demagogue on 18/5/2011 at 22:50

---

A number of games give you the same key information whatever conversation path you take (to a point), but the flavor will be different. So then it's not so much of a gameplay mechanic (so you don't end up gaming it in ways different from how you actually want to talk, like you say) but it's an atmosphere or immersion element to get you into the world and your preferred character.

Nameless Voice on 19/5/2011 at 00:18

---

The idea of spendable "points" that forces you to decide which pieces of information you want to gain from an NPC is interesting - it forces you to choose what you want from that NPC most, and spend the points there first, even though that means you might not have enough points to then use later for another piece of information of lesser importance.

I also like the idea of your clothing and equipment affecting conversations a bit more, if nothing else than people treating you differently if you're wearing normal clothes or armed to the teeth with heavy armour. It really depends on the setting, but in real life you'd probably react a bit differently to a soldier with a machinegun than you would to a well-dressed gentleman in a suit than you would a drunk in a ragged coat. Being able to exploit peoples' opinions by changing your outfits, e.g. acting like a poor beggar so that people don't consider you a threat (without getting into Hitman levels of disguise ridiculousness) would certainly be a good mechanic to see.

Yakoob on 19/5/2011 at 06:08

---

Thanks for the responses everyone. I was secretly expecting TTLG's reaction to be either "you overthink too much, just make good dialogs" or "ok lets micro-analyze the shit out of this" and I guess its turning to be the former. To address a few points:




I know this is the last post, but it's one I wanted to address first as it goes back to the core of the issue.

See, when a player chooses a play style in a game, he is presented with a multitude of options. A "fighter" can use (even craft) many weapons, both melee, long range and throwables, use special skills, such as parry, dodge or knock-out, and use helping items, such as potions or magic scrolls. A "thief" can usually sneak around, hide in shadows, use sound cues, throw items to distract the enemy, turn off lights or use gadgets as rope-arrows, lock-picks or cloaks of invisibility. Even in non-RPGs such a variety is present - in a racing game, one can change their car, tweak the engine, swap out tires before the race, and ram into enemies, find shortcuts, drop an oil spill, or even shoot a gun during. And he can do it all dynamically, at any point in the game, constantly adjust and re-adjusting his strategy.

But a "conversationalist" - what can he do besides merely picking from one of the few predefined responses?

With all other gameplay options you have tons of variety choices, and you are free to experiment and switch them around as you please. With conversations, you have none of this. And, if we were to make a game that relies heavily on conversation, we will need to change it diversify it somehow in order to keep it engaging.

But reading through your replies leads me back to another lingering though I had but refrained from mentioned it until it was confirmed by others - perhaps implementing more complex conversation systems doesn't make the conversation more engaging, but in fact cheapens it, making the inherently complex idea of human interactions just too fake and gamey. Perhaps, then, it's not really the conversation mechanic that needs to be altered, but rather, the story and game flow that surrounds it? Just as game designers have to carefully plan ventilation shafts, mark doors as locked and give keys to some characters who can be pick-pocketed in order to create a "stealthy" gameplay, perhaps a conversation centric game, using just simple trees, needs to be design in such a way that certain conversations reveal key information, which can then be used and built upon in other conversation (perhaps with combination of some "appearance," or collecting special items we may need, as we can learn from yet other dialogs). All this akin to traditional adventure games, or what Rockstar's L.A. Noir is getting at?


Theoretical Example:
As an example, imagine good old JC Dentont from Deus Ex trying to break into VersaLife. Now the game already covers the "stealthy" and "guns blazing" approach well. What if we wanted to use a conversationalist approach? Maybe we could have JC enter the building and be denied access by the receptionist. Exploring around the surrounding park area, he stumbles upon a worker on his lunch break and finds out about an ongoing "investor presentation." He goes back to the receptionist only to learn that a special pass is required for the presentation. Exploring on, he reaches the "relaxation center" of the facility, equipped with a cafeteria and even a bar. He manages to chat up a disgruntled businessman there who left the presentation early. With enough persuasion, and perhaps easing things up with a few drinks (hello "The Witcher"!) or even a small bribe, he manages to convince the guy to give him his pass. Now just a short trip to the nearest clothes store to purchase a tux. With the pass in hand and a convincing disguise, he manages to make his way into the restricted area of the facility. Now, to get access to the laboratory floors...


(note: ammended this to the original post)





Really? Well, think of your favorite RPG of all time. Now, strip all the fighting, all the sneaking, all the magic, everything. Just leave walking around and conversations. Would you still play it? And I mean honestly not some half arsed "well I liked the game and story so of course I would play it" response, but a genuine thought experiment. Was the story only good because it was interspersed with constant well paced gameplay that never overloaded you with too much text and reading, and diverted your attention away when the plot was at its weak points?

... or just play Planescape Torment and imagine all conversations meshed into a single one and call it a "game."



And none of them could be completed via a diplomatic approach alone.



Fair enough, but I don't want to start off with "i made a video game before therefore my opinion is important and you should listen to me." Instead, following the mantra "do don't tell," I instantly started off on a higher, more formal note, to show I am going for a more serious and retrospective thread than just another "game publishers are evil and also mass effect 5 is kewl" threads.



No doubt that WOULD be awesome, but way beyond our technical capabilities at the moment sadly. Hence why we are stuck with crude dialog trees and minigame-esque approximations.




Oooh that's actually really cool. Kinda makes me think of Inception or Psychonauts, where there is a piece of information (or perhaps several pieces, or optional side-pieces) hidden somewhere in a "subonscious" world you need to get. But as you said yourself, it would take a lot of resources to create a single "conversation-world" (unless it's somewhat generated procedurally) and so it's not really suitable for a big-scale game, but rather, indeed as a special system.



Fair enough, but that's just semantics (and everything has been done to death really). Also, there is a big difference between "if weapon in hand say A else B" and having several "appearance dimensions" that affect the conversation in a more diverse way.



Good point, and a balance must be struck. While I am not asking you show the player / let him replay every possible conversation scenario, I personally would have appreciated it if conversation trees showed me ALL my dialog options with costs, even if my "skill" wasn't high enough to select them. Otherwis, as I said, I don't even know how my skill affects it, or if it's affecting it, or how much I need to improve it to become a better speaker?

Contrast it with basically every other gameplay mechanic which always gives you the feedback to let you know how well you are performing and how much room there is for improvement.

Or what Shakey-Lo said.



Hmm I guess I shouldn't have called this "RPG Conversations" but just general conversation then? There really is nothing here that neccessiates these ideas to only be used within an RPG.



It's not about the feedback that calling someone a poo poo head makes him angry. It's about getting a feedback saying "if my persuassion was better by three points I could try to convince this guy I am working for the Agency." Otherwise, you are missing on gameplay options because you don't even know they are there (which is *not* the same as optional, second-playthrough content; the optional content is what happens after you select that response, not the response itself).



Indeed, and I noted that in the problems section.



That's exactly why I said - imagine you want to make a game you can beat purely via dialog. No fighting, no sneaking around. Just clever wordplay. Is that enough to put significant development time in?



I agree. Another thing that bugs me is that the game always assumes everything you say is literal. There is no concept of "lie" or "ulterior motive." Some games try to do it clunkily by adding "[LIE] Ok I will help you" but it always felt to me like it only highlights the issue, not makes it disappate. Hence why I am against Karma - just because you helped someone doesn't mean you did it because you are a good guy. Like you say yourself, I usually play the good guy too - not because I *want* to, but because it's usually the most rewarding path in terms of XP, gear, story options, etc.



Because, as you succinctly point out, they won't know unless all your characters are extreme exaggarations and carricatures. Why do you think 99% of RPGs render characters as extremes or even cliches? Because it makes it easy for the player to understand and identify them.

Also, way to completely miss the point of the thread.

Yakoob on 21/5/2011 at 06:53

---

Speculations on LA noire on another forum have actually struck my imagination in a type of a "network clue" system for solving cases. While not strictly for conversations, it could be applied to it. Basically, as you find "clues" (which can be of many times, like "evidence," "testimonial," "phone records," "someone's appearance" etc.) you are able to create a network, linking them together in different ways ("causation," "correlation," "contradiction" etc.) until you manage to solve the puzzle somehow. Very idea in rough, but could be developed further.

Another thing I was started thinking of... someone before rightly mentioned that in all games, (and in all the systems above), conversations are always represented as a tree. This is inherently very limiting, does not support dynamic flow very well, and encourages repetition. So why don't we try to redefine THAT? How conversations are represented *under the hood*

One idea I've been pondering, after all of your guys' replies is partially procedural conversations (not fully cause that is a task for the next few decades). Instead of trees, just a loose pool of thoughts, attitudes, ideas, wants, needs, and key points. The character can "procedurally" draw from these depending on circumstances.

This could have a cool effect of (as someone mentioned before as well) the AI actually saying stuff to manipulate you based on his own wants and needs. And of course, talking to the same character twice would always produce slightly different results, than the same convo tree.

Of course, stringing all those elements into cohesive response, making them flow, and interpreting a player response is a whole different problem that would have to be tackled.

Here's a few very preliminary ideas I am typing out as they come:

----------------------------------

Template sentences with replaceable tokens:
Per-NPC template Example: "You're late for the meeting, where [optional_insult] have you been, [player_referrer]" -> "Where the fuck have you been, dipshit" vs "Where have you been, sweetheart?") ?

Each NPC would have its own pool of tokens for each token, so that we could customize speaking styles (i.e. an abrasive character would only have very insulting tokens for [player_referrer], whereas a charming lady would only define sweet, flirty words)

Global-Template: Example: express_want("I [npc_degree_of_wanting] [npc_desired_item]" -> "I could use a bunny" or "I deeply crave a chocolate bar!")

Similar to above (and totally mergable), but allowing us to dynamically generate simple sentences based on need and wants.

----------------------------------

That would, of course, be great for simple sentences we could predefined (as the two examples), but what about long, illustrous conversations, characteristic to a given NPC? Those would probably always have to be in the realm of the game writers due to their complexity. But while we could keep key NPCs with custom well written dialogs, we could potentially use this sort of procedural system to spice up the conversations with random NPCs that wander around the world and act as a "filler". Particularly, if we make more complex expressions as well by merging multiple simple sentence templates, such as:

"I heard that [time_stamp] there was [random_event] at [random_place]"
- [time_stamp] = {"yesterday", "last week", "few days ago"}
- [random_event] = {"some random [bad_person_descriptor] apparently [evil_action_past], and harmed [random_npc_name]"}
[INDENT]
- [bad_person_descriptor] = {"thug", "hilbilly", "robberrer"}
- [evil_action] = {"stole a painting", "robbed a commoner", "set the barn on fire"}
- [random_npc_name] = generated dynamically from list of NPCs
[/INDENT]
- [random_place] = generated dynamically from list of locations

These could then be tied with some procedural event-generation system (that actually spawns a random thug that mugs poor Mrs. Bittertits) and you could start getting an interesting, non-repetitive rumor-mill backed up by *actual events* (rather than the usual predefined "I heard bla bla..." you get when you ask someone "what is the latest rumor?")

Of course, this would totally rape any form of localization :P There's probably tons about this written from the old text-based adventure game days and linguistics studies, so I'll need to do some research!


EDIT: (http://koobazaur.com/temp/keng/convomatrix.php) EARLY PROTOTYPING (constantly updating)

demagogue on 24/5/2011 at 13:18

---

I studied cognitive science and philosophy of language in undergrad and have kept up with it since, so I think about the kinds of stuff you're talking about all the time ... like how to set up bots that can communicate information they know (or learn in-game) more naturally. Anyway, I'll tell you a few things I know.

The whole reason I came up with my dream-gameplay was because then events & nodes didn't have to be structured in trees like traditional games, but webs that could go backwards on themselves ... Then rather than trying to get to one or a few top points of a tree (collapsing it into semi-linearity anyway), you're trying to get to the center of the web and can come in from any direction. So I worked it out originally for gameplay-logic reasons. But maybe that kind of web-thinking could work for time-linear situations as well, where different paths can meander around and get you to the win-node in the middle of the web.

ArmA2 is a recent game using the template method. Watch any YouTube video if you haven't already played it. It's clunky and not very robust outside a really narrow range, like the tactical use in that game.

IMO the most clever free-range language puzzle was in an IF called (http://ifdb.tads.org/viewgame?id=4tb9soabrb4apqzd) The Edifice, where you literally had to pick out a new language word by word from a foreigner, the basic grammar verbs and nouns, until you knew enough to ask for what you needed. Really well done, but maybe hard to use that setup without just copying the whole puzzle.

Now back to letting AI give information they know or learn freely, there's lots of interesting things going on in AI on that ... like search engine tech & automated phone customer services (which also have much better meta-templates, and voices synthesized down to the phoneme, not just phrase templates like Arma2, so aren't nearly as clunky), where the whole idea is the user asks a bot for information, and the bot knows what they want and how to give it. I think some of that tech could be adapted for games. Only big thing missing is the bot motivation & utility; sometimes it may want to hide info or distort it for its own utility. But as for free question answering and presentation, this is the cutting edge stuff.

As far as something already designed for game use that does track bot motivation, state, & utility, the most interesting things I've seen lately is (http://www.storytron.com/) StoryTron, which does "interactive free storytelling". (VirtualStoryTeller is another system for natural language story-telling, but it's not interactive like StoryTron). What makes StoryTron cool is that all the bots have their motivations, personality, current emotion/state, & utilities simulated, and they (and the player) act, which changes different states (world states, other bots react based on their motivations & utilities, etc), which all feeds back into the bots' states in the next round, leading to new moves. To me this is on the best track to really cool and (more) natural NPC interaction. So I think this is the model that I'd want to emulate in trying to get something like you're mentioning, where NPCs can act a bit more autonomously in going after what they want generally. Then you can build a conversation system as part of that, so they say how they're really feeling based on what's going on in the game and in their head, and how they want to give you info or help you or mislead you, etc. That is, to really do this well, I think you need a general motivation & emotional-state & utility tracker for NPCs, and then a conversation system would be just one part of that.

Edit: Total aside, but there was a cool news story on (http://www.bbc.co.uk/news/technology-13510988) robot language published today.