Shortening text (Cut down text without meaning loss)
Human Macro Task (Define/change tense of an article, find references which are out of date, find images which complement text)
In order to do this Soylent utilizes Turkit, an API for running iterative tasks on Amazons Mechanical Turk service. For anyone unfamiliar service Amazons Mechanical Turk is a service where humans(Turkers)
perform small tasks for small amount of money. Soylent uses the Turk service to perform the three main tasks previously mentioned but instead of one “Turker” reviewing and rewriting the document it distributes the work amongst a group of workers, then also utilizes a group of workers to unify and check the distributed work, back into a cohesive solution. The video below illustrates how this is implemented within word;
The students behind Soylent believe that through the use of their system a more efficient and effective message can be created than that of a single writer.
By crowdsourcing these small tasks Soylent is really just applying a behavior which already exists. Think about how many times you’ve asked a colleague or friend to perform the same tasks above. Granted there are major confidentiality issues associated with outsourcing/crowdsourcing editing but as an implemented embedded solution in the word processing program I think it provides an interesting stop gap solution until machines can truly understand content.