so my name is a given to be some degree c and i'm currently a
postdoctoral researcher
and i'm going to present this work with a great level and phonetic nonbackchannel
and
first let me see if you were the buttons and the context of this work
so this work is part of the european project
i have a spell
which aims that the designing artificial which we get of information and assistance
and this assistant a on the form of the actual agents
but are it can that ever to engage in a pretty model interaction
involving verbal and nonverbal behavior
there's agents also aim at its adapting to the user
and adapting to for instance in expected situations such as interaction
as well as to this to show emotional state of the u
and
in these projects and that's to that interested in a convergence and that better alignment
as shown by the communication accommodation sorry
can value convergence of behaviour is a very important features of you menu many interaction
that occurs both at low level such as pos true accent speech right and that
high-level such as of the mental emotional and cognitive label
and in particular
human
human the participant
align the mb at all at many linguistic level such as the lexical syntactic and
semantic ones
and one consequence of successful alignments in dialogue i is a set and a repetitive
nice
and
as a consequence there are there is going to be a
some of dialog regions that are going to imagine between the dialogue participant
under the form of lexical items for instance
so on the slide you can see two example of a dialog which represent the
same face aging introduction every face of a negotiation
and in this
in this examples
the dialogue buttons
a core roles and their buttons are the main focus of this work
so on the left you can see that they are very few buttons
in this case we says that the available alignment is very low on the contrary
on the right example you can see that
that a participant's aligned us to may need that of routines
such as nice to meet you how are you good
in this case we are going to say that the better a alignment is higher
so the main focus on this work is to propose measures of the of alignment
based on this data which
so what you think about alignment for human machine interaction so first
we can see from human interaction and that's this is a subconscious phenomenon that naturally
appears and it has been shown by previous work
that speakers we use lexical as well as syntactic structures from previous utterances
on top of that
double and temporal alignment may facilitate successful taskoriented the conversations
however in human machine interaction
it has been shown that linguistic alignment cultures
and in particular are users at the lexical items and syntactic structures from the system
but this is only one way
in most of the system the user aligned with the system is not able to
like
so in this work all goal is to provide a virtual agent with the ability
to detect the alignment behavior of its human participant of each from an interlocutor
and to align or not depending on the strategy with the user
so them in which iteration
of using the about alignment for an agent
is set provide a natural source of evaluation in dialogue and in particular for the
natural language generation that
it also makes it possible to take into account the social emotional behavior of the
behaviour and works
as a social blue
and
it's also way of adapting results the need of an extensive user profile
and what we expect from
providing an agent with the ability of the body a line is to and this
agents ability likability and friendliness to improve
interaction naturalness as wavelet to maintain and for still user engagement
finally we aim at improving collaboration in taskoriented that
so
in this work or approach is to provide the majors a characterizing babble alignment
that are going to be based on the transcript on dialogue and on the shared
expression at the lexical
and a proposition stands on
i was stream in past
the first one is to extract
the dialogue routines other justices the shared expression from the dialogue transcripts
the second part is to be an expression lexicon from this shared expression a as
that's keep track of the expression and some features of these expressions
and then they're deriving measures of that better alignment from the data transcript and the
expression icsi
let me so if you word about the automatic building at the expression a lexicon
so in this work we provide a model where we define
a surface text but then at the utterance a shine expression as a surface text
but then at the utterance level that has been produced by both speakers in dialogue
so for instance you can see
i and example of dialogue
on the left of the slide that in the middle
where there is are shown expressions that's not gonna work for me
which is used to reject a proposition in a negotiation dialogue is that is used
by the interlocutor at
in it in that first term and by interlocutor b in the first
don't
so is a shared expression is part of the expression lexicon
and has been initiated by eight
and so in this paper we present a framework of expressions that maybe and but
the or not
and we also provide
way of automatically extracting is it their shared expression to be done expression next we
can automatically
so this is an instance of sequential best down mining in
and it involves the use of by you informatics algorithms that are usually used to
my in dna sequences
so in short
it is involve zeros are the reserving of the multiple common subsequence problems for the
generalize to fix tree data structure
and through this
base of sequential pattern mining we can be from the transcript of dialogue d v
a dialog lexical
then from the data transcript and the expression lexicon we derive some aspects for one
measures
to characterize verbal alignment
so the first measures a global on the single dialog
and now the expression lexicon size that is this is a number of a unique
shown expression other to establish between dialogue participant
and the expression by a variety which is the expression lexicon size a normalized by
the length of the not a given but as a number of to the total
number of token in the day
we also derive
measure that a specific to the speakers
first the expressed in the expression repetition measure
which
measure which gives the amount of token that is dedicated
to the repetition of an expression by the user
over the total amount of token
and the initiated the expression racial which determines for a given speaker the number
of expression that has been a initiated by him
so to study the proposed from a we present in this paper copies based contrastive
study
that stands on a real interaction copper well involving you menu man and you man
agent but
as well as artificial cover all which
and used as a baseline
and in this work we provide several a study comparing
the real interaction corpora right to our baseline
comparing a double alignment in you menu men covers and human-agent copies and also studying
some condition on the am an agent copy such as a negotiation
so let me so if you will about
the negotiation corpora that we are using this work
so this negotiation corpora
involve two participants is that are required to find an agreement
over the of the amount of
okay they are they have to share
and this negotiation task can be is a integrative that is to say that can
jana to be a wean for bus participant
all completed you
and
this couple right available in that you monuments aiding continue in the human agents sitting
you consume the slide an image from the human agent corpora
in the human-agent sitting
the agent is controlled by you are without of course system
that has been designed to be as natural as possible
and this was system involves more than a eleven thousand possible you challenge is so
the agent as a wider variety of you terence to express it's a
the human colour i never eighty four that a white the human-agent corpora
involve one hundred then fifty four down
from these a couple are we constructed all based about a baseline the showing it
corpora
which have been designed to break the dynamic of us interactive alignment protocol
and to do that we decided to break the cooking between you differences
so starting from a real interaction dialogue
what we have done is that we have k
all the utterances from a speaker
where substituting all the user utterances from the speaker from the others a speaker
by you two entities should which was an from one concludes
from sorry from there are several pull
but utterances are chosen randomly
and the prove a specific
for the human participant
the human participant facing an agent and for the agent
system
so on the slide you can see an example of real dialogue on the colour
and of the left
and one randomized version where all the utterances from the human participant had been that
subject you to buy a randomly choose an and jones
so the main idea of these corpora used to break the dynamic of interactive alignment
process
so the first one of the first hypothesis is that we are investigating in this
work
is that it's the dialogue participants should constitute a richer expression lexicon
in the real interaction call logs and what would happen incidentally industrial get corporal
in the artificial or
and so to investigate this it was hypothesis we looked at the expression very variety
measure from all model
and
what we found
is that there is a significant shift different difference between the you menu man
and so the it's at if you can talk about as well as for human
agent as in as and it's
artificial can talk about
in the sense that is expression body right variety is higher in the real interaction
copper wire than in the signal string will get one
so what we have observed is that's or it was is we have a provided
some arguments to can for this is this hypothesis is that in the sense that
we have observed a richer expression lexicon in the real interaction couple and then the
in the artificial ones
which have been designed to avoid
the interaction process the interactive alignment process and thus the constitution of expression mexico
then we have been interest the in the comparison of that better alignments shows a
measure that we propose a
between the human corpora corpus and the agent corpus
so here what we expected that we expected that moldable alignment from the human
in the human-agent interaction
then the agent the main reason is that
the agent even if it even if it's a was it has not been designed
to be able to align
and the second reason is that
the human participant may be influenced by the belief about the limitation of the communicative
get abilities of the agents
so to us to this i prissy six we looked at the initiated expression right
sure that we propose in a model as well as the expression repetition ratio
and in the human interaction
in terms i would that there are no differences between the two speakers in that
it's there is a symmetrical that by alignments
regarding of these two measures
bus dialogue participants initiate
approximately the same amount of expression
and they repeat also the same amount of
of expression
however
is this is not the case in the human agents and sitting
and we observe here
and estimate
so
this estimator e a
is
this end symmetry happened and
can be is summarized by the fact that
the human participants adopt more was initiated expression
which is not surprising because the which cannot
a adopt easy to use a human participant expression the human participants also they did
get small talk into the repetition of expression
so a here
this give some
arguments to say that the human participant
is influenced by its belief about the limitations
of the communicative capabilities of the agents
and it should be stressed that lets us test image three a does not appear
when considering the number of the can produce by each speaker or when considering the
change proportion
is the proportion of vocabulary
finally we looked at some conditioned on the human agent corpus and
we have mainly focus on the negotiation type
in we wanted to see if there was an impact
on the verbal alignment indicators
given the type of negotiation so integrative negotiation which i don't know to be a
wean a distributive
negotiation
which is a competitive one
and what we found is that
both negotiation type have as a similar amounts the c is a similar value for
the expression for it
that is to says that down
the same amount of expression
that are created in both dialogues but there is a clear difference in the text
prediction repetition ratio
which shows that's
in the competitive in the negotiation
dialogue participants
repeats
all and their body allowing more
then in wean negotiation
so
the fact what we provide here is arguments to us about the fact that it's
competitive negotiation
due to more rubber alignment and one it was this is that
the participants a need to be already allowing more on control proposition
so to conclude on in this work and we have proposed automatic and generic measures
of the other alignment based on sequential pattern mining at the level of stuff first
of texture differences
that makes it possible to characterize
interesting aspect of that by law alignment such as the reading position process
the degree of repetition between that a participant and the orientation of the about that
alignment
we have contrast construe a contrastive then you menu man and you men agent that
better alignment showing us that there is a symmetry in babble alignment
when a given now indicators on
in human interaction why there is an asymmetry in human-agent interaction
and this touch we wanted to evenly comfy m some hypothesis is from they need
to ensure
and the perspective that we want to explore used to used as a measure that
we propose in a dialogue system and should be stressed that the major based on
very efficient algorithm is to say
linear complexity algorithms
we would like also to investigate this
more the query and to do a qualitative analysis of that but alignments between a
human interaction in human-agent interaction
such as a function and analysis of the repetition
and finally we would like to investigate
that was are comparable here menu man and human-agent gabor
to confirm or reasons
thank you for your attention and i'm now ready to answer your question ratio image
thanks for the top i was i was wondering several things about adopt actually one
of them is i i'm not quite sure you said something that on
way of the machine adapting to
to the user there's nothing out there
you have any idea why is nothing out there because when i looked into
it was like slot filling kind of dialogue and that was difficult because you don't
have a lot of data about user
to make the system about two but in this kind of data it might be
different and also the second question is whether
the measures that you come up with would work got for turn level
so if you have the decision to change from mexico expression
with those words to make changes that the turn level rather than
several turn
but like rather than taking into consideration example for
so for the first question about that there are systems that are able to align
as in some interesting work and they are pointed out in the in our paper
the main disadvantages that most of the system i'll based out rule based
and specific to some domain
all of some tasks
and the idea and providing measures and used to go towards more data driven way
an automatic way of aligning
but there are some system that i module
and the second question
so if i understand you where is that if we change the granularity of where
we've well where we look for expression
so we can not be over your problem i
don't see the em program in using all means of to be just changing that
when you're writing and ueller richie
of the units
which we
do you think you would get this you would keep the same accuracy
i don't know we have one check because here we go to variable for a
couple always very when the limited you challenge is
if we look at
i'm not sure to understand their we will your point in fact
we can talk a yes
hello i am here but talk about how you are looking on the degree of
repetition and what i didn't you are looking as repeated
i think perhaps not counting
probably so you get things like
i'm interested in shares or whatever was and in the next one you're getting a
time
in content items as being the repetition
in terms of being you know sort of
alignment which i think in this case where the participants don't really have so much
like what they say that phone first-person pronoun there is only one
and
you have similar ones for me i think that if you were doing alignments
on the on that might also be the same sort of a problem
what the in i think that it's just one of the difficulty to work when
we walk misalignment is that it can be very
you can very specific
words such as the difference between what time i used adding all at what time
and it is going to be very important in that case
and in this work we have chosen to
select all the expression
and to can everything even though we are probably counting some
expression that in around and that are still going to happen even without that but
alignment
but what we show in the by comparing to the strongest cultural
i think is that
when people line
they will create mall expression
so
i just if you were telling
information for
right
i think you would want to understand some of these things are alignment in some
ways
so that you would be producing delays
thinking
and regarding that
since
the expression mexican keeps track of expression instead our future such as the frequency
such as a recent c of an expression we can use it is it's is
it is features to feature out
an interesting expression
but can you because i could be extremely free
and it could be very recent as well
mm
i can just two
to copy this behaviour
we can choose to stop my sentences by the same expression that we use for
instance i want to align
thank you very much nothing to speaker again