However, as highlighted in Cognitive Load Theory, the information in working memory is rarely, if ever, 100% new. Just as we now know that past experience guides how we perceive the world around us (top-down processing, see Gilbert & Li, 2013, for a review), working memory is also influenced by our experiences, as stored in long-term memory. Baddeley (2000) acknowledged this, adding an “episodic buffer” to his model to integrate information from long-term memory with that held in working memory. Other authors, such as Cowan (1988) and Ericsson and Kintsch (1995), suggest that working memory should be conceptualized as the activated part of long-term memory, rather than seeing them as two separate stores. In this view long-term and working memory are not located in different parts of the brain; long-term memories are stored across neurons and synapses and information is represented in working memory in their patterns of activation. In Cowan’s (1988) model, long-term memory is activated by stimulus information (from our external environment or internal thought processes). When we voluntarily attend to such information, effortfully and consciously processing it (such as when we detect a change in our environment), this attended-to, activated long-term memory IS what we call working memory. In this model, it is the attentional control that is limited in capacity to processing a few items or chunks. With repeated experience of an activity, as we automatize how to perform it and we become an expert at it, subsequent performances activate long-term memory without requiring attentional control, meaning that our working memory capacity for that activity becomes virtually unlimited. This aspect of human cognition is the central tenet underlying CLT. When we start to learn a new and complex concept or skill, the constituent elements of that concept or skill are all new and must be attended to and processed separately, creating huge cognitive load for novices.

But can memory really be the main (only?) thing separating experts from novices?

Ericsson and Kintsch (1995) suggest that as we acquire expertise at a task, information is stored in higher order structures or representations in long-term memory, which are reactivated when performing the task. The acquisition of a concept or skill is the process of forming a template or schema for it in long-term memory, in which the multiple elements get gradually integrated into a smaller number of more complex elements that we can activate automatically[1] (Sweller & Chandler, 1994). Expertise is task-specific, but for any activity in which expertise is acquired, long-term working memory yields huge processing capacity, in contrast to the severely limited working memory capacity of novices. The seminal study that suggested that experience (or long-term memory) distinguished experts from novices was a 1946 doctoral thesis by De Groot (published in English translation in 1965) about the thought processes of chess players of various proficiency levels. He gave them various states-of-play in chess and asked them to decide what move they should make. Using a think-aloud protocol, he was unable to find significant differences between masters, experts, and novices in how they approached the task. The only differences between (grand)master players and novice players were that the masters could utilize automatic routines and strategies and could immediately perceive the abstract or deep structure of the game, capabilities beyond those of novices. In De Groot’s words:

His extremely extensive, widely branched and highly organized system of knowledge and experience enables him, first, to recognize immediately a chess position as one belonging to an unwritten category (type) with corresponding board means to be applied, and second, to “see” immediately and in a highly adequate way its specific, individual features against the background of the type (category). (p. 306)

Experience, as represented in long-term memory, enables experts to see the solution(s), typically in an automatic or intuitive manner, whereas novices struggle to see how the parts relate to each other.

Chase and Simon (1973) extended De Groot’s findings by testing chess players’ recall of chess boards in various configurations. Whereas the novice player could only recall the positions of a few pieces, the master could recall almost the entire board. However, this was true only when the configuration of pieces simulated a state of actual game play. When the pieces were randomly placed on the board, the master’s recall was similar to that of the expert and novice. These studies suggest that masters don’t chunk four or five pieces, perceiving them as if they were one piece, but rather that they had formed subconscious long-term memory representations of numerous states of game play, enabling them to see the underlying structure of the configurations and intuitively determine the best move. Memory allows experts to perceive the solution; novices have to test a multitude of possible solutions and can only hope they are lucky enough to land on one that isn’t too bad.

But how does this relate to language learning?

Well, interestingly, Ericsson and Kintsch (1995) base part of their argument that the performance of experts can only be explained by long-term working memory on our ability to understand written texts. They claim that when reading a text, we construct an integrated representation of the gist of the text in long-term working memory in order to comprehend it. In this way, we are able to store a lot of information about the text, much more than a few chunks of information. In terms of CLT, if the text is in our native language (or a language we are highly proficient in), processing the sentences and their meaning is largely automatic and does not impose much cognitive load, provided we are familiar with the topic (a point I will return to later). In the same way master chess players perceive chess pieces, we automatically perceive words and phrases and their meanings and have access to a range of routines for how they relate to each other within larger texts.

 These internal models are like templates, enabling us to easily remember conversations and stories. The movie Memento, about a man with short-term memory loss, is acclaimed partly because it reversed the narrative template so that viewers were as confused as the lead character.

For language learners, reading and listening in the target language are not grounded in long-term working memory, processing the words automatically and with low attentional demands. Instead, comprehending a text relies much more on the ability to hold new information in working memory, with its severe limitations and susceptibility to cognitive overload. Although learners may struggle to decode the words and their meanings, understanding individual words is not the sole problem. According to CLT, it is maintaining the relations between individual items in working memory (element interactivity) that causes cognitive overload for novice learners (Sweller & Chandler, 1994). Understanding what each word means is not so difficult, but understanding how the words’ meanings relate to each other to produce the complex meaning of the whole text is beyond the capacity of their working memory. If a learner has to decode each sentence word by word, this places a heavy demand on working memory, making it very difficult to create a gist representation of the whole sentence, to guide interpretation of the following sentence. With spoken texts the difficulties are compounded by the transience of the input and the inability to control the word rate or to re-listen to a word or phrase. Like a novice chess player, a language learner “has to build up from the ground – if such a thing is possible at all” (De Groot, 1965, p. 305).

^[1] As a result, instructional materials that are optimally designed for learners (novices) are detrimental to experts’ performance, known as the expertise-reversal effect in CLT (Kalyuga, 2007).