In the fourth century BC our ancestors found the so called reflexive paradoxes. And all through the time these paradoxes have embarrassed our rational thinking persistently. The Liar paradox, among them, and Russell's paradox, a late comer, have had the crucial role when we think over the concept of truth and the foundation of modern mathematics. As a matter of fact, Russell's paradox was the first step to modify set theory and to construct several axiomatic set theories, despite his type theory could not get a honor of the theory of foundation. As to the Liar paradox, Tarski taught us convincingly that we were not able to represent 'true in a language L' in L. Here we will pay attention on these two papadoxes and find their common logical structures. And then by using them, we will try to construct a theory of truth, by which we mean a way of defining the predicate 'true (false)'. Our main interest is how to define 'true in a language L' in L. However, what we will find is that we have to restrict Convention (T) to avoid the inconsistency of our system like the limitation of the abstraction principle in the case of Russell's paradox. Moreover, the restricted Convention (T) forces us to use 'true' and 'false' strangely. This means that we are still very far from possessing the true theory of truth. However, we suppose our step here in this paper is certainly one step toward the theory.