I made an animation for my recent neuroscience article on dual-task interference. It was nice to finally visualize a scientific idea, after making it my new year’s resolution for so long. Graphical summary of the paper in my Twitter thread.
Recent events reminded me of the flight of an insect-like drone reported last year in Nature. As I watched the small machine flapping its delicate wings toward the light, its energy source (YouTube link), I felt as though I was watching the first steps of a baby. At the same time, I couldn’t help but wonder what it would grow into: a nimble rescuer, or a killer under unexpected circumstances?
This time, I received answers from four people, and while some names matched mine, no one got everything matched!
The original names that inspired me and the reasons were, in order:
- Flissil: sounds fluid and at a loss (1/4 matched this name)
- Horgous: sounds hollow and lonely (2/4 matched this name)
- Axittic: sounds like an axis + excited (0/4 matched this name)
All in all, this week’s results showed that there are cases when it’s difficult to tell the names (3/12 answers matched the original), unlike last week’s (all 12/12 answers matched the original).
Here are some observations on individual responses:
- It was interesting to note that when you focused on features similar to me, you also guessed the same word. For example, there was an answer that said Flissig is like “Flüssig, can’t stand, erect, define”, which was very similar to my original intention.
- Also, when you focused on different features of either the word or the drawing, you guessed different words. For example, while the second drawing was also said to be “Horgous” because it was a “male name”, it was also named “Axittic” by some of you because it has a stick – an axis.
- Finally, there were times that the drawing itself felt different to you than to me, like the last one. I was thinking of a whirling axis with flickering lights, and while one of you suggested it looks “proud and gorgeous”, many of you instead suggested it looked “watery”, “impermanent, changing being, like flowers wilting in a pot”.
In the original study, participants chose among 4 names. The original name was chosen ~31% of the times, which was higher than the opposite (~21%), unrelated (~24%) or similar (~24%) names. So the difference was statistically significant but not huge (~1.3x to 1.5x more likely to choose the original name than others), which was consistent with what we have seen in our two surveys.
What connects the nonword names to made-up drawings at all? The study suggests that certain properties, like
- round vs. spiky
- large vs. small
- masculine vs. feminine
are inferred consistently across participants from either a word or a drawing, and those properties may have helped people identify the original name that inspired a drawing. For example, words like “heonia” were rated feminine, and elicited drawings that looked feminine. Also, in other studies,
- “mal” was more associated with a large table than “mil”
- “bouba” was associated with a round shape whereas “kiki” was associated with an angular shape.
Some sound-meaning associations are found across different language families, which is called “absolute iconicity“. For example:
Not every language has such associations, but those are found more often than by chance, even across different language families.
Also, many languages use characteristics of sounds to convey relative meanings, which is called “relative iconicity“. To list some in the paper:
- “p” vs “b” and “t” vs “d”: mass (Siwu: tsaratsa, “a light person walking quickly” vs. dzradzra, “a heavy person walking quickly”)
- “ε” vs “o”: size (Ewe: lεgεε, “slim” vs. logoo, “fat” )
I imagine it would be nice if there were a systematic map showing the relative iconicity found across the world. That would not only be fascinating to behold, but also be useful for naming and drawing new characters for a picture book!
If you are reading a picture book with creatures named Axittic, Flissil, and Horgous, would you be able to recognize the creatures just from the names? Let’s see if you can: Pick a name for each creature in the picture among Axittic, Flissil, and Horgous, and let me know a short reason why!
As I did last week, I drew each of these creatures for one of the names. As mentioned, this activity is inspired by a recent study which showed that drawings from made-up names can be matched with the original names better than chance. So you may as well be able to read my mind here!
Last week’s answers showed that you indeed could. All 6 answers about Heonia and Cruckwic I got (on WordPress and Instagram) were correct. So again, this is not impossible as some linguists suggested. Maybe matching two names was even too easy, so let’s try matching three!
How can names be connected to creatures? The study suggests that certain sounds (such as “k”) in the name evoke certain attributes (sharpness), which in turn are reflected in the drawing based on the name. Intriguingly, some of such associations are found across cultures. I will summarize the associations next week.
Until then, please pick a name for each creature in the picture among Axittic, Flissil, and Horgous, and let me know a short reason why!
All six of you got it right! “Cruckwic” inspired the horned creature, and “Heonia” inspired the one with the dark blue background.
Among the interesting reasons for your guesses, my favorite was @riniscreature’s:
Heonia sounds like a dainty plant to me, like a flower. And then Cruckwic sounds more like a mischievous character. Like a creature hiding in a forest and giving people riddles and wrong directions.
That was a fun story I haven’t thought about!
Originally, I imagined Cruckwic as a gowned professor because that was shortly after I learned how to pronounce a professor’s name with a similar ending (according to which I would pronounce the final “c” as “ch” in “speech”). With “Cruck-“, it somehow sounded like an old mythical creature with rough skin, so that’s how it came to be. Surprisingly close to reasons some of you gave.
With Heonia, I imagined a creature that lasts eons, producing rings signaling time, deep under the sea.
My reasoning sounds like this association between the sound and the drawing would be language specific, but surprisingly some of them hold across cultures. I will discuss this next week. Until then – try this week’s, where I now put three creatures as a challenge!
In your reply, let me know which one you think is Heonia, and which one is Cruckwic, and a short reason why you think so!
These are portraits of two creatures that I conjured up from the two made-up names. Next week, I will reveal which name originally inspired which portrait.
If you feel like it, also try drawing Heonia & Cruckwic youself (in a random order) and let me know! I will try to guess yours. I tried this with my family and they loved it.
This activity is inspired by a recent study in cognitive science. It found that, when people are asked to draw creatures with made-up names, they nonetheless make drawings with consistent features. Furthermore, another set of people who viewed the drawings could match them with the original names better than chance. There are suggestions that this may be due to the “iconicity” of the words – the similarity between the words and its meaning. But that position goes against the conventional wisdom:
with the exception of words that directly imitate sounds, the relationship between word-forms and meanings is arbitrary: “There is no reason for you to call a dog ‘dog’ rather than ‘cat’ except for the fact that everyone else is doing it” (Pinker & Bloom 1990, p. 728)
More on this next week.
Until then, let me know which one you think is Heonia and which one is Cruckwic, and a short reason why you think so. If you feel like it, also draw them yourself in a random order and let me know!
I originally planned the mobile as a piece on metacognition, as it apparently is. Then I was about to accompany it with an essay about how we deliberate and reach a decision, which I studied during my PhD.
But while I was making it, I also realized that I had ideas like this, and others, from when I was little. I felt like I’ve only made it now because now I have more experience executing such ideas.
In that sense, the mobile could be also about growth: a larger self looking back on the little selves. Indeed, metacognition, or evaluation of one’s own thinking, can be one way for learning and growth for humans and machines. For example, if you were sure you would be invited to a friend’s birthday and if you were not, you would wonder about the reason and might learn more about what happened to the friend or to your friendship. That’s different from when you were unsure about getting the invitation to begin with, in which case being not invited wouldn’t mean much. Therefore, the sense of being sure, or confidence, is a form of metacognition that can help learning. Machines use confidence to learn as well: agreement between the graded sense of confidence and the all-or-none outcomes can be mathematically expressed as “cross entropy“, which is a standard measure used in training machines.
Back to the mobile, I debated whether to use wires or paper, but chose paper because each figure is planar. I glued several sheets together to reinforce them. If someone wants it in a more permanent form, I’d like to try 3D printing it.
하지만 모빌을 만드는 동안, 어릴 때도 이런 아이디어를 포함해 여러 아이디어들이 있었던 것이 기억났다. 이제야 이 모빌을 만들게 된 것은, 이제서야 그 아이디어를 실현할 수 있게 되어서라는 생각이 들었다.
그런 의미에서, 이 모빌은 성장에 관한 작품일 수도 있겠다: 큰 자신이 작은 자신들을 반추하는 것이다. 메타-인지, 혹은 스스로의 생각에 대한 판단은 실제로 사람이나 기계가 배우고 성장하는 데 도움이 된다. 예컨대 어떤 친구의 생일파티에 초대될 거라고 확신했는데 초대받지 못했다면, 그 친구나 친구와의 우정에 대해 다시 생각해보고 뭔가를 더 배울 수 있을 것이다. 하지만 초대받을지 여부가 애초에 불확실했다면, 초대를 못 받아도 별 뜻이 없었다고 여길 것이다. 이렇게 자신감은 배움을 도울 수 있는 메타–인지의 한 형태이다. 자신감은 기계들의 훈련에도 쓰인다: 자신감과 실제 결과 사이의 일치도는 수학적으로 “크로스 엔트로피“로 표현될 수 있으며, 이는 기계들을 훈련할 때 일상적으로 쓰이는 척도이다.
모빌로 돌아와서, 철사를 쓸지 종이를 쓸지 고민하다가, 각 인물의 디자인이 평면적이기 때문에 종이를 쓰기로 했다. 종이 몇 장을 함께 붙여서 단단하게 세울 수 있도록 만들었다. 더 튼튼한 버전을 원하는 사람이 있다면 3D 프린팅으로 만들어 보고 싶다.
Diggin’ into neural nets.
This illustration mixes images of a stack of sandwiches with a ResNet. The protagonist is adding a “skip connection” (blue wire) between neural layers (slices of bread). Such skip connections, inspired by biology, are the characteristic of ResNets. The skip connections allowed ResNets to be deeper than its predecessors (have a tall stack), and helped them to recognize more complex images.
Why is it good for a network to be deep? And why do skip connections help? I find it a stretch to explain them with the analogy of a sandwich, so I defer the answers to later posts. Let’s say for now that the ResNet has skip connections repeated every 2-3 layers, so it’s easy to make it deeper by stacking the same structure multiple times. That makes it look like the stacked sandwiches (see Figure 3 of the original paper).
Well, this was my first attempt at a neuroscience/AI-inspired illustration. The analogy leaves a lot to be desired, but hopefully it will get better as I try more. At least this fulfills my new year’s resolution—to post about neuroscience—on the new year’s day! 🙂
신경망 속 파고들기.
그림에서는 높이 쌓은 샌드위치와 레스넷의 이미지를 합했습니다. 주인공은 “건너뛰는 연결”(파란 전선)을 신경망 층 (빵) 사이에 추가하고 있습니다. 건너뛰는 연결은 실제 뇌에서 영감을 얻은 구조로, 레스넷의 깊이를 더하는 데 도움을 주었습니다. (층을 높이 쌓을 수 있게 해주었습니다.)
왜 신경망이 깊으면 좋은지, 왜 건너뛰는 연결이 도움이 되는지, 샌드위치의 비유로 설명하기에는 한계가 있어서, 대답은 이후 꼭지들로 미루도록 하겠습니다. 여기서는 레스넷이 건너뛰는 연결을 2-3층마다 반복해서 가지고 있어서, 샌드위치를 쌓듯이 같은 구조를 반복만 하면 쉽게 깊어지도록 할 수 있다는 점만 언급하겠습니다. (원래 논문의 그림 3을 참고하세요.)
뇌과학/AI에 관한 첫 꼭지였습니다. 비유가 아직 어색하지만, 더 시도하면서 개선해 보겠습니다. 적어도 뇌과학에 관해 써 보겠다는 제 새해 다짐을 새해 첫날에 실천한다는 의미는 있겠네요. 🙂