Dataset Viewer
id
stringlengths 6
12
| text
stringlengths 1.53k
102k
|
|---|---|
Bdb001
|
grad c: Yeah , we had a long discussion about how much w how easy we want to make it for people to bleep things out . phd d: It it doesn't grad c: Did did did it ? I didn't even check yesterday whether it was moving . phd d: So I don't know if it doesn't like both of us grad c: Channel three ? Channel three ? phd d: You know , I discovered something yesterday on these , wireless ones . grad c: - ? phd d: You can tell if it 's picking up breath noise and stuff . So if you yeah , if you breathe under breathe and then you see AF go off , then you know it 's p picking up your mouth noise . phd f: In fact , if you listen to just the channels of people not talking , it 's like " @ @ " . It 's very disgust grad c: What ? Did you see Hannibal recently or something ? phd f: Sorry . So , grad c: phd f: I was gonna try to get out of here , like , in half an hour , cuz I really appreciate people coming , and the main thing that I was gonna ask people to help with today is to give input on what kinds of database format we should use in starting to link up things like word transcripts and annotations of word transcripts , so anything that transcribers or discourse coders or whatever put in the signal , with time - marks for , like , words and phone boundaries and all the stuff we get out of the forced alignments and the recognizer . So , we have this , I think a starting point is clearly the the channelized output of Dave Gelbart 's program , which Don brought a copy of , grad c: Yeah . phd f: which phd d: Can I see it ? grad c: And so the only question is it the sort of thing that you want to use or not ? Have you looked at that ? , I had a web page up . So , grad c: So phd f: I actually mostly need to be able to link up , or I it 's it 's a question both of what the representation is and grad c: You mean , this I guess I am gonna be standing up and drawing on the board . grad c: and then you can have lots of different sections , each of which have I Ds attached to it , and then you can refer from other sections to those I Ds , if you want to . I don't e I don't remember exactly what my notation was , phd a: Oh , I remember seeing an example of this . grad c: Yeah , " T equals one point three two " , And then I I also had optional things like accuracy , and then " ID equals T one , one seven " . And then , I also wanted to to be i to be able to not specify specifically what the time was and just have a stamp . grad c: Yeah , so these are arbitrary , assigned by a program , not not by a user . And then somewhere la further down you might have something like an utterance tag which has " start equals T - seventeen , end equals T - eighteen " . grad c: Right ? But it ends at this T - eighteen , which may be somewhere else . We don't know what the t time actually is but we know that it 's the same time as this end time . Right ? So you could you could have some sort of other other tag later in the file that would be something like , oh , I don't know , " noise - type equals door - slam " . You know ? And then , you could either say " time equals a particular time - mark " or you could do other sorts of references . So or or you might have a prosody " Prosody " right ? D ? T ? D ? T ? T ? phd f: It 's an O instead of an I , but the D is good . grad c: you know , so you could have some sort of type here , and then you could have , the utterance that it 's referring to could be U - seventeen or something like that . So , that seems that seems g great for all of the encoding of things with time and , grad c: Oh , well . phd f: I I guess my question is more , what d what do you do with , say , a forced alignment ? phd a: How - how phd f: you 've got all these phone labels , and what do you do if you just conceptually , if you get , transcriptions where the words are staying but the time boundaries are changing , cuz you 've got a new recognition output , or s sort of what 's the , sequence of going from the waveforms that stay the same , the transcripts that may or may not change , and then the utterance which where the time boundaries that may or may not change ? phd a: Oh , that 's That 's actually very nicely handled here because you could you could all you 'd have to change is the , time - stamps in the time - line without without , changing the I Ds . And you 'd be able to propagate all of the the information ? grad c: Right . phd a: You 'd have you 'd have phd f: The we we have phone - level backtraces . grad c: Yeah , this I don't think I would do this for phone - level . I think for phone - level you want to use some sort of binary representation phd f: grad c: because it 'll be too dense otherwise . So , if you were doing that and you had this sort of companion , thing that gets called up for phone - level , what would that look like ? phd a: Why grad c: I would use just an existing an existing way of doing it . But but why not use it for phone - level ? phd f: H h phd a: It 's just a matter of it 's just a matter of it being bigger . But if you have you know , barring memory limitations , or I w this is still the m grad c: It 's parsing limitations . I don't want to have this text file that you have to read in the whole thing to do something very simple for . You would use it only for purposes where you actually want the phone - level information , I 'd imagine . phd f: So you could have some file that configures how much information you want in your in your XML or something . , you 'd y phd f: phd a: You grad c: I I am imagining you 'd have multiple versions of this depending on the information that you want . grad c: I 'm just what I 'm wondering is whether I think for word - level , this would be OK . grad c: For lower than word - level , you 're talking about so much data that I just I don't know . I don't know if that phd f: we actually have So , one thing that Don is doing , is we 're we 're running For every frame , you get a pitch value , phd d: Lattices are big , too . phd f: and not only one pitch value but different kinds of pitch values grad c: Yeah , for something like that I would use P - file phd f: depending on grad c: or or any frame - level stuff I would use P - file . phd d: But what what 's the advantage of doing that versus just putting it into this format ? grad c: More compact , which I think is is better . grad c: if you did it at this phd f: these are long meetings and with for every frame , grad c: You don't want to do it with that Anything at frame - level you had better encode binary phd f: grad c: or it 's gonna be really painful . , b you can always , G - zip them , and , you know , c decompress them on the fly if y if space is really a concern . phd d: Yeah , I was thi I was thinking the advantage is that we can share this with other people . grad c: Well , but if you 're talking about one per frame , you 're talking about gigabyte - size files . These are really grad c: Right ? Because you have a two - gigabyte limit on most O Ss . But for phone - level stuff it 's perfectly phd f: And th it 's phd a: Like phones , or syllables , or anything like that . So , you know , people don't v Look at it , words times the average The average number of phones in an English word is , I don't know , five maybe ? phd f: Yeah , but we actually phd a: So , look at it , t number of words times five . That 's not that not phd f: Oh , so you mean pause phones take up a lot of the long pause phones . grad c: So I think it it 's debatable whether you want to do phone - level in the same thing . grad c: But I think , a anything at frame - level , even P - file , is too verbose . phd f: I haven't seen this particular format , phd a: I 've I 've used them . phd a: I 've forgot what the str phd d: But , wait a minute , P - file for each frame is storing a vector of cepstral or PLP values , grad c: It 's whatever you want , actually . grad c: So that what 's nice about the P - file It i Built into it is the concept of frames , utterances , sentences , that sort of thing , that structure . So , the only problem with it is it 's actually storing the utterance numbers and the frame numbers in the file , even though they 're always sequential . Is there some documentation on this somewhere ? grad c: Yeah , there 's a ton of it . I I was just looking for something I 'm not a database person , but something sort of standard enough that , you know , if we start using this we can give it out , other people can work on it , grad c: Yeah , it 's not standard . phd f: or Is it ? grad c: it 's something that we developed at ICSI . But , phd f: But it 's been used here grad c: But it 's been used here phd f: and people 've grad c: and and , you know , we have a well - configured system that you can distribute for free , and phd d: it must be the equivalent of whatever you guys used to store feat your computed features in , right ? phd f: OK . phd a: Yeah , th we have Actually , we we use a generalization of the the Sphere format . phd a: but Yeah , so there is something like that but it 's , probably not as sophist grad c: Well , what does H T K do for features ? phd d: And I think there 's grad c: Or does it even have a concept of features ? phd a: They ha it has its own , Entropic has their own feature format that 's called , like , S - SD or some so SF or something like that . grad c: I 'm just wondering , would it be worth while to use that instead ? phd d: Yeah . Th - this is exactly the kind of decision It 's just whatever phd d: But , people don't typically share this kind of stuff , right ? phd a: Right . phd f: Actually , I I just you know , we we 've done this stuff on prosodics and three or four places have asked for those prosodic files , and we just have an ASCII , output of frame - by - frame . phd f: Which is fine , but it gets unwieldy to go in and and query these files with really huge files . I was just thinking if there 's something that where all the frame values are grad c: And a and again , if you have a if you have a two - hour - long meeting , that 's gonna phd f: ? They 're they 're fair they 're quite large . phd f: and So it 's doable , it 's just that you can only store a feature vector at frame - by - frame and it doesn't have any kind of , phd d: Is is the sharing part of this a pretty important consideration phd f: phd d: or does that just sort of , a nice thing to have ? phd f: I I don't know enough about what we 're gonna do with the data . But I thought it would be good to get something that we can that other people can use or adopt for their own kinds of encoding . phd f: And especially for the prosody work , what what it ends up being is you get features from the signal , and of course those change every time your alignments change . So you re - run a recognizer , you want to recompute your features , and then keep the database up to date . phd f: Or you change a word , or you change a utterance boundary segment , which is gonna happen a lot . And so I wanted something where all of this can be done in a elegant way and that if somebody wants to try something or compute something else , that it can be done flexibly . , it doesn't have to be pretty , it just has to be , you know , easy to use , and grad c: Yeah , the other thing We should look at ATLAS , the NIST thing , phd f: Oh . phd f: grad c: I 'm not sure what to do about this with ATLAS , because they chose a different route . Your your file format can know about know that you 're talking about language and speech , which is what I chose , and time , or your file format can just be a graph representation . So what it looked like ATLAS chose is , they chose the other way , which was their file format is just nodes and links , and you have to interpret what they mean yourself . phd f: And why did you not choose that type of approach ? grad c: because I knew that we were doing speech , and I thought it was better if you 're looking at a raw file to be t for the tags to say " it 's an utterance " , as opposed to the tag to say " it 's a link " . grad c: So , but phd f: But other than that , are they compatible ? , you could sort of grad c: Yeah , they 're reasonably compatible . phd f: Yeah , that 's w So , grad c: So , well , the other thing is if we choose to use ATLAS , which maybe we should just do , we should just throw this out before we invest a lot of time in it . phd f: just sort of how to , cuz we need to come up with a database like this just to do our work . And I actually don't care , as long as it 's something useful to other people , what we choose . phd f: So maybe it 's maybe oth you know , if if you have any idea of how to choose , cuz I don't . phd a: Do they already have tools ? grad c: I I chose this for a couple reasons . phd f: And you can have as much information in the tag as you want , right ? grad c: Well , I have it structured . So what What NIST would say is that instead of doing this , you would say something like " link start equals , you know , some node ID , phd f: Yeah . So grad c: end equals some other node ID " , and then " type " would be " utterance " . phd f: So why would it be a a waste to do it this way if it 's similar enough that we can always translate it ? phd d: It probably wouldn't be a waste . It would mean that at some point if we wanted to switch , we 'd just have to translate everything . But it se Since they are developing a big phd f: But it but that sounds phd d: But that 's I don't think that 's a big deal . And so it seems to me that if if we want to use that , we might as well go directly to what they 're doing , rather than phd a: If we want to Do they already have something that 's that would be useful for us in place ? phd d: Yeah . , how stable is their Are they ready to go , grad c: The I looked at it phd d: or ? grad c: The last time I looked at it was a while ago , probably a year ago , when we first started talking about this . Since then , they 've developed their own external file format , which is , you know , this sort of s this sort of thing . , and apparently they 've also developed a lot of tools , but I haven't looked at them . phd f: would the tools would the tools run on something like this , if you can translate them anyway ? grad c: th what would would would what would worry me is that maybe we might miss a little detail phd a: It 's a hassle phd f: that I guess it 's a question that phd a: if phd f: yeah . phd a: I I think if it 's conceptually close , and they already have or will have tools that everybody else will be using , it would be crazy to do something s you know , separate that phd f: OK . phd f: Actually , so it 's that that would really be the question , is just what you would feel is in the long run the best thing . phd f: Cuz once we start , sort of , doing this I don't we don't actually have enough time to probably have to rehash it out again grad c: The Yep . The other thing the other way that I sort of established this was as easy translation to and from the Transcriber format . But , I suppose that as long as they have a type here that specifies " utt " , grad c: It 's almost the same . phd f: it 's yeah , close enough that grad c: The the the the point is with this , though , is that you can't really add any supplementary information . Right ? So if you suddenly decide that you want phd f: You have to make a different type . phd f: So Well , if you look at it and , I guess in my mind I don't know enough Jane would know better , about the types of annotations and and But I imagine that those are things that would well , you guys mentioned this , that could span any it could be in its own channel , it could span time boundaries of any type , grad c: Right . And then at the prosody - level we have frame sort of like cepstral feature files , grad c: Yep . And that 's sort of the world of things that I And then we have the aligned channels , of course , grad c: Right . phd a: And then phd f: I I definitely agree and I wanted to find actually a f a nicer format or a maybe a more compact format than what we used before . phd f: Just cuz you 've got ten channels or whatever and two hours of a meeting . phd a: Now now how would you how would you represent , multiple speakers in this framework ? Were You would just represent them as grad c: phd a: You would have like a speaker tag or something ? grad c: there 's a spea speaker tag up at the top which identifies them and then each utt the way I had it is each turn or each utterance , I don't even remember now , had a speaker ID tag attached to it . grad c: And in this format you would have a different tag , which which would , be linked to the link . grad c: Let 's see , would it be a node or a link ? And so so this one would have , an ID is link link seventy - four or something like that . grad c: And then somewhere up here you would have a link that that , you know , was referencing L - seventy - four and had speaker Adam . phd f: Actually , it 's the channel , I think , that phd a: Well , channel or speaker or whatever . phd f: w yeah , channel is what the channelized output out phd a: It doesn't grad c: This isn't quite right . phd f: Yeah , but phd a: But but so how in the NIST format do we express a hierarchical relationship between , say , an utterance and the words within it ? So how do you tell that these are the words that belong to that utterance ? grad c: you would have another structure lower down than this that would be saying they 're all belonging to this ID . And then each utterance could refer to a turn , phd d: So it 's it 's not hi it 's sort of bottom - up . phd f: And what if you actually have So right now what you have as utterance , the closest thing that comes out of the channelized is the stuff between the segment boundaries that the transcribers put in or that Thilo put in , which may or may not actually be , like , a s it 's usually not , the beginning and end of a sentence , say . phd f: So , I assume this is possible , that if you have someone annotates the punctuation or whatever when they transcribe , you can say , you know , from for from the c beginning of the sentence to the end of the sentence , from the annotations , this is a unit , even though it never actually i It 's only a unit by virtue of the annotations at the word - level . grad c: And , what phd f: But it 's just not overtly in the phd a: OK . phd f: cuz this is exactly the kind of phd a: So phd f: I think that should be possible as long as the But , what I don't understand is where the where in this type of file that would be expressed . phd f: S so it would just be floating before the sentence or floating after the sentence without a time - mark . grad c: You could have some sort of link type type equals " sentence " , and ID is " S - whatever " . phd a: grad c: Can you can you say that this is part of this , phd f: See , cuz it 's phd a: Hhh . phd f: it 's phd d: You would just have a r phd f: S grad c: or do you say this is part of this ? I think phd d: You would refer up to the sentence . phd f: But they 're phd a: Well , the thing phd f: they 're actually overlapping each other , sort of . grad c: So phd a: the thing is that some something may be a part of one thing for one purpose and another thing of another purpose . phd a: s , well , s let 's let 's ta so let 's grad c: Well , I think I 'm I think w I had better look at it again phd f: Yeah . phd a: y So for instance @ @ sup grad c: There 's one level there 's one more level of indirection that I 'm forgetting . phd a: Suppose you have a word sequence and you have two different segmentations of that same word sequence . phd a: I don't know if that 's true or not but let 's as phd f: Well , it 's definitely true with the segment . phd f: That 's what I exactly what I meant by the utterances versus the sentence could be sort of phd a: Yeah . So , you want to be s you want to say this this word is part of that sentence and this prosodic phrase . grad c: I I 'm pretty sure that you can do that , but I 'm forgetting the exact level of nesting . phd a: So , you would have to have two different pointers from the word up one level up , one to the sent grad c: So so what you would end up having is a tag saying " here 's a word , and it starts here and it ends here " . grad c: And then lower down you would say " here 's a prosodic boundary and it has these words in it " . phd f: So you would be able to go in and say , you know , " give me all the words in the bound in the prosodic phrase grad c: Yep . The the o the other issue that you had was , how do you actually efficiently extract , find and extract information in a structure of this type ? phd f: OK . phd a: So you gave some examples like phd f: Well , and , you guys might I don't know if this is premature because I suppose once you get the representation you can do this , but the kinds of things I was worried about is , phd a: No , that 's not clear . phd f: phd a: yeah , you c sure you can do it , phd f: Well , OK . So i if it phd a: but can you do it sort of l l you know , it phd f: I , I can't do it , but I can , phd a: y y you gotta you gotta do this you you 're gonna want to do this very quickly grad c: Well phd a: or else you 'll spend all your time sort of searching through very complex data structures phd f: Right . But an example would be " find all the cases in which Adam started to talk while Andreas was talking and his pitch was rising , Andreas 's pitch " . , that 's gonna be Is the rising pitch a feature , or is it gonna be in the same file ? phd f: Well , the rising pitch will never be hand - annotated . So the all the prosodic features are going to be automatically grad c: But the , that 's gonna be hard regardless , phd f: So they 're gonna be in those grad c: right ? Because you 're gonna have to write a program that goes through your feature file and looks for rising pitches . So normally what we would do is we would say " what do we wanna assign rising pitch to ? " Are we gonna assign it to words ? Are we gonna just assign it to sort of when it 's rising we have a begin - end rise representation ? But suppose we dump out this file and we say , for every word we just classify it as , w you know , rise or fall or neither ? grad c: OK . grad c: r phd f: So we would basically be sort of , taking the format and enriching it with things that we wanna query in relation to the words that are already in the file , grad c: Right . phd a: You want sort of a grep that 's that works at the structural on the structural representation . There 's a standard again in XML , specifically for searching XML documents structured X - XML documents , where you can specify both the content and the structural position . phd a: Yeah , but it 's it 's not clear that that 's That 's relative to the structure of the XML document , phd f: If phd a: not to the structure of what you 're representing in the document . grad c: It 's it 's you would use that to build your tool to do that sort of search . phd f: But as long as the grad c: It 's a graph , but phd a: That 's different from searching through the text . phd f: But it seems like as long as the features that grad c: Well , no , no , no . phd a: grad c: So that th phd f: That 's true if the features from your acoustics or whatever that are not explicitly in this are at the level of these types . phd f: That that if you can do that grad c: Yeah , but that 's gonna be the trouble no matter what . Right ? No matter what format you choose , you 're gonna have the trou you 're gonna have the difficulty of relating the the frame - level features phd f: That 's right . phd f: You know , it Or another example was , you know , where in the language where in the word sequence are people interrupting ? So , I guess that one 's actually easier . phd d: What about what about , the idea of using a relational database to , store the information from the XML ? So you would have XML basically would , you you could use the XML to put the data in , and then when you get data out , you put it back in XML . phd d: but then you store the data in the database , which allows you to do all kinds of good search things in there . grad c: The , One of the things that ATLAS is doing is they 're trying to define an API which is independent of the back store , phd f: Huh . grad c: so that , you could define a single API and the the storage could be flat XML files or a database . grad c: My opinion on that is for the s sort of stuff that we 're doing , I suspect it 's overkill to do a full relational database , that , just a flat file and , search tools I bet will be enough . phd a: But grad c: But that 's the advantage of ATLAS , is that if we actually take decide to go that route completely and we program to their API , then if we wanted to add a database later it would be pretty easy . phd f: It seems like the kind of thing you 'd do if I don't know , if people start adding all kinds of s bells and whistles to the data . And so that might be , it 'd be good for us to know to use a format where we know we can easily , input that to some database if other people are using it . grad c: I guess I 'm just a little hesitant to try to go whole hog on sort of the the whole framework that that NIST is talking about , with ATLAS and a database and all that sort of stuff , phd f: So grad c: cuz it 's a big learning curve , just to get going . grad c: Whereas if we just do a flat file format , sure , it may not be as efficient but everyone can program in Perl and and use it . grad c: Right ? phd a: But this is grad c: So , as opposed to phd a: I I 'm still , not convinced that you can do much at all on the text on the flat file that that you know , the text representation . e Because the text representation is gonna be , not reflecting the structure of of your words and annotations . It 's just it 's grad c: Well , if it 's not representing it , then how do you recover it ? Of course it 's representing it . You you have to what you have to do is you have to basically grad c: That 's the whole point . grad c: Right ? So what I was saying is that phd a: But that 's what you 'll have to do . And it 's a set of tools that let you specify given the D - DDT DTD of the document , what sorts of structural searches you want to do . So you want to say that , you know , you 're looking for , a tag within a tag within a particular tag that has this particular text in it , and , refers to a particular value . And so the point isn't that an end - user , who is looking for a query like you specified , wouldn't program it in this language . phd f: Is a See , I think the kinds of questions , at least in the next to the end of this year , are there may be a lot of different ones , but they 'll all have a similar nature . They 'll be looking at either a word - level prosodic , an a value , grad c: But you know , we 'll do something where we some kind of data reduction where the prosodic features are sort o , either at the word - level or at the segment - level , grad c: Right . They 're not gonna be at the phone - level and they 're no not gonna be at the frame - level when we get done with sort of giving them simpler shapes and things . , one that Chuck mentioned is starting out with something that we don't have to start over , that we don't have to throw away if other people want to extend it for other kinds of questions , grad c: Right . phd f: and being able to at least get enough , information out on where we condition the location of features on information that 's in the kind of file that you put up there . grad c: And so it seems to me that , I have to look at it again to see whether it can really do what we want , but if we use the ATLAS external file representation , it seems like it 's rich enough that you could do quick tools just as I said in Perl , and then later on if we choose to go up the learning curve , we can use the whole ATLAS inter infrastructure , phd f: Yeah . phd f: I I don't So if if you would l look at that and let us know what you think . phd f: I think we 're sort of guinea pigs , cuz I I want to get the prosody work done but I don't want to waste time , you know , getting the phd a: Oh , maybe phd f: Yeah ? phd a: grad c: Well , I wouldn't wait for the formats , because anything you pick we 'll be able to translate to another form . phd a: Well Ma well , maybe you should actually look at it yourself too to get a sense of what it is you 'll you 'll be dealing with , phd f: OK . phd a: because , you know , Adam might have one opinion but you might have another , so grad b: Yeah . phd f: Especially if there 's , e you know , if someone can help with at least the the setup of the right grad c: Hi , Jane . phd f: the right representation , then , i you know , I hope it won't We don't actually need the whole full - blown thing to be ready , grad c: Can you Oh , well . , so maybe if you guys can look at it and sort of see what , grad b: Yeah . phd f: I think we 're we 're we 're actually just grad c: We 're about done . phd f: wrapping up , but , Yeah , sorry , it 's a short meeting , but , Well , I don't know . Is there anything else , like that helps me a lot , grad c: Well , I think the other thing we might want to look at is alternatives to P - file . phd f: but grad c: th the reason I like P - file is I 'm already familiar with it , we have expertise here , and so if we pick something else , there 's the learning - curve problem . phd a: Is there an is there an IP - API ? grad c: And so Yeah . And , phd a: There used to be a problem that they get too large , grad c: a bunch of libraries , P - file utilities . phd a: and so basically the the filesystem wouldn't grad c: Well , that 's gonna be a problem no matter what . phd a: Maybe you could extend the API to , support , like splitting up , you know , conceptually one file into smaller files on disk so that you can essentially , you know , have arbitrarily long f grad c: Yep . That that most many of them can s you can specify several P - files and they 'll just be done sequentially . phd f: So , I guess , yeah , if if you and Don can if you can show him the P - file stuff and see . grad c: if you do " man P - file " or " apropos P - file " , you 'll see a lot . phd f: Yeah ? phd d: I don't remember what the " P " is , though . grad c: But there are ni they 're The Quicknet library has a bunch of things in it to handle P - files , phd a: Yeah . phd a: phd f: And that isn't really , I guess , as important as the the main I don't know what you call it , the the main sort of word - level grad c: Neither do I . , so grad c: Yeah , I 've been meaning to look at the ATLAS stuff again anyway . I guess it 's also sort of a political deci , if if you feel like that 's a community that would be good to tie into anyway , then it 's sounds like it 's worth doing . grad c: Yeah , I think it it w phd a: j I think there 's grad c: And , w , as I said , I what I did with this stuff I based it on theirs . So now that they have come up with a format , it doesn't it seems pretty reasonable to use it . grad c: As I said , that phd f: Cuz we actually can start grad c: There 's one level there 's one more level of indirection and I 'm just blanking on exactly how it works . phd f: we can start with , I guess , this input from Dave 's , which you had printed out , the channelized input . Cuz he has all of the channels , you know , with the channels in the tag and stuff like that . And so then it would just be a matter of getting making sure to handle the annotations that are , you know , not at the word - level and , t to import the grad b: Where are those annotations coming from ? phd f: Well , right now , I g Jane would would grad c: postdoc e: Are you talking about the overlap a annotations ? phd f: Yeah , any kind of annotation that , like , isn't already there . And since we w we I I think it 's important to remain flexible regarding the time bins for now . And so it 's nice to have However , you know , you want to have it , time time , located in the discourse . So , if we if we tie the overlap code to the first word in the overlap , then you 'll have a time - marking . It won't it 'll be independent of the time bins , however these e evolve , shrink , or whatever , increase , or Also , you could have different time bins for different purposes . And having it tied to the first word in an overlap segment is unique , you know , anchored , clear . postdoc e: Or the ? phd d: I 'm not sure what that @ @ grad c: Well , is that phd d: It probably doesn't matter . phd d: No , I d postdoc e: We don't have to go into the codes . W the idea is just to have a separate green ribbon , you know , and and and let 's say that this is a time bin . This is the first word of an overlapping segment of any length , overlapping with any other , word , i segment of any length . And , then you can indicate that this here was perhaps a ch a backchannel , or you can say that it was , a usurping of the turn , or you can you know , any any number of categories . But the fact is , you have it time - tagged in a way that 's independent of the , sp particular time bin that the word ends up in . postdoc e: we sh change the boundaries of the units , it 's still unique and and , fits with the format , phd f: Right . phd a: it would be nice , eh , gr this is sort of r regarding , it 's related but not directly germane to the topic of discussion , but , when it comes to annotations , you often find yourself in the situation where you have different annotations of the same , say , word sequence . phd a: And sometimes the word sequences even differ slightly because they were edited s at one place but not the other . phd a: So , once this data gets out there , some people might start annotating this for , I don't know , dialogue acts or , you know , topics or what the heck . And the only thing that is really sort of common among all the versi the various versions of this data is the word sequence , or approximately . But , see , if you 'd annotate dialogue acts , you don't necessarily want to or topics you don't really want to be dealing with time - marks . phd a: You 'd it 's much more efficient for them to just see the word sequence , right ? phd f: phd a: most people aren't as sophisticated as as we are here with , you know , time alignments and stuff . So So the the the point is grad c: Should should we mention some names on the people who are n ? phd a: Right . So , the p my point is that you 're gonna end up with , word sequences that are differently annotated . And you want some tool , that is able to sort of merge these different annotations back into a single , version . OK ? , and we had this problem very massively , at SRI when we worked , a while back on , well , on dialogue acts as well as , you know , what was it ? , phd f: Well , all the Switchboard in it . phd a: Because we had one set of annotations that were based on , one version of the transcripts with a particular segmentation , and then we had another version that was based on , a different s slightly edited version of the transcripts with a different segmentation . So , we had these two different versions which were you know , you could tell they were from the same source but they weren't identical . So it was extremely hard to reliably merge these two back together to correlate the information from the different annotations . But once you have a file format , I can imagine writing not personally , but someone writing a tool that is essentially an alignment tool , that mediates between various versions , phd f: phd a: and , sort of like th , you know , you have this thing in UNIX where you have , diff . phd a: There 's the , diff that actually tries to reconcile different two diffs f based on the same original . phd a: Something like that , but operating on these lattices that are really what 's behind this , this annotation format . phd a: So grad c: There 's actually a diff library you can use to do things like that that so you have different formats . phd f: You could definitely do that with the phd a: So somewhere in the API you would like to have like a merge or some some function that merges two two versions . phd f: But the one thing that would work here actually for i that is more reliable than the utterances is the the speaker ons and offs . So if you have a good , grad c: But this is exactly what , is that that the problem i phd f: Yeah . The problem is saying " what are the semantics , phd f: And grad c: what do you mean by " merge " ? " phd f: Right , right . So so just to let you know what we where we kluged it by , doing , by doing Hhh . phd a: Both were based on words , so , bo we have two versions of the same words intersp you know , sprinkled with with different tags for annotations . phd a: And that 's how grad c: That 's just wh how I would have done it . But , you know , it had lots of errors and things would end up in the wrong order , and so forth . phd a: it it was a kluge because it was basically reducing everything to , to , to textual alignment . grad c: A textual phd a: so phd f: But , d isn't that something where whoever if if the people who are making changes , say in the transcripts , cuz this all happened when the transcripts were different ye , if they tie it to something , like if they tied it to the acoustic segment if they You know what ? Then Or if they tied it to an acoustic segment and we had the time - marks , that would help . phd f: But the problem is exactly as Adam said , that you get , you know , y you don't have that information or it 's lost in the merge somehow , postdoc e: Well , can I ask one question ? phd f: so postdoc e: It it seems to me that , we will have o an official version of the corpus , which will be only one one version in terms of the words where the words are concerned . We 'd still have the the merging issue maybe if coding were done independently of the phd a: And you 're gonna get that postdoc e: But but phd a: because if the data gets out , people will do all kinds of things to it . And , s you know , several years from now you might want to look into , the prosody of referring expressions . And so that 's exactly what we should somehow when you distribute the data , say that you know , that have some way of knowing how to merge it back in and asking people to try to do that . postdoc e: Well , then the phd d: What 's what 's wrong with doing times ? I postdoc e: I agree . phd f: yeah , time is the grad c: Well , postdoc e: Time is unique . You were saying that you didn't think we should phd f: Time is passing ! phd a: Time time times are ephemeral . grad c: what if they haven't notated with them , times ? phd f: Yeah . postdoc e: But then couldn't you just indirectly figure out the time tied to the word ? phd f: But still they Exactly . phd d: But can they change the words without changing the time of the word ? grad c: Sure . The the point is , that that they may have annotated it off a word transcript that isn't the same as our word transcript , so how do you merge it back in ? I understand what you 're saying . grad c: And I I guess the answer is , it 's gonna be different every time . grad c: I it 's exactly what I said before , phd f: You only know the boundaries of the grad c: which is that " what do you mean by " merge " ? " So in this case where you have the words and you don't have the times , well , what do you mean by " merge " ? If you tell me what you mean , I can write a program to do it . phd f: And beyond that , all you know is is relative ordering and sometimes even that is wrong . grad c: So so in so in this one you would have to do a best match between the word sequences , phd f: So . grad c: extract the times f from the best match of theirs to yours , and use that . postdoc e: But it could be that they just , it could be that they chunked they they lost certain utterances and all that stuff , grad c: Right , exactly . phd f: Well , I guess , w I I didn't want to keep people too long and Adam wanted t people I 'll read the digits . phd f: if not , I guess phd a: For th for the for the benefit of science we 'll read the digits
|
Bed002
|
grad a: And you should be able to see which one which one you 're on by , watching the little bars change . grad a: So , actually , if you guys wanna go ahead and read digits now , as long as you 've signed the consent form , that 's alright . grad e: Are we supposed to read digits at the same time ? grad a: No . We 're talking about doing all at the same time but I think cognitively that would be really difficult . grad a: So , when you 're reading the digit strings , the first thing to do is just say which transcript you 're on . You can see the transcript ? There 's two large number strings on the digits ? So you would just read that one . And the pause is just so the person transcribing it can tell where one line ends and the other begins . And I 'll give I 'll read the digit strings first , so can see how that goes . Well , why don't I go ahead and read digit strings and then we can go on from there . All of you look like you 're doing it reasonably correctly , but you want it about two thumb widths away from your mouth , and then , at the corner . And that 's so that you minimize breath sounds , so that when you 're breathing , you don't breathe into the mike . And the short form , you should read the consent form , but , the thing to notice is that we will give you an opportunity to edit a all the transcripts . So , if you say things and you don't want them to be released to the general public , which , these will be available at some point to anyone who wants them , you 'll be given an opportunity by email , to bleep out any portions you don't like . , should I Do you want me to talk at all about why we 're doing this and what this project is ? professor c: yeah . Oh grad e: Does Nancy know that we 're meeting in here ? grad b: I sent an email . So are the people going to be identified by name ? grad a: Well , what we 're gonna we 'll anonymize it in the transcript . So , then in terms of people worrying about , excising things from the transcript , it 's unlikely . Oh , I see , but the a but the but the grad a: Right , so if I said , " Oh , hi Jerry , how are you ? " , we 're not gonna go through and cancel out the " Jerry "s . grad a: so we will go through and , in the speaker ID tags there 'll be , you know , M - one O seven , M - one O eight . grad a: it w , I don't know a good way of doing it on the audio , and still have people who are doing discourse research be able to use the data . grad a: And so we don't wanna have to do aliases professor c: Right . grad a: So I think that it 's better just as a pro post - process to edit out every time you bash Microsoft . The idea is that you 'd be able to put a PDA at the table at an impromptu meeting , and record it , and then be able to do querying and retrieval later on , on the meeting . So that 's my particular interest , is a portable device to do m , information retrieval on meetings . And so what we wanted is a room that 's instrumented with both the table top microphones , and these are very high quality pressure zone mikes , as well as the close talking mikes . What the close talk ng talking mikes gives us is some ground truth , gives us , high quality audio , especially for people who aren't interested in the acoustic parts of this corpus . So , for people who are more interested in language , we didn't want to penalize them by having only the far field mikes available . So that 's why we 're recording in parallel with the close talking and the far field at the same time . And then , all these channels are recorded simultaneously and framed synchronously so that you can also do things like , beam - forming on all the microphones and do research like that . Our intention is to release this data to the public , probably through f through a body like the LDC . So because the general environment is so challenging , we decided to to do at least one set of digit strings to give ourselves something easier . And it 's exactly the same digit strings as in TI - digits , which is a common connected digits corpus . grad a: OK , so when the l last person comes in , just have them wear a wireless . So , the most important form is the consent form , so just be s be sure everyone signs that , if they consent . grad b: I 'm sure it 's pretty usual for meetings that people come late , grad a: Yeah . And , just give me a call , which , my number 's up there when your meeting is over . grad a: And I 'm going to leave the mike here but it 's n , but I 'm not gonna be on so don't have them use this one . So you guys who got email about this oh f , Friday or something about what we 're up to . grad e: What was the nature of the email ? professor c: Oh , this was about , inferring intentions from features in context , and the words , like " s go to see " , or " visit " , or some grad b: Wel - we I I I professor c: You didn't get it ? grad e: I don't think I did . We could pursue , if we thought it 's it 's worth it but , I think we we will agree on that , to come up with a with a sort of very , very first crude prototype , and do some implementation work , and do some some research , and some modeling . So the idea is if you want to go somewhere , and focus on that object down Oh , I can actually walk with this . Now , we found in our , data and from experiments , that there 's three things you can do . If you want to actually go up or into the tower , you have to go this way , and then through some buildings and up some stairs and so forth . If you actually want to see the tower , and that 's what actually most people want to do , is just have a good look of it , take a picture for the family , you have to go this way , and go up here . grad b: or so That 's ab er , i the street network of our geographic information system . It would always use the closest point to the object , and then the tourists would be faced , you know , in front of a wall , but it would do them absolutely no good . grad e: What 's it what 's it made out of ? grad b: r red limestone . Okay , I This , These intentions , we w w we could , if we want to , call it the the Vista mode , where we just want to eh s get the overview or look at it , the Enter mode , and the , well , Tango mode . So But sometimes the the Tango mode is really relevant in the in the sense that , if you want to , If you don't have the intention of entering your building , but you know that something is really close to it , and you just want to approach it , or get to that building . Consider , for example , the Post Office in Chicago , a building so large that it has its own zip code . So , I 've looked , through twenty some , I didn't look through all the data . , and there there 's , a lot more different ways in people , the ways people phrase how to g get if they want to get to a certain place . Maybe I should go back a couple of steps and go through the professor c: No , OK come in , sit down . grad b: Is I I think grad e: No , that one 's already on , I thought he said . , people , when they w when they want to go to a building , sometimes they just want to look at it . I I gave an example where the point where you end up if you want to look at it is completely different from where if you want to enter it . So , this is sort of how people may , may phrase those requests to a a a mock - up system at least that 's the way they did it . And we get tons of of these " how do I get to " , " I want to go to " , but also , " give me directions to " , and " I would like to see " . And , what we can sort of do , if we look closer a closer at the the data That was the wrong one . This is of course a crucial factor , " what type of object is it ? " So , some buildings you just don't want to take pictures of . Sometimes I found in the , looking at the data , in a superficial way , I found some s sort of modifiers that that m may also give us a hint , " I 'm trying to get to " Nuh ? " I need to get to " . Sort of hints to the fact that you 're not really sightseeing and and just f there for pleasure and so forth and so on . That whatever it is you 're doing at the moment may also inter influence the interpretation of of a phrase . What we do know , is that the parser we use in the SmartKom system will never differentiate between any of these . So it 's it 's it 's way too crude to d capture those differences in intentions . So , I thought , " Mmm ! Maybe for a deep understanding task , that 's a nice sort of playground or first little thing . " Where we can start it and n sort of look " OK , we need , we gonna get those M - three - L structures . We may need additional part of speech , or maybe just some information on the verb , and modifiers , auxiliaries . And I will try to to sort of come up with a list of factors that we need to get out of there , and maybe we want to get a g switch for the context . So this is not something which we can actually monitor , now , but just is something we can set . And then you can all imagine sort of a a constrained satisfaction program , depending on on what , comes out . We want to have an a structure resulting if we feed it through a belief - net or or something along those lines . We 'd get an inferred intention , we we produce a structure that differentiates between the Vista , the Enter , and the , Tango mode . So we think it 's a well - formed , starter task for this , deeper understanding in the tourist domain . grad f: So , where exactly is the , deeper understanding being done ? Like , s is it before the Bayes - net ? Is it , professor c: Well , it 's the it 's it 's always all of it . But it 's deep enough that you can distinguish between these th three quite different kinds of , going to see some tourist thing . And , so that 's that 's the quote " deep " that we 're trying to get at . And , Robert 's point is that the current front - end doesn't give you any way to Not only doesn't it do it , but it also doesn't give you enough information to do it . It isn't like , if you just took what the front - end gives you , and used some clever inference algorithm on it , you would be able to figure out which of these is going on . So , and this is Bu - I in general it 's gonna be true of any kind of deep understanding , there 's gonna be contextual things , there 're gonna be linguistic things , there 're gonna be discourse things , and they gotta be combined . And , my idea on how to combine them is with a belief - net , although it may turn out that t some totally different thing is gonna work better . , the idea would be that you , take your You 're editing your slide ? grad b: Yeah . So the thing is , i , d naively speaking , you 've you 've got a for this little task , a belief - net , which is going to have as output , the conditional pr probability of one of three things , that the person wants to , to View it , to Enter it , or to Tango with it . And , then the question is there are two questions is , one , where do you get this i information from , and two , what 's the structure of the belief - net ? So what are the conditional probabilities of this , that , and the other , given these things ? And you probably need intermediate nodes . So it may well be that , for example , that , knowing whether Oh , another thing you want is some information abou I think , about the time of day . And , if things are obviously closed , then , you grad b: People won't want to enter it . And , if it 's not obvious , you may want to actually , point out to people that it 's closed you know , what they 're g going to is closed and they don't have the option of entering it . grad b: s b professor c: So another thing that can come up , and will come up as soon as you get serious about this is , that another option of course is to have a more of a dialogue . So one thing you could do is build a little system that , said , " whenever you got a question like that I 've got one of three answers . grad b: But maybe that 's a false state of the system , that it 's too close to call . You want the you want the ability to a You want the ability to ask , but what you don't wanna do is onl build a system that always asks every time , and i That 's not getting at the scientific problem , grad b: professor c: and it 's In general you 're you know , it 's gonna be much more complex than that . , I think also the the the deep understanding part of it is is going to be in there to the extent that we , want it in terms of our modeling . We can start , you know , basic from human beings , model that , its motions , going , walking , seeing , we can mem model all of that and then compose whatever inferences o we make out of these really conceptual primitives . S so so the way that might come up , if you wanna Suppose you wanted to do that , you might say , " , as an intermediate step in your belief - net , is there a Source - Path - Goal schema involved ? " OK ? And if so , is there a focus on the goal ? Or is there a focus on the path ? or something . And that could be , one of the conditiona you know , th the In some piece of the belief - net , that could be the the appropriate thing to enter . grad f: So , where would we extract that information from ? From the M - three - L ? professor c: No . See , the M - three - L is not gonna give th What he was saying is , the M - three - L does not have any of that . grad e: The M - three - L is the old SmartKom output ? professor c: Right . professor c: So we have th w we we we have to have a better w way of referring to grad b: The parser output ? professor c: grad b: " Analyzed speech " I think it 's what they call it , professor c: Yeah . grad b: o th No , actually , intention lattices is what we 're gonna get . professor c: Is - i but they c they call it intention lattice , but tha grad b: In - in a intention lattice k Hypothesis . So , th they 're gonna give us some cr or We can assume that y you get this crude information . And they don't give you the kind of object , they don't give you any discourse history , if you want to keep that you have to keep it somewhere else . grad e: So , if someone says , " I wanna touch the side of the Powder - Tower " , that would basically , we need to pop up Tango mode and the and the directions ? professor c: If i if Yeah , if it got as simple as that , yeah . But that doesn't necessarily But we 'd have to infer a Source - Path - Goal to some degree for touching the side , right ? grad b: Well , th the there is a p a point there if I understand you . " Where is the city hall ? " And this do they don't wanna sh see it on a map , or they don't wanna know it 's five hundred yards away from you , or that it 's to the your north . Where is that damn thing ? grad e: And the parser would output grad b: Well , that 's a a question mark . sh A lot of parsers , just , That 's way beyond their scope , is of interpreting that . You know ? But , still outcome w the outcome will be some form of structure , with the town hall and maybe saying it 's a WH focus on the town hall . grad e: I 'm just trying to figure out what the SmartKom system would output , depending on these things . grad b: it will probably tell you how far away it is , at least that 's That 's even what Deep Map does . Because i we can not differentiate , at the moment , between , you know , the intention of wanting to go there or the intention of just know wanting to know where where it is . grad d: People no might not be able to infer that either , right ? Like the fact Like , I could imagine if someone came up to me and asked , " Where 's the city hall ? " , I might say , g ar " Are you trying to get there ? " Because how I describe , t its location , p probably depend on whether I think I should give them , you know , directions now , or say , you know , whatever , " It 's half a mile away " or something like that . grad b: because where people ask you , " Where is New York ? " , you will tell them it 's on the East Coast . grad b: Y y eh you won't tell them how to get there , ft you know , take that bus to the airport and blah - blah - blah . grad b: But if it 's the post office , you will tell them how to get there . professor c: But i Go go back to the the , th grad b: So I w this is " onto " is is knowledge about buildings , professor c: Yeah , that slide . grad b: their opening times , and then t coupled with time of day , this should You know . grad d: So that context was like , their presumed purpose context , i like business or travel , as well as the utterance context , like , " I 'm now standing at this place at this time " . professor c: Yeah , well I think we ought to d a As we have all along , d We we 've been distu distinguishing between situational context , which is what you have as context , and discourse context , grad b: And , so what we were talking about doing , a a as a first shot , is not doing any of the linguistics . So , the the the reason the belief - net is in blue , is the notion would be , this may be a bad dis bad idea , but the idea is to take as a first goal , see if we could actually build a belief - net that would make this three way distinction , in a plausible way , given these We have all these transcripts and we 're able to , by hand , extract the features to put in the belief - net . Saying , " Aha ! here 're the things which , if you get them out of out of the language and discourse , and put them into the belief - net , it would tell you which of these three , intentions is most likely . " And if to actually do that , build it , you know , run it y y run it on the data where you hand - transcribe the parameters . th th i i if you can't do this task , grad b: We need a different , engine . Well it i I if it if it 's the belief - nets , we we 'll switch to you know , logic or some terrible thing , but I don't think that 's gonna be the case . I think that , if we can get the information , a belief - net is a perfectly good way of doing the inferential combination of it . The real issue is , do what are the factors involved in determining this ? And I don't know . grad d: I missed the beginning , but , I guess could you back to the slide , the previous one ? So , is it that it 's , These are all factors that , a These are the ones that you said that we are going to ignore now ? or that we want to take into account ? You were saying n professor c: Take them into account . And and it 's clear from the data , like , sorta the correct answer in each case . professor c: Let 's go back to th Let 's go back to the the the slide of data . grad d: That 's that 's the thing I 'm curious ab grad b: grad d: Like do we know from the data wh which OK . But , since we are designing a a a an , compared to this , even bigger data collection effort , we will definitely take care to put it in there , grad d: grad b: in some shape , way , form over the other , grad d: grad b: to see whether we can , then , get sort of empirically validated data . grad b: from this , we can sometimes , you know an and that 's that but that isn't that what we need for a belief - net anyhow ? is sort of s sometimes when people want to just see it , they phrase it more like this ? But it doesn't exclude anybody from phrasing it totally differently , even if they still grad d: grad b: But then other factors may come into play that change the outcome of their belief - net . And I 'm sure even i the most , sort of , deliberate data collection experiment will never give you data that say , " Well , if it 's phrased like that , the intention is this . grad b: You know , because then , you grad d: u u , the only way you could get that is if you were to give th the x subjects a task . Right ? Where you have where your , current goal is to grad b: We Yeah ! That 's what we 're doing . grad d: grad b: But but we will still get the phrasing all over the place . So , I think you all know this , but we are going to actually use this little room grad d: professor c: and start recording subjects probably within a month or something . So , this is not any lo any of you guys ' worry , except that we may want to push that effort to get information we need . If it turns out that we need data of a certain sort , then the sort of data collection branch can be , asked to do that . And one of the reasons why we 're recording the meeting for these guys is cuz we want their help when we d we start doing , recording of subjects . No , you you will not have , and there it is , and , But you know , y y the , grad d: And I think the other concern that has come up before , too , is if it 's I don't know if this was collected what situation this data was collected in . Was it is it the one that you showed in your talk ? Like people grad b: No , no . So was this , like , someone actually mobile , like s using a device ? grad b: N no , no not i it was mobile but not not with a w a real wizard system . But , is it I guess I don't know The situation of of collecting th the data of , like Here you could imagine them being walking around the city . And then you have all sorts of other c situational context factors that would influence w how to interpret , like you said , the scope and things like that . grad d: If they 're doing it in a you know , " I 'm sitting here with a map and asking questions " , I I would imagine that the data would be really different . But It was never th th the goal of that data collection to to serve for sat for such a purpose . So that 's why for example the tasks were not differentiated by intentionality , grad d: I 'm sure we can produce some if we need it , that that will help us along those lines . So , to Finding out what , you know , situational con what the contextual factors of the situation really are , you know is an interesting s interesting thing . grad b: u u Sort of I 'm , at the moment , curious and I 'm I 'm s w want to approach it from the end where we can s sort of start with this toy system that we can play around with , grad d: grad b: so that we get a clearer notion of what input we need for that , grad d: And then we can start worrying about where to get this input , what what do we need , you know Ultimately once we are all experts in changing that parser , for example , maybe , there 's just a couple three things we need to do and then we get more whatever , part of speech and more construction - type - like stuff out of it . grad e: How exactly does the data collection work ? Do they have a map , and then you give them a scenario of some sort ? grad b: OK . You 're gonna be in here , and somebody And and you see , either th the three - D model , or , a QuickTime animation of standing u in a square in Heidelberg . So , just off a textbook , tourist guide , to familiarize , yourself with that sort of odd - sounding German street names , like Fischergasse and so forth . Part two is , you 're told that this huge new , wonderful computer system exists , that can y tell you everything you want to know , and it understands you completely . And so you 're gonna pick up that phone , dial a number , and you get a certain amount of tasks that you have to solve . First you have to know find out how to get to that place , maybe with the intention of buying stamps in there . Maybe So , the next task is to get to a certain place and take a picture for your grandchild . It crashes , And grad d: a At the third ? Right then ? grad b: After the third task . And then , a human operator comes on , and and exp apologizes that the system has crashed , but , you know , urges you to continue , you know ? now with a human operator . And so , you have basically the same tasks again , just with different objects , and you go through it again , and that was it . Oh , and one one little bit w And , the computer you are you are being told the computer system knows exactly where you are , via GPS . And so you have to do some s tell the person sort of where you are , depending on what you see there . , this is a a a a a bit that I d I don't think we Did we discuss that bit ? , I just sort of squeezed that in now . grad d: So , in the display you can Oh , you said that you cou you might have a display that shows , like , the grad b: Yeah . grad d: And so , as you grad b: n grad d: Oh , two - D . grad d: So as you move through it that 's - they just track it on the for themselves grad b: Yeah . So grad b: Yeah ? that would be an an an enormous technical effort , unless we would We can show it walks to , you know . grad b: And you see the label of the name So we get those names , pronunciation stuff , and so forth , and we can change that . So your tasks don't require you to , yo you 're told So when your task is , I don't know , " Go buy stamps " or something like that ? So , do you have to respond ? or does your , what are you ste what are you supposed to be telling the system ? Like , w what you 're doing now ? or grad b: Well , we 'll see what people do . grad d: There 's no OK , so it 's just like , " Let 's figure out what they would say under the circumstances " . grad b: in both cases it 's gonna be a human , in the computer , and in the operator case . grad b: And we will re there will be some dialogue , you know ? So , you first have to do this , and that , grad d: Yep . But , maybe the maybe what you 're suggesting Is what you 're suggesting that it might be too poor , the data , if we sort of limit it to this ping pong one t , task results in a question and then there 's an answer and that 's the end of the task ? You wanna m have it more more steps , sort of ? grad d: Yeah , I I don't know how much direction is given to the subject about what their interaction , th they 're unfamiliar w with interacting with the system . , we we have to have this discussion of th the experiment , and the data collection , and all that sorta stuff grad d: - huh . Sh - Is sh grad d: She started taking the class last year and then didn't , you know , didn't continue . So , anyway , she 's looking for some more part time work w while she 's waiting actually for graduate school . So we may have someone , to do this , and she 's got you know , some background in in all this stuff . That 's So , Nancy , we 'll have an At some point we 'll have another discussion on exactly wha t t you know , how that 's gonna go . professor c: And , Jane , but also , Liz have offered to help us do this , data collection and design and stuff . professor c: So , when we get to that we 'll have some people doing it that know what they 're doing . I guess the reason I was asking about the sort of the de the details of this kind of thing is that , it 's one thing to collect data for , I don't know , speech recognition or various other tasks that have pretty c clear correct answers , but with intention , obviously , as you point out , there 's a lot of di other factors and I 'm not really sure , how how e the question of how to make it a t appropriate toy version of that , it 's ju it 's just hard . So , obviously it 's a grad e: Yeah , actually I guess that was my question . Is the intention implicit in the scenario that 's given ? Like , do the grad d: It is , if they have these tasks that they 're supposed to grad e: Yeah , I just wasn't sure to what level of detail the task was . professor c: the The problem that I was tr gonna try to focus on today was , let 's suppose by magic you could collect dialogues in which , one way or the other , you were able to , figure out both the intention , and set the context , and know what language was used . The issue is , can we find a way to , basically , featurize it so that we get some discrete number of features so that , when we know the values to all those features , or as many as possible , we can w come up with the best estimate of which of the , in this case three little intentions , are most likely . grad d: w What are the t three intentions ? Is it to go there , to see it , and grad b: To come as close as possible to it . professor c: Th - the terminology we 're using is to grad d: Yeah , it 's @ @ . " Take a picture of it " you you might well want to be a really rather different place than entering it . professor c: And , for an object that 's at all big , sort of getting to the nearest part of it , could be quite different than either of those . professor c: Just sort of grad d: OK , so now I understand the referent of Tango mode . grad b: S To " Waltz " it ? grad d: Yeah , like , how close are you gonna be ? professor c: Well . So grad f: All these So , like , the question is how what features can like , do you wanna try to extract from , say , the parse or whatever ? professor c: Right . grad f: Like , the presence of a word or the presence of a certain , stem , or certain construction or whatever . Is there a construction , or the kind of object , or w , anything else that 's in the si It 's either in the in the s the discourse itself or in the context . So if it turns out that , whatever it is , you want to know whether the person 's , a tourist or not , OK ? that becomes a feature . But fo for the current problem , it would just be , " OK , if you can be sure that it 's a tourist , versus a businessman , versus a native , " or something , that would give you a lot of discriminatory power and then just have a little section in your belief - net that said , " pppt ! " Though sin f in the short run , you 'd set them , grad f: professor c: and see ho how it worked , and then in the longer run , you would figure out how you could derive them . So , how should What 's the , plan ? Like , how should we go about figuring out these professor c: OK . So , first of all is , do e either of you guys , you got a favorite belief - net that you 've , you know , played with ? JavaBayes or something ? grad f: Oh . OK ? So y so one of th one of the things we wanna do is actually , pick a package , doesn't matter which one , presumably one that 's got good interactive abilities , cuz a lot of what we 're gonna be d You know , we don't need the one that 'll solve massive , belief - nets quickly . Because i that 's A lot of what it 's gonna be , is , playing with this . So that if if we have all these cases OK ? So we make up cases that have these features , OK , and then you 'd like to be able to say , " OK , here 's a bunch of cases " There 're even ones tha that you can do learning OK ? So you have all their cases and and their results and you have a algorithms to go through and run around trying to set the the probabilities for you . , my guess is we aren't gonna have enough data that 's good enough to make the these data fitting ones worth it , but I don't know . OK , and you wanna it s You know , the standard things you want it stable , you want it yeah , @ @ . And , as soon as we have one , we can start trying to , make a first cut at what 's going on . OK ? We we have a we know what the outcomes are gonna be , and we have some some data that 's loose , we can use our own intuition , and see how hard it is , and , importantly , what intermediate nodes we think we need . So it if it turns out that just , thinking about the problem , you come up with things you really need to You know , this is the kind of thing that is , you know , an intermediate little piece in your belief - net . grad b: And it and it may serve as a platform for a person , maybe me , or whoever , who is interested in doing some linguistic analysis . , w we have the For - FrameNet group here , and we can see what they have found out about those concepts already , that are contained in the data , you know , to come up with a nice little set of features and , maybe even means of s , extracting them . And and that altogether could also be , become a nice paper that 's going to be published somewhere , if we sit down and write it . And When you said JavaBayes belief - net you were talking about ones that run on coffee ? or that are in the program language Java ? professor c: No , th It turns out that there is a , The new end of Java libraries . I have no idea whether that 's The obvious advantage of that is that you can then , relatively easily , get all the other Java packages for GUIs or whatever else you might want to do . professor c: So that i that 's I think why a lot of people doing research use that . But it may not be I have no idea whether that 's the best choice an and there 're plenty of people around , students in the department who , you know , live and breathe Bayes - nets . So , grad d: There 's the m tool kit that , Kevin Murphy has developed , professor c: Right . I don't know I don't know whether you guys have met Kevin yet or not , grad b: grad b: But i But since we all probably are pretty sure that , the professor c: Yeah . And the ontology that , the student is is constructing for me back in in EML is in OIL and that 's also in XML . And so that 's where a lot of knowledge about bakeries , about hotels , about castles and stuff is gonna come from . grad b: so , if it has that IO capability and if it 's a Java package , it will definitely be able We can couple . grad b: Who isn't , nuh ? professor c: So , in terms of of interchanging in and out of any module we build , It 'll be XML . And if you 're going off to queries to the ontology , for example , you 'll have to deal with its interface . But that 's that 's fine an and , all of these things have been built with much bigger projects than this in mind . It 's kind of blackboards and multi - wave blackboards and ways of interchanging and registering your a And so forth . if we can get the core of the thing to work , in a way that we 're comfortable with , then we ca we can get in and out of it with , XML , little descriptors . Yeah , I like , for example , the what you said about the getting input from from just files about where you h where you have the data , have specified the features and so forth . professor c: I don't I don't see grad b: That 's , of course , easy also to do with , you know , XML . grad b: So r professor c: That that , you know , feature value XML format is probably as good a way as any . So it 's als Yeah , I guess it 's also worth , while you 're poking around , poke around for XML packages that , do things you 'd like . grad f: Doesn't does SmartKom system have such packages ? grad b: Yeah . It 's also professor c: And the question is , d you c you you 'll have to l We 'll have to l That should be ay We should be able to look at that grad b: No , u u y the What I What sort of came to my mind i is was the notion of an idea that if if there are l nets that can actually lear try to set their own , probability factors based on on on on input professor c: Yeah . grad b: which is in file format , if we , get really w wild on this , we may actually want to use some some corpora that other people made and , for example , if if they are in in MATE , then we get X M L documents with discourse annotations , t you know , t from the discourse act down to the phonetic level . grad b: Michael has a project where you know , recognizing discourse acts and he does it all in MATE , and so they 're actually annotating data and data and data . So if we w if we think it 's worth it one of these days , not not with this first prototype but maybe with a second , and we have the possibility of of taking input that 's generated elsewhere and learn from that , that 'd be nice . professor c: It 'd be nice , but but I I I do I don't wanna count on it . , you can't you can't run your project based on the speculation that that the data will come , grad b: No , no , just for professor c: and you don't have to actually design the nets . So in terms of of the , the what the SmartKom gives us for M - three - L packages , it could be that they 're fine , or it could be eeh . professor c: it 's , It doesn't control what you do in you know , internally . grad b: grad e: What 's the time frame for this ? grad b: Two days ? Two , three days ? professor c: Huh ? Yeah bu w I 'd like that this y yeah , this week , to ha to n to have y guys , you know , pick the y you know , belief - net package grad b: No . professor c: and tell us what it is , and give us a pointer so we can play with it or something . professor c: And , then as soon as we have it , I think we should start trying to populate it for this problem . Make a first cut at , you know , what 's going on , and probably the ea easiest way to do that is some on - line way . , you can f figure out whether you wanna make it a web site or You know , how grad b: I I I , OK , I t Yeah . grad b: But , maybe it might be interesting if if the two of you can agree on who 's gonna be the speaker next Monday , to tell us something about the net you picked , and what it does , and how it does that . grad b: So that will be sort of the assignment for next week , is to to for slides and whatever net you picked and what it can do and and how far you 've gotten . Pppt ! professor c: Well , I 'd like to also , though , ha have a first cut at what the belief - net looks like . OK ? So , you know , here a here are grad e: So we 're supposed to @ @ about features and whatnot , professor c: Right . professor c: And , as I said , what I 'd like to do is , what would be really great is you bring it in If if if we could , in the meeting , say , you know , " Here 's the package , here 's the current one we have , " , you know , " What other ideas do you have ? " and then we can think about this idea of making up the data file . Of , you know , get a t a p tentative format for it , let 's say XML , that says , l you know , " These are the various scenarios we 've experienced . " We can just add to that and there 'll be this this file of them and when you think you 've got a better belief - net , You just run it against this , this data file . grad e: And what 's the relation to this with Changing the table so that the system works in English ? grad b: OK . I 've downloaded them both , and I started to unpack the Linux one , the NT one worked fine . and I started unta pack the Linux one , it told me that I can't really unpack it because it contains a future date . Now , Then it will be my job to get this whole thing running both on Swede and on this machine . And then Hopefully that hoping that my urgent message will now come through to Ralph and Tilman that it will send some more documentation along , we I control p Maybe that 's what I will do next Monday is show the state and show the system and show that . , what one hopes is that when we understand how the analyzer works , we can both worry about converting it to English and worry about how it could ex extract the parameters we need for the belief - net . So we 're gonna do belief - nets this week , and then professor c: Oh , yeah . n None of this is i n Neither of these projects has got a real tight time - line , in the sense that over the next month there 's a there 's a deliverable . If if you know , if we don't get any information for these guys f for several weeks then we aren't gonna sit around , you know , wasting time , trying to do the problem or guess what they You know , just pppt ! go on and do other things . grad b: Yeah , but but the This point is really I think very , very valid that ultimately we hope that that both will merge into a harmonious and , wonderful , state where we can not only do the bare necessities , IE , changing the table so it does exactly in English what it does in German , but also that we can sort of have the system where we can say , " OK , this is what it usually does , and now we add this little thing to it " , you know ? whatever , Johno 's and Bhaskara 's great belief - net , and we plug it in , and then for these certain tasks , and we know that navigational tasks are gonna be a core domain of the new system , it all all of a sudden it does much better . Nuh ? Because it can produce better answers , tell the person , as I s showed you on this map , n you know , produce either you know , a red line that goes to the Vista point or a red line that goes to the Tango point or red line that goes to the door , which would be great . So not only can you show that you know something sensible but ultimately , if you produce a system like this , it takes the person where it wants to go . So this was actually an actual problem that we encountered , which nobody have has because car navigation systems don't really care . grad b: If you go d If you wanna drive to the SAP in Waldorf , I 'm sure the same is true of Microsoft , it takes you to the the address , whatever , street number blah - blah - blah , you are miles away from the entrance . professor c: Probably not then , cuz y you probably can't drop the mail there anyway . So , you two , who 'll be working on this , li are are you gl will you be doing Well , are you supposed to just do it by thinking about the situation ? Can you use the sample data ? professor c: Of course they use the sample data . grad d: Is it like Yeah , ho is there more than Is there a lot s of sample data that is beyond what you what you have there ? grad b: There there 's more than I showed , but , I think this is sort of , in part my job to look at that and and to see whether there are features in there that can be extracted , grad d: Yeah . grad b: and to come up with some features that are not you know , empirically based on on a real experiment or on on on reality grad d: Right . grad b: but sort of on your intuition of you know , " Aha ! This is maybe a sign for that , grad d: We can end the meeting and call Adam , and then we wanna s look at some filthy pictures of Heidelberg . professor c: is that OK ? grad b: And that 's why , when it was hit by , a cannon ball , it exploded . I first thought it had something to do with the material that it w that 's why I asked
|
Bed003
|
grad b: Could I hit hit F - seven to do that ? on the Robert ? grad a: I 'm grad b: Oh , the remote will do it OK . grad b: Cuz I 'm already up there ? grad a: in control here . grad b: So , we were Ah ! grad c: Johno , where are you ? grad b: OK . grad c: Should you go back to the first one ? grad b: Do I wanna go back to the first one ? grad c: Well grad b: OK . grad d: I 'm sorry I grad c: Well , just to grad b: OK . It 's basically talks about It just refers to the fact that one of main things we had to do was to decide what the intermediate sort of nodes were , grad d: I can read ! I 'm kidding . grad a: But if you really want to find out what it 's about you have to click on the little light bulb . grad b: Although I 've I 've never I don't know what the light bulb is for . grad a: Do you wanna try ? grad d: Ach u grad b: I 'd prefer not to . Is that the idea ? grad a: Why are you doing this in this mode and not in the presentation mode ? grad d: OK . grad b: Because I 'm gonna switch to the JavaBayes program grad a: Oh ! OK . grad b: You want me to Wait , what do you want me to do ? grad c: Can you maximize the window so all that stuff on the side isn't doesn't appear ? grad a: No , It 's OK . grad b: Well I can do that , but then I have to end the presentation in the middle so I can go back to open up grad c: OK , fine . grad b: So then the features we decided or we decided we were talked about , right ? the the prosody , the discourse , verb choice . Whether the and this i we actually have a separate feature but I decided to put it on the same line for space . " Nice walls " which we can look up because if you 're gonna get real close to a building in the Tango mode , right , there 's gotta be a reason for it . And it 's either because you 're in route to something else or you wanna look at the walls . The context , which in this case we 've limited to " business person " , " tourist " , or " unknown " , the time of day , and " open to suggestions " , isn't actually a feature . can I just ask the nice walls part of it is that , in this particular domain you said be i it could be on two different lines but are you saying that in this particular domain it happens the that landmark - iness cor is correlated with grad b: Oh grad c: No . grad b: I either could put " nice walls " on its own line or " open to suggestions " off the slide . grad c: Like you could have a p grad d: And and By " nice " you mean grad c: You Like you could have a post office with you know , nice murals or something . grad b: Or one time I was at this grad d: So " nice walls " is a stand in for like architecturally it , significant grad b: But see the thing is , if it 's grad c: Architecturally appealing from the outside . grad b: Yeah but if it 's architecturally significant you might be able to see it from Like you m might be able to " Vista " it , grad a: grad b: Yeah , versus , like , I was at this place in Europe where they had little carvings of , like , dead people on the walls or something . grad b: But if you looked at it real close , you could see the the in intricacy of the of the walls . grad a: The grad d: Something you want to inspect at close range because it 's interesting . grad b: Robert ? grad a: Well there there is a term that 's often used . And I was just wondering whether that 's the same as what you describe as " landmark - iness " . There 's landmark for , touristic reasons and landmark for I don't know navigational reasons or something . Tourist - y landmarks also happen to be Wouldn't couldn't they also be They 're not exclusive groups , are they ? Like non - tourist - y landmarks and grad a: Or it can be als grad b: direct navigational grad d: They 're not mutually exclusive ? grad b: Yeah . grad b: OK , So our initial idea was not very satisfying , because our initial idea was basically all the features pointing to the output node . grad b: And , so we Reasons being , you know , it 'd be a pain to set up all the probabilities for that . If we moved onto the next step and did learning of some sort , according Bhaskara we 'd be handicapped . grad c: Well usually , you know , N If you have N features , then it 's two to the N or exponential in N . So then our next idea was to add a middle layer , right ? So the thinking behind that was we have the features that we 've drawn from the communication of some Like , the someone s The person at the screen is trying to communicate some abstract idea , like " I 'm " the the abstract idea being " I am a tourist I want to go to this place . " Right ? So we 're gonna set up features along the lines of where they want to go and what they 've said previously and whatnot . Right ? but the middle thing , we were thinking along the lines of maybe trying to figure out , like , the concept of whether they 're a tourist or whether they 're running an errand or something like that along those lines . So then the hidden variables hair variables we came up with were whether someone was on a tour , running an errand , or whether they were in a hurry , because we were thinking , if they were in a hurry there 'd be less likely to like or th grad c: Want to do Vista , grad b: Right . grad c: right ? Because if you want to view things you wouldn't be in a hurry . grad b: Or they might be more likely to be using the place that they want to go to as a like a navigational point to go to another place . right now it 's still kind of in a toy version of it , because we didn't know the probabilities of or Well I 'll talk about it when I get the picture up . " Verb used " is actually personally amusing mainly because it 's it 's just whether the verb is a Tango verb , an Enter verb , or a Vista verb . grad c: Yeah , that one needs a lot of grad d: And are those mutually exclusive sets ? grad b: No . grad c: But that would 've made the probably significantly be more complicated to enter , grad d: Got it . grad c: so we decided that for the purposes of this it 'd be simpler to just have three verbs . Why don't you mention things about this , Bhaskara , that I am not that are not coming to my mind right now . grad c: OK , so Yeah , so note the four nodes down there , the sort of , the things that are not directly extracted . The " closed " is also not directly extracted I guess , from the grad b: Well i it 's grad c: . grad d: From the utterance ? grad b: it 's so it sort of is grad c: Actually , no , wait . grad b: because it 's because have the the time of day grad c: It is . grad c: Right , so f Right , but the other ones , the final destination , the whether they 're doing business , whether they 're in a hurry , and whether they 're tourists , that kind of thing is all sort of you know probabilistically depends on the other things . So we haven't , managed Like we don't have nodes for " discourse " and " parse " , although like in some sense they are parts of this belief - net . grad c: But The idea is that we just extract those features from them , so we don't actually have a node for the entire parse , grad d: grad d: So some of the the top row of things What 's what 's " Disc admission fee " ? grad c: whether they discuss the admission fees . So we looked at the data and in a lot of data people were saying things like " Can I get to this place ? " grad d: Oh . So that 's like a huge clue that they 're trying to Enter the place rather than to Tango or Vista , grad d: - huh . grad b: There were there 'd be other things besides just the admission fee , but you know , we didn't have grad d: So there are certain cues that are very strong either lexical or topic - based , concept cues grad b: From the discourse that Yeah . And then in that second row or whatever that row of Time of Day through that So all of those Some of them come from the utterance and some of them are sort of either world knowledge or situational things . grad d: " Unmark @ @ Time of Day " grad c: Yeah , I m grad a: Yeah . I would actually suggest we go through this one more time so we we all , agree on what what the meaning of these things is at the moment and maybe what changes we grad b: Yeah , th OK . so one thing I I 'm you know unsure about , is how we have the discus the " admission fee " thing set up . So one thing that we were thinking was by doing the layers like this , we kept things from directly affecting the mode beyond the concept , but you could see perhaps discus the " admission fee " going directly to the mode pointing at " Enter " , grad a: grad b: right ? Versus pointing to just at " tourist " , grad d: grad b: But we just decided to keep all the things we extracted to point at the middle and then down . That 's because we 're talking about landmarks as touristic landmarks not as possible grad b: Right . grad c: Disc - " admission fee " is a binary thing , " time of day " is like morning , afternoon , night . grad b: That 's how we have it currently set up , grad a: Yep . grad b: but it could be , you know , based upon hour grad c: Yeah . Normally context will include a huge amount of information , but , we are just using the particular part of the context which consists of the switch that they flick to indicate whether they 're a tourist or not , I guess . grad c: So Right , grad d: Right ? grad c: so it 's not really all of context . Similarly prosody is not all of prosody but simply for our purposes whether or not they appear tense or relaxed . grad a: The the So the context is a switch between tourist or non - tourist ? grad c: and grad a: Or also unknown ? grad b: Or un unknown , grad a: OK . Unknown , right ? grad d: So final dest So it seems like that would really help you for doing business versus tourist , grad c: Which is th Which one ? grad d: but OK . so the the context being , e I don't know if that question 's sort of in general , " are you " the ar ar are do they allow business people to be doing non - business things at the moment ? grad c: Yeah , it does . So then you just have some probabilities over grad c: Everything is probablistic , and There 's always grad d: OK . " Verb used " is like , right now we only have three values , but in general they would be a probability distribution over all verbs . grad c: " nice walls " is binary , " closed " is binary " final destination " , again Yeah , all those are binary I guess . grad c: Yeah , anything with a question mark after it in that picture is a binary node . Right ? grad c: Which things ? grad a: Nice walls ? grad b: Wi grad d: grad b: It is binary but it doesn't have question mark because it 's extracted . grad a: So we can either be in a hurry or not , but we cannot be in a medium hurry at the moment ? grad c: Well , we To do that we would add another value for that . grad c: And that would require s updating the probability distribution for " mode " as well . grad d: So , of course this will happen when we think more about the kinds of verbs that are used in each cases grad a: Yeah , yeah . grad d: but you can imagine that it 's verb plus various other things that are also not in the bottom layer that would that would help you Like it 's a conjunction of , I don't know , you know , the verb used and some other stuff that that would determine grad c: Right . grad a: well the the sort of the landmark is is sort of the object right ? the argument in a sense ? grad d: Usually . I I don't know if that 's always the case I I guess haven't looked at the data as much as you guys have . grad a: that 's always warping on something some entity , grad d: grad a: and maybe at this stage we will we do want to sort of get modifiers in there grad b: . grad a: because they may also tell us whether the person is in a hurry or not grad b: I want to get to the church quickly , grad c: Yeah . Do we have anything else to say about this ? grad c: We can do a little demo . grad a: No , then it wouldn't be a demo I was just gonna s grad c: We can do a demo in the sense that we can , just ob observe the fact that this will , in fact do inference . grad c: So we can , you know , set some of the nodes and then try to find the probability of other nodes . grad c: just I don't know , say they discussed the admission fee grad b: OK . grad c: and the place has nice walls grad b: I love nice walls , OK ? I 'm a big fan . grad d: it 's starting to grow on me grad b: And the time of day is night ? grad c: Yeah , no wait . grad b: One thing that bugs me about JavaBayes is you have to click that and do this . grad c: So that is the probability that they 're Entering , Vista - ing or Tango - ing . grad c: And grad d: So slightly biased toward " Tango " ing grad c: Yeah . grad b: If it 's night time , they have not discussed admission fee , and the n walls are nice . The reason I say the demo doesn't work very well is yesterday we observed everything in favor of taking a tour , and it came up as " Tango " , right ? Over and over again . grad c: Like , we totally hand - tuned the probabilities , grad d: Yeah . We were like " , well if the person does this and this and this , let 's say forty percent for this , grad d: OK . grad a: Yeah but it it grad d: Maybe the bias toward " Tango " ing was yours , then ? grad b: Yeah , grad c: Yeah . grad b: that 's that 's at grad c: It 's So we have to like fit the probabilities . grad d: So , the real case ? grad a: However you know , it The purpose was not really , at this stage , to come up with meaningful probabilities but to get thinking about that hidden middle layer . grad a: And grad b: We would actually I guess once we look at the data more we 'll get more hidden nodes , grad a: grad b: No , I think we should have exponentially more middle nodes than features we 've extracted . Whether you 're It 's whether It 's not grad d: And are th grad c: I think it 's more like " Are you are tourist ? are you in Ham - like Heidelberg for a " grad d: Oh , so , I thought that was directly given by the context switch . What if the context , which is not set , but still they say things like , " I want to go , see the the the castle and , et cetera . " grad a: Is it grad b: Well the I kind of thought of " doing business " as more of running an errand type thing . grad a: So if you run out of cash as a tourist , and and and you need to go to the AT grad b: So i wi th grad d: OK . grad a: " How do I get to the bank ? " grad d: I see . grad c: And that 'll affect whether you want to enter or you if you kinda thing . grad c: Yeah , I think this context node is a bit of a I don't know , like in d Do we wanna have Like it 's grad d: Are you assuming that or not ? Like is that to be if that 's accurate then that would determine tourist node . grad c: If the context were to set one way or another , that like strongly , says something about whether whether or not they 're tourists . grad c: So what 's interesting is when it 's not when it 's set to " unknown " . grad a: We - what set the they set the context to " unknown " ? grad d: OK . grad c: Right now we haven't observed it , so I guess it 's sort of averaging over all those three possibilities . grad a: And if we now do leave everything else as is the results should be the same , grad b: Oops . grad c: Well no , because we Th - the way we set the probabilities might not have Yeah , it 's it 's an it 's an issue , right ? Like grad a: Pretty much the same ? grad c: Yeah , it is . So the issue is that in belief - nets , it 's not common to do what we did of like having , you know , a d bunch of values and then " unknown " as an actual value . What 's common is you just like don't observe the variable , grad d: Yeah . grad c: But We didn't do this because we felt that there 'd I guess we were thinking in terms of a switch that actually grad b: We were thi Yeah , grad a: grad b: We were th grad c: But I don't know y what the right thing is to do for that . grad a: Why don't we Can we , How long would it take to to add another node on the observatory and , play around with it ? grad c: Another node on what ? grad b: well it depends on how many things it 's linked to . If we create something that for example would be So th some things can be landmarks in your sense but they can never be entered ? So for example s a statue . grad a: So maybe we wanna have " landmark " meaning now " enterable landmark " versus , something that 's simply just a vista point , for example . grad a: Yeah ? , a statue or grad c: So basically it 's addressing a variable that 's " enterable or not " . grad b: Also you know , didn't we have a size as one ? The size of the landmark . grad c: What ? grad b: Cuz if it 's grad c: . grad b: For some reason I had that OK , that was a thought that I had at one point but then went away . grad c: So you want to have a a node for like whether or not it can be entered ? grad a: Well , for example , if we include that , yeah ? grad c: Yeah . grad a: accessibility or something , yeah ? " Is it Can it be entered ? " grad c: . In the sense that , you know , if it 's Tom the house of Tom Cruise , you know , it 's enterable but you may not enter it . grad a: Yeah ? and And these are very observable sort of from the from the ontology sort of things . grad b: Way Does it actually help to distinguish between those two cases though ? Whether it 's practically speaking enterable , or actually physically enterable or not ? grad a: y y If If you 're running an errand you maybe more likely to be able to enter places that are usually not al w you 're not usually not allowed to m grad d: It seems like it would for , determining whether they wanna go into it or not . grad b: Well I can see why grad d: Cuz they grad a: Let 's get this b clearer . grad b: Whether it 's a Whether it 's a public building , and whether it 's actually has a door . grad a: This is sort of grad b: So Tom Cruise 's house is not a public building grad d: grad b: OK , sh explain to me why it 's necessary to distinguish between whether something has a door and is not public . Or , if something It seems like it 's equivalent to say that it doesn't have a door a and it grad a: grad b: Or " not public " and " not a door " are equivalent things , grad a: Yeah . So we would have What does it mean , then , that we have to we have an object type statue . grad a: And then we have , for example , an object type , that 's a hotel . It 's the hotel Zum Ritter , which is the only Renaissance building in Heidelberg that was left after the big destruction and for the Thirty Years War , blah - blah - blah . - And lots of detail , c and carvings , engravings and so forth , grad b: Excellent . So I guess your question is so far I have no really arg no real argument why to differentiate between statues as statues and houses of celebrities , from that point of view . Let Let 's do a Can we add , just so I can see how it 's done , a " has door " property or ? grad b: OK . grad c: What would it , connect to ? Like , what would , it affect ? grad a: I think , it might affect Oh actually it 's it it wouldn't affect any of our nodes , right ? grad c: What I was thinking was if you had a like grad a: Oh it 's it affects th The " doing business " is certainly not . grad b: You could affect Theoretically you could affect " doing business " with " has door " . grad a: right ? grad c: Yeah , I don't know if JavaBayes is nice about that . It might be that if you add a new thing pointing to a variable , you just like it just overwrites everything . Whew ! grad c: Well that 's fine , but we have to see the function now . grad b: This grad c: What would be nice if it is if it just like kept the old function for either value but . grad b: Oh wait , it might be Did we w Yes , that 's not good . grad a: Maybe you can read in ? grad c: Ha - So have you used JavaBayes a lot ? grad d: Yes . Really I ha I 've I haven't used it a lot and I haven't used it in the last you know many months so grad c: OK . grad c: Like , we looked at sort of a page that had like a bunch of grad d: Yeah . grad c: in a way this is a lot of good features in Java it 's cra has a GUI and it 's grad d: grad c: What ? grad b: Maybe it did a little bit of learning , grad c: OK . grad a: What is the c code ? Can w can we see that ? How do you write the code grad b: The c grad a: or do you actually never have to write any code there ? grad c: Yeah . grad b: Oh man , grad c: Like , there 's the grad b: I didn't n Is there an ampersand in DOS ? grad c: Nope . grad c: It 'll ask you what you what it wants what you want to open it with and see what BAT , I guess . grad c: That 's Oh ! grad b: Maybe it was just grad a: Oh . grad b: I like I like Word Pad because it has the the returns , grad a: Wordpad ? I grad b: the carriage returns on some of them . grad b: You know how they get " auto - fills " I guess , grad a: Mmm grad b: It just basically looks like it just specifies a bunch of grad a: grad b: It just that it 's grad c: But they 're not very friendly . grad b: Yeah the ordering isn't very clear on grad c: So you 'd have to like figure out Like you have to go and grad d: Right . grad c: it 's not grad b: We were doing it grad c: Yeah we can maybe write an interface th for entering probability distributions easily , something like like a little script . I actually seem to recall Srini complaining about something to do with Entering probability so this is probably grad c: The other thing is it is in Java grad d: Yeah , it 's Yeah . grad b: Or grad a: Do you have the true source files or just the class ? grad b: I don't know if he actually grad c: Yeah . we do grad b: Does he grad c: I I saw directory called " source " , grad b: Oh . grad c: I think it might it might be simpler to just have a script that , you know It 's , like , friendly , grad d: The d the data tables . grad a: But if th if there is an XML file that or format that it can also read it just reads this , right ? When it starts . grad b: Yeah I know there is an I was looking on the we web page and he 's updated it for an XML version of I guess Bayes - nets . grad c: The JavaBayes guy ? So but , e he doesn't use it . So in what sense has he updated it ? grad b: Well th you can either you ca or you can read both . grad b: Because Well at least the I could have misread the web page , I have a habit of doing that , but . grad a: So you got more slides ? grad b: Do I have more slides ? yes , one more . E That 's maybe , I don't know If grad b: that 's future future work . grad b: And of course if you have a presentation that doesn't have something that doesn't work at all , then you have " What I learned " , as a slide . grad b: I know what I like about these meetings is one person will nod , and then the next person will nod , and then it just goes all the way around the room . So this means grad b: Should I pull up the net again ? grad d: Yeah . grad d: So a more general thing than " discussed admission fee " , could be I I 'm just wondering whether the context , the background context of the discourse might be I don't know , if there 's a way to define it or maybe you know generalize it some way , there might be other cues that , say , in the last few utterances there has been something that has strongly associated with say one of the particular modes , I don't know if that might be grad a: I think we grad d: and and into that node would be various various things that that could have specifically come up . grad a: I think a a sort of general strategy here You know , this is this is excellent because it gets you thinking along these terms is that maybe we ob we could observe a couple of discourse phenomena such as the admission fee , and something else and something else , that happened in the discourse before . And maybe there are two So maybe this could be sort of a separate region of the net , which has two has it 's own middle layer . Maybe this , you know , has some kind of , funky thing that di if this and this may influence these hidden nodes of the discourse which is maybe something that is , a more general version of the actual phenomenon that you can observe . So things that point towards grad b: So instead of single node , for like , if they said the word " admission fee " grad d: Exactly . grad b: " admission fee " , or maybe , you know , " how much to enter " grad d: Yeah . Yeah ? And then maybe there are some discourse acts if they happened before , it 's more for a cue that the person actually wants to get somewhere else and that you are in a in a in a route , sort of proceeding past these things , so this would be just something that where you want to pass it . ? Is that it ? However these are of course then the the nodes , the observed nodes , for your middle layer . So this again points to " final destination " , " doing business " , " tourist hurry " and so forth . we have a whole region " in a e grad d: That 's a whole set of discourse related cues to your middle layer . grad d: Right ? grad a: So e because at the end the more we add , you know , the more spider - web - ish it 's going to become in the middle and the more of hand editing . They ra may have there own hidden layer that points to some of the the real hidden layer , or the general hidden layer . grad a: And the same we will be able to do for syntactic information , the verbs used , the object types used , modifiers . grad b: One thing that 's kind of been bugging me when I more I look at this is that the I guess , the fact that the there 's a complete separation between the observed features and in the output . grad b: For instance if the discourse does grad d: What do you mean by that ? grad b: well for instance , the " discourse admission fee " node seems like it should point directly to the grad d: - huh . grad b: or increase the probability of " enter directly " versus " going there via tourist " . Like we could add a node like do they want to enter it , which is affected by admission fee and by whether it 's closed and by whether it has a door . And if it if you do it If you could connect it too hard you may get such phenomenon that like " So how much has it cost to enter ? " and the answer is two hundred fifty dollars , and then the persons says " Yeah I want to see it . " Yeah ? meaning " It 's way out of my budget " grad b: There are places in Germany where it costs two hundred fifty dollars to enter ? grad a: nothing comes to mind . But i you know , i we can Something Somebody can have discussed the admission fee and u the answer is s if we , you know , still , based on that result is never going to enter that building . So the discourse refers to " admission fee " but it just turns out that they change their mind in the middle of the discourse . you have to have some notion of not just there 's a there 's change across several turns of discourse grad b: Right . grad d: so I don't know how if any of this was discussed but how i if it all this is going to interact with whatever general , other other discourse processing that might be happen . grad b: What sort of discourse processing is are the How much is built into SmartKom and grad a: It works like this . The first thing we get is that already the intention is sort of t They tried to figure out the intention , right ? simply by parsing it . And this m won't differentiate between all modes , yeah ? but at least it 'll tell us " OK here we have something that somebody that wants to go someplace , now it 's up for us to figure out what kind of going there is is is happening , and , if the discourse takes a couple of turns before everything all the information is needed , what happens is you know the parser parses it and then it 's handed on to the discourse history which is , o one of the most elaborate elaborate modules . It 's it 's actually the the whole memory of the entire system , that knows what wh who said what , which was what was presented . It helps an an anaphora resolution and it and it fills in all the structures that are omitted , so , because you say " OK , how can I get to the castle ? " Oh , how how much is it ? " and " yeah I would like to g let 's do it " and so forth . So even without an a ana anaphora somebody has to make sure that information we had earlier on is still here . so whenever the , person is not actually rejecting what happened before , so as in " No I really don't want to see that movie . I 'd rather stay home and watch TV " What movie was selected in what cinema in what town is is going to be sort of added into the disc into the representations every di at each dialogue step , by the discourse model discourse model , Yeah , that 's what it 's called . and , it does some help in the anaphora resolution and it also helps in coordinating the gesture screen issues . So a person pointing to something on the screen , you know , the discourse model actually stores what was presented at what location on the s on the screen grad b: . grad a: so it 's a it 's a rather huge huge thing but we can sort of It has a very clear interface . We can query it whether admission fees were discussed in the last turn and and the turn before that or you know how deep we want to search grad b: OK . How deep do we want to sear , you know ? but we should try to keep in mind that , you know , we 're doing this sort of for research , so we we should find a limit that 's reasonable and not go , you know , all the way back to Adam and Eve . You know , did that person ever discuss admissions fee fees in his entire life ? And the dialogues are pretty pretty you know concise and Anyway . grad d: So one thing that might be helpful which is implicit in the use of " admission fee discussion " as a cue for entry , is thinking about the plans that various people might have . This person is , finding out information about this thing in order to go in as a tourist or finding out how to get to this place in order to do business . , because then anything that 's a cue for one of the steps would be slight evidence for that overall plan . They 're in in non in sort of more traditional AI kinds of plan recognition things you sort of have you know , some idea at each turn of agent doing something , " OK , wha what plans is this a consistent with ? " and then get s some more information and then you see " here 's a sequence that this sort of roughly fits into " . grad d: I I don't know how you know you 'd have to figure out what knowl what knowledge representation would work for that . grad a: You know ? and it it 's fifty steps , grad d: grad a: huh ? just for buying a ticket at a ticket counter , you know , and and maybe that 's helpful to look at it to look at those . W when we talked we had the example , you know , of you being a s a person on a ticket counter working at railway station and somebody r runs up to you with a suitcase in his hands , says New York and you say Track seven , huh ? And it 's because you know that that person actually is following , you know You execute a whole plan of going through a hundred and fifty steps , you know , without any information other than " New York " , huh ? inferring everything from the context . , even though there is probably no train from here to New York , right ? grad d: Mmm . Right ? Is that t San Francisco , Chicago ? grad b: I think grad a: Is that possible ? grad b: One time I saw a report on trains , and I think there is a l I don't know if I thought there was a line that went from somewhere , maybe it was Sacramento to Chicago , grad a: grad d: The Transcontinental Railroad , doesn't that ring a bell ? grad b: Yeah but I don't know if it 's still grad d: I think it has to exist somewhere . grad a: Well it never went all the way , right ? you always had to change trains at Omaha , grad d: Well most of the way . grad a: right ? One track ended there and the other one started at five meters away from that grad d: . grad a: yeah ? Has anybody ever been on an Amtrak ? grad d: I have . grad c: What ? Why ? grad b: I just They seem to have a lot of accidents on the Amtrak . grad a: But you know , I don't know whether it 's which ones are safer , you know , statistically . Yeah , they 're Yeah , it 's way better grad a: yeah I used Amtrak quite a bit on the east coast and I was surprised . grad a: ? grad c: I I want to see what it does with " landmark - iness " . grad d: So by the way tha that structure that Robert drew on the board was like more , cue - type - based , right , here 's like we 're gonna segment off a bit of stuff that comes from discourse and then some of the things we 're talking about here are more you know , we mentioned maybe if they talk about , I don't know , entering or som you know like they might be more task - based . grad d: So I I don't know if there There 's obviously some m more than one way of organizing the variables into something grad a: I think that What you guys did is really nicely sketching out different tasks , and maybe some of their conditions . grad a: One task is more likely you 're in a hurry when you do that kind of s doing business , grad d: grad a: and and less in a hurry when you 're a tourist tourists may have never have final destinations , you know because they are eternally traveling around so maybe what what what happened what might happen is that we do get this sort of task - based middle layer , grad d: grad a: and then we 'll get these sub - middle layers , that are more cue - based . So , I suggest w to for to proceed with this in in the sense that maybe throughout this week the three of us will will talk some more about maybe segmenting off different regions , and we make up some some toy a observable " nodes " is that what th grad b: Refined y re just refine the grad a: What 's the technical term ? grad c: OK . For which ? grad a: For the nodes that are observable ? The " outer layer " ? grad c: Just observable nodes , grad b: The features , grad c: evidence nodes ? grad b: I don't know , whatever you grad a: Feature ma make up some features for those Identify four regions , grad c: Yeah . grad a: maybe make up some features for each region and and , and middle layer for those . And then these should then connect somehow to the more plan - based deep space grad c: Yeah . grad c: Yeah , this is totally like The probabilities and all are completely ad - hoc . but , they 're even like like , close to the end we were like , you know we were like really ad - hoc . grad c: Right ? Cuz if it 's like , If it 's four things coming in , right ? And , say , some of them have like three possibilities and all that . So you 're thinking like like a hundred and forty four or something possible things numbers to enter , grad d: And That 's terrible . grad b: Some of them are completely absurd too , like they want to enter , but it 's closed , grad d: That 's Well grad b: it 's night time , you know there are tourists and all this weird stuff happens at the line up and you 're like grad c: Yeah , the only like possible interpretation is that they are like come here just to rob the museum or something to that effect . grad d: In which case you 're supposed to alert the authorities , and see appropriate action . Yeah , another thing to do , is also to , I guess to ask around people about other Bayes - net packages . grad d: Sorry , Wednesday , grad b: Who 's talking on Wednesday ? grad c: Maybe we can ask him about it . grad b: I haven't J Jerry never sent out a sent out an email , did he , ever ? grad c: No . grad d: Ben ? grad a: Ben , then , grad d: I think it 's Ben actually , grad a: Ben . I actually , have , also we can , start looking at the SmartKom tables and I will grad b: Right . grad b: Do you want to trade ? grad a: no I I actually made a mistake because it it fell asleep and when Linux falls asleep on my machine it 's it doesn't wake up ever , so I had to reboot grad d: Oh , no . grad a: And if I reboot without a network , I will not be able to start SmartKom , because I need to have a network . grad b: grad a: So we 'll do that t maybe grad c: But . But once you start sart start SmartKom you can be on You don't have to be on a network anymore . grad b: Why does SmartKom need a network ? grad a: it looks up some stuff that , you know , is is that is in the written by the operating system only if it if you get a DHCP request , so it you know , my computer does not know its IP address , you know ? grad b: Ah . grad a: And I don't have an IP address , they can't look up they don't know who localhost is , and so forth and so forth . grad a: She 's willing to do it , meaning be the wizard for the data collection , also maybe transcribe a little bit , if she has to , but also recruiting subjects , organizing them , and so forth . Jerry however suggested that we should have a trial run with her , see whether she can actually do all the spontaneous , eloquent and creativeness that we expect of the wizard . And I talked to Liz about this and it looks as if Friday afternoon will be the time when we have a first trial run for the data . grad c: So who would be the subject of this trial run ? grad a: Pardon me ? grad c: Who Will there be a Is one Is you one of you gonna be the subject ? Like are you grad a: Liz also volunteered to be the first subject , which I think might be even better than us guys . grad a: If we do need her for the technical stuff , then of course one of you has to sort of jump in . grad c: Well I just figured it has to be someone who 's , familiar enough with the data to cause problems for the wizard , so we can , see if they 're you know good . that 's what we wanna check , right ? grad a: grad d: Well , in this case it 's a p it 's a sort of testing of the wizard rather than of the subject . grad c: Isn't that what it is ? grad d: It 's grad a: yes w we we would like to test the wizard , but you know , if we take a subject that is completely unfamiliar with the task , or any of the set up , we get a more realistic grad c: I guess that would be reasonable . grad d: I 'm sure if we , You think there 's a chance we might need Liz for , whatever , the technical side of things ? I 'm sure we can get other people around who don't know anything , if we want another subject . So , is it a experimental setup for the , data collection totally ready determined ? grad b: I like that . grad a: I think it 's it 's it 's experimental setup u on the technical issue yes , except we st I think we still need a recording device for the wizard , just a tape recorder that 's running in a room . grad a: But in terms of specifying the scenario , we 've gotten a little further grad d: grad a: but we wanted to wait until we know who is the wizard , and have the wizard partake in the ultimate sort of definition probe . So so if if on Friday it turns out that she really likes it and and we really like her , then nothing should stop us from sitting down next week and getting all the details completely figured out . So the ideal task , will have whatever I don't know how much the structure of the evolving Bayes - net will af affect Like we wanna we wanna be able to collect as much of the variables that are needed for that , grad a: Mmm - yea - some . grad d: right ? in the course of the task ? Well not all of them but you know . grad a: Bu - e e e I 'm even This this Tango , Enter , Vista is sort of , itself , an ad - hoc scenario . So we wanted just to collect data , to get that that that elicits more , that elicits richer language . grad a: And we actually did not want to constrain it too much , grad d: And then maybe we 'll discover the phenomenon the phenomena that we want to solve , you know , with whatever engine we we come up with . So this this this is a parallel track , you know , there they hopefully meet , grad d: OK . grad a: but since grad d: It could it could be used for not just this task . grad a: It should tell us , you know , what kind of phenomenon could occur , it should tell us also maybe something about the difference between people who think they speak to a computer versus people who think they speak to a human being grad d: So it may get us some more information on the human - machine pragmatics , that no one knows anything about , as of yesterday . And secondly , now that of course we have sort of started to lick blood with this , and especially since Johno can't stop Tango - ing , we may actually include , you know , those those intentions . So now I think we should maybe have at least one navigational task with with sort of explicit grad d: grad a: not ex it 's implicit that the person wants to enter , grad d: grad a: and maybe some task where it 's more or less explicit that the person wants to take a picture , grad d: grad a: Whereas , you know , if we 'd just get data we 'd never know what they actually wanted , we 'd get no cues . grad b: So is this the official end of the meeting now ? grad c: Yep . grad c: So what 's " Economics , the fallacy " ? grad a: Ma grad b: I just randomly label things . grad c: Oh , really ? grad a: Maybe we ought to switch off these things before we continue
|
Bed004
|
grad d: I tried to go for the EE Cummings sort of feeling , but grad a: Three three six zero zero . grad a: You ever seen " So I married an axe murderer " ? grad c: parts of it . grad a: There 's a part wh there 's parts when he 's doing beat poetry . is when he 's he works in a coffee shop , in San Francisco , and he 's sitting there on this couch and they bring him this massive cup of espresso , and he 's like " excuse me I ordered the large espresso ? " grad d: . grad a: Wait do are y So you 're trying to decide who 's the best taster of tiramisu ? grad d: No ? . There was a a a fierce argument that broke out over whose tiramisu might be the best and so we decided to have a contest where those people who claim to make good tiramisu make them , grad a: Ah . grad d: and then we got a panel of impartial judges that will taste do a blind taste and then vote . grad a: Seems like Seems like you could put a s magic special ingredient in , so that everyone know which one was yours . Well , I was thinking if y you guys have plans for Sunday ? We 're we 're not it 's probably going to be this Sunday , but we 're sort of working with the weather here because we also want to combine it with some barbecue activity where we just fire it up and what whoever brings whatever you know , can throw it on there . grad a: Well , I 'm going back to visit my parents this weekend , so , I 'll be out of town . grad d: So you 're going to the west Bay then ? No , grad a: No , the South Bay , grad d: south Bay ? grad a: yeah . Wonder if these things ever emit a very , like , piercing screech right in your ear ? grad d: They are gonna get more comfortable headsets . I actually , even though Liz was kind enough to offer to be the first subject , I sort of felt that she knew too much , so I asked Litonya . grad d: So , this is what she saw as part of as for instr introduction , this is what she had to read aloud . , that was really difficult for her and grad c: Because of l all the names , you mean ? grad d: The names and this was the first three tasks she had to to master after she called the system , and then of course the system broke down , and those were the l I should say the system was supposed to break down and then these were the remaining three tasks that she was going to solve , with a human . And now comes the This is the phone - in phase of grad c: Wait , can I I have a question . So there 's no system , right ? Like , there was a wizard for both both parts , is this right ? grad d: Yeah . grad d: One time , pretending to be a system , one time , to pretending to be a human , which is actually not pretending . Isn't this kind of obvious when it says " OK now you 're talking to a human " and then the human has the same voice ? grad d: No no no . And the wizard sometimes will not be audible , Because she was actually they there was some lapse in the wireless , we have to move her closer . grad a: Is she mispronouncing " Anlage " ? Is it " Anlaga " or " Anlunga " grad d: They 're mispronouncing everything , grad a: OK . Well , if we we professor b: So , are are you trying to record this meeting ? grad d: There was a strange reflex . , that was already anticipated by some people suggested that if we just have bullets here , they 're gonna not they 're subjects are probably not gonna going to follow the order . professor b: S so if you just number them " one " , " two " , " three " it 's grad d: Yeah , and make it sort of clear in the professor b: OK . That is something that Fey actually thought of a in the last second that sh the system should introduce itself , when it 's called . grad d: And , another suggestion , by Liz , was that we , through subjects , switch the tasks . So when when they have task - one with the computer , the next person should have task - one with a human , and so forth . , we have to refine the tasks more and more , which of course we haven't done at all , so far , in order to avoid this rephrasing , so where , even though w we don't tell the person " ask blah - blah - blah - blah - blah " they still try , or at least Litonya tried to repeat as much of that text as possible . grad d: And my suggestion is of course we we keep the wizard , because I think she did a wonderful job , professor b: Great . grad d: in the sense that she responded quite nicely to things that were not asked for , " How much is a t a bus ticket and a transfer " so this is gonna happen all the time , we d you can never be sure . grad a: I wasn't wasn't sure whether wizard was the correct term for " not a man " . grad c: There 's no female equivalent of grad d: But grad a: Are you sure ? grad c: No , I don't know . grad d: Well , there is witch and warlock , grad a: Yeah , that 's so @ @ . grad c: Yeah , that 's what I was thinking , but grad d: and professor b: Right . And So , some some work needs to be done , but I think we can And this , and in case no you hadn't seen it , this is what Litonya looked at during the while taking the while partaking in the data collection . , do you know exactly how to do that , or is Lila , you know what exactly do we do to to put her on the payroll in some way ? grad d: I 'm completely clueless , but I 'm willing to learn . So anyway , grad d: N professor b: So why don't you ask Lila and see what she says about you know exactly what we do for someone in th grad d: Student - type worker , professor b: Well , yeah she 's un she 's not a a student , grad d: or ? professor b: she just graduated but anyway . professor b: So i if Yeah , I agree , she sounded fine , she a actually was , more , present and stuff than than she was in conversation , so she did a better job than I would have guessed from just talking to her . grad d: This is sort of what I gave her , so this is for example h how to get to the student prison , professor b: Yeah . grad d: and I didn't even spell it out here and in some cases I I spelled it out a little bit more thoroughly , professor b: Right . grad d: this is the information on on the low sunken castle , and the amphitheater that never came up , and , so i if we give her even more , instruments to work with I think the results are gonna be even better . professor b: Oh , yeah , and then of course as she does it she 'll she 'll learn @ @ . And also if she 's willing to take on the job of organizing all those subjects and stuff that would be wonderful . professor b: And , she 's actually she 's going to graduate school in a kind of an experimental paradigm , so I think this is all just fine in terms of h her learning things she 's gonna need to know , to do her career . professor b: So , I my guess is she 'll be r r quite happy to take on that job . grad d: And I told her that we gonna figure out a meeting time in the near future to refine the tasks and s look for the potential sources to find people . She also agrees that you know if it 's all just gonna be students the data is gonna be less valuable because of that so . professor b: Well , as I say there is this s set of people next door , it 's not hard to grad d: We 're already Yeah . professor b: grad d: However , we may run into a problem with a reading task there . We could talk to the people who run it and see if they have a way that they could easily tell people that there 's a task , pays ten bucks or something , grad d: Now , I signed us up for the Wednesday slot , and part of what we should do is this . professor b: So , my idea on that was , partly we 'll talk about system stuff for the computer scientists , but partly I did want it to get the linguists involved in some of this issue about what the task is and all you know , what the dialogue is , and what 's going on linguistically , because to the extent that we can get them contributing , that will be good . professor b: maybe we can get some of the linguists sufficiently interested that they 'll help us with it , other linguists , if you 're a linguist , but in any case , grad d: Yep . So my idea on on Wednesday is partly to you , what you did today would i is just fine . You just do " this is what we did , and here 's the thing , and here 's s some of the dialogue and and so forth . " But then , the other thing of course is we should give the computer scientists some idea of of what 's going on with the system design , and where we think the belief - nets fit in and where the pieces are and stuff like that . So , I don't I don't think it 's worth a lot of work , particularly on your part , to to to make a big presentation . I don't think you should you don't have to make any new PowerPoint or anything . The other two things is we 've can have Johno tell us a little about this professor b: Great . grad d: and we also have a l little bit on the interface , M - three - L enhancement , and then that was it , I think . grad a: So , what I did for this this is , a pedagogical belief - net because I was I I took I tried to conceptually do what you were talking about with the nodes that you could expand out so what I did was I took I made these dummy nodes called Trajector - In and Trajector - Out that would isolate the things related to the trajector . And then I did similar things for our our net to with the context and the discourse and whatnot , so we could sort of isolate them or whatever in terms of the the top layer . Let 's go Slide all the way up so we see what the p the p very bottom looks like , or is that it ? grad a: Yeah , there 's just one more node and it says " Mode " which is the decision between the grad d: Yeah . grad a: So basically all I did was I took the last belief - net professor b: So grad a: and I grouped things according to what how I thought they would fit in to image schemas that would be related . And the two that I came up with were Trajector - landmark and then Source - path - goal as initial ones . grad a: And then I said well , the trajector would be the person in this case probably . grad a: you know , we have we have the concept of what their intention was , whether they were trying to tour or do business or whatever , professor b: Right . And then in terms of the source , the things the only things that we had on there I believe were whether Oh actually , I kind of , I might have added these cuz I don't think we talked too much about the source in the old one but whether the where I 'm currently at is a landmark might have a bearing on whether grad d: And " usefulness " is basi basically means is that an institutional facility like a town hall or something like that that 's not something that you 'd visit for tourist 's tourism 's sake or whatever . " Travel constraints " would be something like you know , maybe they said they can they only wanna take a bus or something like that , right ? And then those are somewhat related to the path , professor b: grad a: so that would determine whether we 'd could take we would be telling them to go to the bus stop or versus walking there directly . Similar things as the source except they also added whether the entity was closed and whether they have somehow marked that is was the final destination . , and then if you go up , Robert , Yeah , so , in terms of Context , what we had currently said was whether they were a businessman or a tourist of some other person . , Discourse was related to whether they had asked about open hours or whether they asked about where the entrance was or the admission fee , or something along those lines . grad a: Prosody I don't really I 'm not really sure what prosody means , in this context , so I just made up you know whether whether what they say is or h how they say it is is that . grad a: the Parse would be what verb they chose , and then maybe how they modified it , in the sense of whether they said " I need to get there quickly " or whatever . grad a: And , in terms of World Knowledge , this would just basically be like opening and closing times of things , the time of day it is , and whatnot . grad d: What 's " tourbook " ? grad a: Tourbook ? That would be , I don't know , the " landmark - iness " of things , grad d: So let me see if I can ask grad a: Well , this is not a working Bayes - net . No , I understand that , but but So , what Let 's slide back up again and see start at the at the bottom and Oop - bo - doop - boop - boop . So , you could imagine w , go ahead , you were about to go up there and point to something . grad a: I I 'd No , I was gonna wait until professor b: Oh , OK . So , so if you if we made if we wanted to make it into a a real Bayes - net , that is , you know , with fill you know , actually f , fill it @ @ in , then grad a: So we 'd have to get rid of this and connect these things directly to the Mode . And and Bhaskara and I was talking about this a little earlier today is , if we just do this , we could wind up with a huge , combinatoric input to the Mode thing . And grad a: Well I oh yeah , I unders I understand that , I just it 's hard for me to imagine how he could get around that . Let me just mention something that I don't want to pursue today which is there are technical ways of doing it , I I slipped a paper to Bhaskara and about Noisy - OR 's and Noisy - MAXes and there 're ways to sort of back off on the purity of your Bayes - net - edness . If you co you could ima and I now I don't know that any of those actually apply in this case , but there is some technology you could try to apply . grad a: So it 's possible that we could do something like a summary node of some sort that OK . grad a: So in that case , the sum we 'd have we , these wouldn't be the summary nodes . We 'd have the summary nodes like where the things were I guess maybe if thi if things were related to business or some other professor b: Yeah . professor b: So what I was gonna say is is maybe a good at this point is to try to informally , not necessarily in th in this meeting , but to try to informally think about what the decision variables are . So , if you have some bottom line decision about which mode , you know , what are the most relevant things . professor b: And the other trick , which is not a technical trick , it 's kind of a knowledge engineering trick , is to make the n each node sufficiently narrow that you don't get this combinatorics . So that if you decided that you could characterize the decision as a trade - off between three factors , whatever they may be , OK ? then you could say " Aha , let 's have these three factors " , OK ? and maybe a binary version f for each , or some relatively compact decision node just above the final one . professor b: And then the question would be if if those are the things that you care about , can you make a relatively compact way of getting from the various inputs to the things you care about . So that y so that , you know , you can sort of try to do a knowledge engineering thing grad a: OK . professor b: given that we 're not gonna screw with the technology and just always use sort of orthodox Bayes - nets , then we have a knowledge engineering little problem of how do we do that . and grad a: So what I kind of need to do is to take this one and the old one and merge them together ? professor b: " Eh - eh - eh . , so , Robert has thought about this problem f for a long time , cuz he 's had these examples kicking around , so he may have some good intuition about you know , what are the crucial things . professor b: and , I understand where this the this is a way of playing with this abs Source - path - goal trajector exp abstraction and and sort of sh displaying it in a particular way . professor b: I don't think our friends on Wednesday are going to be able to Well , maybe they will . This is sort of th the second version and I I I look at this maybe just as a , you know , a a whatever , UML diagram or , you know , as just a screen shot , not really as a Bayes - net as John Johno said . grad a: We could actually , y yeah draw it in a different way , in the sense that it would make it more abstract . But the the the nice thing is that you know , it just is a is a visual aid for thinking about these things which has comple clearly have to be specified m more carefully professor b: Alright , well , le let me think about this some more , grad d: and professor b: and see if we can find a way to present this to this linguists group that that is helpful to them . grad d: ultimately we we may w w we regard this as sort of an exercise in in thinking about the problem and maybe a first version of a module , if you wanna call it that , that you can ask , that you can give input and it it 'll throw the dice for you , throw the die for you , because I integrated this into the existing SmartKom system in in the same way as much the same way we can sort of have this this thing . So if this is what M - three - L will look like and what it 'll give us , And a very simple thing . We have an action that he wants to go from somewhere , which is some type of object , to someplace . grad d: And this these this changed now only , It 's doing it twice now because it already did it once . , we 'll add some action type , which in this case is " Approach " and could be , you know , more refined in many ways . grad d: Or we can have something where the goal is a public place and it will give us then of course an action type of the type " Enter " . So this is just based on this one , on this one feature , and that 's that 's about all you can do . And so in the f if this pla if the object type here is is a m is a landmark , of course it 'll be " Vista " . And this is about as much as we can do if we don't w if we want to avoid a huge combinatorial explosion where we specify " OK , if it 's this and this but that is not the case " , and so forth , it just gets really really messy . You 're you 're grad d: ? professor b: It was much too quick for me . So , I I do understand that you can take the M - three - L and add not and it w and you need to do this , for sure , we have to add , you know , not too much about object types and stuff , and what I think you did is add some rules of the style that are already there that say " If it 's of type " Landmark " , then you take you 're gonna take a picture of it . Ev - every landmark you take a picture of , grad d: Every public place you enter , and statue you want to go as near as possible . grad d: W professor b: that 's a that 's another kind of baseline case , that 's another sort of thing " OK , here 's a another kind of minimal way of tackling this " . Add extra properties , a deterministic rule for every property you have an action , " pppt ! " You do that . , then the question would be Now , if that 's all you 're doing , then you can get the types from the ontology , OK ? because that 's all you 're all you 're using is this type the types in the ontology and you 're done . grad d: ? professor b: Right ? So we don't we don't use the discourse , we don't use the context , we don't do any of those things . professor b: Alright , but that 's but that 's OK , and it it 's again a kind of one minimal extension of the existing things . And that 's something the SmartKom people themselves would they 'd say " Sure , that 's no problem you know , no problem to add types to the ont " Right ? grad d: Yeah . And this is just in order to exemplify what what we can do very , very easily is , we have this this silly interface and we have the rules that are as banal as of we just saw , and we have our content . grad d: Now , the content I whi which is sort of what what we see here , which is sort of the Vista , Schema , Source , Path , Goal , whatever . grad d: This will be a job to find ways of writing down Image schema , X - schema , constructions , in some some form , and have this be in a in a in the content , loosely called " Constructicon " . And and here is exactly where what 's gonna be replaced with our Bayes - net , which is exactly getting the input feeding into here . This decides whether it 's an whether action the the Enter , the Vista , or the whatever professor b: " approach " , you called it , I think this time . This is so what we 'd be generating would be a reference to a semantic like parameters for the for the X - schema ? professor b: For for for Yes . So that that i if you had the generalized " Go " X - schema and you wanted to specialize it to these three ones , then you would have to supply the parameters . professor b: And then , although we haven't worried about this yet , you might wanna worry about something that would go to the GIS and use that to actually get you know , detailed route planning . professor b: But , presumably that that that functionality 's there when when we grad a: So the immediate problem is just deciding w which grad d: Aspects of the X - schema to add . professor b: Yeah , so the pro The immediate problem is is back t t to what you were what you are doing with the belief - net . professor b: You know , what are we going to use to make this decision grad a: Right and then , once we 've made the decision , how do we put that into the content ? professor b: Yeah . professor b: The harder problem is we decide what we want to use , how are we gonna get it ? And that the the that 's the hardest problem . So , the hardest problem is how are you going to get this information from some combination of the what the person says and the context and the ontology . The h So , I think that 's the hardest problem at the moment is is grad a: OK . , and that 's so , getting back to here , we have a d a technical problem with the belief - nets that we we don't want all the com grad a: There 's just too many factors right now . professor b: So we wanna think about which ones we really care about and what they really most depend on , and can we c you know , clean this this up to the point where it grad a: So what we really wanna do i cuz this is really just the three layer net , we wanna b make it expand it out into more layers basically ? professor b: Right . , it 's true that the way you have this , a lot of the times you have what you 're having is the values rather than the variable . So instead of in instead it should really be just be " intention " as a node instead of " intention business " or " intention tour " . professor b: OK ? So you Yeah , right , and then it would have values , " Tour " , " Business " , or " Hurried " . professor b: But then but i it still some knowledge design to do , about i how do you wanna break this up , what really matters . grad a: I think what was going through my mind when I did it was someone could both have a business intention and a touring intention and the probabilities of both of them happening at the same time professor b: Well , you you could do that . And it 's perfectly OK to insist that that , you know , th , they add up to one , but that there 's that that it doesn't have to be one zero zero . So you could have the conditional p So the each of these things is gonna be a a a probability . So whenever there 's a choice , so like landmark - ness and usefulness , grad a: Well , see I don't think those would be mutually professor b: OK grad a: it seems like something could both be professor b: Absolutely right . professor b: And so that you might want to then have those b Th - Then they may have to be separate . professor b: So that 's but again , this is this is the sort of knowledge design you have to go through . It 's you know , it 's great is is , you know , as one step toward toward where we wanna go . grad d: Also it strikes me that we we m may want to approach the point where we can sort of try to find a , a specification for some interface , here that takes the normal M - three - L , looks at it . Then we discussed in our pre - edu EDU meeting how to ask the ontology , what to ask the ontology the fact that we can pretend we have one , make a dummy until we get the real one , and so we we may wanna decide we can do this from here , but we also could do it you know if we have a a a belief - net interface . But this information is just M - three - L , and then we want to look up some more stuff in the ontology and we want to look up some more stuff in the maybe we want to ask the real world , maybe you want to look something up in the GRS , but also we definitely want to look up in the dialogue history some s some stuff . Based on we we have I was just made some examples from the ontology and so we have for example some information there that the town hall is both a a a building and it has doors and stuff like this , but it is also an institution , so it has a mayor and so forth and so forth and we get relations out of it and once we have them , we can use that information to look in the dialogue history , " were any of these things that that are part of the town hall as an institution mentioned ? " , professor b: grad d: " were any of these that make the town hall a building mentioned ? " , grad c: Right . So this may be a a sort of a process of two to three steps before we get our vector , that we feed into the belief - net , professor b: Yeah . grad d: and then professor b: There will be rules , but they aren't rules that come to final decisions , they 're rules that gather information for a decision process . So they 'll they presumably there 'll be a thread or process or something that " Agent " , yeah , " Agent " , whatever you wan wanna say , yeah , that is rule - driven , and can can can do things like that . And there 's an issue about whether there will be that 'll be the same agent and the one that then goes off and carries out the decision , so it probably will . My guess is it 'll be the same basic agent that can go off and get information , run it through a a c this belief - net that turn a crank in the belief - net , that 'll come out with s more another vector , OK , which can then be applied at what we would call the simulation or action end . So on once you pull that out , it could be that that says " Ah ! Now that we know that we gonna go ask the ontology something else . " OK ? Now that we know that it 's a bus trip , OK ? we didn't We didn't need to know beforehand , how long the bus trip takes or whatever , but but now that we know that 's the way it 's coming out then we gotta go find out more . So this is actually , s if if we were to build something that is , and , I had one more thing , the it needs to do Yeah . I think we I I can come up with a a code for a module that we call the " cognitive dispatcher " , which does nothing , professor b: OK . grad d: but it looks of complect object trees and decides how are there parts missing that need to be filled out , there 's this is maybe something that this module can do , something that this module can do and then collect sub - objects and then recombine them and put them together . So maybe this is actually some some useful tool that we can use to rewrite it , and get this part , professor b: Oh , OK . In particular see what we 'd like to do , and and this has been implicit in the discussion , is to do this in such a way that you get a lot of re - use . What you 're trying to get out of this deep co cognitive linguistics is the fact that w if you know about source source , paths and goals , and nnn all this sort of stuff , that a lot of this is the same , for different tasks . And that there 's there 's some some important generalities that you 're getting , so that you don't take each and every one of these tasks and hafta re - do it . grad d: There 're no primitives upon which professor b: u u What are the primitives , and how do you break this grad d: yeah . professor b: So I y I 'm just just there saying eee well you I know how to do any individual case , right ? but I don't yet see what 's the really interesting question is can you use deep cognitive linguistics to get powerful generalizations . professor b: grad d: Maybe we sho should we a add then the " what 's this ? " domain ? N , we have to " how do I get to X " . Then we also have the " what 's this ? " domain , where we get some slightly different professor b: Could . grad d: Johno , actually , does not allow us to call them " intentions " anymore . professor b: Well , I I don't like the term either , so I have n i i i y w i i It grad d: But , I 'm sure the " what 's this ? " questions also create some interesting X - schema aspects . I 'm not a I 'm not op particularly opposed to adding that or any other task , grad d: So . professor b: I 'm just saying that I 'm gonna hafta do some sort of first principles thinking about this . Well , no the Bayes the Bayes - nets The Bayes - nets will be dec specific for each decision . But what I 'd like to be able to do is to have the way that you extract properties , that will go into different Bayes - nets , be the general . So that if you have sources , you have trajectors and stuff like that , and there 's a language for talking about trajectors , you shouldn't have to do that differently for going to something , than for circling it , for telling someone else how to go there , grad d: Getting out of professor b: whatever it is . So that that , the the decision processes are gonna be different What you 'd really like of course is the same thing you 'd always like which is that you have a kind of intermediate representation which looks the same o over a bunch of inputs and a bunch of outputs . So all sorts of different tasks and all sorts of different ways of expressing them use a lot of the same mechanism for pulling out what are the fundamental things going on . And pushing it one step further , when you get to construction grammar and stuff , what you 'd like to be able to do is say you have this parser which is much fancier than the parser that comes with SmartKom , i that that actually uses constructions and is able to tell from this construction that there 's something about the intent you know , the actual what people wanna do or what they 're referring to and stuff , in independent of whether it about what is this or where is it or something , that you could tell from the construction , you could pull out deep semantic information which you 're gonna use in a general way . You might be able to to say that this i this is the kind of construction in which the there 's Let 's say there 's a cont there the the land the construction implies the there 's a con this thing is being viewed as a container . So just from this local construction you know that you 're gonna hafta treat it as a container you might as well go off and get that information . So if you say " how do I get into the castle " OK , then Or , you know , " what is there in the castle " or so there 's all sorts of things you might ask that involve the castle as a container and you 'd like to have this orthogonal so that anytime the castle 's referred to as a container , you crank up the appropriate stuff . professor b: Alright , so that 's that 's the that 's the thesis level grad d: professor b: grad d: It 's unfortunate also that English has sort of got rid of most of its spatial adverbs because they 're really fancy then , in in for these kinds of analysis . professor b: Well , you have prepositional phrases that grad d: Yeah , but they 're they 're easier for parsers . grad d: Parsers can pick those up but but the with the spatial adverbs , they have a tough time . Oh yeah , b But an architecture like this would also enable us maybe to to throw this away and and replace it with something else , or whatever , so that we have so that this is sort of the representational formats we 're we 're we 're talking about that are independent of the problem , that generalize over those problems , and are oh , t of a higher quality than an any actual whatever belief - net , or " X " that we may use for the decision making , ultimately . So , are we gonna be meeting here from now on ? I 'm I 'm happy to do that . We we had talked about it , cuz you have th th the display and everything , that seems fine . grad d: so far I think it was nice as a visual aid for some things and and professor b: Oh yeah . No I I think it 's worth it to ass to meet here to bring this , and assume that something may come up that we wanna look at . grad d: Yeah ? The , she w she was definitely good in the sense that she she showed us some of the weaknesses professor b: Right . grad d: and also the the fact that she was a real subject you know , is is professor b: Right . Yeah , and and and yeah and and she took it seriously and stuff l No , it was great . grad d: So I think that , w Looking just looking at this data , listening to it , what can we get out of it in terms of our problem , for example , is , you know , she actually m said you know , she never s just spoke about entering , she just wanted to get someplace , and she said for buying stuff . grad d: and in the other case , where she wanted to look at the stuff at the graffiti , also , of course , not in the sentence " How do you get there ? " was pretty standard . Nuh ? except that there was a nice anaphora , you know , for pointing at what she talked about before , and there she was talking about looking at pictures that are painted inside a wall on walls , so grad c: Right . , because graffiti is usually found on the outside and not on the inside , grad c: Yeah
|
Bed005
| "So , before we get started with the , technical part , I just want to review what I think is happen(...TRUNCATED) |
Bed006
| "I think we 've met before , like , I remember talking to you about Aspect or something like that at(...TRUNCATED) |
Bed008
| "grad a: Alright , so I 'm - I should read all of these numbers ? professor b: OK . professor b: OK (...TRUNCATED) |
Bed009
| "professor f: So the what w we h have been doing i they would like us all to read these digits . gra(...TRUNCATED) |
Bed010
| "grad b: grad d: How many batteries do you go through ? grad b: Thank you . My suggestion is that R(...TRUNCATED) |
Bed011
| "grad c: Now can you give me the remote T ? professor d: OK , so Eva , co could you read your numb(...TRUNCATED) |
End of preview. Expand
in Data Studio
MSRS: Evaluating Multi-Source Retrieval-Augmented Generation
This paper introduces a scalable framework for constructing evaluation benchmarks that challenge RAG systems to integrate information across distinct sources and generate long-form responses. Using our framework, we build two new benchmarks on Multi-Source Retrieval and Synthesis: MSRS-Story and MSRS-Meet.
🚀 Quickstart
Load the corpora for MSRS-Story and MSRS-Meet:
from datasets import load_dataset
story_corpus = load_dataset("yale-nlp/MSRS", "story-corpus", split="corpus")
meeting_corpus = load_dataset("yale-nlp/MSRS", "meeting-corpus", split="corpus")
Corpus Dataset Example:
{
"id": // Unique ID for the document
"text": // Document text
}
Load the query-answer pairs for MSRS-Story and MSRS-Meet (available splits: train, test, and validation):
from datasets import load_dataset
story_qa = load_dataset("yale-nlp/MSRS", "story-qa")
meeting_qa = load_dataset("yale-nlp/MSRS", "meeting-qa")
QA Dataset Example:
{
"id": // Unique ID for the query
"query": // Query text
"gold_documents": // List of gold document IDs
"answer": // List of answer summaries
}
- Downloads last month
- 177