STORYLINE AND DRAMA Mike Chu Joey Blekicki Stephen

STORYLINE AND DRAMA Mike Chu, Joey Blekicki, Stephen Kish

STORYLINE AND DRAMA History of Narrative Development Dialogue Management Gameplay and Story New Research in Storyline Development

HISTORY OF NARRATIVE DEVELOPMENT

PROPP AND THE FORMALIST Morphology of the folktale � Discovered stable structures in Russian folktales � Transgression, deception, struggle, punishment, wedding, etc. � A sequence of 31 narrative functions

PROPP’S APPROACH Narrative functions are invariant elements and are independent from the characters that execute them There a limited number of narrative functions Functions always occur in the same order � No “backtracking” allowed

LIMITS TO PROBB’S MODEL No “branching” to allow for alternate stories It is not interactive

GREIMAS Introduced the role-based analysis of narratives � Define characters “not for who they are, but for what they do” � Defined roles: subject vs. object � Also identifies semantic fields: love, greed, etc

BARTHES AND THE INTERPRETATIVE CODES Defined scenes in terms of action ramifications Actions have the dimension of semantic field Not restrained to a specific occurrence

BATHES CONT. Some actions do not require a specific sequence � The result of a murder and the reason for the murder Theory based on five codes: � ACT (ACTion) � REF (REFerence) � SYM (SYMbolic) � SEM (SEMantic) � HER (HERmaneutic)

BREMOND His theory centered around the concept of character roles � Opposition between Agent and Patient

BREMOND CONT. Characters can alternate between the agent and patient role �A patient can be prompted into taking action A narrative process affects a patient by: � Influencing their awareness of a situation � Altering the situation (improving or worsening) Agents � Voluntary-purposefully initiates a goal-oriented process � Involuntary-narrative impact derives from unintended side-effects of actions

BREMOND CONT. …AGAIN Also introduces character psychology � Characters � The Matrix have beliefs, motivations, and goals

MORE Patient influencing process � Their Portal actions can influence the outcome of the agent

DIALOGUE MANAGEMENT

DIALOGUE MANAGERS History and beginnings � Colossal Cave Adventure Progressions � Neverwinter Nights Dialogue Manager � Center of interactive language systems Speech or text based � Responsible for what is said and is talked about Based on historical information, goals, possible actions

DIALOGUE MANAGER (DM) Finite state machines, frames, stacks, inference engines, planners Several definitions historically � Process decide what is said at time steps � Similar to goal oriented control structures DM favored system initiative vs. user initiative vs. mixed initiative � System: system drives dialogue Final Fantasy Tactics � User: user drives dialogue, system responds � Mixed: user chimes in, system responds and redirects

FINITE STATE MACHINES Most popular Dialogue Management Paradigm and implementation technique of commercially spoken dialogue systems Neverwinter Nights

FINITE STATE MACHINES Distinctive advantage with spoken language � Use specially tuned acoustic language model for each dialogue state � Know what “utterances” to expect Makes auto speech recognition task easier Suited for special situations � System has dialogue initiative, dialogue states, dependencies between are defined and not many

(FSM) EXAMPLE CONT. Milk Br Swee t SA Tea Choos e Food Choos e Coffee No Milk Br Br Swee tt Milk Br Latte No Milk Br Br Pure e Sugar South African Cake Br Br Swee tt Latte Br Brazilian No Sugar Italian It Br Not Swee t

FINITE STATE MACHINES Provides straightforward way from task breakdown to DM implementation Easy to check uncovered conditions, shortest paths, cycles etc… Difficult lies in growth � Problematic with interrupts to system directed dialogue � Standard: ignore -> steer user back to original

(FSM) SUMMARY Good for simple, informative characters Implementing task oriented subdialogues Familiar to game developers Good starting point when implementing characters with dialogue capabilities Limitations � Need all data to adhere to ordering constraints � Any new information not expected is discarded Examples: Text based

FRAMES Second most popular dialogue modeling technique Foundational paradigm of Voice. XML (Expand) Widely used in commercial dialogue systems Used to fill a form and populate and/or query a database Typical applications � Transport timetable info, call routing (expand)

FRAME BASED DIALOGUE SYSTEM (FBDS) Gets its name from way info is gathered from user Frame is viewed as object in object oriented paradigm with no defined methods Frame keeps track of wanted info Algorithm determines what to do to fill missing items � Prompts initial question, fill as many slots as possible with user’s current “utterances”, ask questions to clarify and fill remaining slots

FBDS Necessary to keep track of info and create clarification questions � Need to keep confirmation flag � Need confidence values associated with slots Similar to learning tables Stronger in case of spoken dialogue systems Need to address ambiguity and interpretation problems

FBDS Allow more efficient and natural interaction System able to use info not explicitly asked for but relevant to the frame � Ease burden to software engineers � Allowed to specify dialogue rules for each frame Management algorithm generates appropriate dialogue moves dynamically Limitations � � To use automatic speech recognition component must be robust Needs to be able to deal with “utterances” used to describe everything in any given frame Goes for natural language understanding module and embedded clarifications Unable to deal with info that falls out of current frame but is still relevant and supported Not used in video games more so commercial products � Human and computer interactions

STACKS Provide natural way to change topic of conversation and resume halted one � Any new conversation is pushed over the old and then the old is popped once the new is done Can be integrated or independent COMIC system � Uses stack and augmented FSM as basis of DM � AFSMs called Dialogue Action Forms (DAFs) Has ability to execute arbitrary action in state transitions Wait for arbitrary and external info Indexing terms like keywords

STACKS DM changes topics through DAF creation, indexing and selections DAFs created with properties such as verbs, nouns, entities and restrictions on world properties Combo of keys are put in index When system receives “utterance” makes key from bits of info from sentence Then selects DAF most closely matched and is put on top of current one When finished it is popped and pervious one resumes

STACKS Limitations � Auto speech recognition needs to have one general layer capable of identifying all “utterances” that lead to a topic or task shift � Natural language understanding module needs to be able to spot keywords, dependencies, entities that signal topic shifts � Tuning the indexing and retrieval mechanisms is challenging task in itself � Needs more sophisticated language generation module capable of summarizing what was said before and introducing appropriate cues and intros to resume previous converstations

STACKS Implementation � Most games use this type of dialogue manager � Tutorials � Sequences for more information Ex) Dragon Age, Brave Fencer Musashi

INFERENCE BASED SYSTEMS Has four components � Knowledge base is composed of declarative rules in logical formulism � Propositional logic, first order logic Inference engine responsible for finding valid proof of a fact � Knowledge base, inference engine, working memory, facts selector Supports unification, forward/backward chaining Working memory is where facts of current interests are kept Facts selector is an algorithm that chooses and combines facts of interest before put in the inference system NICE game system – hybrid system

IBDS Limitations � Difficult to use tuned auto speech recognition model for different dialogue parts with an IBDS � Language understanding module needs to provide enough information to populate working memory with relevant facts Having good knowledge base information to guide interpretation of “utterance” Mass Effect � Blue: Charm, Red: Intimidate

PLAN BASED SYSTEMS Integral part of research and cutting edge commercial dialogue management systems Basic structure � Set of operators and procedures to find sequence of operations that achieve one or more goals � Operations usually specified in terms of preconditions and effects Two basic common uses � Encoding speech acts/DM output directly to operators’ actions � To select facts of interest to be fed into the system

PLANNED BASED SYSTEMS Ordering exists between actions to organize; need to know how many people and where will they sit Offers same complications for auto speech recognition and natural language understanding as an interface based system TRIPS Final Fantasy 9: Cooking (1: 39)

GAMEPLAY AND STORY Brief history of story in games Modeling faction Interactions Applications to RPGs Faction modeling in action

STORY IN GAMES More of a recent development Earlier games relied heavily on gameplay (Megaman, Pong, Super Mario Brothers) Still plenty of games with limited base story: Super Mario Galaxy, Katamari, Call of Duty, Sports Games, Fighting Games.

MODERN GAMES Games are now expected to have at least some type of story. RPG storylines have been growing in scale, and just keep getting larger.

MODEL PARAMETERS Assuming two factions (X, Y), each faction gets a parameter (x, y) where x or y is that faction’s level of cooperation towards the other. Higher parameter value means more cooperative. Factions may also have other parameters for belligerence or pacifism. All the parameters are evaluated to decide the factions’ behavior towards each other.

EQUATIONS TO USE PARAMETERS (BASED ON MODELING AN ARMS RACE) Equations describing intended behavior: �x = ky – ax + g � y = lx – by + h � k and l are fear constants (mutual) � a and b are restraint constants � g and h are grievance terms If ab > kl, there is equilibrium. If ab < kl there is unstable equilibrium.

REINTERPRETED FOR RPGS Same basic equations �X = Ky – Ax + G �Y = Lx – By + H K and L are belligerence factors A and B are pacifism factors G and H are friendliness towards the other faction If the result is above equilibrium, they are in cooperation, if they are below, they are in competition, and on or around eqilibrium is neutrality.

BEHAVIOR IN ACTION Parameters may be applied toward random encounters with other factions. � More hostility = more difficult battles. � Allied factions assist in battles � If factions are cooperative, maybe a negotiation encounter will occur. Also effect conversations as shown in dialog trees.

EXAMPLES IN MODERN GAMES Fable Fallout (Good/Evil) Neverwinter Nights Series Oblivion � Almost every NPC belongs to some faction: Fighter’s guild, Mage’s Guild, Thief’s Guild, Dark Brotherhood, Arena. � Also minor guilds: Blackwood Company, Knights of the Nine, The Blades, Order of the Dragon, etc. � The faction’s standing towards the player effects dialogue and aggressiveness.

ENDINGS OF GAMES Some games use character or faction relations to change the endings(Star Ocean Series, Heavy Rain). Evaluate relationship values and present different scenarios based on those values, allowing different endings for multiple playthroughs.

STORY IN GAMEPLAY Normally, player expects story to effect gameplay, newer games make it work so that gameplay also effects the story. Example: http: //www. youtube. com/watch? v=SQCBLs. Jhc. Do#t =02 m 35 s