Information Technology Reference
In-Depth Information
the characters that are in the story, then about the events that take place in the story,
hence about temporal relations among the events, and finally about causal-temporal
relations among events (see taxonomy in Fig. 1). Accordingly, factual, temporal,
and causal smart games are the actual implementation of the corresponding com-
prehension tasks.
Fig. 1
Smart games taxonomy
In TERENCE, smart games are automatically generated as follows (see Fig. 2):
Phase A.
Firstly, from a story text contained in the story repository, an NLP mod-
ule generates a story annotated with a variant of the TimeML language, that was
extended in [9] with tags that are relevant for the TERENCE smart games. For
instance, the ENTITY and CLINK tags aim, respectively, at representing the en-
tity related to an event, and the causal-temporal relations between two events.
The annotated story is then stored in the same repository.
Phase B.
Then, a reasoner checks the consistency of the annotations, detects the
eventual temporal inconsistencies, and enriches the annotations by adding de-
duced temporal relations as further TLINK tags [6]. This new consistent and
enriched story is also stored in the story repository.
Phase C.
Starting from the consistent and enriched story, the reasoner module
generates automatically instances of smart games. For instance, to create a WHO-
game related to a certain event [7, 5]:
•
the ENTITY that participates in the event with a role of protagonist is selected
as the correct answer;
•
other two entities that are not related to the event and are different from the
entity selected above are added as wrong answers;
•
the question asked to the learner is generated through a text-generation mod-
ule (e.g. if the event is that “Ernesta is riding
1
a bike”, the question will be
“Who is riding a bike?”).
The resulting games are then stored into the game repository.
1
The verb “to ride” is detected as an event.
Search WWH ::
Custom Search