This squib shows that *Map faithfulness constraints, which have been
used to model saltations (Hayes & White 2015), make problematic predictions.
They predict that input-output changes may happen without any markedness
constraints motivating them: this derives unattested phonological patterns such as intervocalic devoicing. By contrast, an approach based on Segment Faithfulness
(Burzio 2000) is more restrictive and therefore preferrable: it predicts that
input-output changes must always be motivated by markedness constraints.