There’s also an issue here concerning how people project their intuitions about phonotactics into the magic region of zero-frequency forms. Note that final [mb] occupies the empty field of a so-called “L-shaped region”, thus:

nt nʧ ŋk mp

nd nʤ — –

I believe that such L-shaped regions are ripe territory for generalization, filling in the gap in the L. The reason, perhaps, is that the simplest characterization that avoids logical disjunction (”final N + stop ok”) projects the missing forms. To characterize only the attested forms of the L, one has to use disjunction (”in final N + stop, the stop must be either voiceless or coronal”).

I believe that the L-shaped region phenomenon was first pointed out by Edward Sapir, who (somewhere) noticed the L-shaped region for word-initial fricatives in English:

f θ s ʃ

v ð z –

The existence of this region tends to make English speakers comfortable with borrowings like soup du jour, Zhivago, Zhirinovsky.

–Bruce Hayes

The reason I got interested (and asked my friend to repeat it) is because I thought that I finally had something other than the old bomb [ba:m] ~ bombard example — and something novel and fun-in-a-geeky-way besides — to use as evidence for the relevant deletion process. But given that I don’t have recorded evidence, and insistence from my “consultants” that they think it’s [za:mb], I’m discouraged …