Abstract. Recently, there was defined a fragment of LTL (containing fairness properties among other interesting formulae) whose validity over a given infinite word depends only on an arbitrary suffix of the word. Building upon an existing translation from LTL to Büchi automata, we introduce a compositional approach where subformulae of this fragment are translated separately from the rest of an input formula and the produced automata are composed in a way that the subformulae are checked only in relevant accepting strongly connected components of the final automaton. Further, we suggest improvements over some procedures commonly applied to generalized Büchi automata, namely over generalized acceptance simplification and over degeneralization. Finally we show how existing simulation-based reductions can be implemented in a signature-based framework in a way that improves the determinism of the automaton.