You are about to hear the sounds of the largest-toothed predator on the planet: an animal bigger than a school bus with perhaps the most sophisticated form of communication that has ever existed.
你將要聽到的聲音,是來自 地球上牙齒最大的捕食性動物: 這種動物比校車還大, 其溝通方式可能是 有史以來最複雜的。
(Video: whale clicking)
(影片:鯨魚發出卡嗒聲)
These are the sounds of the mighty sperm whale, a fellow mammal that can dive almost a mile, hold its breath for more than an hour and lives in these amazingly complex, matriarchal societies. These clicks you heard, called codas, are just a facet of what we know of their communication. We know these animals are communicating, we just don't yet know what they're saying.
這些是巨型抹香鯨發出的聲音, 這種哺乳類夥伴可以下潛近一哩深, 閉氣超過一小時, 生活在這種極複雜的母系社會中。 你聽到的這些卡嗒聲 叫做「尾聲音節」, 只是我們所了解的抹香鯨 溝通的其中一個面向。 我們知道這些動物正在溝通, 我們只是還不知道牠們在說什麼。 CETI 計畫的目標 就是解答這個問題。
Project CETI aims to find out. Over the next five years, our team of AI specialists, roboticists, linguists and marine biologists aim to use the most cutting-edge technologies to make contact with another species, and hopefully communicate back. We believe that by listening deeply to nature, we can change our perspective of ourselves and reshape our relationship with all life on this planet.
在接下來的五年間, 我們的團隊,包括 人工智慧專家、機器人專家、 語言學家,及海洋生物學家, 打算要用最尖端的技術 與其他的物種接觸, 希望牠們會回應溝通。 我們相信藉由深刻傾聽大自然, 我們就能改變我們對自己的看法, 並重新塑造 我們與地球上所有生命的關係。
This of course seems like an impossible goal. People have been trying to make contact with other animals for hundreds of years. How could we do what others could not, especially given that I'm sitting here on my couch in New York City in the middle of a pandemic and protests?
當然,這似乎是個 不可能達成的目標。 數百年來,人類一直 在嘗試和動物溝通。 我們要怎麼做到別人做不到的事? 尤其是,我還在紐約市裡, 坐在我家的沙發上, 現在還正值疫情和抗議? 以海洋生物學家和海洋學家的 身分,我在過去二十年間,
I've spent the last 20 years as a marine biologist and oceanographer, studying the ocean from all different perspectives, from microbes to sharks. I've assembled interdisciplinary teams that have built the first shark-eye camera to see the world from a shark's perspective, and have collaborated with engineers to design robots so gentle that they don't even stress a jellyfish. But it wasn't until 2018 when I was on fellowship at the Radcliffe Institute for Advanced Study that I realized that perhaps the best way to understand the ocean and its inhabitants wasn't just by seeing the world through their eyes, but by listening -- by really, deeply listening.
從各種不同的角度研究海洋, 從微生物到鯊魚。 我組織了跨領域團隊, 打造了第一個鯊魚眼攝影機, 讓我們從鯊魚的視角看世界, 我也與工程師合作, 設計非常溫柔的機器人, 溫柔到甚至不會驚動水母。 但直到 2018 年, 我在拉德克利夫研究所 擔任研究員時, 我才了解,也許,了解海洋 及其居民的最佳方式, 不是僅透過牠們的視角看世界, 而是要傾聽—— 真正的、深刻的傾聽。
I became interested in sperm whales when I heard their sounds. They sounded like they were coming from another universe; a siren song being broadcast from the darkest reaches of the sea. These weren't the typical harmonious whale songs that I had been accustomed to. These sounded more like digital data transfer. We assembled the future Project CETI team and began discussing how to use the most advanced technologies to communicate with whales. One of the principal conclusions was that machine learning had a really good chance of understanding the patterns of sperm whale communication. And the time to apply these technologies was now. Cracking the interspecies communication code didn't just seem possible, it almost seemed inevitable. But how can analyzing patterns help us converse with whales and other animals?
我聽見抹香鯨的聲音後, 便對牠們十分感興趣。 牠們聽起來就像是來自另一個宇宙; 從海洋最黑暗的深處 傳出來的美妙歌曲。 這些聲音並不是我熟悉的 典型和諧鯨魚歌聲。 它們聽起來更像數位資料傳輸。 我們為未來的計畫 CETI 組織了一個團隊, 開始討論要如何運用 最先進的科技與鯨魚溝通。 主要的結論之一是: 機器學習很有機會 了解抹香鯨溝通的模式。 而現在該是使用這些技術的時候了。 破解物種之間的溝通密碼 不僅是有可能的, 似乎還是無可避免的。 但,分析模式怎麼能協助我們 與鯨魚或其他動物交談? 這個嘛,第一步是要了解 抹香鯨溝通的元素。
Well, step one is to understand the elements of sperm whale communication. These codas you heard don't appear to be sentences as we know them, but there's clear structure in how these animals communicate. Sperm whales send codas back and forth to each other in sequences, and there are regional dialects like British and Australian accents. This is exactly why machine learning is such a powerful tool. These approaches analyze patterns in relationship and map meaning to them. Just a few years ago, scientists used machine learning to translate between two totally unknown human languages. Not by using a Rosetta Stone or a dictionary, but by mapping them on patterns in higher-dimensional space. But for machine learning to work effectively, it needs data -- it needs lots and lots of data.
你聽到這些的尾聲音節 似乎不是我們所知的句子, 但這些動物溝通的方式 有著清楚的架構。 抹香鯨之間會來回傳送 一連串尾聲音節, 牠們也有區域性的方言, 就像英國和澳洲口音。 這正是為什麼機器學習 是如此強大的工具。 這些方法可以用關聯性來分析模式, 並找到對應的意義。 幾年前,科學家用機械學習 將兩種完全未知的 人類語言互相翻譯。 不是用羅塞塔語言學習軟體或字典, 而是把它們投射到 更高維度空間中的模式上。 但若要讓機械學習能發揮效果, 就需要資料—— 需要很多很多資料。
In the past half-century, marine researchers have painstakingly collected and hand annotated just a few thousand sperm whale vocalizations, but in order to learn sperm whale communication, we'll need to collect millions, if not tens of millions of carefully annotated sperm whale vocalizations correlated with behaviors. We'll do it with noninvasive, autonomous, free-swimming robots, aerial-aquatic drones, bottom-mounted hydrophone arrays and more.
在前五十年間, 海洋研究者煞費苦心地收集了 幾千筆抹香鯨的發聲資料 並手動加上註記, 但,為了學習抹香鯨的溝通, 我們需要收集到 數百萬筆,甚至數千萬筆, 加上仔細註記的抹香鯨發聲資料, 搭配和行為的關聯。 我們用非侵略性、自動化、 自由游動的機器人、 空中-水底無人機、 底部水下麥克風陣列 等工具來做這項工作。 我們將會與多明尼克抹香鯨 計畫的密切夥伴合作,
We'll work with our close partners at the Dominica Sperm Whale Project to cover a 20-square-kilometer area that is frequented by over 25 well-known families of sperm whales. We're going to put specific focus on the interactions of mothers and calfs, training our machine learning algorithms to learn whale language from the bottom up. All this data we'll have sent through a pipeline and analyzed by the Project CETI translation team. The raw audio and context data will go through our machine learning engine where it's going to be first sorted by structure. The linguistics team will then search for things like syntax and time displacement. For example, if we find an event where a whale was talking about something yesterday, that alone would be a major finding, something that has thus far only been shown in humans. And once we've really mastered listening, we're going to try really carefully to talk back even on the most simplistic level.
以涵蓋二十平方公里的區域, 有至少二十五個知名的抹香鯨 家庭經常出入這塊區域。 我們將把焦點明確放在 母鯨和幼鯨的互動上, 訓練我們的機械學習演算法, 由下而上學習鯨魚的語言。 這些資料會透過一條管道線傳輸, 交由 CETI 計畫的翻譯團隊來分析。 原始聲音以及情境資料 則會交給我們的機械學習引擎, 首先會依結構來分類。 語言學團隊接著會試著尋找 如語法及時間移位。 比如, 如果哪一次我們發現鯨魚 在談昨天發生的事, 光這一點就是重大的發現, 因為到目前只有人類可以做到。 一旦我們真正精通了傾聽, 我們會試著非常謹慎地回話, 即使是過度簡單的層級。
Finally, Project CETI will build an open-source platform where we will make our data sets available to the public, encouraging the global community to come along on this journey for understanding. These animals could be the most intelligent beings on this planet. They have a neocortex and spindle cells -- structure that in humans control our higher thoughts, emotions, memory, language and love. And all the platforms that we develop can be cross-applied to other animals: to elephants, birds, primates, dolphins -- essentially any animal.
最後,CETI 計畫將會建造 一個開放原始碼的平台, 我們的把我們的資料集 公開給大眾使用, 鼓勵全球大眾 一同參與這趟「了解」之旅。 這些動物可能是地球上 最有智慧的生物。 牠們有大腦新皮質以及梭狀細胞—— 人類身上的這種結構 是用來控制更高階的思想、 情緒、記憶、語言,及愛。 而我們開發的所有平台 都能交叉應用到其他 動物身上,如大象、鳥類、 靈長類、海豚—— 基本上,任何動物皆可。
In the late 1960s, our team member, Roger Payne, discovered that whales sing.
1960 年代末, 我們的團隊成員羅傑佩恩
(Recording: whale singing)
發現鯨魚會唱歌。 (錄音:鯨魚唱歌。)
A finding that sparked the Save the Whales movement led to the end of large-scale whaling and prevented several whale species from extinction just by showing that whales sing. Imagine if we could understand what they're saying. Now is the time to open this larger dialogue. Now is the time to listen deeply and show these magical animals, and each other, newfound respect.
這項發現促成了「拯救鯨魚」運動, 導致大規模捕鯨的終止, 並預防數個鯨魚物種絕種, 就只靠著展示出鯨魚會唱歌。 想像一下,如果我們 能了解牠們在說什麼。 現在該是開啟 更大規模對話的時候了。 現在該是深刻傾聽的時候了, 讓這些魔法般的動物及彼此見識一下 新發現的尊重。
Thank you.
謝謝。