One of my most joyful moments is waking up early on summer mornings to the chirping and singing of my best neighbors: songbirds. As an amateur piano player, I often feel frustrated that I cannot carry my instrument with me wherever I go. This only deepens my fascination with these winged musical machines, which carry their instruments within their bodies and make up half of the total bird population. Of course, it is far less romantic to realize that their songs are not the analog of an artistic solo in the shower or a soprano recital at Carnegie Hall; instead, they are simply a way of marking territory or a scheme to flirt.
How wonderful would it be if men on the streets of Paris adopted the seduction strategies of songbirds, rather than their usual whistling or out-of-context dirty talk? But unfortunately, we are far from singing like them (sorry, Anna Netrebko). Unlike us, songbirds possess a specialized vocal organ called the syrinx, which can produce multiple tones simultaneously thanks to its two sound sources. Songbirds control this syrinx through the coordination of respiratory muscles, syringeal muscles, and airflow. They also have a dedicated brain circuitry that involves motor areas, and something called the anterior forebrain pathway, which enables them to learn songs and adjust them.
But songbirds are not superior to us in every way. In fact, we share with them surprising similarities, particularly in language acquisition through imitation. In our first year of life, humans are natural producers of drool and snot, and we are crying and babbling professionals. Little by little, we start imitating the sounds in our surroundings like birds do, and after that we start uttering our first words. There is nothing banal about this imitation during language acquisition. In fact, even mammals very close in evolution to humans such as chimpanzees, gorillas, and orangutans, are unable to do this. In turn, other mammals like cetaceans (killer whales, bottlenose dolphins, sperm whales…) can reproduce sounds. But for some reason, researchers have decided to study this ability with songbirds instead of killer whales, as if birds were not dangerous predators.
Anyway, songbirds are an excellent animal model for studying and understanding how humans develop the ability to speak, and they are very cool. However, I can’t tell you much more about it. I’ll leave that to the real expert on the subject: Richard Hahnloser. He’s a full professor at the ETH Zurich, at the Department of Electrical Engineering and Information Technology. He studied theoretical physics and did a PhD in Computational Neurosciences. Then he realized that experiments were still dominating the field of neurosciences, so he got started on that. He was interested in working with songbirds to study their song system to better understand the neural mechanisms underlying song production, the neural codes of a given song, or how a song is learned.
Juan GarcĂa Ruiz: What can we learn from songbirds as humans?
Richard Hahnloser: They have fantastic vocal skills. Even though they have a larynx like we do, they don’t sing with it but instead they use a specialized singing organ called the syrinx, and maybe that’s why they are so fantastically good at imitating sounds. Maybe we cannot learn from them how they produce the sound because we don’t have the same vocal organ they have. But what we can learn from them is what is the evolutionary algorithm that has allowed them to imitate each other, how does this vocal imitation work, and how do they learn by listening to adult singers. Some of these things can be hard to study in humans, because we cannot really experiment with infants. Ethically it is not justifiable to use babies. Some of the answers about what is going on in the first years of life of an infant can be found in birds.
JGR: Is learning a song for a songbird always a matter of imitation? Could a bird start singing if it is completely isolated from other members of the colony?
RH: They just want to sing something, and they just need a kind of model. They want to be taught. Without the template, they will still sing something, but it will sound strange, not stereotyped, and not clean.
JGR: Could you briefly describe the song system of songbirds in a way a child would understand?
RH: The sound system is a set of brain regions with the sole purpose of producing a song. If a bird has an accident and these song areas are damaged, then the bird would be able to do everything a normal bird can do except produce the songs it has learned as a young bird. It’s a very specific system dedicated to singing and nothing else.
JGR: Humans possess both language and the ability to sing, while birds’ productions are limited to singing but, this is commonly associated with bird language. My question is: are bird songs analogous to human language or is it something very different in nature?
RH: People typically compare more birdsong with speech, but there’s way less combinatorial complexity in their songs than there is in language. The syllables they sing do not have so much meaning, so I wouldn’t say it’s comparable to human language. They mostly sing to attract females or to defend territory. But you can find some analogies between human language and birdsongs. In fact, what these birds produce until they are adult is the same as a baby does until it’s one year old. First a baby will cry, then it starts bubbling, and later it starts to combine different syllables until they produce the first words after they are one year old. This process of crying and bubbling until the first words arrive is very similar to the birds’ process of learning.
JGR: What is the natural language processing and how is it related to your research?
RH: Natural language processing is about how a computer process text. We study how birds learn songs, right? The idea is that we teach them two songs. We first give them one song until they can sing it and then we give them the second song that we reproduce through a loudspeaker. Then the question was: when a bird can already sing one song, how will it change this song to go towards the second song? And we figured out this mathematical way that they use to achieve that, and we put it in a simple algorithm. After that we found out that computational linguists, people who work with text, use the same formula that birds use. Birds came up with this millions of years ago while computational linguists did it only a few years ago, and that’s why I got interested in the analogy between biology and text processing.
JGR: What are the main discoveries made in your team?
RH: People in neuroscience use rewards or punishment. A bird learns a song without being externally rewarded for that, it’s like a little baby. We were the first group to study the way birds reward themselves, or what we call intrinsic reward. We have found that in the end what the bird wants is to sing a song, which is a sequence, but what rewards them is the vocabulary. For instance, take the sentence “hi, how are you?”. What I want to say is the sequence: “hi, how are you?” but the bird will care about the sounds, about the “hi”, the “are”, the “you”, and the “how”. It does not matter how the words come, what matters for them is getting the vocabulary right and then the order of it is a separate problem.
JGR: What are the frontiers of this field?
RH: There are three types of vocal song learning. I only talked to you about one of them, the template-based learning. The second one is the reinforcement learning, when you get instantaneous punishment or reward during the production of sounds. And the third one is the feedback-driven learning, that explains how birds learn their songs by comparing their vocalizations to an auditory feedback that you can manipulate, for instance by switching the pitch up to see how they adjust their productions. I would love to come up with a theory that can explain all three at the same time.
JGR: Do you have a message to share with the readers?
RH: Sometimes the biggest discoveries come from what we would call a stupid question, so do not be afraid of asking them.
