Despite our vast capacity to hold information in long term memory; our working memory is extremely limited and becomes overloaded very easily. Greater insight into these problems and some practical ideas about what to do about them comes from the research of John Sweller. Sweller was interested in how teachers could structure their lessons in order to minimise this problem of overload. From the results of numerous experiments, he developed Cognitive Load Theory (CLT) which explains how teachers might manage the ‘load’ they place on working memory and help students learn more readily. The theory divides up the different kinds of loading on working memory:
Intrinsic load represents the inherent difficulty of the material and is related to their levels of element interactivity. This is limited to between 3-5 items. There’s not much we can do about this as teachers (multiplying 5×8 will always be easier than 5x8x3). However, for some materials, it may be possible break material up into simpler sub-components which can be tackled separately at first and recombined later.
Not all material is equally intrinsically difficult. Where materials are related to what David Geary calls ‘biologically primary knowledge’ the load on working memory appears to be greatly reduced. Our brains are adapted to solve complex problems related to survival and reproduction (e.g. reasoning tasks related to social cheating are much easier than formal syllogistic logic).
Another way we can ‘cheat’ working memory limitations is by exploiting the fact that visual and auditory information can be processed simultaneously without creating additional load. For example, Sweller, Van Merrienboer and Paas (1998) report that where material has high intrinsic load, using visual/audio presentations was far more effective than where text and explanation (which both require verbal processing) was used.
Intrinsic load is also reduced where individuals have a strong background of prior knowledge. Familiar information is said to be organised in our long-term memory as a schema – essentially allowing us to work with a sizeable ‘chunk’ of information as if it were one item. By having automatic access to these schemas, it allows us to overcome something of the limitations of our working memory. This is why, for instance, many people argue for the memorisation of multiplication tables. For example, if the student doesn’t have to mentally calculate 5×8 this will reduce the load on working memory and they will find 5x8x3 easier to ‘hold in mind’.
Extraneous load is the load generated by the way that material is presented to the learner. For example, Kirschner, Sweller and Clark (2006) suggest that where the intrinsic load of material is high, presenting new material through minimally-guided activities like problem-solving creates an additional, unhelpful load on working memory. One of the issues is that when faced with a novel problem, students tend to use a ‘processing intensive’ general strategy called means-end analysis in order to find a solution. Sweller, Van Merrienboer and Paas (1998) suggest that ‘goal-free’ problems can avoid this issue by forcing the student to rely upon strategies other than the load-intensive means-end approach. A second strategy to overcome means-end searches discussed in that paper is the use of worked examples as a substitute for solving problems.
Another further source of extraneous load is attention switching. For example, Mousavi, Low, and Sweller (1995) suggest that rather than having labels alongside a diagram – which requires the student to switch attention between the text and the visual image – placing the labels at appropriate locations on the diagram can dramatically facilitate learning. In essence, we should seek to minimise extraneous cognitive load in order to best facilitate learning.
Just taking these two types of cognitive load, the implication might appear to be that eliminating extraneous load and organising instruction so that sub-components of a complex task are automated would be sufficient for the learning of new material.
However, Sweller, Van Merrienboer and Paas (1998) reported that encouraging learners to engage in conscious cognitive processing that is directly relevant to the construction of schemas benefits learning. For example, varying the conditions of practice appears to have beneficial effects upon learning, despite the fact that the presence of that variety would raise the loading on working memory. They called this germane cognitive load.
Germane cognitive load
Van Merrienboer, Kester and Paas (2006) suggest that whilst load reducing methods, such as low variability and explicit guidance and feedback, are effective in producing high retention of the material – that these techniques hinder the transfer of learning. They argue that there is a need to vary the conditions of practice and only give limited guidance and feedback in order to induce germane cognitive load and improve transfer.
It’s tempting to connect this to Robert Bjork’s ideas about ‘desirable difficulties’. Bjork makes the argument that things that make learning ‘easy’ during instruction do not always lead to long-term learning. He argues that by creating conditions which are difficult and appear to impede immediate performance lead to greater long-term retention and better transfer. David Didau summarises the idea like this:
“I love Bjork’s coining, ‘desirable difficulties’ because it gets to the very heart of the counter intuitive nature of learning. It turns out that making it more difficult for students to learn means that they actually learn more!”
There are lots of examples across in psychology where introducing additional difficulty appears to facilitate learning. For example, it has been shown that making font more difficult for the learner to study improves memory performance. Solving anagrams involves more effort than simply copying words, but this additional effort appears to facilitate recall (for easy anagrams at least).
It may seem that there isn’t a problem. Perhaps, Bjork’s desirable difficulties are merely examples of germane cognitive load. However, it does create an issue for the theory – mainly because there’s no easy way to experimentally measure each type of load and this risks making the theory impossible to falsify. Debue and van de Leemput (2014) explain the problem:
“In the absence of reliable measurements for each load, the CLT cannot ever be refuted because it is always possible to attribute variation in the overall cognitive load to a source that corroborates the initial assumptions. For example, assuming that the overall load is kept constant, a decrease in performance will be attributed to a rise in extraneous load that impairs germane cognitive processes. Conversely, if the performance increases it will be attributed to a germane load enhancement made possible by a drop in extraneous load.”
Perhaps the solution is simply to get rid of the notion of germane load. However, it seems that cognitive load theory needs some sort of component which represents the fact that some kinds of mental effort lead to improved long term memory for material. However, unless there’s a way to measure this (and I suggest self-report measure are unlikely to convince critics of the theory), it risks making the theory effectively unfalsifiable.
What is the right sort of mental effort?
The relationship between some sort of mental effort and learning isn’t terribly controversial. For example, most readers of this blog will recognise the quotes below:
Memory is the residue of thought Dan Willingham
But what does thinking or ‘thinking hard’ mean? Is it just the quantity of thinking or some aspect of the quality of thinking which leads to learning?
Well, one way of thinking about the ‘right kind of thinking’ might be to borrow the concept of ‘depth of processing’ first posited by Craik and Tulving (1972). I describe a bit about their ideas in more detail here. In brief, they suggested that mental effort might comprise of more shallow or deeper processing.
“For example, in shallow processing, the subject answered questions concerning the word’s typeface (for example, is the word “HOUSE” written in capital letters?); in intermediate processing, the subject answered questions about rhyme (for example, does the word “house” rhyme with “pencil”?); and in deep processing, the questions were directed toward the word’s semantic content (for example, does the word “house” fit into this sentence: “The ______ has a beautiful window”?).”
They suggested that retention in long-term memory depends on the depth to which new information is analysed. However, they argued that the system stops processing the information once the analysis relevant to the task has been carried out, so if a task merely requires shallow processing of the material then deeper processing will not occur.
A simple way to illustrate this is consider the difference between two fairly common classroom activities – the word search and the crossword – when familiarising students with new terminology. Word search puzzles are a great example of ‘structural processing’ – they can be completed with no understanding of the key words but simply pattern matching the first few letters. Although such an activity might require mental effort (e.g. some of the words are presented as anagrams or are arranged diagonally or backwards in the grid) it’s not the right sort of mental effort for effective learning. A better ‘quiz’ type activity might be to use a crossword – perhaps with the definitions of words as the clues – as at least any mental effort expended will lead students to attend to the deeper, semantic properties of the key terms.
This may help explain why the ‘testing effect’ is a more effective method of encouraging reliable recall than restudying. Testing encourages ‘semantic searches’ in order to retrieve information from long-term memory and that sort of mental effort facilitates future attempts. It’s interesting to note that the testing effect disappears where there is no mental effort involved in the retrieval. A recent study by Endres and Renkl (2015)examined the testing effect under a range of conditions and concluded:
“Overall, our findings on mental effort and non-tested items support the elaborative retrieval hypothesis, including the interpretation of mental effort as an indicator of semantic elaboration.” …
“Our results suggest that testing tasks should be used that require learners to invest substantial mental effort. A more difficult task leads to more elaboration as long as it can be solved (more or less) successfully.”
We can also relate this to the benefits of spacing – the ‘spacing effect’ – where practice is spread over time rather than condensed over a short period. We’ve known since Ebbinghaus that information is lost fairly rapidly from memory – but that reviewing the material periodically (e.g. through a quiz) leads to better recall over time. It seems plausible that the period of delay increases the semantic focused mental effort required to retrieve the information, whereas immediate testing when the information is freshly retained is too effortless to promote much learning.
This might also help explain why varying the conditions of practice (arguably the key component of germane load), whilst more difficult in the short-term tends lead to better long-term recall. When talking about the problems associated with assessment rubrics, Greg Ashman makes the point that students focus exclusively on the elements required by the rubric and ignore the deeper structure related to the problem.
Again, might this be argued to be a problem related to shallow processing. By varying the conditions of practice, the student is encouraged to engage in deeper semantic processing rather than rely upon fairly superficial automatic recall.
However, does recasting germane load as mental effort related to semantic processing solve the problem of measurement? Well, not yet – but as brain imaging becomes cheaper and more available to psychologists, it’s a possibility. There are certainly studies looking for these neural correlates – for example, Otten, Henson and Rugg (2001) report the results of an fMRI study examining the neural correlates of memory encoding.
15 volunteers were presented with a series of 280 words and (depending on a pre-stimulus cue) had to make a decision based on either a semantic process (was it alive) or a non-semantic process (position in the alphabet). Afterwards they were presented with a recognition task, where they had to pick out the words they had seen (mixed in with 140 others they had not). They found there was anatomical overlap in the fMRI scans for semantically and non-semantically processed items, but the non-semantic items appeared to activate a sub-set of the semantically processed ones. They conclude:
“The overlap between regions activated by the depth of processing and deep subsequent memory effects implies the existence of cognitive operations that are engaged differentially both by semantic versus non-semantic processing and by effective versus less effective episodic encoding in a semantic task.”
It’s a small scale study – like many involving neuroimaging – but might it provide a possible way to eventually anchor a concept of germane load by relating it to semantic processing?