Nothing for Dinner (Nicolas Szilas et al, IDtension)

IDtension_conversation_menu.png

“Nothing for Dinner” is an interactive drama released last fall by Nicolas Szilas and collaborators, using a tool called IDtension. Szilas works out of the TECFA Lab at the University of Geneva. If you read my writeup on the book Interactive Digital Narrative, you’ll have seen a mention of Szilas’ article there. Though it would have been out of place in the book overview, I wanted to come back and look more closely at what “Nothing for Dinner” accomplishes.

The premise of the story is that you’re a young man whose father has suffered a stroke that affects his behavior and memory. You need to get something ready for dinner, but your father keeps getting in the way, and other events spontaneously happen — a school friend coming over to get back a textbook she left at your house, your sister’s DVD player breaking, a phone call from your mother with extra chores — to add blocks to your progress.

The system is clearly quite dynamic: I played three times, and the sequence of events was very different each time, with some blockers appearing only in one of the playthroughs. Also, the conversation menus are dynamically generated to let you try various approaches to any of the currently-active problems, or to give emotional feedback to the other characters about what they’ve just done.

IDtension_grandma.png

If you try to cook dinner alone, your father resentfully complains that you never want to do anything with him; if you try to involve him, he may get annoyed and refuse to help you; if you let him cook by himself, he’ll break things and make a mess. And whenever your father gets upset, your grandmother comes over to chide you for not looking after him.

It’s a very effective mechanism for making me rapidly resent my entire family for offloading all the emotional and practical labor onto me: like a time management game, but with more passive-aggressive commentary, and less opportunity to get anything done.

The best success I had was when I first let Paul try to cook on his own, let him break a bowl, and then offered to cook alongside him. Here we are in a charming father/son moment waving knives above our bowls:

IDtension_vegetables.png

Taken either as a game or as a drama, I found this a bit frustrating. It was easy to get stuck in a loop of trying the same or related tactics over and over again — on my second playthrough, for instance, I spent a lot of time trying to get Paul to cook with me but before he’d had a chance to try by himself and screw up. He would refuse, I would try to cook by myself instead, he would get angry that I didn’t want to spend time with him, Grandma would intervene and tell me to be nice to him, we’d calm Paul down, I’d ask him to work on dinner… repeat, repeat, repeat. The system didn’t seem to react to the fact that we’d been through this whole cycle many times before. (No doubt it works that way sometimes in real life.)

When I got frustrated with that cycle and decided to act less productively, though, I still couldn’t break out into a new kind of experience. I yelled at Paul, he yelled back at me… and still nothing really changed. In my internal narrative of the protagonist, something significant had just happened: I’d just lost my patience, stopped accommodating Paul, and started interacting with him negatively. It was a character transition point. But the game had no way to acknowledge that.

I think part of the issue here is that often the obstacles introduced would slow progress, but not in a way that either changed the overall stakes of the situation. Ask Paul to make dinner, he breaks a bowl, I have to go clean up the bowl… and once that’s over, the world state is the same as it ever was. Nothing has been learned and nothing has significantly changed. If there was an element of the engine behind the scenes that was doing overall tracking of dramatic tension, I didn’t detect it.

Another issue was that even the characters without memory problems seemed not to preserve a sense of context for long. I could compliment my sister Lili on an action and then turn around and immediately criticize her for the same action: the game didn’t seem to notice, and it didn’t appear to make any difference either way.

So in general, the cumulative effect of all this was to make me feel as though, even though I was operating in a rich world and drama simulation, most of my actions had little perceivable consequence.

There were also some lower-level issues around polish: the English isn’t always very idiomatic, and the dialogue generated can feel quite stilted.

On the positive side, though, one can definitely tell that the system is generating and directing NPC actions in order to produce particular dramatic effects, and that the characters are constantly reacting to changes in world state. Sometimes the two characters I wasn’t currently talking to would be having their own conversation in the other corner of the room. Sometimes the conversation generator would offer me some pretty complex (but still generated) options about what to say: for instance, asking my grandmother to help me convince Paul to work in the kitchen. There’s a lot of potential expressiveness there, if only the effects of actions were stronger.

3 thoughts on “Nothing for Dinner (Nicolas Szilas et al, IDtension)

  1. This was very interesting. I didn’t have as much variety in my playthroughs as you, maybe because I somehow found a way to get Paul to help me with dinner the first time through without getting any bowls broken.

    With the language it was maybe a bit obscured by the non-idiomatic English, but the dialogue was often right on the nose in a way that’s pretty common in this kind of project, it seems, where the dialogue straightforwardly reproduces the name of the event being discussed. Like, if your option is “praise Olivia for dealing with the alcohol problem,” your quip will be “Olivia, that was a good thing you did, dealing with the alcohol problem.” Sometimes it reminded me of the survey for Heart of Shadows that asked you to pick among machine-generated quips, which might be:
    “1. By the world of ice and death, what do you think about the fact that Fahn has run away?
    2. Um… what do you think about the fact that Fahn has run away?
    3. Yeah, what do you think about the fact that Fahn has run away, alright?
    4. Yeah, what do you think about the fact that Fahn has run away?”
    as though it’s plopping the name of a stored action right into the middle of the sentence. I feel like for things like this there needs to be an extra layer of generation between the concept and its representation in the quip, so that you can contextually represent the thing you’re expressing in terms of other facts that are involved. So something like “Good job. The alcohol interferes with Paul’s medication” (contextualized by the medicine issue) or just “Thanks for talking to Paul” (contextualized by previously asking Olivia to talk to Paul about alcohol) could work. But not only is this very hard to do, it’d be hard to make sure that the generated language actually succeeds in communicating what it’s supposed to; sometimes the system might think that “Thanks for talking to Paul” clearly invokes “about alcohol” when the player doesn’t.

    I was impressed at the number of different ways of expressing the ideas the game had. Some of the layered ones like “suggest to Lili that maybe it would be a good idea for Paul to go shopping” seemed a bit odd, but when Lili came up to you and said something like “Maybe it would be a good idea for Paul to go shopping” it seemed apparent that the NPCs were playing by the same rules as you, which was nice.

    Also the animation for Paul was really expressive, I thought.

    (For anyone looking, the link to the game is hard to find even on the IDTension page; it’s at nothingfordinner.org; scroll down for links to Play in French or English.)

    • Thanks for the detailed comment! It’s very interesting to hear about someone else’s play experience.

      Re. finding a better, less on-the-nose representation for information — I believe James Ryan‘s “Talk of the Town” project does some text generation/searching work where it determines what fact it wants to convey and then seeks an appropriate context in which it can embed that information as part of a larger anecdote. I don’t think that project is available to the public yet — I’ve only heard about it in the context of academic papers — but I could see something similar (if probably smaller-scale) working here.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s