My Word! Machine versus Human Computation Methods for Identifying and Resolving Acronyms

Abstract

Acronyms are commonly used in human language as alternative forms of concepts to increase recognition, to reduce duplicate references to the same concept, and to stress important concepts. There are no standard rules for acronym creation; therefore, both machine-based acronym identification and ac-ronym resolution are highly prone to error. This might be resolved by a human computation approach, which can take advantage of knowledge external to the document collection. Using three text collections with different properties, we compare a machine-based algorithm with a crowdsourcing approach to identify acronyms. We then perform acronym resolution using these two approaches, plus a game-based approach. The crowd and game-based methods outperform the machine algorithm, even when external information is not used. Also, crowd and game formats offered similar performance with a difference in cost.