Chapter 3 A General Model Supporting Tutorial Dialogues
3.3 Model for adaptive tutorial dialogue support
3.3.2 Tutorial dialogues
Error remediation is facilitated through tutorial dialogues. A dialogue is designed for each error type (i.e. each leaf node in the hierarchy). Prompting students about their mistakes is taking advantage of a good learning opportunity as prior research suggests that successful learning occurs when students make an error (Ohlsson, 1996; VanLehn, Siler, Murray, & Baggett, 1998).
There are two types of dialogues: (i) Single-level dialogues and (ii) Multi-level dialogues. Single-level dialogues handle errors associated with simple syntax errors for which a detailed feedback message is sufficient to explain the error. Hence these dialogues are limited to a single feedback message. An example of a single level dialogue is You have connected an entity A to entity B directly. Entities cannot be directly connected to each other. These are similar to single- turn dialogue moves supported by AutoTutor (Person et al., 2001). Multi-level dialogues handle more complex errors for which a series of prompts is necessary to guide students to self-explain both domain concepts and solution steps. These dialogues consist of four levels:
1. Conceptual prompt: informs the student about the corresponding domain concept that he/she has difficulty with and asks the student about the particular aspect of the domain concept,
2. Reflective prompt: prompts the student to understand why his/her problem-solving action is incorrect,
3. Corrective action prompt: prompts the student to specify the correct problem-solving action and
4. Conceptual reinforcement prompt: helps the student to review the domain concept learnt via a question.
Reflective and the corrective action prompts discuss the error within the context of the current problem. In contrast, the other two prompts (conceptual and conceptual reinforcement prompts) focus on the particular aspect of the domain concept associated with the error. As mentioned in Section 2.4, there are two exceptions to the structure of multi-level dialogues. The former refers to those dialogues that discuss errors related to cardinality and participation; they consist of only three levels. The second exception is the different focus of the reflective prompt when focusing on missing constructs. Both these exceptions will be discussed in detail at the end of this section.
97
We now discuss each prompt using the dialogues presented in Figures 3.5 and 3.6. The hypothetical dialogue in Figure 3.5 is initiated when a regular entity is used to represent a weak entity, an error in conceptual database design. The complete dialogue is given in Appendix B.1.
Figure 3.6 discusses an error in data normalization. It is initiated when the specified candidate key is incorrect because it has extra attributes and is not a minimal key. The complete dialogue is given in Appendix B.2.
Each prompt is associated with three possible options. One of the options is correct and one is incorrect. Options such as “I need more help”, “I’m not sure” provide the opportunity for students to specify that they do not know the correct answer and ask for assistance.
The first level, the conceptual prompt, specifies the domain concept relevant to the most frequently made error type and then asks the student for a particular aspect of the corresponding domain concept. For example, “You seem to be having some difficulty with regular entities. Let’s look at regular entities in detail. Can you tell me the general rule to decide whether something is a regular entity?” (EER-Tutor1 in Figure 3.5) is an example of a conceptual prompt. Here the
EER-Tutor1: You seem to be having some difficulty with regular entities. Can you tell me the general rule to decide whether something is a regular entity?
Student1: I'm not sure
EER-Tutor2: Let’s try a simple thing. What is used to uniquely identify instances of a regular entity?
Student2: partial key attribute
EER-Tutor3: That’s not quite right. A key attribute is used to uniquely identify instances of a regular entity. EER-Tutor4: Now tell me why modelling CHAPTER as a regular entity is incorrect?
Student3: I’m not sure
EER-Tutor5: Let’s see …Check carefully whether you can find a key attribute for CHAPTER. {After Continue button is pressed}
Now please select the key attribute for CHAPTER.
Student4: No
EER-Tutor6: Good job! So now can you tell me best way to model CHAPTER? Student5: weak entity
EER-Tutor7: Well done! Now use the correct word to fill in the blanks so that you’ll remember what you learnt.
A ……….. uniquely identify instances of a ………. Student8: key attribute, regular entity
EER-Tutor8: Great job! Please make the changes and submit the solution when you’re done.
98
concept that the student has difficulty with is regular entities. This concept is covered in several constraints. This dialogue is associated with one particular aspect of the domain concept: i.e. the general rule to decide whether something is a regular entity. Explicitly indicating the domain concept that students have difficulty with provides an opportunity to reflect on them. Even though the first level prompt focuses on the conceptual prompt, the offered options are not strict definitions from the textbook. Instead the student needs to reason about how corresponding domain concept is related to current problem-solving action. This prompt does not reduce the task to recognition, but requires the student to re-examine his/her domain knowledge in order to answer the question. If the student fails to select the correct answer, he/she is given a simpler question. In this dialogues the simpler question used is “What is used to uniquely identify instances of a regular entity?” (EER-Tutor2 in Figure 3.5). If the student still has difficulty answering the simpler question, the system provides the correct answer. A correct answer by a student at any stage moves the dialogue to the next stage and the next prompt is displayed.
The reflective prompt, the second level prompt provides an opportunity to reflect on why the current problem-solving action is incorrect. This prompt requires the use of both procedural and conceptual knowledge. In contrast to the conceptual prompt, the reflective prompt discusses the selected error within the current problem-solving context. For instance, “Tell me why modelling
CHAPTER as a regular entity is incorrect?” (EER-Tutor4 in Figure 3.5) is an example of a reflective prompt in this dialogue. If a student fails to answer the reflective prompt correctly (as shown by Student3 in Figure 3.5), he/she will be given assistance to understand why the current problems-solving action is incorrect (EER-Tutor5 in Figure 3.5). Here the student is expected to understand that CHAPTER should not be modelled as a regular entity because CHAPTER does not have a primary key attribute.
The third level prompt, the corrective action prompt assists the student to understand how to correct the mistake. The prompt “Can you tell me the best way to model CHAPTER?” (EER-
Tutor6 in Figure 3.5)” is an example of a corrective action prompt in this dialogue.
The conceptual reinforcement prompt, the final prompt is another opportunity to learn the corresponding domain knowledge. The prompt presented as EER-Tutor7 in Figure 3.5 helps the student to understand that a key attribute uniquely identifies instances of a regular entity.
99
We developed 31 single-level and 112 multi-level dialogues and these dialogues cover all the errors in conceptual database design. Thus we showed that the proposed structure is expressive enough to discuss all the errors in conceptual database design.
We used the same structure to develop dialogues for data normalization. Based on the hierarchy we developed 33 single-level and 45 multi-level dialogues. A hypothetical dialogue in Figure 3.6 is presented as an example.
This dialogue is initiated when the specified candidate key is incorrect because it is not a minimal key i.e. the specified candidate key has extra attributes. If the specified candidate key is AD (one of the correct candidate keys is A) for step 1 in the example presented in Data Normalization (Section 3.1.3), then it is a non-minimal key (i.e. super key). This is because of the extra attribute D in the specified candidate key. This error violates constraint number 11 which is
NORMIT1: You seem to be having some difficulty with candidate keys. Let's look at candidate keys in detail. What is a candidate key?
Student1: I'm not sure
NORMIT2: Let’s try a simple thing. What do we mean by a candidate key being a minimal set of attributes?
Student2: A candidate key: is the smallest set of prime attributes that determines all the attributes
NORMIT3: That’s not quite right. A candidate key: is the smallest set of attributes that determines all the attributes.
NORMIT4: Now can you tell me what is wrong with AD as a candidate key? Student3: I’m not sure
NORMIT5: Let’s see …Check carefully whether all the attributes of AD are needed to determine all the other attributes.
{After Continue button is pressed}
Do we need all the attributes of AD to determine all the other attributes?
Student4: No
NORMIT6: Good job! We need only some of the attributes of AD to determine all the other attributes. So now tell me the extra attribute(s) for AD?
Student5: D
NORMIT7: Well done! Now try to answer this question so that you'll remember what you learnt. A ... is a ...key with extra ...
Student8: superkey, candidate key, attributes
NORMIT8: Great job! Please make the changes and submit the solution when you’re done.
100
categorised under the node Specifying a non-minimal key as a candidate key, which is a child of the node Candidate keys (Figure 3.5).
The conceptual prompt (NORMIT1 in Figure 3.6) focuses on candidate keys, the corresponding domain concept for the current error. If a student fails to answer this prompt correctly, he/she is given simpler question (NORMIT2 in Figure 3.6). Failure to answer the simpler question results in presenting the correct answer by the tutor. Providing the correct explanation at any stage, moves the dialogue to the next stage and the next prompt is displayed.
The reflective prompt “Now can you tell me what is wrong with AD as a candidate key?” (NORMIT4 in Figure 3.6) assists the student to reflect on why the current-problem solving action is incorrect. Failure to answer this prompt correctly, results in a prompt that guides the student to reflect on the current-problem solving action (NORMIT5 in Figure 3.6). In this dialogue, the tutor helps the student to understand that all the attributes in the specified candidate key are not needed to determine all the other attributes in the given relations. The corrective action prompt provides an opportunity to understand how to correct the mistake (NORMIT6 in Figure 3.6). Here the tutor helps student to clarify the extra attributes in the specified candidate key. Finally the conceptual reinforcement prompt provides another opportunity to learn the corresponding domain concept (NORMIT7 in Figure 3.6).
Dialogues with three levels: As mentioned earlier, dialogues that discuss cardinality and participation consist of only three levels. This is because there are the only two possible values for cardinality (i.e. 1 or N). These are the only ones allowed by the interface (Figure 2.2). As soon as the student is made aware that the cardinality specified is incorrect, the correct answer becomes obvious. Asking students to specify the cardinality using another prompt might potentially demotivate them if they want to resume problem solving straightway. When they correctly explain why the specified cardinality is incorrect, they are given the final prompt of the dialogue. For example, the response will be “Great Job! I guess you know how to correct the mistake now. Before starting to make changes, try to answer this question. What is the correct question to ask when deciding the participation between entities E1 and E2 in a binary relationship?”. Similarly participation also has two possible values: total and partial. As soon as the ITS tells the student that the specified participation is incorrect, then the correct answer becomes apparent. Thus the
101
dialogues that discuss errors related to participation also have only three levels. There are four dialogues that focus on cardinality and four dialogues that discuss errors related to participation.
Different level-2 prompt when discussing missing constructs: The reflective prompt focuses on why the current problem-solving action is incorrect. However, this is not applicable when the error is about missing solution components. In such situations, the student is asked to specify the type of construct that is missing.
Our paper-based investigations focus on developing tutorial dialogues for two different types of instructional tasks: conceptual database design and data normalization. Thus we showed that the proposed structure is expressive enough to discuss all the errors in both domains.