Paper presented at International Conference on Formal Ontology in Information Systems (FOIS'98) Trento, Italy, 6-8 June, 1998. In N. Guarino (ed.) Formal Ontology in Information Systems IOS-Press (Amsterdam) 207-217.
Completeness and Quality of an Ontology for an Information
System
Robert M. Colomb Department of Computer Science and Electrical Engineering Ron Weber Department of Commerce The University of Queensland QLD 4072 Australia
Abstract
We examine the problems of completeness and quality in design of information systems. Taking the view that an information is a representation of a social reality created by genres of speech acts, we view the state of an information system as a text, and the dynamics of the system as essentially the dynamics of a text editor. This view enables us to make use of a generalised ontology developed by Bunge to get a clear picture of the functions of an information system, and therefore a set of criteria for ontological completeness. Further, quality in an information system is seen as a matching between the semiotics of the system and the semiotics of the organisation in which the system is embedded, allowing us to make use of the quality principles advocated by Debenham. The value of these results is essentially that they validate the large body of existing information systems, and also validate the basic approach used to construct them, although suggesting some improvements. We can build and use information systems confident that they will be valid under changes in the understanding of meaning and also changes in the understanding of the metaphysics underlying physical and social reality.
Keywords: the ontology of information and information processing, top-level ontological taxonomies, foundations.
- Introduction
- Completeness and quality
- What is an information system, anyway?
- How do we evaluate ontological completeness?
- How do we evaluate information system quality?
- We can evaluate adequacy and quality, but what about completeness of instances?
- Conclusions
- References
1. Introduction
Ever since the introduction of the entity-relationship model [1], the construction of specialised ontologies (often called conceptual models) has been an essential part of the information systems design process. Historically, these ontologies have been partial. They have not provided a model of all phenomena of interest in the domain of discourse. Rather, they have been used to describe only those phenomena that can be represented conveniently in database schemas. Other phenomena (eg., events and transformations) have been represented via programming language statements, initially using languages like COBOL and more recently using fourth-generation languages like Oracle Forms and Powerbuilder.
Recently, specialised ontologies that are more complete have begun to appear. For example, workflow modeling grammars have now been developed that allow scripts to be generated which describe processes in the domain of discourse (eg [2]). In conjunction with the scripts generated via conceptual modeling grammars, these scripts provide a more complete coverage of the phenomena of interest in the domain of discourse.
The usual goal of constructing specialised ontologies has been to represent the semantics of the domain of discourse. It now seems, however, that they can also be used as a basis for generating rough-cut designs for user interfaces (eg [3]). These types of advances raise the possibility of being able to completely specify an information system at the level of abstraction of a specialised ontology and then having the specification operationalised via high-level translators and interpreters.
Given the possibility, therefore, that an information system could be finished once its specialised ontology has been constructed, in this paper we focus on two questions:
Ä How do we know that our ontology is complete?
Ä How can we know if we have a high-quality ontology?
In the sections that follow, we first explore the concepts of completeness and quality of a specialised ontology. We then present a view of the nature of an information system that enables us to give a concrete expression of the problem of ontological completeness and quality. Finally, we indicate a method of answering the two questions in particular domains of discourse.
2. Completeness and quality
The completeness under consideration in this paper is ontological completeness. Part of the issue is to understand what we mean by ontological completeness both in theory and in practice, and to understand what kinds of claims can be credibly made about ontological completeness. We at least want to say that a system is ontologically incomplete if it is not completely specified, so a system having for example only an ERA conceptual model would not be ontologically complete.
We know about other kinds of completeness that are related to computing. For example, logical completeness is a property associated with combining a procedure for constructing well-formed formulas (wffs), a definition of truth that relates to interpretations and models of logical systems, and a proof procedure that allows new wffs to be derived from old wffs. A logical system is logically complete if every true wff can be derived.
The other side to logical completeness is consistency. If falsity can be derived, then any wff can be derived, so trivially all true wffs can be derived.
Godel's theorem states that a logical system in which arithmetic can be imbedded (for example, the first-order predicate calculus with the Peano axioms) is complete only if it is inconsistent.
A second kind of completeness is computational completeness, which is a property of a programming language. Consider the programming language called a Turing machine. No one has ever invented a programming language in which a program could be written that also could not be represented as a Turing machine. Indeed, the Church-Turing thesis is that it is impossible to have a programming language more powerful than a Turing machine. This thesis has not been proved, but it has never been seriously challenged.
Programming languages exist that are not complete (eg SQL). In other words, it is possible to formulate programs that these languages cannot represent. For example, SQL does not support recursion, which makes it unsuitable for a range of problems including bill-of-materials applications.
The other side to computational completeness is non-termination. Programs can be written that go into an infinite loop or fail to terminate for some more complex reason. It is impossible to write a terminating program, however, that tests other programs for termination. This outcome is called the unsolvability of the halting problem.
Both logical and computational completeness are standard results that are widely known (eg [4]). Ontological completeness, however, is neither logical completeness nor computational completeness. Two examples will clarify the distinction among the three uses of completeness. Both relate to the development of an information system using the ER model as a basis for providing a (partial) specification of the domain of discourse. Therefore, by our preliminary definition above, neither are ontologically complete.
A is a standard business information system, implemented in a relational database system using SQL. It is logically complete (every SQL statement has a well-defined result), but it is computationally incomplete (there are queries we might want to make, which can be computed using a Turing machine or its COBOL equivalent, but which cannot be expressed in SQL).
B is a system to record observations about the world. The observations are expressed in the first-order predicate calculus. Moreover, they are stored in a persistent data system supporting a resolution-based theorem prover. This system is computationally complete but logically incomplete, because Godel's theorem applies to it (there are true statements which cannot be proved).
Now consider again ontological completeness. To improve our preliminary definition, we need to know of what "thing" in the world is it a property. Both logical completeness and computational completeness are properties of the languages and reasoning platforms used to express statements, not properties of the statements themselves. We would therefore expect that ontological completeness would be a property of the language and reasoning platform used to construct an information system.
But what is an information system? One conception we might have is that it is simply a computer program, or at least it is a bureaucratic procedure that is essentially a computer program. A computer program, in turn, can be conceived as a function that produces a particular well-defined output from a particular well-defined input. Viewed this way, the details of how the computation is performed are not relevant (the black-box view).
If the details of how the computation is performed are not relevant, then most likely the computation can be performed in many different ways. The existence of alternatives gives rise to the possibility of design&endash;the choice of one of them. All sorts of designs can be proposed based on tradeoffs made among various non-functional properties of the information system (the engineering approach).
Many non-functional properties might be considered: cost of construction, cost of installation, response times, cost of achieving a particular maximum allowable response time, etc. These quasi-physical aspects of a system, however, do not seem to be relevant to ontological completeness.
Other non-functional properties are associated with the relationship of the system to the people who operate it: ease of use, trainability, consistency of interface, suitability for long-term continuous use, etc. These ergonomic aspects also do not appear to be relevant to ontological completeness.
A further non-functional property is maintainability: the cost of making changes to the system over the part of its life cycle beyond its first commissioning. A view given by Debenham [5] bases maintainability on a relationship between the semiotic system realised in the specification of the system and the semiotic system of the organisation in which the information system is used. (A semiotic system is a characterisation of a language in terms of its grammar, the collection of words in it, how the words are related to each other, and the conventions governing the use of the grammar in creating texts. See [6] or [7].) Debenham's approach to maintainability offers considerable insight into the problem of ontological completeness. He argues:
- An information system is built from components.
- Components have associations among them which are derived from the fact that together they form an interpretation of some aspect of the system's environment (semantic links).
- Components also have associations that are derived from the method of constructing the information system. These associations are naming conventions, referential integrity constraints, relationships of implementation components&endash;say, database column names&endash;to high level design constructs&endash;say, an ER diagram. These associations are called structural links.
- Debenham's maintainability quality principle is that every semantic link must be supported by structural links. Moreover, every semantic link must be represented in the implementation in only one way.
If an information system conforms to Debenham's conditions, he argues that maintainability is improved because:
- An information system functions by interacting with its environment.
- Its environment consists largely of people interacting in a social system.
- The people interact via a semiotic system.
- A request for change in the information system arises either from a desire to implement a class of interaction not heretofore implemented or from a change in the semiotic system. (Requests for change also come from changes to underlying technology, but these are not considered.)
- The major costs of implementing a (changed) information system are human costs (changed procedures, training, etc.).
- Therefore, changes tend to reflect minimum change to the organisation's semiotic system. (Major changes to the organisation are so expensive that the cost of changing the information system per se is negligible. Thus, the tendency is to re-implement the information system anyway.)
- If the organisation's semiotic system is reflected in the internal structure of the implementation, the cost of change is likely to be minimised.
(Note that Debenham's work is not cast in this semiotic language, but we believe our interpretation accords with his in spirit.)
This analysis leads us to suggest that ontological completeness is associated with the relationship between how an information system is specified and the semiotic system in which it is operated.
What, then, can we say about the semiotic system? We are looking for an operationalisable definition of ontological completeness. One form this might take is some sort of checklist&endash;a relatively short (certainly finite) set of properties which, if the semiotic system is found to have, allow it to be declared complete. Our problem, however, is that a semiotic system is (at least part of) a natural language. In general, we can say little about natural language at a coarser level of granularity than its grammar rules. Grammar rules are not sufficient to evaluate ontological completeness. Clearly we want our specification to be syntactically correct (obey the grammar rules of the semiotic system from which it is constructed). To be of any use, however, a definition of ontological completeness must go much further.
If we are looking for a checklist at the level of generality of a semiotic system in general (natural language), then we are doomed to failure. To see the difficulty, note that the first-order predicate calculus is part of natural language, and the first-order predicate calculus is not finitely axiomitisable. (The proof procedure of Godel's theorem guarantees this outcome: see Hofstadter [4].) Thus, we would not expect to find a small number of concepts that would allow us to characterise everything one might want to say in the first-order predicate calculus. Moreover, because the predicate calculus is a subset of natural language, a fortiori we would not expect to find a checklist of properties that would characterise anything said in a natural language.
This analysis leads us to hope that we do not have to be concerned about all of natural language if we are to come to grips with ontological completeness. Indeed, we would like to restrict our attention to semiotic systems that can be characterised by a small number of concepts at a coarser level of granularity than words and grammar rules. Even though we would be able to say little about our semiotic system, we would not be concerned with what we couldn't say. We believe we have achieved this aim, and in this light we will articulate our analysis below.
3. What is an information system, anyway?
A distinction can be made between information system development methods and the functions performed by an information system. For the purpose of ontological completeness, the process of developing an information system is irrelevant, except in the sense that the process either produces or does not produce an information system that has ontological completeness. Furthermore, we have argued in the preceding section that the implementation of an information system can be factored out because ultimately it might be automated. Thus, it has no bearing on the problem of ontological completeness. The focus, instead, is on how an information system interacts with its users and how its specification relates to the semiotic systems employed by its users, with particular reference to the language and reasoning system used to represent the information system.
We argue that an information system is used to store and manipulate records of socially constructed reality. In many cases, the information system's contents are social reality. (Is a student enrolled in a subject if their student identifier does not appear on the classlist?) Social reality is constructed by people using speech acts [8]. In a speech act, the situated production of a language object is an utterance of a locution. Some locutions make consequential changes in social reality. The quintessential such locution is, "I pronounce you husband and wife," made by a marriage celebrant to a couple who fulfil a number of requirements. There are a many others, however&endash;for example, naming something, making agreements, promises, or threats, and giving permission or prohibiting. A locution of this sort is said to have illocutionary force from the speaker's viewpoint and perlocutionary effect from everyone else's viewpoint. An utterance with illocutionary force and perlocutionary effect is called a speech act, which changes social reality for both speaker and listeners.
Speech acts are organised into systems of genres with sometimes strict rules that define what can be done, by whom, and under what circumstances. These rules are called framing rules. For example, consider a customer's placement of an order as a speech act. Part of its framing is an acknowledgment sent to the customer by the seller to indicate receipt of the customer's order. In short, a customer has not placed an order with say Amazon.com (AC) until AC has sent an e-mail message confirming receipt of the order message. Social reality is now changed. The customer has agreed to become indebted to AC following AC's shipping a specified book. The change of state in AC's information system is a representation of this change in social reality.
If an information system is a representation of social reality, then a state of an information system can be considered as a text ([9] is based on this sort of idea). To see the state of an information system as a text is an extremely abstract view. Nonetheless, consider how this view might be justified.
We generally think of an information system as a set of applications centred around a database. The state of the database is the state of the information system. We therefore need to justify viewing the population of a database as a text. Consider that a structured database record might allow one to state "the part identified by 123456 is stored in warehouse bin identified by AA in quantity 231, and is purchased from the supplier identified by ZX443, whose name is Acme Manufacturing". Both this statement and the database record together with the database's conceptual model are representations of the same fact. (The structured database conceptual modelling technique called Object-Role Modelling, eg [10], is based on verbalisation of this sort.) The verbalisation of a database record is a perfectly respectable piece of text. In principle, the entire population of all the tables in the database could be represented as text in this way.
If we accept that the state of an information system is a representation of the state of social reality for (an aspect of) an organisation, then there are at bottom two things we can be interested in: maintaining the state, and finding out things about the state.
An action undertaken by an information system results in a change in its state. If we accept that the state of the system can be represented as a text, then changing the state can be viewed as editing the text. In principle, therefore, changing the state of an information system could be managed with a word processor. Queries can also be made with a word processor. Admittedly, complex queries require a sophisticated word processor that supports SQL or some other query language. Such a tool currently does not exist, but in principle it could be built.
Moreover, an important benefit of an information system is that complex updates are performed automatically. Thus, our hypothetical sophisticated word processor has to be able to recognise the event leading to the need for a change of state in the information system. Given the state can be represented as a text, however, so can the event. Thus, our sophisticated word processor has to be able to construct the updated text from the original texts (which requires it to support a programming language of some kind). These sorts of facilities already exist in embryonic form in the macro languages of current word processors.
We can think of the database, communications, user interface and programming language platforms used to construct information systems as extremely sophisticated text editors. Note that a query such as used in information systems, expressed in a language like SQL, is essentially a computational process on the syntactic elements of the text.
Note, also, that the state of an information system and events in the information system are not the only texts that we have. Because the macros and other parameters of a word processor are text, the information system itself is also a text. Ontological completeness is going to be a property of the language and reasoning system used to specify the information system. We can now see that the language and reasoning system are going to need to be sufficient to develop a text editor, with the capability of editing the text of the information system's own specification as well as the text of the representation of the organisation's social reality.
4. How do we evaluate ontological completeness?
To evaluate an information system for ontological completeness, we have argued above that we have to find a simple characterisation of the relationship between the information system's specification and the semiotic system employed by the organisation whose social reality is represented in the information system. The problem we identified was that it is impossible, in general, to axiomatise a semiotic system. Thus, we have to seek a more limited, but adequate, goal. We reached the point of associating the state of an information system with a representation of the organisation's social reality. In addition, we conceived of the dynamic part of the information system as a text that specifies predicates and operations. In short, the information system itself is a text.
Recall, we pointed out earlier that tools now exist which automatically develop an information system from a conceptual model [3]. The user's model of an information system is constructed from the elements of the ontology. If the implementation is automatic, the script that represents the implementation of the user's model becomes invisible because in effect the user's model becomes the code in a very high-level programming language. This outcome hides physical structure phenomena from the information system, thus allowing it to be the province of the information technology branch of computer science. Instead, the importance of deep structure phenomena are emphasised, which are associated with the meaning of the information system. The deep structure is the specification of the system in terms of input-output behaviour under all circumstances: a detailed and specific recording of the framing rules for the speech acts implemented in the information system.
A text, however, is a statement in a language that is intended to have meaning. In this regard, much ontological research in the computer science and information systems fields focuses on understanding the meaning of information systems texts. Our overarching concern is to place information systems on a sound foundation so that we are confident that our systems are doing what we think they are doing. We would worry, however, if the soundness of information systems depended on the soundness of our understanding of the meaning of texts. After all, millions of information systems exist. Suppose someone discovered that we had been wrong all along in our understanding of meaning. It would be catastrophic if this new understanding suddenly called into question the soundness all existing information systems built with the now-discredited understanding.
The possibility of discovering that our understanding of meaning is incorrect is not idle speculation. At present, great controversy on this point is occurring in a number of areas. Perhaps the most critical is the debate between the postmodernists and the traditionalists as to whether there is any sense in which a text actually has a definite meaning. This debate is widespread and has a large number of participants. Many postmodernists refer to the philosopher Jacques Derrida (eg [11]). A strong spokesperson for the traditionalists is John Searle (eg [12]). Worse for our purposes is that the postmodernists debate among themselves about whether there are any limits on the interpretation of texts, eg., the debate in [13]. We as information systems researchers are not central players in the effort to understand meaning - we must adopt and adapt theories of meaning from these researchers. Since there are many different and strongly argued positions, if we select one and build on it, we run a serious risk of making the wrong choice.
The problem of texts possibly having indeterminate meanings is not new. For example, it was faced by mathematicians in the 18th and early 19th centuries in their attempts to prove the parallel postulate of Euclid. These attempts resulted in their realising that mathematics was not necessarily connected with physical reality and the subsequent invention of many non-Euclidian geometries (see, eg [4], p. 91). Similarly, physicists long ago came to terms with this problem. In a famous essay, "The unreasonable effectiveness of mathematics in the natural sciences," Wigner ([14], ch 17) showed how the mathematical formulations of the solutions to many physics problems can survive great changes in how we interpret them as representations of the physical world.
In many ways, computer programming is a branch of mathematics. Moreover, recall we argued above that the use of mathematics to understand reality is somewhat independent of what we think reality actually is. Thus, if the state of an information system can be described as a text, and its operation as either editing the text or making queries on it which can be computed by essentially syntactic analysis, then the information system itself never has to pass beyond the text to whatever meaning it may have. This approach is closely analogous to the method of phenomenological reduction advocated by Husserl [15]. Phenomenological reduction restricts study to the content of consciousness, eliminating from consideration the (assumed to exist) relationship between what is in consciousness and the external world. The restriction process is called bracketing. Our contention is that we can restrict attention to the representations of organisational reality, bracketing the meanings of these representations, or indeed whether various parts of the organisation assign different meanings to the same representation. For example, signing a contract can represent a triumph to the marketing department, but an impossible workload to the implementation department. The only concern proper to the information system is that both departments agree on the representation, not that they agree on the meaning.
In this view, we do not have to worry about the meaning of the text but only how it is edited. Editing a text involves inserting, deleting, or changing chunks of text. Therefore, we need to care about chunks of text and what can be done with them.
Our central problem of finding some sort of check list which can test an information system for completeness is now very close to being operationalisable. We now know what an information system is. The standard procedures of systems analysis are aimed at elucidating the semiotic system in which the organisation's social reality is expressed, and the semiotic system in which the information system is to be specified is a matter of engineering choice. We bring into play at this point an ontology (in the present context it is a generalised ontology) presented by Bunge [16, 17, 18]), which is summarised by Weber [19]. Bunge's ontology can be encapsulated by the idea that the world is made up of things that have properties. A thing can be complex&endash;for example, a system is a complex thing made up of interacting simpler things. Properties of things include events that have occurred to the thing and laws that govern the states or events that can occur in things.
Bunge's ontology is very much more elaborate than described above, but it is only one of many different and sometimes incompatible ways of conceiving the world. Thus, its validity is a moot point. Nevertheless, whatever comments might be made about the applicability of Bunge's generalised ontology to reality apply also to texts. The reason is that a text is a thing in the context of Bunge's ontology, and accordingly the properties of texts are by definition properties of things. We claim that Bunge's ontology gives a good account of things, their properties and what can be done to them, so that ipso facto it gives a good account of texts, their properties, and what can be done to them.
We have now reached the point of being able to give an operational definition of ontological completeness. We have argued that ontological completeness must be defined in terms of the relationship between an organisation's semiotic system (OSS) and the information system that describes it. The information system is a text constructed using a different semiotic system&endash;the information systems semiotic system (ISSS). The information system is a representation of (aspects of) the social reality of the organisation. Part of the social reality is a collection of genres of speech acts, which are the means by which the social reality is changed. A speech act genre consists of a set of framing rules and the specification of the consequent change in social reality. A speech act generally takes place at least partially outside the information system, although the information system might be intimately involved (as in the Amazon.com example discussed above). The record of the speech act having taken place is a text in the ISSS. Part of the framing rules might also be expressed as predicates on the information system, which are also texts in the ISSS. Note that these framing rules can include access control lists, authentication procedures, and other ways of determining that the speech act has been performed by an authorised agent.
The task of Bunge's generalised ontology is to categorise all possible aspects of the record of social reality and how that record can be changed, allowing us to check whether the specification is at least potentially complete. By potentially complete, we mean that every necessary class of semiotic element is present in the ISSS. This potential for completeness, which as we have seen above is analogous to logical and computational completeness, has been called ontological adequacy by Guarino [20]. It is, of course, impossible to determine a priori whether an instance of an information system (a specialised ontology) actually completely matches the social reality of the organisation, because that judgment depends on how successfully it interacts over time with the organisation in practice (This issue will be developed further below.) Nevertheless, we assert that the Bunge generalised ontology is sufficient to evaluate the ontological adequacy of the ISSS.
Ontological adequacy has been defined operationally as a property of the semiotic system used to specify the information system, evaluated with respect to Bunge's generalised ontology. In particular, every element of Bunge's generalised ontology must be represented in the generalised ontology governing the construction of the ISSS.
To test whether a particular ISSS is ontologically adequate is a complex task. This task has been elaborated and undertaken for the entity-relationship model (ERM) as an example by Weber [19]. We sketch here a test the ontological adequacy of Bunge's generalised ontology for our text editor interpretation of an information system.
First, consider a state of social reality. The state is typically expressed in a more or less specialised subset of natural language. The grammar rules governing the subset of natural language are laws that are properties of the information system thing. In the ISSS, these laws are represented as statements in a meta-ISSS like the ERM. In practice, the ERM appears to be capable of representing these laws. There is a great variety of such meta-ISSSs in use, which are more-or-less equivalent to the ERM.
Second, the record of the speech act is also a thing. This record includes text, which can be represented in the same meta-ISSS as the state (can be taken as part of the state), and also possibly the result of the evaluation of a predicate on the state representing part of the framing rules. The former is covered by the grammar rule laws, while the latter is also a law typically expressed in a query language such as SQL. This second law is also a property of the information system-thing.
Finally, the procedure by which the record of social reality is updated is also a law, generally expressed as a set of statements in some programming language. The meta-ISSS must support a sufficiently complete programming language.
The only other aspect of the information system is the ability of a user to examine the state of the record of social reality, which is the same as the ability to construct a predicate. The meta-ISSS's query language serves this purpose, too.
We conclude that the Bunge generalised ontology is ontologically adequate for information systems.
5. How do we evaluate information system quality?
We have proposed a method of determining whether an information system development environment is ontologically adequate. We will address ontological completeness further below. We now return to the issue of quality.
Essentially, quality was defined above as the "visibility" of the OSS in the ISSS. It is therefore a property of a specific information system rather than an information system's development environment. The Debenham criteria for maintainability&endash;namely, that each element of the OSS be represented in the ISSS in one structure and one structure only&endash;can be applied in a straightforward way.
Debenham's knowledge normalisation (what we call knowledge quality) principles fundamentally are metarules for avoiding redundancy in the specification of framing rules in the deep structure. They have no bearing on physical structure phenomena. If we conceive of the collection of framing rules as being made up of phrases, sometimes the same phrase can occur more than once.
For example, AC both sells books and takes orders for out-of-print books to be filled if a second-hand copy can be located. Both applications require verification of credit card details. The business rule is that verification of credit card details is the same in both cases. If that business rule is represented by copying the credit card details verification from the book order to the out-of-print book order, then the fact that the rules are the same in both cases is not represented in a strong way. Either copy could be modified in the future without modifying the other. Debenham's metarule is that the requirement that the rules are the same in each case should be represented explicitly, thereby making unsynchronised changes impossible.
The other side of Debenham's metarule is that different phrases should be represented separately, even if two phrases always occur together at the present time, on the grounds that a future change might involve only one. For example, the interaction of the book-ordering process with the credit card company involves both verifying the card details and making the billing. The two sub-processes could be represented in the specification by a complex combination of statements that makes it difficult to identify the two phrases. Alternatively, they could be represented by combinations of statements that clearly differentiate the two phrases. In the latter case, it is much easier to represent the out-of-print book process which verifies details without billing in such a way that verification of details is the same in both cases.
In short, Debenham's metarules are encapsulated by the slogan "represent one phrase of the business/framing rules in one structure and one structure only."
6. We can evaluate adequacy and quality, but what about completeness of instances?
Whether an information system completely matches the ontological system of a particular organisation is distinct from the issue of ontological adequacy. Seeing an information system as a text editing system suggests a perspective on the problem of information systems development.
An analogy of writing might make things clearer. There is a view that a person "has a book inside them" and that writing "gets it out." In this view, the book always exists in some sense. The process of writing is a representation of a pre-existing book in a different form. An alternative view, which is gaining strength, is that the book comes into existence in the process of writing it. The process of reviewing and revising drafts is seen as evaluating aspects of the book according to various criteria (aesthetics, consistency, etc.) and making changes to improve it, rather than making it conform more closely to the (unseeable) "inner book."
From this latter viewpoint, a consistent and ontologically adequate specification of an information system is always complete, because social reality would be the way the information system is&endash;in other words, the information system is social reality. The process that stakeholders undertake to make the system "more complete" can be described as evaluating the framing/business rules for workability according to criteria such as interaction with previous habits, the consequences of the treatment of unusual conditions, auditability, etc. An information system is complete, therefore, if it works in practice. It is possible to test the information system's development environment for ontological adequacy, and also the information system for quality, but it is not possible in principle to determine a priori whether an information system is complete. Furthermore, since an information system is complete if it works in practice, it may be considered complete by some people (for whom it works) and not others (for whom it is lacking).
Our approach contrasts with that of Hirschheim et al. [21] who examine a much broader issue: the processes of developing an information system and their relationship with the ways an information system is used. Because of their breadth of perspective, they (necessarily as we have argued) do not attempt anything like our exploration of ontological completeness, adequacy or quality. Their elucidation of different schools of thought on how meaning can be understood, however, underscores our argument that the issue of meaning is not settled, and therefore our argument that it would be dangerous to base the soundness of information systems on a particular approach to meaning. Furthermore, their discussion of fact-based versus rule-based approaches to conceptual modeling is very close to our analysis of an information system. Specification of the state text requires a fact-based generalised ontology, while the framing rules and update procedures (laws in Bunge's generalised ontology) require rule-based epistemological level structures [20] for their specification. We claim therefore that our argument addresses a different problem from that addressed by Hirschheim et al. [21]. To the extent that our argument overlaps with that of Hirschheim et al., we are consistent with them.
7. Conclusions
We have examined the problems of completeness and quality in design of information systems. We have concluded that completeness of an instance of an information system is not well defined, since it depends on the views of stakeholders as to whether the system works in practice. However, a more abstract property ontological adequacy, analogous to logical and computational completeness, is well-defined. Taking the view that an information is a representation of a social reality created by genres of speech acts, we see the state of an information system as a text, and the dynamics of the system as essentially the dynamics of a text editor. This view enables us to make use of a generalised ontology developed by Bunge to get a clear picture of the functions of an information system, and therefore a set of criteria for ontological adequacy. Furthermore, quality in an information system is seen as a matching between the semiotics of the system and the semiotics of the organisation in which the system is embedded, allowing us to make use of the quality principles advocated by Debenham.
The value of these results is that they validate the large body of existing information systems. They also validate the basic approach used to construct them, although suggesting some improvements. We can build and use information systems confident that they will be valid under changes in the understanding of meaning and also changes in the understanding of the metaphysics underlying physical and social reality.
References
[1] P.P.S. Chen "The Entity-Relationship Model - Towards a Unified View of Data" ACM Transactions on Database Systems 1(1) (1976) 9-36.
[2] W. Sadiq and M. Orlowska "On Correctness Issues in Conceptual Modeling of Workflows," Proceedings of the 5th European Conference on Information Systems, Cork, Ireland, (1997) 943-964.
[3] A. Berglas Non-monotonic Reasoning and End-User Conceptual Modelling PhD thesis, The University of Queensland 1997.
[4] D. Hofstadter Godel, Escher, Bach: An Eternal Golden Braid Penguin 1980.
[5] J.K. Debenham Knowledge Systems Design Prentice Hall 1989.
[6] F. de Saussure Course in General Linguistics, Duckworth : London 1983.
[7] C.S. Peirce, Elements of Logic Collected Papers, Volume 2, Harvard University Press, Cambridge MA 1958.
[8] J.L. Austin How to Do Things with Words Cambridge, Mass, Harvard U.P 1962.
[9] R.M. Colomb "Impact of Semantic Heterogeneity on Federating Databases" The Computer Journal 40(5) (1997) 235 -244.
[10] T. A. Halpin Conceptual Schema and Relational Database Design, 2nd edition Prentice-Hall 1995.
[11] J. Derrrida Writing and Difference, University of Chicago Press 1978.
[12] J. Searle "Rationality and Realism", Chapter 3 in J.R. Cole, E.G. Barber and S.R. Graubard, editors The Research University in a Time of Discontent Johns Hopkins University Press 1994.
[13] U. Eco The Limits of Interpretation, Indiana University Press, Bloomington 1990.
[14] E. Wigner Symmetries and Reflections Indiana University Press 1967.
[15] E. Husserl Ideas : general introduction to pure phenomenology London : Allen & Unwin 1931.
[16] M. Bunge Treatise on Basic Philosophy: Volume 2: Semantics II: Interpretation and Truth Reidel, Dordrecht, Holland 1974.
[17] M. Bunge Treatise on Basic Philosophy: Volume 3: Ontology I: The Furniture of the World Reidel, Dordrecht, Holland 1977.
[18] M. Bunge Treatise on Basic Philosophy: Volume 4: Ontology II: A World of Systems Reidel, Dordrecht, Holland 1979.
[19] R. Weber Ontological Foundations of Information Systems Coopers & Lybrand 1997.
[20] N. Guarino "Formal Ontology, Conceptual Analysis and Knowledge Representation" International Journal of Human and Computer Studies 43(5/6) (1995).
[21] R. Hirschheim, H. Klein, and K. Lyytinen Information Systems Development and Data Modelling: Conceptual and Philosophical Foundations Cambridge University Press 1995.