XML Production Workflows

This article is not an argument for why you should have an XML workflow. Those exist, as well as excellent arguments for why you might not want one. One of the points made by Booknet Canada's Michael Tamblyn earlier this year[4] is that XML editorial and production workflows do exist—indeed, as SGML, they have done so since the 1980s—but that they tend to be technically or operationally complex and require a considerable amount of programmer support.

O’Reilly, the world’s leading computer publisher, has been making books using SGML and XML for decades. Their famous colophons, which also provide interesting tidbits about the animals that grace their covers, usually give some detail about the toolchains employed to produce the current volume. Here are two of my favourites:

The systems described in these colophons are not for the faint of heart! Presumably, a technical publisher like O'Reilly has access to the technical talent to make such toolchains work for them. Big textbook publishers like Wiley and Pearson have XML toolchains that allow them to manage complex, multi-participant workflows too; they don’t typically publish the gory details on the last page, however.

Practical Challenges

In 2009, there is still no truly likeable—let alone standard—editing and authoring software for XML. For many (myself included), the high-water mark here was Adobe’s FrameMaker, substantially developed by the late 1990s. With no substantial market for it, it is relegated today mostly to the tech writing industry, unavailable for the Mac, and just far enough afield from the kinds of tools we use today that its adoption represents a significant hurdle. And FrameMaker was the best of the breed; most of the other software in decent circulation are programmers’ tools—the sort of things that, as Michael Tamblyn pointed out, encourage editors to drink at their desks.

The labour question represents a stumbling block as well. The skill-sets and mind-sets that effective XML editors need have limited overlap with those needed by literary and more traditional production editors. The need to think of documents as machine-readable databases is not something that comes naturally to folks steeped in literary culture. In combination with the sheer time and effort that rich tagging requires, many publishers simply outsource the tagging to India, drawing a division of labour that spans oceans, to put it mildly.

Once you have XML content, then what do you do with it? How do you produce books from it? Presumably, you need to be able to produce print output as well as digital formats. But while the latter are new enough to be generally XML-friendly (e-book formats being largely XML based, for instance), there aren’t any straightforward, standard ways of moving XML content into the kind of print production environments we are used to seeing. This isn’t to say that there aren’t ways of getting print—even very high-quality print—output from XML, just that most of them involve replacing your prepress staff with Java programmers.

Why does this have to be so hard?

It’s not that XML is new, or immature, or untested. Remember that the basics have been around, and in production, since the early 1980s at least. But we have to take account of a substantial and long-running cultural disconnect between traditional editorial and production processes (the ones most of us know intimately) and the ways computing people have approached things.

Interestingly, this cultural divide looked rather different in the 1970s, when publishers were looking at how to move to digital typesetting. Back then, printers and software developers could speak the same language. But that was before the ascendancy of the Desktop Publishing paradigm, which computerized the publishing industry while at the same time isolating it culturally. Those of us who learned how to do things the Quark way or the Adobe way had little in common with people who programmed databases or document-management systems. Desktop publishing technology isolated us in a smooth, self-contained universe of toolbars, grid lines, and laser proofs.

So, now that the reasons to get with this program, XML, loom large, how can we bridge this long-standing divide?

Is the Web made of Real XML?

At this point some predictable objections can be heard: wait a moment, the Web isn’t really made out of XML; the HTML that makes up most of the Web is at best the bastard child of SGML, and it is far too flaky/unstructured/underpowered to be taken seriously.

We counter by arguing that although HTML on the Web exists in a staggering array of different incarnations, and that the majority of it is indeed an unstructured mess, this does not undermine the general principle that basic, ubiquitous Web technologies can make a solid platform for content management, editorial process, and production workflow.

With the advent of a published XML standard in the late 1990s came the W3C’s adoption of XHTML: the realization of the Web’s native content markup as a proper XML document type. Today, its acceptance is almost ubiquitous, even while the majority of actual content out there may not be strictly conforming. The more important point is that most contemporary Web software, from browsers to authoring tools to content management systems (from blogs to enterprise systems), are capable of working with clean, valid XHTML. Or, to put the argument the other way around, clean, valid XHTML content plays absolutely seamlessly with everything else on the Web.[7]

The objection which follows, then, will be that even if we grant that XHTML is a real XML document type, that it is underpowered for “serious” content because it is almost entirely presentation (formatting) oriented; it lacks any semantic depth. In XHTML, a paragraph is a paragraph is a paragraph, as opposed to a section or an epigraph or a summary. In contrast, more “serious” XML document types like DocBook[8] or DITA-derived schemas[9] are capable of making semantic distinctions about content chunks at a fine level of granularity and with a high degree of specificity.

This may be true, but this is precisely where we should be questioning the cost-benefit relationship of XML more generally. More semantic richness is of course a good thing—much like more apple pie is a good thing—but in order to achieve it you must make that much more of an investment up front. XHTML, though simple, is a mature and fairly comprehensive system for representing the basic building blocks of prose. So there is an argument for recalling the 80:20 rule here. If XHTML can provide 80% of the value with just 20% of the investment, then what exactly is the business case for spending the other 80% to achieve that last 20% of value? We suspect the ratio is actually quite a bit steeper than 80:20 for most publishers.

Furthermore, just to get technical for a moment, XHTML is extensible in a fairly straightforward way, through the common “class” attribute on each element. Web developers have long leveraged this kind of extensibility in the elaboration of “microformats” for semantic-web applications.[10] There is no reason why publishers shouldn’t think to use XHTML’s simple extensibility in a similar way for their own ends.

Finally, the best reason of all to take XHTML seriously is its unparalleled ubiquity. My quick glance at Wikipedia shows more than 180 different XML markup languages,[11] about 175 of which most of us have never heard of—and neither has the software we use daily. XHTML, on the other hand, is supported by a vast array of quotidian software, starting with the ubiquitous Web browser. For this very reason, XHTML is in fact employed as a component part of several more specialized document types (ONIX and ePub among them). Why re-invent a general-purpose prose representation when XHTML already does the job?

It is worth pausing for a moment to consider the role of XHTML in the ePub standard for ebook content. An ePub file is, anatomically, a simply disguised zip archive. Inside the zip archive are a few standard component parts: there are specialized files that declare metadata about the book, and about the format of the book. And then there is the book’s content, represented in XHTML. An ePub book is a Web page in a wrapper. This simple fact needs to be recognized, if only for the sake of demystification; this is not rocket science.

To sum up the general argument: the Web as it already exists presents incredible value to publishers, as a platform for doing XML content management with existing (and often free) tools, and without having to go blindly into the unknown. At this point, we can offer a few design guidelines:

• prefer existing and/or ubiquitous tools over specialized ones wherever possible;
• prefer free software over proprietary systems where possible;
• prefer simple tools controlled and coordinated by human beings over fully automated (and therefore complex) systems;
• play to our strengths: use Web software for storing and managing content, use layout software for layout, and keep editors and production people in charge of their own domains.

Tracing the steps

To begin with, we worked backwards, moving the book content back to clean XHTML. The simplest method for this conversion—and if you want to create Web content, this is an excellent route—was to use Adobe’s “Export to Digital Editions” option, which creates an ePub file. Recall that ePub is just XHTML in a wrapper, so within the ePub file was a relatively clean XHTML document. It was somewhat cleaner (that is, the XHTML tagging was simpler and less cluttered) than InDesign’s other Web-oriented exports, possibly because Digital Editions is a well understood target, compared with somebody’s website.

In order to achieve our target of clean XHTML, we needed to do some editing; the XHTML produced by InDesign’s “Digital Editions” export was presentation-oriented. For instance, bulleted list items were tagged as paragraphs, with a class attribute identifying them as list items. Using the search-and-replace function, we converted such structures to proper XHTML list and list-item elements. Our guiding principle was to make the XHTML as straightforward as possible, not dependent on any particular software to interpret it.

We broke the book’s content into individual chapter files; each chapter could then carry its own basic metadata, and the pages conveniently fit our Web content management system (which is actually just a wiki). We assembled a dynamically generated table of contents for the 12 chapters, and created a cover page. Essentially, the book was entirely Web-based at this point.

When the book chapters are viewed online, they are formatted via a CSS2 stylesheet that defines a main column for content as well as dedicating screen real estate for navigational elements. We then created a second template to render the content for exporting; this was essentially a bare-bones version of the book with no navigation and minimal styling. Pages (or even the entire book) can be exported (via the “Save As...” function in a Web browser) for use in either print production or ebook conversion. At this point, we required no skills beyond those of any decent Web designer.

Integrating with CS4 for Print

Adobe’s IDML language defines elements specific to InDesign; there is nothing in the language that looks remotely like XHTML. So a mechanical transformation step is needed to convert the XHTML content into something InDesign can use. This is not as hard as it might seem. Both XHTML and IDML are composed of straightforward, well-documented structures, and so transformation from one to the other is, as they say, “trivial.” We chose to use XSLT (Extensible Stylesheet Language Transforms) to do the work. XSLT is part of the overall XML specification, and thus is very well supported in a wide variety of tools. Our prototype used a scripting engine called xsltproc, a nearly ubiquitous piece of software that we found already installed as part of Mac OS X (contemporary Linux distributions also have this as a standard tool), though any XSLT processor would work.

Rather than create an entire IDML document, which specifies every facet of an InDesign file (spreads, masters, styles, etc.), we opted for a somewhat simpler route. InCopy, Adobe’s distributed editing extension for InDesign, allows a number of editors to feed stories into a single InDesign layout simultaneously (think of a newspaper production workflow, with different editors working on copy right up to the press deadline).[17] Prior to the release of CS4, working this way would require purchasing separate licences for the InCopy client software. But CS4 allows the creation of InCopy files as XML—in fact, InCopy CS4’s native file format is a subset of IDML called ICML, including only a story’s basic content and stylesheet components. In other words, we don’t need to buy InCopy, because we just replaced it with the Web. Our wiki is now plugged directly into our InDesign layout. It even automatically updates the InDesign document when the content changes. Credit is due at this point to Adobe: this integration is possible because of the open file format in the Creative Suite 4.

We wrote an XSLT transformation script[18] that converted the XHTML content from the Web into an InCopy ICML file. The script itself is less than 500 lines long, and was written and debugged over a period of about a week by amateurs (again, the people named at the start of this article). The script runs in a couple of seconds, and the resulting .icml file can then be “placed” directly into an InDesign template. The ICML file references an InDesign stylesheet, so the template file can be set up with a house-styled layout, master pages, and stylesheet definitions for paragraphs and character ranges.

The result is very simple and easy to use. Our demonstration requires that a production editor run the XSLT transformation script manually, but there is no reason why this couldn’t be built directly into the Web content management system so that exporting the content to print ran the transformation automatically. The resulting file would then be “placed” in InDesign and proofed.

Our project aimed to be a proof of concept, but is robust enough to handle embedded images and reasonably complex tables; in short, we were able to essentially duplicate the production of the entire Book Publishing 1 volume.

It should be noted that the Book Publishing 1 proof-of-concept was artificially complex; we began with a book laid out in InDesign and ended up with a look-alike book laid out in InDesign. But next time—for instance, when we publish Book Publishing 2—we can begin the process with the content on the Web, and keep it there throughout the editorial process. The book’s content could potentially be written and edited entirely online, as Web content, and then automatically poured into an InDesign template at proof time. “Just in time,” as they say.

This represents an entirely new way of thinking of book production. With a Web-first orientation, it makes little sense to think of the book as “in print” or “out of print”—the book is simply available, in the first place online; in the second place in derivative digital formats; and third, but really not much more difficult, in print-ready format, via the usual InDesign CS print production system publishers are already familiar with.

Creating Ebook Files

Creating electronic versions from XHTML source is vastly simpler than trying to generate these out of the existing print process. The ePub version is extremely easy to generate; so is online marketing copy or excerpts for the Web, since the content begins life Web-native.

Since an ePub file is essentially XHTML content in a special wrapper, all that is required is that we properly “wrap” our XHTML content. Ideally, the content in an ePub file is broken into chapters (as ours was) and a table of contents file is generated in order to allow easy navigation within an ebook reader. We used Julian Smart’s free tool eCub[19] to simply and automatically generate the ePub wrapper and the table of contents. The only custom development we did was to create a CSS stylesheet for the ebook so that headings and paragraph indents looked the way we wanted. Starting with XHTML content, creating ePub is almost too easy.