Dealing With Errors

There is no perfect way to deal with an "error" in training. Once the dog is doing something other than what you were hoping to see--whether you are building a new behavior or practicing a "known" one--every strategy has potential pitfalls.

Let's look at the main options.

There's punishment. Hurting or scaring a dog are off the table for me--but I might close a door that a dog is walking toward so they can't go through it, or move food away from the edge of the counter so a dog can't reach it, or slow my pace on leash so a dog can't charge forward. (Or I might not . . . keep reading.) Those are examples of "negative punishment," where the consequence for a behavior is the removal of access to positive reinforcement. But even mild penalties can poison cues or dampen the dog's enthusiasm for working with you--particularly if you find yourself penalizing over and over.

And then there is extinction--aka nonreinforcement. You have probably heard the common advice to just reward what you like and ignore what you don't. But it's not quite that simple, and dosage is key: if the rate of reinforcement in your training drops too low for too long, your dog may just leave.

Of course, that's not usually the first thing that happens when you hold out for too much too soon. When a previously reinforced behavior doesn't work, what you will see is other behavior that has worked in similar situations (the technical term for this is "resurgence"). If you don't respond to an error, and the dog has a strong or recent reinforcement history for the thing you were looking for in this situation, then it may shuffle back to the top quickly, and you can reinforce it. But if not, you may see what's called an "extinction burst," a flurry of variability and intensity that might include some behavior you would probably call "frustrated." Maybe in that burst you'll also see the behavior that you were hoping for, but it may come with outriders like snorting, huffing, sneezing, barking, or tippy tappy feet. Both punishment and extinction can even provoke aggression.

Even if your dog is not easily frustrated, you may want to avoid repeated "ignoring" of errors. Say you're training your dog to do both a paw target and a nose target. You are working on the paw target today, but they target with their nose instead. When you don't click the nose target, then they try the paw (resurgence!), which you reinforce. When that kind of thing happens, it's key to watch carefully what they do on the next rep. If they just do the paw again on the next round, you might be on the right path, but if you find yourself repeatedly swiping left on the "wrong" behavior to get to the "right" one, you may be starting to train the dog to do the whole sequence--nose then paw, nose then paw.

Redirection/resetting is often the best of our imperfect options. It keeps the dog engaged in the training session, and you can use it to get the dog into a position from which they are more likely to be successful. But do it too often and you may end up reinforcing the error more than the "right" behavior with your "reset cookies" or your go-to cues. Example: your dog lays down when you give the sit cue, so you toss a treat or give him a hand target to get him up, which you maybe also reinforce with a click and a treat. Do this once or twice, occasionally, and it's not likely to create a real problem. But do it over and over and over again and you are doing a nice job training your dog to lie down on the sit cue.

What you really want is not just a high rate of reinforcement but a high rate of reinforcement for mostly the desired behavior. If you are frequently repeating any of the strategies above, that is not what is happening.

Though I have my obvious preferences, the reality is that I might do any of these imperfect options depending on the specifics of the situation. But so long as you're not hurting or scaring your dog, then what you do at the moment an error is already occurring probably does not matter as much how often you do it or what you do next.

So, TL;DR, are some things you might do in the moment when a dog does something other than what you hoped:

• Remove, diminish, or delay the reinforcer if safety requires, or if the behavior getting reinforced even once would really set things back. But then adjust your plan.
• Wait a few seconds and see what else the dog offers, if you're pretty sure it will be what you want. If it's what you want, or something that sets the dog up to be more successful next time, reinforce that, but then adjust your plan. (Many trainers like to teach a "default" behavior of looking at you for further instruction--default meaning that it is so heavily reinforced in so many situations that it is likely to resurge first.)
• Don't give the reward, but neutrally bring them into position to try again, if you think they're likely to get it right. (No guarantees that this won't reinforce the error, but you can take an educated guess.) If they don't, adjust your plan.
• Just reset promptly, without waiting or dithering about whether you're reinforcing the error--but do it strategically. Do it fast, so you are likely to reinforce less of the "wrong" behavior. And do it so that when the dog looks up again, they are in a better position to start the "right" behavior than they were last time. But then adjust your plan.
• If you're not able to make these calls in real time, end the training session, give the dog something else to do, and take a longer break while you . . . you guessed it, adjust your plan.

Note the common theme here is "but then adjust your plan." Here are some starting points for thinking about how:

• How can I tweak the environment to better facilitate the "right" behavior?
• What might have cued the "wrong" behavior? Do the conditions look like they did when that behavior was reinforced? (I see this often when people try to capture two different behaviors without changing their position, the props, the dog's starting position, or where they are delivering their treats.)
• Are your criteria too high for the situation? (Spoiler alert: If your rate of reinforcement for the right thing is low, it's a good bet they are. Change the situation, or change your criteria.)
• If you're cuing the behavior, is your cue unclear, or perhaps not what you think it is?
• Is there another cue in the environment for a behavior that has been reinforced more than the one you are trying to ask for or teach?
• Is your reinforcer valuable enough for the situation? (But note that there are many other things to look at before upping treat value! If the dog doesn't understand what to do, he still can't get the reinforcer.)
• Is the way I'm delivering my reinforcer, or where I'm delivering it, helping or hurting my cause?
• Does your dog have a competing need (e.g., to eliminate, to eat, to relieve pain or discomfort)?
• Are there any prerequisite skills or pieces of the desired behavior you need to pop into the dog's repertoire on before you can expect them to pop out when you don't reinforce what they tried? What do you want to see resurge?

I may have missed some considerations in trying to recreate my thought process, but the most important question of all is: How are you going to use this error as information to change what you are doing?

What is "Differential Reinforcement of Other Behavior"?

You keep using that word. I do not think it means what you think it means.

At the time of this posting, there were a couple different explanations of this concept floating around dog-pro social media recently, all of which perpetuated common misconceptions. Having written a chunk of my thesis about what a differential reinforcement of other behavior (DRO) procedure looks like, why it's thought to work, and what best practices for using it may be, I thought I'd throw this into the mix.

Although DRO stands for "differential reinforcement of other behavior," it does not involve reinforcing "any other behavior" that occurs instead of the unwanted behavior. In fact, it doesn't necessarily involve reinforcing any behavior at all.

In DRO, reinforcement is contingent only on the passage of a certain amount of time without the occurrence of the unwanted behavior.

There are two main flavors of DRO, both of which can be split into subtypes depending on how the time intervals are set (e.g., the same every time or varied from rep to rep).

In "interval DRO," which is more common, the whole interval of time must go by without the behavior that you want to reduce. When time is up, reinforcement is delivered, regardless of what else the learner is doing.

If the learner does the unwanted behavior, typically the time requirement starts over. Sometimes, if the unwanted behavior occurs right as time runs out, extra time is added to avoid reinforcing the unwanted behavior.

In "momentary" DRO, reinforcement is delivered if the behavior is not occurring at the moment that an interval of time ends, regardless of whether the unwanted behavior occurred at any other time during that interval. If the unwanted behavior is occurring, the reinforcer is not delivered. This appears to be mostly used when it would be too hard to monitor the learner for the whole interval (i.e., if you have a classroom with 30 students and can't watch them all at once). It's also probably less effective than interval DRO.

Delivering the reinforcer when you see any of multiple acceptable alternative behaviors occurring--which is how I have seen DRO described in training circles sometimes--would be differential reinforcement of alternative behavior (DRA). That's a different procedure, in which reinforcement is contingent on the occurrence of specific behaviors, and withheld or diminished somehow when the unwanted behavior occurs.

Although it's referred to as a differential reinforcement procedure, reinforcement may not always be at work in DRO. There are at least four mechanisms thought to make it go; these are some of the most commonly discussed:

1. Satiation. Best practices in DRO are to start with very low time criteria, so the start of DRO usually involves reinforcement presented densely, which may temporarily diminish the value of the reinforcer and therefore make behavior whose purpose is to get that reinforcer less likely to occur.

2. Extinction, or the elimination/weakening of the relationship between the unwanted behavior and the reinforcer that maintains it. What we traditionally think of as extinction is the non-availability of reinforcement for a previously reinforced behavior, and DRO typically involves some of that. But noncontingent reinforcement (giving the reinforcer on a time-based schedule whether or not the behavior occurs) can also weaken the contingency between the behavior and the reinforcer, in essence making the behavior seem unnecessary.
   In fact, as Susan Friedman pointed out once during a joint presentation, DRO without extinction is identical to noncontingent reinforcement. Which is an antecedent intervention--something you do before the behavior occurs--not a reinforcement procedure.

3. Punishment. Yes, that's right--DRO might run on punishment, particularly if the learner can tell that you are resetting the time. A signal that predicts an upcoming delay to reinforcement can become an aversive stimulus (one a learner will behave to avoid or escape). If something tells the learner that reinforcement has been delayed (a timer is reset) contingent on their behavior, that something can become a conditioned punisher, or punisher by association.

4. And, finally, DRO may work because of accidental reinforcement of alternative behaviors. Whatever is happening when reinforcement gets delivered might get strengthened. But it takes consistency to develop a contingency (a relationship) between a behavior and a reinforcer, and that doesn't always happen when you're delivering reinforcement contingent on time elapsed. Evidence for reinforcement of other behaviors during DRO is sparse, and in my research and my experience with running DRO with clients, the alternative response that emerged during DRO was not one that was occurring with any regularity when the timer went off/reinforcement was delivered in the early stages.

This is not a complete investigation of DRO--I just wanted to clarify the definition for my fellow pros and other interested parties. But we could also talk about why it's often slower than DRA, why some people call it labor intensive or think it's a procedure to be avoided, the potential advantages of interval vs. momentary, best practices for its use, etc.

You can get more deets in any applied behavior analysis textbook. Here I used Cooper, J. O., Heron, T. E., & Heward, W. L. (2020). Applied behavior analysis (3rd ed.). Pearson.

But here are other sources related to the "mechanisms" section: Jessel, J., Borrero, J. C., & Becraft, J. L. (2015). Differential reinforcement of other behavior increases untargeted behavior. Journal of Applied Behavior Analysis, 48(2), 402–416. https://doi.org/10.1002/jaba.204

Jessel, J., & Ingvarsson, E. T. (2016). Recent advances in applied research on DRO procedures. Journal of Applied Behavior Analysis, 49, 991-995. https://doi.org/10.1002/jaba.323

Langford, J. S., Pitts, R. C., & Hughes, C. E. (2019). Assessing functions of stimuli associated with rich-to-lean transitions using a choice procedure. Journal of the Experimental Analysis of Behavior, 112(1), 97–110. https://doi.org/10.1002/jeab.540

Poling, A., & Ryan, C. (1982). Differential-reinforcement-of-other-behavior schedules: Therapeutic applications. Behavior Modification, 6(1), 3–21. https://doi.org/10.1177/01454455820061001

Rey, C. N., Betz, A. M., Sleiman, A. A., Kuroda, T., & Podlesnik, C. A. (2020). The role of adventitious reinforcement during differential reinforcement of other behavior: A systematic replication. Journal of Applied Behavior Analysis, 53(4), 2440–2449. https://doi.org/10.1002/jaba.678

Thompson, R. H., & Iwata, B. A. (2005). A review of reinforcement control procedures. Journal of Applied Behavior Analysis, 38(2), 257–278. https://doi.org/10.1901/jaba.2005.176-03

Minimizing Frustration When Reinforcement Is Not Available

One of my “particular interests” is in how to minimize extinction (where behavior that has worked before stops working), and its unpleasant side effects (often labeled “frustration”), in situations where a lack of reinforcement is pretty much unavoidable.

Lots of long duration, persistent barking--if we can agree to call that "frustration"--has been shaped up by a combination of ignoring (extinction), either intentionally or simply because people are busy with something else, and then finally redirecting (reinforcement). The dog learns that the situation (i.e., person is talking to a computer or phone, guests are over, people have their hands full cooking) is when this type of barking is needed or will work to produce an outcome they want or need.

As some know, I’ve written about, talked about, and even done a little research toward the question of how we might reduce the amount of frustration that a dog will experience when, inevitably, a person can’t respond to their requests right away.

One way I think holds great promise (and it has a research base with humans already) is to teach the dog signals for when the behavior will work and when it won’t, or when there will be a delay that they need to wait through--sort of like telling a kid "I'll be on the phone for 20 minutes; do what you want within reason but I won't be able to talk to you or get you a snack until I'm off unless it's an emergency."

Since we can’t simply state “rules” to a dog, we have to teach them what the conditions for reinforcement are (and aren't) through experience. That means looking at all the “cues” that may currently tell the dog barking will work and changing them in some salient way before changing what we reinforce and when. (A lightbulb about changing the environment first went off for me during a talk by Jesus Rosales-Ruiz called "Don't Fight Extinction" at ClickerExpo a few years back.)

In the conditions under which the problem already occurs, you are likely to see the dog keep trying even after you stop reinforcing, because when their behavior has worked on a slot-machine type schedule, it’s hard for them to tell that it’s no longer going to be effective even sometimes.

One approach I've tried in the past is to add something to the problem situation for the new rules to get attached to. The change in the antecedent conditions can help clarify when reinforcement for the unwanted behavior is not available, and also ideally become the cue that new, more desirable behavior will be reinforced a little later.

But another, arguably less intrusive approach is to look for conditions that already cue the behavior you want, and see if you can transfer them into the situation where the dog barks.

Earlier this year, I discussed this idea with a new client whose dog persistently barked at, climbed on, pawed, and licked her while she was trying to talk on Zoom in her dining room. He had apparently done well as she Zoomed through COVID lockdown, but when Zooms became a less common occurrence again, this behavior started. The first thing we tried was removing all the toys, treat containers, etc. from the Zoom setting, but the behavior started right away and went on for probably half an hour during our evaluation, growing more frantic.

I should add that this pup leads a lovely and enriched life, with lots of off leash time in the country and attention and training from his human. She had tried preventively reinforcing a settle during Zoom with treats, and giving food toys or chews in advance of calls, but as soon as the goodies ran out, the dog would go back to these behaviors, and because she was trying to work on Zoom she would feel compelled to quiet him by feeding again or refilling the toys.

After our discussion, she came up with the critical information we needed: When she cooked or baked in the kitchen, she had never reinforced her dog for settling with food or done much training at all; she simply had gone about her business regardless of what the dog was doing, and the dog had typically laid down near the warm bottom of the fridge or on a doormat while she worked. I loved this, because she didn’t take away a recipe from our discussion—she took away the core concept and was able to apply it to identify a situation in which the behavior she wanted was already occurring.

So we moved her laptop and a chair to her kitchen utility cart, and in 12 minutes his barking and climbing decreased quickly in intensity and he laid down on the doormat. I’m kicking myself for not recording that session, but above is a screenshot.

Doing Zooms at the kitchen table instead of the dining room table was a comfortable alternative for this client right away. And eventually, she was able to transfer the scene back into the dining room, where she has so far been able to take short Zoom meetings again.

How to Raise Criteria in Shaping

Shaping—teaching a complex behavior by reinforcing behaviors that successively look more and more like your final goal--is hard to describe, and to prescribe. When it goes well, it can feel like magic--whoa, she started clicking and treating for just looking and now the dog is going over and climbing onto the thing they were looking at and sitting on it! And if you're a casual observer, it can seem very mysterious how the trainer got from phase 1 to phase 3 (to borrow a metaphor from South Park):

It's often mysterious even to the trainer who did it, leading some to refer to setting criteria as the "art" part of the "art and science" of shaping. But science informs even art--blue paint looks blue because of its chemistry, and behavior moves because of environmental conditions.

There are some guidelines floating around out there about when to start waiting for more behavior to reinforce--like when you’re reinforcing the current approximation at a high rate, say, 10-15 clicks per minute, or when your learner has offered the current approximation on some high percentage of opportunities to do so--but the problem with these is that they don’t tell you how many opportunities to give or how many minutes to train at each level, or make it clear what is supposed to make the learner go from offering the current approximation at a high rate to offering something beyond it just because you decided it's time for them to do so.

So here are some thoughts on why the next approximation might happen--and how to make it more likely to happen if it's not.

Use a high rate of reinforcement to engender “discretionary effort.” Discretionary effort is a phenomenon observed by Aubrey Daniels and others (more research needed!) in the context of workplace behavior. Daniels describes it as going beyond the minimum required for reinforcement, and attributes it to frequent positive reinforcement. A high rate of reinforcement requires easily achievable criteria, so this is the good ol' practice of splitting so you can click/treat often and keeping your eyes open for what you want to start happening.

Reinforce variability itself. Most if not all behavior is performed with some naturally occurring variability between performances (something I learned from Susan Friedman, who sends along this food for thought), perhaps the better for the environment to be able to select some variations over others. Sometimes our high rate of reinforcement approach can narrow this variability—if you reinforce a very specific approximation a ton, you’ll see more of that and less of other variations. But some very good shapers I know (cough Hannah Branigan cough) often look like they are reinforcing absolutely everything during shaping. What they may be doing is reinforcing variability itself, which gives you more to choose from--including some reps that start to look more like the final behavior. So what you can do is basically reinforce both a little above and a little below your current criterion until you see a variation you want starting to be offered. You can then start to reinforce that one more, or exclusively. (I think this is the technique that led my friend Julianna DeWillems to use a metaphor I love, "sweeping up a pile of dust," to describe shaping.)

Use treat placement strategically. I’ve written a whole other post on this, but briefly, think of your click as the next cue in a chain, rather than just as a conditioned reinforcer. Then think of the behavior that the learner will do to collect the treat as being cued by the click and reinforced by the treat. Then think about how you would chain two behaviors together if you wanted them to eventually blend: You would cue one, and when it was performed, cue the next, and then c/t after the second behavior, then repeat. Then when you saw the dog anticipating the second cue, and starting to jump ahead to the second behavior on finishing the first, you would fade the second cue out. What initially happens after the click, with repetition, will start to creep before it, and when you see that, you can delay your click until the learner has done the second behavior as well. (Note: this effect can also happen when you don’t want it to, another reason to constantly be thinking about what is happening between the click and the treat and adapting your treat placement with your end goal in mind during training, especially shaping.)

Change the antecedent conditions. Big, purposeful examples of this are introducing lures or arranging props like gates or platforms to encourage certain paths of movement, but even tiny changes can get you slightly different behavior from what you're getting now. If you have been sitting down the whole time you've been reinforcing the same approximation over and over, that context is probably cuing what you are getting now. So try kneeling or standing up, or moving to a different chair--these small changes may provoke just enough variability for you to start seeing, and reinforcing, something further along the path toward your goal, or at least to start being able to reinforce variability.

Surf the extinction bursts. This is an older technique and involves stopping reinforcement for the current criterion, or doing “twofers” (reinforcing only every other one) to spur variability. The variability here is a predictable by-product of extinction (nonreinforcement), which can also be a product of intermittent reinforcement (because it involves partial nonreinforcement). This can be done really well, but in my experience it’s harder on the dog if you have already narrowed variability too tightly and you don't change anything else about the context, or if you are waiting for too big a leap. Even if you get a variation you want to reinforce, you’re likely to get some emotional behavior (think “frustration”) along for the ride that you probably don’t want to bake in to your training by reinforcing it along with the bigger effort on the goal behavior. If you get stuck, think about changing treat placement or antecedent conditions first.

Step outside the "staircase" and teach component skills. This is an approach I first learned from Mary Hunter, and it has proven invaluable. Not much that a dog offers during shaping is going to be truly new; it's often going to be behavior that is currently in their repertoire but that (as far as they are concerned) there is no reason to try here. Or, it may be something that simply isn't in their repertoire at all. Either way, you can identify that missing piece and separately teach it, or just build reinforcement history for it using the reinforcers you'll be using in shaping. Then when you go back to shaping, that behavior will be more likely to be offered.

For example: let's say you're trying to shape a dog to go to a bed and lie down. You've got the dog walking onto the bed and sitting but he's just not doing anything remotely like lying down, and even though you deliver treats on the mat, he just stand ups to eat and then sits and looks at you again. In my experience, this is often a dog whom nobody has ever given a treat for lying down before. So when you quit reinforcing sit, which will cause the dog to offer other behaviors that have been reinforced in similar conditions, down isn't among the choices. But if you spend a couple days marking and reinforcing whenever the dog lies down on its own, anywhere, and then try shaping it on the bed again, the dog will be more likely to try it when you delay the reinforcer for sitting. If you do plan to use extinction, or withholding the reinforcer slightly, to move through a shaping plan, understanding the influence of prior reinforement history on what is likely to emerge is a must.

Thanks for Barking: Addenda

This 2021 blog post I wrote to elaborate on a 2017 quickie for clickertraining.com continues to be the most read thing on my website. (For the actual steps in the protocol, go to that post. Then come back and read this one.) And I get a lot of heartening messages about how it has helped people live more harmoniously with their dogs. But like most responsible content producers, I also worry a great deal about sending “recipes” out into the wild without enough context--be it through social media or presentations at conferences that are always shorter than I’d like. So here are some additional thoughts that have been percolating over the past couple years.

The number one thing people ask, with concern, about this protocol is: aren't you reinforcing barking? Initially yes. And, maybe forever. It should, however, be less barking, and less intense barking, and if you follow the protocol all the way through, you might get closer to none.

The number two thing people ask is "can I use it for x," and this is related to the number one thing. I think it's important to emphasize that this protocol is designed to address barking at stimuli that surprise both you and the dog, like stuff outside your window, in your apartment hallway, or outside a fence. If the things your dog barks at are not surprises, and you can see them coming and get ahead of barking consistently to reinforce something else, definitely do that instead. Leslie McDevitt's "Look at That" from Control Unleashed is great for this.

This protocol works best with initial preventive management, pretrained skills, and consistency. I don’t know about you, but I don’t want to do any of those things if I don’t have to. To potentially save yourself some work, if you have a brand-new dog and they are barking at stuff outside the window or fence, or your dog just started barking at sounds when you moved . . . then if your neighbors won't murder you, try doing nothing about it for a week and seeing what happens before starting any other type of protocol. I gave this advice to trainer friend Christie Catan when she moved to the country and you can read about it here.

This protocol initially uses food (usually) to reinforce barking. Less barking, and then if carried out to completion, the reinforcer is used to select a different part of the chain that results. I don’t worry about reinforcing the barking, though, because if you are a place where you are really motivated to do something about the barking, the barking is occurring regularly, which means something is reinforcing it already. The reinforcer is probably the sound or sight going away, or possibly the way you react to the barking. These explanations need to be ruled out before deciding that the barking is “self-reinforcing" as I often hear people do.

Because something in the environment is reinforcing (which means "strengthening") the barking, preventive management is key. Without it, the behavior will continue to be reinforced, and you may find yourself weakening or poisoning the cue you are trying to train (read on). Your management may not be perfect, but don’t let perfect be the enemy of better. If you are not prepared to train, do your best to minimize exposure, and by extension, opportunities to keep getting reinforced by the environment for barking. Use window film or white noise, close blinds, close the back door, put slats through or tarp over the chain link fence, etc. You may find this solves your problem without further training, but maybe you also like to be able to look out the bottom half of the windows and still want to move ahead with some training.

This may sound similar to advice you have heard about training a recall--don't put yourself in a position where you'll need it if you don't think it will work, right? Well, a clean recall cue is the main prerequisite for this protocol. And “thank you” is that clean recall cue. It doesn’t have to be “thank you,” but most people have ruined their first recall cue already (by not reinforcing, reinforcing poorly, punishing late arrivals, only using it when it will cost the dog something to come, etc), and most people haven’t ever said “thank you” to their dog yet, so it usually has no prior associations. It also tends to work as a mindset changer for the humans; it’s hard to say in a mad tone. But you could use anything that is a 100% reliable predictor of stuff your dog loves--even, gasp, your marker signal. A marker is just a cue that means come get your reinforcer.

If your recall isn’t working yet and your management fails, go to your dog and deliver the treats anyway. Drop them, many of them, in a straight line to the ground right past her eyeball. (Don't try to shove them in the dog's mouth; dogs are often less likely to take them that way, and you may get bitten.) When I started this protocol with my own dog Pigeon, my recall cue was not working in this situation, and I hadn’t thought of retrraining a new one “thank you” yet. When she tore down our gangway to hurl herself snarling at the gate, I hustled after her and stood right behind her, gave the cue, and then fed handfuls of Stella and Chewy’s if she so much as gave me a dirty look. Now, I would feed regardless of whether I got the dirty look. After a few sessions, she was paying much more attention to me when I followed her, both her barking and her visible physiological reactions (e.g., piloerection) got less intense, and I was able to start successfully recalling her out of the barking from farther away.

As with any other recall, I don’t recommend adding punishment—intentional or inadvertent—after the "thank you," as you risk poisoning the cue. If your dog won't take the treats even at the location of barking and you absolutely must bring them away, do it—but then adjust your plan, including preventive management, so you are not routinely hauling them off by the collar after you have said “thank you,” which could turn “thank you” into signal for avoidance. If you think "thank you" won't work, don't say it before you go get them.

I like to have the treats (and any other reinforcers) that follow “thank you” come in a predictable location so that (a) you don’t have to have treats on or near you all the time and (b) the dog will anticipate the location and start heading there—but if that’s not practical, the location can just be you.

Reinforce the earliest response your dog has to the stimulus. Don’t wait for barking if you can catch the head swivel or ear flick toward the sound.

Don’t reinforce the same type of (same look, same sound, pointing in the same direction) barking if it is offered when there is no stimulus to bark at. I once met a dog who barked at people outside the window, but also appeared to have learned to go around barking at windows even with nothing outside in order to get his person to get his toys out of hiding. That’s the sort of problem you may create if you reinforce the barking when there is no stimulus outside.

If the dog really is barking at surprise stimuli to get treats, congratulations! Many people seem to be happy to just have a nice way to stop barking quickly, before the thing goes away, but to me the coolest part is that now you can use the treats to shape less barking and faster turning. With Pigeon, I had trained up a predictable pattern of start toward the fence, bark a few times--here's the recall!--then turn around and start looking for the treats. So I began to delay the recall until I saw that turn. She started to bark even less and turn faster—skipping ahead to the behavior closest to the reinforcer.

The other things that may be happening are that the aversiveness of the stimulus may be reduced through the pairing with food (which you may not see if you are also routinely hauling your dog away by the collar after the cue); that when the stimulus starts to go away even without the continued barking, it may reveal barking to be unnecessary to remove the stimulus; and that if your response was reinforcing barking, the dog now gets attention or treats from you for less barking or for other behaviors like turning.