<s>
Teacher	B-Algorithm
forcing	I-Algorithm
is	O
an	O
algorithm	O
for	O
training	O
the	O
weights	O
of	O
recurrent	B-Algorithm
neural	I-Algorithm
networks	I-Algorithm
(	O
RNNs	O
)	O
.	O
</s>
<s>
The	O
term	O
"	O
teacher	B-Algorithm
forcing	I-Algorithm
"	O
can	O
be	O
motivated	O
by	O
comparing	O
the	O
RNN	O
to	O
a	O
human	O
student	O
taking	O
a	O
multi-part	O
exam	O
where	O
the	O
answer	O
to	O
each	O
part	O
(	O
for	O
example	O
a	O
mathematical	O
calculation	O
)	O
depends	O
on	O
the	O
answer	O
to	O
the	O
preceding	O
part	O
.	O
</s>
<s>
The	O
promise	O
is	O
,	O
that	O
teacher	B-Algorithm
forcing	I-Algorithm
helps	O
to	O
reduce	O
the	O
training	O
time	O
.	O
</s>
<s>
The	O
term	O
"	O
teacher	B-Algorithm
forcing	I-Algorithm
"	O
was	O
introduced	O
in	O
1989	O
by	O
Ronald	O
J	O
.	O
Williams	O
and	O
David	O
Zipser	O
,	O
who	O
reported	O
that	O
the	O
technique	O
was	O
already	O
being	O
"	O
frequently	O
used	O
in	O
dynamical	O
supervised	O
learning	O
tasks	O
"	O
around	O
that	O
time	O
.	O
</s>
