|
Videogames are, generally speaking, about a few simple interactions. Fighting is the one that undoubtedly shows up most often, but of course, there are plenty of other sorts - racing cars, playing sports, and so on. But one thing that developers have yet to really master is something which is arguably far more pertinent to our interests as human beings - "simply" talking with other people.
There are all sorts of distinct challenges when attempting to simulate conversation. Most games have distinct failure and success states, so dialogue has to be able to fit into our preconceived rules about winning or losing the game. It has to give players a feeling of reactivity, otherwise it comes across as unnatural and stilted or even non-interactive. Although there are manifold ways in which developers attempt to circumvent these problems, few are ever really been successful in creating dialogue interactions that feel realistic and believable.
In this article, I'll be breaking down some of those challenges in more detail and providing a few examples of techniques used to create compelling conversations, as well as a number of recommendations that can be used to approach tired dialogue systems from new angles. Although I use role-playing games almost exclusively as examples here, it's only because there are so few examples of other games, especially in the mainstream, that revolve so heavily around conversation. Also note that this article assumes a large amount of dialogue - most of the concerns here regarding content bloat and dialogue obviously only apply when a game demands thousands upon thousands of lines, rather than a few hundred.
The Uncanny Valley of Dialogue
Although the much-abused term uncanny valley is almost always used in the context of graphics when it comes to videogames, it's in dialogue where the uncanny valley becomes most obvious. Just think about all the times you've been playing a game and a character's said something that doesn't make sense for a certain situation. Perhaps you heard a dialogue line repeat itself. Maybe a line was even cut off mid-sentence in favor of another one. Or, after your first play-through of a game, maybe you found that most of the conversation options all caused characters to respond in the same way.
This goes doubly true for the player as well. Even though games will often provide radically different choices in dialogue, or allow for multiple lines of inquiry, those lines always equate to simple choices between A, B, and C - there's little in the way of nuance and expression. In Mass Effect, I can't choose to say a line in a cocky manner, or a respectful manner, or a snide one - I have to choose between the options given to me, and if I'm not given an option I like, then I'll have to pick another one. For all those dialogue choices out there, a lot of time could be saved by realistically boiling everything down to the essential choices.
|

|
|
Even when we're given the option to choose the tone of our replies, the meaning can never stray from what's explicitly defined by the developers in advance.
|
The reason for this is obvious - short of creating some sort of super-intelligent AI to parse text and provide responses that make sense in context, and then coming up with technology to voice-act those lines in real-time, it's pretty much impossible to have dialogue that is truly adaptive to what the player can say. Dialogue content is completely finite. Although many interactions can be governed by sets of rules (wherein we abstract our conceptions of reality), speaking to another human being is something that is not so easily predicted. People have personalities, they make rash and unpredictable decisions, they let emotions guide them. Think back on all the times, even just today, when you said or did something based on impulse - while there might be some rules behind how you behave, they're certainly not things that you can divine, and you likely know yourself better than anyone else!
While we've had years and years of training to abstract activities like violence, and others, such as a football game, can be understood in terms of the rules governing the sport and simple approximations of physics, we have little understanding of other people, and even if we get to know someone very, very well, we can't predict complete strangers - in other words, the same rules don't apply. Most videogame content works by producing sets of common rules that different actors can interact within - in a shooter, you interact with the world by shooting at enemies, who behave in predictable and finite ways no matter how many you fight; in a racing game you interact by piloting a vehicle in competition with others to reach a final point - but that doesn't work very well for talking.
Approximating Humanity
The first solution to this problem is to create a system for interacting with characters that deals less with providing highly specific content, and more with content that is highly reactive to the player's input. Since developers can't write unique lines for absolutely every single possible outcome to a situation, this usually means going with templates, or reusing dialogue lines, but on paper, does this seem like such a bad idea? After all, we already accept a great deal of approximation from videogames. We understand that quips like "low on ammo!" serve as auditory feedback on another game mechanic, and we also accept that, under the hood, the game only has so many lines available to offer.
Smart developers know how to space out these lines to get the most mileage out of them - such as recording 10 different variations on the same line and making sure that players hear repeats as little as possible. With a smart implementation, it's impressive how often players' brains will trick them. We don't have extremely long-term memories for most things, so creating dialogue content that exploits the limitations of short-term memory will do a lot of the work for them. But when it comes to more significant, memorable dialogue, that approach quickly becomes useless as long-term memory takes over.
There's really only two solutions to the problem - either create a dialogue system wherein most dialogue only plays once, but some will repeat indefinitely based on the needs of gameplay, or create a dialogue system wherein the amount of available responses increases dramatically. Specifically, this means building a set of rules just like any other for the game, ones that approximate human interaction, and provide gameplay flexibility, even if it means giving up the "immersion factor." Obviously these will vary based on the game being created - the needs of a shooter are very different from those of a dating sim, for instance.
|

|
|
Dialogue in Morrowind lacked the emotional resonance that unique responses could provide, but made up for it with huge amounts of detail and universal mechanics that determined how individuals reacted to the player.
|
The latter solution used to be very common in older role-playing games. The Elder Scrolls series, up until fairly recently, used keywords to simulate paths of inquiry. Players could have a list of ten, twenty or more inquiries, and would largely receive the same responses, with variations for specific cases (like plot-advancing dialogue, or minor variations such as changing pronouns and gendered terms around). In Morrowind, for instance, most characters would recite the same lines over and over again, but mechanics took over - a character's individual reaction modifier would change based on the player's behavior (maybe asking about a taboo subject would net you a -20 reaction penalty), and racial or cultural background would also affect what a person could offer information on (so asking people about events on the other side of the world would rarely be worthwhile).
The big issue with such a system is that, now that you're dealing with a ruleset, suddenly designers and programmers have to start thinking about dialogue in terms of gameplay rules. No longer can conversations be about conveying lots of unique emotions and subtleties to the player - now they're about cause and effect, winning and losing - and have to be crafted with that in mind. And, like any game, those rules have to be consistent, predictable, and simple enough to understand. By effectively turning dialogue into a mini-game, it becomes subject to all of the same constraints any other game is.
Dialogue by the Tree
Dialogue trees are a staple of role-playing games as well, and by far the most popular way to simulate conversation. Dialogue trees, true to their namesake, take the form of a number of topics of inquiry, which then branch out into more paths. For example, a dialogue tree structure might consist of: general inquiry -> clarification -> opinion, with the final option taking the player back to the "root" of the conversation.
Obviously, developers can do a lot with dialogue trees, and games like Planescape: Torment are testament to that, with thousands of unique lines of dialogue and dozens of paths of inquiry that, in themselves, make up much of the gameplay. At the same time, however, the key limitation becomes abundantly clear: while you can take a few shortcuts with a dialogue tree format, pretty soon you're going to end up with huge, sprawling conversations. The content bloat mentioned earlier is felt very quickly when using dialogue trees, and if players notice too many shortcuts being taken (like identical responses to two radically different dialogue options) then the sense of realism that dialogue trees usually go for is completely shattered.
Despite the bloat, however, it's clear that dialogue trees have a big advantage - they do a much better job at simulating the act of conversation. Even if you need (mostly) unique lines for every inquiry the player makes, the big benefit of that is that characters can have much more personality, the player is able to express an opinion in more nuanced ways, and, most importantly, that the mechanical side of dialogue disappears. While there's always going to be a binary yes or no choice, many games do an excellent job of obscuring exactly where the variables in conversations are flipped - Dragon Age: Origins, for example, will show the amount of influence earned or lost after a conversation, but deliberately hides what dialogue options actually affect it, to better simulate the act of talking to a person rather than picking responses from a list for best effect.
|

|
|
Dialogue trees can provide much more detail than almost any other dialogue mechanic short of a graphic novel, but the amount of writing required can become prohibitive to presentation.
|
Additionally, the reuse lines as "cheating" on the part of the developers might be a little unfair. Usually, it's actually quite acceptable to reuse responses and add separate lines in that help redirect the conversation. Most players never really notice these - next time you're playing a dialogue-heavy role-playing game, take note of how many times you hear connecting statements like "anyway", "however", "meanwhile", and so on - in almost every instance, they're being used to disguise points where the conversation has branched and needs to re-converge. Once you pick up on this, it's surprising just how scripted many conversations really are, but so long as there are enough unique lines to keep the illusion going, it works splendidly.
There's also the player's own emotional impact and investment to consider. Gamers are usually not developers, and they're not going to be keen to the tricks and shortcuts used - they're often in a very different mindset when playing games, enjoying the content as it's presented to them rather than fussing over the details or looking for holes to poke. Therefore, as much as I want to complain about Commander Shepard's heroic speech influencing absolutely nothing in Mass Effect, I have to admit that, as a player, it's still pretty cool to be able to give that speech in the first place - to me, it feels like a real decision, and the way that it influences the tone of the narrative can't be denied. Even if it doesn't really matter to the game whether Shepard's an idealist or pragmatist, it matters to me.
Closing Thoughts
The unfortunate fact is that there are very few other ways that developers have actually handled dialogue - although I could bring up text parsers, they're really not much different from the keyword system mentioned above, except that the possible inquiries are kept ambiguous. For all the games industry has managed to so effectively simulate the act of killing another human being, or driving a car, the more complex and subtle, less predictable and deterministic act of talking to another person is something that's still up in the air.
That said, there's a few ways that existing dialogue systems can be enhanced to produce conversations that feel more natural and realistic:
- Abandon the UI. Sometimes, it's better to produce results by getting the player to actually do something rather than picking options off of a list or typing them into a dialogue box. A great example of this can be seen in Half-Life 2, where Alyx Vance will react to all the things the player can do in the game world - they're not deep interactions, sure, and the game is a shooter so there's little meaning behind them as far as gameplay goes, but there's something far more satisfying about using the tools available in one's arsenal to provoke a response rather than selecting "(Shine flashlight in Alyx's eyes)" in a menu. The upcoming Naughty Dog action game, The Last of Us, looks to be taking this model to new heights.
- Emotional impact doesn't mean detail. It's a common misconception that players need voice acting, an orchestral score and Hollywood-style direction to connect with the characters in games. Our brains are willing to fill in an exceptional number of gaps in presentation, and sometimes the most engaging experiences are the ones that exist in our own heads rather than in code. The emotions that players bring to the mechanics of a game are far more potent than the ones that writers try to squeeze out of an audience via sympathetic techniques.
- Remember what the player's done. This is a little thing, but one of the key ways to make characters feel real in a game is to make them reference past events to provide a sense of continuity to the world. These don't have to be dialogue choices either - one of Deus Ex's most remembered moment is Paul Denton scolding or praising the player depending on the level of lethality used in the game's opening mission, while other characters have the opposite reaction. This simple distinction, which ties in with the broader mechanics of the game and was probably tracked in all of one or two variables, does more to draw the player into the game than all the expensive cinematic sequences in the world.
As usual, thanks for reading, and please feel free to leave comments below!
|
"it's clear that dialogue trees have a big advantage - they do a much better job at simulating the act of conversation."
imho dialog trees simulate a "choose your own adventure" book, and (unfortunately for the player) you often have to cycle through much "mechanical" robotic/repeated dialog since they follow a rigid tree structure (and the good stuff is at the leaves).
"....the player is able to express an opinion in more nuanced ways, and, most importantly, that the mechanical side of dialogue disappears."
i haven't ever been able to be express any opinion in a video game in a nuanced way (so perhaps my experience is different from yours) (are you perhaps referring to Mass Effects Paragon/Renegade options, or like Fallout's [persuade] [intimidate] .. options? Those are nuanced?) I find dialog trees to be an absolutely mechanical option, and to be frank they seem like "unfun" minigames ram-rodded into an otherwise fun game.
"Dragon Age: Origins, for example, will show the amount of influence earned or lost after a conversation, but deliberately hides what dialogue options actually affect it, to better simulate the act of talking to a person rather than picking responses from a list for best effect."
this exactly describes my point, you go into the "conversation mini-game" and try to see if you can gain +5 (or whatever). To me this is absolutely mechanical, and leads to me not choosing conversation options that i would choose if i were the character, but rather choosing whatever options will give me the best result on the "conversation mini-game". (immersion breaker)
(..sorry for the long post, respect for the article btw, i very much liked the discussion and points you bring up, just wanted to throw my hat in the ring... cheers)
"imho dialog trees simulate a "choose your own adventure" book, and (unfortunately for the player) you often have to cycle through much "mechanical" robotic/repeated dialog since they follow a rigid tree structure (and the good stuff is at the leaves). "
I agree, although with good construction of a dialogue tree this is less a problem. For my Dragon Age mod, Thirst, I made most key conversations (especially those integral to the plot) avoid the menu format as much as possible by avoiding the more typical category/subcategory/choice structure and focusing more on linear flow. Part of this could also be understanding dialogue as a tree to begin with. Skyrim's dialogue editor, for instance, is functionally identical to your standard dialogue tree, but the presentation makes it easier for designers and writers to avoid the typical tree structure issues.
"i haven't ever been able to be express any opinion in a video game in a nuanced way (so perhaps my experience is different from yours) (are you perhaps referring to Mass Effects Paragon/Renegade options, or like Fallout's [persuade] [intimidate] .. options? Those are nuanced?) I find dialog trees to be an absolutely mechanical option, and to be frank they seem like "unfun" minigames ram-rodded into an otherwise fun game."
To each his own. Here's a question - how do you feel about keeping skill checks but removing the indicators that you're actually using them? You can bullshit your way through and maybe convince someone, or you can be called out a liar based on your Deception ksill or whatever, for instance, but the game will never tell you SUCCESS/FAILURE/etc. Fallout the First actually does this a lot with high-intelligence dialogue and similar.
"this exactly describes my point, you go into the "conversation mini-game" and try to see if you can gain +5 (or whatever). To me this is absolutely mechanical, and leads to me not choosing conversation options that i would choose if i were the character, but rather choosing whatever options will give me the best result on the "conversation mini-game". (immersion breaker)"
This is a good point and I fully admit that I've gamed the system on occasion. The real issue isn't the system itself though, it's that it set up to be gamed in the first place. Once again, referring to my own mod, I recently implemented an influence system for followers, but it's almost impossible to gain influence with one without losing another's. For example, one might like fighting and hate sneaking around, while another prefers the opposite - you'll end up earning or losing respect accordingly, and there's no way around it. By staying away from the petty "suck up" conversation choices, gifts and other items, dialogue becomes less a mini-game and more a barometer for NPC personalities. Neverwinter Nights 2 and Mask of the Betrayer did this very, very well, both by having companions with contrary opinions and breaking the "skill check = win" rule by having some dialogue options backfire and piss off your followers.
"(..sorry for the long post, respect for the article btw, i very much liked the discussion and points you bring up, just wanted to throw my hat in the ring... cheers) "
No problem, it's always the contrary comments that are the most interesting. Thanks for the reply!
thanks for being civil... (sometimes it's hard to get across a disagreement over one's opinion about a certain point without inciting a riot in comments... i much prefer debating ideas and especially opinions without getting sidetracked with "i'm right, you are wrong" posturing)
...also, it's cool you've created a mod (which is more than i can say for myself) i appreciate the fact that you not only talk the talk, but also "walk the walk".
concerning skills and in reference to structure of dialog trees i suppose my ideas are more
"scorched earth" (get rid of em entirely) than most when it comes to dialog trees, I understand why we have em (limited buttons on the controller) but i greatly prefer something that is keyword based (at minimum) but more natural conversation based (in a perfect world). so my opinion on how to handle skill based dialog options is to not have 'em at all.
(and btw i appreciate the fact you are trying something new in your mod, i think it's cool people continue to innovate in this field, i don't mean to be discouraging, just my uneducated opinion)
i've probably got a lot of crazy ideas (alternatives to dialog trees) which i should compile in a blog... ( i think i commented a few on a Kinect post earlier
http://www.gamasutra.com/view/feature/172917/charging_at_the_chall enges_of_.php
) think i'll do that rather than try to muck up and side-track your blog post with marathon comments. again thanks for the courtesy.
Nothing breaks the immersion factor quicker in a game, than when an npc, one which you just saved half his family not more that a few minutes ago, now seemingly not recognizing you. The web uses cookies, developers need to use the same philosophy in game design. Certain developments within the game Must change dialogue.
Second, dialogue has to be written in a manner consistent in how people speak. In real life noone goes off very often on long tangents and speeches, its more of a constant back and forth, most of the time, what is said is already obvious and predictable. In real life people often repeat themselves, they often miss what you say, they often may contradict what you say, and so on. Its rare I see this in video games. More often we are treated to Lucas Starwars 1-3 type garbage dialogue. ugh..
Its funny, at this years E3, The Last of US, really jumped out at me for the very reason that its dialogue rang very true and naturally flowing. Its amazing how immersive good dialogue can be.
Personally I would love to see what someone like Larry David would do with dialogue in a game, but then again, most of his dialogue is not scripted, as he understands how hard it is to write .
Conversations were also much more integrated into the gameplay and narrative; what you did, what you said, had significant bearings on how elements of the story played out. Even the order in which you did missions would influence conversations and change who you meet/what you talk about. It was much like the choose your own adventure analogy but with a vastly higher degree of branching and interconnectivity.
The hitch was the sheer amount of data. Something like 12 hours of dialogue was recorded and for any given play through, you'd only hear about 4 of it.
One of the things we're trying with the Storybricks system is to make dialogue one form of interaction among many. "Give ObjectX to PersonY" is one kind of interaction, and so is "Say [sentence] to PersonY."
While this makes it easy to create parrot-like repeating dialogue exchanges in Storybricks, it's also possible to create patterns of interactions that don't repeat. Maybe more interestingly, the effects of interactions -- including particular conversation paths -- can be both remembered by NPCs (by setting emotional states) and transferable to other NPCs (who can pick up on another NPC's emotional state as a trigger for state changes of their own).
There are a lot of limits right now as we're still iterating on features. But we're finding that the idea of treating dialogue as one kind of state-changing interaction among many offers more freedom for interactive storytelling than just the dialogue tree.
It's not a complete solution. There's no rules-based system for generating emotionally-plausible NPC conversation strings on the fly. So some designer-writer still has to hardcode all possible dialogue strings. But abstracting conversations as interactions that can drive state changes does enable a kind of "NPC memory," which is at least a step in the direction of a more natural-feeling storytelling environment.
FWIW, Chris Avellone (of Planescape: Torment and Alpha Protocol) has had some nice things to say about Storybricks.
Part of it just comes down to the amount of content required - I think dialogue trees work much better for large games where emphasis is more on quantity of total experienced content than nuance and detail - but it's truly impressive how much you can accomplish with just a few variables per character if you set a system up in an intelligent way.
Alpha Protocol was also a really great dialogue experiment. While I wasn't the biggest fan of some of the interface stuff (timers on certain choices, lack of clear feedback on what each choice specifically led to), the conversations were probably some of the most naturally flowing of any game I've ever seen, and there was a very minimal amount of repeated dialogue. Envisioning conversations less as trees to be explored and more temporally-based can lead to some interesting results, for sure, but of course, it also can limit the amount of interactions possible.
"Sure, what do you want to talk about?"
I hate that so much. Such a conversation hardly happens in real-life.
Anyway, I can't remember where I heard it from but there's this idea where, instead of choosing which line of dialogue to say and figuring out if that choice was sarcastic, a lie, etc. your options would instead be how you want the flow of conversation to go.
So the options would be "A. Inquire about... B. Befriend C. Aggravate" etc. then you would watch the conversation go its way there.
You could say this lessens the role-playing aspect. Clearly this is macromanaging the conversation instead of micromanaging each small line of dialogue.
You might check her thought on modelin conversation flow:
http://emshort.wordpress.com/?s=modeling+conversation+flow
http://emshort.wordpress.com/how-to-play/writing-if/my-articles/conversation/
While it's for if (with text parser) she also touch up other type of interface/interactions but more importantly most of her advice work for any conversation.
http://emshort.wordpress.com/category/characterization-in-games/
Check for beyond eliza talk to, and marvel at the cotillon system she had create with Evans:
http://www.gdcvault.com/search.php#&category=free&firstfocus=&type%5B1%5D=1&type
%5B2%5D=2&type%5B3%5D=3&keyword=beyond+eliza&conference_id=
One problem I often face when dialog in game is discussed is the gameplay function of such mechanics, having information gating de facto reduce the amount of "life" we can input into a conversation, the reason being that the conversation is instrumental to progression. Because of this gating function, flavor came at odd as a distraction from the purpose by upsetting progression, function and flavor interaction stretch the conversation too much more than often. Gating also does not allow deep relationship with npc either. Conversation can also be an interface to gating many other function (trigger for plot advancement, gameplay buff, etc...) but then it became a cumbersome investment/progression mechanics.
To shine dialog should be the gameplay focus, when flavor is actually the goal, but then it might come at odd with the main gameplay (generally destroying everything) as the game try to stretch across different focus. Generally to avoid the gameplay fall apart game made this "diversion" parallel from the main gameplay. In mass effect the fighting mechanics can be rip off without the dialog suffer (nothing important happen during, except it allows to pace more one way information during fight) as the game structure is entirely directed by dialog, the reverse is also true. "Dating sims" whose focus is entirely on relation would made a better investment in complex dialog system beyond information feeding since the relation and its mechanics (conversation) is the main focus. It's all down of the progression, more than often feeder are better than dialog. Of course I leave the immersion factor, especially in open world game where progression has a backsit from exploration, but when the main mean of progressing in the game and overcoming obstacle is "fight" it's hard to fit functional good "sidekick gameplay" that only slow down things. Ultimately, the segregation of dialog choice vs fight in bioware game helpt streamline conversation in their game as they are milestone in progression.
The problem of dialog is therefore also a problem of what function (progression or access) it had in the game, like any gameplay feature, its limitation should fit the actual function.