0:00:16 | and then it means |
---|
0:00:17 | but not accommodating rumbled university japan |
---|
0:00:20 | i think i try to talk about alex context kind of ipd strong impression upon |
---|
0:00:24 | is shown over multiple thereof |
---|
0:00:27 | and this is done during the lack of all university and the quantity fact leads |
---|
0:00:31 | to the japan |
---|
0:00:34 | okay so |
---|
0:00:35 | she using a dialogue systems and should allow a dyad |
---|
0:00:40 | and must scotty and that's system should apply a |
---|
0:00:44 | under a |
---|
0:00:45 | accumulated knowledge during their ideas design example |
---|
0:00:49 | i used to a safe you know they are all that and that i-vector line |
---|
0:00:53 | don't take shown you to pay once for try to write |
---|
0:00:57 | and another a new that it i know that dialogue i like to fly to |
---|
0:01:02 | right after the tightest that the other station |
---|
0:01:05 | a three d the user utterances meetings that systems should apply and this kind of |
---|
0:01:11 | information about the upright airline pilots |
---|
0:01:15 | nice |
---|
0:01:16 | and this can be used for the future recommendation |
---|
0:01:21 | not correctly this kind of tum-initial is time we prepared by fifteen db don't a |
---|
0:01:27 | but do you think it should to be applied here and you adding the dialogue |
---|
0:01:32 | and the we also i'm really a closed domain that talked about |
---|
0:01:36 | and it's you think know ledgebased if necessary records |
---|
0:01:40 | assuming the dialogue corpus including all lexical item is unrealistic |
---|
0:01:46 | so i'll talk about i we want to and make that's talked about |
---|
0:01:52 | i two i buy a new concept dialogue and that will reduce cost to manually |
---|
0:01:59 | and knowledge base and now we are building a chat about in the food and |
---|
0:02:05 | the restaurant domain |
---|
0:02:06 | a laconic target is to apply two and a |
---|
0:02:10 | could than a hundred and they worry |
---|
0:02:13 | and the subset tractable us should be able to continue type dialog even for unknown |
---|
0:02:18 | talent |
---|
0:02:20 | the idea example source that you that's it i try to cook now supporting today |
---|
0:02:26 | and those is that it is not supporting it and on down for that system |
---|
0:02:31 | and simplest cradle simplest case john is |
---|
0:02:35 | so what even actually boring but is incredibly meetings such simple hundred maybe abrupt question |
---|
0:02:42 | if the variable that's |
---|
0:02:44 | so on a local it is in the lexicon are shown to implicit confirmation and |
---|
0:02:50 | here that system tried to acquire that on court one protocol cut their worries over |
---|
0:02:56 | unknown time to form a major |
---|
0:02:59 | i don't see is also an example of a single example i will try to |
---|
0:03:04 | class you're going to the and that's a great unknown time interval you |
---|
0:03:09 | o the system predicted |
---|
0:03:12 | okay we use one protocol category although this i don't amnesty one |
---|
0:03:18 | this is done by a british previous method |
---|
0:03:23 | our previous method |
---|
0:03:25 | i thought to use these |
---|
0:03:28 | channel that in the grammar and the four types of japanese a character types |
---|
0:03:33 | and a |
---|
0:03:35 | it can really |
---|
0:03:37 | and this may be a intimacy |
---|
0:03:40 | note that that's is then generate implicit confirmation request we lacked predictive category c o |
---|
0:03:48 | for example a good initial restaurant how or not on the geography |
---|
0:03:53 | and if the user on like i think so |
---|
0:03:58 | they probably user response |
---|
0:04:00 | and that extend it only if predicted how they what we see if correct why |
---|
0:04:06 | not for all a user is this use a response |
---|
0:04:11 | that system a quick story did i mean that |
---|
0:04:16 | these categories seem to be to show are quite then that system and a quite |
---|
0:04:21 | that's not supporting the only two in canadian category |
---|
0:04:26 | and this and it makes k y ix speech components are all mutually it's not |
---|
0:04:31 | a good for this task |
---|
0:04:34 | so |
---|
0:04:35 | this is mister example of explicit confirmation but |
---|
0:04:40 | there we need was incorrect a format on be the only one could argue that |
---|
0:04:45 | system if the model really an italian |
---|
0:04:48 | so we think that this kind of a comparison we case degraded that you that |
---|
0:04:53 | extra experience |
---|
0:04:55 | it's like to hand if also express |
---|
0:04:58 | confirmation and it is correct |
---|
0:05:01 | but very well yes so i sometimes have a much more mushroom are not purely |
---|
0:05:07 | and that system asks if the mushroom are rumoured audio the italian |
---|
0:05:11 | relative to yes and i think the sets explicitly expression degrade the user experience so |
---|
0:05:18 | we are now a lot wrote is using a implicit confirmation |
---|
0:05:25 | okay but so and to determine with that they were either correct or not if |
---|
0:05:31 | this card |
---|
0:05:33 | i because you use that the response |
---|
0:05:35 | you use that is boundaries b is expression |
---|
0:05:38 | and that includes not only simple affine what people and negative responses |
---|
0:05:43 | okay these lists either you very edus samples so you that it is picked up |
---|
0:05:48 | and or yesterday and the that's if their masks i want to one if you |
---|
0:05:52 | to japanese food |
---|
0:05:54 | they that users did not exactly well what are you talking about |
---|
0:05:58 | the by using these as a responsive that system can easily recognise that of the |
---|
0:06:03 | predicted after what do you want incorrect |
---|
0:06:06 | on the other hand in it you example or you that it is based upon |
---|
0:06:10 | goal yesterday under that extensive i wanted to stop and you the food |
---|
0:06:14 | and they do that |
---|
0:06:16 | i likely to |
---|
0:06:18 | so this is a difficult to determine |
---|
0:06:21 | and italian so |
---|
0:06:23 | this is not while it does not want a cue to did i mean that |
---|
0:06:28 | but it is to predict it cut they were collected or not |
---|
0:06:31 | so a lot of a obvious out a our problem without is to take various |
---|
0:06:37 | features into consideration and before and after that increase the complementarity with |
---|
0:06:43 | and i another i |
---|
0:06:45 | about the other programming to be sort of it in this one that are you |
---|
0:06:49 | that do not always respond to correctly |
---|
0:06:53 | so that i don't that means that there are sometimes inconsistent |
---|
0:06:58 | so this is also incorrect |
---|
0:07:01 | confirmation via estimate of the onion started around and the japanese with the |
---|
0:07:06 | and this user response i've just |
---|
0:07:10 | so |
---|
0:07:12 | if a guy's if there are you gotta that this |
---|
0:07:15 | is this indicates that |
---|
0:07:20 | activity correct |
---|
0:07:21 | they are incorrect nor its will be added into the system not it's |
---|
0:07:27 | so i left second problem is old is to exploit responses over multiple dialogues |
---|
0:07:35 | okay so let me someone it our proposed method is a first one is to |
---|
0:07:39 | design a deficit |
---|
0:07:41 | a whole machine learning based classification |
---|
0:07:44 | and that consider expression i dunno simple of comedy or negative responses |
---|
0:07:51 | and that miss out that exploits user utterances around i don't know that the mean |
---|
0:07:57 | before and after |
---|
0:07:58 | implicit confirmation request |
---|
0:08:01 | and all i think on the proposed method is to exploit the determination without although |
---|
0:08:06 | multiple dialogue |
---|
0:08:07 | and this become possible if other system is deployed on sr for this is a |
---|
0:08:14 | conventional one-to-one dialogue |
---|
0:08:16 | but now building but at about and it is the ensemble |
---|
0:08:20 | so that system channel interact use a multiple you that's |
---|
0:08:26 | if the same content |
---|
0:08:27 | so we integrate that is out and |
---|
0:08:30 | user that without |
---|
0:08:34 | for determining |
---|
0:08:36 | without that |
---|
0:08:37 | quite a pretty story |
---|
0:08:38 | predicted cutter will easily collect one |
---|
0:08:42 | okay so this is overview of our method o cost |
---|
0:08:46 | i've that i think i explained |
---|
0:08:48 | are you that some unknown town |
---|
0:08:51 | i do that system generate a implicit confirmation with a |
---|
0:08:56 | predicted category c and the now use a i thought this so |
---|
0:09:01 | utterance |
---|
0:09:02 | and it's and that system calculate the probability p w three from a single user |
---|
0:09:07 | response at this point i |
---|
0:09:10 | and the next that system |
---|
0:09:13 | and according to their responses problem and use that's |
---|
0:09:18 | that is or like this |
---|
0:09:20 | then after that we calculate a major role these a probability i'm so by integrating |
---|
0:09:29 | believe probabilities are be forwarding to find out of confidence major to detach with it |
---|
0:09:35 | to collect one |
---|
0:09:38 | okay so this evaluation so i explain to the background on the proposed method and |
---|
0:09:43 | the problem now i am you explain got a log files may result in more |
---|
0:09:48 | detail |
---|
0:09:49 | and the data for extra and experiment |
---|
0:09:51 | and the next i explained that our |
---|
0:09:54 | second propose a result |
---|
0:09:56 | and without on the computer we computed my talk |
---|
0:10:01 | for five middle part of our proposed with |
---|
0:10:04 | well |
---|
0:10:06 | so you calculate the probability that the response is that it is i believe and |
---|
0:10:11 | without category c for unknown time w if the covariance relative to collect a note |
---|
0:10:16 | but a |
---|
0:10:18 | we i introduce our notation for you wanted to |
---|
0:10:24 | so you don't is therefore user utterance |
---|
0:10:28 | you are containing the unknown time w |
---|
0:10:31 | and it's |
---|
0:10:34 | increase the controversial |
---|
0:10:36 | group-based |
---|
0:10:37 | including the predicted included category and you do is that this response to |
---|
0:10:43 | and the here we use |
---|
0:10:45 | using logistic regression for |
---|
0:10:50 | pretty for determining made it is predicted category the correct or not |
---|
0:10:56 | and we incorporated in table p g s |
---|
0:11:00 | so for the loop is expressed you do so |
---|
0:11:06 | not only affirmative or negative expression but also some of the expression and we also |
---|
0:11:13 | see this you expression and its relationship with what do you wanna under u two |
---|
0:11:20 | and finally we also incorporated a relationship between you want you to |
---|
0:11:26 | and to decide are listed and the teachers |
---|
0:11:30 | so this part of the six |
---|
0:11:33 | are constant six speech as |
---|
0:11:35 | under these that |
---|
0:11:38 | expression in u two |
---|
0:11:40 | for the for the two is a complex wanted to the baseline and we also |
---|
0:11:46 | you can incorporate it either voltages |
---|
0:11:50 | and the second group represent a that express shown you to an adaptive a user |
---|
0:11:56 | utterance before all have actually correct |
---|
0:11:58 | and that |
---|
0:12:00 | the last along its relationship between you and we used that means |
---|
0:12:04 | are you a way that you want you to contain |
---|
0:12:08 | the same one and whatnot |
---|
0:12:10 | and also featured by |
---|
0:12:14 | what is a before the result hundred |
---|
0:12:17 | but data collection |
---|
0:12:19 | so we collected a user utterance it's before and after implicit confirmation request |
---|
0:12:25 | a fast by of clauses fourteen |
---|
0:12:27 | the first we ask a walk while "'cause" to encode a think this is really |
---|
0:12:33 | about a specified by i |
---|
0:12:37 | for example i eight by how to fold up for that |
---|
0:12:41 | and so then that system responded |
---|
0:12:45 | i generate input is to call have initialized an implicit confirmation request |
---|
0:12:51 | that it is that correct or incorrect |
---|
0:12:54 | so |
---|
0:12:56 | these are requests correspond to a this specify the time so we pretty be able |
---|
0:13:01 | to increase to confirmation requests a for each specified that are |
---|
0:13:07 | for example italian particle for data twenty well i mean the dishes |
---|
0:13:13 | and then we ask the user we ask |
---|
0:13:17 | the walker to respond to do this |
---|
0:13:19 | a confirmation request |
---|
0:13:22 | so we pretty the other a twenty channels under their corresponding correct and incorrect if |
---|
0:13:28 | which to cover image only based |
---|
0:13:30 | and the we asked |
---|
0:13:32 | although one hundred workouts |
---|
0:13:34 | and the quality a lot of two thousand and of their own |
---|
0:13:38 | and we after that we excluded embodied utterances |
---|
0:13:44 | what is so this is the result of user logistic regression only ten fold cross |
---|
0:13:49 | validation |
---|
0:13:51 | and the we gotta that can cut their policy is correct if the probability was |
---|
0:13:57 | like larger and larger than zero point five |
---|
0:14:01 | and of |
---|
0:14:02 | this low so the baseline this really the proportion result |
---|
0:14:06 | and this |
---|
0:14:07 | table shows out a confusion matrix |
---|
0:14:10 | and we can see that a classification accuracy improved |
---|
0:14:14 | and |
---|
0:14:17 | especially no precision of the detection of the product cut a woody |
---|
0:14:24 | improved |
---|
0:14:25 | and this the most significant feature was if able that the you might include the |
---|
0:14:31 | cut they were eating use it is one |
---|
0:14:33 | so that means and that same topic if the shared what the shared |
---|
0:14:38 | in |
---|
0:14:39 | the u one and this one |
---|
0:14:42 | and that it is you insignificant a feature but not that it's a user included |
---|
0:14:48 | start of it |
---|
0:14:50 | then it in this result shows that proposed the p to improve the detection of |
---|
0:14:54 | incorrect categories |
---|
0:14:57 | what is needed to move along the next |
---|
0:14:59 | that second problem in front |
---|
0:15:04 | for this is a position she are so we take great it's the |
---|
0:15:10 | probabilities and the |
---|
0:15:12 | i integrate |
---|
0:15:13 | that |
---|
0:15:14 | probabilities |
---|
0:15:17 | so this continuous major is to determine collect cut they what is wrong in the |
---|
0:15:21 | user responses |
---|
0:15:23 | so easy a also used a logistic regression |
---|
0:15:27 | so we actually we test it as a regression function such as a random forest |
---|
0:15:32 | additional buttons we showed that it up it out of the logistic regression |
---|
0:15:38 | i don't we use this by the feature list at each year |
---|
0:15:42 | and this undercutting what we see a very valid have correct one time with w |
---|
0:15:47 | a range that computers the major xt does three shows so we change it is |
---|
0:15:51 | a shorter the value and the if corpus it |
---|
0:15:57 | exceeds a threshold |
---|
0:15:59 | the system channel at the same |
---|
0:16:01 | to that's if they would knowledge base |
---|
0:16:05 | what is so this is a conditional so we use of that same data |
---|
0:16:10 | as i explained before |
---|
0:16:12 | and the we divided them into to rate the training and test with it |
---|
0:16:17 | to make that experience perfectly open use a block on the request |
---|
0:16:22 | and the we selected in this policy is happening with that probability a longevity from |
---|
0:16:28 | a forty nine or forty eight one of the discourse in that we have all |
---|
0:16:33 | a lot of data fit |
---|
0:16:34 | and the daily all that in the feature value problem that in response it according |
---|
0:16:39 | to the computer vision |
---|
0:16:42 | and this |
---|
0:16:44 | i really show the result |
---|
0:16:46 | and this me six we present a fast and that the recognition performance improvement by |
---|
0:16:52 | multiple user responses the second one is how many sports event need to acquire these |
---|
0:16:57 | are correctly |
---|
0:16:58 | they were |
---|
0:16:59 | the third one is how to fit furniture for constant |
---|
0:17:04 | well this is that it out for the last question |
---|
0:17:08 | so we introduced |
---|
0:17:12 | e but so that the meantime break even point until indicating |
---|
0:17:16 | it indicates that parting mean precision rate is equal to the recall rate |
---|
0:17:22 | so i received operational and recall car and the we can see that p b |
---|
0:17:28 | e p value all and do not have to while larger than |
---|
0:17:35 | the top and recall while so |
---|
0:17:39 | that point and the diva any the larger than two |
---|
0:17:43 | so this means that added to the user response is i had able to improve |
---|
0:17:48 | the logistic regression functional |
---|
0:17:51 | performance |
---|
0:17:52 | and the two on the to determine if the predicted categories collect on |
---|
0:17:59 | under the second question is how many user responses are needed |
---|
0:18:03 | so i wrote it |
---|
0:18:07 | the increase in to be p-value i do that |
---|
0:18:11 | a horizontal axis is done about in |
---|
0:18:15 | so this pdf on this graph shows that increases in viterbi peabody while that's right |
---|
0:18:22 | in the in the one small |
---|
0:18:24 | so this indicates that it is worthwhile to ask them why users |
---|
0:18:29 | for |
---|
0:18:30 | that's a that's what you that's |
---|
0:18:33 | implicit confirmation request |
---|
0:18:36 | and the we all we can also see that the deep they diminished in |
---|
0:18:41 | become between class |
---|
0:18:43 | so this means that the d needed to be done problem asking what use |
---|
0:18:49 | the final question is how can we fit to the threshold |
---|
0:18:52 | so we think that hype original data are required because |
---|
0:18:57 | that systems should avoid applying incorrect information |
---|
0:19:01 | so we set high spatial so that the pressure data becomes also almost a one |
---|
0:19:06 | and the we predict recall rate in this day |
---|
0:19:11 | so we can see that the recall rate for indy five |
---|
0:19:15 | well as a zero point two one table one seven five |
---|
0:19:20 | so it is another all but we think but the old because |
---|
0:19:25 | we want to avoid a writing imported incorrect information |
---|
0:19:30 | and we also see that article recall rate but it only increase if we've |
---|
0:19:35 | so this means that substandard high threshold you know that system to a quite more |
---|
0:19:40 | categories along with high pressure right |
---|
0:19:44 | okay so let me somewhat i to this talk |
---|
0:19:47 | so a lot of timit goal is enabling with a realise that system that allowed |
---|
0:19:54 | to do that is dialogue |
---|
0:19:56 | and the tackled in this paper is to determine you stuck at a forty if |
---|
0:19:59 | correct or not a sort of that you wish to complement your process |
---|
0:20:04 | and the we propose to the middle part by dividing a feature set and that's |
---|
0:20:09 | the kind of it integrates the probability o into one complex vector |
---|
0:20:14 | and the result so that performance was improved |
---|
0:20:19 | so our future work is o |
---|
0:20:21 | two for the party line you are used to compare the implicit and explicit confirmation |
---|
0:20:27 | when you quit |
---|
0:20:28 | so we assume that an implicit confirmation it's a bit |
---|
0:20:33 | in that in the viewpoint of the user experience but we need to verify |
---|
0:20:38 | under the second one is to incorporate the proposed effort into it prototype |
---|
0:20:43 | the taurus it for your attention |
---|
0:20:51 | okay so we have about that so many possible questions |
---|
0:21:04 | and it's |
---|
0:21:08 | okay |
---|
0:21:10 | in future work |
---|
0:21:12 | i |
---|
0:21:17 | g u i |
---|
0:21:24 | i |
---|
0:21:25 | e |
---|
0:21:33 | i |
---|
0:21:45 | i think |
---|
0:21:50 | right |
---|
0:21:54 | you |
---|
0:22:03 | three |
---|
0:22:06 | so you |
---|
0:22:09 | once |
---|
0:22:18 | yes it's incentive for your comment and the i think we you |
---|
0:22:22 | and you want to use it to undo a we need to |
---|
0:22:28 | carefully designed that |
---|
0:22:30 | com experiment and |
---|
0:22:32 | i only that we needed to compare the just techniques speech type and but you |
---|
0:22:37 | proceed |
---|
0:22:38 | i and also |
---|
0:22:40 | explicit and |
---|
0:22:42 | not a kind of intuitive and based centre for document |
---|
0:22:47 | a question |
---|
0:22:49 | so one to the to do so it is to have the system so |
---|
0:22:55 | is you |
---|
0:22:58 | rubber goal is to tell you not just cool |
---|
0:23:01 | so that gives the user chance to say no stop |
---|
0:23:05 | or just |
---|
0:23:06 | not do anything so it is not as intrusive so you don't do not problem |
---|
0:23:11 | but sometimes very clear that this make this assumption |
---|
0:23:15 | so |
---|
0:23:16 | one point is don't being with this method in that sense about |
---|
0:23:22 | so intent to talk about that |
---|
0:23:26 | one cory to |
---|
0:23:28 | enjoying the conversation |
---|
0:23:30 | and you said that is we think that this kind of are repeated based on |
---|
0:23:35 | you very annoying |
---|
0:23:36 | four we want to |
---|
0:23:39 | and on the cheap conversational a continue and it |
---|
0:23:43 | as a to do that we are introducing a implicit confirmation |
---|
0:23:49 | you just a i mean i do so it is not a question is just |
---|
0:23:52 | the state so the user doesn't a response you don't to the dropped it is |
---|
0:23:58 | you that's |
---|
0:23:59 | most of your research on |
---|
0:24:01 | there are lots of the target restaurant |
---|
0:24:20 | so |
---|
0:24:22 | that combining the implicit and explicit intuitively if you have a high confidence |
---|
0:24:29 | understanding you could use the implicit and it very well conference |
---|
0:24:34 | my is the explicit |
---|
0:24:38 | you know that seems a little more natural maybe getting over some of the deception |
---|
0:24:41 | issues |
---|
0:24:43 | the first question |
---|
0:24:44 | consider you could also they can |
---|
0:24:46 | these things got nothing to annoy keeping a threshold how many questions are allowed to |
---|
0:24:50 | ask but it those kind of techniques |
---|
0:24:55 | so i already that under a at various you on the data that dialogue strategy |
---|
0:25:02 | so we just think that repeating with this kind of expressions to computationally very annoying |
---|
0:25:07 | so but we need to |
---|
0:25:10 | but i thought you could speechto coverage on implicit confirmation |
---|
0:25:15 | so i don't |
---|
0:25:17 | for convenience if one to control that confirmation process |
---|
0:25:27 | okay so it is just about times so that each sensor speaker |
---|