<s>
Natural	B-Language
language	I-Language
processing	I-Language
(	O
NLP	B-Language
)	O
is	O
an	O
interdisciplinary	O
subfield	O
of	O
linguistics	O
,	O
computer	B-General_Concept
science	I-General_Concept
,	O
and	O
artificial	B-Application
intelligence	I-Application
concerned	O
with	O
the	O
interactions	O
between	O
computers	O
and	O
human	O
language	O
,	O
in	O
particular	O
how	O
to	O
program	O
computers	O
to	O
process	O
and	O
analyze	O
large	O
amounts	O
of	O
natural	O
language	O
data	O
.	O
</s>
<s>
Challenges	O
in	O
natural	B-Language
language	I-Language
processing	I-Language
frequently	O
involve	O
speech	B-Application
recognition	I-Application
,	O
natural-language	B-General_Concept
understanding	I-General_Concept
,	O
and	O
natural-language	B-General_Concept
generation	I-General_Concept
.	O
</s>
<s>
Natural	B-Language
language	I-Language
processing	I-Language
has	O
its	O
roots	O
in	O
the	O
1950s	O
.	O
</s>
<s>
Already	O
in	O
1950	O
,	O
Alan	O
Turing	O
published	O
an	O
article	O
titled	O
"	O
Computing	O
Machinery	O
and	O
Intelligence	O
"	O
which	O
proposed	O
what	O
is	O
now	O
called	O
the	O
Turing	O
test	O
as	O
a	O
criterion	O
of	O
intelligence	O
,	O
though	O
at	O
the	O
time	O
that	O
was	O
not	O
articulated	O
as	O
a	O
problem	O
separate	O
from	O
artificial	B-Application
intelligence	I-Application
.	I-Application
</s>
<s>
The	O
premise	O
of	O
symbolic	O
NLP	B-Language
is	O
well-summarized	O
by	O
John	O
Searle	O
's	O
Chinese	O
room	O
experiment	O
:	O
Given	O
a	O
collection	O
of	O
rules	O
(	O
e.g.	O
,	O
a	O
Chinese	O
phrasebook	O
,	O
with	O
questions	O
and	O
matching	O
answers	O
)	O
,	O
the	O
computer	O
emulates	O
natural	B-General_Concept
language	I-General_Concept
understanding	I-General_Concept
(	O
or	O
other	O
NLP	B-Language
tasks	O
)	O
by	O
applying	O
those	O
rules	O
to	O
the	O
data	O
it	O
confronts	O
.	O
</s>
<s>
1950s	O
:	O
The	O
Georgetown	B-General_Concept
experiment	I-General_Concept
in	O
1954	O
involved	O
fully	O
automatic	B-Application
translation	I-Application
of	O
more	O
than	O
sixty	O
Russian	O
sentences	O
into	O
English	O
.	O
</s>
<s>
The	O
authors	O
claimed	O
that	O
within	O
three	O
or	O
five	O
years	O
,	O
machine	B-Application
translation	I-Application
would	O
be	O
a	O
solved	O
problem	O
.	O
</s>
<s>
However	O
,	O
real	O
progress	O
was	O
much	O
slower	O
,	O
and	O
after	O
the	O
ALPAC	B-General_Concept
report	I-General_Concept
in	O
1966	O
,	O
which	O
found	O
that	O
ten-year-long	O
research	O
had	O
failed	O
to	O
fulfill	O
the	O
expectations	O
,	O
funding	O
for	O
machine	B-Application
translation	I-Application
was	O
dramatically	O
reduced	O
.	O
</s>
<s>
Little	O
further	O
research	O
in	O
machine	B-Application
translation	I-Application
was	O
conducted	O
in	O
America	O
(	O
though	O
some	O
research	O
continued	O
elsewhere	O
,	O
such	O
as	O
Japan	O
and	O
Europe	O
)	O
until	O
the	O
late	O
1980s	O
when	O
the	O
first	O
statistical	B-General_Concept
machine	I-General_Concept
translation	I-General_Concept
systems	O
were	O
developed	O
.	O
</s>
<s>
1960s	O
:	O
Some	O
notably	O
successful	O
natural	B-Language
language	I-Language
processing	I-Language
systems	O
developed	O
in	O
the	O
1960s	O
were	O
SHRDLU	O
,	O
a	O
natural	O
language	O
system	O
working	O
in	O
restricted	O
"	O
blocks	B-General_Concept
worlds	I-General_Concept
"	O
with	O
restricted	O
vocabularies	O
,	O
and	O
ELIZA	B-Application
,	O
a	O
simulation	O
of	O
a	O
Rogerian	O
psychotherapist	O
,	O
written	O
by	O
Joseph	O
Weizenbaum	O
between	O
1964	O
and	O
1966	O
.	O
</s>
<s>
Using	O
almost	O
no	O
information	O
about	O
human	O
thought	O
or	O
emotion	O
,	O
ELIZA	B-Application
sometimes	O
provided	O
a	O
startlingly	O
human-like	O
interaction	O
.	O
</s>
<s>
When	O
the	O
"	O
patient	O
"	O
exceeded	O
the	O
very	O
small	O
knowledge	O
base	O
,	O
ELIZA	B-Application
might	O
provide	O
a	O
generic	O
response	O
,	O
for	O
example	O
,	O
responding	O
to	O
"	O
My	O
head	O
hurts	O
"	O
with	O
"	O
Why	O
do	O
you	O
say	O
your	O
head	O
hurts	O
?	O
</s>
<s>
During	O
this	O
time	O
,	O
the	O
first	O
chatterbots	B-Protocol
were	O
written	O
(	O
e.g.	O
,	O
PARRY	O
)	O
.	O
</s>
<s>
1980s	O
:	O
The	O
1980s	O
and	O
early	O
1990s	O
mark	O
the	O
heyday	O
of	O
symbolic	O
methods	O
in	O
NLP	B-Language
.	O
</s>
<s>
Focus	O
areas	O
of	O
the	O
time	O
included	O
research	O
on	O
rule-based	O
parsing	B-Language
(	O
e.g.	O
,	O
the	O
development	O
of	O
HPSG	O
as	O
a	O
computational	O
operationalization	O
of	O
generative	O
grammar	O
)	O
,	O
morphology	O
(	O
e.g.	O
,	O
two-level	O
morphology	O
)	O
,	O
semantics	O
(	O
e.g.	O
,	O
Lesk	B-General_Concept
algorithm	I-General_Concept
)	O
,	O
reference	O
(	O
e.g.	O
,	O
within	O
Centering	O
Theory	O
)	O
and	O
other	O
areas	O
of	O
natural	B-General_Concept
language	I-General_Concept
understanding	I-General_Concept
(	O
e.g.	O
,	O
in	O
the	O
Rhetorical	O
Structure	O
Theory	O
)	O
.	O
</s>
<s>
Other	O
lines	O
of	O
research	O
were	O
continued	O
,	O
e.g.	O
,	O
the	O
development	O
of	O
chatterbots	B-Protocol
with	O
Racter	B-Protocol
and	O
Jabberwacky	B-Protocol
.	O
</s>
<s>
Up	O
to	O
the	O
1980s	O
,	O
most	O
natural	B-Language
language	I-Language
processing	I-Language
systems	O
were	O
based	O
on	O
complex	O
sets	O
of	O
hand-written	O
rules	O
.	O
</s>
<s>
Starting	O
in	O
the	O
late	O
1980s	O
,	O
however	O
,	O
there	O
was	O
a	O
revolution	O
in	O
natural	B-Language
language	I-Language
processing	I-Language
with	O
the	O
introduction	O
of	O
machine	O
learning	O
algorithms	O
for	O
language	O
processing	O
.	O
</s>
<s>
1990s	O
:	O
Many	O
of	O
the	O
notable	O
early	O
successes	O
on	O
statistical	O
methods	O
in	O
NLP	B-Language
occurred	O
in	O
the	O
field	O
of	O
machine	B-Application
translation	I-Application
,	O
due	O
especially	O
to	O
work	O
at	O
IBM	O
Research	O
,	O
such	O
as	O
IBM	B-General_Concept
alignment	I-General_Concept
models	I-General_Concept
.	O
</s>
<s>
Research	O
has	O
thus	O
increasingly	O
focused	O
on	O
unsupervised	B-General_Concept
and	O
semi-supervised	B-General_Concept
learning	I-General_Concept
algorithms	O
.	O
</s>
<s>
Generally	O
,	O
this	O
task	O
is	O
much	O
more	O
difficult	O
than	O
supervised	B-General_Concept
learning	I-General_Concept
,	O
and	O
typically	O
produces	O
less	O
accurate	O
results	O
for	O
a	O
given	O
amount	O
of	O
input	O
data	O
.	O
</s>
<s>
In	O
the	O
2010s	O
,	O
representation	B-General_Concept
learning	I-General_Concept
and	O
deep	O
neural	O
network-style	O
machine	O
learning	O
methods	O
became	O
widespread	O
in	O
natural	B-Language
language	I-Language
processing	I-Language
.	O
</s>
<s>
That	O
popularity	O
was	O
due	O
partly	O
to	O
a	O
flurry	O
of	O
results	O
showing	O
that	O
such	O
techniques	O
can	O
achieve	O
state-of-the-art	O
results	O
in	O
many	O
natural	O
language	O
tasks	O
,	O
e.g.	O
,	O
in	O
language	B-Language
modeling	I-Language
and	O
parsing	B-Language
.	O
</s>
<s>
This	O
is	O
increasingly	O
important	O
in	B-Application
medicine	I-Application
and	I-Application
healthcare	I-Application
,	O
where	O
NLP	B-Language
helps	O
analyze	O
notes	O
and	O
text	O
in	O
electronic	B-Application
health	I-Application
records	I-Application
that	O
would	O
otherwise	O
be	O
inaccessible	O
for	O
study	O
when	O
seeking	O
to	O
improve	O
care	O
.	O
</s>
<s>
In	O
the	O
early	O
days	O
,	O
many	O
language-processing	O
systems	O
were	O
designed	O
by	O
symbolic	O
methods	O
,	O
i.e.	O
,	O
the	O
hand-coding	O
of	O
a	O
set	O
of	O
rules	O
,	O
coupled	O
with	O
a	O
dictionary	O
lookup	O
:	O
such	O
as	O
by	O
writing	O
grammars	O
or	O
devising	O
heuristic	O
rules	O
for	O
stemming	B-General_Concept
.	O
</s>
<s>
Despite	O
the	O
popularity	O
of	O
machine	O
learning	O
in	O
NLP	B-Language
research	O
,	O
symbolic	O
methods	O
are	O
still	O
(	O
2020	O
)	O
commonly	O
used	O
:	O
</s>
<s>
when	O
the	O
amount	O
of	O
training	O
data	O
is	O
insufficient	O
to	O
successfully	O
apply	O
machine	O
learning	O
methods	O
,	O
e.g.	O
,	O
for	O
the	O
machine	B-Application
translation	I-Application
of	O
low-resource	O
languages	O
such	O
as	O
provided	O
by	O
the	O
Apertium	B-Language
system	O
,	O
</s>
<s>
for	O
postprocessing	O
and	O
transforming	O
the	O
output	O
of	O
NLP	B-Language
pipelines	O
,	O
e.g.	O
,	O
for	O
knowledge	O
extraction	O
from	O
syntactic	O
parses	B-Language
.	O
</s>
<s>
Since	O
the	O
so-called	O
"	O
statistical	O
revolution	O
"	O
in	O
the	O
late	O
1980s	O
and	O
mid-1990s	O
,	O
much	O
natural	B-Language
language	I-Language
processing	I-Language
research	O
has	O
relied	O
heavily	O
on	O
machine	O
learning	O
.	O
</s>
<s>
Increasingly	O
,	O
however	O
,	O
research	O
has	O
focused	O
on	O
statistical	O
models	O
,	O
which	O
make	O
soft	O
,	O
probabilistic	O
decisions	O
based	O
on	O
attaching	O
real-valued	O
weights	O
to	O
each	O
input	O
feature	O
(	O
complex-valued	O
embeddings	B-General_Concept
,	O
and	O
neural	B-Architecture
networks	I-Architecture
in	O
general	O
have	O
also	O
been	O
proposed	O
,	O
for	O
e.g.	O
</s>
<s>
Some	O
of	O
the	O
earliest-used	O
machine	O
learning	O
algorithms	O
,	O
such	O
as	O
decision	B-Algorithm
trees	I-Algorithm
,	O
produced	O
systems	O
of	O
hard	O
if	O
–	O
then	O
rules	O
similar	O
to	O
existing	O
handwritten	O
rules	O
.	O
</s>
<s>
However	O
,	O
part-of-speech	O
tagging	O
introduced	O
the	O
use	O
of	O
hidden	O
Markov	O
models	O
to	O
natural	B-Language
language	I-Language
processing	I-Language
,	O
and	O
increasingly	O
,	O
research	O
has	O
focused	O
on	O
statistical	O
models	O
,	O
which	O
make	O
soft	O
,	O
probabilistic	O
decisions	O
based	O
on	O
attaching	O
real-valued	O
weights	O
to	O
the	O
features	O
making	O
up	O
the	O
input	O
data	O
.	O
</s>
<s>
The	O
cache	B-General_Concept
language	I-General_Concept
models	I-General_Concept
upon	O
which	O
many	O
speech	B-Application
recognition	I-Application
systems	O
now	O
rely	O
are	O
examples	O
of	O
such	O
statistical	O
models	O
.	O
</s>
<s>
Since	O
the	O
neural	O
turn	O
,	O
statistical	O
methods	O
in	O
NLP	B-Language
research	O
have	O
been	O
largely	O
replaced	O
by	O
neural	B-Architecture
networks	I-Architecture
.	O
</s>
<s>
Since	O
2015	O
,	O
the	O
field	O
has	O
thus	O
largely	O
abandoned	O
statistical	O
methods	O
and	O
shifted	O
to	O
neural	B-Architecture
networks	I-Architecture
for	O
machine	O
learning	O
.	O
</s>
<s>
Popular	O
techniques	O
include	O
the	O
use	O
of	O
word	B-General_Concept
embeddings	I-General_Concept
to	O
capture	O
semantic	O
properties	O
of	O
words	O
,	O
and	O
an	O
increase	O
in	O
end-to-end	O
learning	O
of	O
a	O
higher-level	O
task	O
(	O
e.g.	O
,	O
question	B-Algorithm
answering	I-Algorithm
)	O
instead	O
of	O
relying	O
on	O
a	O
pipeline	O
of	O
separate	O
intermediate	O
tasks	O
(	O
e.g.	O
,	O
part-of-speech	O
tagging	O
and	O
dependency	O
parsing	B-Language
)	O
.	O
</s>
<s>
In	O
some	O
areas	O
,	O
this	O
shift	O
has	O
entailed	O
substantial	O
changes	O
in	O
how	O
NLP	B-Language
systems	O
are	O
designed	O
,	O
such	O
that	O
deep	O
neural	O
network-based	O
approaches	O
may	O
be	O
viewed	O
as	O
a	O
new	O
paradigm	O
distinct	O
from	O
statistical	O
natural	B-Language
language	I-Language
processing	I-Language
.	O
</s>
<s>
For	O
instance	O
,	O
the	O
term	O
neural	B-General_Concept
machine	I-General_Concept
translation	I-General_Concept
(	O
NMT	O
)	O
emphasizes	O
the	O
fact	O
that	O
deep	O
learning-based	O
approaches	O
to	O
machine	B-Application
translation	I-Application
directly	O
learn	O
sequence-to-sequence	B-Algorithm
transformations	O
,	O
obviating	O
the	O
need	O
for	O
intermediate	O
steps	O
such	O
as	O
word	O
alignment	O
and	O
language	B-Language
modeling	I-Language
that	O
was	O
used	O
in	O
statistical	B-General_Concept
machine	I-General_Concept
translation	I-General_Concept
(	O
SMT	O
)	O
.	O
</s>
<s>
The	O
following	O
is	O
a	O
list	O
of	O
some	O
of	O
the	O
most	O
commonly	O
researched	O
tasks	O
in	O
natural	B-Language
language	I-Language
processing	I-Language
.	O
</s>
<s>
Though	O
natural	B-Language
language	I-Language
processing	I-Language
tasks	O
are	O
closely	O
intertwined	O
,	O
they	O
can	O
be	O
subdivided	O
into	O
categories	O
for	O
convenience	O
.	O
</s>
<s>
Speech	B-Application
recognition	I-Application
Given	O
a	O
sound	O
clip	O
of	O
a	O
person	O
or	O
people	O
speaking	O
,	O
determine	O
the	O
textual	O
representation	O
of	O
the	O
speech	O
.	O
</s>
<s>
This	O
is	O
the	O
opposite	O
of	O
text	B-Application
to	I-Application
speech	I-Application
and	O
is	O
one	O
of	O
the	O
extremely	O
difficult	O
problems	O
colloquially	O
termed	O
"	O
AI-complete	B-General_Concept
"	O
(	O
see	O
above	O
)	O
.	O
</s>
<s>
In	O
natural	O
speech	O
there	O
are	O
hardly	O
any	O
pauses	O
between	O
successive	O
words	O
,	O
and	O
thus	O
speech	B-General_Concept
segmentation	I-General_Concept
is	O
a	O
necessary	O
subtask	O
of	O
speech	B-Application
recognition	I-Application
(	O
see	O
below	O
)	O
.	O
</s>
<s>
Also	O
,	O
given	O
that	O
words	O
in	O
the	O
same	O
language	O
are	O
spoken	O
by	O
people	O
with	O
different	O
accents	O
,	O
the	O
speech	B-Application
recognition	I-Application
software	I-Application
must	O
be	O
able	O
to	O
recognize	O
the	O
wide	O
variety	O
of	O
input	O
as	O
being	O
identical	O
to	O
each	O
other	O
in	O
terms	O
of	O
its	O
textual	O
equivalent	O
.	O
</s>
<s>
Speech	B-General_Concept
segmentation	I-General_Concept
Given	O
a	O
sound	O
clip	O
of	O
a	O
person	O
or	O
people	O
speaking	O
,	O
separate	O
it	O
into	O
words	O
.	O
</s>
<s>
A	O
subtask	O
of	O
speech	B-Application
recognition	I-Application
and	O
typically	O
grouped	O
with	O
it	O
.	O
</s>
<s>
Text-to-speech	B-Application
can	O
be	O
used	O
to	O
aid	O
the	O
visually	O
impaired	O
.	O
</s>
<s>
Sometimes	O
this	O
process	O
is	O
also	O
used	O
in	O
cases	O
like	O
bag	B-General_Concept
of	I-General_Concept
words	I-General_Concept
(	O
BOW	O
)	O
creation	O
in	O
data	O
mining	O
.	O
</s>
<s>
Lemmatization	B-General_Concept
The	O
task	O
of	O
removing	O
inflectional	O
endings	O
only	O
and	O
to	O
return	O
the	O
base	O
dictionary	O
form	O
of	O
a	O
word	O
which	O
is	O
also	O
known	O
as	O
a	O
lemma	O
.	O
</s>
<s>
Lemmatization	B-General_Concept
is	O
another	O
technique	O
for	O
reducing	O
words	O
to	O
their	O
normalized	O
form	O
.	O
</s>
<s>
Stemming	B-General_Concept
yields	O
similar	O
results	O
as	O
lemmatization	B-General_Concept
,	O
but	O
does	O
so	O
on	O
grounds	O
of	O
rules	O
,	O
not	O
a	O
dictionary	O
.	O
</s>
<s>
Parsing	B-Language
Determine	O
the	O
parse	B-Language
tree	O
(	O
grammatical	O
analysis	O
)	O
of	O
a	O
given	O
sentence	O
.	O
</s>
<s>
The	O
grammar	O
for	O
natural	O
languages	O
is	O
ambiguous	O
and	O
typical	O
sentences	O
have	O
multiple	O
possible	O
analyses	O
:	O
perhaps	O
surprisingly	O
,	O
for	O
a	O
typical	O
sentence	O
there	O
may	O
be	O
thousands	O
of	O
potential	O
parses	B-Language
(	O
most	O
of	O
which	O
will	O
seem	O
completely	O
nonsensical	O
to	O
a	O
human	O
)	O
.	O
</s>
<s>
There	O
are	O
two	O
primary	O
types	O
of	O
parsing	B-Language
:	O
dependency	O
parsing	B-Language
and	O
constituency	O
parsing	B-Language
.	O
</s>
<s>
Dependency	O
parsing	B-Language
focuses	O
on	O
the	O
relationships	O
between	O
words	O
in	O
a	O
sentence	O
(	O
marking	O
things	O
like	O
primary	O
objects	O
and	O
predicates	O
)	O
,	O
whereas	O
constituency	O
parsing	B-Language
focuses	O
on	O
building	O
out	O
the	O
parse	B-Language
tree	O
using	O
a	O
probabilistic	B-General_Concept
context-free	I-General_Concept
grammar	I-General_Concept
(	O
PCFG	B-General_Concept
)	O
(	O
see	O
also	O
stochastic	O
grammar	O
)	O
.	O
</s>
<s>
Named	B-General_Concept
entity	I-General_Concept
recognition	I-General_Concept
(	O
NER	O
)	O
Given	O
a	O
stream	O
of	O
text	O
,	O
determine	O
which	O
items	O
in	O
the	O
text	O
map	O
to	O
proper	O
names	O
,	O
such	O
as	O
people	O
or	O
places	O
,	O
and	O
what	O
the	O
type	O
of	O
each	O
such	O
name	O
is	O
(	O
e.g.	O
</s>
<s>
Although	O
capitalization	O
can	O
aid	O
in	O
recognizing	O
named	B-General_Concept
entities	I-General_Concept
in	O
languages	O
such	O
as	O
English	O
,	O
this	O
information	O
cannot	O
aid	O
in	O
determining	O
the	O
type	O
of	O
named	B-General_Concept
entity	I-General_Concept
,	O
and	O
in	O
any	O
case	O
,	O
is	O
often	O
inaccurate	O
or	O
insufficient	O
.	O
</s>
<s>
For	O
example	O
,	O
the	O
first	O
letter	O
of	O
a	O
sentence	O
is	O
also	O
capitalized	O
,	O
and	O
named	B-General_Concept
entities	I-General_Concept
often	O
span	O
several	O
words	O
,	O
only	O
some	O
of	O
which	O
are	O
capitalized	O
.	O
</s>
<s>
Word-sense	B-General_Concept
disambiguation	I-General_Concept
(	O
WSD	O
)	O
Many	O
words	O
have	O
more	O
than	O
one	O
meaning	O
;	O
we	O
have	O
to	O
select	O
the	O
meaning	O
which	O
makes	O
the	O
most	O
sense	O
in	O
context	O
.	O
</s>
<s>
from	O
a	O
dictionary	O
or	O
an	O
online	O
resource	O
such	O
as	O
WordNet	B-General_Concept
.	O
</s>
<s>
Entity	B-General_Concept
linking	I-General_Concept
Many	O
words	O
—	O
typically	O
proper	O
names	O
—	O
refer	O
to	O
named	B-General_Concept
entities	I-General_Concept
;	O
here	O
we	O
have	O
to	O
select	O
the	O
entity	O
(	O
a	O
famous	O
individual	O
,	O
a	O
location	O
,	O
a	O
company	O
,	O
etc	O
.	O
)	O
</s>
<s>
Relationship	O
extraction	O
Given	O
a	O
chunk	O
of	O
text	O
,	O
identify	O
the	O
relationships	O
among	O
named	B-General_Concept
entities	I-General_Concept
(	O
e.g.	O
</s>
<s>
Semantic	B-Application
parsing	I-Application
Given	O
a	O
piece	O
of	O
text	O
(	O
typically	O
a	O
sentence	O
)	O
,	O
produce	O
a	O
formal	O
representation	O
of	O
its	O
semantics	O
,	O
either	O
as	O
a	O
graph	O
(	O
e.g.	O
,	O
in	O
AMR	O
parsing	B-Language
)	O
or	O
in	O
accordance	O
with	O
a	O
logical	O
formalism	O
(	O
e.g.	O
,	O
in	O
DRT	O
parsing	B-Language
)	O
.	O
</s>
<s>
This	O
challenge	O
typically	O
includes	O
aspects	O
of	O
several	O
more	O
elementary	O
NLP	B-Language
tasks	O
from	O
semantics	O
(	O
e.g.	O
,	O
semantic	O
role	O
labelling	O
,	O
word-sense	B-General_Concept
disambiguation	I-General_Concept
)	O
and	O
can	O
be	O
extended	O
to	O
include	O
full-fledged	O
discourse	O
analysis	O
(	O
e.g.	O
,	O
discourse	O
analysis	O
,	O
coreference	O
;	O
see	O
Natural	B-General_Concept
language	I-General_Concept
understanding	I-General_Concept
below	O
)	O
.	O
</s>
<s>
Given	O
a	O
single	O
sentence	O
,	O
identify	O
and	O
disambiguate	B-General_Concept
semantic	O
predicates	O
(	O
e.g.	O
,	O
verbal	O
frames	O
)	O
,	O
then	O
identify	O
and	O
classify	O
the	O
frame	O
elements	O
(	O
semantic	O
roles	O
)	O
.	O
</s>
<s>
One	O
task	O
is	O
discourse	O
parsing	B-Language
,	O
i.e.	O
,	O
identifying	O
the	O
discourse	O
structure	O
of	O
a	O
connected	O
text	O
,	O
i.e.	O
</s>
<s>
Given	O
a	O
single	O
sentence	O
,	O
identify	O
and	O
disambiguate	B-General_Concept
semantic	O
predicates	O
(	O
e.g.	O
,	O
verbal	O
frames	O
)	O
and	O
their	O
explicit	O
semantic	O
roles	O
in	O
the	O
current	O
sentence	O
(	O
see	O
Semantic	O
role	O
labelling	O
above	O
)	O
.	O
</s>
<s>
The	O
goal	O
of	O
argument	B-Algorithm
mining	I-Algorithm
is	O
the	O
automatic	O
extraction	O
and	O
identification	O
of	O
argumentative	O
structures	O
from	O
natural	O
language	O
text	O
with	O
the	O
aid	O
of	O
computer	O
programs	O
.	O
</s>
<s>
Automatic	B-Application
summarization	I-Application
(	O
text	B-Application
summarization	I-Application
)	O
Produce	O
a	O
readable	O
summary	O
of	O
a	O
chunk	O
of	O
text	O
.	O
</s>
<s>
As	O
far	O
as	O
orthography	O
,	O
morphology	O
,	O
syntax	O
and	O
certain	O
aspects	O
of	O
semantics	O
are	O
concerned	O
,	O
and	O
due	O
to	O
the	O
development	O
of	O
powerful	O
neural	O
language	B-Language
models	I-Language
such	O
as	O
GPT-2	B-General_Concept
,	O
this	O
can	O
now	O
(	O
2019	O
)	O
be	O
considered	O
a	O
largely	O
solved	O
problem	O
and	O
is	O
being	O
marketed	O
in	O
various	O
commercial	O
applications	O
.	O
</s>
<s>
This	O
is	O
one	O
of	O
the	O
most	O
difficult	O
problems	O
,	O
and	O
is	O
a	O
member	O
of	O
a	O
class	O
of	O
problems	O
colloquially	O
termed	O
"	O
AI-complete	B-General_Concept
"	O
,	O
i.e.	O
</s>
<s>
Natural-language	B-General_Concept
understanding	I-General_Concept
(	O
NLU	O
)	O
Convert	O
chunks	O
of	O
text	O
into	O
more	O
formal	O
representations	O
such	O
as	O
first-order	O
logic	O
structures	O
that	O
are	O
easier	O
for	O
computer	O
programs	O
to	O
manipulate	O
.	O
</s>
<s>
Natural	B-General_Concept
language	I-General_Concept
understanding	I-General_Concept
involves	O
the	O
identification	O
of	O
the	O
intended	O
semantic	O
from	O
the	O
multiple	O
possible	O
semantics	O
which	O
can	O
be	O
derived	O
from	O
a	O
natural	O
language	O
expression	O
which	O
usually	O
takes	O
the	O
form	O
of	O
organized	O
notations	O
of	O
natural	O
language	O
concepts	O
.	O
</s>
<s>
An	O
explicit	O
formalization	O
of	O
natural	O
language	O
semantics	O
without	O
confusions	O
with	O
implicit	O
assumptions	O
such	O
as	O
closed-world	B-Application
assumption	I-Application
(	O
CWA	O
)	O
vs.	O
open-world	B-Application
assumption	I-Application
,	O
or	O
subjective	O
Yes/No	O
vs.	O
objective	O
True/False	O
is	O
expected	O
for	O
the	O
construction	O
of	O
a	O
basis	O
of	O
semantics	O
formalization	O
.	O
</s>
<s>
Natural-language	B-General_Concept
generation	I-General_Concept
(	O
NLG	O
)	O
:	O
</s>
<s>
Not	O
an	O
NLP	B-Language
task	O
proper	O
but	O
an	O
extension	O
of	O
natural	B-General_Concept
language	I-General_Concept
generation	I-General_Concept
and	O
other	O
NLP	B-Language
tasks	O
is	O
the	O
creation	O
of	O
full-fledged	O
books	O
.	O
</s>
<s>
The	O
first	O
machine-generated	O
book	O
was	O
created	O
by	O
a	O
rule-based	O
system	O
in	O
1984	O
(	O
Racter	B-Protocol
,	O
The	O
policeman	O
's	O
beard	O
is	O
half-constructed	O
)	O
.	O
</s>
<s>
The	O
first	O
published	O
work	O
by	O
a	O
neural	B-Architecture
network	I-Architecture
was	O
published	O
in	O
2018	O
,	O
1	O
the	O
Road	O
,	O
marketed	O
as	O
a	O
novel	O
,	O
contains	O
sixty	O
million	O
words	O
.	O
</s>
<s>
Both	O
these	O
systems	O
are	O
basically	O
elaborate	O
but	O
non-sensical	O
(	O
semantics-free	O
)	O
language	B-Language
models	I-Language
.	O
</s>
<s>
Unlike	O
Racter	B-Protocol
and	O
1	O
the	O
Road	O
,	O
this	O
is	O
grounded	O
on	O
factual	O
knowledge	O
and	O
based	O
on	O
text	B-Application
summarization	I-Application
.	O
</s>
<s>
A	O
Document	B-Application
AI	I-Application
platform	O
sits	O
on	O
top	O
of	O
the	O
NLP	B-Language
technology	O
enabling	O
users	O
with	O
no	O
prior	O
experience	O
of	O
artificial	B-Application
intelligence	I-Application
,	O
machine	O
learning	O
or	O
NLP	B-Language
to	O
quickly	O
train	O
a	O
computer	O
to	O
extract	O
the	O
specific	O
data	O
they	O
need	O
from	O
different	O
document	O
types	O
.	O
</s>
<s>
NLP-powered	O
Document	B-Application
AI	I-Application
enables	O
non-technical	O
teams	O
to	O
quickly	O
access	O
information	O
hidden	O
in	O
documents	O
,	O
for	O
example	O
,	O
lawyers	O
,	O
business	O
analysts	O
and	O
accountants	O
.	O
</s>
<s>
Question	B-Algorithm
answering	I-Algorithm
Given	O
a	O
human-language	O
question	O
,	O
determine	O
its	O
answer	O
.	O
</s>
<s>
Text-to-image	B-General_Concept
generation	I-General_Concept
Given	O
a	O
description	O
of	O
an	O
image	O
,	O
generate	O
an	O
image	O
that	O
matches	O
the	O
description	O
.	O
</s>
<s>
Text-to-video	B-General_Concept
Given	O
a	O
description	O
of	O
a	O
video	O
,	O
generate	O
a	O
video	O
that	O
matches	O
the	O
description	O
.	O
</s>
<s>
Based	O
on	O
long-standing	O
trends	O
in	O
the	O
field	O
,	O
it	O
is	O
possible	O
to	O
extrapolate	O
future	O
directions	O
of	O
NLP	B-Language
.	O
</s>
<s>
Interest	O
on	O
increasingly	O
abstract	O
,	O
"	O
cognitive	O
"	O
aspects	O
of	O
natural	O
language	O
(	O
1999	O
–	O
2001	O
:	O
shallow	O
parsing	B-Language
,	O
2002	O
–	O
03	O
:	O
named	B-General_Concept
entity	I-General_Concept
recognition	I-General_Concept
,	O
2006	O
–	O
09/2017	O
–	O
18	O
:	O
dependency	O
syntax	O
,	O
2004	O
–	O
05/2008	O
–	O
09	O
semantic	O
role	O
labelling	O
,	O
2011	O
–	O
12	O
coreference	O
,	O
2015	O
–	O
16	O
:	O
discourse	O
parsing	B-Language
,	O
2019	O
:	O
semantic	B-Application
parsing	I-Application
)	O
.	O
</s>
<s>
Most	O
higher-level	O
NLP	B-Language
applications	O
involve	O
aspects	O
that	O
emulate	O
intelligent	O
behaviour	O
and	O
apparent	O
comprehension	O
of	O
natural	O
language	O
.	O
</s>
<s>
More	O
broadly	O
speaking	O
,	O
the	O
technical	O
operationalization	O
of	O
increasingly	O
advanced	O
aspects	O
of	O
cognitive	O
behaviour	O
represents	O
one	O
of	O
the	O
developmental	O
trajectories	O
of	O
NLP	B-Language
(	O
see	O
trends	O
among	O
CoNLL	O
shared	O
tasks	O
above	O
)	O
.	O
</s>
<s>
Especially	O
during	O
the	O
age	O
of	O
symbolic	O
NLP	B-Language
,	O
the	O
area	O
of	O
computational	O
linguistics	O
maintained	O
strong	O
ties	O
with	O
cognitive	O
studies	O
.	O
</s>
<s>
As	O
an	O
example	O
,	O
George	O
Lakoff	O
offers	O
a	O
methodology	O
to	O
build	O
natural	B-Language
language	I-Language
processing	I-Language
(	O
NLP	B-Language
)	O
algorithms	O
through	O
the	O
perspective	O
of	O
cognitive	O
science	O
,	O
along	O
with	O
the	O
findings	O
of	O
cognitive	O
linguistics	O
,	O
with	O
two	O
defining	O
aspects	O
:	O
</s>
<s>
The	O
intent	O
behind	O
other	O
usages	O
,	O
like	O
in	O
"	O
She	O
is	O
a	O
big	O
person	O
"	O
,	O
will	O
remain	O
somewhat	O
ambiguous	O
to	O
a	O
person	O
and	O
a	O
cognitive	O
NLP	B-Language
algorithm	O
alike	O
without	O
additional	O
information	O
.	O
</s>
<s>
Assign	O
relative	O
measures	O
of	O
meaning	O
to	O
a	O
word	O
,	O
phrase	O
,	O
sentence	O
or	O
piece	O
of	O
text	O
based	O
on	O
the	O
information	O
presented	O
before	O
and	O
after	O
the	O
piece	O
of	O
text	O
being	O
analyzed	O
,	O
e.g.	O
,	O
by	O
means	O
of	O
a	O
probabilistic	B-General_Concept
context-free	I-General_Concept
grammar	I-General_Concept
(	O
PCFG	B-General_Concept
)	O
.	O
</s>
<s>
Ties	O
with	O
cognitive	O
linguistics	O
are	O
part	O
of	O
the	O
historical	O
heritage	O
of	O
NLP	B-Language
,	O
but	O
they	O
have	O
been	O
less	O
frequently	O
addressed	O
since	O
the	O
statistical	O
turn	O
during	O
the	O
1990s	O
.	O
</s>
<s>
Nevertheless	O
,	O
approaches	O
to	O
develop	O
cognitive	O
models	O
towards	O
technically	O
operationalizable	O
frameworks	O
have	O
been	O
pursued	O
in	O
the	O
context	O
of	O
various	O
frameworks	O
,	O
e.g.	O
,	O
of	O
cognitive	O
grammar	O
,	O
functional	O
grammar	O
,	O
construction	O
grammar	O
,	O
computational	O
psycholinguistics	O
and	O
cognitive	O
neuroscience	O
(	O
e.g.	O
,	O
ACT-R	B-Language
)	O
,	O
however	O
,	O
with	O
limited	O
uptake	O
in	O
mainstream	O
NLP	B-Language
(	O
as	O
measured	O
by	O
presence	O
on	O
major	O
conferences	O
of	O
the	O
ACL	O
)	O
.	O
</s>
<s>
More	O
recently	O
,	O
ideas	O
of	O
cognitive	O
NLP	B-Language
have	O
been	O
revived	O
as	O
an	O
approach	O
to	O
achieve	O
explainability	O
,	O
e.g.	O
,	O
under	O
the	O
notion	O
of	O
"	O
cognitive	O
AI	B-Application
"	O
.	O
</s>
<s>
Likewise	O
,	O
ideas	O
of	O
cognitive	O
NLP	B-Language
are	O
inherent	O
to	O
neural	O
models	O
multimodal	B-Application
NLP	B-Language
(	O
although	O
rarely	O
made	O
explicit	O
)	O
.	O
</s>
