Abstract
This thesis explores different strategies for constructing robust, inexpensive and empirically-derived word sense inventories and the corresponding sense-annotated\r corpus. All strategies explored rely on non-expert linguistic annotations collected\r through the use of the Amazon Mechanical Turk crowdsourcing marketplace. Experiments using implementation strategies with different quality control mechanisms are reported on in detail. Described herein are multiple best practices discovered through extensive system testing that are required to obtain high quality\r data given the challenge of using non-expert annotations. Results indicate that it is possible to obtain sense inventories that correlate with the gold standard,\r extending it in ways that may prove useful in a variety of other Natural Language\r Processing tasks.