I spent the first part of this week at a workshop entitled "Contextual Metadata and the Teaching and Learning Context" organised by the MURLLO project, which is funded by the Eduserv Foundation and led by the eLanguages team based in the Centre for Language Study at the University of Southampton.
MURLLO is examining (amongst other things) the significance of "context-rich" metadata in supporting the discovery and selection of learning objects. The project's literature review, by Ann Jeffery, provides an overview of the topic and highlights a tension between, on the one hand, an approach to learning objects which emphasises "de-contextualisation" - a separation between the object and its context(s) of use (intended or actual) - with the intent of facilitating greater reuse, and, on the other, some evidence that the availability of information about the context(s) of use of an object is vital in enabling a potential user to find and choose objects suitable for some learning activity.
Of course this rather begs the question of what we mean by "context", and that was the topic of our opening discussion. And to be honest, while there seemed to be general acceptance that we could distinguish information about a resource which was "contextual" from information which was "context-independent", I'm not sure we really articulated very clearly what we really did mean by "context"! (Having pondered a bit more on the train back to Bath, I think the best definition I could come up with - and I think this would more or less fit retrospectively with our deliberations during the workshop - would be something like, "A set of circumstances in which a learning object is used or may be used.") There seemed to be an acceptance that contexts may be intended/projected/"designed for" (i.e. the provider of a learning object might specify that it is intended or designed to support some particular purpose and/or audience) or they may be actual (a teacher or learner makes use of a learning object in some real-world situation). Ideally there would be some overlap between those two categories!
A number of participants argued that some specification of context is an essential characteristic of a learning object: without some association with a context, the resource is not a learning object. But a single learning object may be deployed within several different contexts, some of which may be anticipated by the creator or provider, some of which may be quite unforeseen at the time the object is created or published.
This fairly open-ended notion of what "context" is perhaps inevitably leads to a somewhat fuzzy view of what might constitute "contextual metadata". It was suggested that any of the following (and this isn't intended to be an exhaustive list!) might be considered contextual metadata:
- information about the purpose or objectives associated with a learning object
- information about the instructional methods associated with a learning object
- ratings and reviews based on the use of the learning object
- structural relationships between the object and other resources, e.g. sequence in a learning design etc.
- data derived from tracking the use of the object (numbers of downloads, time spent reading/using/playing/interacting with the object etc)
- information about the accessibility of the object (I think the contextual element here is probably in the requirements/preferences of different users, or of a single user working in different environments; the characteristics of the resource against which those user preferences are matched are, it seems to me, not context-specific)
- information about the role of the user
- information about "user state" during their use of the object, from their previous learning experiences through to aspects of their physical environment
A full description of context, then, might well involve the description of several different resources and the relationships between them.
The workshop was fairly informal and was structured to emphasise discussion rather than presentations, and with the small number of invited participants, the format worked well - thanks also to some very effective facilitation by Hugh Davis and Dave Millard from Southampton. There were a few short presentations of which I'll mention only two:
- Erik Duval (Katholieke Universiteit Leuven) contributed remotely (initially over a somewhat crackly Skype connection, and then over a considerably clearer - but probably rather more expensive! - mobile phone line) and gave a short description of his current work on attention metadata. The approach focusses on capturing data reflecting what users do with digital resources, both from the logs of server-side applications and from desktop tools. Erik pointed to the successful LastFM service as an example of what can be achieved through such approaches. (LastFM aggregates information provided by its members about the music they play using plug-ins in their desktop MP3 players, and then uses that information as the basis of personal/group histories and a filtering/recommendation system. My LastFM profile is here!) Potentially very large amounts of low-level/fine-grained data can be collected and the analysis of such data might may provide the basis for a better understanding of resource usage and, in turn, enhanced retrieval methods based on inferences drawn from that data. A couple of years ago, I saw Erik give a presentation in which he exhorted us to consider the proposition that "Forms [for the entry of metadata] must die!" and this time he threw out the suggestion that "If you can't automate it, it won't work!" While I'm not sure I would go that far - after all, it does seem that in some contexts, people are motivated to put considerable effort into, say, writing reviews for services such as Amazon or Rate Your Music - I certainly agree that we should use tools to exploit useful data that can be gleaned efficiently from the environment.
- Christoph Richter (University of Hagenberg) summarised his approach to the question of context, illustrating how a single object might be used in the context of different learning activities. Although Christoph didn't have time to expand in detail on the role-based model he had developed, I found his graphics particularly helpful in clarifying some of the distinctions which other participants had been discussing from perhaps slightly different perspectives
Given this broad notion of what "contextual metadata" might cover, it becomes clear that this sort of metadata does not fit well within a framework where "the learning object metadata" is conceived as something which is generated once, typically by the provider or distributor of the learning object (or a cataloguer working on their behalf) as some sort of complete, authoritative and more or less static "document" or "record". On the contrary, this information is, almost by definition, provided from multiple independent and possibly quite diverse sources, using different technological systems, and over an undefined period of time. In such a scenario, the capacity to capture and disclose information about who (or what, in the case of algrithmically inferred data) is providing such information - the provenance of this metadata - becomes significant.
Further, any effective aggregation and merging of this distributed data depends on the consistent use of resource identifiers which are global in scope and persistent through time. It seems to me that the diversity of the information itself - information about activities/events, people, places, etc - and of the potential metadata providers requires a metadata architecture which is designed to support flexibility, extensibility and distributed metadata creation - and indeed Mikael Nilsson and his colleagues wrote about exactly these challenges for learning object metadata some four years ago.
Having said all this, I'm also conscious of Scott's recent note of caution about competencies and "using ontologies and schemas to try to pin down human capability in all its dimensions". One challenge is that of determining/assessing what contextual metadata may be required to deliver useful functions to the learner and teacher, and this was one of the areas that emerged from the workshop as worthy of further investigation, perhaps by using some of the contextual data that the eLanguages team have already accrued in the course of this and related projects to conduct observations of how teachers and learners actually use that data, whether and how it helps them in the discovery and selection process, and what other data might usefully be acquired/provided.
Thanks to Kate Dickens and the MURLLO team for an enjoyable event and a stimulating couple of days.