EQUINE CLICKER TRAINING.....
using precision and positive reinforcement to teach horses and people
Clicker Expo 2011: Chicago, Illinois
In March 2011, I traveled to Chicago for the east coast Clicker Expo where I got to meet with other clicker trainers and get new ideas from the excellent Clicker Expo faculty. Clicker Expo is always a great opportunity to really immerse myself in clicker training for three days and my brain is usually running on high speed for the whole thing. When I get home, I try to make sense of it all and share some of the information I received by writing a Clicker Expo report. Unfortunately this year I arrived at Clicker Expo with a slight case of laryngitis and went home even sicker three days later.
While I did manage to attend the sessions, I think it's going to take me a while to really process what I learned because my thinking was so foggy while I was there. In previous expos, I have had new insights and gotten a sense of what's new in clicker training and where it's going, but since I was not feeling well, I didn't have those kinds of "ah-ha" moments. But I do have some new ideas for training and once I have played with them some, I hope to share more about what I learned at the expo. In the meantime I will just share some of the important points that I got out of the sessions I attended.
Friday Session 1: Ken Ramirez - Aggression: Treatment and Context
Ken's talk was a look at the most common training strategies for dealing with aggression, not about the details of each strategy, so it was more of an overview than anything else. He explained the science behind each method and offered suggestions for how to choose what might be the right approach for various situations. I think one reason this topic was chosen is that aggression is a hot topic among dog trainers. In recent years, there have been a number of "new" approaches to aggression and there is a lot of controversy over which is "better," or what to use when. Ken's goal was to present information about the different methods so that dog owners and trainers could make educated choices about which method might work for them, or be more informed when dealing with professionals who promote a certain approach to aggression.
Before I get into his analyses about the different methods, I want to share some of his more general thoughts on aggression. All animals have an aggressive repertoire so aggression should be considered normal animal behavior. Behaviors that we consider to be signs of aggression are reactivity, resource guarding, possessive behavior, protective behavior, biting, barking, growling, lunging and so on.
In animals where aggression is a problem, it is usually a genetic, reactive, or learned behavior. For some reason, aggression has been reinforced in that animal so our job as trainers is to look for ways to reduce aggression. To do this we need to understand scenarios, recognize precursors, use redirection and appropriate training strategies, stop it before it starts, and keep records. These rules for dealing with aggression are credited to Turner and Tompkins,1999. Ken emphasized that record keeping is VERY important in aggression cases.
He started by saying there were 4 general strategies for dealing with aggression (I think this comes from Jean Donaldson.) Most training strategies use one of more of them. They are:
change the consequences = operant conditioning
He went on to identify which strategies were based more on classical conditioning (habituation, flooding, counter-conditioning) and which were based more on operant conditioning (reinforcement, punishment, redirection, CAT, click to calm, BAT, abandonment training, Watch-me, U-Turn, Recall and Look at That). He did emphasize that most strategies use both classical and operant conditioning so this is not a clean division, just a way of separating out which ones are based more on classical conditioning vs. operant conditioning.
Then he explained the science behind each strategy and when it seemed most effective. Some important considerations that might affect your choice are whether or not the trigger is unavoidable, how quickly the dog goes over threshold, what is reinforcing the behavior, the experience level of the trainer and the animal's past history. Many of the strategies he discussed were ones that would be used as part of a larger training plan. For example, a lot of the redirection exercises are helpful but not complete solutions in themselves. The most effective strategies were ones that actually changed the animal's emotional response to the trigger so that the animal formed a new, positive association.
I found Ken's analysis of each method to be really helpful. I was familiar with most of them (just from reading) but had not compared them to each other. At the end of the lecture, I definitely had a better sense of the different ways people deal with aggression as well as what many of the approaches have in common. Rather than try to repeat this information here in a very abbreviated format, I am going to just suggest that if you are dealing with aggression, you take your time to explore the different options.
I haven't really had time to think a lot about how to apply Ken's talk to horses. My only experience with aggressive horses has been with Rosie who came to me as weanling with a serious dislike of most people. Her policy was that "the best defense is a good offense" so I had to deal with biting, charging, kicking and other related behaviors. When he gave the talk, I did recognize that with Rosie, I used the "click to calm" approach which was to click for moments when she was less aggressive and over time, I was able to shape her behavior into something more pleasant. I also did a lot of redirection in that I taught her other more appropriate behaviors so that she had alternate behaviors to offer instead and used management to keep me safe.
I have to say that it was a long, slow process and while she is fine with me now, she still has an deep suspicion of new people. I think part of it is that we don't get that many visitors so this is not something we can practice a lot. But I am seeing small changes recently in that she will offer behaviors I like to visitors, as long as I am present.
Friday Session 2: Michele Pouliot - Anticipation
Michele Pouliot works both in guide dog work and competes in freestyle. I find that she gives out really practical advice and a lot of her information is easily transferrable to horses. I attended this talk because I wanted to see what she said about anticipation. I think anticipation is a normal part of clicker training and I have used it to my advantage, but it has also sometimes gotten me into trouble.
She started out by saying that most people view anticipation as a negative thing. In competition, anticipation can ruin a performance and a dog that anticipates can be difficult to manage in lots of different situations. But she said that she LOVES anticipation. For her, anticipation means that the animal knows the behavior, loves the behavior, and can't wait to perform the behavior. These are all good things and she doesn't want to squash them. What she wants to do is keep that energy and attitude, while teaching the dog to wait for the cue.
A lot of people think of anticipation in regard to trained behaviors and cues, but she gave some examples of using anticipation to train a behavior. With service dogs, they want them to touch a pole with a traffic button. They teach this by reinforcing the dog for touching the pole when guided there by the handler. But what they really want is for the dog to take them to the pole and they get this by rewarding when the dog anticipates and starts to pull a bit toward the pole. They use anticipation to get the behavior of taking the dog taking the handler to the pole instead of the handler taking the dog to the pole.
In training, she finds that anticipation gives her a lot of information about the dog and specific behaviors. Whether or not the dog anticipates gives her information about the status of the behavior (how well does the dog know it?, how does the dog feel about it?). Anticipation gives her information about how to proceed toward the goal with any individual dog. And lack of anticipation also gives her information. A dog that anticipates in one context, but not another is providing useful information. She also found that she could use anticipation to improve latency because if the dog was keen to do the behavior but learned to wait until cued, it would do the behavior more promptly than a dog that was not eagerly waiting for the cue.
Even though she views anticipation as a good thing, most behaviors do eventually have to come under good stimulus control, so it is good to have some strategies for what to do when anticipation is interfering with performance. She wants to address the physical anticipation (the dog actually doing the behavior before cued) so that while the dog is mentally anticipating and eager to do the behavior, it still waits for the cue. There is a balance here between maintaining the dog's enthusiasm and getting reliable behavior. So what do you do when the dog anticipates in a non-productive way? She presented four training options which were:
ignore it - ignore the uncued behavior, pause and then cue the behavior and reinforce for the cued behavior.
reward it - accept and reward the behavior, but make a plan for working on stimulus control.
ignore it - make a plan to prevent it (management, change of routine, etc...) without removing the desire to anticipate
ignore it - talk the dog through it. This means teaching some kind of "wait" cue so that if the dog wants to anticipate, you have a way to communicate "not yet."
She mentions that a lot of anticipation is just a cueing problem. Either the dog is cueing off something unintentional or the handler has been sloppy about reinforcing a behavior off cue or has been allowing the dog to do the behavior for multiple cues. Looking at cueing is usually the first step toward solving anticipation problems. With my own horses, I have found that horses are very quick to anticipate if you are working on a pattern or you have a set sequence of events that lead up to a specific behavior. The horses will start to cue off elements that are part of the set-up so I have to be very clear about what part of the sequence means "do it now."
She had some other interesting points about using anticipation to increase a dog's interest in a work session about which it had previously been ambivalent. The example she used was a dog that had gotten ambivalent about doing scent work with objects. The dog liked training sessions so by taking her time to set up the training session in a predictable way, the dog had time to get excited about the session and this carried over to the scent work.
Overall I found it a pretty thought provoking session. I learned early on that offered behavior (uncued) is often better than cued behavior, at least at some stages of the training. I was thinking about this earlier this year because I was wondering why horses sometimes offer so much behavior when we don't want it. We might not be asking for anything at all, or looking for something else, or actively working on stimulus control and the horse is sure it knows the right answer so it keeps on trying. This talk got me thinking more about the balance between anticipation and stimulus control. I did chat with her a bit at the end and she said the energy associated with offered behaviors is one reason she puts behaviors on cue really late in the process. It keeps the animal more involved and enthusiastic. I know with horses, cues often evolve out of how we shape behavior and unless I am free-shaping, I tend to have some kind of cue and some level of stimulus control early on. I guess the question is, do we lose something by doing it this way?
Or do we have to do it differently with horses for safety reasons? I will often let a horse practice something once I have cued and rewarded it in a session, so I relax stimulus control for a period of time. I think this helps and builds enthusiasm, but then I do have to put it back under stimulus control at some point. What is the effect on the horse of having stimulus control come and go? I find that tightening up stimulus control can have a somewhat dampening effect on enthusiasm. The horses seem to go into a more directed mode where they readily respond to cues, but I don't have that same feeling of "look what I can do." But maybe that's ok. There are differences in how dog and horse trainers structure their sessions and I would find it exhausting if my horses behaved like a lot of dogs for the duration of their training sessions. It seems to me there is some kind of balance here to create a level of enthusiasm that both horse and trainer are comfortable with and this is going to vary from horse to horse and session to session.
Friday Session 3: Kay Laurence - Microshaping
Kay Laurence is one of my favorite presenters at Clicker Expo because she is so creative in how she gets behaviors and she often has a different take on how to do things. I had attended her talk on Microshaping at a previous expo so this was not a new topic to me, but I wanted to see it again. I have also found that even if the topic is the same from year to year, the presenters add new information.
Even though the topic was Microshaping, Kay started off by saying that the most important thing is that clicker training has to be done for the good of the dog, not for the trainer's ego, or because the trainer is in a hurry to get to the end goal. For her, the process is what is important. From previous sessions with her and general conversation, I think she is constantly looking for new and better ways to use positive reinforcement so that even beginner trainers can develop correct behavior in their dogs and she is very careful to choose tools that are appropriate for the skill level of the dog and trainer.
She started by defining Microshaping which is "the development of accurate and refined muscle movements, which can be layered, chained or merged into complex behaviors." In both this lecture and the labs, she emphasized that we click for muscle movement. By focusing on reinforcing muscle movement, she got cleaner behavior and the dog ended up on a higher rate of reinforcement because muscle movement is a very clean and well defined criteria so it is easy to see and mark. I think she uses the term "Microshaping" to separate this way of training from the more general shaping. With Microshaping, she also found that the dogs developed more body awareness and a clearer understanding of what the click meant. Instead of randomly trying behaviors, the dogs themselves learned to be more deliberate learners.
She showed a movie clip of a dog being trained to touch an object through "normal" shaping and with Microshaping. In the "normal" shaping, the dog was reinforced for looking at, then touching the object (with both nose and paw). The goal was a paw touch, but in the early part of the session, any touch was reinforced. You would think that having a somewhat relaxed criteria would make the dog more successful, but instead the rate of reinforcement was lower. I don't remember the exact numbers but the success rate for the dog was below 50%. Then she Microshaped the dog to do the same thing. Instead of reinforcing for interacting with the object in any way, she reinforced the correct leg lift the dog would have to do to touch the object. At this point the dog was not even near the object. The dog figured out the leg movement quite quickly and then she just moved the object closer so the dog touched it as it lifted its leg. Once the dog got reinforced for lifting the leg and touching the object, the dog consistently performed the behavior. The success rate for this session was closer to 75% and she got a cleaner behavior. This was just an example of how by focusing on the muscle movement, she was able to make her shaping more effective and easier on the dog.
Microshaping refers not only to the process of shaping the dog to do the behavior, but also to how she adds in small changes in criteria at each stage of training the behavior so that by the time the behavior is finished, the dog is fluent in the behavior under a variety of conditions. She showed movie clips of dogs being trained for other behaviors and explained how she takes time to generalize early on so that the dog is really solid in each behavior before she adds the next step. She wants to create confident dogs and develop the strength of the behavior through flexibility during acquisition, not on completion. So she doesn't wait until the behavior is done before she adds distractions, changes in the environment, or other small changes. She introduces those as part of her regular training.
She spent some time talking about how her training sessions are structured and the different ways of teaching dogs. She doesn't use any one method for all her training, but instead chooses the best method for any given situation, and she thinks of these methods as falling on a continuum from luring to targeting to free shaping. Luring is the most directed and free shaping is the least directed. From one end of the spectrum to the other, there is an increase in degree of guidance and mental stamina required. Free shaping is hard work for the animal and you must pay attention to when the animal gets tired and loses focus.
She ended with some comments on the benefits of Microshaping which were that it develops a structured teaching approach, develops self-directed learning skills in both trainer and learner, and said success is measured by the rate of reinforcement, not the speed at which final behavior is obtained> In addition, error reduced learning leads to reduced stress.
I am going to talk a bit more about how she gets behaviors when I write about the labs (I went to two of hers). For now, I will just leave you with her final quote which is "Microshaping is not about shaping of impressive behaviors. It is about impressive shaping of behaviors."
Saturday Session 1: Kay Laurence - Mine
This was the first lab I attended at this expo. For those of you who have not been to an expo, the labs are a chance to see actual training with the faculty presenter coaching an attendee through an exercise or demonstrating it her/himself with one of the dogs. Some labs have more talking than others. Kay tends to jump right in and explain as she teaches so I am just going to describe the various exercises she used and how they changed the dog's behavior.
The theme for this lab was using play to modify the dog's emotional state and teach self control. The lab is titled "mine" which makes sense to me because one of her points is that the trainer controls the play. In order for play to be an effective training tool, the trainer is responsible for teaching the dog how to play safely and appropriately. The dog's mental state of the dog is under consideration at all times. Many trainers use play to increase a dog's energy. I would say that Kay uses play to teach a dog to work calmly but with energy and also to teach the dog to raise and lower its energy level.
One of the reasons play is useful is because dogs learn naturally through play. The learning could be about body movement, social interactions, or emotional control. One element of emotional control is learning to bring your energy levels up and down so they are appropriate for the situation. I wrote down that Kay says a dog can go from calm to arousal in 3 seconds, but can't do the reverse without training. The trainer has to teach the dog to go from arousal to calm. I didn't write down a time for this, but what I remember is that she doesn't expect to get from arousal to calm in 3 seconds in most dogs. She starts by reinforcing steps of self control, which is the first step in going from arousal to calm.
The first dog was an Australian Shepherd that gets overexcited at competition. Kay set up trotting poles and had an assistant put a piece of food between each one, so the dog went over a pole with its front feet, stopped, ate the food, and then continued over the next pole. First, they had food between every pole. Next they had food between every other pole, and then food only at the end of the line of poles. When they got to just having food at the end, they added the criteria that the dog had to be trotting the poles at an even pace. This was a reasonable change in criteria because the food placement in the earlier trials had encouraged the dog to evenly trot the poles, so it had been practicing this behavior all along. If the dog cantered or jumped them, it didn't get rewarded and was sent back the other way. The dog seemed to figure this out and by the end of the session, it stayed in trot even when it was excited about running to the end to get its treat.
The second dog was used to demonstrate how to use tug to teach self control. This dog was really into tugging and would jump up on her handler to try and get the game started. Kay had the owner be more clear about when the tug game was starting by changing her mechanics (body language/placement of the tug toy). She also had the owner be the one to end the tug game by a slight upward tug on the collar. The dog's reward for releasing was that she got to play the game again. If the dog didn't release, the owner just waited, holding the collar, so the dog wasn't really getting to continue the tug game because the owner was not participating. Kay also spent a few moments working on helping the owner to be more assertive about her role in the game. Kay views tug as a game between partners who both have jobs to do. The owner's job is to be clear about when the toy is hers (remember the lab name "mine" and when she is sharing with the dog. The dog's job is to show self control so that when tug is not an option, it waits until it is offered again.
The third dog was a little one (havanese?) that was preoccupied with food. I'm not sure what the training issue was, but from the start the dog was clearly orienting towards things other than her handler. She spent a certain amount of time at the beginning of her session on her hind legs trying to see up on to the table where the food was. Kay started throwing food for the dog to chase. She would whistle so the dog knew food was coming and then throw a sausage. I think the idea was to get the dog's attention and involvement in the game by showing her that it wasn't worth spending time on food that was unavailable, and that paying attention to the trainer was a better bet.
The initial throws were all in locations where it was easy for the dog to find the food. Then she threw some where the dog couldn't get it. In some cases, she threw towards someone and asked them to stand on the food. When the dog tried to get the food and couldn't, Kay whistled and threw more food in the opposite direction. The idea was to teach the dog to leave the unavailable food and go after the one she could get. I suppose in some ways, it was a version of "leave-it." Kay pointed out that this is also a great exercise for a dog that runs away from you. The dog learns you might have something better than what is after, and is more likely to turn and come back.
There was another dog who came in and worked on tug. In this dog's case, the problem was that the dog would rebite after she had released the tug. Rebites are a problem because dogs tend to rebite closer to the hand. Kay had the owner run away with the tug as soon as the dog released, a management strategy to avoid allowing the dog to rebite. Then she had the owner use the clicker to mark the dog waiting for the tug.
This was the first time where I think she used the clicker in the lab. She might have used it at the end with the Aussie, but I can't remember. How Kay decides when to use the clicker and when she doesn't is not quite clear to me. I haven't seen her train enough to be able to explain it, but I think she adds the clicker if she wants that precision or if she thinks the clicker plus food is going to motivate the dog more than food alone. She did say that if she can deliver food as the behavior occurs, she doesn't need to click. She wants to manage instincts in a natural way so she uses movement through play and natural reactions to get the behavior she wants. Once she is getting the behavior, she might add in the click to mark it. With this dog, she felt it was very important to reinforce that moment of waiting before the tug was presented again.
The lab was very interesting to watch and I think part of the reason Kay is so successful is that she is very creative, but also quite analytical about how she does things. And she knows dogs. I mean she really knows dogs. She seems to know what motivates different breeds of dogs and which exercises will work for one dog but not for another. The challenge for me is trying to figure out how to take what I learn from her to the horses. She uses a lot of play and food chucking and those both seem easier to do with dogs and horses.
At this point, the biggest take-away I am getting from her is thinking a bit more about when I need to use the clicker and when I don't. Are there ways we can jump start behaviors with horses (in a positive way) so that we don't use the clicker until we actually are getting clean and natural behavior? In some ways, that seems counter to the point of clicker training. But for some behaviors that an animal does naturally, it does seem like marking those first hesitant steps means the animal continue to offer them (they have a reinforcement history) whereas if you only clicked once the animal was confidently offering you more of the behavior, maybe you would create a more confident animal or more fluent behavior?
I don't know, I'm just thinking here, but it did make me think of how I have changed how I teach mat work. I now teach matwork as a directed activity because I never want to reinforce pawing, banging, etc.. at the mat. None of those are part of the final behavior. Instead I focus on teaching the horse to step solidly on the mat as it walks over it and right from the start, the horse learns that the behavior is to put the foot down on the mat and put weight on it. I find that I get horses on mats more quickly, more confidently, and I get less unwanted behavior by reinforcing only clean and solid steps. If you want to read about how I teach mat work, look in the articles section for one titled "Matwork and its Applications" which describes what I call the "walk-over" method.
I would be really interested to hear back from anyone who thinks they use some of Kay's ideas in their work and how they do it. I just have the feeling she is a great untapped resource for horse trainers and I'd love to hear if other people have explored her work or used similar strategies.
Saturday Session 2: Julie Shaw - Behavior Modification, Teaming up with Vets, Part 1
Julie Shaw has been a presenter at the last few Clicker Expos, but I had not previously attended any of her lectures. She is a Senior Animal Behavior Technologist at Purdue University and she works with vets there to evaluate and come up with training plans for dogs that are referred to Purdue for behavior issues. Please note that she gave two lectures (part 1 and part 2) and I only attended part 1.
She started by talking about how important it is for trainers, vets, owners and other professionals to work together. In her job, she works as part of a team and the success of each case depends upon teamwork. In these situations, it is very important that everyone knows and respects each other's role. The role of the vet is clearly defined as the person who rules out health issues, diagnoses behavior problems, prescribes treatment plans and medications, gives a prognosis, and adjusts the treatment plan as needed. The role of the animal behavior technologist is to advocate for the patient and owner, be the team leader to make sure everyone is communicating, assess (be the eyes and ears of the team), assist the client in applying the vet's treatment plan, guide, organize, clarify, and empathize.
She went on to talk about procedure with examples of cases studies and how they were resolved. Case studies are always fascinating because I think we all find it easy to relate to stories about real life and see connections to our own situation. One case study was about a dog that was anxious and this was resolved with medication and training. Another dog had developed the habit of biting his owner when asked to do something he didn't like. In this case, the owner learned to use positive reinforcement to train the dog to be more responsive to cues and the dog no longer had to be coerced to do things. Both those dogs were treated and able to stay with their owners. She had a story about another dog that was clearly in the wrong home and it was causing behavioral issues because his needs were not being met. In this case, the dog was rehomed and did well with the new owner who was able to use behavior modification techniques to work through the dog's issues.
The case studies did make me think about how complicated our relationships with our animals can be. There are so many factors that are important. The success of the relationship can depend upon the environment (busy vs. quiet house, kids?, amount of exercise, etc...), personalities of both owner and dog (anxious, timid, outgoing, high energy/low energy, etc...), knowledge of training, time spent on training, and so on. Horses are not the same as dogs, but we also have to juggle many possible factors when looking at behavioral issues.
I do want to emphasize two comments she made. The first one is that her primary goal is to promote and protect the human-animal bond. It's not about saving the animal's life. Working in a vet's office, there are a certain number of animals that come in with questions about if they can be rehabbed or if it is kinder/safer (to the public) to euthanize the animal. Her goal is to be realistic about what is possible. Can this animal be rehabbed and remain in the same home? Would it be better off in a different home? Is it too emotionally damaged to bond with anyone?
I think one reason this comment resonated with me is that I do encounter horses and people who are not good matches for each other. And while I want to help people work through the issues they have with their current horses (this can be a great learning experience), there are times when I just think it might be better to find the horse a more suitable person and the person a more suitable horse. I used to feel this was "failure" of some kind, but I don't feel that way anymore. I think we all accumulate baggage in working with individual horses and there is a time when you just have to recognize that the best solution is to give both horse and owner a fresh start.
Her other comment was that with every health issue, there is a change in behavior. I think at some level, we all know that. We often notice health issues because there are changes in behavior. But somehow just seeing this written out so simply, it made me realize how close the connection is between how we behave and how we feel. I think it's important to recognize that changes in behavior can be related to health issues and also to expect that health issues are going to lead to changes in behavior. It's important to be observant and understanding when a horse is dealing with a health issue, and if you have a persistent behavioral issue, I think it is important to actively pursue that there might be a health issue behind it. I know people how have had a gut feeling there is something wrong with their horse and it has taken multiple visits with different professionals to find out what is going on. Be persistent!
When I left the lecture, I wasn't quite sure what to do with the information she presented. I enjoyed the talk as she is an enthusiastic and entertaining speaker, but my own work with horses seemed pretty far removed from working in a vet office. However, I have had some time to think about it and I would say that the idea of teamwork and promoting the animal-human bond are two things that we do need to be aware of as clicker trainers, with our own horses and even more so if we are teaching others.
Most horse owners are part of a team that can include various members. If you board, the barn owner/manager/help are part of your team. Your vet is part of your team. Your hoof care professional is part of your team. Your riding instructor is part of your team. It is really important that they all work together to ensure the health of your horse and the health of your relationship with your horse.
When I go out and teach someone new, I am now part of their team. I may think of myself as a professional with a job that is separate from the rest of the horse's life, but what I choose to teach and how I choose to get involved can affect the whole team. One of the challenges of teaching is that sometimes I am asked to offer clicker instruction and I can see other areas that need improvement. If the hoof care is marginal, do I comment on that? What about saddle fit? Maybe we are only doing groundwork and they have a riding instructor, do I want to comment on saddle fit? How do I gently offer alternate views without alienating my new student or making other professionals feel threatened?
I think this is a tricky area. Some students are clearly open to suggestions and others are not. Sometimes I will wait until I have more information and I can re-evaluate to see if my initial assessment is correct. At other times, just waiting means looking for the right opportunity to present itself. I always feel bad for the student who is now put in the position of having to evaluate one professional opinion against another. I think it would be much nicer if there was a sense of teamwork between all the professionals involved with each horse. There are cases where this does happen and it really does work for everyone's common good.
It is not my intention to sound negative here. I do think there are situations where everyone does work together well so I am not saying all situations are difficult. But I think that what I got out of Julie's talk is that as an instructor, my job is to help my students work with me as part of their team of professionals. I also need to keep in mind that in addition to all the medical and husbandry work that is important, it is also important to see how the horse and owner are doing as a team. Are they working well together? Is this relationship positive for both of them? Clicker training is about building positive relationships and if there is undue frustration on one end or the other, then that needs to be addressed.
Saturday Session 3:
Jesus Rosales-Ruiz - Behavior chains
Jesus Rosales-Ruiz - Behavior chains
My notes on this session were not very complete so I wrote
up this little report using Dr. Rosales-Ruiz's PowerPoint presentation, my own
memory and what I did have written down. There are some parts of his PowerPoint
that I don't remember at all so I have not included them here. I'm not sure if I
missed information because I was not feeling well, or if he changed some of his
PowerPoint, but if anyone has anything to add or I have gotten information
wrong, please let me know.
Most clicker trainers are familiar with the idea of
creating behavior chains which are behaviors linked together so that the animal
performs multiple behaviors (usually in a set order) before getting clicked and
treated. Dr. Rosales-Ruiz's lecture was a closer look at behavior chains
including the components of chains, building chains, interrupting chains,
forward vs. backward chaining and behavior cycles.
Most clicker trainers are familiar with the idea of creating behavior chains which are behaviors linked together so that the animal performs multiple behaviors (usually in a set order) before getting clicked and treated. Dr. Rosales-Ruiz's lecture was a closer look at behavior chains including the components of chains, building chains, interrupting chains, forward vs. backward chaining and behavior cycles.
Behavior chains follow the general formula that one response leads to a stimulus in whose presence another response leads to another stimulus. For example, cue1 -> behavior1 -> cue2 -> behavior2 ->cue 3 -> behavior3. The first example of a behavior chain that I heard of in clicker training was teaching an animal to fetch. You can break fetch down into 3 gross behaviors which are chasing the object, picking up the object and bringing it back. The dog has to do each step in sequence and completion of each step cues the dog to do the next step.
As noted above, behavior chains are composed of behaviors and stimuli (a cue is a stimulus). Each stimuli in a chain functions as both a conditioned reinforcer (to reinforce the preceding response) and a discriminative stimulus when it occasions another response in its presence. He had examples of behavior chains that were broken down to identify the stimulus, response, and reinforcer. In a chain, the reinforcer is part of the chain (not a separate event like a click and treat) and usually the reinforcer is getting to do the next part of the chain, or something that is intrinsically rewarding about that behavior.
I always think of chains as being composed of multiple trained behaviors such as come, sit, look at me, but even a simple 3 behavior chain like that is really composed of lots of behaviors. It all depends upon how minutely we are looking at what is happening to maintain the chain. As an example, he showed how a simple behavior such as "sit" is actually part of a behavior chain if we ask the dog to sit, click, and treat. The cue (discriminative stimulus) is the trainer saying "sit," which is reinforced by the click. The click is a cue for the dog to go to the feeder. The proximity and sight of food at the feeder reinforces going to the feeder, and the food in the mouth reinforces taking and biting the food. You could write this out as cue1 (sit) -> dog sits -> cue2 (click) -> dog goes to feeder ->cue3 (proximity and sight of food) -> eating food. The click, proximity and sight of food all function as both cues and reinforcers.
While we don't always have to look at every tiny behavior within a chain, I think if you understand the dual nature of cues and how a chain can be divided into even smaller chains, then you will find it easier to understand why chains sometimes fall apart as well as how to interrupt chains if you create unwanted behavior chains. Dr. Rosales-Ruiz showed the example of the dog from the jackpot study who got confused by a change in food delivery. The dog did not get reinforced for targeting because he didn't understand the new food delivery system (the food went into his bowl from a tube instead of being tossed by the trainer) . This disruption in food delivery not only caused the dog to stop targeting, but the dog actually left the training session. You can read more about this experiment in a previous ClickerExpo report on jackpots.
He had more examples that showed chaining, backchaining, adduction, and Microshaping. Then near the end he showed some examples of unwanted chains where a behavior has unintentionally gotten included in a chain. This is easy to do if the animal is redirected to the correct behavior and then reinforced. You end up with a chain that is the unwanted behavior and then the wanted behavior. In one example he showed a horse that had some stress behaviors that had gotten reinforced as part of a chain. By having the handler wait for longer before asking for a different behavior, or cueing the horse before he started the stress behaviors, the chain could be interrupted.
One other topic he covered was the concept of movement cycles. If you look at behavior as a movement cycle, then the behavior is not done until the animal is in position to do another one. He used the example of sitting. You can't ask someone to sit multiple times without asking them to get up in between. When we train our animals we need to be aware of setting up movement cycles so that the animal is in position to start again as soon as reinforcement is delivered. If you have watched trainers who are clever with food delivery, one thing they will do is use the food delivery to reposition the animal so it can immediately perform the behavior again.
The last example was of a dog being trained to go out to a stool and back. For the first sessions, the trainer (Mary Hunter) reinforced the dog for going out and she clicked when the dog was doing the correct behavior. When the dog was reliably performing the behavior, the click was moved to earlier in the chain so that Mary was clicking when the dog was three quarters of the way out.
If the click marks the behavior, the dog should have stopped and turned at the click, but in most cases, the dog continued with the previously reinforced behavior of going all the way out and around. After enough trials, she did sometimes turn before reaching the stool, but every now and then she would go the way out and around. Then Mary started clicking for only half way out. Again the dog continued to go out even after it was clicked, but they did see it start to turn earlier over time. What was interesting was that by moving the click, the initial behavior deteriorated, but they started to see other behaviors so they actually got more variability in the behavior. In the last clip he showed, the dog would sometimes go all the way out, sometimes go half or three quarters and I think they were even seeing some variations on the turn as that was getting clicked some of the time too.
I think creating chains is one of the easiest ways to get more behavior for each click and treat. I have played around with chains with my own horses and I think the horses are quite happy to do more behaviors if I build it so that they know what they need to do before they get clicked and treated. But chains do sometimes seem to create some interesting problems and while they are easy to build, I sometimes find them hard to maintain. It is not uncommon to have elements drop out of a chain because the horse anticipates or something subtle changes that affects the cues or reinforcers. As an instructor, I find that most people's introduction to chains is when they inadvertently create chains they don't want or when the horse creates a chain to manipulate the handler into doing what it wants. I think horses are good at picking up on patterns and trainers need to have at least a basic working knowledge of chains so they can keep one step ahead of their horses.
Sunday Session 1:
Sunday Session 1:Kay Laurence - Super Skills (lab)
This was my second lab with Kay Laurence. In this lab she showed how to use the dog's natural response to food or a toy to start behaviors and also offered some help on how to set criteria and shape behavior while keeping the dog successful.
She has some general rules for training and how to evaluate your dog's progress and your training plan. First you need to decide upon your criteria before you start training the dog. I know it is easy to go into a training session with the idea of seeing what the animal offers and going from there, but when she is doing a shaping session, she is very clear about the criteria ahead of time. This is important because it allows her to evaluate how the dog is doing. If the dog hasn't met her criteria in 2 repetitions, then she knows she needs to change the criteria. How does she know when the dog is ready for a change in criteria? When it meets criteria 5 times in a row without hesitation. She wants to build confidence into her dogs so she emphasized that it is really important to make each step solid before changing the criteria.
When she talks about changing criteria, she is not necessarily talking about asking the dog to do "more" behavior. A change in criteria could be an environmental change such as the handler moving, a new location, an added person or object and so on. She starts with simple changes and this helps the dog gain confidence in doing the new behavior under a variety of conditions. This idea of making each step in the behavior very solid was something she talked about in the Microshaping lecture so this was a chance to see how to do this during a training session.
The first dog she used as a demonstration dog and after that she had 6 dog and trainer teams working on separate behaviors on their own (with some coaching help). At one point she noted that at her training classes, they have teams of 6 handlers for one dog. Each person on the team has a job (one working the dog, one taking notes, one observing the dog, one observing the handler and so on....) I have to say that when I heard that, I was envious. Most of the horse trainers I know work alone. The idea of having even one extra person seems like it would be a huge help.
The first dog was going to be taught to go out and come back around a cone. She started by throwing food to get the dog nicely trotting out and back. At one point they had to change to a different type of food as it was breaking into little pieces which was causing too much searching behavior. Once the dog was reliably going out and back, she put a cone up near her chair. At this point, she actually wanted the dog to ignore the cone. If the dog orients to the cone with its nose or front end, it won't go around the cone in a natural manner so step 1 is teach the dog to ignore the cone. Then the cone is moved slightly over time so that the dog has to go around it to get back to the trainer, but as far as the dog is concerned, the exercise is not about the cone. The cone just becomes something it has to go around to get to the trainer.
The dog was progressing nicely, but at some point it started stopping before getting to Kay. It was not clear why, so Kay moved her chair back to see if the dog was stopping a certain distance from her, or if there was something else. After a few more repetitions, it became clear that the dog was stopping next to the cone. Kay suggested that there might be something in the dog's previous training that was making the dog more likely to stop at the cone so they finished the training by reinforcing that (rather than work against what the dog already knew).
The next part of the lab was with the teams of dogs and handlers. I couldn't see all the teams working but one was training a dog to drop a toy in a hula hoop, one wanted a dog to come toward her and lie on a mat, one was doing paw targeting, and one was being taught to stand on a platform (I think). There was another dog that the owner wanted to teach to back up and put its foot through the hole in a cone. I found this sort of funny because the owner described the behavior and Kay looked skeptical and said "would you want to put your foot through there?" in a sort of disbelieving tone. The owner said "sure" and moved on to training it. A while later Kay came to see how they were doing. They were not having much success as it was too hard for the dog to position itself correctly and the seemed reluctant to place its paw through the hole. I have to confess that if Kay questioned my training choice in that tone of voice, I would have changed behaviors pretty quickly.
Kay went around and coached the teams. She helped them break down criteria (there were quite a few lumpers) and think about how to set the dog up to be successful. One session I did watch was on teaching a foot target and it was similar to what she showed in the video on Microshaping. She got the dog stepping forward and back through food placement and then added the target by placing it where the dog had been stepping. Within a really short amount of time, she had a reliable and clean foot target. She said she gets the muscle pattern first and then adds the click.
She had some other comments on clicker training and what's new in clicker training. She said that it is always better to not click if you are not sure. Knowing when NOT to click is as important a skill as knowing when to click. She now starts students without a clicker and they don't get one until they show her they are ready for it. When you do use the clicker, don't click until you get what you want. When you do get what you want, ask yourself, "do I want to click this or can I improve upon it?"
If you read my report on the other Kay Laurence lab, you will see a lot of similarities. In both labs, she used food or play to get a behavior started instead of just free-shaping from nothing. But she does free-shape from nothing when she is microshaping, so I don't want to leave you with the idea that she always uses food or play to get the dog started. I think in some ways, the food chucking is a way to see what the dog is going to give you to work with. Yes, she comes in with specific criteria, but I think that is because she has a lot of experience and knows what kinds of things she can look for.
I think what I might do is try to find ways to generate behavior that allow me to observe and experiment a bit with my horse's behavior without getting into the kind of training mode where the animal is already focused on moving a certain body part a certain way. I am sure that in some cases, the dogs Kay are training have no idea what behavior is being reinforced. They just know they are being reinforced and are willing to play along.
Sunday Session 2: Jesus Rosales-Ruiz - Generalization: Scientifically explored
When training behaviors, we want animals to learn to generalize so that the same cue will produce the same behavior in many different situations. This lecture focused on how we teach our animals to generalize and what generalization means from a scientific viewpoint. Dr. Rosales-Ruiz started with a movie clip about the importance of generalization and talked about how generalization occurs.
Dr. Rosales-Ruiz spent a bit of time on how scientists study and measure generalization. It gets a bit technical and I can't say I followed all of it, but there were a few interesting points that stuck with me. One is that scientists will draw a generalization gradient which measures how an animal responds to similar stimuli and can be used to measure how much a stimulus has to change before it is no longer recognized and the animal's behavior changes. I think this is an important concept because it means there is variation in how animals respond to similar stimuli.
They are going to respond to the "correct" stimulus with the most, but they are also going to respond to similar stimuli at least some of the time. This means we can teach an animal to "expand the stimulus class" by reinforcing correct responses to stimuli that are slightly different, but that we consider part of the same stimulus class. So we have choices. We can train an animal to respond to a very specific, well defined stimulus or we can teach the animal that even though there can be variation in the stimulus, the stimulus should prompt the same response.
There is another kind of generalization which is based on functional classes so animals learn to group stimuli together by the associated behaviors. The example he used for this was that we all learn to associate "stop" with multiple stimuli such as stop signs, red lights, train crossings and policemen. I'm not quite sure how this ties in with animal training except that it is the same idea as having multiple cues for the same behavior.
In generalization, the animal has to learn to recognize the cue despite other changes in the environment. We often think of this as happening naturally through the training process, but it is much more likely to happen if we train specifically for generalization. To do this, we want to train so that the animal has the opportunity to learn which stimuli are irrelevant and which are important. He offered some guidelines which were to train loosely, teach sufficient exemplars, and program common stimuli.
By "train loosely," He means to train in a way that exposes the dog to minor changes in the environment so the dog learns that these minor adjustments are not meaningful cues about which behavior is expected. This is what Kay Laurence does when she makes small changes in criteria at each step of learning a behavior. For example, if you train a behavior by standing in different positions around the dog, the dog will learn that your position is not part of the cue.
It is easy to get stuck in patterns of how and when you teach behaviors and this can make the dog less likely to generalize. But even if you are stuck in one training environment, you can vary the environment by changing the decorations, furniture, locations of other items, order in which you ask for behaviors, temperature, smells, time of day - really anything you can think of. I find that the presence of other people or animals is often a huge distraction so it is a good idea to try to vary who you train with or who is present.
Other things you can do are to use two or more teachers, teach in two or more places, teach from a variety of positions relative to the dog, vary your tone of voice, dress differently, train in places that are darker/brighter, or noisier/quieter. That doesn't mean you can't have cues where position or tone are part of the cue, but if they are not, then you want to go through the training process of showing the dog that they are not part of the cue.
He talked about "programming common stimuli" which means that as you vary parts of the environment, you want to make sure that you are consistent about keeping some stimuli (those that are part of the cue) the same. In a way generalization is about teaching the dog what is part of the cue and what is not. Your job is to teach the dog that the same cue means the same thing even if other things change. This may mean rethinking your cue to make it more easily transportable to new environments.
He had a series of movie clips where a dog was taught to stand on 4 pods. A pod is round like a ball on one side, but flat on the other, so it can sit flat on the floor with the round side up. The dog was taught to put a foot on each pod, first in one location and then in other locations that required slightly different body positions. Regardless of where the trainer placed the pods, the behavior was always the same and the movie showed that the dog figured out how to get all 4 feet on the pods more quickly in each new location.
I found this lecture was a good reminder of how important it is to practice behaviors under lots of different conditions before you assume the dog knows them. I have a tendency to train the same things in the same places and I find that in the beginning of the training process, this actually speeds things along as horses are very context sensitive. If I ask for the same behavior in the same place a few times, that location becomes part of the cue. Later I will move to new locations. Now I am wondering if I should be making more small changes as part of the training process, at least for some behaviors that I will want in lots of different places.
Sunday Session 3: Kathy Sdao - The All-Seeing I: Improve Your Observation Skills
This was the last session of the expo and I debated about which session to attend. I chose to go to Kathy's presentation because she is such a good speaker and I had not been to anything of hers yet. The subtitle for this session was "improve your observation skills" and I was curious to see what she would say about that.
She started by talking about the importance of timing. She repeated something Karen Pryor described to her about how behavior is like a river, constantly changing and flowing along with one behavior connecting seamlessly Into the next one. When you try to mark a specific behavior, it is like shooting an arrow into the river. You have to be able to see the flow of behavior in order to have the right timing to mark the behavior you want. If your timing is slightly off, you end up marking a different behavior. Kathy said this image has always stuck with her and reminds her that not only is timing important because it is important to mark the correct behavior, but because it reminds her that even if the click is "off target," she is still marking behavior.
We would all like to have good timing, but the reality is that most of us have a limited reaction time. There are various games you can play on the internet that check your response time when the screen changes color or something like that. While most people can improve their timing, there is always going to be a lag between the event and when we click. She shared some numbers about reaction times on these tests for college students vs. elite athletes and so on, but the bottom line was that there is a limit to how fast our nervous system can send the impulse from our eye to brain, and from our brain to our fingers. In other words, there is a point beyond which we don't get more accurate timing by learning to click faster.
So how do we get more accurate timing? We get it by observing and learning to predict when the behavior is going to happen. Timing improves if you can predict behavior. Most clicker trainers with good timing are reinforcing behaviors they have seen many times before, in many animals and they have learned to recognize the precursors to the behavior they want. For example, Kathy is good at clicking dogs for sneezing because she recognizes the way a dog's face changes right before it sneezes. If you want to reinforce leg movement, you learn to be good at recognizing weight shifts or the first sign of leg movement so you click as the leg starts moving. This goes back to the importance of being good observers. We learn to predict behavior by being learning to see patterns in behavior
Kathy now emphasizes the importance of seeing with her mnemonic for good training which she calls "Get SMART." SMART training stands for See, Mark And Reward/Reinforce, Training. She said she added the "see" part when she realized that many novice trainers actually can't see behavior happening because they haven't learned what kinds of things to look for. Or sometimes they have some kind of mental block against seeing it. She shared some stories about owners seeing what they expected to see and not what was really happening. I think this happens all the time. We see what we expect to see and that is why it is so helpful to have a fresh set of eyes when we get stuck.
She shared a few movie clips to test how well we saw. One was a clip I had seen before which shows kids playing basketball when a person in a gorilla suit walks through. If you are given a task such as counting throws to do while you watch the clip, you won't even notice the gorilla. I had seen this before and I found it actually took a lot of effort to count and look for the gorilla and I couldn't really do both. She had a number of other visual tests to see if we could notice changes in a picture over time or differences between two pictures.
We tend to assume that what we see is accurate, but it turns out that what we see is influenced by lots of factors. Some of the ones she identified are preconceptions and labels, judgments and analysis, talking and prompting, and the audience effect. In order to see well, you have to turn off the part of your brain that makes judgments based on previous information. You also have to turn off the part of your brain that analyzes things. You have to just concentrate on seeing. It turns out that even talking makes you see less well.
She ended with some more visual tests and a bit about how we can help our students learn to see. On her power point, she says:
Looking is not the same as seeing.
I enjoyed the lecture and it reminded me again of how hard it is to time your click correctly when you are first starting out or trying to shape a new behavior for the first time. I even notice that with new animals, it can take me a few tries to get my clicks in synch with the new animal. Some move faster than others and they all have slight variations in how they do each behavior. I have also noticed that the longer I have been clicker training the same animal, the easier it is to be precise about my clicking because I know that animal so well.
Karen Pryor talks about needing "10,000 hour eyes" (I think it was 10,000 - a big number at any rate) before you can really see behavior clearly. This is just experience, but it is also the learned skill of being able to really observe without expectation, just an awareness of all the possible things an animal can do.
If you have questions on something that I have covered here, you can email me for more information (email@example.com). If you want more information on a specific topic or faculty member that I have mentioned here, most of them have web sites and/or written resources available. A list of the faculty members is available at www.clickertraining.com.
Katie Bartlett, 2011 - please do not copy or distribute without my permission