<s>
Semantic	B-Application
parsing	I-Application
is	O
the	O
task	O
of	O
converting	O
a	O
natural	O
language	O
utterance	O
to	O
a	O
logical	O
form	O
:	O
a	O
machine-understandable	O
representation	O
of	O
its	O
meaning	O
.	O
</s>
<s>
Semantic	B-Application
parsing	I-Application
can	O
thus	O
be	O
understood	O
as	O
extracting	O
the	O
precise	O
meaning	O
of	O
an	O
utterance	O
.	O
</s>
<s>
Applications	O
of	O
semantic	B-Application
parsing	I-Application
include	O
machine	B-Application
translation	I-Application
,	O
question	B-Algorithm
answering	I-Algorithm
,	O
ontology	B-General_Concept
induction	I-General_Concept
,	O
automated	O
reasoning	O
,	O
and	O
code	B-Application
generation	I-Application
.	O
</s>
<s>
The	O
phrase	O
was	O
first	O
used	O
in	O
the	O
1970s	O
by	O
Yorick	O
Wilks	O
as	O
the	O
basis	O
for	O
machine	B-Application
translation	I-Application
programs	O
working	O
with	O
only	O
semantic	O
representations	O
.	O
</s>
<s>
In	O
computer	B-Application
vision	I-Application
,	O
semantic	B-Application
parsing	I-Application
is	O
a	O
process	O
of	O
segmentation	B-Algorithm
for	O
3D	O
objects	O
.	O
</s>
<s>
Shallow	O
semantic	B-Application
parsing	I-Application
is	O
concerned	O
with	O
identifying	O
entities	O
in	O
an	O
utterance	O
and	O
labelling	O
them	O
with	O
the	O
roles	O
they	O
play	O
.	O
</s>
<s>
Shallow	O
semantic	B-Application
parsing	I-Application
is	O
sometimes	O
known	O
as	O
slot-filling	O
or	O
frame	O
semantic	B-Application
parsing	I-Application
,	O
since	O
its	O
theoretical	O
basis	O
comes	O
from	O
frame	O
semantics	O
,	O
wherein	O
a	O
word	O
evokes	O
a	O
frame	O
of	O
related	O
concepts	O
and	O
roles	O
.	O
</s>
<s>
Slot-filling	O
systems	O
are	O
widely	O
used	O
in	O
virtual	B-Protocol
assistants	I-Protocol
in	O
conjunction	O
with	O
intent	O
classifiers	O
,	O
which	O
can	O
be	O
seen	O
as	O
mechanisms	O
for	O
identifying	O
the	O
frame	O
evoked	O
by	O
an	O
utterance	O
.	O
</s>
<s>
Popular	O
architectures	O
for	O
slot-filling	O
are	O
largely	O
variants	O
of	O
an	O
encoder-decoder	O
model	O
,	O
wherein	O
two	O
recurrent	B-Algorithm
neural	I-Algorithm
networks	I-Algorithm
(	O
RNNs	O
)	O
are	O
trained	O
jointly	O
to	O
encode	O
an	O
utterance	O
into	O
a	O
vector	O
and	O
to	O
decode	O
that	O
vector	O
into	O
a	O
sequence	O
of	O
slot	O
labels	O
.	O
</s>
<s>
This	O
type	O
of	O
model	O
is	O
used	O
in	O
the	O
Amazon	B-Application
Alexa	I-Application
spoken	O
language	O
understanding	O
system	O
.	O
</s>
<s>
Deep	O
semantic	B-Application
parsing	I-Application
,	O
also	O
known	O
as	O
compositional	O
semantic	B-Application
parsing	I-Application
,	O
is	O
concerned	O
with	O
producing	O
precise	O
meaning	O
representations	O
of	O
utterances	O
that	O
can	O
contain	O
significant	O
compositionality	O
.	O
</s>
<s>
Shallow	O
semantic	B-Application
parsers	I-Application
can	O
parse	O
utterances	O
like	O
"	O
show	O
me	O
flights	O
from	O
Boston	O
to	O
Dallas	O
"	O
by	O
classifying	O
the	O
intent	O
as	O
"	O
list	O
flights	O
"	O
,	O
and	O
filling	O
slots	O
"	O
source	O
"	O
and	O
"	O
destination	O
"	O
with	O
"	O
Boston	O
"	O
and	O
"	O
Dallas	O
"	O
,	O
respectively	O
.	O
</s>
<s>
However	O
,	O
shallow	O
semantic	B-Application
parsing	I-Application
cannot	O
parse	O
arbitrary	O
compositional	O
utterances	O
,	O
like	O
"	O
show	O
me	O
flights	O
from	O
Boston	O
to	O
anywhere	O
that	O
has	O
flights	O
to	O
Juneau	O
"	O
.	O
</s>
<s>
Deep	O
semantic	B-Application
parsing	I-Application
attempts	O
to	O
parse	O
such	O
utterances	O
,	O
typically	O
by	O
converting	O
them	O
to	O
a	O
formal	O
meaning	O
representation	O
language	O
.	O
</s>
<s>
Early	O
semantic	B-Application
parsers	I-Application
used	O
highly	O
domain-specific	O
meaning	O
representation	O
languages	O
,	O
with	O
later	O
systems	O
using	O
more	O
extensible	O
languages	O
like	O
Prolog	B-Language
,	O
lambda	B-Language
calculus	I-Language
,	O
lambda	O
dependency-based	O
compositional	O
semantics	O
(λ-	O
DCS	O
)	O
,	O
SQL	B-Language
,	O
Python	B-Language
,	O
Java	B-Language
,	O
the	O
Alexa	B-Application
Meaning	O
Representation	O
Language	O
,	O
and	O
the	O
Abstract	O
Meaning	O
Representation	O
(	O
AMR	O
)	O
.	O
</s>
<s>
Some	O
work	O
has	O
used	O
more	O
exotic	O
meaning	O
representations	O
,	O
like	O
query	B-Library
graphs	O
,	O
semantic	O
graphs	O
,	O
or	O
vector	O
representations	O
.	O
</s>
<s>
Most	O
modern	O
deep	O
semantic	B-Application
parsing	I-Application
models	O
are	O
either	O
based	O
on	O
defining	O
a	O
formal	O
grammar	O
for	O
a	O
chart	B-Application
parser	I-Application
or	O
utilizing	O
RNNs	O
to	O
directly	O
translate	O
from	O
a	O
natural	O
language	O
to	O
a	O
meaning	O
representation	O
language	O
.	O
</s>
<s>
Examples	O
of	O
systems	O
built	O
on	O
formal	O
grammars	O
are	O
the	O
Cornell	O
Semantic	B-Application
Parsing	I-Application
Framework	O
,	O
Stanford	O
University	O
's	O
Semantic	B-Application
Parsing	I-Application
with	O
Execution	O
(	O
SEMPRE	O
)	O
,	O
and	O
the	O
Word	O
Alignment-based	O
Semantic	B-Application
Parser	I-Application
(	O
WASP	O
)	O
.	O
</s>
<s>
Datasets	O
used	O
for	O
training	O
statistical	B-Application
semantic	I-Application
parsing	I-Application
models	O
are	O
divided	O
into	O
two	O
main	O
classes	O
based	O
on	O
application	O
:	O
those	O
used	O
for	O
question	B-Algorithm
answering	I-Algorithm
via	O
knowledge	O
base	O
queries	B-Library
,	O
and	O
those	O
used	O
for	O
code	B-Application
generation	I-Application
.	O
</s>
<s>
A	O
standard	O
dataset	O
for	O
question	B-Algorithm
answering	I-Algorithm
via	O
semantic	B-Application
parsing	I-Application
is	O
the	O
Air	O
Travel	O
Information	O
System	O
(	O
ATIS	O
)	O
dataset	O
,	O
which	O
contains	O
questions	O
and	O
commands	O
about	O
upcoming	O
flights	O
as	O
well	O
as	O
corresponding	O
SQL	B-Language
.	O
</s>
<s>
Another	O
benchmark	O
dataset	O
is	O
the	O
GeoQuery	O
dataset	O
which	O
contains	O
questions	O
about	O
the	O
geography	O
of	O
the	O
U.S.	O
paired	O
with	O
corresponding	O
Prolog	B-Language
.	O
</s>
<s>
The	O
Overnight	O
dataset	O
is	O
used	O
to	O
test	O
how	O
well	O
semantic	B-Application
parsers	I-Application
adapt	O
across	O
multiple	O
domains	O
;	O
it	O
contains	O
natural	O
language	O
queries	B-Library
about	O
8	O
different	O
domains	O
paired	O
with	O
corresponding	O
λ-DCS	O
expressions	O
.	O
</s>
<s>
Popular	O
datasets	O
for	O
code	B-Application
generation	I-Application
include	O
two	O
trading	O
card	O
datasets	O
that	O
link	O
the	O
text	O
that	O
appears	O
on	O
cards	O
to	O
code	O
that	O
precisely	O
represents	O
those	O
cards	O
.	O
</s>
<s>
One	O
was	O
constructed	O
linking	O
Magic	B-Application
:	I-Application
The	I-Application
Gathering	I-Application
card	O
texts	O
to	O
Java	B-Language
snippets	O
;	O
the	O
other	O
by	O
linking	O
Hearthstone	B-Application
card	O
texts	O
to	O
Python	B-Language
snippets	O
.	O
</s>
<s>
The	O
IFTTT	B-Application
dataset	O
uses	O
a	O
specialized	O
domain-specific	O
language	O
with	O
short	O
conditional	O
commands	O
.	O
</s>
<s>
The	O
Django	B-Language
dataset	O
pairs	O
Python	B-Language
snippets	O
with	O
English	O
and	O
Japanese	O
pseudocode	O
describing	O
them	O
.	O
</s>
