Revenge of the Finnish Chicken

Earlier in the year I wrote a short article about my findings with the Finnish Chicken game. You can read the whole thing here, F*ck The Finnish Chicken – A Case Study of CLA Games Implementation, but if you haven’t done so, here is a quick summary:

Last Episode

Rules of the Game

Both fencers start out of distance in right Vom Tag. As they move closer to each other they are allowed to pick one of two actions:

  1. Direct cut to the head from the right.
    • If they both attack, look at who was moving forward and who was standing still. In just about all cases one person didn’t move their feet and thus were clearly attacking in response to their opponent instead of parrying. Auto-lose.
  2. Parry against the cut to the head from the right.
    • If you can’t tell if someone was parrying or attacking it is also an auto-lose. Make it clearer. This isn’t sparring, it’s a game with rules that you need to follow.


  1. When playing the Finnish Chicken game people consistently attack from too far away, and lose by being parried. They are also aware that 
  2. From a logical game theory point of view it would make sense for neither participant to attack, and only seek to get closer. Either to draw the parry or get so close that they don’t have any chance of defense.
  3. No matter how much you tell people “get closer before attacking” it doesn’t really change the results.

(Note that this is not a SwordSTEM article. I have been counting during the times I run the game, but not keeping records.)


I was able to get people to attack at an appropriate distance by first “priming” them by doing Direct Attack game before, and switching the available target areas mid-game.

And Now The Continuation…

Of course I didn’t stop tinkering here. While the initial experiment with doing an extensive session of Direct Attack worked, it didn’t result in improved overall performance on subsequent attempts without the Direct Attack proceeding it.

Unequal Losses

It’s been well documented in many, many studies that humans do not make rational choices when looking at outcomes. (The Wikipedia section on Risk Aversion is a good read.) A simple way to see this in action is to do a simple training exercise.

  1. Have [Z] hold a staff and swing it slowly back and forth on a horizontal plane.
  2. Have [A] stand outside the range of the staff, and instruct them to land a cut to the head and retreat without getting hit by the staff.

So the course of action is pretty simple, wait until the staff passes, move forward and strike, and then move backwards. All fast enough you don’t get hit by the staff. (This is actually the very first thing I have a new person do before even giving them any instruction.)

When using this on more experienced students (and swinging a bit faster) it can still be a challenge, as they have to get in and out quickly. And if it’s truly a challenge then they will occasionally not make it out fast enough. Any drill without the challenge isn’t a very good drill, so this is expected and good. 

But then the fun part.

“how many times did you get hit because you didn’t back out fast enough?”

*Everyone raises hands*

“How many times did you get hit because you went too early and ran into the staff before it cleared you?”

*Almost no one raises hands*

“If you’re having trouble getting out in time, why aren’t you starting earlier?”

Because if you don’t get out fast enough you feel you just need to do a little better. If you go too early you feel dumb. Thus even though both are a “fail”, they don’t feel equally bad. And the flip side of this is you don’t get much learning on exactly just how early you can go.

Back to the Finnish Chicken

One of my ideas was that people feel that “getting hit in the head” and “getting parried” are not equal outcomes, despite the game rules saying they are. Which makes perfect sense.

I tried the game again (on a new group) with instructions:

“we are trying to figure out how close we need to be, so moving to the right place is of absolute importance. If you lose because you swung too early and you parried you need to do two jumping jacks, if you lose because you got hit you don’t need to do anything.”

This is not exactly a harsh punishment for losing, but I at least made it clear that one was worse than the other. And, excuse me if you’re seen this episode before, it didn’t help. 

I’m sure that making the consequences rather harsh would eventually lead to a behavior change, but that’s kind of defeating the whole point of having some sort of rock-paper-scissors meta game baked in.. The idea is that they are learning to figure this out themselves.


The most success I’ve ever had with this is through reframing the instructions in terms of the concept of “affordances”. I won’t go into the details here (currently working on a SwordSTEM covering the subject), but we can examine two different instructions:

  1. “Don’t attack until you are at the correct distance.”
  2. “Don’t attack until you are sure that you can hit with it.”

The first is prescriptive on where they need to be, it is getting them to focus on the distance between the two. And humans are really bad at estimating physical measurements like distance. The second is an instruction to use an affordance – the affordance of their face for receiving your sword. And humans are good at doing affordances.

It sounds crazy, but in multiple attempts telling people “get closer” doesn’t help at all. But “don’t attack unless you’re very confident you can actually hit” does help. Crazy!

Higher Order Information

If baseball players only train to hit balls off of a constant pitching machine they will likely learn to time their actions to something like “when the ball gets <x> big, swing”. Because the size of the ball increases as it gets closer, so once you see it at a certain size you know it is a certain time away, and that is when you swing.

But if you are instead learning to hit at random speeds you must instead develop a more robust control law, using specifying information like the ratio of its size to its increase in size. (This gives you a direct result for time-to-contact, regardless of the size, speed, or distance away. Known as “tau”.)

This suggests that most HEMA fencers are using non-specifying information to determine when to attack. The subconscious equivalent of “When I get this close to someone, I attack”. All while not not actually taking the circumstances into account. If the fighters instead start to develop control laws based on more robust information about circumstances they wouldn’t have such difficulty. 

Which suggests that this is why the first attempt with “priming” them for the correct distance worked. Not because I was teaching people that they needed to be closer. But because I was teaching them to break out of their default way of estimating attack range. 

Actionable Items

In the end this is about much more than the Finnish Chicken. Teaching people at what range to attack in a very fixed and abstract scenario isn’t helpful. What is helpful is getting people to base their attack distance on whatever affordances are or aren’t offered for attack and defense – rather than just attacking at the same range every time.

This is best accomplished by making sure that they are constantly doing activities which force them to attack from different distances in order for them to be successful. And in doing so learn to figure out when to attack, rather than just dialing in the correct “optimal difference” for all circumstances.