0:00:16 | if we could have all the position paper presenters up at the front |
---|
0:00:21 | and i think we have this microphone and the lapel mic there |
---|
0:00:26 | so our first question is for format |
---|
0:00:31 | who pointed out that there is already a commercial application |
---|
0:00:34 | involving negotiation dialogs so what other kinds of application involving negotiation or non-cooperative dialogue can |
---|
0:00:40 | you in addition |
---|
0:00:44 | okay so |
---|
0:00:46 | basically this is machines don't have any preferences |
---|
0:00:50 | but they can express constraints so the negotiation between a user and a machine |
---|
0:00:56 | general user that comes with preferences and machine that have |
---|
0:01:02 | that has some constraints |
---|
0:01:04 | and that the user is not aware |
---|
0:01:09 | so than the most commonly deployed the negotiation dialogue system nowadays it's probably the appointment |
---|
0:01:16 | scheduling |
---|
0:01:18 | there has been a lot of work on the on this task |
---|
0:01:21 | and |
---|
0:01:24 | and it still i still walking quite pretty well |
---|
0:01:31 | and but for the future even human operators are not too low to really negotiate |
---|
0:01:39 | with the customers |
---|
0:01:41 | so |
---|
0:01:42 | i |
---|
0:01:44 | i guess it's we want to load that for the four systems right now and |
---|
0:01:49 | neither |
---|
0:01:50 | but we can we can imagine that in the game industry there that would be |
---|
0:01:55 | a lot of a lot of by gaining a stuff like this are trying to |
---|
0:02:00 | trying to find a way to take information from characters |
---|
0:02:08 | and stuff like this |
---|
0:02:14 | the second question is for antenna how can we disentangle stance from sentiment emotion and |
---|
0:02:20 | negotiation dialogue |
---|
0:02:27 | i think there is like there's a lot of work and sentiment analysis and stands |
---|
0:02:31 | recognition in every type of like text but defined a want to make is that |
---|
0:02:36 | actually negotiation dialogue and especially argumentation |
---|
0:02:40 | is that different scenario because actually liking this kind of context also like expression on |
---|
0:02:47 | sentiment is used to express that stance |
---|
0:02:50 | so it's more difficult to disambiguate need to |
---|
0:02:54 | that's for sure and also like it could also be the case that you can |
---|
0:03:00 | have like express no emotion can be used what than as an argument not claim |
---|
0:03:07 | so i don't have a solution i just think it's more when we have to |
---|
0:03:11 | deal will like negotiation and argumentation discourse in general this equation is much more complex |
---|
0:03:17 | and so you have to model different dimension and not just like the pure semantics |
---|
0:03:23 | of i sentence |
---|
0:03:30 | the third question is for all good |
---|
0:03:32 | how can well how can we collect |
---|
0:03:36 | useful to date datasets for modeling negotiation dialogue |
---|
0:03:42 | so actually i was examining how many negotiation dialogs we already corpora were already half |
---|
0:03:47 | and we have pretty |
---|
0:03:49 | pretty much or a well in different domains but my question was more a big |
---|
0:03:55 | sv in this section session of you also have argumentative dialogues and what i recently |
---|
0:04:01 | found out and which can be a quite useful and explore exceed the out the |
---|
0:04:07 | visual content that we already have like |
---|
0:04:10 | presidential debates |
---|
0:04:12 | they are online you to put it up first i started to collect the design |
---|
0:04:18 | scenarios for metal a corpus i branch true debates in interviews new yorker anti smoking |
---|
0:04:27 | campaign and |
---|
0:04:29 | good music is a most of them are already transcribed it's automatic speech recognition but |
---|
0:04:36 | we can post correct can be save time on these |
---|
0:04:39 | also a lot of gaming a lot of v t p d at the received |
---|
0:04:43 | a bit db p at there is that i b m corpus features huge what |
---|
0:04:49 | we |
---|
0:04:50 | saved a lot of time on v gave to our professional school experts could meet |
---|
0:04:56 | these negotiation dialogue and actually be a working a lot of these doctors who need |
---|
0:05:01 | to convenes and negotiate is that patients shouldn't treatments |
---|
0:05:05 | so or be designed for them some kind of content creating tools where they can |
---|
0:05:12 | correct so information that the use and we can use these as a as a |
---|
0:05:20 | is the two for collecting data |
---|
0:05:23 | for simulating then extending down to retraining than |
---|
0:05:29 | using this a real users |
---|
0:05:31 | and also what they want to address is actually it's very sad everything kiss in |
---|
0:05:36 | english |
---|
0:05:38 | so there are |
---|
0:05:40 | very few corpora for german there is a bit of dutch there is a bit |
---|
0:05:44 | the french but not out the polish iso polish but well we all represents the |
---|
0:05:50 | multilingual community it would be nice because |
---|
0:05:53 | people need this actually in their own languages |
---|
0:05:57 | for the of |
---|
0:06:01 | thank you for that great point and the first question is for beth and what |
---|
0:06:05 | would your like ideal shared task negotiation dialogue if we collected all this |
---|
0:06:10 | things you suggested look like |
---|
0:06:15 | well i think that the sort of corpora we've been talking about |
---|
0:06:20 | be a little more challenging |
---|
0:06:22 | and then have systems run on all sorts of corpora that would be the obvious |
---|
0:06:26 | thing to do right |
---|
0:06:28 | so |
---|
0:06:30 | i mean even like the cut on corporate you could do that or some of |
---|
0:06:33 | the corpora are the christian i we're suggesting if we had some of those collected |
---|
0:06:38 | some of the existing corpora i think |
---|
0:06:42 | would be best if they were actually negotiation oriented rather than just argumentative |
---|
0:06:47 | because i think that we know less about what the language looks like in that |
---|
0:06:53 | negotiation scenario |
---|
0:06:56 | and particularly i think we know less about what it looks like in terms of |
---|
0:07:00 | the prosody and the various ways that people actually express emotions and stance and things |
---|
0:07:10 | like that are not |
---|
0:07:12 | they're not well captured by just transcribing usually and so you really need to have |
---|
0:07:17 | corpora that incorporate that and make you able to get it those things |
---|
0:07:25 | we invite questions from the audience |
---|
0:07:41 | i had a question about the collection based on jobs and emotional investment that |
---|
0:07:46 | since people know that it's kind of just plucked is the more real jobs i |
---|
0:07:52 | do still going to be emotionally invested |
---|
0:08:02 | went out before we went out for real tool conference so that we wouldn't embarrass |
---|
0:08:06 | anyone |
---|
0:08:08 | that were pretty intense am i wrong way |
---|
0:08:12 | where you would |
---|
0:08:16 | i also from a particular experience i think if the simulation if the practise is |
---|
0:08:24 | realistic enough |
---|
0:08:26 | i think you're emotional apparatus kicks in pretty well and it will be identical to |
---|
0:08:32 | the real thing but it is probably |
---|
0:08:35 | pretty close |
---|
0:08:37 | and you know would give you |
---|
0:08:40 | it would give you this other scenario right which is more of a real-world scenario |
---|
0:08:45 | i non gaming sort of real world scenarios |
---|
0:09:00 | so this is kind of a comments and i guess not try to make it |
---|
0:09:04 | into a question at some point but |
---|
0:09:06 | so we |
---|
0:09:08 | at usc we built this negotiation system on |
---|
0:09:14 | looking at multi-issue bargaining with records and lance and creates a nice a little bit |
---|
0:09:19 | about and one of the talks earlier |
---|
0:09:21 | and one of the things we found is that you know well that kind of |
---|
0:09:25 | system is easy to define are having apples and bananas since one we talk with |
---|
0:09:29 | people to actually train others and how to negotiate |
---|
0:09:33 | and what we found this the lot of the students people who are like doing |
---|
0:09:39 | in b as learning how to really negotiate real world deals found these system so |
---|
0:09:47 | simple as to seem almost irrelevant to the kinda negotiation they have to do and |
---|
0:09:54 | and when you look at the sorts of things come up in real world negotiation |
---|
0:09:59 | a lot of it is |
---|
0:10:01 | it's very open ended in it's a very creative exercise a lot of times the |
---|
0:10:07 | challenge one of the challenges is like figuring out what the even care about twenty |
---|
0:10:11 | start out you don't know if it's apples or bananas maybe what they care about |
---|
0:10:15 | is like hedging a certain kind of risk or something like that and you have |
---|
0:10:19 | to do a lot of work to discover that |
---|
0:10:22 | and then they can be quite creative and how they you know resolve those issues |
---|
0:10:28 | so it may be able to insert some weird legal terms in the contract that |
---|
0:10:34 | allow them to kinda |
---|
0:10:35 | and meted out that risk that they perceive and so i think a challenge for |
---|
0:10:40 | us as the |
---|
0:10:42 | community trying to model the stuff computational use you know how do we get how |
---|
0:10:46 | do we get into the space of all kind of the complexity of real world |
---|
0:10:51 | kind of discussions that come up it and negotiation and i think in some cases |
---|
0:10:56 | we can get there by looking at |
---|
0:10:59 | things like job interviews where you it is a real-world scenario because you have pretty |
---|
0:11:05 | standard things a get negotiated like salary in vacation and can i work remotely and |
---|
0:11:10 | all those sorts of things |
---|
0:11:13 | but i think are six sn kind of like having these things have impacts you |
---|
0:11:18 | know depends on us kinda really engaging beyond |
---|
0:11:22 | you know the bananas and apples and kind of getting in and two terms that |
---|
0:11:26 | people that really need to learn these skills you know can benefit from |
---|
0:11:32 | i guess my question is |
---|
0:11:34 | do you need you guys |
---|
0:11:37 | have ideas about kind of how do you how do you make this stuff appeal |
---|
0:11:42 | you know and the real world to people that really need to learn to negotiate |
---|
0:11:46 | at super important and really valuable to people |
---|
0:11:49 | you know that i think that's important for the community |
---|
0:11:54 | well it's very good question i i'm not claiming i can on so this hundred |
---|
0:11:59 | percent but i think negotiation skills are quite complex and for some people a pot |
---|
0:12:04 | of negotiation scale well what this negotiation skills what's involved is the presentation is that |
---|
0:12:10 | approximation is that's up |
---|
0:12:13 | money touring what the other is doing is it's reflecting go on what we try |
---|
0:12:17 | to not to lock it was met the cognitive skills training so it's money during |
---|
0:12:21 | what department is doing to find out trying to find out it's pretty his preferences |
---|
0:12:28 | and act accordingly cooperatively non-cooperative lee and depending on your preferences to send to you |
---|
0:12:34 | so |
---|
0:12:36 | but did for politicians and that that's it and do formant not be dialogue system |
---|
0:12:42 | in a sense of speech and vol still in perfect it was a bit though |
---|
0:12:46 | for |
---|
0:12:48 | annoying long but the apt negotiation aptitude built on they found the to very useful |
---|
0:12:55 | and they like to use it to posts the game it was very engaging and |
---|
0:13:01 | this so the poignant they try to train to calm to parental efficient outcome we |
---|
0:13:08 | are no improvement is possible unless you hard out there |
---|
0:13:12 | so and it was very useful for them are when they got what is the |
---|
0:13:17 | sns |
---|
0:13:19 | so i |
---|
0:13:21 | i would say i wouldn't break down the negotiation skills in two |
---|
0:13:26 | set of skills that are necessary and train maybe |
---|
0:13:31 | we also have |
---|
0:13:32 | i'm would be to |
---|
0:13:34 | jumping from the also for people |
---|
0:13:37 | to select themselves what be seeing for them is important to train |
---|
0:13:42 | so i want to train joss presentation or i want to train a quality of |
---|
0:13:48 | the arguments are present different my position or i want to train other |
---|
0:13:54 | so maybe |
---|
0:13:59 | to elaborate on that a little bit i think i one of the things that |
---|
0:14:02 | attractive to me about negotiation is it's one of these domains where you can't do |
---|
0:14:08 | it the way that you just rain through and do the |
---|
0:14:12 | what rest italian restaurant is downtown you know it's complex and i think what we |
---|
0:14:19 | need here is that classical we and in et al i which is to do |
---|
0:14:23 | abstraction |
---|
0:14:25 | right so i think that |
---|
0:14:26 | the whole |
---|
0:14:27 | the way you put together a system's gotta be at a different level of abstraction |
---|
0:14:32 | then some of the systems that we've been seeing and i think that's the attraction |
---|
0:14:46 | so we know that negotiation even if the simple a kind of what we've been |
---|
0:14:51 | doing right now it's already very difficult because it's |
---|
0:14:55 | it requires game theory elements is very difficult to train very difficult to have a |
---|
0:15:01 | something that is not easily exploitable by their whether it with the player |
---|
0:15:09 | so |
---|
0:15:11 | it's a bit like a poker game somehow and |
---|
0:15:15 | so you have to hide you information but at the same time you want to |
---|
0:15:18 | you want to you want to get what you want |
---|
0:15:22 | and so even it's |
---|
0:15:25 | no very soon |
---|
0:15:26 | simple form it's |
---|
0:15:28 | it's already very difficult to solve |
---|
0:15:30 | and if you add to this very complex so |
---|
0:15:34 | right context make an is |
---|
0:15:36 | it's very difficult to deal with it at all at once |
---|
0:15:41 | well as that's why we built simulator that abstracted everything that is not the negotiation |
---|
0:15:47 | game itself |
---|
0:15:53 | i always thought that actually like a good weight to improve negotiation skills i like |
---|
0:16:00 | the remediation so starting mediation unlike have a mediator of would i eight |
---|
0:16:08 | actually it's i think it's a good way to understand what are like a controversial |
---|
0:16:14 | fines and how to lexical what difference of opinion or how to being perceives e |
---|
0:16:19 | and so on |
---|
0:16:30 | alright well we are almost at the end of our time so i just want |
---|
0:16:34 | to reiterate that |
---|
0:16:35 | the position papers and links to the same down signal papers for the session are |
---|
0:16:40 | on the sessions website as well as pointers to resources and a |
---|
0:16:44 | bibliography |
---|
0:16:45 | so thank you all for coming and i think the next session started three thirty |
---|