<s>
Bidirectional	B-Algorithm
recurrent	I-Algorithm
neural	I-Algorithm
networks	I-Algorithm
(	O
BRNN	O
)	O
connect	O
two	O
hidden	O
layers	O
of	O
opposite	O
directions	O
to	O
the	O
same	O
output	O
.	O
</s>
<s>
For	O
example	O
,	O
multilayer	B-Algorithm
perceptron	I-Algorithm
(	O
MLPs	O
)	O
and	O
time	B-Algorithm
delay	I-Algorithm
neural	I-Algorithm
network	I-Algorithm
(	O
TDNNs	B-Algorithm
)	O
have	O
limitations	O
on	O
the	O
input	O
data	O
flexibility	O
,	O
as	O
they	O
require	O
their	O
input	O
data	O
to	O
be	O
fixed	O
.	O
</s>
<s>
Standard	O
recurrent	B-Algorithm
neural	I-Algorithm
network	I-Algorithm
(	O
RNNs	O
)	O
also	O
have	O
restrictions	O
as	O
the	O
future	O
input	O
information	O
cannot	O
be	O
reached	O
from	O
the	O
current	O
state	O
.	O
</s>
<s>
For	O
example	O
,	O
in	O
handwriting	B-Application
recognition	I-Application
,	O
the	O
performance	O
can	O
be	O
enhanced	O
by	O
knowledge	O
of	O
the	O
letters	O
located	O
before	O
and	O
after	O
the	O
current	O
letter	O
.	O
</s>
