| Hi Ed,
Seems to me the problem as described is not about different apps with different views of Customer, that’s well understood and things like namespaces handle it sufficiently.
The problem was that the Customer was changed in an existing single app to change its membership in that app. Therefore, any processing (e.g. monthly report) of Customer that depended on then “correct” definition was now invalid (e.g. comparing monthly reports is now meaningless). That’s what is objectionable about this scenario - changing the definition and membership of an already defined class in an operational app.
Cheers, David
UK +44 7788 561308
US +1 336 283 0606
I would like to take this discussion back two steps to where it started, i.e., with Tom’s example of a manager electing to redefine the term “customer”, either to align the usage in his division with that of other divisions (a good practice), or to achieve some deception for upper management (a dubious practice).* The fact of the matter is that in large business organizations and partnerships, the same term has multiple meanings with a large overlap in the referents. So the “socio-spatial” scope of a term is just as important as the temporal scope. (I recently argued in another forum that) business partners often agree on the referents of a term as it relates to their partnership, without agreeing on the definition of the term at all. In effect, there are two or more different predicates that are expected to have a common extension in the partnership space. The second problem is that the intensions (definitions) of business terms DO change over time, precisely because the business environment and the business practices change. The underlying intent of the term does not usually change (much), but any kind of formal operating definition gets modified to deal with the new and evolving elements of the actual business world. Even biological taxonomy changed significantly in 1970; the definitions of the “second” and the “inch” changed in 1972; and no one had to protect the data on his phone in 1995. So, changing the meaning of the term ‘customer’ to match a change in business practice is a common occurrence. -Ed *Back in my youth, the McDonald’s restaurant practice was to “make to stock” the dozen menu items. When you stepped up to the counter, the clerk opened the transaction on his “register”, took the order, and went to the hot table, pulled the order items, returned with the filled order, took the payment and closed the transaction. But when (c. 1975) management installed new “point of sale” machines that clocked elapsed time per transaction per station, the clerk opened the transaction, took the order, accepted payment, closed the transaction, and then went about fulfilling the order. The point-of-sale terminal reported “serving time: 26 seconds”, but the customer did not experience the performance that the management statistics indicated. The meaning of “serving time” – that which the management intended to measure -- had been changed by changing the business practice. I'm ok with <Class> being defined as a set whose set membership criterion does not change -- which is how you have defined it. But I'm not ok with the membership of a <Class> never changing, and with your claim that apparent change is simply a matter of our discovering a new member that we hadn't recognized as a member before. My reason is that even if set membership criteria do not change, those things which make up the universe of discourse for a <Class>/set do change. For example, if a customer of a tobacco company has to be eighteen years old, then every year there will be a new crop of people (the universe of discourse) who are eligible to become customers; and probably some of them will. This isn't a merely apparent change; it is a change in what you called "true membership”.
The way this issue is addressed in ISO 15926/4-dimensionalism is that there is considered to me an all-knowing system which does know all members of a class for all all time and that’s how one identifies the class by its extension in all of space/time. In your local application, you can often only record a specific subset of that full membership at any point in time, but that lack of data locally does not change the fact that the all-knowing system is what is used to identify the class.
Any software engineer/Semantic Web expert would not allow “So the A boss has the distinction changed so that people who have NEVER bought, but who are known by A, are now treated as customers just like the B division mgrs do it.” to occur. In any serious organisation, the change management of the software system containing Customer would reject this proposal. The actual requirement would be analysed and a reasonable change made the application. For many Semantic Web developers, a key characteristic of a “Class” is that their definition does not change. If a change is required with different membership is the desired outcome, then another Class is created that is a superclass of Customer, subclass of Customer or overlaps with Customer. Philosophy and technology cannot actually stop people from doing dumb things, but they do provide guidelines and rules that if followed result in fewer dumb things happening. In an engineering standard ISO 15926 used as OWL in Oil and Gas, classes are defined as follows: <class> is a <thing> that is an understanding of the nature of things and that divides things into those which are members of the class and those which are not according to one or more criteria. The identity of a <class> is ultimately defined by its members. No two classes have the same membership. However, a distinction must be made between a <class> having members, and those members being known, so within an information system the members recorded may change over time, even though the true membership does not change. NOTE 1 The membership of a <class> is unchanging as a result of the spatio-temporal paradigm upon which this schema is based.
and as you can see it specifically stated that classes are unchanging. 4D is hard so not that many modellers follow it rigorously, but the idea that you cannot change the definition of a class is pretty typical. The OWL 2 specification says “Classes can be understood as sets of individuals” and “Class expressions represent sets of individuals by formally specifying conditions on the individuals' properties; individuals satisfying these conditions are said to be instances of the respective class expressions.”, which if used properly would lead to rejecting the change proposed by this “boss”. In Semantic Web languages, the concept of namespace is used to disambiguate cases where people use the same term to mean different things, where Tank (military) vs.Tank (storage) is the example people often provide. As it happens, we use Customer as the example in our training material although the people in our example are more responsible than the “boss” in your example. David I agree, Rich. So since so-called Customer tables will represent different collections of things, albeit all called customers, what are we to do when trying to compare different Customer tables from different databases? Give up? Appeal to Mirriam-Webster? I take it that the whole idea of a Semantic Web is that software can recognize in what respects similar tables are similar but not identical, or by virtue of what identical set of criteria identical tables are identical (whether or not their names are). If this can't be done by software, it has to be done by people, interpreting definitions written down in data dictionaries. Definitions, I will add, that in thirty years plus of IT database modeling and design in the world of commercial IT, I have never found to be anything more than vague glosses on a statement of the criteria that define those tables and that account for their similarities and differences. In this forum, I believe Pat Hayes to be the most informed expert on the Semantic Web that we have. Perhaps we need to hear from an expert like him, or others, rather than additional comments from an opinionated ersatz academic like myself. This is what we need a true Semantic Web for. We can't just look for tables in different databases that have the same name, to know we will be accessing tables that represent the same things. While that statement is true, the name of the table is not as important as its functionality, which is more a conceptual thing about who pays the DB users money, versus who gets paid by the DB users. When I talk to a developer of systems using different tools and equipment than I use, that developer will immediately know what I mean by the "Customer" table. So it's more about inflow, outflow and accumulations within categories. Two tables represent the "same things" if and only if (i) they share a universe of discourse, and (ii) they use identical set membership criteria to partition that universe of discourse into things that aren't represented in those tables and things that are. It's perhaps more philosophically correct to go that way. But in real situations (as you probably know), the software does what the software customer wants it to. It doesn't follow some one-size-fits-all conceptual structure, but it is custom tailored to the owner, the business, the products and services, financing, and all kinds of constraints like that. So a simple, consistent set of definitions would likely cause disaster in many applications. Chief Technology Officer, MetaSemantics Corporation MetaSemantics AT EnglishLogicKernel DOT com For naming database tables, what the English language says is irrelevant or, at best, nothing more than suggestive. The table Customer for division A is whatever division A says it is, and so too for division B. "Customer" is a name that each division gives to a table it has. If they wanted to call the table "XYZ", there is no reason why they can't. The table, and its set membership criteria, are one thing; the name given to it is something else. The tables -- the one for A and the one for B -- if they are relational tables, are sets. As sets, they have set membership criteria which say which, for all the possible persons who might be or become customers, what criteria they have to meet to become customers, and to remain customers. And as I said, across a couple of dozen clients, I have never found any two of them who had Customer tables with exactly the same set membership criteria. And, to look at it from the opposite end of things, suppose A and B did define their tables, in set membership terms, identically, but that A called its table a "Customer" table and B called its table an "Obligated Party" table. The different names wouldn't matter. The two tables would still be identical. This is an example of one of the important things ontologies can do. They can tell us when two identical names are homonyms and, conversely, when two different names are synonyms. If you are worried about the "customer" vs. "prospect" distinction in ordinary English, let me change the example. Let's say that for division A, someone stops being a customer if they haven't purchased anything in the last three years; but for division B, someone stops being a customer if they haven't purchased anything in the last two years. Surely no English dictionary would say that either or both of these rules created an "untruth". And note the most important point. These are differences which make a difference. Even though both tables are given the name "Customer", they mean different things. A SQL Count statement against the two tables, to decide which division had the most customers, could not provide an answer, because the statement will be counting different definitions of "customer". The statement will be counting apples and oranges. Reading up on dictionaries will not solve this problem. First, as I said, dictionaries are irrelevant. Secondly, no ordinary language dictionary would ever provide a definition accurate enough to tell us what rules have to be satisfied for somebody to be represented in any Customer table. This is what we need a true Semantic Web for. We can't just look for tables in different databases that have the same name, to know we will be accessing tables that represent the same things. Two tables represent the "same things" if and only if (i) they share a universe of discourse, and (ii) they use identical set membership criteria to partition that universe of discourse into things that aren't represented in those tables and things that are. I would like to recapitulate your example of the company that has two divisions A and B. B treats all prospects as customers while A distinguishes customers as those who have actually bought something in the past. The boss tells A division mgrs to up the customer count. So the A boss has the distinction changed so that people who have NEVER bought, but who are known by A, are now treated as customers just like the B division mgrs do it. While that change seems very normal and natural for a business to do that in trying to wrap its processes around the tax and risk constraints it has to deal with, it also seems like an untruth, since the English language says Prospects are not the same as Customers. How do you philosophers in the crowd deal with that kind of change of definition into something every business knows is just plain incorrect? Chief Technology Officer, MetaSemantics Corporation MetaSemantics AT EnglishLogicKernel DOT com
|