When the host check for a topological host enters a failure state (i.e. WARN or CRIT), the service monitors for hosts that depend on that element will have their alerts / actions suppressed and enter an UNKN state until the host check on the topological host recovers.  However, there are a few scenarios where you may still notice alerts being sent when it appears that the topological host has failed:

As a general rule, your topological parent's host check should check as often as the most checked service on any of the child elements and have a re-check interval / max re-checks value that is shorter than the most checked service.