so you know everyone my name is german i am up used to in a
nice as a fine so today i would like to present my i will we
tried to new ways neck to neck which and ways in dialogue using an encoder
decoder with a semantic relation
so is my present a little bit on the technical know report still based on
a don't know four point to sleep okay that's a
so i my representation i personally i and is to use some of brief introduction
about the not fast so followed by a no general a model for or for
the re recording you are then we see that then write the and
with we of these and my a mean look my main contribution of in which
we is to use a new architecture we called it an ankle do calculate the
decoder
from the use of forms you agree that so we you're going to prison they
have not to kind of a new class and the first part is lively and
the second pass and we find a and a lastly we
only
give some experimental setup and results can lose it
so let's start with the introduction of nlg task
so is the n is that many
i just e
okay i don't convert the a meaning representation to an actual
and english on a sentence
for example we a given but i don't which is a combination of bow
the tax and i've we have since i've here for example the inform and the
list of slot value pairs for example we have to do a to a to
slot value pairs here for someone to new with the value of the hybrid the
second one is a poor with c at a value my script
so well the generators should and generates the not a sentence to test for example
we have a previous is basque restaurant all the second one we had not then
with the pirates of last for so what
that is to a brief introduction of and as the last
so one
i believe but you one
the
new approach based on the
on and on a
neural architecture so
follow from the button to the top given the dialogue act as a pair of
dialogue act and system than a to be learned so what the course of a
natural language generator is at the lexical i and sentences with a list of example
we have three slot names serves as a lot for
so for the known choice of open of the acoustic so we can based on
the man and
i am model lstms you bought some kind of encoder-decoder model so that powerful of
c is then we don't use "'em"
on also a sentence and after that we have a descendant can be lexical like
to perform to form the required sentence
so
here i and you give museum this general on a general model for the on
the record in a new language generator so
is an overview of the i and bodies newly a new maybe that generate the
which can be divided into two pass the pos on this and go to encode
all select and realise how much information it was information
the i z the core to the upper side is you could do it uses
use the we use the as you can and base model language model
so well here is our how many than on a minute contribution in this quilt
in ways we a propose a new model colson it and echoed actually very to
decode the
so
yep
here's a whole model
the
and go to know with a new architecture can be divided into a three pass
three components so first only and corner to end goal are all you compress the
target mini dropped and that a representation
this second one is the and then you a new proposed on a component we
call to a greater to a lie control the semantic and
to refine easier the input sequence
and
and the c d is you have said one is the decoder used and i
and you could or would you see a reply sentences
so let's move to a further up to the but so in the decoder side
we use the be directly know the are you with and course the separated a
parameterization of slots and values
and
where e the reader consists of two last
and a lilac to but and the what was it i don't run light on
a representation and the re final to cancun calculated see a new input token able
to do the decoder
all user is you
so let's move to further into the at a later in you know how a
model so the a photo one z the lighter calculate c
dynamite through a representation with cca a concatenation of slot of accent i
and z
and as easy the pants and you can isn't based on the absolute value representations
and here the refinancing second pass and we finally
we'll
calculators z new we would x the and put it then i into the
and the are using our
for the language of then there is an of the sentences
and
in how what we
we further apply the not i don't have a representation to the that you put
into the other u c l so firstly the it is you we set and
a big it can be a normally five to one
to use a on the on
as the not i don't know a representation
and
a sickly z can you did a activation is also modified to depose the influenced
by the as you like about a representation
so let's move to the we find a so in a in how well a
refined as well and examine the choices for the refined of for example we can
now we can use the cans and based on see a getting algorithms to apply
to the refined a so
actually the refine it is highly refined have work week or refinement from seven
of cr
i don't a web sense it's and dg and the origin to a token that
but
so for this we use the tense and look as a here is a higher
attendant algorithm and is the second one is a getting because the is that the
dense and get is them with less and has advanced and apply for the refined
of
of course the a lighter used so it how what we used a
the first attention because it
and for the kids and look as and we just apply there's assume a simple
one m is wise at least inanimate a multiplication for getting the reply to
so that more to further into know how can a
how can contrast the not depends in we get them
so firstly we just use a simple back to you attend to wait see a
see that i
and
we further we can we'll but with both files that you may be lies in
metrics can be back to do
another two where the
to get information and
lastly to not in order to be
of course the no
put a further than in the context information we propose here to what we can
see a pretty as a here
a recent i
active have a previous know if you story of the to the two of the
not dance and
so here you we use them getting the guys in which is use the two
guy a simple and simple way to study the multiplication and addition
so left to the experimental setup
is a well
we know only
we can that the use we only on the
under the dataset for model
which as the rest of one hotel a laptop and t v
we implemented by not using the tensor for all
and all of the generators what chain a we see back propagation through time
or
a stochastic gradient ascent with early stopping we had a l two regularization up to
four forty five ginny a examples and the hidden sty dataset is that the c
and
we said to keep drawing dropped lower rate seventy no for the initialize what is
right and position we use a group of a glue
and
for the evaluation of we use a blues and
slot error rate no discourse
do you evaluate
i will work
so
here use our result we compare our work with it was intended to form the
no
which we represent what here and we politically get a got and own something a
nice result here we go back to the out the whole models outperform the previous
one
and
use we propose a and because maybe a how work cannot and can varies based
on them together and can be varies by on the etsi so what
we have is we take i comments and that takes place of five are randomly
to supply and that's well
so here is the result so use a few go three yes
we can see that actually a whole just performs a real someone so in is
in the is a peak a three which is a on that so it's beyond
by
increase mantle step on the other the
the a put percent this of training data from what can withstand the training data
to the one hundred percent
so no the not enough figure four we just a and conduct the a general
models in which we pools of we most only the owns it for five i'm
not do means and the arch in which a gender issue proposed model
now
yes on individual domain in here we that's only if it or restore and hotel
laptop t v
so here is very a little bit nice no is that so is the dense
and on the behavior of three models from this we can now sees at how
the pole model with the context with second day
can
can
can not as intense in can
as a model can at hand and in the am going consecutive of acoustic the
tokens
so what symbol we can
a list of spoken here
the phrase okay
so here is no with that of the top generated a post from
form
and no on a
compare the from our model with the previous the one syllable
so that company would we just and presented no our new model coder and go
to a greater decoder in which we z is the with the and can see
it up to new but the first one politely used and attention over the input
meaning representation
the second part is the refined a with the danced and all getting a mechanism
to revise the input tokens and on which a model and we can generate or
do you see up also model and then we use the evaluation metric a bus
and
thus
score and a slot array as to what you various how well and take us
to send
thank you we have again
six minutes questions
i think you're much of the joke
can you please maybe i just didn't really
see something about the size of your training data and the number of difference
to predict looks like compare or liam
how much as
in addition
how many
you see a
and you replaced all so question
it is on a system
about
initial dataset size
how much
the image slice
which instance
again
actually i'm sorry the us government and
about now five thousand a sentence is the hotel almost seven thousand a lot of
and tb is must big with all who in a
a route
a thirteen a thirteen thousand synthesis
that's this aside dataset
okay i think the size
number of the predicates and model
things like compare
conditional move the
well slowed the legislate
one before lost
without
last slide
example low
okay sorry
so you see the predicted compare
and all of the first right
compare we aim
yes
how many things like this predicate are in the mobile
alright sites
i'm sorry for a not for this for the laptop sale mean
well on
and
and i don't a here is a appear only one time in the dataset
for example we can have a man to compare name
with the and but we have the same i think i but the is a
list of slot value pair is different for example we had in this we have
a companion screen size residues and in as a one week have
maybe we have compared name is i knew squeezed i also
so
and not that i don't i hear
appear in this latest and laptop and t v just one time so is that
stuff up for is such a nice the a model but i have to learn
the new one
the new the new and
i have to learn how to how to
applied to the new lose sequence of
a slot value pair
okay we assume
if the input dialogue act with just cry
those to a entities what would be the difference in the output
or actually in our work here we also a follows that no
in the
we also follows that uphold okay and up would use the is then routed out
as a sentence we the
firstly i complete which we can generate owns the of these slot requires a lot
for you pass and a six second one also as the is the list of
syllable one so well in the country all correct although for example you know
in is this one
by sampling is one where we have like in input dialogue here
so up the incorrect output can be this just output can upload use all the
information from the slot value pairs but
we can see that the l a own night and the l seventies
into in-correct so
but not
as it problem but you know in our model we can now and generates the
do you at the correct all of the week lies a reply
plentiful of the that say for example in that example you just had
what if the screen sizes range with the same value for both entities
so they both have a large would you then leave large out because it's not
it's not different so there's no comparison
actually in how well the g different value of screen
is not is not problem it but
also we came back here
the same the same value is not know
about
the value of slot followed by us a lot and of a span of
so the value of slot is not know
a
is not important in it is cool because
in here example
we have at least of celebrity by so we
not really skyline the and as the
i understand that you delexicalise but a human we do something different if the slot
values of the same versus a slot values are different
so it doesn't make sense if you comparing two things and they both have the
same slot value
that is it makes sense to say
just say this one's large and batman's large
instead of say for example they both are large or not mention that slot at
all because that's the same value for both one
how well actually is exactly z is the a post processing when
which is
and
system is an for the first he
just a the post processing after we have a and i was no candidates the
that is collect a sentence and we lexical i descendant to inform the oars in
the one
so that's why insist that
so thank you very
i presenter again