Outline

• Interactive Fiction
  • History
  • Essential features
  • Existing UI limitations
• Skald
  • What problems I expected Skald to fix
• Evaluation
• Results

Colossal Cave Adventure

• 1976: First text adventure game.
  • aka ADVENT or Adventure, by Will Crowther.
  • Based on Flint-Mammoth Cave system in KY, with a few fantasy elements
• Spread among university mainframes
  • Expanded by Don Woods (1977); prompted many ports, variants, and imitators
  • Inspired MUDs (Multi-User Dungeons)

Zork

• Inspired by Adventure
• Zork by students at MIT (1977 - 1979)
• Creators then formed a company named Infocom

A PC release of Zork I by Infocom

Commercial Boom

• Infocom (and others) boomed in the 1980s
• Moved beyond dungeon exploration -> richer plots and more genres (sci-fi, mystery, etc).
• Term interactive fiction coined (partly to elevate the genre)
• Text adventure games -> graphical adventure games (such as by Sierra)
• Commercial text games ended in 1993 with Legend Entertainment's Gateway II: Homeworld

Gateway II: Homeworld

Interactive Fiction

• A small community of independent authors and developers continues
  • Yearly IF competition; archives of IF games; tutorials and advice
  • 2 major free authoring systems
• Inform
  • Since 1993; compiles to Infocom's old ZMachine VM format; new formats too
  • Recent: Browser/JavaScript-based interpreters (Parchment, Quixe)
  • Recent: Inform 7 (2006) uses natural language-like programming syntax
  • Simpler, more popular; sometimes technologically cruder
• TADS
  • Since 1987; version 3.0 (2006)
  • Richer world model and more programming language features than Inform
  • Recent: Includes server within VM; can serve custom HTML/JavaScript UI
• (handful of other IF authoring systems)

Serious Applications of IF

• (Artistic Expression)
• Education
• Good for interactive drama/storytelling research
  • Easy simulated world model; rapid prototyping
  • Weyhrauch - Tea for Three (1997)
  • Nelson and Mateas - Anchorhead (2005)
  • Sharma et al - Anchorhead (2010)
  • Tomaszewski - Demeter (2011)

Essence of IF

• An explicitly-modeled narrative world with a text-based user interface
  • explicitly-modelled world
  • narrative (at least entertainment) purpose
  • text-based UI (output, input; usually turn-based)
• Nick Montfort (2003) says something similar:
  • displays text, accepts text, responds with more text
  • a world model simulated in software
  • a natural language parser for input

Borders of IF

• Hypertext fiction (no modelled world; no parser)
  • Twine (fuzzy)
• Graphical adventure games (not text-based)
• MUDs (fuzzy; less narrative focused)
• Simulations (not narrative/entertainment driven)

Experience of IF

Can consider as (or compare to):

• computer program
• narrative, story, or novel
• roleplaying game
• puzzle
• virtual world
• riddle (Montfort 2003)
  • systematic world, presented for "solving"
  • balance of challenge and difficulty
  • literary, puzzling, intriguing
  • often marginalized as trivial, despite a rich history

IF's UI Limitations

Often unclear:

• which objects exist and support interaction
• which actions each object supports
• which words or synonyms are allowed
• what grammar or syntax is required
• (what the game objectives are)

Essence of IF's UI Limitations

The author (and IF system) can't support everything the user might want to do

System Image and Affordances

Norman, Design of Everyday Things (1988, 2002)

Poetics for Interactive Narrative

Tomaszewski & Binsted (2006)
Based on Mateas (2004), Laurel (1991), Smiley (1971), and Aristotle

Clearly Affording Objects + Actions

Designs considered:

• real-time parser
  (possibly with auto-complete)

Clearly Affording Objects + Actions

Designs considered:

• real-time parser
• lists of actions and objects

Clearly Affording Objects + Actions

Designs considered:

• real-time parser
• lists of actions and objects
• inverse parser / command builder

Clearly Affording Objects + Actions

Designs considered:

• real-time parser
• lists of actions and objects
• inverse parser / command builder
• object-based command menus

Clearly Affording Objects + Actions

Designs considered:

• real-time parser
• lists of actions and objects
• inverse parser / command builder
• object-based command menus
• in-text action menus on specific objects

Skald

Skald Implementation

• Backend
  • TADS (for server + game engine)
  • TADS plugin (custom processing)
• Frontend
  • Google Web Toolkit (JavaScript browser UI)

Effects on IF Experience?

• Assumed gains
  • No input errors
  • Clearer afforances (verbs, objects, commands)?
  • Easier to use?
• Possible losses
  • Less "open-ended" (no illusion of NL input)
  • Less "puzzling" (now multiple choice)
  • Less dialog-like (passive point & click)
• Outcomes?
  • More "fun"? Preferred by users?
  • Change how users play?

Experimental Setup

• Online study, recruited by email
• Consent form + background survey
• Play two short games (constant order)
  • Captain Fate
  • The Queen's Heart
• Two random groups
  • Group CS - Traditional CLI followed by Skald (11)
  • Group SC - Skald followed by traditional CLI (9)
• Collected
  • survey responses - after each game + final comparison summary
  • metrics (time spent, commands entered)

Response Survey Questions

• Scale: 5 (strongly agree) to 1 (strongly disagree)
• Affordances
  • Object: It was clear which objects I could interact with in the game world
  • Action: I knew which actions I could perform in the game
  • Command: I was able to able to construct commands that the game understood
• World-level agency
  • I was sufficiently able to direct my character's actions in the game world--such as move from place to place, manipulate objects, affect other characters, etc

Response Survey Questions

• Scale: 5 (strongly agree) to 1 (strongly disagree)
• Story-level agency
  • I usually knew what I was expected to do in the game (even if sometimes I had to figure out exactly how to accomplish it)
• IF Experience
  • Narrative: The game session had a story-like structure
  • Puzzles: The game contained puzzles or situations that required some thought or exploration to overcome
  • Freedom/Signficant Choices: I believe that the game probably would have had a different outcome had I performed different actions.
• Overall
  • I enjoyed playing this game
  • (Open-ended: Most and least enjoyable aspects)

Comparison Survey Questions

Which interface...

• Easier?
• Faster?
• More enjoyable?
• Made the game more interesting or engaging?
• Would you prefer to use to play a third game?

Scale: CLI (definitely) - CLI (slightly) - No Pref - Skald (slightly) - Skald (definitely)

IF is not for everyone

• 30% of those who completed background (41) failed to finish first game (31)
  • No clear difference across UI on this
  • Lost 12% more on first response survey (26 complete)
• Some participant comments suggested they did not enjoy the games regardless of UI

Traditional IF UI has high error rate

• 25% of user inputs did not map to author-supported action
  • 21% produced error messages
  • Similar numbers (20%) seen in previous study (Tomaszewski 2011)
• Mean error rate: 36% of inputs
  • Some users entered a lot of errors
• Every CLI user entered at least one invalid input

Skald is easier to use

• All^* inputs mapped to author-supported actions
• Affordances rated as significantly more obvious
  • Objects: 3.95 Skald vs 2.7 CLI (p < .0001)
  • Actions: 4.15 Skald vs 2.85 CLI (p < .001)
  • Command construction: 4.45 Skald vs 3.55 CLI (p < .001)
  • 70% explicitly stated that Skald was slightly or definitely easier to use than CLI

Skald encourages more exploration

• 2x the number unique game world objects used (both games)
• Some objects used to a greater extent
  • Particularly background or scenery objects
• Wider range of unique verbs (Captain Fate only)
  • Average number of unique verbs used: 13.3 Skald vs 10.1 CLI, p < .01
• Examine verb used ~3x as often (both games)

Skald does not signficantly affect play time

• Not significantly different between UIs:
  • Time spent playing
  • Number of inputs entered
• So effectively turning errors into exploration

Skald does not eliminate affordance issues

• "Figuring out what to do" still often listed as least enjoyable aspect
  • Available options unclear?
  • Too many options to choose from?
  • Unclear story objectives?
• "Figuring out what to do" also listed as most enjoyable aspect by others
  • So Skald not completely undermining "puzzle" aspect of IF

World-level agency requires more than material affordances

• Poetics theory of IN:
  • Material for action (affordances)
  • Formal constraints (guidance) on appropriate choices
  • Balanced → user agency
• Skald improved object, action, and command affordances...
  • But did not improve world-level agency!
• Did not correlate:
  • Story-level ("knew what I was expected to do")
  • World-level ("sufficiently able to direct my character's actions in the game world")
• Something more going on here...

The game matters

• Game = content
  • Narrative: characters, story, narration/description
  • World: verismilitude, completeness
  • Game: puzzles, complexity, subtlety
• On measures used here (regardless of UI):
  • Time required: 9.1 min Fate vs 13.0 min Queen (p = .004)
  • Object affordances: 3.0 Fate vs 3.65 Queen (p < .001)
  • Action affordances: 3.15 Fate vs 3.85 Queen (p < .001)
  • Story-level agency: 3.65 Fate vs 2.6 Queen (p < .001)

UI has only subtle effects on story experience

• No significant user-reported differences at story level
  • Story-level agency
  • Story-like experience
  • Contained puzzles that required some thought
  • Appearance of significant user choices
• IF experience apparently preserved...
  • Did increase the range of possible game endings encountered
  • Fewer QUIT endings
  • Fewer comments praising the puzzle challenges

Skald not a clear winner overall

• 30% would still strongly prefer traditional CLI for future games
• Similar spread to extremes for
  • Easier to use
  • Faster to use
  • More enjoyable
  • More engaging
• Similar spread seen previously in other research
  • NL vs menu-driven dialog systems in games (Mehta et al 2010; Sali et al 2010)
  • NL input more difficult, opaque, error-prone
  • Also has greater potential, feeling of freedom
  • Engagement depends on user

User background has some effect

• When using CLI, previous CLI experience corresponds with higher ratings for
  • Constructing valid commands
  • Exerting world-level agency
  • Enjoyment overall
  • Preference for CLI for future games
    • Faster, more engaging, more enjoyable
    • But not easier
• Previous CLI experience did not have a significant correspondence with Skald ratings
• When using Skald (only), previous IF experience corresponded with greater affordances
  • Object, action, command affordances; world-level agency
  • (I can't really explain this)

Conclusions

• IF has a long ongoing history
  • Valuable applications beyond entertainment
  • Education and research, where often exposed to new users
• Provided Skald UI as reusable library
  • Modern, web-based
  • Works with one of main two IF platforms
• Empirical data on its effects
  • Easier to use
    • Eliminates invalid user input
    • Clearly affords objects, actions, and commands
  • Encourages greater exploration of the world (and thus of the story)
    • May be a trade-off for puzzle-based games
  • Not every player's preference

Future Work

• Expand Skald to include a (non-default) command line option
  • Probably need to rebuild it
• Re-examine the interactive narrative poetics and what it claims about user agency

Questions?

ztomasze@hawaii.edu

http://www2.hawaii.edu/~ztomasze/argax/slides/