CityLights: A Method for Music Metadata Validation - Semantic Acquisition Games: Harnessing Manpower for Creating Semantics - page 80

Fig. 6.1 An example of the CityLights game interface
1. (a “correct” choice composed of tags) lovely , 19th century and classic .
2. (a choice composed of tags belonging toBeethoven's Egmont overture) Beethoven ,
classic and dramatic .
3. (a choice composed of tags belonging to Mozart's Turkish march ) march , funny
and piano .
The setup of this question is a bit tricky for the player. At first, the “correct” choice
is composed of quite “unfortunate” tags. The tag lovely is very subjective and objec-
tively has no distinctive power. The 19th century tag is correct, but very general and
the classic tag, though correct in a broader sense (classical music), is in fact wrong,
because the piece belongs to the romantic period (rather than classical).
On the other hand, the second choice looks much more promising. The Beethoven
is a strong clue, valid not only for the second choice, but also for the first. Also, the
tag dramatic characterizes the 5th symphony as well as Egmont overture . The third
choice is in this case a very distinct to what the player hears, but the second choice
looks most promising, so player picks it.
After the wrong choice pick, the player looses some points and can pick again.
Now, he has a good chance of picking the “correct” (i.e. the first) choice. The player
might be confused by the outcome of his first answer, but he knows about the nature
of the game, e.g. that he is dealing with a noisy dataset. He simply follows the game
task: identifying the original set. By choosing the wrong choice first, however, he
indirectly indicates that the tags present in the wrong choice are comparably better
describing the played track that those in the first choice. If the player, on the other
hand, would pick the correct option first (perhaps in a different question), he indirectly
states that out of the present choices, these tags represent the music track best.