Learning theory of performance errors
The Learning Theory of Performance Errors emphasizes the role that errors play in learning and how knowledge can be modeled to explain how errors are made, detected, and corrected by the learner. Through this process, learners progress from being novices, making numerous errors at new, unfamiliar tasks, to being experts who rarely make errors. The process of gaining and strengthening new skills and knowledge can be traced as the learner gets better and better at the task (Ohlsson 1996; 2016).
This article outlines the types of scenarios that this theory applies to, the idea of production rules that describe how learners are able to apply knowledge, the Perceive-Decide-Act Cycle that learners use to select production rules to apply, the nature of constraint-based knowledge that is encoded as production rules, and how errors are detected and corrected after a production rule has been applied.
The Learning Theory of Performance errors applies to situations where learners are trying a new, complex task, which is defined as a task which has steps that are carried out in sequence and each step has more than one choice that the learner can make. An example of a complex task would be driving a car from one location to another. It is limited to scenarios where the learner is practicing unsupervised (Ohlsson 1996:242).
In order for a learner to have an idea of the best action to take in order to complete a task, they must have some practical knowledge (Ohlsson 1996:243) that they can apply. This knowledge is comprised of three parts:
- the goal (G), or the end state the learner wishes to achieve;
- the situation (S), or the set of conditions that describe the perceived scenario; and
- the action (A) to be taken in order to make progress toward the goal.
This relationship, called a production rule, can be understood in plain language as: When I have G goal in situation S, I should take action A. In other words, the combination of a goal and a given situation imply some action that should be taken in order to get closer to the overall goal. For example, a production rule for driving an automatic vehicle might be:
- Goal: drive the vehicle forward
- Situation: in an automatic vehicle which is started but parked
- Action: hold down the brake pedal and shift the car into drive
When a learner has a production rule that matches their current goal and situation, it is likely to be applied to the current task, because they have learned it from prior experience (Ohlsson 1996:243). When there are multiple production rules that fit the current situation, the rule that was most successful in the past is applied (Ohlsson 1996:244). As a learner gains more experience in a certain subject, they will have more production rules in their memory, and the rules will also be more specialized. On top of that, more specific tasks require more production rules in order to properly execute them. But when a learner is presented with a new situation that has no production rules that apply to it, they will apply general methods which are broad production rules that could apply to a number of situations (Ohlsson 1996:247). For example, a learner may know how to drive an automatic vehicle, but not know how to drive a manual vehicle. They do not have production rules for getting a manual vehicle into gear. Not knowing what to do in this specific situation, they may try to apply their automatic vehicle knowledge as a general method and try to shift the car without knowing how to use the clutch, resulting in grinding the gears. However, knowing how to drive one automatic vehicle will form a good general method which can be applied to many different kinds of automatic vehicles.
When learners carry out a task, they first perceive the current situation, decide on which action to take, and then act on that decision. The perceive portion of the cycle informs the learner's understanding of the situation, the decide portion determines which production rule they will use based on the situation, and the act portion is the application of the action component of the chosen production rule. When they carry out their action, the situation is affected, bringing them back to the top of the Perceive-Decide-Act cycle (Ohlsson 1996:243). It is by the repeated use of this cycle that the learner performs their task sequentially until it is completed.
Declarative and Practical Knowledge
In many other theories of learning, declarative knowledge is defined as what a person knows. In this theory, declarative knowledge is more specific: it describes what should be instead of what is. For example, a person may know that “when I subtract x from y, the number I get should be smaller than y". This type of declarative knowledge can be used to make a judgement, or an evaluation based on something that a person knows should be the outcome. Therefore, it is possible for a learner to have enough declarative knowledge to know that they have done something incorrectly, but not enough practical knowledge to not have made the error in the first place or to perform better (Ohlsson 1996:246).
By this definition, some declarative knowledge is encoded in memory as constraints, or rules that can be violated. In the subtraction example, the constraint would be that the result of a subtraction should be smaller than the original number that was subtracted from. This is the way that a learner is able to use their declarative knowledge to figure out when they have made an error -- when constraint violations occur, creating conflict with their declarative knowledge (Ohlsson 1996:245).
The objective method of error detection says that when a learner applies the action from a production rule and sees that the outcome was not the best decision they could have made in the situation, they have committed an error. However, this objective method means that the learner must know all the consequences of their possible decisions; otherwise, they could not know whether it was not the best choice (Ohlsson 1996:245). Since this theory is about situations where the learner is unsupervised and has no expert to compare against, the objective method is not a good fit for this kind of learning. Instead, error detection in this theory happens in a subjective way: based on the knowledge that the learner has, they have some idea of an expected result. When the outcome of their action does not match with what they expect based on what they know, something along their path to the current state must have been wrong. It is possible for the learner's expectation to be incorrect and their result to be correct, but for the learner to believe they have made an error because of their result does not match their expectations. It is also possible for the learner to believe they have made no errors when they have chosen an action based on incorrect perceptions (Ohlsson 1996:245).
When general methods are applied in new situations because the learner lacks specific production rules about the subject or task to apply, constraint violations will happen because those general methods, which have very few restrictions, will not be broadly applicable to all situations. When such errors occur and are detected, error correction happens by specializing the rules to the relevant specific situation.
Given a faulty general rule like: to move a car forward when it is started and parked, hold down the brake pedal and shift it into drive, which does not apply to manual transmission vehicles, specialization happens by recognizing that the action component was an incorrect one to take in the situation because of the grinding gears and lurching car. The learner can then identify some features of the current situation and specialize the rule to something like: to move an automatic car forward when it is started and parked, hold down the brake pedal and shift it into drive. Given this new specialized rule, no constraint violation exists as it did with the previous generalized rule, because it no longer applies to trying to drive a manual vehicle. It is through specializing rules over and over again through error detection and correction and strengthening then through successfully applying existing production rules that a learner passes from novice to expert and gains the declarative and practical knowledge required to become skilled at a task.
- Ohlsson, S. (1996). Learning from Performance Errors. Psychological Review, 103(2), 241-262.
- Ohlsson, S. (2016). Constraint-Based Modeling: From Cognitive Theory to Computer Tutoring – and Back Again. International Journal of Artificial Intelligence in Education, 26(1), 457–473.