Thursday, April 30, 2015

Hiking Through Hell Valley: Episodic vs. Semantic Memories

Let me tell you a short story. After you read it, see how many details you can recall.
In the summer of 2013, I embarked on a grand adventure. I traveled halfway across the globe to Sapporo, Japan, which is home to the ramen noodle. From there, I took a train to Hell Valley where I hiked up the side of a mountain. Midway though the hike, me and my three friends stopped to soaked our feet in a 110-degree mountain stream, which was fed by a hot spring.  We also saw wasps the size of your thumb. When we got back to Sapporo, we went out to dinner, sat on the floor, and ate sushi made from yellow-fin tuna.
Now take a few minutes to repeat back my story with as many details as you can recall.
  • What details did you omit (or insert)? 
  • Which parts were the easiest (or hardest) to remember?
  • Were you able to get the chronology right?

Thirty-one Flavors of Memory

My little story might leave an impression, but I guarantee that you will remember it in much more detail, and for a longer period of time, if it actually happened to you.

This demonstration is meant to underscore the fact that we have the ability to learn vicariously from other people. We can also learn from our own histories and from our own autobiographical memories. The memories that we create directly from our experiences are called episodic memories. These memories are like own own personal narrative. They consist of an event that takes place in a particular time and place (i.e., the "setting"), and potentially with other people (i.e., the "characters"). They might also contain an emotional element (e.g., laughing with your friends when you discover how incredibly hot the hot springs truly are). 

Episodic memories are often contrasted with semantic memories, which don't really have a personal connection to your life. Reading about my trip to Japan results in a semantic memory for you, but it is an episodic memory for me. Because semantic memories are not tied to a specific time or location, they are general knowledge about the world. This might sound familiar because both episodic and semantic memories are different subtypes of declarative memory.

Is the distinction between episodic and semantic memories real, or is it merely a convenient theoretical distinction? There is suggestive evidence from neuroscience that these two types of memories are represented in different areas of the brain. For example, individuals who have damage to the hippocampus or left prefrontal cortex lose the ability to form new episodic memories; however, their ability to learn new facts is left (mostly) intact.

The STEM Connection

How does this connect to education? A large percentage of information that a student must learn is semantic. In the end, we want students to remember facts like: plants take in carbon dioxide and expel oxygen. Moreover, we want our students to remember the lessons that they learned in school for a very long time. It is probably a stretch to expect our students to remember everything we teach them; however, given our knowledge about the way memory works, we can structure information in a way that resonates with the brain.

When something is personally relevant, it is easier to learn and remember. As we saw earlier, memories are easier to retrieve when we have multiple routes to that information. It's also the case that our semantic network is densely packed with information about our own lives. Thus, if we can embed scientific, technological, engineering and/or mathematical information in the context of our lives, then it is likely that we will own that information. 

The challenge for instructional designers is to figure out a way to make STEM lessons come to life. One potential method is to make lessons more like episodic memories and less like semantic memories. We could teach kids a strategy for mental arithmetic. That would provide the students with a strategy that will be devoid of any episodic memory. Alternatively, we can ask students to work in small groups to invent their own a strategies [2]. If they attempt to solve this on their own (with sufficient scaffolding and resources, of course), then the event itself will become part of the memory. They will remember working with other students, in a particular context, and maybe even remember some of the wrong avenues they explored and how they were able to overcome those obstacles. 

Ultimately, the goal is to help students bind content knowledge to the vast repository of other episodic memories. Asking students to invent their own strategies is immensely gratifying, and it can serve as potential gist for long-lasting, personal memories. To conclude, I will leave you with this quote from Benjamin Franklin:

   Tell me and I forget,
   teach me and I may remember,
   involve me and I learn.

Share and Enjoy!

Dr. Bob

For More Information

[1] Tulving, E. (1972). Episodic and semantic memory. In E. Tulving and W. Donaldson (Eds.), Organization of Memory (pp. 381–402). New York: Academic Press.

[2] Schwartz, D. L., & Martin, T. (2004). Inventing to prepare for future learning: The hidden efficiency of encouraging original student production in statistics instruction. Cognition and Instruction, 22(2), 129-184.

Thursday, April 23, 2015

Just a Reflection of a Reflection: Hierarchical Retrieval

When I started this blog, my goal was to publish one post a week for a full year. Since I am officially halfway to my goal, I thought this would be a good time to pause and reflect on what we've learned so far. If you've been following along, almost all of my posts thus far have been about two things: how knowledge is acquired and how it gets represented. These two themes were further broken down into smaller pieces so that they could be understood in isolation. In so doing, however, it is likely that we miss the forest for the trees. So let's take a step back and see where we are and how far we've come. 

To motivate this reflection, I would like to introduce a new concept. As we discussed previously, working memory only holds a limited amount of information. We also learned a couple of different strategies to expand our working-memory capacity. One of them was called chunking. The concept for today's reflection is called Hierarchical Retrieval, and it is a way to chunk a lot of information by exploiting the semantic connections between the ideas.

Hierarchical Retrieval Applied to Cognitive Science

As stated previously, the mind is a big fan of both meaning and order. Things that go together tend to get represented together. For example, it is very natural to link together the concept of "cheese" and "mouse" because we believe that mice like to eat cheese. In other words, these two concepts are linked because of the meaning that they have, and we can generate a causal connection that links the two. However, the concept "cheese" and "clown fish" is a less natural connection. It would take a pretty active imagination to link these two concepts. If we are able to tell a story about how two concepts are linked, then it is more likely that we will be able to remember them. Thus, retrieval can be based on meaning (i.e., semantics).

The mind also craves order. Concepts are connected to each other, as we saw in the associative and semantic network representations of knowledge. This network representation was useful because it helped explain cognitive phenomena like priming and how we generate creative ideas. When laid out as a network, there didn't seem to be any discernible order. However, we can explicitly reorganize these network structures so that they are hierarchical. Here's an example that might look familiar.

There are eleven concepts represented in the hierarchical diagram. That means that we are beyond the typical working memory capacity. However, I think it would be trivial for you to memorize this list because you can chunk together the items that are nested under the "parent" concepts (i.e., the concept "Birds" is the parent concept to "Finches" and "Sparrows").

The STEM Connection

Textbooks use the power of a hierarchy all the time. In fact, you might think about the table of contents as a way to hierarchically organize the information in the book. In addition, it might help students to create their own hierarchical diagram of the information they have covered in class because it forces the student to think about the connections between the ideas. Also, it might be helpful, when introducing a new topic, to lay out all of the information so the student can see where the class currently is and where it is going. In educational psychology, they call this an advanced organizer.

If we were to apply that same process to this blog, it might look something like this:

My goal is to keep expanding this diagram until all of the various concepts and topics have been discussed. For example, we talked about the limitations of the attentional system in that we can only process one stream of information at a time. The concept "Attention" does not have a node in the network diagram above. Thus, this representation is incomplete. Moreover, it fails to capture the inner-connections between various concepts. For example, while talking about solving problems, it became evident that we needed to consider the representation that was used during problem solving, and the different effects on working-memory load. Thus, a hierarchical retrieval structure, such as this, fails to capture all of the information. But it does give us a nice, bird's eye view of all the topics we've covered so far. It also points to some of the concepts that are missing, which I can't wait to discuss in future blog posts! 

Share and Enjoy! 

Dr. Bob

For More Information

[1] I was introduced to the lovely idea of a hierarchical retrieval structure in the following paper: Bower, G. H., Clark, M. C., Lesgold, A. M., & Winzenz, D. (1969). Hierarchical retrieval schemes in recall of categorized word lists. Journal of Verbal Learning and Verbal Behavior, 8(3), 323-343.

Thursday, April 16, 2015

Getting Out of Your Head: External Cognition

Here's a fun little game. Enter a number, any number, in the left text box marked "Input." Then look at the right side of the box. What's going on inside the box [1]? 


One of the fundamental concepts in computer science is the distinction between input and output. Input is the information that gets used by a program, and output is the result of any sort of manipulation that happens to that input. At the highest level of description, every computer program can be described as: input => processing => output. The power of a program comes from the ability to link together smaller sub-processes so that the output of one process can become the input for another process [2].

For the past forty years, Cognitive Science has benefited from thinking about human cognition in these terms. In other words, many scientists think about the mind like a computer that can be broken down into components that are responsible for processing certain types of information. It's useful to track the information that is used as input (e.g., stimuli) and what gets generated as output (e.g., behavior).

However, within Cognitive Science, there is a group of people who don't necessarily believe that all cognition needs to take place strictly in the mind. Instead, some of our cognition (i.e., the processing phase) can be offloaded to the outside world. Does this really happen? If so, how can we use this to our advantage?

"Luke, Use the Force(ing Function)" --Ben Kenobi

Probably the most basic and simplistic example of external cognition is writing something down. We take notes for a variety of reasons. First, most of us have an accurate mental model of our memory and realize that it is fallible. Through experience, we realize that we will not remember all of the details of everything that we see or hear. Second, there's evidence that merely taking notes helps us better encode the information; therefore, we externalize our memory in hopes of improving both encoding and storage.

In a previous post, I made the claim that cued recall is almost always easier than free recall. Because of that fact, we can use it to our advantage. One way to exploit cued recall in everyday life is to systematically embed clues in our environment to help us remember easily forgotten tasks. One of my favorite memory tricks is called a forcing function. For example, to start your car, you need your car keys. In other words, you are forced to grab your keys to drive your car. Now suppose you have a letter that you need to mail tomorrow; unfortunately, it's likely that you will forget because mailing a letter isn't part of your normal morning routine. So what should you do with it? Based on the lovely feature of cued recall, I am going to put the letter under my car keys so I am forced to remember it. I could tie a string around my finger (like they did in the old days), but that's not a really great cue because the association between letters and strings is completely artificial. A forcing function is a great way to help you remember. The external environment is now tasked with "remembering" something for you.

Engineers who understand the limitations of memory and attention use forcing functions in their designs. My favorite example of a forcing function was the way I locked my Volvo 240 station wagon. I had to insert my key into the outside lock and twist it to engage the power locks. In other words, I had to stand on the outside of my car, with my keys in my hand, to lock my car. There was no way to lock myself out [3].

"Roger, Roger. What's our vector, Victor?" --Captain Oveur

External cognition can be much more than a memory aid. It can also assist in the actual process of thinking and reasoning. One of the oldest tools for externalizing cognition is the abacus. Humans are perfectly able to carry out mental computations; however, working memory tends to fill up quickly as we chain together calculations. Thus, an abacus can help increase the speed and accuracy of our mental calculations by externalizing the interim numbers.

Here is another, far more complex, example of external cognition: landing an airplane [4]. When a plane is about 30 minutes away from the airport, the pilot and co-pilot begin calculating the speed at which they will eventually land their plane. While this might sound simple, it is far from it. Why? Because the landing speed is largely a function of the airplane's weight. The weight of the plane is influenced by the amount of fuel onboard. As the plane flies and burns fuel, it becomes lighter. Given the association between speed and weight, the pilots need to project the weight and speed of the plane at the time of landing. 

In addition to adjusting the speed, pilots can also reconfigure the shape of the wing by changing the angle of the leading edge, extending the overall length of the wing, and setting the angle of the trailing edge (i.e., the "flaps"). The goal of reconfiguring the wing is to maximize lift while minimizing the plane's speed when landing. 

To configure the optimal settings of the shape of the wing and the plane's speed, the pilot has several artifacts at her disposal. First, there is the instrumentation of the cockpit. Various gauges and meters help pilots understand their current altitude, speed, and bearing. Second, they also have a card that they carry where they can look up the minimum speed for different increments of weight crossed with the flap configuration of the wing. This is a simple representation for humans to use because we are better at visually scanning rows and columns of numerical data than we are at computing the value of a complex, multi-factor function. Finally, they have each other. The pilot and co-pilot work together to make sure they do not skip any steps while making this calculation.

The STEM Connection

How can we use the ideas from external cognition and forcing functions in the classroom? Obviously students would like to externalize their memory by taking notes. Some students take copious amounts of notes. This might help some, but hurt others. The reason why it might hurt others is because we are serial processors, and we can only process one stream of information at a time. For these students, maybe it would be best if the teacher supplied the notes. Alternatively, the student can listen to the lesson and take notes from the textbook outside of class. 

Another application might be to make "external cognition" into an assignment. Some (probably most) students benefit from making the lesson concrete. One way to make something concrete is to create an actual artifact that can be used. Going back to the pilot example, they created a table so that they can look up the speed given the aircraft's weight and wing configuration. Students could construct an analogous representation for complex relationships. For example, in statistics, we had tables that we used to look up the critical value for a t-test given the degrees of freedom and the significance level.

Finally, and this is kind of a stretch, but you can teach your kids about forcing functions, especially if they are particularly prone to forgetting (e.g., remembering to take home their textbook or to return a permission slip).

Humans are extremely smart. One of the reasons why we are smart is because we are able to recognize our limitations and structure the external environment to help us overcome our shortcomings. Knowing about external cognition can help us become even more intelligent! 

Share and Enjoy! 

Dr. Bob

For More Information

[1] Special thanks to Josh Fisher for creating The Blackbox Game. Check out his google+ page for more fun and creative widgets that he has assembled!

[2] One of the strengths of our species is that we don't have to solve the same problems that others have solved in the past. In other words, I don't need to rediscover algebra from scratch to use it to solve problems. Instead, I can rely on the hard work from the people who have come before me. This is very much true in computer science where individuals will solve a particularly difficult problem for the rest of the community and then post their solution in a library for a particular programming language. For example, if you have neither the time nor expertise to write a parser that will pull apart time and dates in Python, that's already been done for you. A similar arrangement could also be done in the classroom. 

[3] Actually, that's not entirely true. If you open the passenger door, walk around to the driver's door, lock all of the doors from the outside, walk back around to the passenger side, throw the keys in, and close the door, I would be locked out. Thankfully, I was never that stupid.

[4] Hutchins, E., (1995) How a cockpit remembers its speed. Cognitive Science, 19, 265-288.

Thursday, April 9, 2015

Rack 'Em and Crack 'Em: Free vs. Cued Recall

Let's start with an activity. Study the word pairs below and try to commit them to memory. Order isn't important, but matching the pairs together is.
  1. bicycle : bear
  2. shoe : pill
  3. face : engine
  4. street : monkey
  5. coffee : satellite
  6. dragon : boat

Memory Test A. Now, scroll your screen so you can't see the original list of words. You goal is to list as many of the pairs as you can remember. 

  1. _______ : _______
  2. _______ : _______
  3. _______ : _______
  4. _______ : _______
  5. _______ : _______
  6. _______ : _______ 

Memory Test B. Let's try again, but this time I will give you one of the words from each pair, and your job is to recall its mate. 
  1. _______ : bear
  2. shoe : _______
  3. face : _______
  4. _______ : monkey
  5. coffee : _______ 
  6. _______ : boat

Memory Test C. I know you've probably have this list memorized by now, but let's try one last time. I will give you one word, and then you have to pick the matching word from a list.

  1. _______ : bear
    1. slow
    2. bicycle
    3. bridge
  2. shoe : _______
    1. pill
    2. plug
    3. capital
  3. face : _______
    1. ladle
    2. battery
    3. engine
  4. _______ : monkey
    1. keyboard
    2. street
    3. insurance
  5. coffee : _______
    1. satellite
    2. heater
    3. shovel
  6. _______ : boat
    1. bread
    2. goggles
    3. dragon

How many did you get right? Which memory test was the easiest? Which test was the hardest?

"'Cause I'm as free as a bird now." --Lynyrd Skynyrd

The purpose of this demonstration is to highlight the distinction between three different types of memory. The first task is called free recall because there isn't anything in your surrounding environment to help you remember the answer. You have to do it straight from memory. If I ask you for your social security number, then that would be another example of a free recall task.

The second task is different because you have a clue. Cued recall is different from free recall because there is some information in the environment that will assist you in remembering. For example, I might ask you, "What was the former name of the city 'Istanbul'?" If you're familiar with the song by the band They Might Be Giants, then my humming the tune might help you remember that Istanbul was once ConstantinopleThe tune of that song is a powerful cue for the lyrics and (thankfully) the lyrics contain an accurate answer to my question. If you're like me, cued recall is almost always easier than free recall.

The third type of memory is related to another useful concept in memory research: Recognition memory. You might not be able to recall the title of the movie starring both Tom Cruise and Paul Newman. But if you see a list of movies (e.g., Days of Thunder, The Color of Money, Rain Man, or Top Gun), then you might recognize the right answer. The reason why this difference is useful is that the knowledge is stored somewhere in long-term memory, but the route to that piece of knowledge is blocked or unavailable. Being able to recognize the right answer demonstrates that the separation of storage and retrieval in our simple model of memory

The STEM Connection

The cued vs. free vs. recognition memory distinctions have interesting implications for the design of educational assessments. When designing an assessment, an educator might ask if it is important for the student to be able to recognize or recall a chunk of knowledge in isolation or during problem solving. The assessment, of course, depends on the educational objective. If we play this out, here is how the same piece of knowledge might be assessed differently, based on the educational objective:

A. Recognition

  1. The measure of a right angle is ______ degrees.
    1. 30
    2. 60
    3. 90
    4. 120

B. Cued Recall

  1. Calculate the measure of the unknown angle. [1]

C. Free Recall

  1. The measure of a right angle is ______ degrees.

In all three cases, the intent of this assessment item is to determine if the student has encoded and stored in long-term memory the declarative chunk: The measure of a right angle is equal to ninety degrees. Because you can assess the availability of this declarative chunk in many different ways, deciding on the format largely depends on how rigorously you want to assess the student's understanding. I think most practitioners would agree that the list above is ordered in terms of rigor. In other words, it is easier to recognize the answer than it is to recall it. Both cued recall and recognition are easier than free recall. 

An even more rigorous method would be to assess the student's understanding during problem solving. This would require that the student show each step of his or her solution. I'm imaging a problem like this:

Designing assessments is difficult work. But knowing the differences in the types of memory needed to answer the question can help in creation of useful assessment items. 

Share and Enjoy! 

Dr. Bob

For More Information

[1] I'm doing a bit of handwaving here. I'm assuming that the shape of the angle and the right-angle symbol is enough of a cue to help the student recall the answer. In other words, during the learning phase, the student made the association between the diagram, its symbol, and the measure of the angle.

Thursday, April 2, 2015

This is This and That is That: Isomorphic Representations

Let's play a little game. If you haven't seen it before, it's called the Tower of Hanoi, and the goal is to move all three of the colored disks from the left-most peg to the right-most peg. But here's the catch: you can only move one disk at a time. Oh, and one more thing: you can't place a larger disk on top of a smaller disk [1]. 
Canvas not supported.
Discs Moves:

In a previous post, we talked about defining a problem space. The problem space for the Tower of Hanoi is shown in Figure 1. At the top of the pyramid is the initial state. All of the colored disks are lined up on the left-most peg. Then each arrow points to a legal move. There is only one operator, which is that you can move one disk to a different peg. There are two constraints. First, you can't just grab the whole stack and move it. You have to move one disk at a time. The second constraint is that you are not permitted to place a larger disk on top of a smaller disk. The goal state is located at the bottom-right side of the pyramid where all of the disks are stacked on the right-most peg. [2]

Figure 1. The problem space for the three-disk version of the Tower of Hanoi

The Tower of Hanoi is like the Drosophila of Cognitive Science. It has been used in countless studies, and is arguably the best understood task in the literature on problem solving. I'm overstating things a bit, but you get the point. It's been around for a while. 

"Play it again Sam"

Since this task is so well known, some creative scientists have toyed around with different versions of the task. Let's play again, but this time, imagine the disks are like different sized cans that have been opened and turned upside-down.

Which version do you like better: the first version with solid disks or the second version with open cans? If you have a preference, what is the reason for your preference? Personally, I prefer the version with the upside-down cans. Why? Because I don't have to remember one of the problem constraints. Instead, the external environment has encoded that information for me. Now, making a move is simply moving something that can withstand the placement of another can. It's so much easier (for me, at least)! 

A Wolf in Wolf's Clothing

The fact that simply changing the representation of the colored bands in the Tower of Hanoi can make the problem easier to solve illustrates an important point about problem solving: the difficulty of a problem can be completely determined by the framing of the problem. Let's take a different example. For this problem, you have to decide if the following rule is valid or not: All cards with a vowel have an even number on the back. I then lay out four cards in front of you. Your task is to flip over the minimum number of cards to test the rule. [3]

Which cards did you turn over? If you're like 90% of the people who have been asked to do this task, then you probably were tempted to turn over "A" and "4." But if you turn over those two cards, then you don't know for sure if the rule is valid. For example, what if you turn over "7" and discover that it has a vowel? Then you know you for a fact that rule is invalid.

When the problem is framed in terms of an abstract (and highly arbitrary!) rule, then most people don't give the correct answer (A and 7). But what happens when we reframe the problem by contextualizing it? 

You're a bouncer at a bar, and you are in charge of finding underage drinkers (i.e., the rule is: If a person is drinking alcohol, then they need to be 21 or over). Which people do you need to interrogate? [4]

  1. You need to see the ID of a man drinking a beer.
  2. You need to see the ID of a woman drinking a coke.
  3. You need to see what a 21 year old is drinking. 
  4. You need to see what an 18 year old is drinking. 

Notice that the deep structure of each of these problems is exactly the same, which makes them isomorphic problems. You don't care what the woman is drinking (which maps onto the "K" card), nor do you care what the 21 year old is drinking (which maps onto the "4" card). Instead, to enforce the rule, you need to see the ID of the man drinking a beer (i.e., the "A" card), and you also need to see what the 18 year old is drinking (i.e., the "7" card).

The point is: the difficulty of the task depends on the way it is framed. Isomorphic problems  are easier to solve when framed in terms of familiar, contextualized rules than when  framed as arbitrary and abstract rules. 

The STEM Connection

The educational implication of how isomorphic problems are framed is pretty interesting. In a previous post, we talked about the educational advantage of using multiple representations. For the Tower of Hanoi, the way to solve the problem is to search through a problem space for the path that connects the initial and goal states. We can extend the concept of a problem space to include a space of representations. As we saw above, the Tower of Hanoi can be represented in (at least) two different ways, although many variants exist [5].

Here's another great example of the search for an isomorphic problem representation. As the story goes, John Von Neumann was at a dinner party and someone posed the following problem:
Two cyclist, 100 miles apart, start riding toward one another at a pace of 10 miles per hour. At the same instant they start, a bee flies from one bicycle toward the other at 20 miles per hour. When he reaches the second cyclist, the bee turns around and flies back toward the first cyclist. He does this until the cyclists meet. How far did the bee fly?
The obvious way to represent the problem is through a series of back-and-forth trips by the bee. However, once you start down this path, you quickly realize that the bee is going to take a lot of trips. Mathematically speaking, you would have to sum the distance of an infinite number of trips. That doesn't sound like fun! Instead, you can re-represent the problem as having two parts. First, figure out how long the cyclists will be riding until they meet. Then, use the simple speed = distance / time formula and calculate the distance the bee flew in the time it took the cyclists to meet. That's a much easier problem to solve! 

One of the lessons we need to teach our students is to reframe or re-represent difficult problems in a way that is much more meaningful, concrete, and easier to answer.

Share and Enjoy! 

Dr. Bob

For More Information

[1] Special thanks to Josh Fisher for making the Tower of Hanoi a fully interactive experience. Check out his google+ page for more fun and creative widgets that he has assembled!

[2] Newell, A., & Simon, H. A. (1972). Human problem solving. Englewood Cliffs, N.J.: Prentice-Hall.

[3] Wason, P. C. (1968). Reasoning about a rule. Quarterly Journal of Experimental Psychology, 20, 273 281.

[4] Cheng, P. W., & Holyoak, K. J. (1985). Pragmatic reasoning schemas. Cognitive Psychology, 17, 391–416.

[5] Zhang, J., & Norman, D. A. (1994). Representations in distributed cognitive tasks. Cognitive Science, 18, 87–122.