A Companion to the History of the Book (70 page)

Front-end systems had their origins in the need to automate newspaper composition, especially in the United States, where there was a large domestic market relatively unfettered by union restrictive practices. The first such system was installed by Xylogics at the Daytona Beach
News-Journal
in 1971. (In the UK, restrictive practices prevented the introduction of such systems in newspapers until Eddie Shah’s
Today
in 1986.) Copy could be entered by journalists, and routed to sub-editors and page-layout operators, thereby achieving the goal of single-keystroking. From 1970, keyboards were combined with visual display units (VDUs) to provide feedback to the keyboarder (Seybold 1977: 281). On multi-terminal systems, these entry terminals only had sufficient memory to store the text being worked on at the time: data was held and processed centrally. Front-end systems such as the Miles 33 400 system and the Ferranti CS7 were used in the composition of the
Oxford Shakespeare
and the second edition of the
Oxford English Dictionary
(1989). These systems were not used in the same way as for newspaper production, to capture new keystrokes: in the case of the
OED,
all text editing was done on specially designed editing software, and then transferred to the CS7 on magnetic tape; there was editorial keyboarding by
Oxford Shakespeare
staff, but it was essentially to amend and add formatting codes to existing texts prepared by Trevor Howard-Hill for the Shakespeare concordance project. Essentially, both systems were used for formatting rather for text entry and editing (Luna 1990; Ragg and Luna 2003).

The second edition of the
OED,
like the
Random House Dictionary,
was conceived as a database that would hold the components of the dictionary. Urdang’s compartmentalization of data into fields had been overtaken by the more sophisticated approach of describing a document’s structure. This was called “standard generalized mark-up language” (SGML) and was developed by Charles Goldfarb and others for IBM in the 1960s. SGML described the underlying structure of a document rather than its visual attributes; it delineated the structural elements, and also described the hierarchical relationships between them (Alschuler 1995). For example, the elements of a dictionary entry might be divided into a headword group, a definition group, and an etymology group; each tagged group would then have its component elements tagged: within the headword group would be the headword, variant spelling, and pronunciation. A document type definition (DTD), separate from the specific data, and usable for as many documents as necessary, describes these elements and relationships. While it seeks to separate structure and content from visual implementation, usefully from the typesetting point of view it allows the definition of entity references, codes which represent any character, such as an accented one, not accessible from a normal keyboard, or a combination of characters, such as a fraction. This was an important step in overcoming the mismatch between the limited capabilities of the typewriter-derived computer keyboard with a hundred or so keys, and the relatively wide character sets of traditional systems such as the Monotype.

The composition of the
OED
confirmed the globalization of text composition and book production in this period: data-capture took place in the US, much of the software was written in Canada, system management and text-processing were done in Oxford, typesetting in the north of England, and printing and binding in the US. In comparison with the production of the first edition, as John Trevitt has written, “what was a connected sequence of interdependent activities carried out under one roof has become a set of discrete functions each of which may be performed anywhere in the world” (Steinberg 1996: 239). Data-capture was done manually, optical character recognition still, as in the 1960s, regarded as having an unacceptably high error rate. SGML tagging was added at the keyboarding stage, based on the logical typographic coding of the original edition’s typography.

For supply to the typesetter, both the
OED
and the
Oxford Shakespeare
text were tagged with proprietary codes describing typographic attributes and also more sophisticated text-processing routines, such as conditional text, line numbering, cues for indexing, and extracting headlines. These systems used were efficient at performing multi-column pagination, implementing complex rules about how columns and pages should be broken, and how footnotes and sidenotes could be inserted. The Miles 33 system, for example, was designed to handle the complex typography of parliamentary papers produced by the Stationery Office. But such systems did not support SGML, and all SGML tags in incoming data had to be converted to typographic tags in a “tags-to-typography” specification. Nor was it possible to interact with the system. Tags were added to text, manually or by search-and-replace routines, and then galley files were sent for batch processing into pages. In the case of a long file, or one with many conditional or variable commands, this processing could run overnight. Errors in tagging were only revealed if the resulting page files were incorrect, or if they failed to emerge. At this stage, there was no “what you see is what you get” (WYSIWYG) functionality, and screens represented all text, of whatever style, in generic founts with visible codes, while proofing prior to output on a high-resolution typesetter was carried out using low-resolution printers with generic founts rather than true type designs.

This process was suited to the text-heavy, multi-column, multivolume
OED,
or the two-column, line-numbered
Oxford Shakespeare.
Editorial work concentrated on the construction and correction of the text; design concentrated on describing rules for the combination of repetitive elements and for page make-up. Automatic line-end word division (which in the
OED
was not used to assist justification) could be handled by the front-end, using either algorithms or an exception dictionary; it could also, in some cases, be handled by logic in the final typesetting device. Fourth-generation phototype-setting devices were capable of imaging a whole page at a time using proprietary page description languages (Seybold 1984: 386–91). A scanning (rasterizing) laser built up a whole-page image in horizontal scan lines at a resolution of at least 1,000 lines to the inch. The page could contain type, line artwork, and halftone images. Monotype’s Lasercomp (1976), used for the
OED
and
Oxford Shakespeare,
was the first such machine: it set a broadsheet newspaper page in less than a minute. The more modest Linotron 202 or Compugraphic 8400, using the previous-generation CRT technology, were more likely to be used for straightforward book composition.

The proprietary nature of the software controlling different machines caused problems in the control that designers and production staff could exercise over their suppliers. Computerized phototypesetting systems worked to many different sets of standards: increments could be points, half-points, decipoints, or eighths of a point. Changing output resolution could change point size. Italic could be created by slanting roman type. It was difficult to keep up with the differences in capability between Compugraphic, Linotron, and Monophoto machines. Partly because versions of the typeface were available on all systems, Times New Roman started its second lease of life: originally a superior newspaper typeface, it became the default for computerized composition, a role it holds to this day.

These production changes did not directly affect the author–publisher relationship (unless the complaints about the quality of computerized word division count). That came when the author’s text, keyed on a word-processor, became a direct source of copy. The 39th edition of
Hart’s Rules
(1983) makes no concession to computerized typesetting: the only hint is a mention of the “96-unit system” used on the Lasercomp that had replaced the Monotype hot-metal 18-unit system. The same year saw the first issue of the
Chicago Guide to Preparing Electronic Manuscripts,
followed in 1984 by
Oxford Rules for the Preparation of Text on Microcomputers.
Both described sets of generic codes that authors were asked to use rather than the (then) variety of word-processor commands for bold, italic, and levels of heading.

The
Chicago Guide
also demonstrates how the use of authors’ keystrokes changed the role of the copy-editor, who now had to check on the author’s presentation of the text, as well as prepare it for the typesetter. When all text had to be re-keyed, the imposition of house style imposed relatively little extra cost. Author-supplied text would now have to be put into house style by the author under instruction from a copy-editor, or else by the typesetter, which would reduce the savings promised. The
Guide
suggested that, “because an electronic version of the manuscript already exists, the printout should be marked more as if it were a proof than original manuscript.” It went on to comment that it might not be appropriate to insist on “certain mechanical or arbitrary things that a copyeditor does to conform to house style – things that, if done in some other consistent reasonable way, should not detract from the quality of the writing.” A list of issues such as the use of serial commas and the use of en-rules rather than hyphens in figure extents, as well as grammatical points, followed. The
Guide
continued: “We try to eliminate as candidates for electronic publishing those manuscripts that are likely to cause eventual problems if they are handled electronically: manuscripts whose authors may balk at being asked to enter changes to their prose.”

While the large publishing, front-end driving, high-speed typesetters were indispensable for projects such as the
OED
and the
Oxford Shakespeare,
the need to reduce the cost of composition meant that relatively few formerly integrated typesetting and printing houses survived by adopting this technology. Economic pressures were forcing a division of labor on the industry. In the early 1970s, the IBM Selectric Composer offered a low-cost alternative to the phototypesetter. It could produce camera-ready copy at low cost, and, most importantly, it could be operated by a typist rather than a trained compositor (Steinberg 1996: 221–2). Printers such as OUP experimented with IBM composition in-house; then, against a background of trade union difficulties, encouraged the start-up of small, independent typesetting bureaus, often run by ex-compositors. The development of small, independent typesetters was important because it allowed specialization. The replacement of large hot-metal composing rooms with skills across a wide variety of kinds of book composition (mathematical, scientific, dictionary, foreign-language) was partly achieved through the fragmentation of these individual skills into a range of small suppliers. These suppliers typically started with IBM composition, moved on to direct-entry photo-composition systems, such as the Linotype CRTronic (1979), and finally to desktop publishing. Academically prestigious but economically problematic publications, such as
The New Testament in Greek: The Gospel According to St. Luke
(Clarendon Press, 1984), showed how complex setting that would previously have been handled on the Monotype could be set on the IBM with efficiency, if with considerable reduction in the aesthetics of the typesetting.

The IBM’s range of typefaces and point sizes was severely limited, and any display type had to be set separately. It is interesting to note that the non-Greek text of
St. Luke
was set in IBM’s version of Times New Roman, hardly ever used in OUP’s hot-metal composing rooms, while display type is phototypeset in a version of Baskerville. IBM composition cut the cost of one-off composition but, as we have seen, did not allow for the manipulation or reuse of data. For this, a new generation of desktop machines would have to be developed.

Another niche market was mathematical composition. Hot-metal composition of mathematics was highly expensive because of the handwork required to convert the elements of equations, which could be keyboarded, into displayed equations. Monotype developed its four-line composition system, first for hot metal (1956), then for photocomposition, and similar capabilities were developed for the Linotron 505, but a radically different approach was taken by Donald Knuth, who between 1977 and 1989 developed his own mathematical typesetting system, TeX. TeX allowed mathematicians to do their own typesetting by using a mark-up system that was conceptually half-way between procedural (system- or machine-specific) and structural (e.g. SGML) mark-up. Aimed at scientists who had access to mainframe computers, it achieved device independence partly through the use of a linked fount description system, MetaFont, and its own Computer Modern Roman founts, which were essentially copies of Monotype’s hot-metal mathematical fount designs. The sophistication of the composition features that Knuth included in TeX, such as its ability to optimize justification and page-breaks across a whole document, did not prevent it from being seen as only a solution for niche scientific typesetting, where it has maintained a following. Knuth’s own
Digital Typography
(1999) displayed its strengths, and an unusual example of TeX being used for non-mathematical setting can be seen in John Sutherland’s
Who Betrays Elizabeth Bennet?
(1999), which used PostScript founts rather than Computer Modern Roman. LaTeX, a development of TeX, was easier to use, and found a niche amongst academic authors from the humanities as well as the sciences.

By 1985, many pieces of the puzzle were in place. Phototypesetters such as the Lasercomp could output any combination of type and graphics over a whole newspaper-size page on film or bromide, using their own proprietary raster image processor (RIP). Personal computers with word-processing applications allowed authors to key and save text to floppy disks that could be transferred from one device to another (Steinberg 1996: 228).

The key development that changed all kinds of text composition was the device-independent page-description language. Adobe’s PostScript language (1984) rapidly became dominant. It interpreted the layout of text and images that could now be created, using Adobe’s PageMaker software on an Apple Macintosh desktop computer (1984), so that it could be printed out on a PostScript-enabled Apple desktop LaserWriter (1985). Promoted as “desktop publishing,” this powerful model of device-independent standards and WYSIWYG software, which enabled complex graphic material to be created interactively, became the new norm for all publishing and typesetting activities. The economies of scale of the personal computer market made it inevitable that desktop composition systems would be far cheaper than dedicated typesetting front-ends; the few that survived reconfigured themselves around now-standard desktop components.