Background reading:
Alderete, J. et. al. (1999) Reduplication with Fixed Segmentism.
Linguistic Inquiry 30, 327-364 (a copy of the article is available at Baris Kabak's mailbox at
Linguistics, 46 E. Delaware for photocopying):

Abstract:

In a recent paper, Alderete et al (1999) proposed that fixed segments in reduplication are derivable from general markedness considerations in the phonology. They argue that the fixed segment in the reduplicant is the (unmarked) epenthetic segment of the language. The classic example is the
fixed [i] in Yoruba deverbal reduplication. In this talk, I discuss two types of asymmetries relating to the Yoruba deverbal reduplication. In the first asymmetry the fixed segment in de-verbal reduplication behaves differently fixed segmentism in epenthesis, as in loan vocabulary. Fixed segmentism in epenthesis shows assimilation to a neighboring labial consonant, while fixed segmentism in deverbal reduplication does not. Since the assimilation is systematic, I conclude that the two types of
fixed segments are not the same. I propose that the fixed [i] in deverbal reduplication is really an input morpheme while the fixed [i]  in loan vocabulary is from epenthesis. The second asymmetry is one which occurs within the "fixed" segment in deverbal reduplication itself. The deverbal "fixed" segment shows variation based on whether the verb base has a high vowel or a nonhigh vowel. The "reduplicant" copies the initial CV of the base completely if the base has a high vowel, otherwise the prefix [i] is retained. The asymmetry in the deverbal reduplication is resolved by assuming that when the verb stem vowel is high the "reduplicant" vowel is faithful to both the underlying morpheme [i] of the prefix, as well as the high vowel of the verb base.