Nonlinear rheological behavior under uniaxial elongation was examined for unentangled melts of polystyrene (PS27; molecular weight M = 27k) and poly(p-tert-butylstyrene) (PtBS53; M = 53k) having nearly the same number of Kuhn segments per chain (n(K) = 30 and 35 for PS27 and PtBS53, respectively). For both materials, the steady state elongational viscosity, eta(E) exhibited strain-rate-hardening and then strain-rate-softening on an increase of the Weissenberg number Wi >= 0.3 (Wi = <(epsilon)over dot>tau(eq)(1), with tau(eq)(1) and <(epsilon)over dot> being the longest relaxation time in the linear viscoelastic regime and the Hencky strain rate, respectively). For the unentangled melts, the hardening and softening were free from any entanglement nonlinearity, so that the hardening was unequivocally related to the finite extensible nonlinear elasticity (FENE) of the chain, and the softening, to the suppression of the FENE effect due to reduction of the segment friction zeta occurring for the highly stretched and oriented chain. Thus, the zeta reduction, speculatively discussed for entangled melts so far, was experimentally confirmed, for the first time, for unentangled melts. Quantitatively, the hardening at intermediate Wi was stronger and the softening at higher Wi was weaker for PtBS53 than for PS27 despite the similarity of their n(K) values, which suggested that the magnitude of zeta reduction depends on the chemical structure of the chains. For estimation of this magnitude, the FENE-PM model (a FENE bead-spring model with a pre-averaged FENE spring strength) was modified for the zeta reduction in an empirical way with an assumption that 4 at a given time is fully determined by the chain stretch and orientation and, thus, by the tensile stress sigma(E) at that time. This modified model was able to mimic the steady state eta(E) data excellently, and the zeta reduction utilized in the modification was weaker for PtBS53 than for PS27 to confirm the dependence of the magnitude of zeta reduction on the chemical structure of the chain. Nevertheless, the same modified model failed to mimic the transient stress growth and relaxation data on start-up and cessation of fast flow (at Wi >= 4) despite its success for the steady state eta(E) data in the entire range of Wi. Specifically, changes of zeta in the unentangled melts with time during the relaxation for large Wi were delayed compared to the model calculation. This result suggests, as one possibility, that zeta in those melts is determined not only by the chain stretch and orientation (i.e., by sigma(E)) at respective times but also by the transient changes of the stretch and orientation (by <(sigma)over dot>(E)), with those changes vanishing in the steady state thereby allowing the model to mimic the eta(E) data. The origin of this change of zeta with the transient changes of the stretch and orientation is discussed in relation to the local motion of the chain necessary for adjusting its friction to the changes of the stretch/orientation environment.