DOCUMENT PROCESSING RESEARCH
April 17, 1987

Document Processing
and
CRG Systems Research

Dan Bobrow, Ed McCreight (chairs)
Kris Halvorsen, Ron Kaplan, David Levy (PARC/ISL)
Bob Flegal, Bob Stults (PARC/SCL)
Rick Beach, Russ Atkinson, Bob Hagmann (PARC/CSL)
Bob Street (PARC/GSL)
Bob Sprague (PARC/EIL)
Jill Miller (WRC/ISL)

April 17, 1987
Table of contents
§1 Introduction
§2 Visions of document processing
§3 Leverage points for document processing technology
§4 Criteria for evaluation of document processing research
§5 Addressing systems research weaknesses
§6 Conclusion
§1 Introduction
Xerox recently announced to the world that its primary business is document processing. To consider how systems research in CRG relates to this new theme, Ron Rider and John Seely Brown commissioned a committee of senior researchers from the three computer research laboratories at PARC, together with representatives from the other systems"related laboratories in CRG. The committee was charged with producing a report that
f outlines its vision of document processing
f identifies science and technology that provide significant leverage for document processing
f lists criteria for judging the current state and future choice of research areas and projects
f proposes steps to enable Xerox to take better advantage of systems research and technology in pursuing document processing
This report concerns itself with systems activities directly involved with document processing. It should not be construed as minimizing the importance of other research activities. It is difficult but important to maintain an appropriate balance both between document processing research and other activities, and among near term, longer term and basic research. Theoretical activities contribute to our understanding of systems issues. Some applied projects that are not directly related to document processing may nevertheless provide essential platforms for document processing.
A number of important messages are woven through the document. Here are the ravel ends of these threads:
f Document processing is a big thing: it is complex and evolving rapidly and is fundamentally concerned with communication among people. Document processing must move beyond the paper metaphor $ not simply be an extension of the copier business; documents will include new media, at least voice and video. Processing must handle large document bases, and move from concern with form to concern with content. Technology, although important, must be focused towards helping individuals and groups achieve their goals.
f Xerox will be a player, but not the playwright. We must use systems and platforms developed elsewhere, and add value to them. However, Xerox must develop some of its own critical technology even to be a player. Our success hinges on our ability to recognize and exploit leverage points specific to document processing.
f CRG has significant strengths that can help Xerox achieve success in document processing. Xerox needs to build on these strengths. It must encourage and support cooperation between laboratories, seek external alliances where it chooses not to support its own critical mass, and strengthen a few enabling areas where depending on external alliances is inappropriate.
f CRG also needs a stable view of who its customers are and what they need. They in turn must recognize that CRG cannot accomplish technology transfer alone. Resources must be available for taking research concepts, prototypes, and technologies and incrementally incorporating them into products.
§2 Visions of document processing
§2.1 The nature of documents
Documents facilitate a broad range of human activities by organizing and preserving information for future use. As physical encodings of information, they are cultural artifacts, so that their use is as much a function of their physical properties as their information content. They have traditionally been realized as marks on paper, but increasingly, computational technology has been altering their ªlook and feel.º
Is the traditional notion of a document giving way to a new conception? Which aspects of documents are changing in response to the new technology, and which remain unaltered? The answer to these questions is to be found in understanding the role and place of information in document processing.
f Encoding of information
Information must be represented or encoded in a physical substrate; it must be embodied. These encodings vary along three important dimensions: their perceptibility, permanence, and accessibility.
Perceptibility: Some encodings are directly available to human perception, while others are not. An encoding of information that can be directly perceived by humans might be called an external encoding, or a presentation. Examples are speech (an acoustic encoding) and marks on paper. Encodings not directly perceptible (internal encodings) include computer files and human thoughts (which are unavailable to all but the thinker).
Permanence: Orthogonal to perceptibility is the durability of an encoding, which is determined by the characteristics of the physical substrate. Paper can last several hundred years, while vellum can last a thousand or more. Speech, at the other extreme, is evanescent. A substrate capable of preserving an encoding can be thought of as storage medium.
Accessibility: As is well known in computer science, every encoding scheme biases the form and speed of access. This is equally true for perceptible and non"perceptible encodings, for permanent and impermanent ones.
f Transformations of information
To this essentially static account should be added a dynamic notion, that of the transformation of information. A transformation is a process that operates on one encoding to produce another. Both encodings may be embodied in the same type of substrate, but some of the more interesting cases are those in which the substrates are different. Of particular interest are those cases in which an internal encoding is transformed into an external one (the case of access, for example), or the inverse (in document capture).
f Documents as encodings of information
Viewed in this light, the traditional paper document can be seen to have a single encoding, which is both external and (relatively) permanent. The paper substrate constitutes the storage medium. This fact would have remained unremarkable, were it not for the changes brought by computational technology. For with computers came the ability $ in fact the need $ to distinguish two classes of document encodings, external and internal, and the need to transform each into the other.
As a result, the external encoding (e.g. marks on paper) is coming to be viewed as transient and subordinate to the internal encodings (e.g. computer files), which are viewed as permanent [O. North, 1986] (even though the former may in fact have the same absolute permanence as the latter). There are two important consequences of this split. First, it becomes possible to generate multiple external encodings (not just different layouts of the same information, but encodings in different media $ speech as well as text) from the same external encoding. Second, while internal encodings are not directly accessible to humans, certain forms are directly accessible to computers, so document processing is no longer inextricably wedded to the external encoding, nor to processing by humans exclusively.
f Constancy and change
To return to the earlier question, what has changed and what has not? Documents, it seems clear, are still (relatively) permanent, external encodings of information. But these are no longer the ªwhole truth,º for in the computational medium they are created by transformation of an internal encoding. The current emphasis on document recognition is exactly the concern for performing the reverse transformation, a need created by the splitting into two encodings, one ªperceptible,º the other ªpermanent.º
Consider the tradeoffs between paper and electronic documents. Just as paper is the dominant medium where it has advantages, electronic forms of documents will predominate where they have advantages.
f Paper documents
In the recent past, document processing was performed by people working with paper documents. This mode of processing is still the way that most people handle documents. The plain paper segment of this medium has been the basis for Xerox's success.
As a document medium, paper has many advantages over alternative media. Paper is relatively cheap (especially in small quantities), is quite portable, relatively permanent, requires no special equipment to read (or to write in small quantities), has a large base of trained users, and has widely accepted standards for written language. Other media (e.g. voice) also have an independent place in the current culture.
f Electronic documents
Electronic forms can combine elements of many media: visual and auditory, static and dynamic, words and images. Electronic documents currently have a substantial cost advantage in the large. But the primary advantages of electronic documents have to do with processing. With the advent of computers it has become far easier and cheaper to modify documents, transmit them from place to place, search for information, extract and merge information, and perform filing and retrieval operations.
Electronic documents still have disadvantages, and the future of document processing will depend on how some of those disadvantages are overcome. Electronic documents currently require a high investment in equipment relative to paper, they are not yet as permanent, they have less social and legal acceptance, their use requires extra training, and embryonic standards limit interchange among disparate systems.
It is inevitable that electronic and paper documents will coexist for quite some time, since each medium has advantages. In fact, the introduction of electronic document processing is reducing the cost of producing paper documents, thereby extending the advantage of paper in some areas. There will be substantial benefit, and substantial business, in converting documents among media.
f Documents vs. data
How will document processing differ from data processing? The two come from different roots, and have differed quite significantly in the past. While they will continue to differ in emphasis, it seems likely that their roots will become more heavily intertwined, and, in particular, that document processing systems will need to interface to data processing systems. As Paul Allaire noted in his remarks to The Research Board in March, 1987, ª ... the data processing world and the document world must and will be brought together. For corporate investments in information processing to become fully productive, these systems must be able to operate on all the information used by the corporation. And since most of that information is in the form of documents, this means connecting the document base to the MIS [ data processing ] environment.º He also observed that the technology to make this possible exists today, and that products are beginning to appear.
The revolution that is taking place in document processing is based on changing technology. It is now vastly easier to transfer information from one medium to another, and to use new media. To understand the future of document processing it is important to consider not only the information encoded in documents, but also their physical media and the social contexts of their use.
§2.2 Visions of document processing in the future
§2.2.1 The paper metaphor and beyond
Our vision begins with the familiar ªpaper metaphor.º Paper will remain a primary means for presenting documents to human readers, and since electronic reprographic products will displace conventional reprographic products, electronic reprographics is a business that Xerox must be in.
Beyond paper our vision expands the notion of document to include forms of information and techniques of information presentation that do not have paper analogs, and it encompasses more capable systems for processing documents and the information that they contain.
f Capture and creation
Information will come from an increasing variety of sources. Simple capture of keystroked text and mouse motions will be supplemented by scanned text and images, by communications with other document and data processing systems, and by voice, audio, and video input.
f Transformation and modification
Documents will be edited and modified by both people and machines. We expect increased technical assistance in such areas as version history and control, annotations, and coordination of changes in documents with several authors.
f Understanding
Documents will be increasingly understood by machines. This will involve understanding their social context and purposes, as well as the information that they express explicitly. Machines will become capable of automatic translation, intelligent retrieval, extraction and reformulation of document content, and recognition of human"oriented forms of documents.
This will enable a proliferation of intelligent applications packages, for example, to assist in decision making, to streamline authorization procedures, and to improve financial accounting and control.
f Storage and retrieval
Documents will be stored in document bases whose size will dwarf today's data bases. Users will be able to access these document bases using loose descriptions of desired information. Documents will contain machine"understood references to other documents, so that users can navigate around in a web of interconnected documents and inquire how they interrelate.
Multiple storage media will be used, since conversion of all information in an office to any single medium is unlikely. Paper, microfilm, magnetic tape, and optical and magnetic disk storage will all be used. Information on all these media will be coordinated through a single coherent retrieval system.
f Dissemination
Much more information will be available online, both locally and through communications links.
There will be on"demand production of abstracted, annotated, or specially"formatted versions of documents at the point of use.
Automatic routing of documents will become commonplace. For example, a document might be routed to a person based on his own interest profile or as required by office work flow procedures.
Workstations will be linked with the telephone system to allow direct access to and from virtually everywhere. Workstations will incorporate multinational character sets, so that documents can be produced in many languages. Much of the day's mail may be heard in the car on the way to work, via a radio telephone link.
f Other features
Color will be commonplace, both in electronic and paper forms of documents.
Documents will be active, and include such computations as simulations, spreadsheets, animations, and data base queries. Their activity may be triggered by changes in the document base, the passage of time, or other external events.
Electronic documents will become legal entities. Digital signatures will be used for authentication and authorization.
Expanded uses of alternative document forms will change the way we use paper. In many cases documents will be generated, serve their purpose, and be destroyed or archived, without ever existing in paper form. On the other hand, inexpensive printers will greatly increase our ability to produce paper when paper is the most suitable medium.
§2.2.2 Competitive environment and influence on Xerox
We expect a continued high level of competition. There will be no ªsafe'' markets for Xerox; even high volume printing will be competitive. This means that Xerox will have to pay great attention to the requirements of its customers. Many customers will simply want present"day functionality at a much lower price. Others will demand faster systems with broader ranges of applications and functionality. Neither group will have much patience with unreliability or complex user interfaces.
There will be an increasing premium on compatibility with external standards.
There will be rapid changes in technology, and therefore rapid changes in customer requirements. We view technology as the enabler of change in document processing. Technology alters the way we do business since it introduces new methods, and alters the relative costs of old methods. Special attention to new requirements will be essential, since old assumptions will be frequently wrong. This places a premium on flexibility, and will force upward compatibility for archival documents. It will be essential to have short development cycles.
We believe, although we have not substantiated the belief in this report, that profits will come from the value added through better systems integration and better understanding of customer requirements.
§2.2.3 Reflections on the future
It is critical that Xerox develop an understanding of the fundamental parameters of document processing so as to move flexibly in this changing world. Basic research can help develop this understanding.
§2.3 The technological base for our vision
This section lists specific research areas that focus on document processing, including both systems issues and components; it is a statement of the application's requirements rather than a list of specific projects.
f Document system architecture and system design
The technical issues of document processing systems are not dramatically different from those of other large distributed interactive computer based systems. A capable document processing system will support the user through the integration of its individual components / capabilities. It must be modular so that customers can choose subsets of features and extensible so that new object types can be added to the system. The architecture must be open and scalable so that it can connect with a heterogeneous and rapidly changing environment of hardware, software, protocols, standards and languages. The system should also support groups of people working together on a given task in real time.
f Document base
There is general agreement that any document system must support storage and retrieval of the large amounts of heterogeneous and heavily linked data that constitute our notion of documents. The system design/architecture must be able to handle very large objects (digital forms of image, voice and video). Electronic storage will be organized in a performance hierarchy with buffered, immediate, and archival levels. Data will migrate automatically among these levels to improve both reliability and access time.
This document base should support massive amounts of hierarchical file storage, with distribution and migration of information across the internet. It will require DBMS retrieval capabilities, extendible to allow retrieval by content and use of the document, and should support autonomous activity by servers that help users achieve standing goals. As described in much more detail in the Spinrad report on document systems, documents must be active and polystructured, with links from a document to its history, and to other (pieces of) documents. Documents should support nontextual information: images, voice annotations, and video segments.
f Communications
A document processing system must support the communication of documents between system elements and users. A document system requires high performance, high bandwidth point to point connectivity with a network infrastructure supporting the distribution of documents: e. g., authentication and access control, and gateways to alternate communication architectures.
The system should also include standard, easily used mechanisms for supporting document"related communications: electronic mail supporting forms and multiple body types, easily processed language (like Courier or more functional) for describing and implementing communications, and ''smart'' routing of documents between people (by role of the intended receiver as well as by interest in content of the document).
f Document transformation processing
Much of the added value of a document processing system over file storage lies in its ability to carry out or assist transformations of the data form and content. It should be able to hide the boundary between inter"document and intra"document structure, and support compression for storage and retrieval. It might be able to index a document by content and attributes, reason about this content, and use its analysis to create alternate, derived representations for different purposes. Servers on the net will allow intelligent processing that otherwise will be too expensive to implement in individual workstations.
f Human interface
Both the richness and the evolution of a document processing system will be most evident in its support for a variety of representations for the system's data and control mechanisms. An effective document processing system must include sophisticated input and output support for hardcopy, voice and video. Users and developers should be able to extend and customize the interface for specific applications, and for level of expertise of the user.
f Social and anthropological issues
To support its users, a document system must incorporate knowledge of how documents are used and processed, both within the system and by its human users. This requires an understanding of current human document processing work flow and use of models that include users and tasks. Tools must support and enhance productivity and understandability of the activities.
§3 Leverage points for document processing technology
A leverage point is a theory, technique, or technology that amplifies efforts and resources. For CRG it is usually a basic intellectual product that enables many applications.
To achieve our document processing vision we need to identify its theoretical and technological requirements, and to define criteria for selecting areas of research that will enable solutions to a broad range of problems that now prevent those requirements from being met. Our document processing strategy pivot on such leverage points to be successful.
§3.1 Leverage points for document processing technology
We have identified a number of characteristics that will help identify high leverage areas of our research program.
f Advantage
A leverage point is an advantage over competition. If everyone has it then it is not a leverage point but simply current technology. A leverage point must be an area of strength. Where we are not strong, we should try to find available outside technology to support our areas of strength.
f Reusability
A leverage point should be reusable in many contexts. This amortizes its cost of development over many applications. Two properties that make this more likely are modularity and extensibility. A concept or technology has modularity when it can be moved from one environment to another at low cost. A leverage point has extensibility when it can be enhanced easily for different requirements.
f Generality
Leverage often comes from a solution to a general problem that has broad enough applicability to justify the cost of finding it. The value of general theories and architectures is amplified by multiple applications and by the enabling of new classes of applications.
f Selectivity
Being selective in choosing areas to attack is essential for attaining leverage with limited resources. Leverage may come from solving a special case of critical importance. The key is the identification of the right special cases.
f Relevance to document processing
The leverage point must be relevant to document processing, which is to be the main business of Xerox. Advances applicable to general computing (e.g. increasing processor speed) may be left to general computing companies. Xerox must focus on fulcra that support activities in its selected niche.
§3.2 Examples of leverage points
The following paragraphs describe selected leverage points from current or recent research. These take the form of individual projects or fairly narrow research areas.
f Large area electronics
The large area silicon efforts at PARC are both general and selective. Because of the size and resolution compatibility between large area electronic and optoelectronic devices and current paper documents, such devices have natural application to a variety of hardcopy output devices, input scanners, and displays. The ability to design and fabricate such components is thus a leverage point. Selectivity applies to this leverage point by narrowing the device efforts to those specific to document processing, avoiding the enormous resource requirements of a less selective general silicon fabrication. The application of large area techniques to packaging of high speed processors (Dragon) is a spinoff of this technology with potential ramifications for document processing.
f VLSI design & tools
Carefully targeted custom VLSI is an increasingly critical part of the complex electronic subsystems required in document processing. Thus the ability to interact with the technology and to turn new designs around quickly is a leverage point because of its broad applicability to document processing electronic systems. Where we have no advantage, we purchase the services. We buy commercial circuits if they meet our needs, and we buy CMOS wafer fabrication from our mask tapes. Some of the VLSI tools are built by CSL, some by outside vendors. The tools are being used in the Dragon project, the Tamarin project, and for large area electronics design.
f Marking technology
Marking technology is an area of Xerox technical and business strength that is expected to be utilized in any future vision we can conceive of document processing. It is thus a leverage point that is integral to the systems view of our future, as well as to extensions of our current reprographics business.
f Natural language research
Natural language is the primary information carrier in unstructured documents. Powerful document processing capabilities require dealing knowledgeably with the form and content of language, whether as written text or speech. PARC's research in natural language provides a strong theoretical and technological platform for the utilization of linguistic knowledge in document processing. Our capabilities span from the theory of finite state morphology, which has resulted in techniques for efficient analysis and encoding of words, to constraint based theories of syntactic, semantic and phonological analysis. Some systems that illustrate the power of these ideas are TypeRight spelling checker, the Lexical Functional Grammar system as applied to phonological, syntactic and semantic analysis, and a variety of dictionary capabilities based on the Dictionary Server concept.
f Representation, interpretation, and automatic reasoning
Documents are used to convey information between humans and computational agents. They do so by means of representations or encodings. Thus document processing systems must also traffic in encodings. Document processing involves transformations among different encodings $ between, for example, marks on paper, page description languages, semantic representations of their content, representations of their linguistic form, and so on $ , and comparisons of the encodings of different documents (as in retrieval requests). PARC's research on theories of representation and interpretation provides a scientific basis for examining information preserving as well as information altering document transformations, and for the design of representation languages capable of satisfying the multitude of requirements of document processing systems.
Intelligent document processing will require not only the extraction of representations of content from the document but the ability to draw inferences from and reason with this information. PARC's research in artificial intelligence provides Xerox with leading expertise and technology in this area. The Assumption"based Truth Maintenance System (ATMS) is an important enabling technology for a variety of tasks requiring automatic inference. This work applies to forms even better than to free text.
f Foundations of documents
We have been studying the structure of documents (their graphical form, their logical structure) in relation to their content. This study includes work on the structure of written natural language. Research of this kind can be expected to contribute to the design of editors, as well as to the work on document recognition and retrieval. This research has already spawned systems such as Trollope, a structure editor for written English, and System Zero, an editor that allows the user to reparse and reclassify the figural aspects of the document.
f Authoring systems: Notecards
Notecards is a state"of"the"art hypertext system, a document system that supports multiple paths through a set of ideas, not just reading from beginning to end. It has been an important vehicle for the exploration of novel approaches to the structuring, organizing and writing documents. It also allows the exploration of techniques for collaborative authoring, and supports experiments on the organization of text, graphics and active (animated) documents. Notecards and related systems at PARC provide an important stepping stone for experimentation and prototyping of the document systems of the future.
f Expert systems
The process of problem solving, as well as the problem of designing many kinds of complex systems, can be viewed as a knowledge guided constraint satisfaction problem. PARC's research on this approach to expert system building provides a unique basis for the use of knowledge based system technology to deal intelligently with the information in documents. Documents are also the output of expert systems (for example, the PRIDE system for designing paper paths in copiers), and provide a natural human interface to control such expert systems.
f Computing with partial information, uncertainty, and ambiguity
Many document processing capabilities depend on making reasonable interpretations of partially specified, uncertain, or ambiguous information. Capturing sensory inputs perceptible to humans (printed documents, voice and video) and making them available for machine processing, discovering the best set of indexing terms for automatic retrieval of documents, computing likely document distribution routing lists, extracting and reformulating a document's information content $ these and other capabilities require inferences to be made on the basis of partial and possibly erroneous evidence. Constraint languages for different domains, algorithms and specialized machinery for constraint satisfaction are of primary importance for solving these problems.
f Color research
CRG research on color has explored ways of generating, transmitting and reproducing color. Color generation includes both synthetic graphics and scanning systems. CSL has produced state"of"the"art systems to generate synthetic 2D and 3D color graphics. WRC/ISL (Webster and Pasadena) has investigated color scanning systems, raster image compression, and color printing. Both groups have worked with computer tools to modify the color of scanned images once they are online.
Representing color for digital interchange, so that it can be accurately reproduced, requires a device independent specification exhibiting both precision and efficiency; CRG work on the Xerox Color Encoding Standard is addressing this issue. The various groups share a commitment to extending these representational issues to digital input and output devices such as scanners, monitors and printers. In the area of marking technology, WRC/XTL is working on xerographic color printing, WRC/EML works on ink jet color printing and XRCC is exploring color materials.
f Fonts
CRG font work encompasses font design and production systems. CSL and PARC/ISL are exploring a constraint based expert system approach to font design, and PARC was responsible for much early work in spline outline fonts. From WRC/ISL, Typefounder is a comprehensive set of Mesa software tools for creating and managing digital fonts and logos $ critical elements in rendering documents on both workstations and printers. Typefounder is modular, extensible, and embodies a general approach to digital typography that is now being transferred to printer ESSs and Font Servers.
f Cedar Imager
The Cedar Imager is a device independent graphics library. Client programs need not be concerned whether their graphics will be produced on CRT displays or on hardcopy printers of various resolutions. It is a procedural interface to Interpress, and a display screen implementation of Interpress. It is a precursor of the Adobe Illustrator, and the Sun NeWs system.
The Imager exemplifies many of the criteria for leverage points. It is modular and extensible, embodies a general approach to imaging, results from our strengths in device independent graphics, and has numerous applications to document processing. The Imager has already been transferred to development via the Master Blaster project, and is the basis of the Interpress Development Environment.
§4 Criteria for evaluation of document processing research
Evaluating research results makes sense only if the customers for the research activity have been identified. CRG has a variety of (sometimes conflicting) current and potential customers for its document processing business. These customers include:
f The Xerox corporation at large.
f Individual Xerox business units.
f Corporate research.
f External scientific community.
There are four primary types of CRG output: fundamental scientific advances, prototypes of future technologies, specific technologies, and customer support. Each of these is important, and often a leverage area will result in all four types. The following subsections describe each of these types of outputs, their specific customers and their criteria for success in more detail.
§4.1 Fundamental scientific advances
Part of the job of CRG is to increase the body of knowledge within particular areas of specialization. This activity takes the form of inventing new concepts or new linkages related to old concepts.
f The customers for this activity are primarily the external scientific community, the Xerox corporation, and corporate research itself. These customers are interested in either the intrinsic intellectual value of the invention or knowledge, or the corporate advantage achieved by developing or owning it.
f This output takes the form of research publications, patents, internal reports and conference presentations.
f This output is judged by the following criteria: scientific merit as indicated by the quality of the journals, conferences, etc. where results are presented; applicability to such corporate strategic directions as document processing; productive contacts with the external community; and increased potential for successful recruiting.
§4.2 Prototypes of future technologies
Another role of corporate research is to carry out experiments within particular research areas, often through the building of working prototypes that then become the embodiment of a particular concept.
f The customers for this activity are primarily the Xerox business units and corporate research itself. The business unit customers are interested in visions of future technology through presentations or demonstrations of prototype systems, and want some assurance that they will not be surprised by competitive uses of a given concept. The research staff is interested in the ability to discover new concepts through use of more sophisticated tools.
f This output primarily takes the form of prototype implementations, the demonstration of such systems, and reports and presentations describing the prototype.
f This activity is judged by the following criteria: the alignment with strategic business directions; the usefulness of the prototype for other researchers; the synergistic ªfitº with other research activities; the preservation of technology and expertise; the probability of intermediate results; and the identification of generally useful components for possible adoption by others.
§4.3 Specific technologies
Prototype development and basic scientific work may each result in the identification of specific ªnuggetsº that are potentially transferable to the product organizations and other labs.
f The customers for this activity are again the Xerox business units, and corporate research itself. Both are looking for small pieces of technology with high leverage that can be incorporated into their own systems.
f This output primarily takes the form of rather isolated prototype implementations that have very detailed and clear specifications of both the interface behavior and the supporting algorithms or technologies.
f This activity is judged by the following criteria: the alignment with product plans; the usefulness of the output for both the business units and other researchers; standards compatibility; timeliness; and the increase in interoperability through proliferation of common components.
§4.4 Customer support
Each of the different CRG customers will occasionally require some special support. This may consist of editing or writing for a scientific journal, working on a conference or reviewing papers; evaluating competitive technologies; supporting a research tool; contributing to project reviews; or even temporarily moving into a business unit to help transfer a key piece of technology.
f The customers for this activity are all of those listed above: corporate Xerox and its business units, corporate research, the external scientific community.
f This output takes different forms, depending upon the customer and the request.
f This activity is judged by the following criteria: good customer relations; and productive contacts.
§5 Addressing systems research weaknesses
As well as building on its existing strengths, as recommended in section 3, CRG must strengthen itself in some critical areas. Section 4 deals with the criteria that management must consider in maintaining an appropriate balance when faced with limited resources. This section proposes actions to increase the effective use of CRG document processing research. CRG needs to do other research as well, but as stipulated in the introduction, this report deals only with document processing.
A particular characteristic of systems research is that it involves the integration of many elements. We as a company cannot develop all of the necessary document processing technologies, and we will succeed only if we can build effectively on external knowledge and products. We must choose what technology we will develop independently, what we will adapt from external sources, and what we will buy. Section 5.1 discusses how we can improve our make/buy decisions.
Certain broad document related technologies cannot be procured externally and yet they are not encompassed within current CRG strengths. Because these technologies are critical to Xerox' future success in document processing, projects in these areas of previous weakness will have to be developed. Section 5.2 identifies three such areas.
The system requirements for integration inevitably means that we must pay close attention to how different technologies relate together. Synergy between systems projects within CRG is thus vital. Similarly, we must carefully consider how the systems research can be combined and targeted for different customer needs. Sections 5.3 and 5.4 address this at length.
§5.1 Improving our utilization of external technology
§5.1.1 Use of commercial processors and operating systems
CRG has the problem that we don't use external technology as much and as effectively as we should. When faced with a make/buy decision we tend to make rather than buy. Examples have included the use of proprietary processors, operating systems and programming languages.
When PARC was founded there was no established base of computers capable of document processing. PARC led the way in this technology by developing our own languages and designing our own processors. Even today many of our proprietary systems are substantially better than commercially popular systems.
However there has arisen a massive external base of microprocessors, software environments and applications packages that we cannot ignore. The disadvantage of incompatibility with this external base has begun to outweigh the advantages of isolating ourselves within our own proprietary systems.
How PARC should adapt to the new situation is a very complex issue, but one that must be addressed. It has many conflicting aspects:
f PARC has made major investments in its different software environments that would be lost if they were abandoned. On the other hand, established external environments offer a greater opportunity to leverage off others' research, and to test the quality of our own work in others' systems.
f Adoption of standard software and hardware allows us to gain from others' investments, but may involve giving up some technical edges present in our proprietary systems.
f Research in programming languages, implementation, and environments can lead to increased programming productivity. It is important to have the users and developers of systems in close contact. Initial research in these areas may be better pursued on nonstandard platforms.
f Standard software and hardware may make it harder to develop new ideas, but it may make technology transfer easier.
§5.1.2 Recommendations
Given the complexity of the problem and the time available, this committee cannot give a definite solution, but it can offer a guide as to how the problem must be solved. The important steps are as follows:
f CRG must re"evaluate whether its own research needs can be satisfied by the functionality offered by external products and standards.
f CRG must explore how we can effectively connect our own technological base with standard platforms.
f CRG must encourage BPSG to decide which families of platforms and standards will be endorsed and supported within Xerox products. This decision must be relatively stable.
f Based on BPSG's decisions and its own needs, CRG must choose the environments and standards that it will support based on two requirements: the effectiveness of the platform as a research tool (how useful it is to CRG), and the ease of transferring the resulting technology (how easy it is for business units to take advantage of the results).
§5.2 Ensuring critical mass in crucial research areas
§2.2 and §2.3 mention many areas of research and technology that are critical to document processing. §3 stresses the need to emphasize our strengths in providing document leverage, and it lists some of those strengths. This section identifies three broad areas that are critical to document processing success for Xerox, but which are not current research strengths. The committee is not unanimous that these three areas are ªmost important.º Rather, we characterize them as being ªvery important and needing attention.º
The recommendation of the committee is that cross"laboratory technical groups evaluate ways of attaining a competence in each of these broad areas. Decisions might be to set up new research projects, to create alliances with outside companies, or to do a combination of these.
§5.2.1 Systems architecture and design
There is a need for research in combining the pieces of document processing systems into well integrated systems. Document processing today is largely characterized by the desktop publishing metaphor. We now see different media being introduced, sharing of documents increasing, and the set of existing documents growing. There is an increasing need for larger systems that can support a community of users, and interoperability to allow transmission of documents among systems. Even within CRG, better means must be found for passing structured documents among systems and for incorporating externally generated documents.
Issues of interoperability and system architecture to support heterogeneous document processing components that must be addressed include:
f Document interchange format
CRG as a single organization should be able easily to exchange documents that are significantly richer than ASCII text. This exchange requires internal standards that cover both communications and document formats. Should one of the internal formats be used? Should we find and adapt an external standard? Should we try to develop a new standard?
f Document processing systems design
There is significant expertise within CRG in various areas related to document processing systems. How can we combine a powerful, active document base, constraint propagation as a fundamental system paradigm, concerns about parallel processing, and anthropological or social models of document usage in a single high performance system?
§5.2.2 Document base
Several activities related to the storage, retrieval and processing of different types of documents have been carried out within CRG. These include Notecards, Cypress, the SCL Object Server, the FolioPub database, the Dictionary server and attempts to build a large/high performance file server. Currently, there is a strong desire for a document base that can be customized or extended to support the VLSI DA (CAD) group in CSL; software storage for the Cedar Programming Environment in CSL; storage of scanned images; the Object Server/Bank project in SCL; document publication, electronic reprographics and traditional library functions in WRC; plus all the document processing functions described in §2.
We have modest expertise about the pieces of such a system, but we must develop this expertise much further and bring it to bear on the problem of providing a single implementation on which all of CRG (and indeed Xerox) can build.
A cross"laboratory technical and planning group needs to look at this fundamental weakness in document bases. In particular the following issues need to be addressed:
f System specification
What are the requirements that are common to document processing applications? What functionality is required from the document base? We see this study as a major effort.
f System implementation
Having established requirements for the system, what is the best approach to obtaining this functionality? We could start with a clean sheet design and implementation. This approach would involve a great deal of effort simply to reproduce functionality already available in commercial systems. Or we could buy a document storage system like Filenet. This would introduce communications incompatibilities almost immediately. An intermediate approach might involve an alliance with Filenet, Oracle or some other vendor that allowed CRG to modify the vendor's code and thus leverage off of his expertise. Still another approach might simply incorporate such vendor products within a research implementation. What is the best approach, and what should be the implementation environment?
§5.2.3 Document Input Technology
Input scanning has been identified as a technology crucial to Xerox' document processing business. In §2 we emphasized that document creation in the future will be multimedia, certainly involving input scanning, but eventually also voice and other media. Although input scanners and character recognition are now available, the technology is new and needs improvement.
This area is already receiving some attention in various CRG projects. But there is very little being done at the hardware level, and more emphasis is needed on inferring document content, structure and style from raw images. This area is particularly attractive for CRG because it offers the chance to combine advances in hardware and software.
f Document recognition algorithms
Image based document recognition involves transforming bit"mapped text, graphics and images into a higher level encoding that captures the structure, content elements and style that the author intended. The Image Systems organization within DSBU is currently developing scanners, and character recognition is being done at Kurzweil. CRG's limited activity consists of WRC/ISL work on document segmentation, PARC/ISL work on document and character recognition, and PARC/EIL work on the Scrapper image scanner.
f Electronic materials
Electronics materials research is needed to support the present work on printers and scanners at PARC. However, beyond these applications there is also an opportunity to create novel electronic devices especially suited to the requirements of document processing. Large area electronic materials provide an opportunity to implement input and output arrays with the required functionality in terms of speed, size, light output and/or sensing. It may also support highly parallel logic architecture designed, for applications in character recognition and information compression; parallel data transfer techniques; and new memory structures designed around array electronics.
f Voice
We anticipate other means of document input in the future, and voice is the most promising. Speech recognition in CRG is currently being supported in part by a DARPA contract.
§5.3 Improving synergy between labs
Improving synergy is important in making effective use of our research. In addition to developing the individual technologies, systems research requires the combination of technologies so that they work together. Improved communication and collaboration is needed both within the systems laboratories, and between the systems and physics laboratories.
§5.3.1 Recommendations
f Interlab communication needs continuing creativity and encouragement.
f Interlab projects deserve special recognition.
f The OCM should reward labs that provide interlab services.
§5.3.2 Proposed interlab research
The list below is a sampling of areas where interest has been noted in more than one laboratory, and sometimes where cooperative projects have already started. They are listed here to indicate the potential for synergy in encouraging interlab cooperation.
f Related to document base
Object server [ISL, SCL]
Gigabyte storage server [CSL, ISL, SCL, WRC/ISL]
Hierarchical file storage service [CSL, WRC/ISL]
Retrieval service [CSL, ISL, WRC/ISL]
f Related to interoperability
Interpersonal Computing [ ISL, SCL]
Structured document interchange [CSL, ISL, SCL, WRC/ISL]
f Related to document transformation processing
Document recognition at the sensor level [CSL, EIL, ISL]
Document recognition at the structural level [ISL, WRC/ISL]
Video document support [CSL, SCL]
Constraint based editing of text [CSL, ISL]
Additional imaging models [CSL, ISL]
Image analysis and processing tools [CSL, GSL]
f Related to document system architecture
Information processing architecture [CSL, ISL]
New basis for parallel distributed processing [ISL, CSL]
Anthropological or sociological studies for creation and use of documents [ISL, SCL]
§5.4 Improving technology transfer
Although CRG has an admirable record of technology innovation, there is the perception that not enough of this technology has been effectively transferred to the product organizations. Notable examples of frustrated transfer include Interpress and color imaging. Several possible causes of this problem have been identified.
§5.4.1 Causes
f Structural problems
Research, by its very nature, is broad based and neither the applications nor the time scale for success are easy to predict. On the other hand, the product groups are narrow in their focus, and work with short time scales. In order for successful transfer to occur, the inevitable mismatch must be well managed.
f Attitude problems
There has historically been a notable lack of cooperation between CRG and the systems business units, with faults on both sides. In general, within CRG invention is rewarded more than transfer. Moreover, memories of past frustrations are a disincentive to further efforts at transfer. CRG's perception of the business units is that they are much less receptive to new technology than external companies have been.
f Technical problems
CRG systems projects are rarely designed for transfer. In particular the problem of interoperability with outside technology has been largely ignored, and open systems design has not been a requirement. Furthermore the existence of proprietary operating environments that are not supported by the business units means that a significant effort is needed to transfer a technology.
§5.4.2 Recommendations
f Better communications
CRG should be closely involved in technical decisions taken by the business units. This process is vital to reestablishing trust and cooperation. CRG and the business units should work together to ensure that the short range plans of the business units extend naturally toward CRG's longer range vision. Mismatches should be explicitly noted, reasons for the different positions recorded, and this record updated as new information becomes available.
f Planning for technology transfer
The individual labs in CRG should plan and budget technology transfer for near term research projects. This includes committing resources to assist the transfer, and developing technology that will aid in the transfer (such as the creation of internal standards, layered architecture, and prototypes).
Complementary resources must be made available in the business units. There should be standard, timely communication between CRG and business units, and extended temporary reassignment of personnel between organizations.
f Finding other routes for transfer
Xerox must anticipate that a significant fraction of CRG's research will result in opportunities outside its narrow business focus. Management should plan to make profitable use of this research by starting joint ventures, funding entrepreneurial activities, broadening the product range, and other actions.
§6 Conclusion
In this report we have addressed the nature of document processing and its desired impact on CRG systems research activities. In analyzing document processing we have identified three major components:
f the document itself, a cultural and physical artifact having both form (graphical, acoustic, linguistic, etc.) and content (or meaning, what the document is about)
f the human activities, the social practices, that are enabled by documents, as well as the human beings and social organizations that use them
f the systems, computational and otherwise, that enable document operations such as creation, filing, distribution, etc.
In laying out these facets, we note that they are highly and necessarily intertwined. For example, the design of document processing systems must necessarily take into account both the nature of the document and the expected forms of use $ the use of the documents that emanate from the system as well as the use of the system itself. We conclude that research is needed into all three facets.
We recognize that the nature of documents and document processing is in flux and consequently that predictions about future technologies are likely to be unreliable. It is exactly in such a situation that we can expect big payoffs from basic research, to the extent that research into the underlying ªinvariantsº $ such as the presentation and use of information in different encodings $ can keep Xerox light on its feet, allowing it to respond quickly to new technological opportunities based on in"depth understanding; and to the extent that continued research into the changing nature of document processing can help us identify trends and opportunities. In both of these cases, we expect that our research presence can give us a substantial edge over competitors, most of whom can be expected to have only selective competence in certain aspects of document processing.
We feel that an in"depth understanding of the nature of document processing, as outlined in the three facets noted above, is our greatest insurance for the future, is essential to long term success in the document processing arena, and can only be achieved by a substantial commitment to basic research. In particular, we have identified a number of research ªleverage pointsº that by their reusability, generality, or specific applicability can be expected to contribute to Xerox' long term health in a market to which Xerox has just made a long term commitment.
Moral:
In announcing to the world that its theme for the late 1980's is document processing, Xerox has seized a tiger firmly by the tail. There will be exciting times before it is clear that we can harness that tiger to work effectively for us. CRG can provide insights about the general nature of tigers and about which direction this particular tiger is likely to turn, and may produce prototype tiger harnesses of suitable materials. But corporate muscle is needed as well, and nothing less than excellent corporate coordination will succeed. Still, the risks of letting go are likely greater than the risks of holding on: the tiger has noticed us.