A Conversation with a Work of Art

Simulating true conversations in a computer game is tough stuff. We're still some way away from effectively applying computational linguistics to game playing in a way that allows a wide range of natural language input, and I'm not even sure that's an entirely desirable goal given the enormous complexity that this would introduce into game design. So for now the general approach is to restrict the range and format of player input using a system that is easily interpreted and applied, such as a "click to talk" or multiple choice dialogue tree system.

The question, however, is this: when you play a game that implements conversation, do you feel that playing through the conversation contributes in any way to gameplay? Or does the conversation feel like more of an afterthought that requires little attention or skill on the part of the player (or the developer, for that matter)?

The answer, I think, largely has to do with the goal of conversation in the game. It can be a method of providing character or game backstory, or maybe a break in the action to advance the storyline. It is also used to present the player with a choice, such as a side quest or a broad game pathway. But more often than not, the conversation is brief and shallow, without requiring a great deal of thought from the player, and without considerable effort put into things like conversation flow, structure, or context. For the most part, given the limited role of conversation in the game, these things are unnecessary and not worth the added developmental effort.

Conversation doesn't have to be limited to these kinds of functions, however. In many interactive fiction games, conversation takes on a much larger role: it directs the narrative; it offers challenges to the player; it contributes to deeper character development -- within player and non-player characters and between them. Conversation might not be nearly as much fun as actions requiring fast-twitch hand-eye coordination which are rewarded with explosions and rag-doll physics, but I have to believe that there are some methods and mechanics in IF games that could translate to non-IF games to enhance the experience, so why not conversation?

I think any discussion of conversation in games, using IF as a foundation, would have to start with Emily Short's Galatea. Of all the IF games of the past decade or so, Galatea is, I would guess, the one most played by people who would not identify themselves as IF players, perhaps because of its visibility and renown. But interestingly, it is also one of the only games I know that consists entirely of a conversation, one that takes place between the player and the one non-player character in a single room.

It's a deceptively simple construct for a game which does an excellent job of hiding the sophisticated mechanics underneath.

Galatea primarily uses the traditional ASK/TELL system for conversation with the NPC, although the majority of the time is spent ASKing rather than TELLing. Because ASK HER ABOUT and TELL HER ABOUT can be shortened to A and T, respectively, most of the player's efforts are essentially thinking up and entering topic words to discuss. Other occasional interactions are allowed, such as HELLO and GOODBYE, which are interesting inclusions; while they do help newer players get started into the game, they also represent direct statements to the NPC (as opposed to commands for the player-character), which might throw off some players into expecting other direct statements to work. Nevertheless, the basics of the ASK/TELL system are familiar to most current IF players and are not difficult, I expect, for newer players to pick up, which is one reason why I think this game succeeds in attracting a broad audience.

Still, if the game consisted only of the basic ASK/TELL system, even with its impressive range of topics to discuss, it would likely fail to keep players interested for long.

Where Galatea succeeds is in Short's dogged attention to the underlying mechanics of the conversation. The NPC in this case is not your typical information vending machine which spits out its one preprogrammed response to each topic. Here Short has taken into account things like mood, flow, and context, such that the same question about one particular topic has a different effect and response depending on the current established relationship with the NPC, what questions were just asked immediately prior, and what topics had already been discussed to that point. It's a mechanic that sounds entirely logical to a normal conversation but one that is rarely attempted in games, probably due to the complexity of design.

But does creating complexity of conversation such as this really come across as far more daunting a task than the prospect of creating hyper-realistic graphics and physics systems?

As a very simple example, in the game you note that the NPC--a statue that has become animated--moves as though she is breathing. If you ask her about her breathing, she will tell you that she taught herself to breathe, and although not necessary she does so because it is soothing. Later on, if you then ask her about sleeping, the question (not the answer, but the question from your player character) is then worded: "And do you have to sleep? Or did you teach yourself, the way you taught yourself to breathe?" But if you hadn't previously asked about breathing, the question is worded only: "So do you sleep?"

The effect on the narrative, in this case, is quite minor. But there is an elegance to it, a quality that makes the conversational narrative flow more smoothly and naturally and that shows that the game isn't merely spitting out canned text but rather is constructing context-sensitive and appropriate responses that give the impression that the game is reacting dynamically to what the player is doing and has done.

The mechanics, of course, extend far beyond decorative adaptations.

In her essay on conversation systems in IF, Short describes Galatea's system as "topic quips with mood tracking and quip-tagging," where quips are the actual snippets of dialogue used in the game. So here we have single topic quips like with most ASK/TELL systems (one predetermined player character snippet for each ASK or TELL command, instead of multiple-choice options), but she includes quip-tagging, so that once quips are used they are not repeated. In addition, some quips can be used only after other quips; in fact, the same command (e.g., ASK HER ABOUT THE ARTIST) can produce different player quips and NPC responses depending on what other topics had been discussed previously.

Mood tracking is also fascinating, in that the mood of the NPC is tracked and changes in response to certain questions and answers, which can then impact responses to subsequent questions later on. The NPC's mood is also reflected in different ways, including gestures, expressions, and tones of voice, and the system was apparently designed to accomodate these dynamically into the text of the response.

There are, of course, some limitations of this type of conversation system. As mentioned in a previous blog, the ASK/TELL topic quip system can be frustrating at times, because the player is only allowed to specify the topic of discussion rather than the specific aspect of the topic to discuss; ASK HER ABOUT THE ARTIST is an extremely broad command that could result in asking her what the artist's name is, what he looks like, or what her relationship with him is like. Generally, Galatea does a nice job keeping the questions within the current context, but there remains a degree of helplessness on the part of the player.

Another limitation is that conversations with this type of system are notoriously one-sided; the NPC in Galatea is almost entirely reactive, rather than active. This makes the resulting conversation seem only like an interview, with the player asking one question after another, but therein lies one of the difficulties of designing a conversation with a system that uses topic quips rather than multiple-choice dialogue trees: how can we easily allow the player to answer questions, in addition to asking them?

Nevertheless, this game is a notable achievement that allows players to have a conversation with an NPC that reads like, and has the feel of, a very dynamic, responsive interaction with a character that is far more three-dimensional than any character from a mainstream game. The amount of effort that Short must have put into this is frightening, although since the source code was never released, most of this is based on interpretations and assumptions in addition to her writings. Still, with all of the effort and money already put into various complicated aspects of mainstream, AAA-quality games these days, I have to wonder: what might a team of Shorts create with a AAA engine, a fistful of investment capital, and a couple years worth of effort, just by utilizing some of the basic mechanics that already exist in games like Galatea?

Conversations with NPCs

As Corvus Elrod likes to say, compelling stories arise primarily from the relationships between characters. Although these relationships can be generated or expressed in different ways, I think it's fair to say that conversation is probably the most obvious and frequently used method in games. Yet it's interesting to note that conversation systems in games are fairly rudimentary and, in many cases, pretty unsatisfying.

There are many reasons for that, of course; human conversation can be horrifically complicated to deconstruct, and dynamically generating realistic and meaningful conversation with computer-controlled characters is still years away, especially when you factor audio into the equation. As a result, most conversation systems in games are simplistic representations that often follow tight scripts and leave little room for exploration, which is probably fine with most developers; it's difficult and time-consuming to create sophisticated interactive conversation, particularly without forcing players to use text-based input, and most text entry in mainstream games was abandoned some time ago. In the end, conversation is typically an insignificant component of gameplay, demanding barely more thought from players than a cutscene might.

I don't have nearly the range of experience with games that many other people have, but I would venture to say that the conversation systems in most graphical, mainstream games fall under one of two general classes:

1. Click to speak. The simplest system, where you just click on an NPC or hit a single key to trigger conversation. The player typically has no control over the topic of conversation or what the player character says, generally or specifically.

2. Multiple choice dialog trees. Players select from a limited list of speech options, which helps direct the conversation to some extent.

I'm sure there are mainstream games with variations on these that I've missed or forgotten, but they are less common. Nevertheless, I have yet to see a conversation system in a graphical, mainstream game that is particularly satisfying. Most games that use #1 generally are just using conversation as a means of advancing a storyline, without really incorporating conversation into gameplay in any significant way. Systems related to #2 do at least involve some thoughtful input on the part of the player, and because the choices presented are often specific player quotations, the resulting conversation can seem much more natural while also providing some degree of characterization for the player character. On the other hand, this type of system also encourages a "lawnmower" approach of trying all options, even if it requires re-initiating the conversation or using the save/restore approaches. There is thus little consequence to making those choices, especially if the different conversation branches just lead to the same end point anyway, as is often the case.

It's probably clear to readers here that I think interactive fiction has a leg up on other game genres in many areas, and conversation is one of those areas. But why is that? Is it because the medium has more conversation tools at its disposal? Is it because text-based input (and output) allows a greater range of conversation possibilities?

I don't necessarily think so.

A while back Emily Short created what is essentially the seminal piece on conversation systems in IF, and there's little reason to rehash what she so comprehensively covered. For those not as familiar with IF or her essay, I'll just summarize the most common IF conversation systems, which mirror those mentioned above in graphical games. The three most common are:

1. TALK. Players just type TALK TO to initiate conversation with the character, which generally results in a single back-and-forth exchange with the NPC (though sometimes more, and sometimes less). Not unlike system #1 above for graphical games, where the player just clicks on an NPC to trigger conversation without any say in what is discussed.

2. Topic-based conversation. The most common form is the ASK/TELL system, where players type ASK ABOUT , or TELL ABOUT . Here, players can direct the conversation by specifying which topic to discuss. In many iterations, players are not prompted with specific topics to choose from, which forces them to think about which topics might be fertile to discuss. Alternatively, some games use a variation on this approach, where conversation topics are highlighted in the text, and players need only type the topic itself to trigger conversation on that topic. That said, often the topics used are too broad; players may intend to ask about a specific aspect of a topic, but are compelled to rely on whatever was programmed by the author. Another issue is that NPCs often end up feeling more like information booths than living individuals, dispensing conversation like a vending machine without much of a two-way dialogue.

3. Multiple choice dialog trees. Same as mentioned earlier. Unlike with the ASK/TELL and TALK systems, players have a pretty good idea what specifically they will be saying or asking with the NPC, rather than generally. Particularly in a text game, this can provide a better degree of characterization. On the other hand, all potential choices are patent, leaving little challenge for the player, particularly if the conversation branches all lead eventually to the same end point. Also, as mentioned above, the UNDO or SAVE/RESTORE commands encourage the lawnmower approach, which can lessen player involvement and remove much of any challenge that is present.

These basic systems are not all that fundamentally different than those used in most mainstream graphical games, and within the IF realm they have their advantages and disadvantages. To me, though, there are three main reasons why I think conversation in IF generally works better than in other game genres: skill, creativity, and patience.

As I've suggested before, I think that perhaps it is a consequence of these games being created largely by writers rather than teams of programmers, designers, and artists, and I'm guessing it also has to do with the fact that most IF games are not commercial and lack the associated deadlines and budget constraints. It takes a lot of skill and time to write good conversation that creates depth for characters and their relationships, and even more to write it within a system where it is often pieced together at different points in time based on player actions. That takes a lot of planning, design, and testing. In some cases it demands some creative modification of the conversation system being used, but for the most part good, meaningful, and challenging conversation in games can be created with the tools and systems that are already out there. The more I play different IF games that have well-crafted conversation, the more I have come to realize this, and the more I believe that mainstream, graphical games can learn a lot from these implementations -- if only the developers would have any interest in it, which is to say, if only they thought that players would have any interest in it.

I also think the nature of the medium and the IF community is to encourage experimentation with different types of systems, and the result is that there have been some pretty creative solutions to what is (still) a complicated and mystifying issue. Sometimes that involves systems that combine different elements; Vespers, for instance, combines the TALK system with the ASK/TELL system, while a game like Short's Pytho's Mask combines topic-based conversation with multiple choice dialog trees. But often there is more that is designed under the hood, so to speak, in order to generate conversations that flow smoothly and are consistent with the events of the game regardless of the player's prior actions.

Over time, I'll try to review some of the games that I think do well at implementing conversation, with a focus on the mechanics, but also with a focus on the question of whether conversation systems like these could have a role in the more mainstream, graphical games industry. In the meantime, I'd be interested to hear if there are any mainstream games that people have encountered in the past that they felt utilized an effective conversation system.

Super Cthulhu Crunch Time

I've been away from the blog for a little bit now, as I've been in super crunch mode at work lately. I've been away at meetings all week, and I have a huge grant due tomorrow evening...which means I'm looking at an all nighter on this last night before submission. It got me thinking about what to best name this crunch time, when I came across this little gem on the net called "Cereal Killers", a blog with spoof cereal box artwork. Some great stuff there, including "Hellfire Blazin' Bran", "Sugar Frosted Skulls", "Shredded Feet (with Frosted Toes)", "Rice Kreepies", and of course, the aforementioned "Cthulhu Crunch".

(Honorable mention to "Tinkles" [Who peed in your cereal this morning? Tinkles did!]).

Given that this mad crunch time has been (and will continue to be) an exercise in extreme horror and evil, I believe Super Cthulhu Crunch can be appropriately named the Official Cereal(®) of my grant application week.

I'll be back next week with more appropriate content. Not too long ago I thought I would start a review of different conversation systems employed by various IF pieces (mostly by Emily Short, with a few others like those used by Aaron Reed) to see which ones I thought worked best at accomplishing different things. I started off with Short's "Pytho's Mask", which is a fascinating piece that has a lot to offer in terms of conversation implementation. I think a lot of other game genres can learn from the techniques used in IF games, which is actually one of the reasons I started the Vespers project.

In other news, I go ten days without blogging and my subscriber number (according to Feedburner) has at last, somehow, broken through that magical glass ceiling of 100. Maybe if I put off the next few blogs for a little while longer I can get my subscribership even higher. Thanks everyone!

I Wanna Hold Your Hand

I Wanna Hold Your Hand: This month’s Blogs of the Round Table invites you to explore a relationship within a game that you found compelling or memorable.

As Corvus has admirably asserted numerous times, "it’s not the characters themselves that make for compelling stories, but character relationships." I still contend that compelling stories owe at least part of their success to interesting and deep characters themselves, although admittedly I lack the skills to make a coherent argument to this effect. Nevertheless, I do agree that character relationships are the core of any good story, and it's a great topic to focus on for the Round Table since so few games have really embraced this concept.

Relationships certainly do exist in games, but they exist in much the same way that characters themselves typically exist in games: as relatively thin implementations that lack any significant depth or complexity. Part of that, I suspect, is because relationships between players and NPCs are very difficult to craft, for a couple of reasons: first, relationships that involve the player have to take into account the fact that the character is really a hybrid of the protagonist in the story and the player himself, which presents all kinds of difficulties in establishing a particular persona while simultaneously allowing the player's own feelings and decisions to be expressed; and second, creating and implementing meaningful relationships that have real impact on the story and the player requires considerable skill, effort, and time, which often doesn't rank up there on the developer's priority list alongside slick graphics and sound effects. Relationships that go beyond a bit of backstory and a few multiple-choice dialog screens is tough work.

Not to sound like a broken record, but I think the game genre that has one up on the others in this area is interactive fiction. In fact, I think what I said for the last Round Table similarly applies: "Perhaps it is the nature of the medium, perhaps it is a consequence of these games being created largely by writers rather than teams of programmers, designers, and artists. I'm not sure, but I know I have yet to play a graphical game that deals with the nuance of character relationships the same way that some IF games have so far been able to." The italics, of course, added now for relevance.

I recall once again some of the best known recent works of IF, pieces like Galatea, The Baron, and Anchorhead, which just serves to reinforce the fact that these games are so well known and worked so well largely because of the attention the authors paid to crafting the characters and their relationships.

Galatea is essentially a piece about relationships: the player's relationship with Galatea, and her relationship with her creator. It's a good example of a relationship that is expressed through gameplay, as the course of the game is really an exploration of the relationship between player and NPC through conversation. That's essentially one of the real challenges of playing Galatea; trying to establish a particular relationship with the NPC, so that you can learn more about her and her relationship with her creator. It's a clever piece that I would say is one of the strongest examples of a game that puts relationships at the center of its design, and provides a multitude of channels through which you can explore them.

The Baron and Anchorhead are two IF games that use particular relationships between the player character and NPCs as the basis for exceptional, well-crafted stories -- the rescue of a kidnapped daughter in The Baron, and the saving of one's husband from a bizarre family "tradition" in Anchorhead. In both cases, the relationship itself is not explored to considerable depth through interaction, but the game itself serves as a mechanism through which one gains an appreciation of the relationship with the family member. In fact, one might argue -- particularly in Anchorhead -- that the sparse interaction with the protagonist's husband is actually an effective means of providing engagement with that relationship, and serves to create tension during gameplay.

There are also a number of IF games that use the exploration of the relationship between the player and the player character as a central focus of gameplay. Three pieces that come to mind are Emily Short's Glass, Adam Cadre's 9:05, and one of Jason Devlin's "other" games, Legion. In these games, it is not clear at the start of the game who (or what) the protagonist in the game is, and a clear focus of gameplay is to figure that out. It is the kind of relationship that I think could only really be effective in a text environment, and these and other games of this kind make for a fun and different kind of experience.

There are many other works of IF that explore relationships in these ways, which is one of the reasons I find the genre so fascinating and fun. Most of the games mentioned, by the way, can be played online -- I suggest trying a new web-based IF platform called "Parchment", and clicking on the game you want from the list.

But -- as I'm sure you're waiting breathlessly for -- what about Vespers?

Vespers does include relationships as a central focus of the game: the protagonist's (the Abbot's) relationship with each of the remaining brothers is key to the game and the story, and it is the exploration of these relationships through interaction that provides much of the gameplay experience. Just before the start of the game, the Abbot chooses to close the monastery to the local villagers in order to try and keep the Plague away, a decision that was not supported by many of the brothers. When it becomes evident that the decision is a poor one, it puts a strain on the Abbot's relationships with some of the brothers, and this provides some of the tension in the game. There is also the relationship between Matteo and Lucca, which provides a great deal of background for the story, and the player's relationship with the young girl who appears at the monastery is, certainly, the central focus of the game. But you'll have to see that for yourself.

This post is a response to the June ‘08 topic from Blogs of the Round Table. You can see other entries on this subject in the drop down box below, which will update automatically with each new post.