My interview of Walter Perry on semantics, parsing, data modeling,object

["Costello, Roger L." <costello@mitre.org>]

Hi Folks,

Recently I had the privilege to interview Walter Perry through a series of email exchanges. Every time I read Walter's writings I feel new connections being made in my brain. His writings are masterpieces and his ideas are mind-blowing. I asked him if it was okay that I share our discussions and he consented. Below is a summary of our discussions. Any errors in the summary are my own.

Roger: Okay Walter, here's the eternal question: does text have semantics?

Walter: A text endures, in its fixed syntactic form. What semantics you might elaborate from processing that text are both ephemeral and utterly contingent upon the circumstances of processing.

Roger: Doesn't parsing give semantics to data?

Walter: The point is that 'given' data can be parsed in differing ways depending on the context (or environment or circumstances) and the purpose (or intent) of the parse.

Roger: Doesn't binary-formatted data give data a specific semantics? 

Walter: This is both why it is a *terrible* idea to ship around binary-format data, or output parse trees, or anything which is the output of a single particular parse rather than potentially the input to any of a range of possible parses. Conversely, this is why XML--as a text format with no pretensions to be an object serialization--is such an excellent interchange format (as Arjun Ray has recently pointed out in his return to the xml-dev list). Good ol' text XML inherently provides common ground (and therefore a common medium of exchange) between different parses (and perhaps the different software in which they are implemented) against the same data but in differing circumstances and with differing intent.

Roger: What about object serialization and data binding tools, don't they yield a well-defined set of, say, Java, objects ... ergo semantics?

Walter: The object serialization guys simply cannot get this:  they want and expect the output of a parsing process against particular data always to be the same result. What could possibly be more brittle than that? Or less useful for exploiting the whole range of usefulness in a given body of data? By contrast, by using XML the *same data* is the *input* to differing processes, which of course render different outputs, each appropriate to its circumstances.

Roger: A 'data model' is always created with a particular processing (semantics) in mind. Right?

Walter: Modelling data is a *process*. In the real world it is done by taking data in the form you have it and re-casting it into a form which is (you hope) best suited to the purpose and circumstances in which you need to use that data. Think of data handling as a pipeline:  you get some data, you model it in a way that most elegantly and succinctly expresses a particular understanding of that data, and then you run it through a chain of subsequent processes which share that model's understanding of that data's structure and logic. Different structures, for different purposes, and therefore different models of the same data are not just possible, but desirable in differing circumstances.

Roger: How does one create an XML document that is not biased toward some particular semantics, that is simply a 'given' for a great variety of processing (semantics)?

Walter: Ultimately, one doesn't, or at least not entirely. But a different user of that same data may well see its usefulness expressed in an entirely different modelling, and will transform the data structure with a 'bias' toward different particular semantics.

Roger: In one sentence, why is data primary?

Walter: The primacy of data is precisely because it is the 'given' upon which all sorts of clever processes have to work.

Roger: Recently I summarized the discussion: Data is Primary. I now realize that my summary isn't the whole story.

Walter: Not the whole story by a long shot. And the glory of  text markup is just how broad and versatile a story it is.

Roger: Thank you very much Walter.

Walter: And thank you.