State-of-the-Art: Semantics Acquisition Games - Semantic Acquisition Games: Harnessing Manpower for Creating Semantics - page 41

3.4 Domain Modeling Games
Apart from resource description, several semantics acquisition games have been
devised for building domain models: ontologies or other more lightweight semantic
structures like taxonomies or folksonomies. These SAGs—usually word games—
collect commonsense facts and transform them to ontology triplets. Some games
are focused rather on validating existing facts, others solve the important task of
connecting ontologies (ontology linking and ontology alignment), which is given
much attention in today's research [ 5 , 25 ].
3.4.1 Verbosity
For building a base of commonsense facts (e.g., “a cow is a mammal”), Luis von
Ahn devised the Verbosity game [ 24 ]. In this game, two random partners collaborate
to win, similarly to the ESP Game or Peekaboom. Also similarly to Peekaboom, the
players have different but complementing roles in the game: player A is given a task
term and player B has to correctly guess it.
Player A gives clues to player B. According to these clues the B has to guess the
task term given to A by the game. A clue is a sentence composed by player A using
one of the predefined “sentence stubs” and a word that player A may freely choose.
For example, if the player A is given a task word “milk”, he can construct a clue as
“It is usually located in the fridge ”, “It is a beverage ” or “It has color white ”. Such
sentences effectively tell some facts about the given term, but from the game purpose
point of view, they can be easily decoupled to the ontological predicates (equivalent
to sentence stubs, e.g., “located near”, “is a”, “has color”) and objects (“fridge”,
“beverage”, “white”). If the player B guesses the given word (which serves as the
subject of the triplet), it is highly probable that the clues (facts) provided by player
A are based on widely accepted truth.
Verbosity collects such assumptions and additionally validates them by “cross-
game” agreement (passing through only assumptions suggested independently by
multiple players). The validated triplets are afterward added to the ConceptNet 2
ontology and available for use. The Verbosity game produces accurate facts, but it is
bound by the limited set of predefined predicates.
3.4.2 GuessWhat!?
Another SAG addressing the issue of ontological relationship acquisition is the
GuessWhat!? game [ 13 ]. The player's task is to correctly guess the concept, for
which the game provides textual clues. The game is played simultaneously by two
2 http://csc.media.mit.edu/conceptnet