State-of-the-Art: Semantics Acquisition Games - Semantic Acquisition Games: Harnessing Manpower for Creating Semantics - page 39

Different to the previous two games is the Listen Game designed by Turnbull
et al. [ 20 ]. The difference lies in the type of the primary action that player does in
the game. Instead of writing his own tags, he has to pick one which best matches the
track from the list (in fact, he also has to pick a tag he considers worst). This has
two effects. First, the gameplay mechanics is simpler, because the player does not
have to type the words anymore, he just selects them (which is easier). Second, this
effectively changes the resource description acquisition from tagging to (multiclass)
categorization (with all its positive and negative effects).
A more social approach in music metadata acquisition game was applied by
Barrington et al. [ 1 ] (members of the Turnbull's group). They too devised a cat-
egorization game similar to Listen Game—the players received points when they
agreed on a certain answers possible to a questions they were given about the music
(e.g. what is the sub-genre of the music, what is the prominent instrument playing).
Moreover, the game was heavily relying on the social aspects of its players. To attract
the players, the game was propagated via social networks. In addition, it kept a track
about players' desires and areas of interests in the music field and matched similar
players together, implying that players liking similar or the same music would also
think similar about it. The authors also used these “player models” for personalized
recommendation of music outside the game.
Morton et al. designed an unique SAG called Moodswings [ 14 ]. As its name sug-
gest, the game focuses exclusively on the acquisition of information about the mood
of played tracks (resp. the mood it evokes upon listening). In particular, the game
collects the changes of the mood—with per-second accuracy. The players play the
game in pairs and they must agree (in regular intervals) on the mood characterizing
the played track (just as in ESP). An interesting feature is that players do not inter-
act with nominal metadata values (e.g. characteristic words they would type-it or
select from a list). Instead, they set the mood they perceive using a two-dimensional
continuous scale on which two axes (horizontal and vertical) represent two mood
dimensions: positiveness/negativeness and emotional intensity (see example in the
Fig. 3.4 ).
Overall, the employment of SAGs in audio resource annotation and description
seems to be a working idea. The approaches we reviewed deliver valid metadata of
many desired types. Many reports on the existing “music SAGs” also reported that
the players much liked the interaction with the music samples and the interaction
with abstract symbols (tags, categories) describing them. Hearing and playing with
music is a relaxing experience for them. Therefore, any music-based SAGs have an
implicit advantage in this from the start.
3.3 Text Annotation Games
Description of textual documents as single entities, in comparison to multime-
dia description, is not so dependent on human mind labor and we are not aware
of any SAGs in this field. After all, the metadata about textual documents (such
as characteristic terms) are being sufficiently extracted by automated approaches.