We have monitored the space-time transformation of 150-fs pulse, undergoing self-focusing and filamentation in water, by means of the nonlinear gating technique. We have observed that pulse splitting and subsequent recombination apply to axial temporal intensity only, whereas space-integrated pulse profile preserves its original shape. PACS numbers: 190.5530, 190.5940, 260.5950, 320.7100 Spatial and temporal transformations of wave packets which undergo self-focusing in transparent media with Kerr nonlinearity attract a great deal of interest from the point of view both of fundamental and applied science. Self-focusing of ultrashort light pulses with power well exceeding that for continuous-wave (CW) beam collapse gives rise to a variety of spatial, temporal and interrelated (spatio-temporal, ST) effects and has been a topic of intense theoretical and experimental research [1,2,3,4,5,6,7,8]. Discovery of a long range propagation (filamentation) of ultrashort laser pulses in air boosted an interest in ST transformation through filamentation dynamics in gasses [9,10], solids [11] and liquids [12, 13].Numerical models of different complexity have been elaborated to study temporal, and, more generally, ST dynamics. Propagation of intense light pulses in media with rather distinct optical properties (amount of normal GVD, nonlinearity, etc; namely gasses, solids and liquids) under various initial conditions (pulse duration, beam width, power, wavelength) has been examined. In spite of different regimes, temporal pulse splitting emerged as a common feature resulting from the interplay between self-focusing, self-phase-modulation and normal dispersion. The open question is then what happens after the pulse splitting occurs. A scenario leading to subsequent multiple splitting has been predicted, but not observed directly (see the discussion in Ref [14] and references therein). In Ref.[7] pulse recombination after splitting has been observed and attributed to possible underlined multiple splitting.More recently, two interpretations have been proposed for understanding filament formation in condensed matter. Some of the present Authors, after having experimentally demonstrated that filaments cannot be treated as a self guided beams, have outlined the key role plaid by non-linear losses in ruling a spontaneous transformation from gaussian to conical (Bessel-like) wave, in the frame of a CW model [15]. In this context, the conical wave provides the beam with a large (and not absorbed) power reservoir that keeps "refuelling" the hot central spot and so ensures stationarity, even in presence of nonlinear losses. Independently from this work, a second approach has been proposed [16] sharing with the first the genuine idea of filaments as non-soliton, but conical waves. The important difference is that here the Authors have outlined the key role of chromatic dispersion, fully neglected in previous case, thus interpreting the filament on the basis the dynamics of non-linear X waves [17,18,19].The major problem concerning the ...