The Lost Art of Conversation In Games

By Oluf Pedersen [09.10.09]


The idea for this paper came about in the most satisfying way: during gaming. There I was, playing oldies from the '80s on a Commodore 64-emulator during the small amount of vacation time I had in the summer, when during a particular game, Law of the West, I encountered dialogue choices that were very interesting and had quite a profound effect on gameplay[1].

I was especially impressed by these choices of interaction, because at the time I was fooling around with Oblivion (yet) again. And this old C64 game from 1985 had much more interesting dialogue choices than the 2006 smash hit (and many other newer RPGs/adventure games).

This experience got me started on a short path to try and discern, first and foremost why this could be? Why wasn't Oblivion full of interesting conversations? And was this a general trend in newer games? And was I the only one who actually cared about this?

The answers to these questions were: Because the developers of Oblivion chose otherwise, due to design issues but also probably because the game was also being developed for the Xbox 360. No, it's not a general trend as such; lots of newer games have conversations (although whether they're as interesting as in LotW is an open discussion). And no, there exist tons of forum messages praising dialogue heavy games.

Having answered these questions I found myself, as most people do when having answered questions they have posed themselves, with more questions. And these questions are the basis for this project/paper:

  1. What happens when we analyze conversational choices in games? Does there appear to be a connection of sorts? A framework?
  2. What, if anything, has been written about the subject (Artificial Conversational Intelligence (ACI) in computer game)?
  3. Is ACI useful at all in enriching the game experience? Since maybe it has not developed significantly in 20-25 years, does this mean it's not that important? The cases for and against will be stated.
  4. Do developers of computer games overlook the single most important method of communication in human history?

Questions 1 and 2 will be answered through analysis of games and papers respectively. Question 3 will be answered through the answers to 1 and 2 plus specific independent research concerning the matter. The last question will be answered as a summation to all the previous questions and as a conclusion to the paper.

I will analyze games known as RPG's/adventure games, because these types of games are usually the only games to actually feature dialogue options.

Law of the West (1985)

Law of the West is quite a simple game wherein you play the role of a sheriff in a Wild West town. The game consists of encounters with different people whom you, the sheriff, have to handle according to your own moral compass. Thus you can choose to shoot the lovely Rose, below.

Through the conversations, which take place with the player choosing from four different responses, how to reply to any given person, the characters in the game really spring to life; there's the seemingly innocent brash young lad with his new shotgun, the lovely, but as we have already seen on the screenshots, defiant Rose, the alcoholic Doc who isn't best pleased with you gunning down people all the time, the very annoying deputy and many more.

At the end of the day's encounters you get a summation telling you how you fared: how many people you killed, how many females you charmed, how many bullets you took and so forth. And then the game restarts with exactly the same encounters -- replayability is obviously not that high. Although given that there are about 10 encounters and each encounter spans about four rounds of replying, and that the dialogue options change according to previous choice, it would seem that there are 4 * 4 * 4 * 4 * 10 = 2560 lines of dialogue all in all. I haven't tried them all out however, so I cannot be sure.

This sort of tree-structured dialogue course will henceforth in this paper be known as the (aptly named) "linear"-sentence-structured-dialogue. It is called linear due to the fact that the dialogue options don't change in a given encounter; they are always the same.

Pool of Radiance (1988)

Pool of Radiance was the first licensed D&D-game in SSI's legendary "Gold Box" series of games. It doesn't have elaborate dialogue choices; it's either yes/no-questions, typing a single word (passwords) or in certain encounters the choosing of the "tone" of the reply (as seen above). The "correct" choice of tone for your reply depends on who is asking (orcs for example need to be told of, any sign of weakness causes them to attack, whereas giants need to hear more diplomatic words).

So you get to choose a word to symbolize your response to any given question.

The dialogue options also sometimes change, depending on whether the player has visited a certain place or acquired a particular item.

This type of dialogue interface will henceforth be referenced as the "Dynamic" keyword-structured-dialogue, dynamic in the sense that it changes according to places/people/items that the player has experienced.

[1] Here literally the way the game is played.

The Secret of Monkey Island (1990)

The Secret of Monkey Island is a classic graphic adventure game from Lucasfilms' SCUMM series (Script Creation Utility for Maniac Mansion).

It contains many dialogue scenes, most notably the swordfights which are decided by who has the most insults and snappy comebacks. These insults and snappy comebacks are learned only through hearing an insult or the comeback to that particular insult. This makes the learning process rather tiresome, and the only fight that rises above average is the one against the Islands sword master, Carla (pictured above), where you hear new insults and then have to match those to your existing comebacks.

The dialogue choices are different than in LotW, in that they change according to several things: places you've been, people you have talked to or items you've acquired. Thus if you've spoken with a person before, or if you've found something which helps solve a quest, they respond differently and you are presented with (slightly) different answering opportunities.

Again we are no longer talking about a strictly "linear" structured sentence-based dialogue, but instead a "dynamic" one. Therefore, this type of dialogue interface will henceforth be referenced as the "dynamic"-sentence-structured-dialogue.

Fallout (1997)

Fallout was a very successful Role Playing Game which spawned an entire franchise.

The dialogue in Fallout is quite unique. It's divided into two different interfaces, and it also features a graphic representation of some of the characters faces along with voice acting. The normal interface is like that of LotW/Monkey Island, with the player choosing sentences from a list. The second is the opportunity to ask people about anything you like. Obviously the word parser is not particularly strong, and you will only get (useful) information if you ask about specific places or groups/individuals within the game. But this is still a brave attempt at combining existing types of dialogue interfaces in computer games.

Another interesting aspect of the dialogue in Fallout is the inclusion of a specific statistic in the player creation, Speech, which when trained (with experience points), gives the player more and more dialogue options, and therefore more options when solving any given quest.

Fallout's dialogue interface is mainly like that of Monkey Island, but with added complexity, as it has another "dimension" (Statistics) in terms of what influences the dialogue options. The word parser is a brave inclusion, but doesn't really influence the game a whole lot.

Planescape: Torment (1999)

Torment is a favourite with many RPG fans, and features tons of dialogue. An unofficial Italian translation recorded 1.4 million words and 68,510 dialogue sentences[2].

It is (like Fallout) also published by Interplay, and in many respects, the game is quite similar to Fallout: the graphics, the music, the environments, the mood of the game and lastly, the dialogue options, which as mentioned in the above paragraph were very extensive.

The word parser from Fallout is gone, but instead it features more statistics that influence the dialogue options. A high intelligence, wisdom or charisma all influence the dialogue options and in the end your path through the game. Dexterity also gives you options, but the choices relating to this statistic are not dialogue-like but more action-like, for example, you get to choose to break a pickpocket's neck.

A new dimension when compared to Fallout is the fact that your party-members now also affect the dialogue options. This means for example, that if you have the character Morte, a floating skull, in your party, this gives you certain extra dialogue opportunities in specific encounters. Sometimes the party members would also have small conversations with each other.

Another new dimension is the fact that you get to choose a "political standpoint". Throughout the game you come across several different groups/factions that differ in their world view. You encounter amongst others half-crazed anarchists and fully-crazed religious zealots and get to choose who, if any, to side with. But siding with one faction will give you problems with others, and therefore also change your dialogue options.

Throughout the game the player finds different magic items that affect the protagonist's statistics. It is also possible to have tattoos made that reflect his adventures. These tattoos also affect his statistics. So overall the player's choice of weapons, "wardrobe" and "body-decorations", might influence the possible dialogue options. This is, if not an entirely new dimension (since it's based on Statistics), certainly another layer to that specific dimension.

The dialogue type in Torment is like that of Fallout, but with extra weight added to the dimension relating to how statistics influence your dialogue options, and two new dimensions in that who you travel with, and your choice of "political standpoint", also influences your dialogue options.


Baldur's Gate II - Shadows of Amn (2000)

Baldur's Gate II was yet another big RPG favourite published by Interplay; it was much more focused on dialogue than the first Baldur's Gate, and gained much acclaim for this. Especially the way the party members interacted with the player and each other will probably strike a cord in the memory of many RPG-fans.

Screenshot taken from Sorcerer's Place (

It also added another three new dimensions to what could influence the dialogue options: at least on one occasion the race of the player could have an impact on a conversation. In the market-square of Athkatla, the player could meet an elven-hating man, who would insult the player if he/she were elven or half-elven or if a party member was (he would also actually attack if one had the dark-elf, Viconia, in the party).

The second new dimension was that the player-characters class (fighter, thief etc.) would determine what possible "stronghold" the player could get, which in turn meant different dialogue options based on class.

And lastly the gender of the player-character determined not only possible love-interests, but also how certain situations in the game played out.

However, the dialogue options were not as diverse or dependent on statistics as in for example Torment. I can only recall a single instant where a statistic influenced the dialogue options (Wisdom, the Spectator Beholder in the Sahuagin City).

The Elder Scrolls III: Morrowind (2002)

The Elder Scrolls III: Morrowind was an epic game and one that took "close to 100 man-years to create"[3]. The sheer size, scope and beauty of the game awed gamers at the time, and made individuals like myself, purchase new computer hardware in order to run it properly.

The game's dialogues consist of two parts, the general one which is a keyword-structured interface, and a second, rarer one (sentence-structured), where actual dialogue is chosen. The options for the second one rarely number more than three (yes, no and maybe later), while the keyword-structured responses rarely number less than ten as the game advances.

The skill "Speechcraft" affected your ability to bribe or simply admire someone, so that they liked you better, and therefore might reveal something they normally wouldn't.

So the dialogue in Morrowind actually works a bit like the dialogue in Fallout: both have two systems in place and both use a "Speech"-skill to open up more conversational opportunities. Here the similarity ends however, as Fallout's sentence-structured-dialogue is much more detailed than Morrowind's likewise. This is on the one hand a logistic necessity due to the sheer number of people populating Morrowind, but also ultimately a design decision. It would not have been impossible for example to have a few characters in each town that stood out more from the general public of Vvardenfell (the island where Morrowind takes place).

Vampire: The Masquerade - Bloodlines (2004)

Bloodlines differs from most other games mentioned here in that it's not just a RPG/Adventure, but also a first-person shooter or more basically, an action game. It was built using Valve's Source Engine, and was released at the same time as Half-Life 2. Unfortunately the game was developed while the engine wasn't finished, and it was also rushed out, meaning the game was full of bugs when it first saw the light of day.

In any case the game wasn't a commercial success, and its developer, Troika Games, closed down soon after. Which was really a shame since the game (patched!) was the best action-RPG since Deus Ex.

The dialogue in the game is quite influenced by statistics and race. Three different forms of influencing the dialogue are available (Persuasion, Intimidation and Seduction), plus one racial trait (Dementation or Domination).

Probably the most interesting way to play the game is as a Malkavian, a race of vampires that are plagued with insanity. This insanity makes the dialogue options completely different than whenever playing the game "normally" with any other race.

Another extreme race to play in the game is the Nosferatu type of vampire, hideously deformed creatures that cannot be seen by mortals, unless they wish to reveal their supernatural heritage (bad idea). In practice this means you cannot initiate dialogue with normal people and that you are constantly travelling and hiding in the dark and in the sewers ... obviously not a choice recommended for socialite gamers.

Overall the dialogue options in Vampire: Bloodlines are heavily influenced by race and statistics, but also one's "political preference" plays a significant role in the dialogue options. The game's first-person view and graphics engine, also offers something different from the rest: the very visual depiction of the characters you encounter, I state very visual, because Bloodlines really lets you see people's nervous eyes or twitches or simply their general state of mind (angry, happy, sad, surprised, unsettled etc.). While this doesn't influence your dialogue options in terms of choices available, it certainly adds a great deal of extra flavour to the dialogues themselves as they play out, and why you choose a certain response rather than another.


The Elder Scrolls IV: Oblivion (2006)

The Elder Scrolls IV: Oblivion is perhaps the most successful RPG of all time; the amount of space, quests and possibilities in the game is quite phenomenal. Mix this with stunning graphics utilizing HDR, a great musical score, the new Radiant AI (allowing inhabitants of the world to seemingly have lives of their own) and other blessings, and it's no wonder the game has reached the level of popularity and infamy it has.

Yet for all these amazing possibilities and technical achievements, the NPC's still feel quite lifeless and bland. I write still because the same was true of its predecessor. Even though the NPC's have needs and schedules, it's just not possible to have an interesting conversation with (almost) any of them. The dialogue structure from Morrowind, where you have two systems in place, a keyword-system and a sentence system, is gone. It has been replaced by a merging of sorts: most of the time you are choosing either keywords or very short sentences. This is probably attributable to the game also being developed for the Xbox 360 console.

But at least there are some decent facial expressions to go with the dialogue, and the game features a mini-game involving this. Oblivion, like Morrowind, only really implements the basic adventure dimension, and to a small degree the Statistics dimension, in its dialogue options.

Neverwinter Nights 2 (2006)

Neverwinter Nights 2 is probably the game with the largest amount of influenceable dialogue ever created. The number of different statistics affecting dialogue options must number at least 10 in the original campaign[4] itself, and they are quite diverse: from trying to bluff or lie your way out of an unhappy predicament, to using your "business sense" in figuring out how to tax merchants near your keep "appropriately".

Whatever one might say about NWN2, it is definitely a step forward for the dialogue options front. The quality of the story or dialogue in general may not be as good as in Planescape: Torment or Vampire - Bloodlines, but the sheer number of influenceable dialogue options available, makes the game's situations and characters feel very "alive" and responsive to player development and individuality, even though the game itself, and its story in particular, is quite bland and linear.

The game is heavily reliant on the statistics and companion dimensions, and also naturally the "classic" adventure game quest dimension as seen in Monkey Island.

Whatever race and gender (and maybe class) you choose for your player-character also influences your dialogue options a little bit. And as a new option you get to choose the characters "background trait", for example if you were a farmer, a militia member or the town bully, and this also influences the dialogue options (at least in the tutorial level).

As in Planescape: Torment, your weapons and wardrobe influence your statistics, which in turn influences your dialogue options: so if you put on a certain hat (as seen in the top screenshot), you might get more dialogue options.

[4] (in the NWN2-toolset the theoretical number is 6 different ability score checks plus 28 skill checks, and they can be combined ...)

A framework for games with dialogue/keyword options

I definitely think that a distinct pattern emerges from all these games. If one takes LotW as the starting point, it becomes clear that dialogues in games have evolved since then. Games with dialogue choices today are usually much more dependent on previous player choices and behaviour. And a game like Oblivion is not representative of the options and possibilities normally given the player in games with dialogue choices. The developers of Oblivion obviously didn't think this part of the game to be worthy of as much attention as for example the graphics, and in looking at the sales and reviews of the game, it's difficult to argue with this.

My research leads me to conclude the following typology of games with dialogue choices:

Type 0 is not included in this analysis, because of its rarity today, because of the fundamental differences from modern "graphic" games (and choices) and lastly because it has already been thoroughly discussed in Nick Montfords book "Twisty Little Passages".

In Type 2 I find the following dimensions influencing the player's dialogue options:

  1. The Adventure Basics - Places/people/items (First observed by myself in Monkey Island).
  2. Stats I: Interior (Fallout).
  3. Companions/Associates (Baldur's Gate I).
  4. Gender (Baldur's Gate II).
  5. Race (Baldur's Gate II).
  6. Class (Baldur's Gate II).
  7. "Political view"/Personal compass expressed through dialogue or actions (Planescape: Torment).
  8. Stats II: Exterior (Planescape: Torment).
  9. Upbringing/Background trait (NWN2)

There are problems with these dimensions. It would be possible to claim for example, that the player's choice of gender, race, class or upbringing are all simply statistics, but then what isn't a "statistic" in a computer game? I think one has to look at it from the gamer's viewpoint and not the programmer's. And the gamer will not view for example his or her player-characters race as similar to what the character does for a living. To the player, and indeed to most people, the two are inherently different. If for example I were to insult someone based on their profession (e.g. "Damn lawyer!" or "Damn electrician!"), this would be acceptable behaviour to most, but if I were to base my insult on race or appearance however ("Damn Jew!" Or "You're fat!"), most people would frown upon it.

The separation of the Statistics dimension into the interior and exterior layers was necessary, because it is different to have a skill as for instance being very persuasive, to say putting on a hat, getting a tattoo or carrying around The Hammer of Seductiveness.

The seventh dimension here is a bit difficult to define properly; it could well be argued that it's a mixture of the basic dimension plus the statistic dimension, but I strongly feel that it should stand alone, as it's more than these two dimensions put together: it's how and why your character acts and lives, it's maybe even their fundamental raison d'etre. It is therefore also very much beyond being simply good or evil (or lawful/chaotic), and thus many computer games do not implement this dimension.

These dimensions are obviously related, but they don't necessarily occupy the same time-frame. I will therefore structure them accordingly:

Before entering game-world:

After entering game-world:

The avatars background, behaviour and development are all clearly linked, but exist in different points in "time and space". The background provides the foundation for the behaviour and subsequent development, but this is clearly quite a one-way relationship.

The behaviour affects the development of the character, which in turn (might) affect future behaviour, in what is a much more dynamic relationship.

In time however, one's background grows (hopefully!), since the present becomes the past, and so it's not an entirely static or one-way relationship between background and behaviour/development, according to this model, the player's Profession, Personal Compass and Stats (I) might change.

There are also a number of "free radicals" floating about here. How to define the accurate graphical depiction of NPC's in for example Bloodlines or Oblivion? Or good voice acting, which may lend an entirely new feel to an otherwise read sentence?

Also, how does one measure each dimension's influence on the gaming experience? Does one use a scale from say 1 to 10? And are all the dimensions equally important, or is the avatars personal compass for example, more important than its race?

Literature on ACI

There is a lot of literature on ACI in computer games. Unfortunately, none of it really deals with what has been done in past games or indeed, what is being done right now in new games (except "Twisted Little Passages" by Nick Montford, where he deals with older text-based games like the Infocom titles). Most articles[5] are about the problems and dilemmas concerning possible future development of autonomous agents, which have a near human intelligence and knowledgebase, and may converse about almost everything.

One such article is the aptly named "Can we talk?" by Bill DeSmedt and Don Loritz. In this article they take it for granted that the development of autonomous agents will revolutionize computer games because of the influence these agents will have on player immersion.

The main problem for this article, and indeed most of the literature on the subject, is what I would dub the "Why?"-factor. Why would highly able autonomous conversational agents improve gaming? Because it seems to me, that people have a hard enough time listening to real people in the real world, so why on earth would they want to discuss for example existentialism with a gnome in a fantasy setting? And moreover, why build/license an autonomous agent to simulate such a conversation, when it's already possible to do so, with existing technology?

Obviously the autonomous agent would also be able to address other interesting subjects the player might want to discuss, but the basic question remains: would the majority of players want to discuss any "deep" subject altogether with any NPC?

Another problem would be how the agent would receive input. I think typing can easily be ruled out, and dialogue choices would necessitate a lot of writing.

It might be solved with speech-interpretation software/hardware, but there is no guarantee this would work in a gaming context (and what about all the natural pauses in conversations?).

I do feel that autonomous conversational agents are an interesting prospect, but I can't help thinking that it's going to be either a novelty that quickly wears off or even worse, the final nail in the coffin of what is basic human-human interaction. Why talk to your friends about daily life for example, when you could design your own friends, and take them adventuring into wonderfully crafted virtual worlds? Why indeed have real life friends at all?

Yet another problem concerning autonomous agents is the time-frame involved. The authors of "Can we talk" mention a number of 25 years from one of the projects (Cyc) involving these agents. This today would mean about 15-20 years from now before we see this technology implemented in games. So 15 to 20 years of waiting for something that might be useful, if even technologically possible or feasible at all ... I'm not impressed.

All in all the literature about ACI to me seems very strange; not that it isn't interesting to speculate and argue about the future, and not that there's a huge general interest in AI-projects in general, but why no-one (publicly anyway) has taken the time to delve into what has been done in games in the past decade, what has worked, and what we can learn from it?

Metaphorically speaking, it seems like people are starring at the sky, when they should be tending the earth.

The only real source of information and opinions on conversations in computer games today, which I've found during my long search, is mainly in the internet forums. So only at grassroots level does one find useful insight into the subject as it stands today.


So what do the forum-members say? Well, I'm in no doubt that they would all unequivocally support as many dimensions influencing dialogue options as possible (my subjective interpretation after reading hundreds of posts on the matter of dialogue in computer games). Why? Because it gives them a feeling of the dialogue adapting to them, as individual (role?-) players, and not the other way around, where the more linear dialogue controls and restrains them.

They also like well written dialogue, and would like quite a lot more of it.

This view is probably also held by readers of Gamasutra who recently held a vote on quantum-leaps in RPG's[6]. The two games at second and first place in the top 5 were Torment and Fallout respectively. Due in no small part I'm sure, to the dialogue and dialogue options available in the two games.

Against this, one could argue that forum members (and Gamasutra readers) only constitute a minority of gamers, and that they are usually the more "hard-core" gamers.

The majority of gamers, it might seem, don't care that much about dialogue choices and options available. This is certainly the conclusion when looking at the success of so-called RPG's like Oblivion or Morrowind before it (the Gothic-series also spring to mind). Indeed most games in general do not include changeable dialogue options or even dialogue options at all.


The conclusion, as a summation to the past three points must be that: no, developers have not overlooked dialogue options in games (at least not to a point that I can prove they have). Certainly the new game Neverwinter Nights 2 proves that there definitely is a large amount of focus (from some developers) on the subject of dialogue choices. The newer games that do not implement much in the way of dialogue choice, obviously do so with the blessing of the majority of their audience. I may lament that fact, but I too was stunned by the wonderful visuals, and the sheer number of quests and opportunities which awaited me in Cyrodiil.

As for the framework, I consider it a good attempt at mapping the player's experience of what influences his or her dialogue options, and why they do so. It is however by no means perfect, as specific dimensions are certainly debatable.

The relevance of ACI is in the eyes of the beholder. I might appreciate lengthy well written dialogue like in Torment, but this is not every gamer's dream. Also, with current consoles and the fact that their displays are often an "ordinary" TV-screen, there are clear problems with long-winded (written) dialogue on them.

So in the end, it boils down to a simple cost/benefit-situation: "Who's your audience?" and "What's the platform(s) of the game?", to name but a few of the players in this equation.

The future

For me personally, I would like to investigate further the findings of this paper, in particular the unanswered questions regarding the framework. I also look forward to some hands-on training in working with the NWN2-toolset.

The framework also opens up for a possible new view of creating stories in games. If one views the dimensions as possible story-building blocks, then maybe the story in a game would feel more "natural", due to it being built from that particular game's dimensions (dialogue-engine). Instead of trying to force a "classic" story onto a game, the storyteller could build the story within the game's limitations so to speak. However, this is obviously not in any way a proven point, but certainly something that I will look into in the future.

For the industry, I think the future will hold, not autonomous agents in quite a while, if ever, but instead more dimensions in general, and a better "flow" and focus on each one of them. Through testing and common sense it should not be difficult to ascertain which dimensions are more important than others, and it should also be possible to detect/feel whenever a particular dimension might be useful to enhance the story or the game-play in general.

Literature List

"Can we talk?" by Bill DeSmedt and Don Loritz:

The Gamasutra Quantum Leap Awards: Role-Playing Games:

Penny Drennan's webpage on "Dialogue in Computer Games":

Game List

Law of the West: U.S. Gold Ltd., 1985.

Pool of Radiance: Strategic Simulations, Inc., 1988.

The Secret of Monkey Island: Softgold Computerspiele GmbH, 1990.

Fallout: Interplay Entertainment Corp., 1997.

Planescape: Torment: Interplay Entertainment Corp., 1999.

Baldur's Gate II: Shadows of Amn: Interplay Entertainment Corp., 2000.

The Elder Scrolls III: Morrowind: Bethesda Softworks LLC., 2002.

Vampire: The Masquerade - Bloodlines: Activision Publishing, Inc., 2004.

The Elder Scrolls IV: Oblivion: Bethesda Softworks LLC., 2006.

Neverwinter Nights 2: Atari, Inc., 2006.

[5] Try this site for typical articles concerning ACI:


Return to the web version of this article
Copyright © UBM TechWeb