The main reason is that the type system of Scala allows sub-typing, which the Hindley-Milner type inference algorithm does not support.
Haskell does not have sub-typing, so the algorithm works much better there, although many popular type system extensions supported by GHC cause type inference to fail again, forcing you to provide explicit type signatures for some expressions.
In the end, it’s a trade-off between the power of the type system and the amount of type inference that can be done. Scala and Haskell have simply made different trade-offs.