In this paper we presented two algorithms for the computation of multivalued dependencies
from relations. The hypothesis space used in the algorithms is dened
in accordance with the theoretical view of the discovery of interesting sentences
proposed in [9]. The top-down algorithm computes the multivalued dependencies
from the input relation by using the generalisation/specialisation hierarchy
of multivalued dependencies to investigate the hypotheses from the most general
towards more specic hypotheses. Our bottom-up algorithm provides an alternative
by rst calculating the invalid dependencies, and later calculating valid
dependencies from the invalid dependencies, rather than from the dataset. The
implementation of the algorithms was analysed and the empirical results of the
algorithms were presented.