In this view, these are derived from bases in -er by replacing the -er affix with -ster (e.
3 Female personal nouns ending in -ster, -es and -in
As can be seen, it seems to be the case that -ster is denominal in (6a), whereas it is deverbal in (6b).
Under this analysis it is purely coincidental that both aar and -ster are separate affixes elsewhere in the system, and second, it fails to explain why the surfacing of the part -aar exactly mirrors the conditions (after schwa followed by a coronal sonorant) under which this allomorph of -er occurs.
Van Marle proposes that the suffix -ster replaces the affix -er.
We would like to argue for an analysis in which -aar and -ster are two separate affixes (contra De Vries & Van Santen 1981), and in which there is no affix replacement (contra Van Marle 1985 and Booij 2010).
However, the sequence -er-ster is ill-formed and a rule of haplology applies that deletes the left affix (-er) in the context of the following -ster.
The first question is why we find -es rather than -ster in these cases.
14) -ster does not attach to -aar precisely in those cases in which -aar idiosyncratically replaces -er.
Nevertheless, the generalization in (14) requires certain information to be available in order to attach -ster, namely that -aar is not a regular allomorph, but is present by exception.
Now observe that -ster attaches to -aar in (13a) and that the prosody is as in (16a) below.
This gives us the prosodic structure of (16a), predicting -ster.