Textual Procedural Generation and Narration: Generalities, Opinions, Tips and Perspectives.

This is the transcription of a lecture given initially at Game Camp France, in Lille. To attract people, I called it "How to get rid of your writer and their huge ego"

Around this provocative idea we will partially explore the world of text generation and see some unexpected perspectives.

Imagine you’re a game creator and you’ve delegated the narrative to a writer. This is a good thing because they’re a professional, but now you have problems because their ego overrules your decisions (as a writer myself, I know all about that).

The role of the writer

What is the purpose of a writer when creating a video game?

The first things that immediately come to mind are: background, scenario, mythology, interaction, dialogues, descriptions ...

But we rarely think about the “textual environment". If you have a sound designer in your team, you won’t  try to ​​make your own sound for a button: you  ask your sound designer to do it.

However, in many indie productions (if not all) in which I’ve participated, all interface texts were written by the UI manager, the graphic designer or the GD. After all, an "OK" button doesn’t need to be handled by a writer, right?

But in truth, why not? It's their job, their skill. For the textual interface environment, there are sometimes conflicts between the interface expert, who wants a clear and meaningful text, and the writer, who will try to stick to a context and try to tell a story.

In Out There that I created with Michael, I designed a button labelled "Encounter Life" for when a player wants to explore a planet and approach its inhabitants.

The wording is rather clumsy and English-speaking players may have guessed that it was simply a bad translation. In truth, in its French version, it’s even more clumsy: "Meet the life". The intention was not to be clear, but to be mysterious, solemn and evocative.

An interface must support the gaming experience but also transcend it. If your game is narrative, give your writer a little freedom with your textual environment.

AAAs, stuck in their quality processes, will not do it. Indies can do it and, as only they can, they must.

A crucial second aspect of the writer’s job is the one of "gameplay texture".

Writers often complain of being relegated to the background in the VG. There is indeed something to complain about. How many times have we been presented with a game, for example a tactical finished at 90% and we are told "well, can you make us some story for this game? ". We would have liked to be there from the beginning, in order to bring some kind of elegance, combining ways and means, instead of just making up a story because a VG is supposed to have one.

I wouldn’t say that when the writer's work is required (we will try to eradicate the writer from the process in a minute), it is more important than the rest, such as the code. I will say it is just as important, no more and no less.

However, incorporating it in the early stages of design can be an asset in developing unique gameplay.

Recently, I had to do the GD and narrative work for a spy game called The Sigma Theory. If I only worked as the GD, what would I have thought? Spies, Jason Bourne, car chases, double agents, secret documents ...

My role as narrative designer has meant I’ve learnt a lot about geopolitics for developing a scenario: I discovered the ambiguous relations of states with terrorism, cryptography, the importance of drones, interpreters, diplomats, the very special personality of each great power ...

Of course, all these new ideas have greatly fueled my GD work.

In GD, the golden rule is: "Know your universe perfectly, the rules will emerge on their own. "

And who knows your world better than the one who’ll be writing the background?

Some aspects of textual narration

Localization

The most important advice I could give is to not to write too much.

I could develop this advice on many philosophical and artistic levels, but the fact is that the deployment of your indie game must take into account the localization costs.

In our productions that are very narrative and textual, we limit the volume to between 35K - 50K words. At €0.10 cts/word, this represents, say, €4000 per country.

Localizing a French game into English, Russian or Chinese is obvious given the size of the markets. But into German, where the majority of population is able to speak English? Into Italian, where the market is not that big?

The excellent game 80 Days does not include a French version. It has 500,000 words, which makes it equivalent in size of Les Miserables by Victor Hugo. If a French (or Italian) translation is ever envisaged, a return of at least € 50K on the territory concerned would be needed. A careful balancing act ...

Text as a vector of subtlety

Many talented creators (that have a fat budget for this) try to convey complex emotions or feelings through 3D cutscenes.

It’s a complicated issue. It’s often unbalanced, more or less mimicking a movie narrative which is a heterogeneous field of storytelling. They pay the price of being pioneers.

Meanwhile, a simple text can convey concepts that are difficult to illustrate otherwise. Text alone can convey a feeling of accomplishment, happiness, determination, and most elusively, the thing around which the whole world is focused and which is so absent from our productions : love.

The text - interface

We saw that an interface included text and that this text could set an atmosphere.

Now the text may be a full-fledged interface component, describing something more appropriately than another means, such as an icon.

The rogue-like ADOM for example suggests you create a hero with a race and statistics. At the beginning of the game, it synthesizes the player's statistical information to make a biography. For example, a character who grew up in the fields will be tough.

Similarly, the Ultima series generated a character and stats based on questions like "what would you do in this context". These questions, asked by a mysterious gypsy, set the atmosphere and combine relevance and elegance.

Plunging deeper into the fractal nature of the meaning of the text, the font also tells a story: Oblivion's Caroline places us in a context of medieval cities and battles of lords, while the icy-white Helvetica condensed from Skyrim reminds us at all times that we are surrounded by snowy mountains.

It's a bit like a Freudian interpretation of a dream: when we talk about text, everything might mean something.

Grammar

American and French players today have different experiences of the same (textual) game, largely due to the grammar.

(I’m not talking about adjectives. I have a note to make on French language: as a French writer I prefer the phrase “orange flowers” (des fleurs orange) to “white flowers (des fleurs blanches)”, because orange is strictly invariable (it doesn’t take s in the plural form), while white (blanc) not only takes a plural s, but in the feminine form it turns into blanche. Red (rouge) does take an s in the plural form but stay rouge in the feminine form. Coding French flowers is coding exceptions!)

We inherited the grammatical forms of the 80’s games : "Talk to" in English and "Parler à" in French.

What does an American think when they click on “Talk To”?

Are they ordering the hero to go and talk to that person? Or are they adding an implicit “attempt” because they more invested in the character in the first-person? I don’t know.

In French we have a very neutral infinitive meaning "talk to" whose translation in English would be "To talk to". We establish a list of neutral instructions: the avatar essentially undergoes a form of programming. We have a more analytical than emotional relationship with the game and it is the result of complex grammar. If we used the imperative "Prends" (Take!) instead of "Prendre" (To take), we would expose ourselves, in text parser games, to difficult situations, simply because the average French player hasn’t  fully mastered the imperative tense for all the verbs.

Using "talk to" and "parler à" in the creation of our games for the last 40 years I sometimes feel like we’ve been eating the same dish or  have been making love the same way for 40 years.

The future, the past, the third person singular ... why not use these conjugations, these modes and these times to offer a unique narrative? There’s still much more to be done!

Procedural generation

Combinatorics

A common strategy for generating text is to use combinatorics, if possible under conditions.

When in Skyrim a guard comes to you saying "Hello Orc, can you help me enchant something?” we have three parts in the sentence:

“Hello” or “good evening”, related to in-game time
“Orc”, related to the player's race
The following sentence is related to a player's skill. If they don’t have one, the guard can always tell them about a knee accident ...

What can we conclude from this example?

First, we can create many sentences using little basic material. Our dream of getting rid of the cumbersome writer is progressing.

There is a real "wow effect" the first few times. The game reacts passively to the identity of our character, and it's impressive.

However, we are talking to players, humans who can read patterns in the feedback of a game, for some it might even be their main skill.

As such, there is a transparent systematism in such a combinatorial application that paradoxically leads to a loss of immersion. This supports a larger idea, which we will see later, which suggests that a character must be written or at least give the illusion of being written, even in a world that wants to be systemic

Finally, and this is perhaps the most delicate point, the information transmitted is weak. The Skyrim guard is a mirror that evaluates my progress, and nothing else. In this particular case, there are many ways to improve it.

In the end, the exercise of the combinatorial sentence can be condensed into a single rule of quality: "Am I transmitting new and coherent information as part of a corpus large enough not to tire the reader? "

Sheherazade

There is a technique based on the balance of the story, which I call "Sheherazade" and which combines soap opera techniques with more classical show architectures (like X-Files).

Imagine that you are a writer for a crazy video game creator and they ask you to write one million quests because, according to them, the balance of the game depends on this number - and you don’t want to go through a procedural generation system.

It’s possible to give the illusion of a game that offers 1,000,000 quests by writing a rather small number of them, let’s say 800, and using various techniques, this number can be lowered even further (400).

X-Files (or Star Trek) has an interesting structure: it’s composed of a main arc whose key episodes are at the beginning and at the end, and there are one or two in the middle. All the rest are independent episodes, which you could also watch in any order.

Many RPGs are based on a similar structure. The main arc is the main quest, and the intermediate episodes are the side quests, which can be done in any order.

Next, imagine that you turn on the TV and it’s airing an episode of The Young and The Restless. A person walks past you and exclaims "Wow, that’s season 16 episode 32! ". You’d certainly be amazed, skeptical or amused, because it’s difficult, if not impossible, to know the episodes of the series well enough to  identify them so precisely.

We can actually play on our brain's inability to perfectly memorize a story or quest if we follow soap opera writing conditions.

The idea is that by the time we complete the 800th written quest, we won’t remember the first one. So, we can, yes, loop on the same body of stories, ad lib.

For that there are 2 rules to follow:

It must be absolutely in the same context: a small American city, science fiction, medieval fantasy. The more generic we are, the more efficient the technique.

You don’t have to be that talented if you do not create granularity in the narration. The quests must follow the same rhythms and be, like the TV episodes not of the main arc, easily interchangeable.

A job offer for this type of work might look like "We are looking for a talented writer who can write the blandest stories possible! "

Can a story series without granularity (except internal episode rhythm) really be of interest? Ask audiences that love soap operas.

But it is pretentious to believe that all players want exciting and engaging stories - many just want to relax, and not add to their thoughts the burden of virtual adventures. And this market is, without doubt, the largest and the largest one yet to conquer.

Turbo Sheherazade

There are ways to shape these techniques more efficiently too, such as using holes in prewritten stories - and then adding some combinatorics. This drastically reduces the already low number of stories needed to be efficient - but here we encounter a financial limitation, because writing 400 stories doesn’t cost so much.

Then there is the pure approach of procedural generation. The techniques are particularly numerous and diverse, but as a whole they simply consist of generating stories using algorithms.

A valid approach, particularly in the context of processing TV series licenses (or something that looks like that), is a simulationist approach. It’s possible to design a screenwriter simulator in the same way we design a missile launcher simulator: they usually have the same reference works, operate on standardized rhythms, and apply the same recipes.

When we reduce this simulation design to the quest design niche, the work is even more reductive: McGuffin, exploration, combat (or alternative test), reward, eventual twist or conclusion.

The independent pen-and-paper RPG scene is very rich in books for organizing the algorithmic generation of quality quests - for example Eureka and Masks published by Gnome Stew.

The promise of infinite quests with a controlled background is appealing but at the end of the day presents the same cultural pitfalls as the AI ​​deep learning generation that we’ll discuss next. I’m pessimistic about their value in the short-term.

Broken hearts

Usually, when I have an experimental game idea, I ask myself two questions:

1 - Is it possible?
2 - Is it fun?

Independent developers are driven by passion and pleasure. The emotional and intellectual attractiveness of the work is important, because they can’t be sure they’ll be paid after all their hard work. 

I’ve got a note of warning about the algorithmic generation previously mentioned.

A few weeks ago, I wanted to write a love story on the following principle: a girl sends me some texts every morning, wishing me a good day, inviting me to talk, to enjoy life, sometimes teasing me, spiced up sometimes with a little cybersex ... and to do so, I wanted to use my techniques to make this an endless game.

Trying to do this was hellish: it’s very painful to write about love, a particularly sensitive and intimate subject, algorithmic piece by algorithmic piece: it’s like slowly completing a puzzle without the big picture. It’s an awful job that takes you away from the initial subject and is absolutely depressing. 

So, when we come to the quantum leap of textual algorithmic generation, we must also ask ourselves the question:

3 - Will I enjoy doing it?

I like to believe that the love we put into the work has a direct influence on the quality of the output. We must therefore be cautious about the methods we use.

Deep Learning AI

The technology of neural networks allows infinite number of texts to be created on a given subject, under some conditions.

Before going further, a warning: many professionals more qualified than I believe that it is a technology of the future. I think they are probably right.

However, I have the opposite opinion. 

Why develop an opinion that is probably wrong, then? Because you’ll need plenty of arguments to back yourself up when your client / boss wants to put everything on the line because it will be a trendy, if not relevant solution.

The technology is quite accessible, with surprising results. You feed a neural network a corpus of texts (we’ll expand on that later), and the neural network returns similar texts on demand.

It’s a balancing act to ensure that the rendered texts are relevant and intelligible without being copies of all or part of the original corpus (“overfitting “) but the results are truly amazing. A machine with higher-than-average specs is also needed: we used a PC with 85 GB of RAM to train a simple text generator neural network. Now’s the time to use all these graphics cards that you mine bitcoin with for something really useful!

For the purposes of a math podcast, we created a small conversational AI that even chose its own name: Yurie. It has an answer to everything, answers that follow an internal logic but that are unfortunately rarely useful.

Yurie was used for questions / answers, rap texts, parts of novels, the reconstruction of a language close to French (We trained it to not consider sentences as strings of words but strings of letters), and even a piece of the script of a comic book that will be published soon. By varying the corpus and training techniques, we can create machines that can produce endless text in any form.

Including lore, dialogues and quests related to VG?

Let's take a little look at the constraints of the corpus.

Our Yurie began to speak a good but not overfitted French with a corpus of 3 billion words. This figure may of course vary depending on the language. 

But this opens up three major challenges:

First of all, a large quantity of words must be obtained, and more importantly, in a specific context: you could feed your AI with all the copyright-free texts written before 1950 that you wanted, but that won’t help it express itself using in the medieval, slangy or technical way you need for your text.

There is no direct control over the results that will be produced. Trained with texts written before 1950, Yurie has become a sexist entity:

Yurie, what is a degree for?

Yurie: A degree allows a man to become a scientist, and a woman to become a nurse.

You might imagine, however, that an AI​ ​would produce 1 million pieces of text and that a team of authors could spend its time selecting the best rather than writing them. But it is to miss a great benefit of this technology.

You cannot translate these texts into another language. You need to create an AI per language. So, you need a corpus per language ...

AI generation allows us to distinguish three levels of text production:

"The background noise": there is a need for “background dialog” in AAA productions, a considerable sum of text that is almost never read or heard: these are the conversations of passers-by in an open world for example.

Short generic dialogs: do you want an NPC that can say "yes" or "hello" in 3000 different ways? This can be done.

The writing of weak lore: so, you need a library containing a thousand stories about dragons? It's entirely possible.

To me, anything more ambitious would be technically difficult to achieve today (but as I said, I'm wrong).

But there is another issue: is an NPC saying "yes" in 3000 different ways a cool thing?

Cool as in 1999

There is a cultural barrier that seems to me to be huge compared to the technological arsenal of procedural quests or the AI ​​generating text.

Promising a unique gaming experience is something that was thrilling in the 20th century but today is just not enough. 

Today a game is consumed by playing it, but also by watching it on YouTube and by streaming.

Just as in the consumption of television shows, the gaming experience is also outside the game, through forums, social networks, and offline player interactions.

An emerging storytelling experience thus needs to be combined with fixed, reproducible markers.

Basically: "I saw this streamer in this game situation doing this and it was cool, I want to do it at home and/or I want to experience it a little differently."

This is more or less the sophisticated version of the experience of several people talking passionately about a sports match they just attended.

At the two opposites of the emergent narrative spectrum we have:

> The premise of a completely rigid game like a point'n'click is weak as 100% of the gaming experience can be consumed through YouTube / streaming.

> The premise of a theoretical game where all the experiences are unique is also weak because what the player sees in a stream is not reproducible and there are no common markers capable of generating a culture outside the game.

Unsurprisingly, it is the classic open worlds that have long offered an optimal emerging narrative experience, where the player had a great degree of freedom in a world that was the same for everyone.

As an aside, I suggest that even if you have an AI that creates 3000 ways to say “yes”, it's better to have 3000 NPCs with only one way to say yes than 1 (or 3000 NPCs) with 3000 ways to say “yes”.

Even a weakly written character will be better than a conversation generated by an AI with the technology at its current stage. A character who repeats the same sentence is sometimes boring, but that characterizes them, it's a gimmick.

More importantly, it retains the mark of the human behind the virtual piece of work, a question we are going to explore now.

The emptiness of infinity

To my knowledge, video games is the only industry that went from automation to manual when it had the means. From the procedural Daggerfall or Arena, the teams of the Elder Scrolls, when they have had the means, have switched to a handmade level design.

Why in the endless frescoes of No Man's Sky is there less of interest than in the more limited cities in the space and diversity of an open world?

The question may be a philosophical one. I would say that when playing we feel a link with the creator of the game: there is a musician, a graphic designer, a screenwriter, a level designer, who with their awkwardness and their talents, transmit something to us.

In essence, as in any work of art, we seek and find ourselves, or at least the trace of a human.

However, a purely procedural world does not show the hand of man.

Was this pebble on a distant planet laid by a human or generated by an equation?

Is there an uncanny valley of artistic design? Here is an arbitrary opinion: I think so.

And I think I can spend hours looking at a virtual pebble laid by a human, because there is a story behind it.

The pebble laid by the equation is not devoid of story either, but it is at a degree of separation that requires a personal investment that not everyone will be ready to give.

Narration resulting from a gameplay imbalance

I would just like to mention some final points before bringing together all the previous points to strengthen a proposal for a perspective in narrative design.

Emerging storytelling, in essence, comes from gameplay. Successful infiltration gameplay, for example, or squad management in X-COM, leads to great stories without imposing a prewritten scenario.

I wanted to talk about unbalanced, (seemingly) ill-conceived or idiotic gameplay that is against design standards.

These gameplays can seed new narratives.

Imagine a standard hack'n'slash where the hero discovers a weapon at the beginning of the game. They have the choice of a normal sword or a rusty dagger. Plunged into the meta-game, the player will tell themselves a story, “If I take this dagger, I might access a hidden narrative or a special reward, even though the game might be harder”. The trick here is to do nothing special, the game remains absolutely identical: at each event, each checkpoint and each decision, the player will be thinking that they may have a unique experience because of their initial decision. So you see, with a simple choice that looks like a gameplay flaw, we can give a new dimension to the story. (It may be a mad vision: this idea is constantly vetoed in my work).

I am sometimes consulted to help create emerging narratives in MMOs. Narration in this context, like war, is often born from population pressure or lack of resources.
Take a game like Eve Online and award 99% of the resources to a single player. And watch what happens. Whatever happens, it will be interesting ... much more in any case than in these balanced games with progression mechanics that are all the same. It’s quite easy to see why this idea keeps being rejected by project managers ...

All this data…

Emerging narrative from gameplay and multiplayer tensions
Increasingly strong real-time interactions between large numbers of players (as in Battle Royale)

...let us consider that the player is just one link in a global game where they indirectly participate in the writing through their own actions.

As such, we can draw a founding idea, that of a game without randomness or procedural generation, but one that is still unpredictable.

Founding principle

The idea is to substitute the work of the writer or the procedural narration with interactions between the players.

Direct example:

Let’s consider the housing in an MMO. A player buys a decorative object. Example: a painting.
We introduce a gameplay rule that on the fly generates a quest in the guild of thieves whose purpose is to steal the painting.

This painting which was a quite basic object of decoration will become the object of a war between two players.

If the painting is stolen, the player it was stolen from might buy the object again, but perhaps set traps around, wait for the thief with friends, or even put the painting at the bottom of a cave full of monsters …

Without intervening, we have here emerging narrative but also a form of iterative and crowdsourced level design.

Indirect example:

As part of an RPG, the loot of a vault is indexed to the player's level modulo a random number.

We substitute the RNG by a Karma value which is the result (a little worked) of another player's actions in another game. So a player is selfish, mean, clumsy in a game? This will result in poor loot, abominable monsters and storms in another. And the actions of that same player will impact those of another ... in a chain of positive or negative karma - each player being the deity of another.

When you are unlucky in a game, it has a very different flavor to know that you can curse someone living for this situation!

The 4 rules

Here are 4 rules for creating a game that corresponds to these ambitions, or simply to prepare a game for a future transition to a narration based on player interactions:

1) All the actions of all the players need to be recorded. In the examples that will follow, when a player picks up a flower, it is a meaningful action that is noted.

2) With each intervention of the RNG, let’s ask the question: "how can I replace the RNG by an action (that may be asynchronous) of another player? 

3) With each intervention of the procedural generation (or of the writer), let’s ask the question: "How can I replace that intervention with the past action of another player?” (Recorded in 1)

4) Finally, with each intervention of the writer, ask the question: "Would it be better to avoid saying anything at all rather than telling them something?” 

I will develop point 3 with a simple example.

I used PHP to create a very humble experimental textual MMO (I’ll speak about this later) in which we can, among other things, discover, name and collect flowers.

In this same MMO, you can ask an NPC to build a house. They can carve the door. When the game has to determine carvings on the door, it will browse the actions of the players to find something that has already occurred.

So, if you pick a flower one day, the next day you might discover an engraved door that represents you in the process of picking this flower.

Emergent Lore

The next stage that follows the emergent narrative is the emergent lore: the creation of a mythological, geographical and cultural background, resulting from the systemic exploitation of the actions of the players and without any direct intervention from a writer.

It has the major advantage of being inexhaustible once the rules are in place but above all, as it stems from rules obeying an objective or arbitrary logic (creating in the latter case an internal metaphysics), it has a coherence and brings meaning.

Moreover, as the emergent lore depends on the player, undiscovered assets or unexplored places are really terra incognita: something exists because there was once a human behind it. We are at the opposite end of the spectrum of the emptiness of the infinite without having hired a writer.

Establishing an emergent lore does not happen ex nihilo. Abundant procedural tools must be created and injected into standard game mechanics (explore, feed, build a village, assign functions to NPCs) for indirect actions that will create this emergent lore.

There is then a cascading effect under Rule 3) above. If you procedurally generate a tree type and give it a name and properties, other players will have to use it.

The goal is not to visit a tavern created from the combined work of a level designer and a writer, but to visit a house whose walls, roofs, doors are of materials that have a history and traceability, whose inhabitants are hybrid breeds of animals that have traceability and have inherited the voice and temperament of the animals of which they are composed, and whose menu is composed of close animals that have the nutritive properties of the ecosystem where they live ... etc. ... nothing new if Dwarf Fortress was multiplayer, but it is not...yet.

Let’s take a look at some small, functional examples.

Practical examples

Last year on Twitch I coded an incomplete text game to show this concept: Maspero Blue. The game is playable here: http://maspero.blue,  alas all in French (but we can translate it on the fly with Google).

I’m not very proud of it but it is an illustration of the subject.

Players explore procedurally generated places and name plants, animals, rivers ...

The encyclopedia of their works is here:

http://maspero.blue/encyclopedie.php?type=plante&id=425013720195191337&new=ok&item=plantes

In the "House" tab, we can see a "Ferme des oiseaux", that can roughly translate to:

Birds Farm

Birds Farm is an elongated orange-colored house with timid fern walls, with a simple arched entrance opening and a green pyramidal green roof. This house can accommodate 4 people.

Here lives Farfelu a Man-louse-on-stilts.

This dwelling is in Mist City.

What is interesting is that this house exists through the indirect competition of several players: a player has one day found and named the "timid fern".

They might have done it because they wanted to eat it.

Later another player picked it up and made the walls of their house.

Farfelu is a hybrid of an animal "the louse on stilts" which was discovered by a player (again, perhaps because they wanted to eat it to survive). The hybrid was generated by the game and then was found and invited into the village of a third player ... and so on.

As the game evolves into higher levels of play, procedural generation disappears in favor of player interactions.

Quantify, and therefore master, lore

I am working on a new application (called "M2") that will make it possible to collaboratively create a universe and more specifically, control the density of a created universe.

In Star Wars, when you’re looking for an individual, you can open the door of any bar and either that person is inside, or there is good friend or enemy of this individual: everyone knows each other. The universe is dense.

In Warhammer 40K, which spans over billions of worlds containing billions of people in a harsher context, it is impossible to find anyone by opening  the first door you come to. The universe is very sparse.

We can imagine a little text multiplayer game following:

> A player creates a character
> They have several choices: travel, choose an occupation, accept a quest ... (in the context of the skeletal creation of the procedural quests seen above)
> The quests cause the meeting of new characters or discovery of new places.
> If they encounter a new place or a new character they can name and describe it and determine its properties through a balanced system (as in Maspero) or use an asset already created by players.
> The X choices offered are composed of N assets already established by the other players
> By varying N (between 0 and X), we can have hyper dense universes where each element is connected to another narratively (when N = X) contrary to very low-density universes (when N tends to 0).

We can therefore control the density of a universe that is created while the players are having fun.

What other parameters can we control? What densities will provide the most amusement (is there perhaps even some connection between fun and density?)

One day I hope I’ll be able to tell you.

Refining a game

Imagine a non-procedural Nethack without RNG: it would be a pure adventure game where anyone could win 100% of the games with a fixed solution.

Today’s point'n'clicks suffer from a deficit of sales and thus of creativity. These are games that are consumed on stream and on YouTube, as opposed to the "endless games" dreamt of by major publishers.

I was recently offered the opportunity to design a new version of a detective game similar to the Lankhor Maupiti Island / Mortevielle Manor games.

I devised a game of pure puzzles with procedural assets, moving the very classic kind of detective point'n'click game towards a form of rogue-like game without statistics: "The Villa on the Sea".

You are on a Greek island for 3 days and a murder has taken place. Each person has varying motive and varying alibis, as well as a hidden agenda and some related adventures to be played out (stories of love and revenge).

The player, who during the game is discovered to have a dissociative personality disorder, is paradoxically one of the suspects.

As in a classic rogue-type game, you can’t resume an old save game: if you haven’t solved the case after three days, it is a failure.

And if you start the game again, the places and their layout, the characters, the motives ... the murderer ... everything changes.

So, you don’t have to memorize a fixed walkthrough as you would for a classic adventure game (e.g.: the butler is the illegitimate son of the victim) but you have to discover and master ways to succeed just as you would for a rogue-type procedural game(e.g.: determine the time the crime took place and check where everyone was at that time).

We create a classic puzzle game that will allow a streamer to say, “I’m good at this game, I'm ready to beat it this time!”. As they might say today of Dark Souls, but less so of a puzzle adventure game.

By introducing a procedural approach to a classic puzzle game, we give it a new, possibly commercially viable, dimension.

We have a lot to discover and reinvent in this fascinating area.

I hope that reading this article helps spark great ideas for beautiful games that will amaze us all.

The lecture (in french) here :