Sequence learning has a variety of different approaches. We can distinguish two fundamental approaches: general-regularity learning and specific sequence learning. Although some studies suggest that it is possible for a single subsystem underlying both general-regularity and specific-sequence learning, it is proved that relatively independent subsystems may execute the two types of learning more efficiently than a single subsystem. In this paper, we propose to implement specific sequence learning in a neural network with no memory and with an implicit representation of time.