<s>
PRAC	O
(	O
Probabilistic	B-Algorithm
Action	I-Algorithm
Cores	I-Algorithm
)	O
is	O
an	O
interpreter	B-General_Concept
for	I-General_Concept
natural-language	I-General_Concept
instructions	O
for	O
robotic	O
applications	O
developed	O
at	O
the	O
Institute	O
for	O
Artificial	O
Intelligence	O
at	O
the	O
University	O
of	O
Bremen	O
,	O
Germany	O
,	O
and	O
is	O
supported	O
in	O
parts	O
by	O
the	O
European	O
Commission	O
and	O
the	O
German	O
Research	O
Foundation	O
(	O
DFG	O
)	O
.	O
</s>
<s>
To	O
this	O
end	O
,	O
PRAC	O
maintains	O
probabilistic	B-Algorithm
first-order	I-Algorithm
knowledge	I-Algorithm
bases	I-Algorithm
over	O
semantic	B-Application
networks	O
represented	O
in	O
Markov	O
logic	O
networks	O
.	O
</s>
<s>
As	O
opposed	O
to	O
other	O
semantic	B-Application
learning	O
initiatives	O
like	O
NELL	B-Algorithm
or	O
IBM	O
's	O
Watson	B-Application
,	O
PRAC	O
does	O
not	O
aim	O
at	O
answering	O
questions	O
in	O
natural	O
language	O
,	O
but	O
to	O
disambiguate	O
and	O
infer	O
information	O
pieces	O
that	O
are	O
missing	O
in	O
natural-language	O
instructions	O
,	O
such	O
that	O
they	O
can	O
be	O
executed	O
by	O
a	O
robot	O
.	O
</s>
<s>
In	O
addition	O
to	O
probabilistic	O
relational	O
models	O
,	O
PRAC	O
uses	O
the	O
principles	O
of	O
analogical	O
reasoning	O
and	O
instance-based	B-General_Concept
learning	I-General_Concept
to	O
infer	O
completions	O
of	O
roles	O
in	O
semantic	B-Application
networks	O
.	O
</s>
