Cognitive Load Theory

Cognitive Load Theory (CLT) can provide us with insights about more general learning and also has a high relevancy: "Our limited processing capacity is one of the most important and well known of our cognitive characteristics." (Sweller, 1994, p. 310). This means that a major constraint, when dealing with L2 acquisition, is our processing capacity, more specifically our short term memory. CLT attempts to explain how different factors explain the varying difficulty of learning new information, from extremely easy to almost impossible (Sweller, 1994, p. 295).

Elements and schemas

The required information, that needs to be learned, can be broken down into pieces called elements. Elements cannot be derived from the information directly, instead they depend on pre-existing knowledge and so are different and individual to each user. A collection of related elements is organized into a cognitive schema, and a schema can be viewed as a cluster of information. Schemas can be linked to other schemas and subsequently consist of other schemas (an element can be a schema). This linking especially happens, when new information is contrasted with familiar schemas (Gobet et al., 2001; Sweller, 1994).

An example for a schema, that Sweller provides, could be a tree. Trees are composed of related elements such as leaves, a wooden trunk and branches. Together these elements form a tree. No tree is exactly like another, so there is an infinitive variation. However, the schema 'tree' applies to all of them, and a user with this cognitive schema has no problem applying it and recognizing a tree as one. Elements of the tree schema could also be schemas. For example, there might be the schema for a leaf. Learning or extending a leaf schema might also influence the tree schema, and the other way around. Thus, learning new things or information means acquisition of schemas (Sweller, 1994, p. 310).

Memory

Users have a short term working memory for processing new information and a long term memory that provides the user with access to a vast amount of knowledge (Gobet et al., 2001). The role of the long term memory is to contain an amount of knowledge, of which the limit is not known, to provide the user with relevant knowledge on his surroundings and dealings with life. Sweller compares the long term memory with DNA: It is difficult to modify (only through random mutation and breeding), but it holds all the information needed for survival. Conversely, irrelevant knowledge is unwanted in the long term memory.

This leads to the role of the short term memory. Its role is to filter out the immediate information to only allow a transfer of useful and relevant information into the long term memory. The useful information in the long term memory should be stable, adding unnecessary things randomly would make it unstable. To do this, its capacity is extremely limited, which only allows relevant information. If its capacity was higher, more irrelevant information could be transferred, which would increase chaos in the long term memory and make it less stable (Van Merriënboer & Sweller, 2005, p. 153 - 154). Schema acquisition is subsequently limited by the capacity of the short term memory. Before a schema is in the long term memory, using the schema is very cognitively taxing. Once a schema is acquired in the long term memory, usage becomes easy. This enables acquisition of additional schemas that before were maybe too difficult. Schemas can therefore also become elements or sub-schemas of other schemas, which further explains why elements and schemas are different for every user: They are modified and change over time. In this sense, a schema is a building block on which can be build on. If, however, the short term memory is overloaded, learning the information becomes impossible and the overload may even be harmful for the learning process (expressed in the acquisition of schemas). CLT deals with avoiding such an overload and finding efficient ways of schema acquisition (Van Merriënboer & Sweller, 2005, p. 148).

Transfer from novice to expert

The acquisition of a schema is, however, not binary. Instead, there is a transfer from controlled and conscious usage of a schema, which has a high cognitive load, to the automatic usage, which has a low load. This transfer process, while being very slow, is key to developing competency (Sweller, 1994, p. 297f). In Sweller's words: "This process of automation is the second major learning mechanism after schema acquisition and affects everything learned, including schemas themselves." (Sweller, 1994, p. 298) - schema automation plays a major role in learning.

Variance of difficulty in schema acquisition

There are three types of cognitive load, which are additive: intrinsic, extraneous and germane. To enable schema acquisition, the total cognitive load, consistent of these three sub-types, should be below the short term memory capacity. If the cognitive load is too high or too low, it will impede learning or even contribute to memory decay. Instead, it should be manageable (Paas, Renkl & Sweller, 2003; Paas, Renkl & Sweller, 2004).

Intrinsic cognitive load

Intrinsic cognitive load refers to the intrinsic difficulty or complexity of a schema. This difficulty is inherent to the schema and cannot really be modified, apart from reducing the complexity. The difficulty is determined by how dependent and interactive (or in other words how interdependent) the elements within the schema are and thus by extension if the elements can be learned incrementally, in isolation, one by one or if all elements and how they interact with each other, have to be learned at once (Paas, Renkl & Sweller, 2003; Paas, Renkl & Sweller, 2004). High interactivity leads to a high cognitive load, and inversely a low interactivity between the elements leads to a low cognitive load, which makes schema acquisition easier. If the interactivity is too high for the user to acquire the schema, the only thing that can be done, is to artificially reduce the interactivity. The resulting schema will be incorrect. However, once it is acquired, the correct schema can be learned (Paas, Renkl & Sweller, 2003; Paas, Renkl & Sweller, 2004; Valcke, 2002). In relation to the schema interactivity of L2 acquisition, Sweller notes that, while the learning of vocabulary is of low interactivity, the semantic meaning of words in relation to their order when in a sequence has high interactivity, because the words interact with each other (Sweller, 1994, p. 304). Moreover, it could be easier to learn only nouns, because they are less dependent on each other, while in contrast, learning a new noun and a new verb might increase intrinsic load, because the noun interacts with the verb (Sweller, 1994, p. 307).

Extraneous and germane cognitive load

In contrast to intrinsic cognitive load, both extraneous and germane cognitive load stem from the presentation and learning activities required from the user and can, therefore, also be directly altered. Any cognitive load imposed by the system, that doesn't result in schema construction and automation, is extraneous and conversely any cognitive load, that does result in schema construction and automation is germane.

Since schemas vary and change based on the user and his knowledge, the cognitive load, that might be germane for a novice, can be extraneous for an expert: "As learner expertise increases, the optimal instructional procedures alter. The types of tasks presented to novices should differ from those presented to more knowledgeable learner" (Paas, Renkl & Sweller, 2004, p. 7). While the extraneous load should be minimized, the germane load should be maximized to a manageable level for optimized learning (Paas, Renkl & Sweller, 2004). Moreover, Sweller notes that the instructional design that is employed in teaching commonly assumes a level of cognitive load in their users, that far exceeds the mental capacity, that humans have (Sweller, 1994, p. 299). Finally, as another study found, classroom anxiety also increases cognitive load and decreases memory effectiveness (Chen & Chang, 2018), so to improve schemas, anxiety should be minimized. The implication thus is that, for one, the teaching system should consider the composition of cognitive load (Sweller, 1994, p. 308), but also that this consideration is not trivial, because the composition is different for each user.

Sweller argues that, the learning system composition is highly relevant for the course designer, but irrelevant to the user. This is because the user only cares about the total sum of his cognitive load and not what this sum consists of (Sweller, 1994, p. 308). However, this might be inaccurate: A more or less advanced user might recognize a teaching or presentation style, that has a high extraneous load. In effect, the user might be frustrated, which would impede his learning success and might reduce his motivation, especially when the user cannot avoid this because, for example, he is stuck in class with such a teacher. In addition to this, Sweller also states, that with low intrinsic load, the role of the extraneous cognitive load becomes far less relevant. If the schema is simple, it does not matter if the extraneous load is high (Van Merriënboer & Sweller, 2005, p. 156). Here we can argue that, while this might be accurate for learning one schema, there will still be some amount of cognitive load. This will limit learning if the user wants to learn more than just one schema. So keeping the cognitive load low in general will be beneficial when trying to maximize the amount of schemas learned. It follows that in this case, even with low intrinsic cognitive load, the extraneous cognitive load should still be minimized. However, while a low cognitive load might be desirable, there is a trade off: As Paas states, to optimize learning, the germane load should be high, while maintaining a manageable degree. In relation to memory retention of vocabs, this is supported by Laufer who found that retention benefited by "task-induced involvement load" (Hulstijn & Laufer, 2001).

Spaced Repetition Systems

A method for schema automation is through a spaced repetition system (SRS). There are many different SRSs, and from the literature it is not clear, if there is a clear superior one. SRSs attempt to represent how elements are stored in long term memory and how they will decay over time, if left unused. Different modifications aim on improving the accuracy in different ways, but the basics (with language learning in mind) of all SRSs go the following way: There are different levels of memorization, as in how well a user can recall a word. If a user has 'learned' a word, it will move up to the next level. The next level could be in one day and after that day, the word is now queued to be tested again. If the user is again able to recall the word, it is moved to the next level, which could be in two days. This goes on and on. If, however, the user does not recall the word correctly, it is moved down a level. Levels go up - in the following example exponentially: 1, 2, 4, 8, 16, 32 (Settles & Meeder, 2016; Raaijmakers, 2003).