ADAPTIVE TUTORS FOR DEVELOPING VISUAL REASONING

SKILLS IN PROGRAMMING COURSES

Rika Yoshii

Computer Science and Information Systems Department, California State University San Marcos, California, U.S.A.

Keywords: Visual reasoning, Programming, Adaptive, Mastery.

Abstract: This position paper argues that students who have difficulty programming often exhibit difficulty in

building correct semantic representation of program statement effects. We argue for the need to develop

and use tutoring software to help students develop visual reasoning skills in programming. Adaptive

tutoring systems will allow students to work privately at their own pace. Example tutoring system designs

incorporating visualization, adaptive learning, conversational tutoring and mastery learning are presented.

1 INTRODUCTION

As early as the late 80’s, difficulties students face in

solving algebra word problems have been discussed

widely. Studies have been published in cognitive

science journals indicating students’ problems in

building correct semantic representation of the

situations depicted in word problems (Hall,

1989a)(Hall, 1989b). The same type of difficulty

has been discussed in the area of programming

(Buchananand, 1995). Many students cannot

envision correct semantic representation of

 the problem statements, and/or

 programming constructs,

and they fail to pass beginning programming classes.

But what are we doing to help students develop

visual reasoning skills?

One possibility is to use example diagrams in

lectures and force students to use those diagrams to

explain their program parts. This approach works

well with students who have little or no ability to

form their own diagrams, but it will inhibit creativity

in other types of students. To help only those who

are having trouble forming semantic representation,

tutoring software to help students in this area will be

valuable, allowing those students to practice

privately at their own pace. Adaptive tutoring

software availability is especially important in

beginning programming classes with a large number

of students, making individualized instruction

difficult.

We have developed two tutoring systems in

second language learning, incorporating the

visualization approach. DaRT helps students

develop visual understanding of English articles

such as “a” versus “the” (Yoshii, 1998). DeCEN

helps students develop visual understanding of

countable versus non-countable nouns (Slott, 2005),

(Yoshii, 2010). Both systems have been used with

actual students and produced good results. We

believe that the same can and should be done for

programming classes.

In this paper, we will describe a sequence of

findings and projects at California State University,

San Marcos (CSUSM) leading to our current project

for creating a highly interactive adaptive tutor to

help students visualize effects of C++ statements.

2 VISUAL REASONING FOR

ENTRY LEVEL

MATHEMATICS

Entry Level Mathematics was the first area in which

we assessed the student difficulty in forming

semantic representation of situations. In our first

programming course for computer science majors,

students often expressed difficulty writing a small

program for solving simple mathematical problems.

After giving a quiz each semester to students

entering the first programming class, we realized

that all students who missed the following problem

drew incorrect diagrams to depict the described

364

Yoshii R. (2010).

ADAPTIVE TUTORS FOR DEVELOPING VISUAL REASONING SKILLS IN PROGRAMMING COURSES.

In Proceedings of the 2nd International Conference on Computer Supported Education, pages 364-367

DOI: 10.5220/0002860703640367

 SciTePress

situation:

“Trains A and B are at the same

station. Train A goes to the east at

X miles per hour. Train B goes to

the west at Y miles per hour. The

trains left the station at the same

time and travelled for Z hours.

What is the distance between

Train A and Train B?”

Students with incorrect answers drew diagrams

showing the trains moving in the same direction or

towards each other.

This experiment was repeated again this year in

both the first and second programming classes. In

both classes, 35% of students had incorrect diagrams

and thus equations. These students were not having

problems with the programming task but were

having problems coming up with the correct

equations because they could not visualize the

situation correctly. This result echoes what was

discussed in the papers (Hall, 1989a) and (Hall,

1989b). Unfortunately, appropriate remedial

instruction for the 35% of the students cannot be

provided within the context of the programming

classes.

3 VISUAL REPRESENTATION OF

LINKED LISTS

The second area in which we discovered the same

difficulty was with linked lists in the second

programming course. After two lectures showing

diagrams for adding and deleting nodes from linked

lists, the students were given an assignment to draw

the effects of C++ statements for manipulating

linked lists before they were allowed to write a

linked list class. An example problem is:

“Why can't we do the following to

add a node to the rear? Draw the

effect. Rear =

new Node;

Rear->Next = Rear;”

Every semester, more than 50% of the class had

great difficulty drawing correct pictures, and the

same group of students had difficulty coding the

linked list class. With a large number of students, it

was not possible to sit with each student to

individually analyze their diagrams.

4 CPP TRACER TUTOR

The students in the first programming course who

had problems with Entry level Mathematics

problems were also resistant to the explanations and

exercises that were typically successful in leading

others to an visual understanding of machine

operation, and the ability to understand their own

programs. Instead of relating their programs to the

sequences of machine actions they cause, the

students seemed to memorize a few stock

“boilerplate” program fragments for the problems

they had seen. They were unable to analyze why a

program works or does not, seeing no connection to

the resulting actions by the machine.

With these students, simply using a debugger did

not help since they needed repeated explanations

with examples. But they were reluctant to ask for

help and preferred to practice without being noticed

by other students. Therefore, we developed a

tutoring system, Cpp Tracer Tutor (Yoshii, 2003).

The goal of the Tutor is to help students develop and

practice their reasoning skills using visual

representations of executing programs. The

following are its main features. The latter three of

these features were adopted from the Irvine-Geneva

strategy (Bork, 1992)(Bork 2001):

 Visual representations of programs: the Tutor

provides a live media visualization of the

actions taken by the machine in executing a

running program.

 Conversational tutoring: the Tutor provides

frequent interactions with free-form student

answers.

 Adaptive learning: the Tutor uses student

answers to provide appropriate hints and to

select the next exercise.

 Mastery learning: the Tutor does not let the

student move onto a different program

example until the student shows no difficulty

understanding the current one.

The Tutor, written as a Java applet is available

via the Internet, allowing students to use it privately

at their own pace. Currently, the Tutor supports five

different types of example C++ programs that are

often seen in the first programming course: 1) a for-

loop with a constant as its upper limit, 2) a for-loop

with the upper limit pre-supplied by the user, 3) a

while-loop that continuously take the user’s input

until a sentinel value is reached, 4) a while-loop that

terminates when a sentinel value is read or a certain

number of iterations has been done, and 5) a while-

loop that terminates when a sentinel value is read or

ADAPTIVE TUTORS FOR DEVELOPING VISUAL REASONING SKILLS IN PROGRAMMING COURSES

365

the sum of input values reaches a certain number.

The Tutor’s design permits addition of more

example programs without making major changes to

the system.

Assessment in five different laboratory sections

of the first programming course indicate that the

students had an average improvement of 15% in

points between the pre and post-tests on reading and

writing C++ loops after using the system for only

two hours.

5 THE NEXT STEP : VISUAL

TUTOR FOR C++

STATEMENTS

A problem with Cpp Tracer Tutor is that it assumes

the students already understand the effects of

individual statements found in loops. With students

who have problems with algebra word problems,

training in visual reasoning must start earlier.

Another problem is that the students are simply

observing the visualization of actions. The students

need to be trained to visualize the actions.

Therefore, we are currently designing a system

for helping students visualize the effects of

individual C++ statements. As with the Cpp Tracer

Tutor, the tutorial style will be based on the Irvine-

Geneva theory emphasizing:

 Conversational tutoring

 Adaptive learning

 Mastery learning

The system will start off by showing correct

visual representation of C++ statements using many

examples. In the tutorial section, the student will be

given a C++ statement. By clicking and dragging

icons the student must create the visual

understanding of that statement. The system will

categorize wrong answers and record them so that

the student can be given an appropriate sequence of

exercises. The system will give appropriate

conversational hints and take the student back to the

same question. When the answer is finally correct,

the system will give the same type of question again

to verify student understanding. For example:

Declaration int A;

The student has to select a memory

box and add labels for the type and

name.

Input cin >> A;

The student has to select the correct

box and drag the user input into the

box.

Output cout << A;

The student has to select the correct

box and copy the value to the output

stream.

Assignment A = A + 1;

The student has to select the correct

box and change the value.

Loops Each part of a loop will be

highlighted in turn in the execution

order, and the student is asked to show its

effect.

In advanced tracks, arrays, stacks, queues and

linked lists will be included.

One of the goals of the project is to create it

based on a language-independent framework so that

the tutor can be used for learning any programming

language. Once the system is complete, we plan to

1) assess its usability with actual students and

teachers, and 2) assess its effectiveness with our

students and students at local community colleges.

6 SUMMARY

We have presented an acute need for developing

tutoring systems to help students develop visual

reasoning skills in order to succeed in programming

classes. Out findings in beginning programming

classes at CSUSM show that many students cannot

envision correct semantic representation of the word

problem statements, and/or programming constructs.

We have shown two example tutor designs

incorporating visual reasoning, adaptive learning,

conversational tutoring and mastery learning. With

these types of tutoring systems, students in large

classes can work at their own pace at home to

develop their visual reasoning skills.

We also believe that such systems should be

developed for Entry Level Mathematics classes to

train students to visualize problem situations before

they come to programming classes.

REFERENCES

Bork, A. et al., 1992. The Irvine-Geneva Development

System. In R. Aiken (Ed.), Education and Society

Vol. II, North Holland: Elsevier Science Publisher.

Bork, A, Gunnarsdottir, S., 2001. Tutorial Distance

CSEDU 2010 - 2nd International Conference on Computer Supported Education

366

Learning: Rebuilding our Educational System, Kluwer

Academic Publishers.

Buchanan, R., Farrand, P., 1995. Can simulations help

students understand programming concepts: a case

study.http://www.ascilite.org.au/conferences/melbourn

e95/smtu/papers/buchanan.pdf.

Hall, R., Kibler, D., Wenger, E., & Truxaw, C., 1989a.

Exploring the episodic structure of algebra story

problem solving. Cognition and Instruction, 6(3), 223-

283.

Hall R., 1989b. Qualitative diagrams: supporting the

construction of algebraic representations in applied

problem solving. In D. Bierman, J. Breuker & J.

Sandburg (Eds.), Artificial intelligence and education,

Proceedings of the 4th International Conference on AI

and Education (116–122). Amsterdam, Netherlands:

IOS.

Slott, K., Yoshii, R., 2005. “DeCEN and Tutor Writer:

Making an Interactive CALL Tutor Available as a

Java Applet”, Proceedings of the ED-MEDIA 2005

Conference, June 2005.

Yoshii, R., 1998. "DaRT: A CALL System to Help

Students Practice and Develop Reasoning in Choosing

English Articles." CALICO Journal, vol. 16, no 2., pp.

121-155, Fall 1998.

Yoshii, R., Milne, A., 2003. “The C++ Tracing Tutor:

Visualizing Computer Program Behavior for

Beginning Programming Courses”, Proceedings of the

ED-MEDIA 2003 Conference, June 2003.

Yoshii, R., Pasrija, N., 2010. “TWIGy: The Tutor Writer

Input file Generator to Help Create an Interactive

Individualized Tutor that Runs Over the Internet”,

Proceedings of the SITE 2010 Conference., March

2010.

ADAPTIVE TUTORS FOR DEVELOPING VISUAL REASONING SKILLS IN PROGRAMMING COURSES

367