A couple of technical references crossed my desktop this evening: Social Bookmarking in the Enterprise (ACM Queue, Nov. 2005); Software untangles chat room debate (NewScientistTech, June 2006). A few thoughts for virtual worlds.
Online conversation is a tricky beast for many reasons. Much of it has to do with context and how it is so difficult to trace through - even for highly caffeinated and well-trained online conversationalists. Why then is the text box in virtual worlds so small (scrollable or not, does it matter)?...
Claim: online chat scales poorly.
How many people can you fit on a single channel? Sure it depends, it is some function of the turnover in speakers and the number and depth (how long running) of the conversation threads and how important the threads are in terms of synchronizing some activity. Yes, RL point-to-point conversations don't scale well either, but the online medium invites an ease of channel aggregation and switching that is hard to engineer in the social spaces of the real world.
The NewScientistTech reported worked by Eduard Hovy (et al.) presented at the recent Human Language Technology Conference (NY, May 2006) based on an adaptation of the Hypertext Induced Topic Selection (HITS) algorithm applied to messaging. The idea here is to identify the most influential post or message in a conversation.
David Millen's (et al.) ACM piece is on one level a conventional spin on social bookmarking systems. However its detail is spun - interestingly enough - in terms of the enterprise space. Content collaboration there has a stronger value-added proposition that clarifies a little more sharply the benefit of shared content.
Both of these pieces, combined suggest that meaning and content may be best organized in non-linear structures, e.g. graphs. Or is there some narrative bug grounded into our biology that gets in the way.
Can you think of a better way to organize an online conversation in the same amount of real-estate now afforded to the chat box? Or is this what we got, to the end of time.
----------------------------------------------------------------------------------------------
From here:
KillSwitch: [Shouting] Does anyone want to join our hunting party?
Farglik: Dude?
KillSwitch: We're gonna go hunt wrixes.
Farglik: Right on.
KillSwitch: Why are you attacking the durneys?
Farglik: Dude?
KillSwitch: The durneys, the little bunny things -- why do you keep shooting at them?
Farglik: Dude?
Troobacca:My weapon powerup expired, I need a new one.
Farglik: Sure.
Troobacca: ...what's this?
Farglik: Dude?
Troobacca: You handed me a melon!
Farglik: Dude!
KillSwitch: Knock it off guys, I see some wrixes up ahead. Let's do this.
Farglik: Right on.
KillSwitch: We rock!
Farglik: Dude!
Troobacca: We so OWNED them!
Farglik: Dude!
KillSwitch: Uh oh, hang on. Up ahead are some Sharnaff bulls. We can't handle them, so don't shoot.
Farglik: Okie.
KillSwitch: You IDIOT! Farglik why did you shoot at them?
Farglik: Lag.
KillSwitch: Well don't do it again.
Farglik: Sure.
It seems that yes, one would need to generate context as if the chat were organized as hypertext, but the current hypertext frameworks I've seen all deal with static content. Sure, you could update definition pages over time, but I don't think that's hypertext's strength. Since situations and information can change on the internet extraordinarily rapidly, it sounds like the problem is how to deal with changes over time.
I didn't quite understand what you meant by "graph" until I thought that perhaps a flowchart would be a decent way of representing the information, context, and assumptions contained in a conversation.
In any case, if time is the problem here, what's the best way of displaying changes in state for one item over time?
Posted by: Alex Wreschnig | Jun 17, 2006 at 00:16
In any case, if time is the problem here, what's the best way of displaying changes in state for one item over time?
I don't think it is only change over time. It seems to me it would be useful to ID and present in some coherent fashion the flow (as you put it) from the 'important' (which may or may not be a proxy for 'influential' in the sense of Hovy, above). In a forum where there are ongoing parallel conversations, I'll assert that some degree of interdependency occurs (hence graph as metaphor). However, I 'll guess that the degree of interdependency likely varies considerably among the different types of online channels - say, threaded conversation on BB versus a guild main chat channel.
Posted by: nate combs | Jun 17, 2006 at 06:54
I've seen channels in Eve Online with over 1000 people in them. Granted, that's not 1000 talking at the same time, but people will tend to join alternative channels more aligned to what they want to talk about after talking about it in a more general one becomes difficult.
You rarely get a problem unless the difficulty involved in forming a more specific discussion is too high, since people will do this themselves, so I'm not entirely sure what the point of this is. Also note tools like IRC, where people are entirely capapable of following a number of seperate conversation channels.
Posted by: Daniel Speed | Jun 17, 2006 at 11:09
Not sure about a graph, but in the past we've tried a few things that have worked well for large shared channels. On one product, an online game lobby with people either hanging around or sitting at tables (think of a game convention or casino), we use different intensities to indicate proximity: if you were at a table, all chat from that table was shown in bold. Chat from those walking around not at a table was shown normally. Chat from those at other tables was shown in gray. This form of visual/spatial segmentation worked very well for our users, and also reduced the degree to which people turned off all "non-table" chat, which had been a social barrier up to that point.
Similarly in an MMOG, I could see trying techniques like dynamically highlighting/dimming conversation threads depending on whom you select or where you hover your mouse, so that you can follow one thread more easily. I don't think we're going to get beyond the linear, scrolling form factor of text because that's what the past few thousand years have handed us. But I think that by using other dimensions -- color, size, brightness, etc. -- we may be able to assist users in extracting conversational threads from a crowded channel, just as we semi-selectively tune out other conversations in a crowded room.
Posted by: Mike Sellers | Jun 17, 2006 at 12:11
>I've seen channels in Eve Online with over 1000 people in them.
I've seen those too. Yes, with enough ediquette and modest enough conversational goals (vs.intricate discussion of something complicated, say) and sufficient attention - it sort of works out. Amazing how one can develop a "stack" of sort of ongoing conversation threads.
Here is the rub - IMO - actually try to do something else (besides following the chat) can lead to mistakes - and forbid you go AFK for a while: who has time to catch-up read all that stuff. Noise in the channel.
Posted by: nate combs | Jun 17, 2006 at 12:34
Dude!
There are lots of ways to organize chat with large numbers of users and make it useful, but yo have to tailor your solution to the nature of what you're doing.
One simple way is have ops voice speakers, only those with voice get to talk, those without voice can message ops for permission. Very common in online interviews or chats with celebrities--AOL does this kind of thing all the time.
But you can certainly get more complicated. Years ago, I tried to get funding to develop a chat system that would enforce Roberts Rules of Order ("everybody keeps talking about electronic democracy, but nobody does anything about it..."). Didn't happen though.
Posted by: Greg | Jun 17, 2006 at 14:16
>based on an adaptation of the Hypertext Induced Topic Selection (HITS) algorithm applied to messaging. The idea here is to identify the most influential post or message in a conversation.
I have to say I'm worried about the implications of more Google-style and Wiki-style "inducing" of what is "influential". Very problematic.
It strikes me that you could stay linear, and just make more lines to the linear stuff. The big problem with this chat referenced here from the quest, and with chat in SL and other games/worlds, is that there's only that one line, or even half line.
I think one of the ways in which Will Wright and co. made the Sims Online such a nicer, socialable world in which strangers so readily collaborated was by making the chat interface be a box that took an entire paragraph at once. That got people to type out a few lines that completed a thought. That meant that their thought spit out one line at a time, but game more like the bubbles around a cartoon character in a comic strip (indeed you could set the graphics to put bubbles like that). That way, there wasn't this constant lagging of comprehension and the intersplicing of lines creating incoherence.
It's just such a simple thing: make a box to fit more than a line. If the could do it in 2-D TSO, why can't they do it elsewhere? Part of the problem is they don't want to take up the space because they have all these other widgets on the user panel. But this should take priority.
People even have meetings sometimes where they ask for the right to make an "all clear" signal and be allowed to type a whole paragraph, which involves pasting line by line.
There is something in SL called the "Soap Box" made for the Thinkers' meetings. It isn't used so much because it had this annoying tic of telling other people who clicked on it to shut up, somebody was already on the box -- instead of just queuing them up. It's just a scripting thing probably easily fixed. So the thing would keep firing off and annoy people during Town Halls, where they tried to use it on those 4-sim crossroads and of course it often didn't work because it would say "you are not in the region of the object you touched".
Posted by: Prokofy Neva | Jun 17, 2006 at 16:57
"Claim: online chat scales poorly."
Compared to what? Verbal chat breaks down at a far lower figure.
Posted by: Raph | Jun 17, 2006 at 17:44
I've often thought that it would be very helpful if there would be a visual way to parse the various conversations going on in a room by either color, indentation, something. Even a number in front of each person's chat name indicating, "This is conversation #1," so that you can track who's talking to whom. In a room with even 2 conversations going on at the same time, it's hard to keep track of who's saying what to whom. But if you could choose to bust your conversation out into its own window (not hidden, like IM, in case others want to join), or at least tag the string somehow, it would be simpler. General comments go to "zero" or something. I don't know. But if I'm talking with Bob and Mary, our three sets of comments could somehow be differentiated visually from the others...
The "Dude!" string above really doesn't phase me. It's more an issue of players being... well... variously engaged. Ahem. You get that in RL in spades. Go to the mall and listen to a gaggle of 13-year-old girls try to organize what store they're going to next, and you'll get a string of live chat much like it.
Posted by: Andy Havens | Jun 17, 2006 at 19:34
"Claim: online chat scales poorly."
Compared to what? Verbal chat breaks down at a far lower figure.
I meant by scale that online chat becomes less efficient as you increase the number of participants, becoming at some point impractical.
I agree with what you suggest - that RL chat scales less well in comparison to online chat (even less efficient and at a nearer point becoming impractical).
However. I think one might be able to argue that RL vs. VW chat comparisons involve a bit of "apple and oranging". RL chat involves a whole different set of constraints (e.g. physics, bodies with mass, and biologically grounded social instincts - 'comfort zones' and eye contact) and just plain practicalities of how life is structured now - that VW chat is exempt from. Consequently I think they become functionally different. Take the earlier Eve 1K chatroom example (with the lessor constraints of the text medium it demands), what analog is there in RL?
Posted by: nate combs | Jun 17, 2006 at 20:54
In the Knowledge Management space, there are various thoughts and ideas how to organize and make useful various chats over various mediums.
Most of the work started with searching and parsing of text, but is now moving toward active agents organizing knowledge. I can see this kind of technology developed for the business environment transfered to the online game environment soon. Autonomy is one of the many companies that is working on natural language parser and organizers.
For example, one method of scaling online chat are:
1. Use time and space and the first order variables. More recent or closer conversations are passively highlighted by proximity (coversation that you overheared during a party).
2. Use relevance as the second order variable. Index relevant terms used and pull the relevant conversations procedurally. A relevance network map can be drawn to allow the reader to time-shift the online conversation and zoom in on particular lines-of-thought.
3. etc. etc.
You see, the key to online chat that enables scalability and also allows for time-shifting is that everything is recorded. The only thing needed to allow scalability is tools. Googles and Wikis may not satisfy, but we're getting there.
I personally think the HITS system need to be married to a conversation summarizer and visual mapping tool. Something that the Web 2.0 trend no doubt will yield a great solution.
Frank
Posted by: magicback (Frank) | Jun 17, 2006 at 21:24
Nate,
One real world analog is the Echelon surveillance system. What happens when everyone can get access to their own Echelon attached to their brain?
So the thought experiment is what tools and functionality do we need to break through the Wall of Babel and allow for a level of chat omnipresence.
Frank
Posted by: magicback (Frank) | Jun 17, 2006 at 21:33
The issues I've been having with coming up with any sort of useful, insightful reply is that it's freaking hard for people to organize their own conversations in the middle of a game or some other attention-intensive activity. The current text-box model with one coherent (or incoherent) conversation does one thing well -- and that thing is "Transmit basic text messages without eating up too much of a user's time or attention." While a lot of these ideas are good for other applications, are any of them really good for improving in-game chat, instead of just a chat room?
Mike Sellers' "tables" idea sounds like it would work well with Second Life and related worlds at least... better than anything I've thought of so far, but if you're moving around that ruins everything but the idea of party-chat, doesn't it? Or am I missing something here?
The constraints I've been using are that I don't want to be fooling with anything extraneous while trying to keep up with a group (or if I'm between aggro'd monsters - I've done a lot of chatting that way).
Posted by: Alex Wreschnig | Jun 18, 2006 at 01:14
it's freaking hard for people to organize their own conversations in the middle of a game or some other attention-intensive activity.
From the discusson above it might be useful to categorize the issues - those related to (1.) "information management" part of the problem versus (2.) "noise suppression" versus the (3.) "real time communication demands" placed on chat.
(1.) is where this post started out with - but they are all related. Greg intriguingly suggested ideas addressing (2.) to get at (1.). (3.) is quite an issue related to everything else.
An interesting question here is how any piece of the solution will change the nature of in-game chat. For example, consider a hypothetical pipe-dream (post) of some graphical organization of conversation based on meta information (what is important/interesting to me etc.). Imagine that is how all chat came across as. I daresay that will offer quite a different (detached?) view of chat than the layered stream-of-consciousness we have now.
Would such a hypothetical be better? Well it seems to me it would really depend on the goals of the chat to the individual at a particular moment in time. If it is manage large amounts of information -- perhaps. Is it to banter on a personal level with someone -- probably nothing will beat the ability to set up a private channel somewhere in cyberspace. etc.
Posted by: nate combs | Jun 18, 2006 at 10:45
Who monkeys with the chat display interface monkeys with human consciousness.
So I'm not sure I just want to hand this function over to some bot or code-is-law program that decides for me, based on Google philosophy, what meta information is "interesting to me" etc.
I don't want the room to look like delicio.us tags with their vacuous non-standardized meanings, large and small typeface, etc. Organizing chat interface should be opt-in/opt-out.
The idea of coloured text really annoys me but that's just because I have synesthesia, I guess, and for me it's already tagged, organized, and coloured, thank you.
Posted by: Prokofy Neva | Jun 18, 2006 at 11:06
Also, I think any analysis of text chat in virtual worlds has to take into account not just the scrolling IMs, but *what else is going on*. How many other windows are open? Is the arrangement of text into windows tiling up from virtual world parcel/virtual world IM/Yahoo Messenger/email/IRC channel/Real Life on f2f, phone...going to arrange or distract???
Tao Takashi has a good summary of this on his blog as he contemplates the confusion and frustration of the now-voice Town Hall meetings with Philip Linden:
"So why I am writing so much about the setup you might wonder? Well, it’s basically because it felt somewhat strange to me. It wasn’t really a meeting going on in Second Life anymore but beside it. You had Philip standing still or getting Away while you listen to him, you had this IM window open where some other discussion was going on and I personally also had IRC open with some more discussion on the topics. So having that, there’s no real reason anymore to come to that meeting as you can do everything also from whereever you are in-world or even if you aren’t in-world at all (just missing the IM chat then).
Which make me think that the amount of actual interaction has been minimized to nearly zero. In earlier meetings there have been at least some questions coming up directly in the audience and Philip was sometimes answering. Right now it would have made no distinction if Philip even would not have been there."
http://taotakashi.wordpress.com/2006/05/19/yesterdays-town-hall-meeting/
Posted by: Prokofy Neva | Jun 18, 2006 at 13:12
My point is that text based communication channels have been about for a while, and in a greater context than just MMOs and virtual worlds. If this sort of technology were critical to communication in those media, I'm sure we would have seen the emergence of some sort of similar thing in, f.ex. IRC clients already.
If anything, the New Scientist article illustrates just how far we are away from good systems for dealing with language, and as people who have tried spoken-word-to-text solutions that don't work very well, it's incredibly frustrating when they don't work properly. If anything, trying to solve a problem that people tend to solve for themselves, and doing it badly, might be entirely counter-productive. Given the correct tools for the job, people will do this themselves - instead of hyperlinking in chat like crazy, why not give people better access to communal wikis, announcement boards, mail, and websites of their own?
It seems to me like this is an attempt to reinvent writing for a textual oral tradition of the internet.
Posted by: Daniel Speed | Jun 18, 2006 at 17:19
From the discusson above it might be useful to categorize the issues - those related to (1.) "information management" part of the problem versus (2.) "noise suppression" versus the (3.) "real time communication demands" placed on chat.
(1.) is where this post started out with - but they are all related. Greg intriguingly suggested ideas addressing (2.) to get at (1.). (3.) is quite an issue related to everything else.
As far as I can tell, the most accurate thing I've read was something Greg said:
There are lots of ways to organize chat with large numbers of users and make it useful, but yo have to tailor your solution to the nature of what you're doing.
I like your categories, Nate, but maybe we should go a little further. Town hall meetings are completely different from chatting in a safe area of an MMO, which is way different from chatting between spawns in the overworld of the same MMO. It's probably a good idea to have some sort of idea of what problem we're trying to solve before we think of any specific solutions!
(1) Town Hall Meeting: I imagine noise reduction is less of a priority than organizing a large quantity of possibly worthwhile information here.
(2) Large number of unmoderated, random conversations, like what you might find in "cities" in a typical MMO: I would suppose that every category you mentioned would have equal importance here.
(3) Conversation in a dangerous area: If it's not ridiculously easy to use, it won't be used. Noise reduction good, but secondary. (categorization of information is probably a non-factor.)
(Those are what I came up with, anyway. All IMHO.)
For what it's worth, the idea I keep thinking of for #1 is some sort of real-time threaded message board system. Players might select statements to reply to, allowing for easy self-threading. You could show bullet points of a size proportionate to the "importance" as an easy way of getting attention, and decide "importance" using any number of algorithms (another "easy way out" would be to use the number of replies). It sounds very doable to me, but maybe not as interesting to talk about as some of the others.
I really like Mike Sellers' "tables" idea for #2. Can't think of anything better, or really even close.
Posted by: Alex Wreschnig | Jun 19, 2006 at 00:05
I think the symbiosis between user and interface to yield the necessary organizational effect is important. Allowing users to customize the tool to fit their own need is probably more important than any standardized methods proposed.
For a guild raid communication tool, I can mentally imagine a few nifty agents/bots that automates and organizes the various lines of communication to add coordination of a raid.
For example, an agent can look for key text commands like “Rally Here” to (1) open a new chat window for everyone within a specified proximity and (2) set the new chat window as the primary chat window. Another command like “Mages Rally Here” could be used to do the same, but only for the mages in the group. To
To keep track of other channels, a healer may program an agent to bring to the forefront the “health” channel whenever someone typed or Hot-keyed “Need Healing”. Moreover, the agent could be sophisticated enough to integrate with the targeting interface to quickly bring forward the caller in to view.
The above are just some of the immediate ideas I have on raid communication tools. With regards to:
1. Organization
2. Filtering
3. Compression/summarization
4. Highlighting/priortizing
the tools that allow each person to hack together their own communication agent is probably more important than any standardized methods.
Frank
Posted by: magicback (Frank) | Jun 19, 2006 at 07:48
>some sort of real-time threaded message board system. Players might select statements to reply to, allowing for easy self-threading.
This format is tolerable precisely because it's *not* in real time and there it is, static, able to be accessed when you have the time and inclination -- and patience. Porting this format to real time might involve lots of clutter and impatience.
Posted by: Prokofy Neva | Jun 19, 2006 at 14:06
I dunno. If you've ever looked at a really active thread -- I've seen a few DeviantArt ones with several thousand people posting at one time -- they somehow manage to work, but you don't have much time to reply before your comment is rendered irrelevant. As opposed to something like TN with its extremely leisurely pace. :)
The reason why I suggested it for town hall meetings was because I imagine something like a series of "core" posts by whoever's speaking followed by hundreds or thousands of replies, all at once and probably well-thought-out, and it seemed like a good way to organize them, that's all.
Posted by: Alex Wreschnig | Jun 19, 2006 at 19:44
I imagine something like a series of "core" posts by whoever's speaking followed by hundreds or thousands of replies,
How do you picture these to be visually represented? As bubbles? As larger and smaller type face hanging in the air? As a big scroll along the left?
Posted by: Prokofy Neva | Jun 19, 2006 at 19:58
What matters most, I think, is that the receiver of messages has the power to "tag" them in real time.
From the point of view of a message receiver, I'd want to quickly/easily tag messages by:
* source (the name, location, type, or other attribute of a message sender)
* sender-defined priority (so that I don't miss useful interrupts from unexpected sources)
* relevance (compared to keywords in a list).
And then I'd want to be able to send messages of certain priority levels to different interfaces (including to the bit bucket).
Even if only three tag priorities were available -- important, background, ignore -- as long as the tagging mechanism is relatively simple/fast, that would go a long way toward helping me extract information from noisy channels.
(Note that this approach might also help in non-real-time systems like email.)
--Bart
Posted by: Bart Stewart | Jun 19, 2006 at 20:00
How do you picture these to be visually represented? As bubbles? As larger and smaller type face hanging in the air? As a big scroll along the left?
The image I've got in my head is that you would be able to completely see one "level" of comments, with importance rated by the visual (pixel) size of the bullet (I was thinking bullet-points in a chat box, with larger bullets indicating more importance, but this is something that would require experimentation).
Of course I haven't got any clue whether it would work, but I think it might. I don't think user tagging of posts is feasible for any virtual worlds I'm familiar with, but there might be some interesting ways of tagging users -- so users you are more familiar with, for example, might appear brighter or larger (but doesn't displaying a username already serve this function? So I might be suggesting something useless here.).
Maybe an easy and effective option would be to let users set certain words to filter for. So if you were using other filtering (by distance or table, etc), words that might be of interest to them could be bolded or color-shifted. I like the idea of walking around with a filter set for, say, "Red Viper" if I'm looking for a red viper bow, and have it drawn to my attention as I wander around. User configuration would also mean you don't have to have it, if it bothers you (or you find it redundant).
Posted by: Alex Wreschnig | Jun 20, 2006 at 01:49
Threading could work- when typing your response (I'm assuming we're still typing, not talking) instead of hitting enter, you click on the statement you were responding to, presumably one that's either in chat history, or floating above someone's head, or whatever.
Clicking or otherwise interacting on someone's line of text might give you an opportunity to "pay attention" to that whole threaded discussion, much like your email clients "show all related mails". You can use a few systems to highlight this interesting topic- from bold type in chat, to a seperate chat window or whatever.
If nothing else, it would make it possible to have a threaded transcript of the session, just like usenet etc. Anything you just "press enter" is a root level statement, anything you identify as a reply (such as by clicking on a statement to reply to), is nested accordingly.
Expanding it, you could even use some kind of hypertext system to provide link reference to previous statements/threads rather than just quoting them via cut n paste.
Posted by: Ace | Jun 20, 2006 at 06:29
Well, hrm. The problem with most of these solutions is they defeat some of the purpose of chat as a medium: Get the message out quickly. Threading tends not to work well in synchronous communication due to the complex ebb and flow of chat topics.
Color and keyword highlighting seems to help on this.
Posted by: kirkjobsluder | Jun 20, 2006 at 15:35
Rather than rely on machines to sort out semantic 'relevance' to the discussion, why not return to the idea that began this thread and introduce a social dimension to the chatroom paradigm?
It occurs to me that even at a dinner table of 12 people, I personally select a handful of people to interact with - based first on their proximity to me, then by other factors such as interest in the topic and personal interest in those people 'following the thread' as it were.
Imagine if, in this single chat window of 1k participants, I could drag the name of a speaker out into empty space and spawn a new window filtered to show only their text (the inverse filter would be placed on the main chat window to avoid presenting duplicate text onscreen).
At the same time, his computer would send me the names of people that he has added to his 'circle' of conversation within the room; that is, presuming that he has dragged multiple names from the general chat into the same sub-chat window. To the casual observer, nothing has changed, and conversation continues in the tangled line-by-line mess.
There is a question of scale and the multiplicity of windows that could get spawned in an unmanaged space, but it seems this could be overcome with some thought.. It would also be interesting to play with the visual/spacial cues afforded by virtual environments: perhaps autospawning new windows grouping chat from avatars in your visual field and returning them one-by-one to the main window as you turn away?
If anyone's interested in pursuing these ideas, feel free to contact me in-world SL or via email.
Posted by: Creative Nadir | Jun 23, 2006 at 05:12