Efficient Disambiguation of Task Instructions in Crowdsourcing

Crowdsourcing allows users to offload tedious work to an on-demand workforce. However, the time saved by the requesters is often offset by the time they must spend preparing instructions and refining them to address the ambiguities that typically arise. If crowdsourcing is to become viable, and result in net gains for requesters, requesters must be able to obtain high-quality results with a low investment of time in writing instructions. That might mean finding ways to accommodate hastily written instructions. Instruction quality could be improved by resolving ambiguities either with help of crowd workers, or by using NLP-based tools. 

In this dissertation, I present 1) a taxonomy of ambiguities that can occur in task instructions, 2) a workflow that enables requesters to resolve ambiguities before posting them to workers, 3) a set of methods to improve the quality of instructions while workers are working on the task, and finally, 4) a system that leverages current NLP technologies to detect ambiguities automatically before they are posted to the workers.