Submitted to HTF7 2 October 1998
Paul Martin
School of Mathematical and Computing Sciences
Victoria University of Wellington
Wellington, New Zealand
Organisational Memory Systems (OMSs) are a research area concerned with providing a way for organisations to capture, manage, and make available to their employees informal and transient knowledge about the organisations’ processes. Organisations without OMSs are destined to repeat their mistakes and waste effort reinventing their wheels. Scholarly Web Sites (SWSs) have been proposed as a general, hypertext-based method for storing, representing concisely, refinding and sharing the information generated and used by scholars and researchers in their day-to-day work. In this paper, we explore the extent to which SWSs function as a kind of OMS and, based on our experience with a prototype SWS, suggest hypertext features that could be usefully incorporated into other OMSs. In particular, we conclude that an emphasis on local rather than global structure is important and that some simple constraints on link placement and link annotation can produce a system that is easier to use for both authors and readers.
Organisational Memory Systems (OMSs) are a research area that aims to enable organisations to capture and exploit knowledge that resides in the collective memories of their employees or members but which is not usually found in any of the formal documents produced by the organisation. Features of this informal or soft knowledge are its method of transmission – primarily by social interaction – its rapid evolution, and its use in the process of making products or carrying out tasks but its absence in the final product or task outcome.
Effective organisational memory is a prerequisite for organisational learning, as described in [Balasubramanian 1995]. An organisation without memory forgets the assumptions, constraints, and design rationale associated with particular tasks or products and is forced to start each new venture from first principles: [Conklin 1993] and [Conklin 1996]. In short, without OMSs the organisation has no way of answering the question, "Hasn’t this been solved before?"
While the term OMS refers to a whole research area, Scholarly Web Sites (SWSs) are proposed as a method for solving the particular problem of turning scholars’ research notebooks into electronic, online resources. In this paper we explore the extent to which SWSs function as a kind of OMS and, based on our experience developing and using a prototype SWS for over a year, we suggest novel aspects of the SWS that could be incorporated into OMSs. In particular, we argue that SWSs – and therefore designs for OMSs – would do better to focus on local structuring of information than on global structuring of information. We also provide practical suggestions for supporting local structure in hypertext.
This paper is organised as follows. In Section 2 we describe the main features of SWSs, our hypertext-based method for implementing online research notebooks, and discuss how SWSs fit within the category of OMSs. In Section 3 we list the key requirements in the literature for effective OMSs so that in Section 4 we can evaluate the SWS method against these requirements. Finally, in Section 5 we present our conclusions and future directions for research.
Scholarly Web Sites were first proposed in [Martin 1998] as a general method for scholars and researchers to put their research notebooks online. The method covers the division of research materials into separate nodes (implemented as Web pages), the presentation and content of the text in hyperlinks, the allowable destinations for links, and the typing of nodes according to whether they are semi-structured or unstructured and whether they are manually or automatically generated. The method is supported by an extensible set of tools for managing links, for inferring new links, and for generating content from external information sources.
The scenario for which SWS were designed is as follows. A SWS is owned by an individual scholar who authors the nodes and creates the links. The SWS is used primarily by the owner but, being part of the WWW, it can also be made visible to anyone else with a Web browser. The owner uses the SWS as a more flexible form of the traditional research notebook: to store ‘basic information’ such as collected data, bibliographic details, biographic details of other scholars, lists of online resources, details of upcoming conferences, and information on research projects. The SWS is also used to store more descriptive material that defines the scholar’s area of research interest: definitions for technical terms, summaries of important concepts, and links into the ‘basic information’. Finally, the SWS is used to store analytic and organisational material, for example drafts of papers, funding proposals, proposals for student projects, and lists of good ideas to chase up.
Given that every successful scholar must store these kinds of information in some storage place, then choosing to store them online, as hypertext, has some significant advantages. First, the electronic format greatly simplifies the refinding of information by browsing or searching. Second, specific items of information can be stored in one place and then cross-referenced (using hyperlinks) as necessary – thus saving the effort of re-entering information and the risk of inconsistencies. Third, provided that the presentation of the hypertext is reasonably consistent, the SWS can be made accessible to other researchers, potential collaborators, and potential graduate students with no additional effort. According to [Schatz 1993] this accessibility is big advantage, allowing a larger number of researchers to keep up to date with the leading edge work and opening research areas up to interdisciplinary collaborations.
In order to assess the extent to which SWSs function as a kind of OMS we list the key requirements for an effective OMS in the next section and then evaluate SWSs against those requirements in the section after that.
From the literature we have identified a list of four key requirements for an effective OMS.
A detailed discussion of each requirement follows.
In order to capture a significant proportion of the organisation’s informal knowledge and in order to ensure that this knowledge is disseminated to those it can benefit, the OMS must be accepted and used widely within the organisation. [Conklin 1993] goes as far as to say that successful integration is the most difficult problem to overcome when building an OMS. Perhaps the best test of integration is to see whether an organisation’s members make more or less use of their OMS when under pressure from deadlines. We also need to bear in mind the fact that many organisations are ‘artefact oriented’ – they reward members who deliver tangible results or products but ignore the quality of process used to derive that result or product.
Capturing and representing informal knowledge is an essential part of the OMS concept, but getting people to write down informal knowledge is not straightforward. Few people have much experience in choosing appropriate formal, rhetorical structures for informal information. Should, for example, knowledge about ‘who really makes decisions round here’ be structured as a technical report, a press release, a training manual, or in some other way? Moreover, informal information is most often communicated during social interaction in which the ‘expert’ communicates directly with his or her audience. Communication via the OMS, however, is indirect and the expert may not know who else will ultimately view the information. As [Ackerman 1994] points out, this situation naturally makes experts nervous about committing their views to an archival format.
Clearly, there is little point in capturing and representing knowledge if that knowledge is not subsequently made accessible to the members of the organisation who could benefit most from it. Making knowledge accessible involves more than simply stockpiling documents and effective dissemination of knowledge is particularly important when tasks are complex or require a number of people to be co-ordinated.
Two necessary conditions for accessible knowledge are: an efficient way of finding information related to some information need, and an efficient way of investigating the ‘context’ of found information in order to establish its relevance and reliability.
While there are significant benefits in having an OMS that captures, represents, and makes accessible knowledge, it is possible to get a further win if we let members of the organisation reuse or re-purpose knowledge in the OMS by allowing them to add to or amend the structure and linking in the OMS. [Schatz 1993], for example, says that allowing users to reassemble knowledge in the way that makes most sense to them is an important way for a community to express its views and value systems.
In the preceding Section 3 we identified and discussed four key requirements for an effective OMS. In this section we evaluate the SWS approach against each of these requirements in turn.
SWSs are authored using standard editors or WYSIWYG editors for HTML, distributed using standard Web servers, and viewed using standard Web browsers. The extensible set of tools is currently implemented in Perl using a small number of freely available Perl modules. Thus, a minimal amount of support is needed to install a SWS and only a modest amount of training need be given to authors and users.
HTML is a relatively undemanding format and it has proven easy to make tools that can translate information in other formats into content for the SWS. As [Schatz 1993] noted, systems for sharing knowledge need to be able to import information from external sources as well as dealing with information that is entered directly. [Conklin 1996] and [Conklin 1993] talk about ‘tapping into’ existing flows as a way of capturing information without burdening the users with additional work.
In order to capture information from external sources, SWSs have ‘interface nodes’. Interface nodes are Web pages that are initially generated from external information sources, but which can then be edited by the author and periodically merged with the external information sources in order to keep the interface nodes up to date. For example, one of the tools in the extensible tool set converts bibliographic details stored in BibTeX format into a Web page that can be included in a SWS. The author can add information to this page – critiques of some papers, perhaps – and then at a later date merge the BibTeX file with the Web page, capturing new entries from the BibTeX file but not losing any of the critiques.
SWSs rely on the expressive capabilities of their authors and do not rigorously impose constraints and checks on user input. Thus there are few opportunities for the tools in the extensible toolkit to output error messages, pop up alert boxes, or display warning messages and, hopefully, few opportunities for users to feel judged and found wanting. The link management tool, for instance, does issue warnings about links that it cannot fix, but these warnings are offered more as flags for jobs ‘to do’ and do not cause the link management tool to abort at the offending line. To some extent, the mere fact that SWSs are implemented as Web pages both reminds the user of the informal nature of the content and means that users are likely to be more tolerant of broken links since they are such a fact of life in the Web at large.
To summarise, SWSs have three features which should help them integrate into the everyday activities of a group: their use of standard technologies and tools means that they are relatively easy to set up and learn to use, their capacity to maintain ‘live’ interfaces with other information sources means that they become a part of the existing information flow, and their unrestrictive nature means that authors can express themselves naturally and without worrying about causing errors.
In Subsection 3.2 we suggested that in order to represent informal knowledge effectively an OMS must overcome two difficulties: it must provide an appropriate formal, rhetorical structure, and it must make ‘experts’ feel safe in committing their knowledge to the OMS.
With regard to the first difficulty, the SWS method includes guidelines concerning appropriate rhetorical structure. These guidelines require the author to divide new information into units with names, to put units that have one of a small set of standard types into the associated ‘directory node’, and to put units that don’t have a standard type into their own ‘content nodes’. In our use of SWSs for online research notebooks we have found the following set of standard types and associated directories to be reasonably robust: a directory for published articles, a directory for conferences and workshops, a directory for journals and publishers, a directory for people (i.e. researchers), and a directory for projects and products. Thus, for example, whenever we discover information about a new project in our area of research we make a new entry in the project directory and place links to that entry as required from the rest of the SWS.
Directory nodes contain lists of objects of the same type ordered by name and are therefore semi-structured nodes. Because of their uniform nature and limited number (six or seven seem sufficient) directory nodes also make effective ‘landmarks’ for users – landmarks are described in [Balasubramanian 1993].
Our experience with the prototype SWS has shown that the rhetorical structure of interlinked directory and content nodes as described above is ‘appropriate’ because it is prescriptive enough to help authors tackle the problem of how to fit new information into an existing Web site without being overly restrictive or complicated to apply and because the resulting structure supports ‘browsing of local structure’, as described in the next Subsection 4.3.
With regard to the second difficulty, we can make no conclusions because the author of a SWS is also the owner and therefore has control over what information is made public.
The SWS approach gives the greatest importance to the problem of making stored knowledge accessible. We believe that there are currently three distinct ways of finding information in a large collection. The first way is by keyword search or by similar technique from the domain of information retrieval. Information retrieval techniques often meet the needs of information seekers very efficiently, particularly when they are lucky in their choice of keyword. However, such techniques are limited in practice by the difficulty of computing the meaning of natural language texts. As [Conklin 1996] puts it, filing documents away in the attic and then throwing a search engine at them is not good enough.
The second way of finding information is by what might be termed ‘browsing of global structure’, meaning that users use their knowledge of the global structure of a Web site to predict where in that space the information they desire is located. This technique can work well, provided that a consistent and permanent global structure can be found for the Web site and provided that the user knows or can learn easily what this structure is. Intranets for organisations often use the managerial structure of the organisation as a basis for the global structure of the intranet, leveraging the users’ familiarity with the managerial structure to help them navigate the intranet hyperspace. Unfortunately, global structures work poorly when there is no widely understood structure that can be leveraged or when information items will not fit naturally in exactly one position in the structure but ‘belong’ in two or more positions.
The third way of finding information is by what might be termed ‘browsing of local structure’, meaning that users rely on local context – what they can see at any instant in their browser window – to choose the next link. The advantage of this way is that users need little training or practice before they can use a Web site. Additionally, authors of material are saved from the significant burden of having to decide where in the global structure a new information item belongs – they simply link it in to anywhere and everywhere that seems appropriate. The disadvantages of this way is that users do not have the reassuring notion of navigating within a fixed space and care must be taken to ensure that all hypertext links provide sufficient context for users to choose between them with accuracy.
In fact, SWSs embrace all three ways of finding information but give primacy to the third way – of browsing local structure. Current implementations of SWSs use the search engine Glimpse [Manber 1993] to provide a simple keyword search facility. They also make limited use of a global structure with the following two constraints. The first constraint is that the familiar structures leveraged in the SWS must be familiar and understood by a wide range of users. For example, the ‘top level’ of structure in our prototype SWS is the set of six directory nodes for articles, conferences, journals, research institutions, researchers, and projects. We explicitly avoid structuring SWSs into, for example, collections of information about adaptive hypertext systems, information about information filtering systems, and information about dynamic hypertext because not all hypertext researchers will agree that the topic of hypertext is best partitioned into these three sub-areas. As [Ackerman 1994] says, "It may be virtually impossible to construct one data or knowledge ontology for a large organisation given the various language worlds that exist." The other constraint is that navigation features based on movement within the global structure are discouraged: there is no ‘home page’ in the SWS and no ‘up’ or ‘next’ buttons.
While authors of SWSs are steered away from the use of global structures, they are encouraged to make more use of local structure. This encouragement is given by the link management tool in the SWS tool kit. This tool converts ‘link hints’, entered by the author, into working links, thus saving the author much of the effort of inserting the HTML tags for the link source and the link destination, of choosing a unique text string for the bookmark at the destination, and of specifying the correct URL at the source. In the terminology of [Ashman 1997], SWSs use pre-computed, hand-made links.
There are, however, two restrictions imposed by the link tool – all hyperlink destinations must be at level 4 headings (as defined by <h4> tags in HTML) – and the text in each hyperlink will be the same text as is found in the level 4 heading at the destination. With link destinations restricted in this way, an author makes a ‘link hint’ simply by enclosing a substring of the level 4 heading at the desired destination in parentheses. For example, if the link tool were used on this document, then it would recognise the link hint "Intro" as the source of a hyperlink to the first section of this document, "1. Introduction" above. Additional functionality in the link tool allows for approximate string matching and enables the tool to suggest alternative destinations in the case that the substring matches more than one level 4 heading.
The restriction that the text in a hyperlink will contain the same text as the heading at the link destination has been imposed partly to ensure that the expanded link hints is meaningful to the readers, partly to help the reader’s eye identify where in the browser window they should start reading from after following a link, partly to ensure that the readers always realises when two links lead to the same destination, and partly to force links into a uniform rhetorical role – that of naming the information at the destination.
One further, important part of the SWS philosophy is that it encourages rich interlinking. We believe that rich interlinking makes finding information by browsing more effective (provided that links are annotated in a way that helps readers choose them with accuracy) because it provides many paths to reach a desired result. For example, a reader interested in a specific project might reach information about that project by first finding a researcher involved in that project, or by finding the institution where the project work was done, or by finding an article describing the project, and so on. Multiple paths are virtually a necessity as long as different users have different mental models of a domain – [Conklin 1996] asserts that the biggest barrier to team work is a lack "of shared understanding, especially about key concepts and terms." Finally, rich interlinking provides the second necessary condition for accessible knowledge – a way for the user to investigate the ‘context’ of found information (by browsing).
SWS support re-purposing of knowledge in two ways. First, with the management tool doing most of the work to create functioning hyperlinks from link hints, the author can concentrate on creating new content without having to be distracted by hacking raw HTML in several files concurrently. Second, the author is relieved of the burden of having to figure out where the new content fits within some global structure – the author simply links the new content in from one or more appropriate locations. Third, it may often prove to be possible to re-purpose knowledge simply by removing old links and adding new links – without having to manually rewrite contents pages, indexes or document summaries.
In Section 2 we summarised the scope and purpose of SWSs. In Section 3 we identified four key requirements for an effective OMS: that it integrate into the everyday activities of the organisation’s members, that it be able to represent informal knowledge, that it be able to make this knowledge accessible to a range of users, and that it support re-purposing of knowledge in the OMS.
In Section 4 we evaluated our experience with a prototype SWS against the four key requirements. This evaluation showed that there was a significant overlap between the scope and purpose of SWSs and the scope and purpose of OMSs. On the basis of this overlap we feel justified in claiming that the following features of SWSs might be usefully incorporated in the design of future OMSs.
The essential difference between the SWSs and OMSs is in the definition of the group from which the memories are collected and redistributed to – with SWSs the group is defined in terms of individuals with a shared interest in a particular research topic, with OMSs the group is defined in terms of employees of an organisation with a shared interest (presumably) in carrying out the tasks and processes of the organisation. SWSs are similar to the ‘electronic community systems’ described by [Schatz 1993].
Clearly, there is still research to be done to refine the SWS method. Our prototype tools are not yet robust enough for general distribution. There are also two important questions we need to explore relating to the use of SWSs as OMS: ‘What happens as the information in a SWSs ages, does it make new information harder to find?’ and ‘Can new knowledge be created by inferring new hyperlinks in a SWS?’
The current literature appears divided on the first question, with several authors citing the Web to support the argument that old information and broken links significantly reduce the utility of hypertext systems. [Conklin 1996], however, draws parallels between OMSs and human memory and then says that the mechanism that allows people to retrieve relevant information from their memories is not dependent on people flushing out old and irrelevant information.
Our experience to date fails to answer the second question conclusively. Our current link management tool inserts ‘reverse links’ – for every link from A to B inserted by the author the tool inserts the reverse link from B to A. Most often these reverse links are useful. Occasionally they are not. For example, every link from an entry in the ‘articles’ directory to the corresponding entry in the ‘journals’ or ‘conferences’ directory causes a reverse link to be created from the journal or conference back to the article. These links, which simply show for each journal or conference every known article published there, we generally find to be so numerous as to be distracting rather than helpful.