Abstract

This paper studies the problem of learning from ambiguous supervision, focusing on the task of learning semantic correspondences. A learning problem is said to be ambiguously supervised when, for a given training input, a set of output candidates is provided with no prior of which one is correct. We propose to tackle this problem by solving a related unambiguous task with a label ranking approach and show how and why this performs well on the original task, via the method of task-transfer. We apply it to learning to match natural language sentences to a structured representation of their meaning and empirically demonstrate that this competes with the state-of-the-art on two benchmarks.