Moe Computing Research Center (MCRC) – Tsundere Computer Development

Posted On May 27, 2008

Filed under Analysis, koneko-chan

Comments Dropped 11 responses

If all of the science and engineering research assistants look as cute as Otome, then everyone would be doing research and we’d have military grade combat mecha by now (or we’d get no work done). Sigh. Such is the unfortunate state of being an engineering and science nerd.

whitehatazn-48.jpgRecently at work, my project has entered another stage of testing. What made this development version interesting was some of the new integration components. Testing of some of these new components requires cooperation with other software programs within the system. This is fine for most systems. But when dealing with the redundancy requirements for a military grade system, both the system and software engineers have to pay special attention to the various states of each computer component in the system. Knowing the possible conditions of the computer component and its connected counterparts makes diagnosing problems in the system easier.

My mind sorta stayed in that mindset in the past few weeks and when I read tjhan’s attempt at tsundere mathematical modeling. After looking at one of the best math graphs I’ve ever seen, one idea popped into my head: I can model the same thing using my short professional career as a systems engineer. Designing the tsundere computer system is what I will discuss today.

In both computer science and computer engineering, a state machine is a system which can be described as a set of states and the transition between these states. It is considered to be finite (FSM) when states in the system are known to be of a limited number or the system is reduced to definite number of states. In both computer fields, state machines are used in a number areas from to modeling programming code flow to creating sequential circuit systems. Computer science and computer engineering students are taught the basics of these systems in the undergraduate level. (In comp sci, I believe it’s taught in the theory of computation which leads to automata theory. In comp eng, state machines are the building blocks for computer organization.)

State machines have also been used to study areas outside of computer systems. Human nature can be described as a form of an emotional state machine. If I asked someone “How was your day?”, the response may be different depending on how that person is feeling. I could ask that same question at 5pm every day for 7 days and I could get a different response each time even though my input stays the same. However, it is widely understood that human nature should be modeled at least as an abstract state machine because of the numerous amount of states one can process from things like gender and race.

Yet, the world of anime has graced us with the tsunderekko which works well with an FSM model. The reason being is that tsundere characters only have two primary states. With such a small number of states, the number of transition states are only a handful which make the state machine easy to model. Below is my attempt to generate such a model:

There are a couple of things to note here:

  • I introduced a normal state into the processing. This is to add some stability into the system. Any interactions outside of the tsundere state set (tsuntsun and deredere) are in this state and most of the time, this is the starting state. There are some shows where the tsunderekko would never reveal the normal state to the audience. Again, this is to make the design complete.
  • Note the two non-normal transitions: The Tsundere Moment transition is the key state interaction needed for the tsunderekko. Without it, a tsundere character would be incomplete. The dashed connector for the normal-deredere transition is also important. I find the best tsundere model is one where the character never enters the deredere state until she enters from the tsuntsun state first. This makes the first Tsundere Moment transition more impactful and allows for any deredere state to be traced back to the original defining tsundere transition.
  • I did not include a transition from deredere to tsuntsun. Some would argue that the transition (possibly modeled with a dashed line) is needed to be complete. I found it unnecessary. The deredere state is too consuming to the tsundere character. Transitioning to the tsuntsun state without traveling through some “wake up” (normal) state is too quick of a change. For some characters, the dere-to-tsun transition is fairly active and works ok (Kumiguya Rie characters come to mind). The model without that transition should describe most tsundere characters.

One characteristic of state machine models is that they can be expandable. Each state can be broken down into smaller sub states with more transitions. Knowing that, I’ve decided to generate a detailed state machine model of one of my favorite tsundere bishoujos: Kyou from Clannad.

To model the Kyou FSM, I used a Mealy machine model which models state transitions based on the current state and the current input. Each connector has both an input causing the transition and an output caused from the transition. I color coded the transitions to make the paths a bit easier to read (starting state = normal state). I also highlighted the major tsundere path that we all know from Clannad TV series.

So what can you do with these models? Like I said before, FSM models are used in a number of areas within computing. As for the general tsundere FSM and the Kyou FSM, you can apply these into many implementations. The one that comes to mind is high level AI generation for various types of computer programs from the fairly simple Ren’ai conversation tree to possibly as complex as an android (which this blog will 100% support any university attempt at MCRC Kyou persocom development).

A visual screenshot representation of an emotional state path. In this case, the scene is the memorable Kyou-Tomoyo confrontation and the screenshot order is the annoyed state, the competitive state, the competitive-surprised transition input, the surprised state and the embarrassed state. Note that being in an emotional state does not have to last long since the last three screenshots occur in about 4 seconds.

There are two other things I’d like to note about tsundere FSM models. First, I don’t think I can replicate a tjhan’s mathematical model of sinusoidal wave. If I place integer outputs on the states or the transitions and attached a counter, the one-way primary transition from the tsuntsun to the deredere state causes a number of problems. The model would end up converging in one direction either towards +infinity or -infinity depending on the viewier’s bias.

Second, looking at this makes me realize that the tsundere archetype is very limited and simplistic. Now, I was able to take only high level estimation of a tsunderekko. However, I modeled about 70% of the Clannad TV’s Kyou in less than 10 states, far from the multitude of complex states in a human ASM. It is disappointing to see how otaku (including me) are attracted to the simplistic and predicatble representations of female personalities in our world.

The MCRC is still looking for new researchers for a number of projects. Among these, tsundere computer development is at full speed with Shana persocom development and the Hinagiku defense system additions. Combine that with the Combat Waitress program and the MCRC is among the elite in moe research. So please submit your CV’s. We will not let the Mecha Research Triangle make the latest developments first!


11 Responses to “Moe Computing Research Center (MCRC) – Tsundere Computer Development”

  1. meganeshounen

    Wow. Just… wow. Totally takes me back to my college days, this does.

    You’re right about all the possible states a normal tsundere (or even a girl) could be in. It’s nigh impossible to map all those states, but it’s still possible to create several normal “starting states” that every girl could experience (boy-meets-girl, embarrassment, happiness), then work on the finer details from there. And worse for real life models, connections can change too, like something that could make a girl happy turning into something that could make her sad the next time it happens. Installing a “growth” module could make them more “realistic” by adding additional states and interlinking them to existing ones, either via self-realization or through real experiences.

    There are a lot of robot girls around, but those of note (who have probably surpassed their mechanical and programming limits) are Chachamaru from Negima and Aigis of Persona 3. Coincidentally (to the opening picture), Da Capo’s Miharu and her descendant, Minatsu are good examples. Especially Minatsu, who is pretty much… the definitive tsundere robot girl.

  2. Baka-Raptor

    That final FSM is my new wallpaper.

  3. Hegemon

    The following contains extremely dry mathematical humor. You have been warned.

    I believe that your difficulty in obtaining the desired function from constant transitional values comes mostly from your attempt to outright model the phenomenon as a deterministic finite automata (DFA). Rather, you might meet with more success and insight into the problem if you instead take the approach of the nondeterministic finite automata (NFA), where it would be possible to have several “Normal” states defined as some superposition of tsuntsun and deredere, most of which would be, of course, unstable. Thus, the model could be formulated, with enough states, so that transitioning from these unstable superpositions would cause multiple transitions of low total value (using equivalent transition conditions on each with nearly equal value and opposite positivity), whereas transition from a pure state would result in a high value from the total sum, mimicking the squared cyclic function exhibited in your cited source. As you are well aware, this can be modeled as a DFA which has at the least three states (much like your original formulation) if we use only the most basic tsuntsun and deredere states in the NFA, or some exponentially increasing number with the number of states you use to model the nondeterministic formulation.
    You might consider this less canonical, but more flexible, approach in your next modeling experiment.

    Next time I hope to hear your thoughts on using a two qubit quantum computational model, representing the tsundere state and environment, respectively, with an unbounded number of unitary gates to represent more compactly and analytically the superposition and entanglement often physically observed in such systems.

  4. Onlooker

    After reading your article, I’m impressed at which you can handle all this data and interpret it. I’d like to know if the same interpretations can be made from other archetypes within anime or can this only apply to the Tsundere archetype. If the flow were to reverse, would it result in an entirely new archetype or reflect the yandere archetype? There are so many different ways, I’m interested in how they are going to be organized.

    P.S. Any chance the the MCRC can work alongside the Mecha Research Triangle? You guys could make the next Evangelion if you wanted to. With even more mood swings!!

  5. koneko-chan

    @ meganeshounen, Hegemon: Having more normal states for an NFA-like model would make this more detailed. This is something I didn’t explore initially since I have never taken a class in the Theory of Computation when going for B.S. in Computer Engineering. However, exploring a more elaborate model would end up exploring more of the general female emotions rather than the tsundere character herself.

    @ meganeshounen: Adding a form of growth module could make things potentially interesting. However, I’ll defer that analysis to someone who studied machine learning or dynamic reprogramming. I could make a small attempt at it since I know quite a bit about Genetic Algorithms (GAs).

    @ Hegemon: I was able to understand more than half of your rant with my knowledge in this field. I’m not too sure if I should be proud about that…

    @ Onlooker: Yes, yandere could be a reverse case. Keep in mind here that this post came about because I was reviewing similar design models at work. I selected tsundere because I knew that it would be fairly easy to model using FSMs (the Kyou FSM took about an hour to design correctly). Other archetypes may need to be designed using a different kind of model. I do not want to spend too much time on this unless I’m really motivated or in the right mindset. Because of work, I was thinking about creating a feasibility white paper on the dynamic configuration change management of Kallen’s Guren in Code Geass R2 episode 6 for my next post (and thus further cementing my nerdiness).

    I do like my moe to be untainted by the likes of mecha. Besides, being that I work for a defense contractor, I find competition to always be good for pushing R&D. Also, I do really like the idea of a Combat Waitress project.

  6. iniksbane

    Wow… I’m not sure what to think there.

    But.. wow.

  7. Moe Computing Research Center (MCRC) - Harem Resource Management « Anime Academy

    […] to be a better way to teach some of this dry stuff.” Remembering that I once wrote about tsundere state machines, I thought of the various ways I can teach computer science/engineering theory with anime again. […]

  8. Internet Deprivation Diary « The Animanachronism

    […] and ‘new tsundere‘ (alternating between cold- and warm-hearted, possibly using some sophisticated charts) ignores the possibility of a combination, a character who alternates between moments of antagonism […]

  9. NegativeZero

    You know you’re a nerd when you stumble upon an entry like this and it gets you thinking about whether it would be possible to structure an entire galge around a few characters with personalities defined using Finite State Automata, rather than the usual flags-and-checks dialogue tree systems.

    After considering this for a while, I think that, on its own, an FSA is really not quite enough. They certainly show a decent mapping of personality, but their main flaw is a lack of memory. Certainly there would be no way to model yandere without some kind of memory involved, You could probably do it with some trickery, but it would be quite inelegant. You could probably model just about any haremette using a pushdown automata, with the stack used as a love-love meter to track her affection towards the Generic Male Protagonist. For yandere you might also need some kind of sanity or jealousy record too,

    And of course, the elephant in the room is that we’re trying to apply some kind of definite, logical, repeatable and predictable model to female behavior. 😛

  10. koneko-chan

    @ I agree that FSMs have limited memory and there are possibly other ways to generate the models. However, the point of this post was to model the tsundere by her “known” attributes through show episodes. Going beyond that will generalize the tsundere model into an attempt to model female emotions as a whole (something I’m not going to bother trying to model for how complicated it will be). Also, complicating the models further will make this post no longer an intro lesson for some of the non-compsci readers out there (yay for readership).

    I think the only predictable and repeatable behavior is the response to the double-edged sword of replying when she asks: “Do I look fat in this?”

  11. Pervy Otaku

    Wow, with all this Data, it is only a matter of time until we have real Persocoms (Read Chobits)… Ah, Just thinking about a real life Chi make me all giddy! Lol, j/k, but for real, it is a distinct possibility that Persocoms may soon be a reality

Leave a Reply

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

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

Google+ photo

You are commenting using your Google+ 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 )


Connecting to %s