Welcome to Memory Mondays, where I read a textbook on memory and talk about what I learned. If you like your cognitive psychology neatly summarized, with a healthy dose of unnecessary commentary and excessive amounts of semi-colons, this is the series for you! Now new and improved, with fewer comma splices.
Long hiatus from this series. It seemed, when I initially started a weekly series, that keeping up would be manageable, but that did not turn out to be the case. I took an online course, pursued other educational psychology interests, and was busy with work. I also have poor impulse control when it comes to spending time online; if I reel that bad habit in, I can make time for more meaningful activities such as researching and writing for my blog.
The break from blogging has been helpful in some ways, as breaks from writing are. Looking back on previous posts, they seem wordy and heavy with technical vocabulary. Considering I am trying to make these posts accessible to laypeople, that is poor writing on my part. I’ll be running the text through the Hemingway App, to simplify it, and Grammarly, to find the comma splices that litter my writing; it will be an interesting experiment on the extent to which artificial intelligence can help with the writing process.
This week’s topic is semantic memory. Without further ado, let’s get started.
Semantic and Episodic Memory
Semantic memory is our general knowledge of the world and is distinct from episodic memory in that it is abstracted from a specific experience. An episodic memory of a dog might be of your neighbour’s dog, small and white, yapping at you on your way to work. A semantic memory would be the concept of a dog: a general understanding of the idea which you would use to identify the different four-legged furry animals you meet as dog or not-dog. Episodic memory develops later in life and is more susceptible to deterioration than semantic memory.
Neuropsychology provides further support for the distinction between episodic and semantic memory; one system can be severely impaired while the other still works just fine. A patient with retrograde amnesia loses episodic memories from before the onset of the amnesia, but general knowledge acquired over the same time period as the episodic memories stays mostly intact. A patient with semantic dementia, on the other hand, loses concept knowledge while keeping memories of life events. Retrograde amnesia is associated with damage to the medial temporal lobe, while semantic dementia is associated with damage to anterior frontal lobes; in simpler terms, these disorders are linked to different parts of the brain.
However, there is still evidence that the systems are interlinked. Autobiographical memory, for example, involves a mix of episodic and semantic memory. For distant autobiographical memories (memories of events that occurred a long time ago and thus are harder to recall), semantic knowledge serves as a framework to help with the retrieval of the episodic information. You have general knowledge of school playgrounds and children play activities, and your episodic memory fills in the details of childhood school memories. And, as was discussed in an earlier post, many semantic memories start their life as specific experiences before being generalized into concepts. Separate but interdependent systems.
Organization of Semantic Memory
There are several models for how concepts are organized in the mind.
The first is the hierarchical network model, developed by Collins and Quillian in 1969. Concepts are represented as nodes, and properties of that concept are stored at that node. So, the ‘dog’ node would have ‘furry’, ‘four-legs’ and ‘tail’ stored at it. The property information for a concept is stored as high up as possible, to minimize the amount of information in the hierarchy. So, at the ‘labradoodle’ node, ‘furry’ and ‘four-legged’ would not be stored, as that is at the ‘dog’ node, but ‘hypoallergenic’ and ‘looks like fried chicken’ would be.
This model has been tested in laboratory settings. As an example, participants would be asked to identify statements as true or false, such as “labradoodles have curly fur”, or “labradoodles nurse their young.” According to the model, sentences where the information is stored at the same node (‘labradoodle’ and ‘curly fur’) will get a quicker response than sentences where the information is stored at different nodes (‘nurse their young’ would be at the ‘mammal’ node).
Experiments have shown this prediction to be accurate; the model identifies how we use semantic memory to infer the correct answer. For example, we don’t store information such as “Aristotle had elbows.” However, we can infer it, because we know Aristotle was (probably) human and (the vast majority of) humans have elbows. So Aristotle had elbows. And we can infer that labradoodles nurse their young, because labradoodles are dogs, and dogs are mammals, and mammals nurse their young. The hierarchical network allows inferences, and inferences mean we don’t have to store as much information.
There are limits to the model. It cannot explain the typicality effect, which is how people take less time to decide if a typical category member belongs to category (e.g. oranges are fruit) than atypical members (e.g. durian are also fruits). According to the model, both typical and atypical members should be the same distance from the category node, and thus should have the same response times for deciding if they belong in the category. Experiments have not found this to be the case. As well, different people have different definitions for categories, or some things do not fit clearly into one category as opposed to another. For example, should synchronized swimming or horseback riding be classified as sports? Ludwig Wittgenstein asked in 1958 what the defining characteristics of a game are. After all, hockey is a game, and so is Dungeons and Dragons. We all seem to have a sense of what a game is, but efforts to make the definition explicit are complex. Categories are not clearly defined.
Collins and Loftus proposed the spreading activation model in 1975 to resolve these problems. Concepts are still at nodes, but are organized by semantic distance, or how closely related the concepts are. One way to determine distance is to get people to list as many members as possible for a category; those named most often are most related to the category. This allows for vague category definitions, as some items that don’t seem to fit the category well can be more distantly related. Point for spreading activation theory.
When a node is activated, activation then spreads to related nodes, and most strongly to those most closely related. That’s the “spreading activation” part of the model. So, when someone says “fruit” and you picture an orange or apple, that’s because “fruit” activated the ‘orange’ and ‘apple’ nodes much more strongly than ‘durian’ was activated. This model, then, predicts the typicality effect, as more typical items are more closely related, are activated more strongly, and thus have a shorter response time. Second point for spreading activation.
Spreading activation has proven to be a better model than the hierarchical network, as it can account for more findings such as the typicality effect. This is because it is a more flexible model, and thus has less precise predictions. Which kind of feels like saying “I’m pretty good at darts, if you make the dartboard twice as big.”
Another problem with the model are the assumptions that a) concepts are found at a single node, rather than being more distributed, and b) concepts have a single fixed definition or representation. We will soon see the problems with these assumptions.
The final model to be discussed has to do with how we name objects. Take, as an example and not through an act of theft, a desk in your classroom. You could call it “furniture”, a “desk” or a “wooden desk”, but you’ll probably call it a desk. Names of things exist at different levels: superordinate (furniture), basic (desk), and subordinate (wooden desk).
We usually use the basic level. It provides the right amount of information and it is usually acquired first by children. But this isn’t always the case. If you’re an expert in something, such as dogs, you’re more likely to use the subordinate level, calling a labradoodle a labradoodle instead of a dog.
This shows a commonality between the spreading activation model and this second hierarchy model. (It merits mention that these are my own thoughts, and not from the chapter.) The information you hold in semantic memory influences how you use and organize concepts. If you have extensive knowledge on a topic, you are more likely to use the subordinate level when naming objects in that topic. The background knowledge also holds influence in the spreading activation model. I intentionally chose durian as an ‘atypical’ fruit example. For people here in Southeast Asia, it would be more typical and more closely related to the fruit node than for people who developed their concept of fruit in Western countries. Like I pointed out in my earlier post on cognitive psychology, the contents of our memory hold a large influence over our cognitive processes.
So, what do these models mean for teachers? Beyond “ain’t this interesting”, I’m not sure. It is valuable to build networks of knowledge in student minds and to create connections between different parts of a subject and different subjects. The goal is, when they think of a concept (e.g. democracy), a tidal wave of related examples and concepts comes rushing in (e.g. Ancient Greece, direct and representational, election spending). Pulling in a diverse range of ideas is what creates interesting answers to complex questions. The spreading activation model, then, helps us understand this process of activation of related ideas.
The influence of background knowledge on the organization of concepts is also important to keep in mind. Teachers need to be aware of the curse of knowledge; our extensive experience with a subject will result in a different understanding of a concept than our students will have, but we are often not aware of this difference. Beyond that, I am not sure what implications these theories have for teaching and learning. Perhaps it will be more apparent as I dig further into how we develop understandings of concepts.
There is still so much left in this chapter to explore! Seeing as this post is already getting lengthy, I’ll save it for another week. 😊
Next time: concepts, our brains, and schema