A splitter is the routine or algorithm used during decision tree training to identify the best condition at a node. It evaluates possible feature tests and thresholds to improve the quality of resulting branches.
While training a decision tree, the routine (and algorithm) responsible for finding the best condition at each node.