunaga, Y., eds. 1991. ing Derived Relations. In Advances in Database Theo- Proceedings of the Second International Conference on ry, Volume 1, ed. H. G. J. M. J.-M. Nicolas, 235–260. Deductive and Object-Oriented Databases (DOOD’91). New York: Plenum. Heidelberg: Springer-Verlag. Chang, C. 1978. DEDUCE 2: Further Investigations of del Val, A. 1995. An Analysis of Approximate Knowl- Deduction in Relational Databases. In Logic and edge Compilation. In Proceedings of the Fourteenth Databases, ed. H. G. J. Minker, 201–236. New York: International Conference on Artificial Intelligence. Plenum. Menlo Park, Calif.: International Joint Conferences on Artificial Intelligence. Chang, C. 1976. DEDUCE—A Deductive Query Lan- Demolombe, R., and Jones, A. J. I. 1996. Integrity guage for Relational Databases. In Pattern Recognition Constraints Revisited. Journal of the IGPL (Interest and Artificial Intelligence, ed. C. Chen, 108–134. San Group in Pure and Applied Logics): An Electronic Journal Diego, Calif.: Academic. on Pure and Applied Logic 4(3): 369–383. Chawathe, S.; Garcia-Molina, H.; Hammer, J.; Ire- Dix, J. 1992. A Framework for Representing and land, K.; Papakonstantinou, Y.; Ullman, J.; and Characterizing Semantics of Logic Programs. In Prin- Widom, J. 1994. The TSIMMIS Project: Integration of ciples of Knowledge Representation and Reasoning: Pro- Heterogeneous Information Sources. Paper present- ceedings of the Third International Conference (KR ‘92), ed at IPSJ Conference, October, Tokyo, Japan. eds. B. Nebel, C. Rich, and W. Swartout, 591–602. Chen, W., and Warren, D. 1993. A Goal-Oriented San Francisco, Calif.: Morgan Kaufmann. Approach to Computing Well-Founded Semantics. Dobbie, G., and Topor, R. 1996. Arithmetic and Journal of Logic Programming 17(2–4): 279–300. Aggregate Operators in Deductive Object-Oriented Chimenti, D.; Gamboa, R.; Krishnamurthy, R.; Databases. In Logic in Databases (LID’96), eds. D. Naqvi, S.; Tsur, S.; and Zaniolo, C. 1990. The LDL Pedreschi and C. Zaniolo, 399–407. New York: System Prototype. IEEE Transactions on Knowledge Springer-Verlag. and Data Engineering 2(1): 76–90. Doyle, J. 1979. Truth Maintenance System. Artificial Chomicki, J. 1995. Efficient Checking of Temporal Intelligence 12(3): 231–272. Integrity Constraints Using Bounded History Encod- Eiter, T., and Gottlob, G. 1995. On the Computation ing. ACM Transactions on Database Systems 20(2): Cost of Disjunctive Logic Programming: Proposi- 111–148. tional Case. Annals of Mathematics and Artificial Intel- Chu, W. W.; Chen, Q.; and Merzbacher, M. A. 1994. ligence 15(3–4): 289–323. COBASE: A Cooperative Database System. In Nonstan- Fernández, J., and Minker, J. 1995. Bottom-Up Com- dard Queries and Nonstandard Answers, Studies in Logic putation of Perfect Models for Disjunctive Theories. and Computation 3, eds. R. Demolombe and T. Journal of Logic Programming 25(1): 33–51. Imielinski, 41–73. Oxford, U.K.: Clarendon. Fernández, J., and Minker, J. 1992. Semantics of Dis- Clark, K. L. 1978. Negation as Failure. In Logic and junctive Deductive Databases. In Proceedings of the Data Bases, eds. H. Gallaire and J. Minker, 293–322. International Conference on Database Theory, 332–356. New York: Plenum. Berlin: Springer-Verlag. Codd, E. 1970. A Relational Model of Data for Large Fernández, J. A., and Minker, J. 1991. Bottom-Up Shared Data Banks. Communications of the ACM Evaluation of Hierarchical Disjunctive Deductive Databases. In Logic Programming Proceedings of the 13(6): 377–387. Eighth International Conference, ed. K. Furukawa, Colmerauer, A.; Kanoui, H.; Pasero, R.; and Roussel, 660–675. Cambridge, Mass.: MIT Press. P. 1973. Un Systeme de Communication Homme- Fernández, J.; Grant, J.; and Minker, J. 1996. Model- Machine en Francais (A System of Man-Machine Theoretic Approach to View Updates in Deductive Communication in French). Technical Report, Databases. Journal of Automated Reasoning 17(2): Groupe de Intelligence Artificielle Universitae de 171–197. Aix-Marseille II, Marseille. Fernández, J. A.; Lobo, J.; Minker, J.; and Subrahman- Computer Corporation. 1967. Relational Structures ian, V. 1993. Disjunctive LP + Integrity Constraints = Research. Technical Report, Computer Corporation Stable Model Semantics. Annals of Mathematics and of America, Cambridge, Massachusetts. Artificial Intelligence 8(3–4): 449–474. Date, C. 1995. An Introduction to Database Systems, Fitting, M. 1985. A Kripke-Kleene Semantics for Log- 6th ed. Reading, Mass.: Addison-Wesley. ic Programs. Journal of Logic Programming 2:295–312. Decker, H. 1991. On the Declarative, Operational, Friesen, O.; Lefebvre, A.; and Vieille, L. 1996. and Procedural Semantics of Disjunctive Computa- VALIDITY: Applications of a DOOD System. In Proceed- tional Theories. In Proceedings of the Second Inter- ings of the Fifth International Conference on Extending national Workshop on the Deductive Approach to Database Technology—EDBT’96, 131–134. Lecture Information Systems and Databases, 9–11 Septem- Notes in Computer Science 1057. New York: ber, Aiguablava, Spain. Springer-Verlag.

42 AI MAGAZINE Articles

Friesen, O.; Gauthier-Villars, G.; Lefebvre, A.; and Gottlob, G. 1994. Complexity and Expressive Power Vieille, L. 1995. Applications of Deductive Object- of Disjunctive Logic Programming (Research Oriented Databases Using DEL. In Applications of Logic Overview). In International Logic Programming Sympo- Databases, ed. R. Ramakrishnan, 1–22. New York: sium ILPS’94, ed. M. Bruynooghe, 23–42. Cambridge, Kluwer Academic. Mass.: MIT Press. Gaasterland, T., and Lobo, J. 1993. Processing Nega- Green, C., and Raphael, B. 1968a. Research in Intel- tion and Disjunction in Logic Programs through ligent Question-Answering Systems. In Proceedings Integrity Constraints. Journal of Intelligent Informa- of the ACM Twenty-Third National Conference, tion Systems 2(3): 225–243. 169–181. New York: Association of Computing Gaasterland, T.; Godfrey, P.; and Minker, J. 1992a. An Machinery. Overview of Cooperative Answering. Journal of Intel- Green, C., and Raphael, B. 1968b. The Use of Theo- ligent Information Systems 1(2): 123–157. rem-Proving Techniques in Question-Answering Sys- Gaasterland, T.; Godfrey, P.; and Minker, J. 1992b. tems. In Proceedings of the Twenty-Third ACM Relaxation as a Platform for Cooperative Answering. National Conference. New York: Association of Journal of Intelligent Information Systems 1:293–321. Computing Machinery. Gaasterland, T.; Godfrey, P.; Minker, J.; and Novik, L. Hammer, M., and Zdonik, S. 1980. Knowledge-Based 1992. A Cooperative Answering System. In Proceed- Query Processing. In Proceedings of the Sixth Inter- ings of the Logic Programming and Automated Reason- national Conference on Very Large Data Bases, ing Conference, ed. A. Voronkov, 478–480. Lecture 137–147. Washington, D.C.: IEEE Computer Society. Notes in Artificial Intelligence 624. New York: Harel, D. 1980. Review Number 36,671 of Logic and Springer-Verlag. Data Bases by H. Gallaire and J. Minker. Computing Gaasterland, T.; Godfrey, P.; Minker, J.; and Novik, L. Reviews 21(8): 367–369. 1992. A Cooperative Answering System. In Proceed- Hill, R. 1974. Lush Resolution and Its Completeness. ings of the Logic Programming and Automated Reason- Technical Report, DCL Memo 78, Department of ing Conference, ed. A. Voronkov, 478–480. Lecture Artificial Intelligence, University of Edinburgh. Notes in Artificial Intelligence 624. New York: Inoue, K., and Sakama, C. 1996. A Fixpoint Charac- Springer-Verlag. terization of Abductive Logic Programs. Journal of Gallaire, H., and Minker, J., eds. 1978. Logic and Logic Programming 27(2): 107–136. Databases. New York: Plenum. Jaffar, J., and Maher, M. 1994. Constraint Logic Pro- Gallaire, H.; Minker, J.; and Nicolas, J.-M., eds. gramming: A Survey. Journal of Logic Programming 1984a. Advances in Database Theory, Volume 2. New 19–20: 503–581. York: Plenum. Jeusfeld, M., and Staudt, M. 1993. Query Optimiza- Gallaire, H.; Minker, J.; and Nicolas, J.-M. 1984b. tion in Deductive Object Bases. In Query Processing Logic and Databases: A Deductive Approach. ACM for Advanced Database Applications, eds. G. Vossen, J. Computing Surveys 16(2): 153–185. C. Feytag, and D. Maier. San Francisco, Calif.: Mor- Gallaire, H.; Minker, J.; and Nicolas, J.-M., eds. 1981. gan Kaufmann. Advances in Database Theory, Volume 1. New York: Kakas, A. C.; Kowalski, R. A.; and Toni, F. 1993. Plenum. Abductive Logic Programming. Journal of Logic and Gelfond, M., and Lifschitz, V. 1993. Representing Computation 6(2): 719–770. Actions and Change by Logic Programs. Journal of Kautz, H., and Selman, B. 1992. Forming Concepts Logic Programming 17(2–4): 301–323. for Fast Inference. In Proceedings of the Tenth Gelfond, M., and Lifschitz, V. 1991. Classical Nega- National Conference on Artificial Intelligence, tion in Logic Programs and Disjunctive Databases. 786–793. Menlo Park, Calif.: American Association New Generation Computing 9:365–385. for Artificial Intelligence. Gelfond, M., and Lifschitz, V. 1990. Logic Programs Kellogg, C.; Klahr, P.; and Travis, L. 1978. Deductive with Classical Negation. In Proceedings of the Seventh Planning and Pathfinding for Relational Data Bases. International Conference on Logic Programming, eds. D. In Logic and Data Bases, eds. H. Gallaire and J. Warren and P. Szeredi, 579–597. Cambridge, Mass.: Minker, 179–200. New York: Plenum. MIT Press. Kero, B., and Tsur, S. 1996. The CALLIGRAPHIC IQ Sys- Gelfond, M., and Lifschitz, V. 1988. The Stable Mod- tem: A Deductive Database Information Lens for el Semantics for Logic Programming. In Proceedings Reasoning about Textual Information. In Logic in of the Fifth International Conference and Symposium on Databases (LID’96), eds. D. Pedreschi and C. Zaniolo, Logic Programming, eds. R. Kowalski and K. Bowen, 377–395. New York: Springer-Verlag. 1070–1080. Cambridge, Mass.: MIT Press. Kiessling, W., and Schmidt, H. 1994. DECLARE and SDS: Godfrey, P.; Gryz, J.; and Minker, J. 1996. Semantic Early Efforts to Commercialize Deductive Database Query Evaluation for Bottom-Up Evaluation. In Pro- Technology. VLDB Journal 3(2): 211–243. ceedings of the Ninth International Symposium on Kifer, M.; Lausen, G.; and Wu, J. 1995. Logical Foun- Methodologies for Intelligent Systems, June, dations of Object-Oriented and Frame-Based Lan- Zakopane, Poland. guages. Journal of the ACM 42(4): 741–843.

FALL 1997 43 Articles

Kim, W.; Nicolas, J.-M.; and Nishio, S., eds. 1990. Levy, A., and Sagiv, Y. 1995. Semantic Query Opti- Proceedings of the First International Conference on mization in Datalog Programs. In Principles of Data- Deductive and Object-Oriented Databases (DOOD’89). base Systems 1995 (PODS95), 163–173. New York: Amsterdam: North-Holland. Association of Computing Machinery. King, J. 1981. QUIST: A System for Semantic Query Levy, A.; Rajaraman, A.; and Ordille, J. 1996. Query- Optimization in Relational Databases. In Proceed- ing Heterogeneous Information Sources Using ings of the Seventh International Conference on Source Descriptions. In Proceedings of the Twenty-Sec- Very Large Data Bases, 510–517. Washington, D.C.: ond Very Large Data Base Conference, 251–262. San IEEE Computer Society. Francisco, Calif.: Morgan Kaufmann. Kowalski, R. 1978. Logic for Data Description. In Logic Levy, A.; Mendelzon, A.; Sagiv, Y.; and Srivastava, D. and Data Bases, ed. H. G. J. Minker, 77–102. New York: 1995. Answering Queries Using Views. In Proceed- Plenum. ings of the 14th ACM SIGACT-SIGMOD-SIGART Symposium on Principles of Database Systems Kowalski, R. 1974. Predicate Logic as a Programming (PODS-95), 95–104. New York: Association of Com- Language. In Proceedings of the International Feder- puting Machinery. ation for Information Processing (World Computer Congress) 4, 569–574. Amsterdam: North-Holland. Ling, T.-W.; Mendelzon, A.; and Vieille, L., eds. 1995. Proceedings of the Fourth International Conference on Kowalski, R., and Sadri, F. 1996. Toward a Unified Deductive and Object-Oriented Databases (DOOD’95). Agent Architecture That Combines Rationality with LNCS 1013. Heidelberg: Springer-Verlag. Reactivity. In Logic in Databases (LID’96), eds. D. Pedreschi and C. Zaniolo, 131–150. New York: Lloyd, J. W., and Topor, R. W. 1985. A Basis for Springer-Verlag. Deductive Database Systems. Journal of Logic Pro- gramming 2(2): 93–109. Kowalski, R. A. 1975. A Proof Procedure Using Con- nection Graphs. Journal of the ACM 22(4): 572–595. Lobo, J.; Minker, J.; and Rajasekar, A. 1992. Founda- tions of Disjunctive Logic Programming. Cambridge, Kowalski, R. A., and Kuehner, D. 1971. Linear Reso- Mass.: MIT Press. lution with Selection Function. Artificial Intelligence Loveland, D.; Reed, D.; and Wilson, D. 1993. SATCH- 2:227–260. MORE: SATCHMO with Relevancy. Technical report, CS- Kuhns, J. 1967. Answering Questions by Computer: 1993-06, Department of Computer Science, Duke A Logical Study. Technical Report, The Rand Corpo- University. ration, Santa Monica, California. McCarthy, J. 1980. Circumscription—A Form of Kuipers, B.; Paredaens, J.; Smits, M.; and den Buss- Non-Monotonic Reasoning. Artificial Intelligence che, J. V. 1996. Termination Properties of Spatial Dat- 13:27–39. alog Programs. In Logic in Databases (LID’96), eds. D. McDermott, D., and Doyle, J. 1980. Nonmonotonic Pedreschi and C. Zaniolo, 95–109. New York: Logic I. Artificial Intelligence 13:41–72. Springer-Verlag. McSkimin, J., and Minker, J. 1977. The Use of a Laurent, D., and Vrain, C. 1996. Learning Query Semantic Network in Deductive Question-Answer- Rules for Optimizing Databases with Update Rules. ing Systems. In Proceedings of the Fifth Internation- In Logic in Databases (LID’96), eds. D. Pedreschi and al Joint Conference on Artificial Intelligence, 50–58. C. Zaniolo, 173–192. New York: Springer-Verlag. Menlo Park, Calif.: International Joint Conferences Lefebvre, L. 1992. Towards an Efficient Evaluation of on Artificial Intelligence. Recursive Aggregates in Deductive Databases. Paper Manthey, R., and Bry, F. 1988. SATCHMO: A Theorem presented at the Fourth International Conference on Prover Implemented in PROLOG. In Proceedings of the Fifth-Generation Computer Systems (FGCS), 1–5 Ninth International Conference on Automated Deduc- June, Tokyo, Japan. tion (CADE), 415–434. Lecture Notes in Computer Leone, N., and Rullo, P. 1992. The Safe Computation Science 310. New York: Springer-Verlag. of the Well-Founded Semantics for Logic Program- Martin, C., and Sistac, J. 1996. Applying Transition ming. Information Systems 17(1): 17–31. Rules to Bitemporal Deductive Databases for Integri- Leone, N.; Rullo, P.; and Scarcello, F. 1996. Stable ty Constraint Checking. In Logic in Databases Model Checking for Disjunctive Programs. In Logic (LID’96), eds. D. Pedreschi and C. Zaniolo, 111–128. in Databases (LID’96), eds. D. Pedreschi and C. Zan- New York: Springer-Verlag. iolo, 281–294. New York: Springer-Verlag. Melton, J. 1996. An SQL3 Snapshot. In Proceedings of Lesperance, Y.; Levesque, H.; Lin, F.; Marcu, D.; Reit- the Twelfth International Conference on Data Engi- er, R.; and Scherl, R. 1994. A Logical Approach to neering, 666–672. Washington, D.C.: IEEE Comput- High-Level Robot Programming—A Progress Report. er Society. Paper presented at the 1994 AAAI Fall Symposium Melton, J., and Simon, A. R. 1993. Understanding the on Control of the Physical World by Intelligent Sys- New SQL: A Complete Guide. San Francisco, Calif.: Mor- tems, November, New Orleans. gan Kaufmann. Levien, R., and Maron, M. 1965. Relational Data File: Miller, R.; Ioannidis, Y.; and Ramakrishnan, R. 1994. A Tool for Mechanized Inference Execution and Data Translation and Integration of Heterogeneous Retrieval. The Rand Corporation, Santa Monica, Cal- Schemas: Bridging the Gap between Theory and ifornia. Practice. Information Systems 19(1): 3–31.

44 AI MAGAZINE Articles

Minker, J. 1996. Logic and Databases: A 20-Year Ret- Naughton, J., and Sagiv, Y. 1987. A Decidable Class rospective. In Logic in Databases, eds. D. Pedreschi of Bounded Recursions. In Proceedings of the Sixth and C. Zaniolo, 3–57. New York: Springer-Verlag. ACM SIGACT-SIGMOD-SIGART Symposium on Minker, J. 1993. An Overview of Nonmonotonic Principles of Database Systems, 227–236. New York: Reasoning and Logic Programming. Journal of Logic Association of Computing Machinery. Programming 17(2–4): 95–126. Ng, R., and Subrahmanian, V. 1993. Probabilistic Minker, J. 1989. Toward a Foundation of Disjunctive Logic Programming. Information and Computation Logic Programming. In Proceedings of the North Amer- 101(2): 150–201. ican Conference on Logic Programming, 121–125. Cam- Nicolas, J.-M. 1979. Logic for Improving Integrity bridge, Mass.: MIT Press. Checking in Relational Databases. Acta Informatica Minker, J., ed. 1988a. Foundations of Deductive Data- 18(3): 227–253. bases and Logic Programming. San Francisco, Calif.: Nicolas, J.-M., and Syre, J. C. 1974. Natural Question- Morgan Kaufmann. Answering and Automatic Deduction in System Syn- Minker, J. 1988b. Perspectives in Deductive Data- tex. In Proceedings of the IFIP Congress 1974, 595–599. bases. Journal of Logic Programming 5:33–60. Berlin: Springer-Verlag. Pearce, P., and Wagner, G. 1989. Logic Programming Minker, J. ed. 1986. Proceedings of the Workshop on with Strong Negation. In Proceedings of the Interna- Foundations of Deductive Databases and Logic Pro- tional Workshop on Extensions of Logic Programming, gramming. College Park, Md.: University of Mary- ed. P. Schroeder-Heister, 311–326. Lecture Notes in land Institute for Advanced Computer Studies. Artificial Intelligence. New York: Springer-Verlag. Minker, J. 1982. On Indefinite Databases and the Piatetsky-Shapiro, G., and Frawley, W. J., eds. 1991. Closed-World Assumption. In Lecture Notes in Com- Knowledge Discovery in Databases. Menlo Park, Calif.: puter Science 138, 292–308. New York: Springer Ver- AAAI Press. lag. Powell, P. 1977. Answer-Reason Extraction in a Par- Minker, J. 1978. Search Strategy and Selection Func- allel Relational Data Base System. Master’s thesis, tion for an Inferential Relational System. Transac- Department of Computer Science, University of tions on Data Base Systems 3(1): 1–31. Maryland. Minker, J., and Nicolas, J.-M. 1982. On Recursive Pradhan, S., and Minker, J. 1995. Combining Data- Axioms in Deductive Databases. Information Systems log Databases Using Priorities. Journal of Intelligent 7(4): 1–15. Cooperative Information Systems 4(3): 231–260. Minker, J., and Rajasekar, A. 1990. A Fixpoint Seman- Przymusinski, T. 1995. Static Semantics for Normal tics for Disjunctive Logic Programs. Journal of Logic and Disjunctive Logic Programs. Annals of Mathe- Programming 9(1): 45–74. matics and Artificial Intelligence 14:323–357. Minker, J., and Ruiz, C. 1996. Mixing a Default Rule Przymusinski, T. 1990. Stationary Semantics for Dis- with Stable Negation. In Proceedings of the Fourth junctive Logic Programs and Deductive Databases. International Symposium on Artificial Intelligence In Proceedings of the North American Conference on and Mathematics, 122–125. Available from Martin Logic Programming, eds. S. Debray and M. Columbic, Bar-Ilan University, Department of Com- Hermenegildo, 40–62. Cambridge, Mass.: MIT Press. puter Science, Tel Aviv, Israel. Przymusinski, T. C. 1991. Stable Semantics for Dis- Minker, J., and Ruiz, C. 1994. Semantics for Disjunc- junctive Programs. New Generation Computing tive Logic Programs with Explicit and Default Nega- 9:401–424. tion. Fundamenta Informaticae 20(3–4): 145–192. Przymusinski, T. C. 1990. Extended Stable Semantics Minker, J., and Sable, J. 1970. Relational Data System for Normal and Disjunctive Programs. In Proceedings Study. Technical Report RADC-TR-70-180, Rome Air of the Seventh International Logic Programming Confer- Development Center, Air Force Systems Command, ence, eds. D. Warren and P. Szeredi, 459–477. Cam- Griffiss Air Force Base, New York. bridge, Mass.: MIT Press. Moore, R. 1985. Semantical Considerations on Non- Przymusinski, T. C. 1988. On the Declarative Seman- monotonic Logic. Artificial Intelligence 25:75–94. tics of Deductive Databases and Logic Programming. Morishita, S.; Derr, M.; and Phipps, G. 1993. Design In Foundations of Deductive Databases and Logic Pro- and Implementation of the GLUE-NAIL Database Sys- gramming, eds. J. Minker, 193–216. San Francisco, tem. In Proceedings of the ACM-SIGMOD’93 Con- Calif.: Morgan Kaufmann. ference, 147–167. New York: Association of Comput- Ramamohanarao, K. 1993. An Implementation ing Machinery. Overview of the ADITI Deductive Database System. In Muggleton, S., and DeRaedt, L. 1994. Inductive Log- Third International Conference (DOOD’93), 184–203. ic Programming: Theory and Methods. Journal of Lecture Notes in Computer Science 760. New York: Logic Programming 19–20:629–679. Springer-Verlag. Mumick, I.; Finkelstein, S.; Pirahesh, H.; and Ramakrishnan, R. 1995. Applications of Logic Data- Ramakrishnan, R. 1990. Magic Is Relevant. In Pro- bases. New York: Kluwer Academic. ceedings of the ACM SIGMOD International Confer- Ramakrishnan, R., and Ullman, J. 1995. A Survey of ence on Management of Data, 247–258. New York: Research on Deductive Database Systems. Journal of Association of Computing Machinery. Logic Programming 23(2): 125–149.

FALL 1997 45 Articles

Ramakrishnan, R.; Srivastava, D.; and Sudarshan, S. Selman, B., and Kautz, H. 1996. Knowledge Compi- 1992. CORAL—Control, Relations, and Logic. In Pro- lation and Theory Approximation. Journal of the ceedings of the Eighteenth International Conference on ACM 43(2): 193–224. Very Large Databases, ed. L.-Y. Yuan, 238–250. San Shepherdson, J. 1988. Negation in Logic Program- Francisco, Calif.: Morgan Kaufmann. ming. In Foundations of Deductive Databases and Logic Raphael, B. 1968. A Computer Program for Semantic Programming, ed. J. Minker, 19–88. San Francisco, Information Retrieval. In Semantic Information Pro- Calif.: Morgan Kaufmann. cessing, ed. M. Minsky, 33–134. Cambridge, Mass.: Sickel, S. 1976. A Search Technique for Clause Inter- MIT Press. connectivity Graphs. IEEE Transactions on Computers Reiter, R. 1990. On Asking What a Database Knows. C-25(8): 823–835. In Computational Logic, ed. J. W. Lloyd. Basic Sistla, P., and Wolfson, O. 1995. Temporal Condi- Research Series. New York: Springer-Verlag. tions and Integrity Constraint Checking in Active Reiter, R. 1988. On Integrity Constraints. In Proceed- Database Systems. In Proceedings of the 1995 ACM ings of the Second Conference on the Theoretical Aspects SIGMOD International Conference on Management of Reasoning about Knowledge, ed. M. Y. Vardi, 97–111. of Data, 269–280. New York: Association of Comput- San Francisco, Calif.: Morgan Kaufmann. ing Machinery. Reiter, R. 1984. Towards a Logical Reconstruction of Snodgrass, R. 1987. The Temporal Query Language Relational Database Theory. In On Conceptual Model- TQUEL. ACM Transactions on Database Systems 12(2): ling, eds. M. Brodie, J. Mylopoulos, and J. Schmit, 247–298. 163–189. New York: Springer-Verlag. Subrahmanian, V., and Jajodia, S. 1995. Multimedia Reiter, R. 1980. A Logic for Default Reasoning. Arti- Database Systems. New York: Springer Verlag. ficial Intelligence 13:81–132. Subrahmanian, V.; Adali, S.; Brink, A.; Emery, R.; Lu, Reiter, R. 1978a. Deductive Question-Answering on J.; Rajput, A.; Rogers, T.; and Ross, R. 1994. HERMES: A Relational Data Bases. In Logic and Data Bases, eds. H. Heterogeneous Reasoning and Mediator System. Gallaire and J. Minker, 149–177. New York: Plenum. Available from www.cs.umd.edu/projects/hermes/ Reiter, R. 1978b. On Closed-World Databases. In Log- overview/paper. ic and Data Bases, eds. H. Gallaire and J. Minker, 55–76. New York: Plenum. Tsur, S., and Zaniolo, C. 1986. LDL: A Logic-Based Data Language. Paper presented at the VLDB Confer- Robinson, J. H. 1965. A Machine-Oriented Logic ence, 25–28 August, Kyoto, Japan. Based on the Resolution Principle. Journal of the ACM 12(1): 23–41. Ullman, J. D. 1989. Principles of Database and Knowl- edge-Base Systems, Volume II: The New Technologies. Rohmer, J.; Lescoeur, R.; and Kerisit, J.-M. 1986. The Principles of Computer Science Series. Rockville, Alexander Method: A Technique for the Processing Md.: Computer Science Press. of Recursive Axioms in Deductive Databases. New Generation Computing 4(3): 273–285. Ullman, J. D. 1988. Principles of Database and Knowl- edge-Base Systems, Volume I. Principles of Computer Ross, K. 1990. Modular Stratification and Magic Sets Science Series. Rockville, Md.: Computer Science. for DATALOG Programs with Negation. In Proceedings of the ACM Symposium on Principles of Database Vaghani, J.; Ramamohanarao, K.; Kemp, D. B.; and Systems, 161–171. New York: Association of Com- Stuckey, P. J. 1991. Design Overview of the ADITI puting Machinery. Deductive Database System. In Proceedings of the Ross, K. 1989. Well-Founded Semantics for Disjunc- Seventh International Conference on Data Engineer- tive Logic Programs. In Proceedings of the First Interna- ing, 240–247. Washington, D.C.: IEEE Computer tional Conference on Deductive and Object-Oriented Society. Databases, 352–369. New York: North Holland. van Emden, M., and Kowalski, R. 1976. The Seman- Sacca, D., and Zaniolo, C. 1986. On the Implemen- tics of Predicate Logic as a Programming Language. tation of a Simple Class of Logic Queries. In Proceed- Journal of the ACM 23(4): 733–742. ings of the ACM Symposium on Principles of Data- Van Gelder, A. 1988. Negation as Failure Using Tight base Systems, 16–23. New York: Association of Derivations for General Logic Programs. In Founda- Computing Machinery. tions of Deductive Databases and Logic Programming, Sandewall, E. 1994. The Range of Applicability of ed. J. Minker, 149–176. San Francisco, Calif.: Morgan Some Non-Monotonic Logics for Strict Inertia. Jour- Kaufmann. nal of Logic and Computation 4(5): 581–616. Van Gelder, A.; Ross, K.; and Schlipf, J. 1991. The Savitt, D.; Love, H.; and Troop, R. 1967. ASP: A New Well-Founded Semantics for General Logic Pro- Concept in Language and Machine Organization. In grams. Journal of the Association for Computing 1967 Spring Joint Computer Conference, 87–102. Machinery 38(3): 620–650. Schlipf, J. 1995. Complexity and Undecidability Van Gelder, A.; Ross, K.; and Schlipf, J. 1988. Results for Logic Programming. Annals of Mathemat- Unfounded Sets and Well-Founded Semantics for ics and Artificial Intelligence 15(3-4): 257–288. General Logic Programs. In Proceedings of the Sev- Seipel, D. 1995. Efficient Reasoning in Disjunctive enth Symposium on the Principles of Database Sys- Deductive Databases. Ph.D. thesis, Department of tems, 221–230. New York: Association of Computing Computer Science, University of Tubingen. Machinery.

46 AI MAGAZINE Articles

Vieille, L. 1986. Recursive Axioms in Deductive Data- bases: The Query-Subquery Approach. Paper present- Call for Papers (IEA/AIE-98) ed at the First International Conference on Expert Database Systems, 1–4 April, Charleston, South Car- olina. 11th International Conference on Vielle, L.; Bayer, P.; and Kuechenhoff, V. 1996. Industrial & Engineering Integrity Checking and Materialized View Handling by Update Propagation in the EKS-V1 System. In Applications of Materialized Views, eds. A. Gupta and I. Mumick. Cambridge, Mass.: MIT Press. Artificial Intelligence & Vieille, L.; Bayer, P.; Kuechenhoff, V.; Lefebvre, A.; and Manthey, R. 1992. The EKS-V1 System. In Proceed- Expert Systems ings of Logic Programming and Automated Reasoning, 504–506. New York: Springer-Verlag. Xerox. 1989. HIPAC: A Research Project in Active, Benicassim, Castellon, Spain Time-Constrained Databases. Technical Report 187, June 1-4, 1998 Xerox Advanced Information Technologies, Palo Alto, California. Sponsored by: International Society of Applied Intelli- Yuan, L., and You, J.-H. 1993. Autoepistemic Circum- scription and Logic Programming. Journal of Auto- gence; Universitat Jaume-I de Castellon; Universidad mated Reasoning 10:143–160. Nacional de Educacion a Distancia (UNED), Madrid Zaniolo, C. 1995. Active Database Rules with Trans- In Cooperation with: AAAI, ACM/SIGART, CSCSI, action-Conscious Stable Models Semantics. In Pro- ECCAI, IEE, INNS, JSAI, SWT ceedings of Deductive Object-Oriented Databases 1995, 55–72. New York: Springer-Verlag. IEA/AIE-98 will focus on methodological as well as practical aspects in the development of KBS’s, knowledge modeling, hybrid techniques that integrate Jack Minker is a professor of computer science in the Department the symbolic and connectionistic perspectives in the industrial application of AI, of Computer Science and the and application of intelligent systems’ technology to solve real life problems. Institute for Advanced Computer Accepted papers, either as oral presentations or as poster panels, will be pub- Studies at the University of Mary- lished at full length in the proceedings. Selected papers will be published in the land. His research areas are deduc- International Journal of Applied Intelligence. tive databases, logic program- Authors are invited to submit by November 7, 1997, five copies of papers, ming, AI, and nonmonotonic double spaced, written in English, of up to 10 pages, including figures, tables, reasoning. He was the first chair- and references. The format should be A4 or 8 1/2 X 12 paper, in a Roman font, man of the Department of Computer Science at the University of Maryland from 1974 to 1979 and 12 point in size, without page numbers, and a printing area of 15.3 X 24.2cm2 chairman of the Advisory Committee on Computing (6.0 X 9.5 sq. in.). If possible, please make use of the latex/plaintex style file at the National Science Foundation from 1979 to available in our WWW site. In addition, one sheet must be attached including: 1982. In 1985, Minker received the Association for title, authors’ names, a list of five keywords, the topic under which the paper best Computing Machinery (ACM) Outstanding Contri- fits, the preferred presentation (oral or poster), and the corresponding author bution Award for his work in human rights. He is a information (name, postal and e-mail address, phone and fax numbers). This fellow of the American Association for the Advance- page must also be sent by e-mail to [email protected] before November 7, ment of Science, a founding fellow of the American 1997. Association for Artificial Intelligence, a fellow of the Institute of Electrical and Electronics Engineers, and Contributions must be sent to the Program Co-chair Prof. Angel P. del Pobil a founding fellow of the ACM. He received the Uni- at the address below. Conference information can be obtained from the General versity of Maryland Presidential Medal for 1996 and Chair Prof. Moonis Ali, the Program Co-Chair Prof. del Pobil, and through our is a distinguished scholar-teacher for 1997 to 1998. web site. His e-mail address is [email protected]: Moonis Ali Angel P. del Pobil IEA/AIE-98 in Cooperation with AAAI General Chair, IEA/AIE-98 Program Co-chair, IEA/AIE-98 Dept. of Computer Science IEA/AIE-98 Sec., Informatics Dept. Southwest Texas State Jaume-I Univ., University Campus de Penyeta Roja San Marcos, TX 78666-4616 USA E-12071 Castellon, Spain Phone: +1 (512) 245-3409 Phone: +34 64-345.642 FAX: +1 (512) 245-8750 FAX: +34 64-345.848 E-mail: [email protected] E-mail: [email protected] http://titan.inf.uji.es/iea98/

FALL 1997 47 Articles

48 AI MAGAZINE