State-of-the-Art: Semantics Acquisition Games - Semantic Acquisition Games: Harnessing Manpower for Creating Semantics - page 40

Fig. 3.4 Example of the Moodswings interface. The players are moving the cursor over the two
dimensions to describe the actual mood of the music track they are listening to [ 14 ]
However, if we leave the summary characteristics of texts, there is a sentence-level
text processing task, where SAGs can be and are successfully exploited: the entity
co-reference identification [ 2 , 7 ].
The co-reference identification has an objective to detect sets of words in the text
referring to a same concept or instance. For example, in text “John was the guilty one,
he threw the stone” the words “John” and “he” refer to the same subject (they are in
co-reference). Such co-references are usually caused by the usage of pronouns and
synonyms (mainly due to the aesthetics of the text). This however, creates an obstacle
for automated text processing approaches, for example those mining the ontological
facts. The co-references may span across larger portions of texts and multiple of
them may overlap. All this makes co-reference identification an extremely difficult
task to be solved by automated methods.
To do the job by human labor, two SAGs were devised: PlayCoref (by Hladka
et al.) [ 7 ] and PhraseDetectives (by Chamberlain et al.) [ 2 ]. In PlayCoref, two play-
ers are racing in marking the co-references by matching nouns with pronouns. The
player score is afterward computed by validating his guesses against the opponent and
also by comparison to results of an automated co-reference detection approach [ 7 ].
PhraseDetectives is slightly different, there, the game is done in two rounds: annota-
tion and validation (in which the opponents validates guesses of the each other) [ 2 ].
These games shows that it is not a completely bad idea to think about games for
playing with text. They support the argument against the premise that there are more
attractive games with rich graphical interfaces and that those games will prevail.
However, the textual games may have a “rule comprehension barrier” (as stressed
by Hladka et al.) because textual games such as PlayCoref or PhraseDetectives are
not conventional and thus not understood intuitively.