In this work, possibility of decreasing active degrees of freedom in the motion learning was investigated. For simplicity, it was focussed on producing the motion of a robot arm with four degrees of freedom only from a three-dimensional time sequence of Cartesian positions generated by movements of a mouse on pad while writing. In this case, the attitude angles(orientation) of the arm's end-effector cannot be directly determined since the mouse trajectory has no information about it. However, this problem is solved by using an artificial neural network which maps a segment of the time sequence onto the attitude angles. Additionally, to confirm the effectiveness of the proposed scheme, successful hand-writing is actually performed by the robot manipulator.