This paper introduces an abstract notion of fragments of monadic second-order logic. This concept is based on purely syntactic closure properties. We show that over finite words, every logical fragment defines a lattice of languages with certain closure properties. Among these closure properties are residuals and inverse C-morphisms. Here, depending on certain closure properties of the fragment, C is the family of arbitrary, non-erasing, length-preserving, length-multiplying, or length-reducing morphisms. In particular, definability in a certain fragment can often be characterized in terms of the syntactic morphism. This work extends a result of Straubing in which he investigated certain restrictions of first-order logic formulae. In contrast to Straubing's model-theoretic approach, our notion of a logical fragment is purely syntactic and it does not rely on Ehrenfeucht-Fraisse games.

As motivating examples, we present (1) a fragment which captures the stutter-invariant part of piecewise-testable languages and (2) an acyclic fragment of Sigma_2. As it turns out, the latter has the same expressive power as two-variable first-order logic FO^2.