Quick hits

Wherein I sling bullshit regarding a few game things, in brief segments.



In the most recent issue of Edge Magazine, Randy Smith's column addresses three approaches to dialogue with AI characters.

1. Natural Language Processing-- speaking to an AI as you would another person, and letting them reason out a response procedurally. He acknowledges that this requires non-existent technology for processing human speech into meaning and Turing test-passing AIs to synthesize that meaning and respond believably. In other words, not in our lifetimes.

2. Dialogue trees. We've all seen these, most often in traditional RPGs like Fallout or Planescape. You have a limited list of authored dialogue options, to which the AI replies with a limited list of authored responses. This is effective for what it is, but very transparent to the player.

3. Single-word query response. The system used in old Ultima games wherein the player could type in any single word they liked and test whether the AI "knew" anything about that (ie, whether a response had been authored for that query.) This allowed the player to feel out the possibility space more organically, but presumably leads to a lot of interactive dead ends ("I don't know anything about that" in response to all but the handful of recognized queries.)

Smith prefers the third option overall, and I think there have been examples of games that mix 2. and 3. fairly successfully: Sam & Max Hit the Road, for instance, presented dialogue topics as single icons, so you knew which topics were valid (no dead ends) but didn't know exactly what your character was going to say about it. As you encountered new concepts they were added to the icon list. Similar are Mass Effect or Chronicles of Riddick Escape from Butcher Bay, wherein the dialogue tree options are single words or short phrases-- you choose the gist of what you want the character to express, and the author takes over from there.

But in any case, I agree that some version of the classic dialogue tree is the best we've got, and that probably finding ways to make it feel more natural is our best bet in the short-term.

I think dialogue trees probably feel least natural in games where the player character is a pure cipher-- in which the character is as close to being "me" as possible. So in Fallout 3, for instance, it feels limiting because the tree doesn't contain all the options that "I" would want to say in that situation. Conversely, in Escape from Butcher Bay, Riddick is an established character with a strong personality. It feels more natural for the dialogue tree to only contain options that "he" would say. (Commander Shephard in Mass Effect is in similar territory.) So, one valid approach when considering a design with dialogue might be to give the player an established persona to inhabit ("I am Jim. It has been demonstrated that Jim is very timid. He would NEVER just walk up to the cute girl and ask for her number!") as opposed to a true blank slate. This might open up more interesting possibilities in the end than the cipher route, depending on the broader design ("I played in such a way that Jim became much more confident over time, and by the end had totally different dialogue options available for that same cute girl.")

Alternately, the stranger in a strange land could be useful. Maybe the player character is a recent immigrant and hardly speaks any of the native language. You can only talk to people who know some bit of your own language (hence why every person in the game isn't a dialogue target) and then can only communicate using rudimentary phrases, pointing, and gestures. It's a familiar real-world paradigm for any player, and could again open up interesting avenues of play involving indirect communication.

Or maybe the problem could be externalized more literally: the player character is the only human left alive in a world filled with robots, and those robots only understand specific inputs depending on their preprogrammed role.

In any case, taking the dialogue tree as a mechanical constraint, perhaps strides could be made by making our chosen fictional context less perilous for dialogue ("I'm whoever I want, talking to whomever I want") and instead couching our limited dialogue systems in situations that complement them. I do wish more games allowed you to converse with characters in the world, and I hope we'll see more player-driven dialogue going on in games outside the traditional RPG space.



Warning: this segment contains spoilers for a very minor subplot in the Grand Theft Auto 4 downloadable content, The Ballad of Gay Tony.

In The Ballad of Gay Tony (TBGT from here on) you play as Luis Lopez, an ex-con and current manager/business partner at a string of nightclubs in Liberty City. It's established early and often that Luis is a promiscuous guy and not interested in commitment to any one woman. When you check your email in the game, you find multiple messages from "Margot," a girl that hooked up with Luis in the past and who has developed an unhealthy obsession with him.

Partway into the campaign, if you drive to a small icon on the map, you can meet Margot. In a cutscene, she expresses her obsession with Luis, and he explains that he's not interested, and that he thinks she's crazy. It's comic relief, but it turns fairly dark when Margot reveals she's taken a whole bottle of pills as a cry for help. You drop her off at the hospital and complete the mission. Still played for laughs.

At another icon, you meet Margot again, this time on the third-floor walkway of a shopping center on the pier. She again expresses her obsession, and another funny exchange goes on between the two characters. Then she steps over the railing and says that if she jumps, everybody will think Luis did it. The scene stays in the black humor zone, until Margot jumps from the railing, falls three stories, cracks her head on a flight of stairs and dies.

This is pretty surprising and unexpected! And there was nothing you could do to prevent it (except not to visit that nondescript little side-mission icon.) Help text pops up onscreen: "Everybody thinks you pushed her! Get out of there!" Civilians in the area start yelling that you pushed her, that you're a murderer. And your only option is to run until you're out of the danger zone, at which point the mission completes successfully.

And while there are a LOT of things the player can do in Liberty City, the entire premise for this side-mission exposes the boundaries of interactions that are possible in a GTA game. For one, you have no control in a cutscene. I'd have loved the ability to talk Margot down, or to jump out and grab her hand and keep her from falling. But, alright, Luis is Luis, not me, and drama is often based on the viewer being unable to avert terrible events, so there you have it. But then you're forced to flee the scene of a suicide that you're the primary witness to. Even if you wanted to, there's no "stop and explain" button. If I didn't know it would've resulted in a failure state, I would've wanted to just turn myself into the police. The scene is begging for an entire sequence wherein Luis is explaining himself to the arresting officer under a hot interrogation light; wherein he's on the stand at trial-- does his testimony outweigh that of the other witnesses that day? What about the forensic evidence? Can the medical examiner prove she was or wasn't pushed? Was she in psychotherapy at the time? Hospital records show she was suicidal-- she was taken to the ER due to an overdose of pills not more than a week or two earlier, and it was you that dropped her off. Isn't that worth something? What will the jury say?

Of all the situations in GTA that result in police sirens, Margot's subplot seems the most defensible-- you didn't shoot a cop, you didn't blow up a plane, hell, for once you didn't do ANYthing. But GTA doesn't support the player being innocent. And so, in raising more questions than it's capable of answering, Margot's sad, surprising little story is an interactive bridge too far. But if it could be supported with mechanics... man, wouldn't that be exciting?


Loss of Progress

Challenge in games is a weird thing. Especially when I tend to find a lot of games too challenging to be fun, even on normal difficulty.

The thing is that the moment-to-moment challenge itself isn't the issue. I enjoy difficult problems and things that I actually have to try hard to figure out, or build skill to overcome. The problem is with loss of progress as punishment, and the drudgery of retrying the exact same sequence over and over again.

Recently, for instance, I've been playing Left 4 Dead 2 in singleplayer mode. Granted, this isn't the central focus of the game (that's co-op or versus.) But friends aren't often playing when I am, and I prefer not to play with random jokers. And I just enjoy singleplayer games-- I like having that time to myself, just me, the gameworld, and the experience. So I pick the Singleplayer menu option and away I go. And I can't clear a single campaign on normal difficulty.

It's not because any one part is too hard on its own. I wouldn't want the enemies to be wimpier or myself to be hardier, as the experience would feel pointless then. And though I do die too often, I think I could clear most chapters after a few retries. But the design of the game's systems actively discourages me from retrying by stealing away a criminal amount of progress if my character dies.

So, part one. Many co-op focused games are made significantly more punishing in singleplayer mode in that the AI is unable to revive you the way another player can. This is true of Left 4 Dead, it was true of Gears of War. Survivability of a multi-human party is increased due simply to the developers being unwilling or unable to account for AI partners reviving the player when killed (either by using defibrillator paddles or freeing them from a spawn closet.) When you're dead, you're dead. Sorry. So the single player is just that much more fragile, and backed up by AI characters that are generally less capable in some ways than a human player.

Part two: each chapter is probably 20 minutes long or more on average, meaning you can die at the end of a level and lose fully 20 minutes of your life. The level restarts in the saferoom and you're staring down 20 minutes of rote retread of the exact same environments and scripted sequences you already cleared, just to get back to the point where you failed. And if you fail again, you're back to square one in every regard. When the player fails in your game, you want them to instinctively feel the desire to jump up and try again. Saying "the last 20 minutes of your life have been erased. Do it over" does not engender this response.

So, mitigation. Maybe each chapter has a mid-mission checkpoint. Or, hell, I'm playing alone, let me save wherever I want. Or maybe when my party is wiped out, all four of us respawn in the saferoom without the level being reset: you're back to basic equipment, but all those hordes you cleared and scripted gates you opened are still in their completed state. It becomes a corpse run across empty ground instead of a full instance reset, which is much more palatable. There's a penalty, but the penalty isn't "do all that over again."

In many games, failing and restarting discrete challenges makes sense-- a racing game or fighting game or puzzle game, where each stage is micro and retryable in a few minutes, or even GTA4 wherein each mission is relatively short, generally has mid-mission saves, and lets the player instantly retry from the last checkpoint on failure via the cell phone. In games that have a more fluid sense of progression this is much less the case.

I don't mean to rag on L4D2 so hard. It's just my current example of a phenomenon I've completely outgrown: a game turning into a joyless then frustrating slog because of lost progress on failure. Even in the core action space, we live in a post-GTA-hospital, post-Vita-Chamber world. It should be as challenging as you want it to be to make material progress, but forcing me to go through the motions when I die, redoing challenges I've already proven I can overcome just to try again, shows an outright lack of respect for the player's valuable time.

It's not to say that loss of progress can't be available to the player as a challenge mechanic: BioShock has an option to turn off Vita-Chambers, and Rock Band offers No Fail Mode if that's what you're looking for. I love games with robust options menus. Let me turn "restart chapter on death" on or off to fit my own playstyle. Plenty of games disable achievements or unlockables when difficulty modifiers are applied; so be it.

Of course, these are pretty pedestrian observations about a fairly specific genre. What about unlimited time rewind? Prince of Persia Sands of Time only seemed to limit rewind as a form of challenge. Halo 3 seemingly saves the state of the game continuously, allowing you to revisit and fly around "footage" of your play session after the fact. Doesn't this imply that the design could reasonably allow for the player to choose to rewind only as far as they want to on failure? Braid supported this, opening up the game's challenge to be entirely based on the player's ability to reason out a solution to move forward, not to restart the level to retry that one tricky jump again, as you had to in the game it takes as its inspiration, Super Mario Bros.

And of course, zooming out further, what about games that have no explicit failure state, where reverting to an earlier state to retry a "challenge" is outside of the player's expectations entirely? Games like The Sims or Animal Crossing, where small failures or setbacks within the interactive frame are generally deemed acceptable by players, because they can always be undone through further play? I'd love to see more games unask the question of checkpoints, quicksaving and loss of progress by removing death and failure from the experience entirely, instead of searching for less abrasive ways to tell the player "you have failed."


Those turned out longer than I'd expected.


juv3nal said...

1. Natural Language Processing-- speaking to an AI as you would another person, and letting them reason out a response procedurally. He acknowledges that this requires non-existent technology for processing human speech into meaning and Turing test-passing AIs to synthesize that meaning and respond believably. In other words, not in our lifetimes.

In this regard, I wonder what you make of Facade. Clearly not "true" NLP, and more like an NLP-looking front end to an option 2/3 hybrid, but IMO still a significant (if woefully un-emulated) alternate approach

Unknown said...

In my opinion, Mateas and Stern made a mistake going down the NLP route with Facade. It's hard enough developing a viable interactive storytelling framework; heaping another unsolved problem on top of it is crazy.

Sam Crisp said...

@juv3nal: http://old.idlethumbs.net/display.php?id=181

Melf_Himself said...

Restarting from the safe room is definitely annoying in Left 4 Dead. Of course, the timing/composition of enemies will not be the same the second time so I think a larger "time rewind" is more justifiable there compared with other games. Any reason why you don't want to play it on easy?

Miyamoto is one of my (and most people's) favourite game designers. Interestingly in interviews he says that it is mandatory that the players slog through *some level* of repeated content after dying. He chooses his checkpoints so that this always follows a nice little difficulty progression. This quickly ensures that the player has their confidence restored before being plunged into the hard part again.

Of course, there is a limit to how much you should have to slog through, which depends on the particular game.

NeoDodge said...

@Melf : I have my own reason, which may or may not be close to Steve's of initially not playing on easy (though I've gotten lax with this rule as of late) and it's quite simple : I've played this kind of game already. I've been there, killed my share of cacodemons, solved so many riddles and crossed a fair number of finish lines first. Isn't if safe to assume that I have what it takes to engage this game the way it was supposed to be played in the first place ? L4D ain't no trailblazer, really. There's no reason not to believe oneself fully able to tackle its challenges with a baggage of more than a few FPSs, survival horrors and Horde modes.

Johnnemann said...

Some quick Googling didn't turn up the game I'm thinking of, but there was a recent indie adventure game that gave the player a "dialog inventory system", in which concepts could be carried around like objects, and used to initiate conversations with NPCs and unlock new conversation options. I like that idea a lot because it fits the mental model I have for conversations, and allows the developer to gate concepts in a natural-seeming way. Of course, it falls prey to some of the same problems as other systems in that it doesn't allow the player's natural reasoning or hunches to be explored, unlike most single-word response systems, since the player has to find the concept in the game world before being allowed to have dialogs about it.

Anonymous said...

But isn't removing the consequences of a failed challenge getting close to removing the challenge itself? Lots of complaints were made about the latest Prince of Persia when it did that.

The Vita-Chambers in BioShock always bothered me. I don't remember turning them off, but I rarely used them. Even if they helped a couple of times, their presence bothered me the rest of the time. They felt like videogamey things artificially added to the world. I can only think of a few games where I liked how savegames or checkpoints were integrated into the lore: "Jesus Saves" and GTA's safehouses in general, the Gamsaav in Outcast, couches in ICO.

Back to removing consequences and Left 4 Dead 2. I think L4D2 is a game that one shouldn't necessarily be play to complete the campaign. I haven't played the it yet, but in the first one multiplayer was clearly the important part and single-player felt like training for the multiplayer. It didn't feel like I should play the single player campaign to progress from the beginning of the game to end. I also didn't play it much: part of the campaign, then a few multiplayer games with friends, then completed the single-player campaign later. I haven't touched it for a long time, mostly because I too have a difficulty finding suitable times to play with friends.

Melf_Himself said...

"L4D ain't no trailblazer, really. There's no reason not to believe oneself fully able to tackle its challenges with a baggage of more than a few FPSs, survival horrors and Horde modes."

The fact that you find the game difficult but have a good horror/FPS C.V. argues the opposite. It clearly does break new ground. What Left 4 Dead did is make a game that is all about team work and where "skillz" such as twitch aiming and such don't come into it much.

Some people don't want to worry all the time about other "people" and didn't enjoy the game. I'm kind of the opposite and really get into the team-oriented games, so I found the going fairly easy.

This gets away through from the original reason that I asked why not just play it on easy... For me, the answer (for any game) is just that it's a psychological kick in the nuts. The easiest difficulty level should be called 'normal' to prevent pride from getting in the way of choosing the best game experience for yourself, imo.