| dc.description.abstract | Natural Language Processing, is an important collection of methods for processing the vast
amounts of available natural language text we continually produce. These methods make
use of supervised learning, an approach that learns from large amounts of annotated
data. As humans, we’re able to provide information about text that such systems can learn from.
Historically, this was carried out by small groups of experts. However, this did not scale. This led
to various crowdsourcing approaches being taken that used large pools of non-experts.
The traditional form of crowdsourcing was to pay users small amounts of money to complete
tasks. As time progressed, gamification approaches such as GWAPs, showed various benefits
over the micro-payment methods used before. These included a cost saving, worker training
opportunities, increased worker engagement and potential to far exceed the scale of crowdsourcing.
While these were successful in domains such as image labelling, they struggled in the domain
of text annotation, which wasn’t such a natural fit. Despite many challenges, there were also
clearly many opportunities and benefits to applying this approach to text annotation. Many of
these are demonstrated by Phrase Detectives. Based on lessons learned from Phrase Detectives
and investigations into other GWAPs, in this work, we attempt to create full GWAPs for NLP,
extracting the benefits of the methodology. This includes training, high quality output from
non-experts and a truly game-like GWAP design that players are happy to play voluntarily. | en_US |
