<s>
Semantic	B-General_Concept
folding	I-General_Concept
theory	O
describes	O
a	O
procedure	O
for	O
encoding	O
the	O
semantics	B-Application
of	O
natural	O
language	O
text	O
in	O
a	O
semantically	O
grounded	O
binary	O
representation	O
.	O
</s>
<s>
Semantic	B-General_Concept
folding	I-General_Concept
theory	O
draws	O
inspiration	O
from	O
Douglas	O
R	O
.	O
Hofstadter	O
's	O
Analogy	O
as	O
the	O
Core	O
of	O
Cognition	O
which	O
suggests	O
that	O
the	O
brain	O
makes	O
sense	O
of	O
the	O
world	O
by	O
identifying	O
and	O
applying	O
analogies	O
.	O
</s>
<s>
The	O
theory	O
hypothesises	O
that	O
semantic	O
data	O
must	O
therefore	O
be	O
introduced	O
to	O
the	O
neocortex	O
in	O
such	O
a	O
form	O
as	O
to	O
allow	O
the	O
application	O
of	O
a	O
similarity	O
measure	O
and	O
offers	O
,	O
as	O
a	O
solution	O
,	O
the	O
sparse	B-Algorithm
binary	B-Data_Structure
vector	I-Data_Structure
employing	O
a	O
two-dimensional	O
topographic	O
semantic	O
space	O
as	O
a	O
distributional	O
reference	O
frame	O
.	O
</s>
<s>
The	O
theory	O
builds	O
on	O
the	O
computational	O
theory	O
of	O
the	O
human	O
cortex	O
known	O
as	O
hierarchical	B-Algorithm
temporal	I-Algorithm
memory	I-Algorithm
(	O
HTM	O
)	O
,	O
and	O
positions	O
itself	O
as	O
a	O
complementary	O
theory	O
for	O
the	O
representation	O
of	O
language	O
semantics	B-Application
.	O
</s>
<s>
Analogous	O
to	O
the	O
structure	O
of	O
the	O
neocortex	O
,	O
Semantic	B-General_Concept
Folding	I-General_Concept
theory	O
posits	O
the	O
implementation	O
of	O
a	O
semantic	O
space	O
as	O
a	O
two-dimensional	O
grid	O
.	O
</s>
<s>
This	O
vector	O
space	O
model	O
is	O
presented	O
in	O
the	O
theory	O
as	O
an	O
equivalence	O
to	O
the	O
well	O
known	O
word	O
space	O
model	O
described	O
in	O
the	O
information	B-Library
retrieval	I-Library
literature	O
.	O
</s>
<s>
The	O
result	O
of	O
this	O
process	O
will	O
be	O
a	O
word-vector	O
containing	O
all	O
the	O
contexts	O
in	O
which	O
the	O
word	O
Y	O
appears	O
and	O
will	O
therefore	O
be	O
representative	O
of	O
the	O
semantics	B-Application
of	O
that	O
word	O
in	O
the	O
semantic	O
space	O
.	O
</s>
<s>
It	O
can	O
be	O
seen	O
that	O
the	O
resulting	O
word-vector	O
is	O
also	O
in	O
a	O
sparse	B-Algorithm
distributed	O
representation	O
(	O
SDR	O
)	O
format	O
[	O
Schütze	O
,	O
1993 ]	O
&	O
[	O
Sahlgreen	O
,	O
2006 ]	O
.	O
</s>
<s>
Some	O
properties	O
of	O
word-SDRs	O
that	O
are	O
of	O
particular	O
interest	O
with	O
respect	O
to	O
computational	B-General_Concept
semantics	I-General_Concept
are	O
:	O
</s>
<s>
The	O
application	O
of	O
semantic	O
spaces	O
in	O
natural	B-Language
language	I-Language
processing	I-Language
(	O
NLP	B-Language
)	O
aims	O
at	O
overcoming	O
limitations	O
of	O
rule-based	O
or	O
model-based	O
approaches	O
operating	O
on	O
the	B-Application
keyword	I-Application
level	O
.	O
</s>
<s>
The	O
main	O
drawback	O
with	O
these	O
approaches	O
is	O
their	O
brittleness	O
,	O
and	O
the	O
large	O
manual	O
effort	O
required	O
to	O
create	O
either	O
rule-based	O
NLP	B-Language
systems	O
or	O
training	O
corpora	O
for	O
model	O
learning	O
.	O
</s>
<s>
Rule-based	O
and	O
machine	O
learning-based	O
models	O
are	O
fixed	O
on	O
the	B-Application
keyword	I-Application
level	O
and	O
break	O
down	O
if	O
the	O
vocabulary	O
differs	O
from	O
that	O
defined	O
in	O
the	O
rules	O
or	O
from	O
the	O
training	O
material	O
used	O
for	O
the	O
statistical	O
models	O
.	O
</s>
<s>
"	O
spider-web	O
"	O
,	O
"	O
lighter-cigarette	O
"	O
,	O
as	O
opposed	O
to	O
synonymous	O
relations	O
such	O
as	O
"	O
whale-dolphin	O
"	O
,	O
"	O
astronaut-driver	O
"	O
)	O
was	O
achieved	O
by	O
explicit	B-General_Concept
semantic	I-General_Concept
analysis	I-General_Concept
(	O
ESA	O
)	O
in	O
2007	O
.	O
</s>
<s>
More	O
recently	O
,	O
advances	O
in	O
neural	B-Architecture
networking	I-Architecture
techniques	O
in	O
combination	O
with	O
other	O
new	O
approaches	O
(	O
tensors	B-Device
)	O
led	O
to	O
a	O
host	O
of	O
new	O
recent	O
developments	O
:	O
Word2vec	B-Algorithm
from	O
Google	B-Application
and	O
GloVe	B-General_Concept
from	O
Stanford	O
University	O
.	O
</s>
<s>
Semantic	B-General_Concept
folding	I-General_Concept
represents	O
a	O
novel	O
,	O
biologically	O
inspired	O
approach	O
to	O
semantic	O
spaces	O
where	O
each	O
word	O
is	O
represented	O
as	O
a	O
sparse	B-Algorithm
binary	B-Data_Structure
vector	I-Data_Structure
with	O
16,000	O
dimensions	O
(	O
a	O
semantic	O
fingerprint	O
)	O
in	O
a	O
2D	O
semantic	O
map	O
(	O
the	O
semantic	O
universe	O
)	O
.	O
</s>
<s>
Sparse	B-Algorithm
binary	O
representation	O
are	O
advantageous	O
in	O
terms	O
of	O
computational	O
efficiency	O
,	O
and	O
allow	O
for	O
the	O
storage	O
of	O
very	O
large	O
numbers	O
of	O
possible	O
patterns	O
.	O
</s>
<s>
The	O
topological	O
distribution	O
over	O
a	O
two-dimensional	O
grid	O
(	O
outlined	O
above	O
)	O
lends	O
itself	O
to	O
a	O
bitmap	O
type	O
visualization	O
of	O
the	O
semantics	B-Application
of	O
any	O
word	O
or	O
text	O
,	O
where	O
each	O
active	O
semantic	O
feature	O
can	O
be	O
displayed	O
as	O
e.g.	O
</s>
<s>
a	O
pixel	B-Algorithm
.	O
</s>
<s>
As	O
can	O
be	O
seen	O
in	O
the	O
images	O
shown	O
here	O
,	O
this	O
representation	O
allows	O
for	O
a	O
direct	O
visual	O
comparison	O
of	O
the	O
semantics	B-Application
of	O
two	O
(	O
or	O
more	O
)	O
linguistic	O
items	O
.	O
</s>
<s>
Image	O
1	O
clearly	O
demonstrates	O
that	O
the	O
two	O
disparate	O
terms	O
"	O
dog	O
"	O
and	O
"	O
car	O
"	O
have	O
,	O
as	O
expected	O
,	O
very	O
obviously	O
different	O
semantics	B-Application
.	O
</s>
<s>
The	O
visualization	O
of	O
semantic	O
similarity	O
using	O
Semantic	B-General_Concept
Folding	I-General_Concept
bears	O
a	O
strong	O
resemblance	O
to	O
the	O
fMRI	B-Algorithm
images	O
produced	O
in	O
a	O
research	O
study	O
conducted	O
by	O
A.G.	O
Huth	O
et	O
al.	O
,	O
where	O
it	O
is	O
claimed	O
that	O
words	O
are	O
grouped	O
in	O
the	O
brain	O
by	O
meaning	O
.	O
</s>
<s>
voxels	B-Algorithm
,	O
little	O
volume	O
segments	O
of	O
the	O
brain	O
,	O
were	O
found	O
to	O
follow	O
a	O
pattern	O
were	O
semantic	O
information	O
is	O
represented	O
along	O
the	O
boundary	O
of	O
the	O
visual	O
cortex	O
with	O
visual	O
and	O
linguistic	O
categories	O
represented	O
on	O
posterior	O
and	O
anterior	O
side	O
respectively	O
.	O
</s>
