When a heterogeneous material is strained, its evolution toward breaking is characterized by the nucleation and the coalescence of micro-cracks before the final break-up. Electromagnetic (EM) emission in a wide frequency spectrum ranging from very low frequencies (VLF) to very high frequencies (VHF) is produced by micro-cracks, which can be considered as the so-called precursors of general fracture. Herein we consider earthquakes (EQs) as large-scale fracture phenomena. We study the capability of nonlinear time series analysis to extract features from pre-seismic electromagnetic (EM) activity possibly indicating the nucleation of the impending EQ. In particular, we want to quantify and to visualize temporal changes of the complexity into consecutive time-windows of the time series. In this direction the original continuous time EM data is projected to a linguistic symbolic sequence and then we calculate the block entropies of the optimal partition. This analysis reveals a significant reduction of complexity of the underlying fracto-electromagnetic mechanism as the catastrophic events is approaching. We verify this result in terms of correlation dimension analysis. We point out that these findings are compatible with results from an independent linear method which uses a wavelet based approach for the estimation of fractal spectral characteristics. Field and laboratory experiments associate the epoch of low complexity in the tail of the precursory emission with the nucleation phase of the impending earthquake.