Tom Martin's Development Log

The following case studies of two recent narrative games (Fallout 4 and Disco Elysium) will analyse how their dialogue systems are rendered through their user interfaces (UI). Of particular interest to me are how a player physically interacts with dialogue, how games visually present text, and how they communicate feedback on player choice.

Fallout 4

Dialogue options are mapped to four on-screen ‘buttons’ which ‘float’ over the conversation. On a controller, these buttons correspond to the A/B/X/Y (or triangle/circle/square/x) buttons; on mouse and keyboard the buttons correspond to up / down / left / right arrow keys, or can be selected by moving the mouse cursor in the general direction of each button. In contrast to Mass Effect, which places its dialogue options around a radial ‘wheel’ at the bottom of the screen, the options are presented in the centre. The Witcher 3 uses a similar system. This has the effect of drawing the eye into the scene’s cinematic presentation, though oddly more towards the chest area.

Once a dialogue option has been selected, the dialogue menu disappears and the characters are animated speaking their lines; voice acting is used throughout. Often there is some basic camera movement during this, but facial animation is severely limited. Emotional responses are communicated through generic gestures (likely motion captured). Options within a dialogue scene are paraphrased significantly, often to a degree so simplistic it’s difficult to predict whether the chosen response will be in line with player expectation.  Dialogue is captioned by default, and appears at the bottom of the screen. Once a dialogue option is available again, the camera refocuses on the player and the 4-option menu appears again, with new options beside each of its buttons.

‘Checks’ – referring to choices that require successful skill ‘rolls’ to complete felicitously – are divided into easy, medium and hard, and rendered in different colours (yellow, orange, red). The systems at work here are more opaque than in previous Fallout games, which would often communicate percentage chance to the player, based off the strength of their relevant skill. In Fallout 4, a player with a ‘low’ Speech skill has a 10% chance of succeeding a hard Speech check, with this number increasing as their skill grows (but never over 100%).

UI feedback is limited; beyond being notified of quest completion, XP gain or change in reputation, there is a single selection noise. The UI design is in keeping with the diegetic ‘Pip-Boy’ inventory and menu system (the player operates a wrist-mounted LCD computer to perform inventory management and character advancement).

Disco Elysium

Disco Elysium presents its dialogue in a vertically stretched text box, uniformly placed on the right side of the screen. New text is loaded in with a swift swiping animation and a typewriter whirr. While the dialogue is present, the game camera can zoom in or out of the game action (generally it is zoomed in, though often some panning is required to properly compose the shot). This places the readable dialogue at the opposite diagonal to the player character and their companions’ portraits, as well as the player’s health and morale readouts in the bottom left of the screen – both these UI elements act as frames for the in-game action. 

Low- and no-consequence dialogue choices are presented in basic orange; choices that rely on or have been unlocked by facility in a particular skill are marked with [brackets] and coloured in the skill’s colour (blue for intelligence skills, purple for psychological, pink for physical and yellow for speed). Higher consequence dialogue choices – those that involve dice rolls – are displayed in different colours. If a roll is repeatable, it is highlighted in white; if a roll is only attemptable once, it is highlighted in red. Chance-based choices are also accompanied by their percentage success rate (16%, 94%, etc), giving the player insight into both the game’s underlying mechanics and the choice’s likely narrative outcome.

Rolls are communicated by a brief animation of dice ‘rolling’, followed by either an orange-red ‘splash’ effect for failure or a blue-green one for success. This splash extends across the screen, briefly obscuring or tinting the action, and unifying both in-game and UI elements in a single colour palette. The failure splash will often result in a loss of health or morale; the UI element in the bottom left ‘flashes’ as this happens.

Portraits of speaking characters or voices are displayed alongside the dialogue box, slightly intruding on the in-game mise-en-scene. They remain there as long as they are speaking, and are replaced when a new character or voice speaks. Some lines (all in the Final Cut release) are voice acted, and character models are animated in response to significant dialogue choices – the player character performing a roundhouse kick, or another character losing faith in the player character.

Takeaways

The immediate physical reactions of characters, though generic, provide moments of compelling feedback. The resultant player experience is much less focused on the game’s writing than previous entries in the series – a source of some consternation for many players. I don’t know that this is all that bad. 

Fallout 4’s reduction of player dialogue to paraphrasings of their ‘intentions’ is certainly an interesting development – I’m attracted to the idea of a player wanting to achieve an effect or take an action, but not necessarily knowing what they’ll say in order to make it happen. It’s slightly botched here by a lack of consistency (some intentions are questions, some instructions, some subjects), and the fact that dialogue is sometimes of the ‘hub’ school and sometimes of the ‘hatch’ school, as per my other case study. With some thought to maintaining consistency in where choices are placed, and an overhaul of when dialogue choices are demanded, this UI approach could help players reach a more instinctual, reactive mindspace.

Disco Elysium is a compelling case study in how to use high-impact visual and aural design, as well as carefully chosen animation, to make player choices feel impactful. It may suffer from the same informational overload as Planescape: Torment and newer games like Tyranny and Pillars of Eternity, but it uses visual design to signpost players towards plot-significant and ‘characterful’ choices much more effectively. Much work has been put into making the dialogue UI as responsive (or ‘alive-feeling’) as the world (and I’ve not even talked about the inventory or objective menu UIs, which are even more stylish!) – it’s genuinely a pleasure to read.

A tension between the two is the degree to which they expose game systems in dialogue choices. DE’s visual gating of high-consequence choice communicates both story progression goals (if I choose this, I will progress the game state significantly) and gives players a visualisation of the game’s structure. The communication of percentage chance also reinforces a sense of character identity and progression. In Fallout 4, this more numerical approach has been abandoned in favour of a more instinctual presentation, with varying results. I like how it allows the player to sit in uncertainty, but the lack of preview concerning even the likely consequences makes for an uneven experience.

In my research for this project, I identified two key approaches to dialogue systems in games. The first system – the so-called ‘hub’ system – has been widely adopted across game genres. The second – which I’ll call the ‘hatch’ system – has been developed almost as a response to the pervasiveness and rigidness of the first. I have also identified contemporary games that use each system, and which provide interesting examples of these systems being used to place players more firmly in the role of ‘actor’ or ‘performer.’

The dialogue ‘hub’

The ‘hub’ system remains the industry standard for designing interactive game dialogue, and has its roots in the earliest instances of branching dialogue – Choose Your Own Adventure books, which presented dialogue options linking to different pages of a printed book. Conversations for the digital adaptation of this system are generally designed to the following rubric:

This system can be characterised by a wealth of apparent choice – the player can pursue many different avenues of inquiry, defining what they want to talk about, with the game (usually) cooperating – but very little actual choice beyond accepting quests or deepening the player’s knowledge of the world and its characters. Options remain mostly static, and any ‘action’-based options are often signposted as ‘leave conversation’ options – ‘begin combat’ / ‘attack’, for example. High-consequence dialogue options (often ‘gated’ behind ‘skill checks’) are similarly bolded out to the player, encouraging them either to act now and skip extraneous information, or hold off until they’ve exhausted all other options. Additionally, the outside game world is often paused or ‘suspended’ while choices are selected.

We can see the power of this system in games like Baldur’s Gate and Planescape: Torment, where the philosophical and emotional beats of the story are delivered through excavating reams of text, both spoken and narrated.

By breaking these up into clear points of investigation, the writers can avoid overwhelming players with information, while also sharpening their attention to key themes or clues; these games also make use of omniscient ‘narrator’ voices on top of character text to add even more context. These detail-oriented, text-heavy games are favourites of players who love reading novels, and owe much to the genre of fantasy fiction and tabletop role-playing from which they draw both mechanical and aesthetic inspiration.

The ‘hub’ system evolved significantly with Bioware’s Mass Effect. This game abandoned visualisation of every line of dialogue in favour of a ‘radial’ system that both paraphrased the content of a player’s chosen line and displayed the likely in-game consequences of that choice (divided mainly between the alignment choices of ‘paragon’ and ‘renegade). This simplification was necessary for a fully-voice-acted game, as well as one that was interested in functioning ‘on cinema time’ rather than ‘on novel time.’

This design represented a significant shift in player role as well. Rather than functioning as the writer of the game’s script, as in earlier Bioware titles, the player now worked more as a director of a film – less concerned with the word-by-word script-work than by triggering character beats and altering the compositional flow of conversation (Mass Effect ran an advanced-for-its-time cinematic camera system). This system has remained largely unchanged for fifteen years, and features in such modern titles as Fallout 4: initiating dialogue triggers a series of shot / reverse-shot camera movements, pausing or slowing events in the game world outside.

This system is obviously great at developing thoughts; it may be less so at developing relationships or emotions, and I would argue that none of the ‘hub’ systems mentioned allow for the player to truly feel like the ‘actor’ of the story.

2019’s Disco Elysium takes a significant step towards correcting that. Though it returns to the word-focused ‘hub’ presentation of Planescape: Torment and Baldur’s Gate, it makes the player feel more like an actor by innovating on the formula in two subtle ways.

Firstly, the game does away with the idea of an omniscient, or at least halfway-reliable, narrator, replacing it instead with an ensemble of voices, representing the protagonist’s fractured internal monologue. Core mental and physical traits like ‘Logic,’ ‘Reaction Speed,’ and ‘Endurance’ are all given voice, but so are more ineffable qualities: ‘Shivers’ and ‘Inland Empire’ representing physical and psychic intuition respectively. Also characterised are the player character’s life experience: ‘Esprit d’Corp’ is the voice of the player character’s level of ‘cop,’ while ‘Electrochemistry’ represents the player character’s facility with, and desire for, drugs.

The more a player relies on or improves these skills, the more that particular voice will be called upon to narrate the action or provide advice. This funnels the player’s experiential attention through the lenses of those voices – touch, hearing, smell or intuition might dominate different players’ stories. This provides for some significantly different playthroughs, but also activates specific imaginative senses. In actor training, the term ‘sense memory’ is often used, and it refers to tying a particular emotional memory onto a sensation (trauma onto the smell of gasoline, attraction onto the texture of chocolate). By experiencing the sense during performance, the actor can relive the emotional memory, and certain actors find different senses more useful for this. DE’s dialogue system replicates this to a compelling degree, providing the player with the experience of being a sensationally specific character.

Secondly, DE introduces time. Mass Effect and Fallout 3 can be rightly critiqued for the dissonance between the fast-paced action gameplay and the glutinous slowdown of their dialogue scenes. The writers and voice actors of these games work hard to convey a sense of immediacy, but no amount of visual or emotional pyrotechnics can distract from the fact that what the player is really engaged in is a Choose Your Own Adventure novel. Even with all the background explosions, climactic music etc, talking still feels like it ‘suspends’ the world. In DE, however, every dialogue choice advances the game’s clock by three ‘minutes’. Bar reading a book or sleeping, this is the only way to advance time, and given that events in the story are tied to particular times of day, and that the central mystery is a time-sensitive affair, your (and your character’s) relationship to time is of extreme narrative significance.

The dialogue ‘hatch’

The second system is slightly more difficult to analyse, as one of its chief components is a lack of concern with showing how it works. As detailed in Jon Ingold’s AdventureX 2018 talk ‘Sparkling Dialogue’, it is concerned with fostering meaningful choice by limiting dialogue options. As such, I have chosen the metaphor of the service hatch to describe it. A waiter receives meals through the service hatch at a restaurant; behind the hatch is the kitchen, and until the meals are ready it remains closed. When the waiter is in position and the correct meals are ready, the hatch opens. The waiter selects the meal they need, and the hatch is closed. It’s difficult to conduct a comprehensive analysis without looking at the system’s underlying code, but it broadly functions broadly like this:

Inkle’s games, like Heaven’s Vault and 80 Days, make heavy use of these weighting systems to provide seamless and context-bespoke options to the player. Dialogue choices are often trinary or binary, and uniformly short in length.

The immediate advantage of this system is that players are no longer choosing between options that might make no sense to them. NPCs ‘act’ towards the player in more complex and subtle ways (certainly more so than Bioware or Bethesda’s alignment / reputation systems). The visual design of the system itself also becomes incredibly flexible, and can procedurally inform other game systems. For example, Heaven’s Vault relies on a procedural camera positioning system which reacts to weights based on player position and dialogue choices in order to determine shot composition; Pendragon reads the state of its game board, including previous player ‘moves’, in order to generate its dialogue options between characters; the player’s moves after a choice further inform the development of that dialogue.

All of this can only be good for story cohesion and player immersion, and the ways in which story systems have been incorporated within games already is inspirational. However, sometimes the exposure of a game’s systems is key to how players ‘grok’ it, and I find that, more often than not, I need someone to explain how the narrative systems in Inkle games are actually working in order to appreciate their elegance and responsiveness. While its intention might be to make the player feel more in line with the character – by limiting player knowledge to the player character’s immediate knowledge – in effect what it does is make the player feel more like a disciplined writer, and rarely (if ever) into the ‘actor’ position.

One game that bucks this trend is Signs of the Sojourner. Now, whether Signs of the Sojourner actually uses an Inkle-style system under its hood I’ve no idea (it’ll be worth contacting the writer and developer should I continue this kind of research for my thesis), but on its surface it shares many similarities. Dialogue responses seem to be heavily context-dependent, and ‘options’ are almost non-existent, beyond who you talk to first in a location. Instead, the player utilises a deck of ‘cards’, each representing different emotional approaches. By matching them with their scene partner’s cards, the player either progresses the conversation constructively, or allows it trail off.

After each conversation, the player must discard one card from their deck, replacing it with one of the NPC’s. This way, their future conversations are informed by their past ones, both mechanically and narratively – most importantly, the player has an easily referenced record of how this has happened. Many ‘hatch’-based games rely upon your ability to retain informational details in order to build up a sense of your character, and expect you to perform consistently with choices you have made during your playthrough. Not every player is likely to be this kind of ‘method player’, and games must provide shortcuts to memory and emotional association in order to remain accessible to attention-poor players – Signs of the Sojourner’s matching system does this with a careful simplicity. Players are reminded of their connections with other characters by the makeup of their deck, and – in an elegant mechanical metaphor – the more it changes the more their relationships with earlier characters shift.

This development log will track and summarise some of the research undertaken during this project.

I began this project with the objective of making a philosophically engaged game, specifically connected to language and dialogue. With most narrative-driven games using dialogue systems to simulate acts of speech, I decided that J. L. Austin’s seminal text How To Do Things With Words would be the start of my research, and might be a source of inspiration in how to approach the creation of alternative dialogue systems. From Austin I would move to John Searle, and then began connecting my own background in actor training and theory to some writings on constructing game dialogue.

How To Do Things With Words (J. L. Austin)

A series of lectures delivered by the philosopher J. L. Austin between 1955 and 1962, this collection forms the basis of much modern philosophy of language.

Central to Austin’s proposition is a categorisation of spoken language into three different acts:

• Locutionary acts, in which the intention or meaning of a speech act is bound up in what is said.
• Illocutionary acts, in which the intention is separated from what is being said. For example, the phrase ‘is there any salt?’ often contains the illocutionary request ‘please pass me the salt.’
• Perlocutionary acts, in which – thanks to particular contexts – a change is made to the world of the speaker. Perlocutionary acts are conditional, in that they only have an effect if specific conditions are met: saying ‘I now pronounce you to be married’ has no effect if the speaker is not legally ordained to marry two people, nor will it have an effect if there is no couple present to be married; if these conditions are met, though, then two people become a legal entity known as a marriage! 

The consequence of a speech act that doesn’t rely on true and false statements can also be defined as either ‘felicitous’ or ‘infelicitous,’ depending on whether the consequence is in line with the speaker’s intention. These will prove useful definitions when thinking about dialogue.

What Is A Speech Act? (John Searle)

Searle extends Austin’s work, breaking down language into rules, propositions and meaning. It is the last of these that most concerns the writing of dialogue. In order to analyse meaning within speech, we must define aspects both intentional (what the speaker is trying to achieve with the speech act) and conventional (what is commonly understood as being meant by the speech act). I would contend that, in the pursuit of directing the player’s attention, game writing has a tendency to over-clarify both of these aspects, and that an opportunity for play might exist in the interplay between them, and perhaps even in their muddying. 

Sparkling Dialogue: A Masterclass (Jon Ingold)

One of many talks and articles written by the lead developer of British game studio Inkle, this one focuses on simple techniques to sharpen up game dialogue.

In his analysis of an excerpt from an Assassin’s Creed game, Ingold makes some gentle criticism of the similarity between the lines – in order to hammer home a few crucial pieces of information, the writer essentially repeats the same line three times. It doesn’t take a huge leap to see some of Austin and Searle in Ingold’s thinking. The dialogue in Assassin’s Creed is locutionary, with its intention almost entirely bound up in its content. The result is dialogue full of information but devoid of subtext and drama. 

Ingold also outlines his own writing approach, which largely consists of three dialogue propositions: accepting, rejecting or deflecting. He maps these to three player types (those who prefer to dig deeper, those who want to get to the point, and those who want to disrupt narrative), but also to what he has found to be historically useful in writing dramatic game dialogue, namely that ensuring every player choice is in direct relation to the previous line of dialogue increases narrative engagement. The other upside to writing in this mindset is that if any programming needs to reference dialogue choices in terms of ‘attitude’, half the work of categorising them is already done!

Instant Acting (Jeremy Whelan)

One of my favourite books on actor technique, this connects elegantly with Ingold’s suggestion of ‘accepting, rejecting and deflecting.’ Whelan maintains that every action on stage (and indeed in life) can be broken down into three reactions – being ‘impelled’ towards someone or something; being ‘repelled’ away from it, or being ‘compelled’ to stay still in relation to it. By slowing down rehearsal (using a technique involving the recording and playback of spoken lines), actors can discover just how much they are impelled / repelled / compelled by each line they say and hear, and so uncover deeper emotions.

While Ingold supposes that certain types of players will ‘tend’ towards particular options (and hence that always writing these three can subtly ensure engagement across all player profiles), Whelan might contend that a player’s choice would always be contingent on what the line meant to them in that moment, and that creating the space for that reading to be fully felt might allow for a deeper, more considered, and more varied player response.

The Actor and the Target (Declan Donnellan) / Actions, The Actors’ Thesaurus (Marina Caldarone & Maggie Lloyd-Williams)

Declan Donnelan’s manual for actors defines that everything a character does (including what they say) is always in relation to another character’s perception of them. We never ‘see’ other people; rather, we see the image of ourselves that they see, projected back at us, and our egos are in a constant battle to change that image. 

Underpinning this work is Max Stafford-Clarke’s technique of ‘actioning.’ Actioning is the process of applying a transitive verb to the delivery of a line of dialogue, referencing both the speaker and the target. For example, a line might sound very differently read with the action ‘I frighten you’ than if it were read with the action ‘I console you,’ yet both might be perfectly legitimate readings of that line. Each action choice creates a new context for performers to react to; the process of actioning suggests agreeing during rehearsal on a pattern of actions to be performed every night. The actions chosen might usefully connect to the theme of the play as well, to provide subconscious emotional and thematic prompts to its audience.

The reason The Actor and the Target is such a seminal manual is because it helps an actor to simplify and clarify what they’re doing when they’re not talking. By specifying what they hope to achieve, and the verb by which they go about trying to achieve it, they open themselves up to the possibility of success (‘felicity’) – or failure (‘infelicity’) – lying in the other actor’s line. Games have rarely engaged with this suspenseful moment between choices; indeed, contemporary titles still struggle with what to do with character animations while waiting for player input, resulting in a ‘waiting at a bus stop’ feel during more fully animated games.

Towards Expressive Input for Character Dialogue in Digital Games (Nick Junius, Michael Mateas & Noah Wardrip-Fruin)

This paper, written by Masters students at UC Santa Cruz, reviews traditional game dialogue inputs and compares them to the work of performance theorists Konstantin Stanislavski and Bertolt Brecht, as well as the traditional Japanese noh theatre, in order to identify avenues for designing dialogue inputs that allowed for greater expression.

The writers share my puzzlement that players, more often than not, are cast in the role of the writers and directors of dialogue scenes, rather than as actors playing a role. Of particular interest to me was their proposition surrounding animation – that, given the advances in procedural techniques, it would not be difficult to have animation responding to different types of Stanislavksian emotional ‘energy’ given off by players due to the quality of their input, in order to create a truly responsive scene. 

Takeaways

This research led me towards some clearer definitions of how I would like to represent or explore Austin’s work in a game. I would like to:

• Open up the ‘illocutionary space’ as an arena for play.
• Reassess the importance of ‘plot’ and ‘progression’ in making a player feel like part of a dialogue.
• Increase player sensitivity to the proximity and energy of their ‘partner,’ using ‘impel / repel / compel’ and ‘accept / reject / deflect’ as rubrics.
• Encourage the player to (either consciously or subconsciously) define the effect they would like to have on their ‘partner’, and find an emotional reaction in whether the partner response is in line with or opposed to that effect. 
• Experiment with a more sensitive and expressive system than ‘clicking.’

With these objectives in mind, I can begin work on the game proper.