ScripTale: generation of procedural narrative

  ScripTale  : Generation of Procedural Narrative  Bachelor’s thesis Design and Development of Video Games    Author: Ripoll Galan, Laura Director: Torelló Oliver, Josep CITM (UPC) - Terrassa Course: 2017 - 2018 Scheme: 2014   Generation of Procedural Narrative Ripoll Galan, Laura Abstract The subject of research is the procedural narrative. Procedural narrative is the way of generating narrative automatically along with the progress of the game. More specifically, the subject of the study will be about exploring the way procedural narrative could be implemented in a project and through with a tool external to the code of the game itself. Key words and notation AI: Artificial Intelligence PN: Procedural Narrative PCG: Procedural Content Generation LL: Linear Logic GOAP: Goal Oriented Action Planning States: Resultant variables of each character after an action Conditions: Rules to be ascertained to develop an action Actions: Performance of an available option of the character Messages: Content of the dialogue said by a character towards another      1  Generation of Procedural Narrative Ripoll Galan, Laura  Index  Abstract 1 Key words and notation 1 Index 2 Index of figures and images 5 Index of tables 6 1. Introduction 7 1.1. Project justification 7 1.1.1. Motivation 7 1.1.2. The problem 7 1.2. Aim of the project 8 1.2.1. General goals of the project 8 1.2.2. Specific goals of the project 8 1.3. Project scope 9 2. State of the art 10 2.1. NPCs. Learning the player’s way of play 11 2.2. Voice recognition and realistic graphics 12 2.3. Procedural Content Generation 14 2.4. ANGELINA 15 2.5. Procedural Narrative 15 2.6. The tool: ScripTale 17 3. Planning 19 3.1. Gantt chart 19 3.1.1. Tasks description 19 2  Generation of Procedural Narrative Ripoll Galan, Laura 3.1.2. Chart 22 3.2. SWOT 24 3.2.1. Analysis 24 3.2.2. Contingency plan 26 3.3. Costs analysis 27 3.3.1 Costs overview 27 3.3.2. Costs breakdown 30 3.4. Management tools 33 4. Methodology 33 4.1. Tracking of the project 34 4.2. Validation tools 34 5. Development 36 5.1. Design of the tool 36 5.2. ScripTale Initial implementation 40 5.2.1. Curve Editor 40 5.2.2. States Editor 42 5.2.3. Relationships Editor 45 5.2.4. Dialogue System 46 5.2.5. Characters 47 5.2.6. Messages and actions 48 5.2.7. Emotions 49 5.3. Initial workflow of ScripTale 52 5.4. Last state of ScripTale 53 5.4.1. Curve editor 53 5.4.2. States Editor 54 5.4.3. Relationships Editor 55 5.4.4. Dialogue System 56 5.4.4. Characters 56 3  Generation of Procedural Narrative Ripoll Galan, Laura 5.4.5. Messages and Actions 60 5.4.6. Emotions 60 5.5. Final workflow of ScripTale 61 6. Conclusions 62 Bibliography 65 Annexes 68            4  Generation of Procedural Narrative Ripoll Galan, Laura Index of figures and images  Monte Carlo Tree Search Algorithm schema Fig. 2.1. Characters’ decision taking schema in The hare and the turtle Fig. 2.5. Gantt diagram Fig. 3.1.2. Design schema of the base classes of ScripTale Fig. 5.1.1. Design shema of the editors of ScripTale Fig. 5.1.2. Curve editor of ScripTale Fig. 5.2.1.a. Curve functions of ScripTale  necessary for the programmers Fig. 5.2.1.b. Schema of the States functionality Fig. 5.2.2. Schema of the Relationships functionality Fig. 5.2.3. Workflow of ScripTale Fig. 5.3. Relationships Editor of ScripTale Fig. 5.4.3. Current workflow of ScripTale Fig. 5.5. Simple example of use of TeLLer  provided by its creators Fig. 6.  5  Generation of Procedural Narrative Ripoll Galan, Laura Index of tables  Work Packages Table 3.3.1. a Partners  Table 3.3.1. b Budget Table 3.3.1. c  Human resources  Table 3.3.2 a  Software  Table 3.3.2 b  Hardware Table 3.3.2 c  Indirect costs Table 3.3.2 d         6  Generation of Procedural Narrative Ripoll Galan, Laura 1. Introduction 1.1. Project justification 1.1.1. Motivation The strongest justification of this project is my intrinsic motivation to link two of my               favourite fields in video games: Artificial Intelligence and narrative. Not only because            of my own interest in these two areas but also due to the high rate of difficulty                 involved and the amazing result that could be achieved by defeating the challenge             to unify both opposed behaviours, as narrative tends to be structured and AI             emergent. 1.1.2. The problem Nowadays, procedural methodologies in video games are spreading, from         animations to terrain creation and so, narrative. However, there is not any global             tool in Unity’s Asset Store to allow procedural narrative generation in games            developed with this popular engine. More investigation in this field is needed and a high demand of procedural narrative              exists in the industry. Some AAA games try to make such an approach but, these               implementations are not public and other developers could find it difficult to include             such a valuable feature on commercial engines like Unity. Some theoretical research has been carried out, however, there is no practical tool             to apply procedural narrative to real projects.   7  Generation of Procedural Narrative Ripoll Galan, Laura 1.2. Aim of the project The goal of the project is to program a tool in Unity that could be implemented in                 any game to generate procedural narrative through a dialogue system with the            existing characters. The tool will be submitted in a Unity package format to make it easier to put it in the                   Asset Store afterwards. If the tool has proved an interest in the community, a Demo               will be delivered as a Unity project so the code will be accessible to anyone and                also in a compiled game version. However, it is not an aim of this project but a                 further possible work beginning with the tool. To develop the entire project, be it the package of the tool, the engine to be used                 will be Unity. On the other hand, for scripts edition and code, Visual Studio will be                the IDE to work with. 1.2.1. General goals of the project -Generate a narrative with different results at each game: Make it possible            to generate different narrative with a same base system dialogue but with            unexpected results at each game. -Approach the tool to independent developers: Make a available tool to           improve narrative systems on independent games in the market. -Get the tool published in Asset Store: Accomplish all the requirements to            publish the result of this research, the tool, to the Asset Store in unity. 1.2.2. Specific goals of the project -Control an AI with unpredictable results: Being capable of controlling most           of the internal paths or processes of the AI without losing the surprise of              getting an emergent narrative situation. 8  Generation of Procedural Narrative Ripoll Galan, Laura 2.4. ANGELINA It can be relatively easily conceived that some areas such as art or sound can be                modulated and created alongside procedural generation but, Design seems         something untouchable that can only be done by human. Thinking up an entire             game from scratch is a task that is usually carried out by a designer. Withal,               ANGELINA is the practical validation of the potential of PCG, being able to             substitute this last task. Michael Cook is the 30-year-old senior research fellow at University of Falmouth            behind ANGELINA. The tool’s name stands for “A Novel Game-Evolving Labrat I’ve            Named Angelina.” He developed ANGELINA in 2011 and since then, Michael has            kept updating the tool. Meanwhile, ANGELINA has developed by itself hundreds of            experimental video games. The senior researcher is now working on Creative Code            Generation for Interactive Media. PCG is normally employed in video game development as a tool to aid in a specific                feature within a large creative vision that comes from a person. What Michael tries              to achieve with ANGELINA is to have a system that procedurally generates the             majority or even all the aspects of a game’s design by itself. 2.5. Procedural Narrative Procedural Narrative is the automatic process that generates narrative over time.           However, narrative is a structured story by definition so, balancing the emergency of             PCG and the structure of the narrative is the main conflict in procedural narrative.              Other procedural contents, have no need of a rule or structure to exist but the ones                imposed by the designer, whereas narrative does. A story needs to have a             beginning and an ending, with its relevant stages and conflict points. To build an emergent story, characters and AI will take more control of the actions               and the main weight of the narrative will be the interaction between characters and              15  Generation of Procedural Narrative Ripoll Galan, Laura their decision taking aspect. In this way, by letting the player some freedom, the              machine and the player will take turns to redrive the story to some desired states               established by the Designer (or not). To illustrate what has been brought up in Practices for Procedural Narrative            Generation  (Martens Chris, 2017), coming up with an schema could make it more            clear. So, a good example of story with a main plot determined by the characters’              decisions is the classical tale: The Hare and the turtle  . As can be seen in the                following schema, all the decisions brings the story to an end. Characters can plot              their own story:       For the implementation of AI there is not a correct set algorithm to follow, each               solution finds its own way: Deep Learning, Machine Learning, Finite State Machines,            Behaviour Trees, States, MCSTA… Theoretical studies from various universities,         such as the Development of an Emergent Narrative Generation Architecture for           Videogames (Schudlo, Nicholas A., 2014) point out one shared method for           procedural narrative: Linear Logic, a type of substructural logic. Substructural logic is the study of correct methods of reasoning characterized by            following three rules: 16  Generation of Procedural Narrative Ripoll Galan, Laura -The contraction rule: It states that any premise used for a valid            deduction, can remains valid only once. -The weakening rule: While a hypothesis of a valid deduction (not a            premise) remains valid, more premises can be added to it. -The exchange rule: Also called the permutation rule, it defines that           the order of premises is irrelevant to the validation of a deduction. Linear Logic is one exception in substructural logic due to its omission of the              contraction rule and the weakening rule.  Linear Logic emphasizes on the role of formulas as resources. To avoid focusing on              the truth or the proof from classic logic, it does not allow the application of the                contraction rule or the weakening rule to all the formulas but only to those marked               with certain modals. This way it has an involutive negation while keeping strong the              interpretation. Provided this focus on resources, Linear Logic is applied in           programming such as, in this case in procedural narrative which, as already seen is              based on rules and algorithms to control it.  2.6. The tool: ScripTale ScripTale  , name of the tool which is object of the current study, will begin from the                point of the theoretical investigations done in Procedural Narrative. That means it            will use the information asserted in these studies to progress. It will go further implementing in Unity what these studies have shown in papers.              Despite some AAA games such as The Witcher 3: Wild Hunt  (CD Projekt RED, 2015)               could have presented similar features or innovative narrative components, none of           them makes the code available and/or scalable, being closed and specific for the             game. The distinctive feature of ScripTale is the freedom, not only for offering the              17  Generation of Procedural Narrative Ripoll Galan, Laura tool to a wider range of studios but also for being able to adapt to any game made                  with Unity. This AI tool’s limitations yet would be positively reviewed for its specific purpose.             ScripTale depends on human input to work, the designer or the screenwriter. The             responsible of the narrative task would have to provide the tool with a minimum              amount of dialogue classified on the type of character that would say that sentence              and in which case. To let the tool work with the provided dialogue system, the               designer has to set the path of the story, the picks of intensity or the curve.                ScripTale will then, adapt the story and decision making to follow the established             rhythm. All this means that this tool can work with a dialogue to generate narrative               whilst the others in the market cannot.          18  Generation of Procedural Narrative Ripoll Galan, Laura 3. Planning 3.1. Gantt chart 3.1.1. Tasks description The description of the tasks chronologically ordered can be found below. All of             them are tagged showing the area they belong to.  Rubric 1 preparation: Prepare the contents of Rubric 1 and write them down on the document. Research of assets: Looking for assets of the project. Research of AI methods: Deep knowledge about the different methods that could be applied to the project. Schema of the tool: Design the functionality of the tool, scratch an UML and make a general schema of               how it will look like and how it will work. Initial tool: Programming the base of the tool by creating a minimum AI for the characters and               the basic interface for the tool.    19  Generation of Procedural Narrative Ripoll Galan, Laura Dialogue system: Create a basic dialogue system. Simple sentences but working with the logic            needed in the project (example of the logic: combine sentences to create new ones              from the features of the characters, for instance, adding tags). Programming of the dialogue system: Improve the base tool by adding the methods chosen. (Possibly, programming of            the methods without including dialogue system yet. That is contemplated in the            following step of fixing the tool). Test and fix the tool: Detect errors, make the code cleaner, and polish the dialogue and narrative. Rubric 2 preparation: Prepare the contents of Rubric 1 and write them down on the document. Improve dialogue: Add more sentences and expand the existing dialogue. Improve the GUI of the tool: Improve the tool’s interface, make the interface usable and the graphics appealing. Improve AI system: Improvement of the AI methods, fix the program’s errors. Project writing: End up with writing the document itself. Further stages of the project will be added               during the project revision part, at the end of it. Separate the tool from the game: Separate the game (in .dll format) from the tool (a unity package). The game should               have the tool applied in it. 20  Generation of Procedural Narrative Ripoll Galan, Laura Decorate the tool: Decide the name of the tool, create an icon, add extra features (such as sending the                script of the final story to the users). Narrative improvements: Last final minor improvements on the dialogue system or narrative. AI improvements: Last final minor improvements on the AI system of the tool. Overall project revision: No more programming or design to add new features will be done during these last               days, only bug fixing. If a task is not finished on time, then, it will be decided where                  and how to cut. Include everything new on the project document and ensure the              quality of the final tool. 21  Generation of Procedural Narrative Ripoll Galan, Laura 3.1.2. Chart     22  Generation of Procedural Narrative Ripoll Galan, Laura    23  Generation of Procedural Narrative Ripoll Galan, Laura 3.2. SWOT 3.2.1. Analysis Strengths: -Attractive offer: The offer itself is appealing. Not only does it provide a             dialogue system to reduce time and increase efficiency but also a narrative            complexity higher than the one that could be achieved without the tool. -Innovation: Innovation attracts the target to try something unknown if it has            been proved that to work. -Accessible: The target knows the place to get it and normally enters the             store. If the tool is attractive enough, they can easily manage to download it              and include it into a project. Target is used to it and that makes the tool                accessible. -User friendly: Easy to understand. There is no need of previous knowledge           to use it. The tool is intuitive and most of the functionality can be applied               through the UI, not through code. Weaknesses: -Unrecognised brand: There is no market behind waiting for the tool because          it is a new unknown brand. That makes it complicated to get popularity. -Innovation: Innovation can be both, a strength and a weakness at the same             time. Being considered innovative in a field where a lot of research has been              done, and is still being, is dangerous because it could mean that the idea              leads to failure. -Oriented to very specific games: Gameplay on games that will use the tool             may offer on an open-world scenario to allow the players to interact. Cannot             be applied in most of the mobile games developed by indies.  24  Generation of Procedural Narrative Ripoll Galan, Laura Software Software Price per unit Units Total cost Unity Plus license  360 €/unit 1 unit 360 € Microsoft Office  149 €/unit 1 unit 149 € Unity assets (Characters & Environments)  12 €/unit 2 units 24 € TOTAL  - - 533 €  -Assumptions:  The developers acquire a license for 1 year and 1 account.           Unity assets bought shall be low price assets but not some of the free ones.               Microsoft Office offers a license for 3 years, so it may be used in further               projects of the studio. Hardware  Hardware Price per unit Unit Lifetime Total cost Amortisation PCs  900 €/h 1 unit 5 years 900 € 60 € TOTAL  - - - 900 € 60 €  31  Generation of Procedural Narrative Ripoll Galan, Laura Indirect costs Type Price per month Amount of months Total cost Office rental  1200 €/unit 4 4800 € Light, water & other supplies  150 €/unit 4 600 € Self-employed quota  264.44/month and person 4 2115.52 € TOTAL  - - 7515.52 €  -Assumptions:  As the proper size of a small office is less than 100 m2, it is               assumed the property will sit on a 90 m2 surface, located in Barcelona.             Self-employed quota must be paid twice, having the developer and the           screenwriter as the main share-holders.     32  Generation of Procedural Narrative Ripoll Galan, Laura 3.4. Management tools Hacknplan and Trello will both be used to keep the track of the development.              Hacknplan tables may change for each milestone or in a short settled period. Trello,              meanwhile, is thought to be used to store a more general picture of the project, in                other words, all the project tasks will be specified in Trello as generic capsules that               could embrace multiple specific assignments. 4. Methodology In the videogames industry, it is common to use a pre-production, production and             post-production model-based methodology. However, this study does not consist         on a videogame itself but on a tool, therefore, this methodology will not work              efficiently on a project of this kind. An agile interactive method would be the most suitable to develop the project. The              different stages could change dynamically as it consists of a research that has to be               iterated, tested, and change continuously. Scrum is one of these methodologies           that would perfectly fit the project’s typology, so that is the one to be used. Scrum, being based on sprints, can be easily managed with such a management             tool like Hacknplan. Weekly stages: -Beginning of the week: A backlog update will be done and the tasks to be               developed over a week will be noted down. -During the week: Daily sprint review of the tasks done, balancing remaining           time/tasks and estimations established to get to know one own’s error rate. -End of the week: Try to close all the tasks. If that is not possible, draw to a                  close by deciding which features are to be cut given that the time is limited.  33  Generation of Procedural Narrative Ripoll Galan, Laura 4.1. Tracking of the project Hacknplan, as used for the management, will be accessible at any time of the              project and it will also show its progress, current, and past. As happens with Hacknplan, GitHub can also provide a good tool for monitoring the              project. Through all the commits, if described properly, it is possible to see the              advance. However GitBash will be used to commit all the project’s changes instead of             GitHub, tool which also allows this action, through GitHub will be used to check the               progress. Github provides an intuitive User Interface for this monitoring feature. 4.2. Validation tools To check the authenticity of the theoric information collected through internet,           consulting to a professional in this field will be done. E-mailing the Doctor in              Artificial Intelligence Chris Martens could be considered by itself a validation tool for             the investigation part. Before testing, because of using Unity as an engine, a suitable debug needs to be               done. Linking Unity to the IDE Visual Studio makes it possible not only to             implement the code but also to check it at the same instant, to debug it. Visual studio will also be used to test, but not as an exclusive tool, because testing                needs to be done by using the program and as such Unity Editor will meet this               needs. After having fixed the bugs found through this tool, another testing round will be              done by building the application. A testing scene will be creating during the testing              session to evaluate its effectivity in runtime, not only within the editors, but also the               AI scripts executing in game. 34  Generation of Procedural Narrative Ripoll Galan, Laura Validation is also required for the narrative aspect of the project so, in this case,               online tools will be used, for instance: correctors,synonym dictionaries and        dictionaries.                35  Generation of Procedural Narrative Ripoll Galan, Laura 5. Development The development process starts by conceiving the structure of the tool, deciding            which parameters to obtain by the input of a designer or which ones should be left                up to the tool thus the planning of further tasks gets easier. However, this led to structural problems on how many variables should be handled             by the designer and once decided and how an usable UI could be displayed leading               to the ease in managing and editing a vast number of parameters. For these reason, parameters to be handled have been divided into three different             fields: the relationships’ editor, the curve editor, the states’ editor. These editors,            however, need other structures to get to work, such as the characters with its              emotions and behaviours or the messages classified from the dialogue. Before the implementation, all of these had to be clearly designed to avoid             inconsistencies or negligible work. 5.1. Design of the tool After seeing other concepts of Procedural Narrative that can be found in the State              of Art section, letting the characters plot the story by their actions was the way to                start with the tool. However, with the new idea of including a dialogue to fit the                narrative, the Linear Logic method used in these other projects was not effective             enough and this other methods to be combined with were required in the             ScripTale  ’s implementation. To behave in a way or another, the characters will be driven by emotions and each                consequent state after an action should affect some characters in some way. So,             the emotional component is added now to the linear logic relatives states. At this point, the necessity of including a grade of importance was needed on how               the characters would be affected. Consequently, the tool should include weights. All            the interactions with the characters would affect them in some way so, all the inputs               36  Generation of Procedural Narrative Ripoll Galan, Laura are designed to control emotions, and emotions will drive the story. States (results             of the actions performed) and messages (what players say) have emotional           implication. This emotional implication is translated into a natural behaviour of the characters in             taking actions. To sum up, in this tool, actions are still plotting the story but, being                affected by the emotions resultant of the original characters’ behaviour and other            characters’ actions, so as to get a fitting dialogue in all this procedural generation of               narrative. The artificial intelligence of the tool works with linear logic and weights. - AI concepts: Working with weights means implementing a Goal Oriented Action Planning          (GOAP). It is a system that allows the agents (characters) to plan a sequence of               actions to satisfy an aim. This sequence of actions depends not only to the goal but                also on the current state of the agent. Each possible action has multiple weights of necessities it will satisfy. However, the             actions available can present a condition. That is the point where the current project              combines Linear Logic (LL) with GOAP. LL will ensure that an action cannot be done until another one has ended, an item                has been consumed or, on the other hand, some actions can happen            simultaneously. Linear Logic states the conditions and the end point of each action. ScripTale  implements LL to decide the available spectrum of actions from the           current states, the possible ones and the demand of the intensity curve. GOAP is              used after to get the most likely action or message to occur given the character               current state and the available actions filtered by the Linear Logic step.    37  Generation of Procedural Narrative Ripoll Galan, Laura - Initial design schemas to start implementing: All the elements in the schemas were added and modified by the schemas’ own              nature if they had two follow the two rules of: emotional implication for the decision               making and actions to plot the story. Base Classes     38  Generation of Procedural Narrative Ripoll Galan, Laura Editors       39  Generation of Procedural Narrative Ripoll Galan, Laura 5.2. ScripTale  Initial implementation 5.2.1. Curve Editor The curve editor provided to the designer is meant to be driving the story through               an intensity path desired. It would allow the designer to control the time the climax               will happen or the more relaxing parts of the story.     - Problems: The curve editor actually did not suppose any striking problem. The only trouble             that was generated had been when ScripTale was applied to a game consisting of              narrative parts (interactive parts where developing the plot of the story) and            gameplay parts, the curvature was still advancing in time. The problem was easily             solved by adding an option to pause the curve by calling a function, if the narrative                needs to be stopped, and resume it when required. However, despite the tool is thought to be used by designers, this lead to the               simplest as well as major inconvenient of ScripTale  . It is the programmer, not the              designer who has to take the already mentioned problem into account. The solution             40  Generation of Procedural Narrative Ripoll Galan, Laura To break the limitations of this kind of software, the Dialogue System was changed              to work, as it currently does, through Google Drive. The Asset Store of Unity and even other platforms on the Internet, provide Plugins              to integrate the use of spreadsheet Excel in a project. However, ScripTale works             with document files, not Excel, and, as it was complicated to find a Plugin with this                functionality, this one was implemented in the code of the tool itself through the              Google Drive API without any other Plugin. The research of Plugins and the final              decision of coding the desired functionality were the two main problems that            appeared after solving the precedent conflict relative to software licenses. 5.2.5. Characters Characters are entities with an id different one to each other to be quickly identified.               They will also own a stack of events, each event consisting on a structure with a                pointer to another character and an array of emotions. The stack pretends to simulate a memory, it keeps track of what has happened to               the character and what could break or modify a relationship. Finally and the most important, the character is constituted by two sets of emotions.              The original ones that define it and determine its way of acting and behaviour. If a                character tends to be aggressive by nature (high level of rage emotion by default),              for instance, a message transmitting rage will increase the rage current state of the              character. - Problems: Characters must have a memory to keep track of what has happened. However, the              knowledge of the events is not taken in consideration. These caused two problems: - Messages did not have any effect on them at any moment if they are not               applied to the character. 47  Generation of Procedural Narrative Ripoll Galan, Laura - The character is affected even if it doesn’t know the source of its emotions              because has not received directly the message or the action. A temporary solution is not to apply an effect of a message that has not been                 received directly by the character and apply the effect if it gets to know it. It is                 complicated, though, to track if it discovers a message that is already in its memory               stack. 5.2.6. Messages and actions Messages and actions are actually the same thing. Both of them consist on             structures containing the id of the character emitter, the one of the receiver and a               dictionary with the ids of the characters affected as keys and the array of emotions               that the message or the actions produced. The difference between messages and actions is the output. Messages are meant            to pick a type of dialogue depending on the emotions of the emitter character and               to create an effect on the receiver. This is, then, the output of the message.               However, an action derives to an state. It is also chosen by the emotional state of                the character, however, in this case, it takes into account the conditions of the              story. To quickly pick up a message from all the loaded ones, these are the helper               functions called by ScripTale  when building the dialogue.    public static List<Message> GetMessagesByEmotion(EmotionType emotion) { return messages.FindAll(m => m.effect.emmotion == emotion); } 48  Generation of Procedural Narrative Ripoll Galan, Laura public static List<Message> GetMessagesByEmotionAndLevel(EmotionType emotion, float level, float threshold) { return messages.FindAll(m => m.effect.emmotion == emotion && (m.effect.level <= (level + threshold) && m.effect.level >= (level - threshold))); }   - Problems: Although actions are messages are conceptually the same type of structures,           messages are not supposed to be able to get to a state and so, it could be difficult                  to reach to a state. Some possible solutions were or choosing an action right after a                dialogue happens, which would still feel antinatural to cut a conversation and start             an action immediately; another one could be to start an action depending on the              resultant states after a conversation. This last one is the approach taken in this tool,               despite there are probably better implementations to solve it. 5.2.7. Emotions As it has been stated multiple times in the current document, characters, messages,             actions and states are based on emotions. In a more technical aspect emotions             consist on structures which simply contain the enum type of emotion and a float              value corresponding to the level.      49  Generation of Procedural Narrative Ripoll Galan, Laura  The basic emotions to define a character are:   public enum EmotionType { Grief, Loathing, Rage, Vigilance, Ecstasy, Admiration, Terror, Amazing, None }  All of them have been chosen to achieve a human-like realistic behaviour and             different archetypes of characters by mixing various levels of emotions. On the            basis of that postulate, the feelings that have been applied in the current project              actually follow the ones that are thought to constitute the human being. Plutchik R.              (1927 - 2006) had developed a wheel of emotions to describe the people with              feelings. This has been the inspiration for the EmotionType choice of options. - Problems: Before realizing some research on psychological papers was needed to develop a            variate and spectrum of emotions to define a character, these basic feelings where             needed. However it was complicated to pick up the right choice of emotions that              50  Generation of Procedural Narrative Ripoll Galan, Laura could be combined to get different behaviours without any psychological          knowledge. The real matter about the inclusion of emotions in the project was the unexpectancy              of the task itself. It was not taken into consideration at the time of the planning.                However, due to the nature of research of this project, these emergent situations             were thought to occur so, some generic tasks include in themselves investigation            and implementation. The Gannt chart can, then, remain the same, as “Emotions”            can be considered a task inside the “Dialogue system  ” one.                51  Generation of Procedural Narrative Ripoll Galan, Laura 5.3. Initial workflow of ScripTale The big picture of ScripTale  would look like this schema.                  52  Generation of Procedural Narrative Ripoll Galan, Laura 5.4. Last state of ScripTale 5.4.1. Curve editor The tool has definitely removed the Curve editor because of design incongruities.            First of all, usage of the tool gets more complicated as more elements are being               added. Moreover, sTaleCharacters have currently been totally programmed and         have a huge impact on design, so more items to control were being developed.              What is more, story had very little influence in the story and its perks were fewer                than the problems it caused. It was only taken into account to get the next possible                state, depending on the intensity on time. The corresponding function in which the curve was involved was the following.            Were the tool just got the intensity directrices from it and combined it with the               intensity requirements of each of the states. Then it took the most suitable state for               the established curve.  private static void GetNextState() { float intensity = sTaleCurve.GetCurveState(); float[] level = new float[current_state.children.Count]; for (int i = 0; i < current_state.children.Count; ++i) { float average = 0; int count = 0; for (int e = 0; e < current_state.children[i].content.returnEffect.Count; ++e) { for (int c = 0; c < current_state.children[i].content.returnEffect[e].conditions.Count; ++c) { average += current_state.children[i].content.returnEffect[e].conditions[c].level; count++; } } 53  Generation of Procedural Narrative Ripoll Galan, Laura average = average / count; level[i] = Mathf.Clamp(current_state.children[i].content.conditionsSet * average, 0.0f, 100.0f); } int following_index = 0; for(int lev = 0; lev < level.Length; ++lev) { if(Mathf.Abs(intensity - level[lev]) < Mathf.Abs(intensity - level[following_index])) { following_index = lev; } } following_state = current_state.children[following_index]; }  The states Editor is a way itself to control the intensity depending on the conditions               set by the designer and no further editors should be needed to accomplish this aim. 5.4.2. States Editor The States Editor functionality remained almost the same as what was initially            designed and implemented. As the decision of removing the intensity curve was            removed, the next state to follow in the tree had to be decided in a self-contained                way. Without the input of the designer to plot the path, the tool had to do it by itself,                   so GOAP takes more relevance in this process. Various goals were planned for each state. From the current state, the story has              different aims. To decide which one to take, the tool checks the conditions in each               of these states and calculates the distance between the current emotional level of             the character and the level set in the condition. The one to be choose is the most                 likely to happen. In technical terms, the cheapest option, the one that would take              less iterations. 54  Generation of Procedural Narrative Ripoll Galan, Laura 5.4.3. Relationships Editor This editor remains the same as was planned. When the editor is first opened, all               the characters appearing in the story have to be already put in the scene. All of                them will appear as a node and when they are connected to each other, their               relationship has to be set. The relationships can be saved in an own extension file               and this file or other files derived from the editor can be assigned to the ScripTale                prefab to be used over the course of the game.         55  Generation of Procedural Narrative Ripoll Galan, Laura 5.4.4. Dialogue System Few changes have been applied to the Dialogue System. The main workflow            through Google Drive has been kept: The document must be uploaded in Google             Drive and it will be downloaded directly throughout Unity, without any need to             download it in the computer and passing it to Unity afterwards. The principal and only improvement is that it is the user who chooses when to               download the dialogue file by checking the toggle provided in the Dialogue Importer             prefab of ScripTale  . The document is not refreshed automatically but only when the             designer wants to and selects the option to download it again. All the instructions of use are specified in the Readme.md  . 5.4.4. Characters Characters’ narrative behaviour is all found in the script sTaleCharacter, that has to             be attached to the Character prefab in the scene. It is one of the most complex                script in ScripTale because it consists in an agent that interacts with the editors to               act or talk depending on the current state and also on the messages and actions               that the character itself owns.   These are all the variables that the character needs or controls:  [SerializeField] public int identifier; [SerializeField] public Effect[] baseBehaviour; [SerializeField] public Message[] characterMessages; public Effect[] currentState; List<Effect> memory; 56  Generation of Procedural Narrative Ripoll Galan, Laura            TeLLer had stopped at this point because their solution was about generating a             syntax for exploring procedurality in narrative with a cleaner and shorter manner            than with the already existing behavior trees. The current project could have ended             up with this anti-innovative solution to solve the same trouble raised by another             study in a more complex way. However, ScripTale  , wanted to justify the use of trees               in the States Editor with a meaningful output in the development of the story. Narrative would be linked with dialogue in my implementation. How to make some             structured field such as the dialogue up to an artificial intelligence was the obstacle              that has been the hardest part of the current study until the last implementation of               63  Generation of Procedural Narrative Ripoll Galan, Laura ScripTale  , and still is despite having found a solution. Classifying all the messages             into different types of emotions could allow the tool to decide which message to              pick up without any knowledge of the content but the emotion the message could              cause to a character. To endow this feature with realism, not all the characters              would be affected equally by a message, the effect had to be modified depending              on the relationship between characters. Realism is the key that is still found a limit for the tool. To achieve emotional realistic                 narrative behavior, the logic of the conversation was not contemplated. So           characters must still have default sentences as variate as the designer wants to set              them. However, they must be generic as conversations may not happen to make             sense. This could be a further point of research for the current project. It was not                implemented because a supposed to be simple tool would have become so            complicated that AI would have less control over the decision taking process of the              messages sent by the characters. To conclude with this project, an observation has to be done. In an uncertain field               such as Artificial Intelligence, the accomplishment of goals becomes dubious. They           were technically achieved, however, better implementations can be done and          studies still have a long way to improve in Procedural Narrative to get better results               and really impressive conversational behavior totally up to the characters or agents            of the AI.        64  Generation of Procedural Narrative Ripoll Galan, Laura  Bibliography - State of the art AI meaning gameplay benefits  [Online]. (2017, 15th September). Article, URL <https://www.forbes.com/sites/quora/2017/09/15/how-will-artificial-intelligence-cha nge-video-games/#6107d8dd2bb5> [Consult 14th February 2018]. NPCs algorithms  [Online]. (2017, 28th August). Blog, URL<http://sitn.hms.harvard.edu/flash/2017/ai-video-games-toward-intelligent-gam e/> [Consult 14th February 2018]. AI’s effect on video games  [Online]. (2018, 10th January). Blog, URL<http://www.techradar.com/news/game-on-how-ai-is-transforming-video-gam es-forever> [Consult 14th February 2018]. Future of AI in the video games field  [Online]. (2017, 27th July). Blog, URL<https://www.theatlantic.com/technology/archive/2017/07/musk-vs-zuck/5350 77> [Consult 15th February 2018]. Visual improvements  [Online]. (2018, 10th January). Blog, URL<https://www.techradar.com/news/tim-sweeney-refreshing-consoles-every-sev eral-years-keeps-everyone-happy/2#article-body> [Consult 17th February 2018]. Voice recognition video game  [Online]. (2017, 2nd February). Article, URL<https://www.bbc.co.uk/taster/pilots/inspection-chamber> [Consult 17th February 2018]. Destiny 2 Alexa  [Online]. (2017, 30th November). Article, URL<https://www.polygon.com/2017/11/30/16719532/destiny-2-amazon-alexa-voi ce-commands-ghost-shell> [Consult 17th February 2018]. -Procedural content generation  : 65  Generation of Procedural Narrative Ripoll Galan, Laura Green Dale. (2016). Procedural Content Generation for C++ Game Development  . UK: Packt Publishing. Community Experience Distilled. Watkins Ryan. (2016). Procedural Content Generation for Unity Game Development  . UK: Packt Publishing. Community Experience Distilled. Areas in which PCG can be applied  [Online]. (2017, 14th July). Video, URL <https://www.youtube.com/watch?time_continue=2&v=0vST4IvG5CE> [Consult 8th February 2018]. PCG implementation  [Online]. (GDC Europe 2014). Conference content, URL <https://www.gdcvault.com/play/1020868/Games-with-Freedom-Programming-Pro cedural> [Consult 8th February 2018]. Angelina  [Online]. (2017, 29th November). Blog, URL <https://www.technologyreview.com/s/609482/ai-is-dreaming-up-new-kinds-of-vid eo-games/> [Consult 9th February 2018]. Games made by Angelina  [Online]. (2018, 15th January). Web page, URL <http://www.gamesbyangelina.org/> [Consult 9th February 2018]. - Procedural narrative information: Narrative Structure  [Online]. (2017, 6th September). Blog, URL <http://www.gamasutra.com/blogs/AndrzejMarczewski/20170609/299705/Narrative _Atoms_and_the_Soap_Heros_Journey.php> [Consult 8th February 2018]. Practices for Procedural Narrative Generation  [Online]. (2017, 12th May). Video, URL <https://www.youtube.com/watch?v=k2rgzZ2WXKo> [Consult 7th February 2018]. Practices for Procedural Narrative Generation  [Online]. (2016, 10th January). Video, URL <https://www.youtube.com/watch?v=p40p0AVUH70> [Consult 7th February 2018].   66  Generation of Procedural Narrative Ripoll Galan, Laura - Development Schudlo, Nicholas A.. (2014). Development of an Emergent Narrative Generation Architecture for Videogames. Bosser A., Ferreira J. F., and Cavazza M., and Martens C.. Linear Logic Programming for Narrative Generation  [Online]. Research, URL <https://www.cs.cmu.edu/~cmartens/lpnmr13.pdf  > [Consult 21st March 2018]. Goal Oriented Action Planning for a Smarter AI  [Online]. (2014, 23rd April). Tutorial, URL <https://gamedevelopment.tutsplus.com/tutorials/goal-oriented-action-planning-for -a-smarter-ai--cms-20793> [Consult 20tht April 2018].               67  Generation of Procedural Narrative Ripoll Galan, Laura Annexes - Annex 1: ScripTale.unitypackage Annex 1, named ScripTale.unitypackage can be found in the same folder as the             current memory. It consists on the tool itself, the resultant package to be imported              to unity so as to make use of the tool. - Annex 2: Readme.md Annex 2, named Readme.md can be found in the same folder as the current              memory. It consists on the README of the tool, the instructions of use.   68