Core-electron excitation spectra are used widely for structural and chemica
l analysis of materials, bur interpretation of the near-edge structure rema
ins unsettled, especially for semiconductors. For the important Si L-2.3 ed
ge, there are two mutually inconsistent interpretations, in terms of effect
ive-mass excitons and in terms of Bloch conduction-hand final states. We re
port ab initio calculations and show that neither interpretation is valid a
nd that the near-edge structure is in fact dominated by short-range electro
n-hole interactions even though the only bound excitons are effective-mass-
like.