composition systems that do provide output of good typographic quality. There are obviously continuing problems of advanced system design, in terms of appropriate comprises between speed and quality.6.12

6.2.1. Line Printers

In general, direct, on-line printouts from information processing systems are made at high speed, but with monocase limitations,6. 6.13 fixed intercharacter and interword spacing, and poor utilization of paper. 6.14 In the case of some chain printers, such as the IBM 1403 6.15 and the Potter Chain Printer,6.16 somewhat larger character sets than the single case set can be accommodated. Then it is to be noted that "use of 2-case print mechanisms on high-speed computer printers for the display of extended chemical character sets has been under joint development by IBM and Chemical Abstracts." (Burger, 1964, p. 2).

In addition, it should be noted that "printing speed for these [chain] printers depends both on the information to be printed and the character arrangement. Since most printing applications use some characters more often than others it may be possible to increase the printing speed by repeating high usage characters on the chain more frequently than low usage ones." (Eichelberger et al., 1968, p. 130). Further, "given character usage statistics we can (1) reliably estimate printing speed for a given character chain arrangement and (2) improve the arrangement so as to increase the average printing speed." (Ibid, p. 136).

There are continuing trends for higher and higher speed of printer output, especially in the area of non-impact printing devices, since, as Becker points out, 1200 lines per minute is "about the maximum speed of a line printer because there is a limit to the firing time available for mechanically operating a print hammer." (Becker, 1967, p. 2).6.17 Advanced hardware techniques for output printing include a thermal imprinting development by NCR for the U.S. Army Electronics Command.6.17a

In a brief 1965 survey of non-impact printers, Webster notes first a facsimile process technique which forms a latent character image on electroconductive paper by pulsing selected wires in a matrix of styli. The visible image is formed by liquid or power toner that adheres only to the image area and is bonded by heat and/or pressure. Representing this approach today are small serial devices such as Motorola's TP-4000, and "high-volume printers such as the Radiation 'Super-Speed' printer-over 31,000 lines per minute".

Planned variations in the TP-4000 teleprinter equipment also illustrate new developments of interest in the multiple access and man-machine interactive systems. "This new teleprinter will . . . provide a single facility for the human operator to make inquiry of a computer at his maximum but plodding rate of 10 to 15 characters per second, and receive his answer back at a rate of 300 characters per second." (De Paris, 1965, p. 49).6.18 Another

example is the development of Miniaturized Techniques for Printing, or Miniprint, by NCR.6.18a On the other hand, higher speeds may result in the loss of visibility for the user of an on-line terminal.6.18b

Nisenoff reports that, as of 1966, "several developments were undertaken to produce ultra-highspeed printing mechanisms of a non-mechanical nature. Radiation, Inc., for example, built an electrostatic printer which required special electrosensitive paper. With the paper moving at nearly 100 inches per second, a maximum printing rate of 60,000 characters per second (500 lines per second or 30,000 lines per minute) was attained. The cost, however, was high for general use of this type of subsystem." (Nisenoff, 1966, p. 1824).6.18c

6.2.2. Photocomposition and Electronic Character Generation Techniques

A major type of non-impact printing technique is represented by the various types of photocomposition equipment. By contrast with impact printers, with other types of non-impact printing, and with hot-lead typecasting machines, the photocomposition techniques involve the transfer of optical images of characters from a negative master to a photographic emulsion, with suitable provisions for illumination, image magnification or reduction, and positioning. Many techniques and configurations of mechanical, optical and electronic devices and developments have been employed for such purposes.

Typographic quality output is achieved in photocomposition techniques by the automatic selection of the right one of several sets of characters, by point size adjustments for the individual character including those necessary for subscripts and superscripts,6.18d by controlled positioning of the character as projected, and by insertion of special symbols and characters. 6.19

The generation, accurate placement, and permanent storage of hundreds or thousands of high-quality symbols per second presently require that opticalphotographic techniques be employed, since the speeds are too great to permit mechanical symbol selection. Photography on film or paper is the currently preferred recording method, although other graphic storage techniques, such as electrostatic, thermographic, thermoplastic, photochromic or video signals on magnetic tape offer alternative recording possibilities.6.20

More than two decades of development in photocomposition techniques (from the Higgonet and Moyroud "Lumitype" and "Photon" patent applications of 1945) have culminated as of today in a wide variety of available devices in a broad range of costs and performance characteristics. At one extreme, we find 6-10 character per second devices. At the other end of the speed spectrum is the Benson Lehner Transdata system with a theoretical text composing speed of 62,500 characters per second, but with a character set (or "vocabulary") limited to less than 100 characters or symbols. The intermediate speed range is bracketed by the Photon

900 series (300-500 cps) and the specified Linotron rates of 1,000-10,000 cps. Hartsuch reports 1967 developments by Photon, Fairchild Graphic Equipment, American Type Founders, Alphanumeric Inc., Harris-Intertype, and K. S. Paul and Associates, among others. 6.20a

Among the electronic character generation techniques noted in Webster's 1965 survey were the A. B. Dick Videograph,6.21 the Stromberg-Carlson Charactron shaped-beam tube,6.22 and "the . . . little-known Xeronic High-Speed Computer Output Printer developed by Rank Data Systems Division

Character formation is by a pair of cathode ray tubes which form the latent image on the charged paper, which is developed by a continuous xerographic system. The 'form' repertoire is held on a strip of film which is superimposed over the frame of variable information." (Webster, 1965, p. 45).

Character generation techniques for high quality as well as high speed composition are employed in the Linotron system,6.23 in RCA developments,6.24 and in the Digiset equipment developed by Dr. Rudolf Hell.6.25 An entry of Bell Laboratories is described by Mathews and Miller (1965) as having the following graphic arts display characteristics: "The unit can produce on a single scope face high quality printing with as much as 200 lines of 350 letters each . . . at speeds of 500 to 5,000 characters/ second. . . . Since the image is described in complete generality digitally, line drawings, mathematical equations, musical scores, and an unlimited number of type fonts can be produced by the same, completely standard, means." (p. 397) More recent developments in high speed character generation for phototypesetting also include the IBM 2680 CRT printer. 6.25a

6.2.3. Computer-Controlled Typographic Composition

Thus, in the past several years, three major, and often closely interrelated, developments in hard copy production technology have gained considerable momentum. These are: the development of techniques for computer-based typesetting, including automatic justification and hyphenation; the further improvement and widespread use of photocomposition techniques, and the development of both off-line and on-line devices for high speed character and symbol generation, including the recent breakthrough in "computer calligraphy", notably, at the U.S. Naval Weapons Laboratory, Dahlgren.6.25b

The automation of the traditional compositor functions is well on its way toward revolutionizing the printing industry. Overviews of this field have been provided by Barnett (1966), by publication of the Proceedings of the two Computer Typesetting Conferences (the Institute of Printing, 1965, 1967), and of a 1965 Symposium sponsored by the Center for Technology and Administration of The American University (Hattery and Bush [eds.], 1965), and by a Symposium held at the National Bureau of Stand

ards (Lee and Worral [eds.], 1968), as well as by the Stevens and Little review (1967).

It is to be noted in particular that "the Government Printing Office was instrumental in the development of computer programs and production techniques which have made it possible to utilize the most recent developments in printing technology and to apply automation to the publication of . . [the] latest edition of the Library of Congress subject heading list. This is the largest single publication which has been produced by the Government Printing Office using computers and photo composition." (L.C. Info. Bull. 25, 612, 1966). Further, there are even computer typesetting procedures designed for publication of computer language specifications, such as that developed for Autocode AB, in Sweden. (von Sydow, 1967).

The field of typographic-quality composition and printing actually comprises several quite different technologies. However, a number of these techniques may be made to converge in the design of a single large-scale system. At one extreme, a skilled, highly trained operator must perform all of the compositor functions (spacing, leading, horizontal and vertical justification, hyphenation, page makeup) required to operate typesetting equipment; at the other extreme, text copy with or without font, style, size, or format specification can be processed with a high degree of automation through all the operations that are necessary to produce type or plates set for printing.

Among the pertinent technologies are: specialpurpose devices or computers to provide justification, re-spacing and hyphenation information automatically; 6.26 photocomposition developments; electronic character generation devices for both on-line and off-line applications, and the programming of general-purpose computers to accomplish the type compositor functions.

Improvements in size of character repertoire and provisions for rapid switching between character sets under automatic control are under active development. Some systems incorporate possiblities for program-controlled, light-pen inserted, or reproduced-from-storage interpolations of graphic material with text, ranging from formatting grids and lines (in order to separate or to emphasize tabular or columnar matter) to the incorporation of reproduced photographs with half-tone fidelity.6.26a Other combinations may permit the intermixture or superpositioning of pre-recorded graphic information (including pictorial data) with that of text or graphics generated effectively on-line.6.26b For example, the new Photon 901 allows variable character density, together with other improvements of the 900 (or Zip), and in addition provides for the insertion of graphics from projected slides.6.26c

Going beyond the direct composition onto the page of two-dimensional symbols, many applications of the future will demand the automatic insertion of previously stored graphic material such as drawings, illustrations and photographs and of

similar material received via communication means such as facsimile. Actually, this type of sophisticated output has been investigated in several quite early projects,6 6.27 and it has been claimed that "there appears to be no real technical barrier to the simultaneous or practically simultaneous compositionimposition of both text and graphics." (Duncan, 1964, p. 135).

The Linotrons and related equipment built for the Air Force will provide for the generation of graphical material, such as line art illustrations, continuous tone illustrations, and screened halftone illustrations in a separate video display system, with optical mixing of the lexical and graphical material.6.28 Beyond such questions are those of "Computer Calligraphy" (Hershey, 1967), including problems of computer-controlled composition of chemical structure diagrams and mathematical formulas, to be considered in a more general discussion of character sets and display terminal design requirements in a later report in this series. As of 1967, then, the computer has demonstrated capabilities for eliminating the need for complex compositor skills on the part of keyboard operators, for providing significant increases in the speed with which type composition functions may be carried out, and for keeping pace with those developments in electronic photocomposers that promise graphic arts quality typesetting at speeds of several thousand characters per second and the superimposition of line art, screened halftones, and continuous tone illustrations.

6.3. Hard Copy Reproduction

It is obvious that direct printout, delayed printout, photocomposition, and symbol and character generation techniques can provide the typical system client with hard-copy versions of the information processed by a system or contained in or retrieved from its files. In other cases, however, the primary system output (whether in Box 5 or Box 14 of Figure 3) is some type of visual display of complete items or their surrogate representations which the user scans. Such displays may also be accompanied by address-locators such that the item in its entirety may be retrieved or reproduced in facsimile from its location in physical storage. In addition, the option of hard-copy or facsimile reproduction may often be desired from the system response and display as such.6.28a

In particular, the provision of hard-copy retrieval options is important in most information selection. and retrieval systems involving storage of documents and data in microforms. While some controversy exists with respect to the extent and urgency of client requirements, user satisfaction considerations would appear to dictate continuing agreement with such commentators as Mikhailov of VINITI in 1962, "the production of a large number of copies directly from microfilm must be made possible..." (1962, p. 51) and of Licklider with respect to an on-line system of the future: "There

should be a way to capture any frame of the display in the form of 'hard copy', and the hard copy should be coded automatically for machine filing, retrieval. and redisplay." (1965, p. 94). It is, however, important to determine the extent to which hard copy options are actually needed, rather than merely desirable 6.29

In terms of technological developments related to hard-copy output options we note various types of electrostatic, magnetic, and smoke photographic methods.6.30 We note further that RADC has awarded Xerox Corp., a contract to build a selective photocopier that will separate printed text from illustrative and pictorial material.6.31 Another example is the Bolt, Beranek and Newman "Symbiont", a computer program system designed to facilitate the use of technical documents by students that includes an automatic note-taking and' copy feature for student-selected portions of text (Bobrow et al., 1966). Thus, photocopying techniques continue to be an important means of system response. However, such techniques are encountering steadily increasing difficulties in terms of coping with the steady growth in volume of demand. There have been complaints, moreover, that continuing development efforts are required in the conversion of microfilm to hard copy, particularly in the areas of speeding up the processes and reducing the cost of copies.6.31a

6.4. Post-Processing Operations and Special Forms of Output

In the case of computer typesetting, photocomposition, and character generation techniques, the' post-processing operations of Box 13, Figure 3, are represented first by formatting processes to provide' horizontal and vertical justification, hyphenation, and the like.

On-line formatting in terms of output through either computer-printer or computer-controlled typesetting devices is typically limited to intra-line. and interlinear spacing control. As we have seen, controlled variation of printing weight may also be, available through repetitive overprinting of the same character in the same spatial position on the output media (e.g., special chain-printer arrangements). In certain ingenious applications even so limited a device as the high-speed line printer has been used to simulate gray-scale variation in twodimensional graphic representations by choice of heavily "inked" character symbols to fill in darker areas and those characters involving relatively less. weight for the lighter areas. With the introduction of electronic generation and display techniques, however, on-line formatting potentialities take on much more versatility.

Other than for formatted printout, output response may be required at several different levels. Postprocessing operations will thus involve re-arrangement, re-formatting and transliteration or translation of the direct results of processing, matching, selection, and/or retrieval operations into forms and into

6.31b

anguages required for subsequent presentation, ransmission, and use to the system clientele." For example, Press and Rogers describe TRACE, which "is a program with extensive capability to edit, reconfigure, and display multivariate data. Press and Rogers, 1967, p. 37),6.31c

In such post-processing operations, for another example, we note a speech recognition requirement: 'The speech-recognition device, after having idenified phonetic stretches on the basis of their perceptual characterisitics, must transform them nto strings of linguistic signs . . ." (Garvin, 1963, . 113). As we have seen in an earlier report, speech compression requirements are directly related to problems of achieving more efficient and nore economical use of communication and data ransmission links.

Postprocessing operations may also be applied In order to provide special forms of output that are lictated for particular potential applications; ransliteration into Braille,6.32 into synthetic or prerecorded spoken messages, and translation from orthographic word records into phonemic form 6.3 or example. Nelson (1965) suggests a relatively simple device for coupling blind users to on-line ext libraries via a Braille display.

Another special form of output for the handicapped is to produce from information processing equipment applied to speech analysis rapid displays of both spectrum and frequency analyses so that a totally deaf child can see the results of his attempts o speak and compare them with desired perormance in order to make rapid adjustments.6.34 'Spelled speech" is another output variant that is

intended to be used in connection with the development of reading aids for the blind.6.35

Post-processing operations will also include, as appropriate, various summarizations and condensations of the information that has been processed or retrieved. For example: "Large-scale data-processing systems have been planned in which the input is in the form of on-the-spot observations in special formats. Retrieval is then required, not just of the observations themselves, but of [inferences] from them, through conditional search, summing, averaging, correlating, finding maxima, reporting sudden changes of state, etc." (Bohnert, 1963, p. 155). Yet, as of today "too many systems are designed to display all the data, and not to display only the data needed for the decisions the system is called upon to make." (Fubini, 1964, p. 2).

Finally we note that system outputs based upon post-processing operations will potentially involve feedback information, such as activity analysis, or the following: "A programming capability is to be provided which, when initiated, will tally the number of requests for specified combinations of data items thereby enabling analysis by operating personnel, leading to refinement of the output formats and output request procedures through experience with the system." (Geddes et al., 1963, p. 23). In particular, output becomes input again in many of the advanced techniques that have been investigated and demonstrated. More expecially, feedback information, whether generated by man or by machine, may control various "threshold" settings as shown in Figure 1 and other figures in this and other reports in the series.

With respect to Box 15 of Figure 3, that of use and evaluation of information processing system products, we will consider many of the current problems and difficulties in more detail in another report in connection with the severe case of evaluaion of information storage, selection, and retrieval ISSR) systems. However, we note here a few examples of current concern with problems of systems evaluation methodology generally.

Basic problems of evaluation methodology relate irst to the size, complexity, and open-endedness of he systems themselves; the number of interdependent variables; the difficulties of defining and solating variables, and the amount of data to be collected and processed. In particular, it is noted that: "Information collected about a very arge system may become obsolete faster than it can be collected." (McCarn, 1965, p. 95).

7.1

It also still appears obvious, as of 1967, that 'comparison and evaluation of systems are still very hazardous, because of the many variables that enter into the design of any particular system and the fact that the design of any system is decided o some extent on specific objectives and somewhat

unique combinations of objectives. It is doubtful that the comparative tests that have been made or that are planned can decide the relative merits of systems for general application. They can only help to decide on the merits of certain particular systems for certain particular applications. Hence there is real danger in drawing general conclusions from experience that is as yet too limited." (Lee and Campbell, 1959, p. 1555).

Current techniques for system evaluation pose many difficult problems and have been severely criticised as to their present inadequacies. For example, to Bryant's complaints about 'lack of well-defined objectives' and 'lack of meaningful models', (1966, presented in 1964) Davis adds the following difficulties: "Uncertainty concerning measures of effectiveness", those of determining system scope, and lack of quantitative evaluation criteria, as well as those of security or proprietary interests and those where disinterest in evaluation results because there are no obvious penalties for poor system design. (Davis, 1965, p. 81).

An obviously difficult problem is that of identifying meaningful evaluation criteria.7.2 Even with obvious