We have obtained high-resolution spectroscopy of ten red giants in the Carina dwarf spheroidal (dSph) galaxy with the ultraviolet and visual echelle spectrograph at the European Southern Observatory Very Large Telescope in order to study the detailed chemical evolution of this Galactic satellite. Here we present the abundances of O, Na, Mg, Si, Ca, Ti, and Fe. By comparison of the derived iron abundances [Fe/H] with metallicities based on the well-established calcium triplet (CaT) calibration, [Fe/H]CaT, we show that the empirical CaT technique yields good agreement with the high-resolution data for [Fe/H] gsim –2 dex, but tends to deviate from these data at lower metallicities. With [Fe/H] ~ –1.7 dex the mean iron abundance of our targets is fully consistent with the peak metallicity of Carina as derived from medium-resolution spectroscopy and previous photometric studies, all calibrated onto iron via Galactic globular cluster scales. We identify two metal-poor stars with iron abundances of –2.72 and –2.50 dex. These stars are found to have enhanced [α/Fe] ratios similar to the elemental ratios of stars in the Milky Way halo. In this context, it is conceivable that the moderately metal-poor halo stars may originate from an early dSph accretion event. The bulk of the Carina red giants exhibit a depletion in the [α/Fe] abundance ratios with respect to the Galactic halo at a given metallicity. One of our targets with a moderately low [Fe/H] of –1.5 dex is considerably depleted in almost all of the α-elements by ~0.5 dex compared to the solar values. Such low values of the ratio of α-elements to iron can be produced by stochastical fluctuations in terms of an incomplete mixing of single type Ia and type II supernova (SN) events into the interstellar medium. Moreover, the system's slow star-formation (SF) rate grants sufficient time for SNe I to occur. Our derived chemical element ratios are consistent with the episodic and extended in Carina previously derived from analyses of its color-magnitude diagram. We find a considerable star-to-star scatter in the abundance ratios. This suggests that Carina's varies with position within the galaxy, with incomplete mixing. In addition, or alternatively, the rate is so low that the high-mass stellar initial mass function is sparsely populated, as expected on statistical grounds in low-mass star clusters, leading to true scatter in the resultant mass-integrated yields. Both ideas are consistent with slow stochastic in dissolving associations or star clusters, so that one may not speak prima facie of a single SF history at a detailed level.