The theory described in this paper can be considered as a theory for design and action in the taxonomy of Gregor (2002). As such, it should be useful in guiding future designs.
One thing the theory does is explain why SQL and other logical databases are overwhelmingly the platform of choice in information systems implementations. This, however, does not seem to be a controversial situation. It is not a matter for concern, and there are no serious proposals for any other kind of platform. So to have value, the theory in this paper must do more.
The success of logical databases in information systems is most apparent in systems that serve highly focused organisational subunits. These are the subunits responsible for limited classes of speech acts, so needing records of limited classes of institutional facts. These are also the levels of institutional structure where the informal behaviour patterns and norms are the most stable, so where the background aspect of the context for the institutional facts is the most uniform.
As a result of success at this scale, there has been for many years a push to tie the information systems together. More recently, the availablity of cheap and powerful communication facilities has led to a push for tying together information systems of separate organisations into what may be thought of as world-scale computing. Although there have been successes at both of these enterprises, there have been many failures, with projects abandoned after vast expenditure. The idea that logical databases work well because they manage institutional facts can explain the successes and failures, and can be used to predict a priori whether a given project proposal has a chance of success.
The first of these enterprises, that of tying together the information systems in a single large organisation, was given a formulation as an extension of logical database technology in the federated database movement whose strategies are summarised by Sheth and Larsen (1990). The idea was that if we had many individual information systems, we could build a single big system by federating the data models and schemas of the local systems without requiring changes in the local systems. These efforts often failed, an example being the CS90 project of Westpac Bank in Australia in the late 1980s, which was abandoned after several years at a cost reported to be about A$500 million. Other major banking projects of the type were similarly abandoned at even higher costs.
In terms of the present theory, the reason these projects failed is that the speech acts performed by the organisation did not extend to the appropriate scale. The organisational subunits are in fact generally created to perform the limited class of speech act, and the framing rules for the speech act are often largely limited to things within the scope of that organisational unit. In a bank of the 1970s the savings accounts would be managed by a department, which would define what a customer was, the rules for interest payments, what addresses were kept, and so on. The home mortgage department would have analogous definitions, but there would have been no mechanisms to synchronise them. Also, different types of speech act have framing rules that take different things into account. A two-year-old might be a valid customer for a savings account, but not for a home mortgage, for example. Large organisations typically support hundreds of separate information systems serving low-level organisational units or specialised staff functions, and the speech acts performed by these subunits are typically uncoordinated. Furthermore, each organisational subunit has its own culture, so contributes a different background to the context of the speech acts for which it is responsible.
To integrate the information systems supporting these organisational subunits required far too much negotiation and resolution of different views of what were in principle common concerns, beyond what was needed to support the speech acts for which these units were actually responsible.
Tying together the information systems of a large organisation turned out not to be primarily a technical problem. It did require a large investment in technology, but was also predicated on extending the scope of the speech acts performed by the organisation to encompass all of the interactions needed to serve particular stakeholders. This involves not only the formal rules but requires creating a common culture so as to create a uniform background. This extension of scope is called business process reorganisation. If a bank wants to provide a web interface integrating all the services it provides to a given customer, the various departments need to come to a common definition of what a customer is, how they are named, what addresses they can have, under what conditions a customer is enabled to access a particular product, and so on. Making these decisions then reorganising the organisational subunits to work from the now larger scale ontology is a major cost to the organisation. Investment in technology is an enabling factor for business process reorganisation, but is not the major cost.
The prediction of the theory is that no proposal to integrate the separate information systems of organisational subunits is likely to succeed unless the organisation is rebuilt so that the speech acts it performs are at the scale of the whole of business interaction with classes of stakeholders. Once the speech acts are at the right scale, the consequently revised schemas and models will be able to be integrated in a relatively straightforward way. So the failure of the federated database approach to information systems integration can be retrodicted by the theory.
A similar problem has arisen more recently with the Internet. Since it has become technically feasible to interconnect systems operated by different organisations, people have been talking about interoperation. Of course people have been able to find resources using text database technology (search engines), and to compose individually selected services for particular purposes, but the dream is to be able to interoperate automatically using logical database technology. (This is often called use of intelligent agents.)
There are a number of manifestations of this dream, the most recent and concrete of which is the semantic web (Berners-Lee and Fischetti, 1999). There are a fair number of developments of what might be thought of as infrastructure for interoperation, for example XML, RDF, OWL, SOAP and WSDL [1] . There is a sometimes not clearly expressed dream that if you represent your web site or database in XML, or if you put descriptors on your site using RDF or OWL, then you can interoperate using logical database technology with anybody else who does so too.
The theory of this paper, that logical databases work because they store institutional facts, leads to the conclusion that interoperability using logical database technology is only possible if the interoperating sites share speech acts and consequently share in the creation of institutional facts. In particular, they must have sufficient shared culture so that the background is sufficiently uniform. (For another view of this issue, see Colomb, 1997.) Some of the kinds of situations where this condition is satisfied include:
The sites do business together. This is what Electronic Data Interchange (EDI) is all about. For example, a group of businesses agree on common terms and common business messages with agreed semantics, and can then buy and sell from each other by the interoperation of their respective purchasing and order entry systems. E-commerce exchanges are built on this basis. The agreement on common terms and common business messages with agreed semantics constitutes the synchronisation of the framing rules for speech acts, so that the interoperation can make speech acts and there is an agreed semantics for the consequent institutional facts. The agreement is a transcendent ontology, supported by a common background. The ontology is transcendent because the only way to change the common world is to change the ontology, which is done by the management body outside of the routine interoperation of the sites.
All sites report to a central body using a common ontology. Tax returns in a given jurisdiction or financial reports to a given stock exchange are examples. The common ontology is the set of regulations and accounting standards established by the tax office or stock exchange and enforced by auditors and the commercial law institutions. This ontology is generally transcendent because it is imposed by the central body, and the participation in the relationship with the central body gives aspects of common culture so there is a stable background.
All sites operate as small players in a dense market. An example is residential property sales in a particular city. There are many agents, many sellers and many buyers, and each has the choice to deal with many of the others. In these markets, conventions develop so that to do business one must do it pretty well the way everyone else does. The speech acts and consequent institutional facts are similar by convention rather than by agreement. Any innovation either dies out quickly or is quickly adopted by everyone else due to competitive pressure. Here, the ontology is not transcendent, but immanent, derived from patterns in the background. It is possible to build, for example, services that will search for a house in many agents’ sites. There are many ways to do this requiring more or less cooperation among the players. An immanent ontology is unstable in that a player may innovate at will, and that innovation may take off unpredictably. Background is the critical factor in this situation.
Unless there is some reason to assume the interoperating sites share institutional facts, there is no reason to think that interoperability using logical database technology is possible.