Monthly Archives: February 2011

Automated video summaries – are they the future?

I think it is sad that there would be a demand for this, but there are sincere efforts underway in automatically creating short summary videos from longer A/V content items in order to save people time and more efficiently learn from it.

Different methods exist for producing video surrogates and were tested for learner comprehension. Fast-forwarding (FF) is one more well-known technique, where the video stream is averagely sampled at higher rate and visual playback occurs at higher speed. To get the verbal information to learners, they can watch all captions. Better is the keyword-based summary method (KVSUM), which uses automated keyword extraction, subtitle detection and a summariser module to compress the verbal information. This is mapped back to key frames of the video, so the audience gets the relevant visual parts. Furthermore, the keywords are presented as additional navigation tool in the shape of a tag cloud linked to respective parts of the film.

Other surrogate methods exist, such as audio extraction, subtitling or captions, but this loses all the visual information. Tests on learner comprehension have shown that video surrogates using FF or KVSUM can be effective with as little as 10% of the original video.

This seems to be a symptom rather than a cure for the lack of time and the search for maximal efficiency that’s plaguing our times. Imagine watching Harry Potter in 10 Minutes. Sure you could go for a surrogate ice-cream in the time you saved, but will you enjoy it the same as the full movie?!


Actor Network Theory: a short critique

A bit belated perhaps, but I now got round catching up a bit on readings and activities in CCK11. One of the topics was Actor Network Theory (ANT). I read the books, I saw the movie, but I still don’t get the point.

The only value I see in ANT is as a sociological perspective for the Sciences in how knowledge emerges and is time and context bound. To this I would agree, but, hey, that wasn’t really enlightening. Only that they use obscure terminology to turn it into a black art. There is a distinction between actant and actor, weird talk about ‘translation’, ‘enrollment’ and other things that are not what they sound like.

Actor Network Theory calls everything ‘everything’. This is what for me depreciates it’s novelty or relevance. Determining that everything is connected, whether it’s a human, an inanimate system agent, or a stone (nature), is already well-known. We call this “context”, so instead of creating an artificial construct around it, we could simply say that everything is context-bound. Change the context and the thing itself will change.

Then, ANT talks about black-boxing, which is when a scientific context or knowledge are dealt with as being stable and accepted. This I actually find wholly unsatisfactory especially from a scientific perspective. It simply means that we take some knowledge for granted and no longer question it. I disagree with this attitude, and would perhaps argue that Quantum Physics has shown that Newtonian Physics (which was a black box before) can and needs to be challenged. The same I count true for all other knowledge workers, and indeed for everyone on the planet – including droids.

Summative evaluation of MOOCs

I decided to do this summative view on MOOCs early, so it can still be part of the LAK11 course which now enters its final week. My impressions are based on LAK11 and CCK11.

The good things first: I really benefitted from the course. The keynotes and peer interactions were superbly enriching, thought provoking, and knowledge building. The organisational structure and curriculum topics were well chosen and, together with commitment and enthusiasm, definitely worth imitating (apart from my complaint about the length of the live sessions). So a round of applause is due to George Siemens and his team!

Apart from these personal feelings, what I identified as a clear strength of MOOCs is that it focuses a world-wide community of practice (CoP) on a particular scientific field for a specified periode of time. In this situation, a lot of knowledge is created and shared between people from different cultures, backgrounds, and levels of expertise. A dense cloud of knowledge emerges, and, best of all, it’s preserved as manifestation on the web for later retrieval. This by far surpasses any Google search on the topic or Wikipedia browsing.

The density of the network is a real benefit, as we get much further in our pursuit of knowledge when we pool resources simultaneously around a common task or theme, as opposed to our usual world-wide-web ramblings here and there. The web connects people, but its powers to amplify knowledge are limited to hyperlinks and drawn out conversations. MOOCs have the possibility do achieve more. Density as a dimension may, therefore, well become a future determinant of digital scholarship.

Additionally, a MOOC creates connections. Participants are free to connect from sparsely to intensely, depending on their social preferences. In any case, there is tremendous social currency floating about in the network. In connectivist terms this leads to new strong (or weak) links between participants that might not have found a reason to connect before.

In both these strengths – knowledge building and connecting – lies sustainability, i.e. something that participants can take away and repurpose. Even people who were not part in the original run, will be able to exploit the MOOC as the knowledge and its creation processes are preserved and further continued.

Now the downsides: knowledge from the MOOC is dramatically fragmented, which despite its preservation would pose serious challenges to reassemble it. This fragmentation leaves things at the mercy of search engines. Try looking for LAK11 on twitter in two months time, what will we find? And on Google? With maybe disappearing from the face of the earth, even shared bookmarks may not be save, or anything that’s stored in the cloud.

CCK09 twitter search - older posts unavailable

CCK09 on older posts unavailable!

Another weakness, but not necessarily a weakness of the MOOC, is lack of referenceability. In (traditional) scholarly publishing practices, we want to reference knowledge sources, but as the debate on digital scholarship showed, referencing to the MOOC knowledge cloud is at least difficult.

Then there is the education aspect: MOOCs are restrictive. They are not intended to be, but, in spite of their openness, they depend on connectivity, skills and time that not everyone is guaranteed to have. Apart from obvious technical literacies, they assume self-motivation, evaluation, and learning-to-learn skills. Developing a personal strategy to get you through the amount of connections, platforms and content is not everybody’s cup of tea, neither is sorting the wheat from the chaff. Stephen Downes would argue that there are no wrong things to learn, but there is at least the strong possibility that you’re wasting your time.

Overall, though, MOOCs have much to offer as an educational model, and in a world of diversified learning, I see a rather bright future for them – provided that free committed facilitation can be upheld.

Nodes in the Knowledge Cloud

Here is what I believe happens to knowledge in a connected world. Connected in this context means the natural world – since by nature humans are connected, social beings.

knowledge node

The abstract vision of collective knowledge is represented here as a cloud in analogy to computer networks, although this misrepresents the collection of connected nodes, some of which are closer, others are further afield (both in a location sense, as well as in a psychological one).

Knowledge taken from the cloud is digested, which means it is not simply engraved and stored, but filtered and manipulated. Parts of the knowledge we consume consciously or unconsciously disappears before we may feed it back to the cloud, possibly enhanced with our very personal additional meaning or experiences.

My poor memory – knowledge lost in space

In a most unscientific way, I tried to analyse my memory and knowledge retrieval powers, when combating distinct signals of growing forgetfulness. This happens at varying degrees in different areas as I found out. So, I reflected at what I find relatively easy to remember and what not, which led me to the following types of memory I seem to have at my disposal:

(1) Factual memory: Remember names, facts and figures. I’m not very good at that and getting worse. Recalling the author and title of an article I read last week I find difficult. It includes these moments, when I just know that I knew, but somehow it got misplaced in the back of my brain, only to be found hours or even days later.

(2) Visual memory: Faces, images, movies, colours. This is often passive, i.e. I recognise that I have seen that face or photo before, but may find it difficult to call it up on demand.

(3) Location memory: Places, spaces, situations, navigation. That’s a clear strength of mine. I might not remember the name of the place, but I can easily find it again. Similarly, when searching a document for something, I will direct my scan to the place on the page I remember having seen the item before. I recall situations much easier when locatable in space: e.g. the content of an exhibition, when thinking of the museum, former colleagues when locating them in their office surroundings, names of students when recalling the seating order.

(4) Time memory: often time and location are put together into a spacial entity, but my memo banks treat it quite differently. It’s hard to recall what I did yesterday or even this morning, if this is contained in a ‘usual’ location, e.g. home or office. When parking my car in a new surrounding, I will always find it; however, when I park it in a similar spot on the university car park, I need to go looking. So it seems that location + same location = no location, i.e. that items lose their reference cue that otherwise helps me.

(5) Emotional/logical memory: I might not remember the arguments in a scholarly article, but I subconsciously remember whether they were any good or bad, convincing or not. Together with other impressions, it is formative of my world views and opinions. In learning, I often refer to this type of sense-making memory.

Clearly, all these different types of recall are connected to a greater or lesser extent. Part of a successful learning strategy is to start with the strong points and to scaffold the weaknesses (in my case this means attaching location data to the knowledge item). Rather than memorising all those stimulating twitter/facebook posts, they become a formative part of my memory, views and perspectives. I guess you can call it constructivist knowledge.

What's in the data?

Data is supposedly neutral, statisticians aren’t. Of course this is wrong, since data collection is carefully designed, hence sensors are put where data designers want to put their emphasis. Many times also concrete expectations rather than just curiosity are included in the design.

So, we might say that interpretation (and manipulation) of data starts before the data is actually there. Reliability of learning and knowledge analytics therefore depend on how much one trusts in the design, the implementation and the analysis of data gathering. This skews the knowledge we can deduct from such data. It’s like looking at the world through a drinking glass – some parts are magnified, but the edges are blurred and disappear into nothingness.

Setting up data sensors is one thing, presenting the message of the data quite another. Infographs are wonderful ways to get messages across. Some are interactive, like this PISA graph, others are not, but like kiddy sweets in the supermarket, they all impress with a colourful exterior.

Clearly, the more colourful the wrapping the tastier the information, or is it? As the data world grows, – and I won’t deny that it has great appeal and holds great and important insights – we may see a requirement for a new core skill – critical data reflection.

Infographs are something of a black box, nice to look at, quick to pick up, but difficult to evaluate. Instead of just relying on trust, we need to ask about the design of the data gathering and analysis processes. Equipping students with skills to do so, is just the next step up from evaluating Google search results versus blind trust. But this step has yet to be taken.

The webconferencing challenge of the MOOC

People argue whether multi-tasking is a myth or not. All I am certain of is that I’m no good at it. From what I gather in discussions like this, most other people feel the same way.

Our live MOOC presentations in CCK11 and LAK11 contain splendid food for thought, but I only get some junks of knowledge and find it tiring and cognitively challenging. Firstly, the live sessions are way too long to sit and stare at a screen. There have been plenty of studies on concentration span in web conferencing and I suggest cutting the session to half an hour max.

Secondly, there is the powerpoint presentation and the presenter’s talk. Interesting as they both may be, sometimes I find myself caught between reading and listening. For lack of other visual cues that are available in face-to-face environments, such as the presenter pointing at things or gesturing, stricter rules for presentation slides should be adhered to. Most unhelpful in this respect are entire paragraphs of text commented with the message “I won’t read this to you,…” and then talking across it, while the audience tries to determine why this slide is there in the first place.

Thirdly, the web conferencing environment offers wonderful add-ons, like text chat. In other cases, there are also whiteboards, private messaging, file sharing facilities, and more. It requires constant scanning to see what’s going on. This distracts greatly from the keynote. Text chat is good for input to the talk, e.g. to ask a question or make a constructive comment. But it is easily abused for “Hi Fred, long time no see” exchanges, or comments like “good point”. In the web conference sessions that I participated in, I enjoyed direct interaction with other participants, but at the same time, I realised that the keynote then faded from my consciousness.

Switch-tasking between the different information streams, I find entirely unsatisfactory, giving me the feeling I’m not getting the best of two worlds, but nothing of any. It exhausts me nevertheless to glue the bits and pieces I pick up together and to fill in the gaps. Frankly, I also have no intention to re-visit the session for another hour later.

One possibility would be a moderated chat, which goes into a queue to be released to the audience only by a moderator person. Obviously, this would offend some people posting comments that won’t make their way onto the stage, but may still find general agreement in easing the cognitive load of the session.

Data – a privilege not a right

In week 5 of LAK11 (Learning and Knowledge Analytics) Linda Baer gave an interesting presentation on organisational implementation of analytics. She presented various examples of institutional and governmental initiatives (US) in learning analytics covering the school sector as well as private initiatives like the Gates foundation.

Data of and about learners, of course, are owned by the institution (or government organisation). Not only that, the questions about success of learners are asked by the very same bodies. This is only natural, and a good deal of examples were given to illustrate how this may help improve the orientation of new students, reduce drop-outs, or guide students to success.

However, top level management typically does not deal with pedagogies, but with budgets and funding streams. It is therefore most likely that analytics will be expected to provide insights into anything and everything to do with financial success of the institution, not only with what’s best for the student.

The other reservation I have is that students maybe deluded and manipulated through data presentation. Data is power, and to see your own data so far is considered a privilege not a right. Just like trust is the main usage driver behind Google search results, learners will have to trust the analytic visualisation of their behaviour by the institution. Because data can be highly persuasive, the effect may be that we become undiscerning of what we see. This may lead to students taking less and less responsibility for their own learning, and instead passing this to the institutions that hold the power of analytics.

Note that I don’t accuse institutions of deliberately producing wrong data sets in order to manipulate students, but that there are fundamental differences in interests and therefore in the criteria and metrics put to the analytic processes.