<s>
Natural-language	B-General_Concept
understanding	I-General_Concept
(	O
NLU	O
)	O
or	O
natural-language	O
interpretation	O
(	O
NLI	O
)	O
is	O
a	O
subtopic	O
of	O
natural-language	B-Language
processing	I-Language
in	O
artificial	B-Application
intelligence	I-Application
that	O
deals	O
with	O
machine	B-General_Concept
reading	I-General_Concept
comprehension	I-General_Concept
.	O
</s>
<s>
Natural-language	B-General_Concept
understanding	I-General_Concept
is	O
considered	O
an	O
AI-hard	B-General_Concept
problem	O
.	O
</s>
<s>
There	O
is	O
considerable	O
commercial	O
interest	O
in	O
the	O
field	O
because	O
of	O
its	O
application	O
to	O
automated	O
reasoning	O
,	O
machine	B-Application
translation	I-Application
,	O
question	B-Algorithm
answering	I-Algorithm
,	O
news-gathering	O
,	O
text	B-Algorithm
categorization	I-Algorithm
,	O
voice-activation	B-Application
,	O
archiving	O
,	O
and	O
large-scale	O
content	O
analysis	O
.	O
</s>
<s>
The	O
program	O
STUDENT	B-General_Concept
,	O
written	O
in	O
1964	O
by	O
Daniel	O
Bobrow	O
for	O
his	O
PhD	O
dissertation	O
at	O
MIT	O
,	O
is	O
one	O
of	O
the	O
earliest	O
known	O
attempts	O
at	O
natural-language	B-General_Concept
understanding	I-General_Concept
by	O
a	O
computer	O
.	O
</s>
<s>
Eight	O
years	O
after	O
John	O
McCarthy	O
coined	O
the	O
term	O
artificial	B-Application
intelligence	I-Application
,	O
Bobrow	O
's	O
dissertation	O
(	O
titled	O
Natural	O
Language	O
Input	O
for	O
a	O
Computer	O
Problem	O
Solving	O
System	O
)	O
showed	O
how	O
a	O
computer	O
could	O
understand	O
simple	O
natural	O
language	O
input	O
to	O
solve	O
algebra	O
word	O
problems	O
.	O
</s>
<s>
A	O
year	O
later	O
,	O
in	O
1965	O
,	O
Joseph	O
Weizenbaum	O
at	O
MIT	O
wrote	O
ELIZA	B-Application
,	O
an	O
interactive	O
program	O
that	O
carried	O
on	O
a	O
dialogue	O
in	O
English	O
on	O
any	O
topic	O
,	O
the	O
most	O
popular	O
being	O
psychotherapy	O
.	O
</s>
<s>
ELIZA	B-Application
worked	O
by	O
simple	O
parsing	O
and	O
substitution	O
of	O
key	O
words	O
into	O
canned	O
phrases	O
and	O
Weizenbaum	O
sidestepped	O
the	O
problem	O
of	O
giving	O
the	O
program	O
a	O
database	O
of	O
real-world	O
knowledge	O
or	O
a	O
rich	O
lexicon	B-Application
.	O
</s>
<s>
Yet	O
ELIZA	B-Application
gained	O
surprising	O
popularity	O
as	O
a	O
toy	O
project	O
and	O
can	O
be	O
seen	O
as	O
a	O
very	O
early	O
precursor	O
to	O
current	O
commercial	O
systems	O
such	O
as	O
those	O
used	O
by	O
Ask.com	B-Application
.	O
</s>
<s>
In	O
1969	O
,	O
Roger	O
Schank	O
at	O
Stanford	O
University	O
introduced	O
the	O
conceptual	O
dependency	O
theory	O
for	O
natural-language	B-General_Concept
understanding	I-General_Concept
.	O
</s>
<s>
Woods	O
introduced	O
the	O
augmented	B-Application
transition	I-Application
network	I-Application
(	O
ATN	O
)	O
to	O
represent	O
natural	O
language	O
input	O
.	O
</s>
<s>
Instead	O
of	O
phrase	O
structure	O
rules	O
ATNs	B-Application
used	O
an	O
equivalent	O
set	O
of	O
finite	B-Architecture
state	I-Architecture
automata	I-Architecture
that	O
were	O
called	O
recursively	O
.	O
</s>
<s>
ATNs	B-Application
and	O
their	O
more	O
general	O
format	O
called	O
"	O
generalized	O
ATNs	B-Application
"	O
continued	O
to	O
be	O
used	O
for	O
a	O
number	O
of	O
years	O
.	O
</s>
<s>
At	O
Stanford	O
,	O
Winograd	O
would	O
later	O
advise	O
Larry	O
Page	O
,	O
who	O
co-founded	O
Google	B-Application
.	I-Application
</s>
<s>
In	O
the	O
1970s	O
and	O
1980s	O
,	O
the	B-Language
natural	I-Language
language	I-Language
processing	I-Language
group	O
at	O
SRI	O
International	O
continued	O
research	O
and	O
development	O
in	O
the	O
field	O
.	O
</s>
<s>
A	O
number	O
of	O
commercial	O
efforts	O
based	O
on	O
the	O
research	O
were	O
undertaken	O
,	O
e.g.	O
,	O
in	O
1982	O
Gary	O
Hendrix	O
formed	O
Symantec	O
Corporation	O
originally	O
as	O
a	O
company	O
for	O
developing	O
a	O
natural	B-General_Concept
language	I-General_Concept
interface	I-General_Concept
for	O
database	O
queries	O
on	O
personal	O
computers	O
.	O
</s>
<s>
However	O
,	O
with	O
the	O
advent	O
of	O
mouse-driven	O
graphical	B-Application
user	I-Application
interfaces	I-Application
,	O
Symantec	O
changed	O
direction	O
.	O
</s>
<s>
A	O
number	O
of	O
other	O
commercial	O
efforts	O
were	O
started	O
around	O
the	O
same	O
time	O
,	O
e.g.	O
,	O
Larry	O
R	O
.	O
Harris	O
at	O
the	B-Application
Artificial	I-Application
Intelligence	I-Application
Corporation	O
and	O
Roger	O
Schank	O
and	O
his	O
students	O
at	O
Cognitive	B-Application
Systems	I-Application
Corp	O
.	O
</s>
<s>
The	O
third	O
millennium	O
saw	O
the	O
introduction	O
of	O
systems	O
using	O
machine	O
learning	O
for	O
text	B-Algorithm
classification	I-Algorithm
,	O
such	O
as	O
the	O
IBM	B-Application
Watson	I-Application
.	O
</s>
<s>
However	O
,	O
experts	O
debate	O
how	O
much	O
"	O
understanding	O
"	O
such	O
systems	O
demonstrate	O
:	O
e.g.	O
,	O
according	O
to	O
John	O
Searle	O
,	O
Watson	B-Application
did	O
not	O
even	O
understand	O
the	O
questions	O
.	O
</s>
<s>
Natural	B-Language
language	I-Language
processing	I-Language
has	O
made	O
inroads	O
for	O
applications	O
to	O
support	O
human	O
productivity	O
in	O
service	O
and	O
ecommerce	O
,	O
but	O
this	O
has	O
largely	O
been	O
made	O
possible	O
by	O
narrowing	O
the	O
scope	O
of	O
the	O
application	O
.	O
</s>
<s>
There	O
are	O
thousands	O
of	O
ways	O
to	O
request	O
something	O
in	O
a	O
human	O
language	O
that	O
still	O
defies	O
conventional	O
natural	B-Language
language	I-Language
processing	I-Language
.	O
</s>
<s>
The	O
umbrella	O
term	O
"	O
natural-language	B-General_Concept
understanding	I-General_Concept
"	O
can	O
be	O
applied	O
to	O
a	O
diverse	O
set	O
of	O
computer	O
applications	O
,	O
ranging	O
from	O
small	O
,	O
relatively	O
simple	O
tasks	O
such	O
as	O
short	O
commands	O
issued	O
to	O
robots	B-Algorithm
,	O
to	O
highly	O
complex	O
endeavors	O
such	O
as	O
the	O
full	O
comprehension	O
of	O
newspaper	O
articles	O
or	O
poetry	O
passages	O
.	O
</s>
<s>
Many	O
real-world	O
applications	O
fall	O
between	O
the	O
two	O
extremes	O
,	O
for	O
instance	O
text	B-Algorithm
classification	I-Algorithm
for	O
the	O
automatic	O
analysis	O
of	O
emails	O
and	O
their	O
routing	O
to	O
a	O
suitable	O
department	O
in	O
a	O
corporation	O
does	O
not	O
require	O
an	O
in-depth	O
understanding	O
of	O
the	O
text	O
,	O
but	O
needs	O
to	O
deal	O
with	O
a	O
much	O
larger	O
vocabulary	O
and	O
more	O
diverse	O
syntax	O
than	O
the	O
management	O
of	O
simple	O
queries	O
to	O
database	O
tables	O
with	O
fixed	O
schemata	O
.	O
</s>
<s>
For	O
example	O
,	O
Wayne	O
Ratliff	O
originally	O
developed	O
the	O
Vulcan	O
program	O
with	O
an	O
English-like	O
syntax	O
to	O
mimic	O
the	O
English	O
speaking	O
computer	O
in	O
Star	B-Operating_System
Trek	I-Operating_System
.	O
</s>
<s>
Vulcan	O
later	O
became	O
the	O
dBase	B-Application
system	O
whose	O
easy-to-use	O
syntax	O
effectively	O
launched	O
the	O
personal	O
computer	O
database	O
industry	O
.	O
</s>
<s>
Systems	O
with	O
an	O
easy	O
to	O
use	O
or	O
English	O
like	O
syntax	O
are	O
,	O
however	O
,	O
quite	O
distinct	O
from	O
systems	O
that	O
use	O
a	O
rich	O
lexicon	B-Application
and	O
include	O
an	O
internal	O
representation	O
(	O
often	O
as	O
first	O
order	O
logic	O
)	O
of	O
the	O
semantics	B-Application
of	O
natural	O
language	O
sentences	O
.	O
</s>
<s>
The	O
system	O
needs	O
a	O
lexicon	B-Application
of	O
the	O
language	O
and	O
a	O
parser	B-Language
and	O
grammar	O
rules	O
to	O
break	O
sentences	O
into	O
an	O
internal	O
representation	O
.	O
</s>
<s>
The	O
construction	O
of	O
a	O
rich	O
lexicon	B-Application
with	O
a	O
suitable	O
ontology	O
requires	O
significant	O
effort	O
,	O
e.g.	O
,	O
the	B-General_Concept
Wordnet	I-General_Concept
lexicon	B-Application
required	O
many	O
person-years	O
of	O
effort	O
.	O
</s>
<s>
The	O
system	O
also	O
needs	O
theory	O
from	O
semantics	B-Application
to	O
guide	O
the	O
comprehension	O
.	O
</s>
<s>
These	O
range	O
from	O
naive	B-General_Concept
semantics	I-General_Concept
or	O
stochastic	B-General_Concept
semantic	I-General_Concept
analysis	I-General_Concept
to	O
the	O
use	O
of	O
pragmatics	O
to	O
derive	O
meaning	O
from	O
context	O
.	O
</s>
<s>
Semantic	B-Application
parsers	I-Application
convert	O
natural-language	O
texts	O
into	O
formal	O
meaning	O
representations	O
.	O
</s>
<s>
Advanced	O
applications	O
of	O
natural-language	B-General_Concept
understanding	I-General_Concept
also	O
attempt	O
to	O
incorporate	O
logical	O
inference	O
within	O
their	O
framework	O
.	O
</s>
<s>
Therefore	O
,	O
systems	O
based	O
on	O
functional	O
languages	O
such	O
as	O
Lisp	B-Language
need	O
to	O
include	O
a	O
subsystem	O
to	O
represent	O
logical	O
assertions	O
,	O
while	O
logic-oriented	O
systems	O
such	O
as	O
those	O
using	O
the	O
language	O
Prolog	B-Language
generally	O
rely	O
on	O
an	O
extension	O
of	O
the	O
built-in	O
logical	O
representation	O
framework	O
.	O
</s>
<s>
The	O
management	O
of	O
context	O
in	O
natural-language	B-General_Concept
understanding	I-General_Concept
can	O
present	O
special	O
challenges	O
.	O
</s>
