«

人工智能通过监听鲸鱼间的交流,或许有了意外的新发现。

qimuai 发布于 阅读:98 一手编译


人工智能通过监听鲸鱼间的交流,或许有了意外的新发现。

内容来源:https://www.sciencenews.org/article/ai-new-sperm-whale-communication

内容总结:

一项突破性研究宣称通过人工智能技术从抹香鲸的"咔嗒"声中发现新型交流模式,这项发表于《开放心智》期刊的成果在学界引发激烈争论。

科研团队利用生成对抗网络对加勒比海抹香鲸的通信编码进行分析,发现传统认知中的"咔嗒"声实则存在"清脆型"与"低沉型"两种声学变体。项目首席科学家沙恩·杰罗表示,通过消除咔嗒声之间的间隔,人类首次能清晰分辨这两种不同频率的发声模式。

语言学家加斯珀·贝古什推测,这种声学差异可能类似人类元音的辨义功能。但圣安德鲁斯大学海洋生物学家卢克·伦德尔直指该类比"完全无稽",强调"没有任何证据表明鲸鱼对这些模式产生反应"。他警告录音设备产生的声波涟漪可能造成分析误差。

研究团队回应称已在不同实验室的设备中重复验证该模式。尽管存在争议,佛罗里达大西洋大学的丹妮丝·赫辛承认这项研究"为鲸类通信探索提供了全新视角"。布里斯托大学专家斯蒂芬妮·金则提出,声学差异更可能反映鲸鱼兴奋程度而非语言特征。

目前研究团队正持续采集鲸群活动时的声学数据,试图破解这些密码背后可能隐藏的行为意义。这场跨越生物学与人工智能的对话,正在重新定义人类对海洋智慧生命的认知边界。

中文翻译:

人工智能窃听鲸语,新发现或将改写认知
抹香鲸的“咔嗒”声可能隐藏着“铿锵音”,这一发现引发学界争议。

海豚嘶鸣,座头鲸吟唱,抹香鲸则通过一连串独特的“咔嗒声”进行交流。11月12日发表于《开放心智》的研究指出,这些被称为“代码”的敲击声序列中潜藏着未被识别的声学规律,意味着鲸群的交流系统可能比想象中更复杂精密。

然而这项研究正遭遇海洋生物学家的尖锐质疑。反对者认为,所谓规律更可能是录音伪影或警觉状态的副产品,而非类语言信号。

数十年来,生物学家已知代码中敲击次数与间隔时间具有意义,甚至能借此辨识抹香鲸所属族群。以多米尼加加勒比海群为例,它们惯用“咔嗒…咔嗒…咔嗒咔嗒咔嗒”的节奏进行交流。

专注抹香鲸沟通研究的非营利组织“CETI计划”生物学家谢恩·杰罗透露,通过人工智能与语言学分析,团队发现某些情境下这段节奏更接近“铿锵…铿锵…铿锵铿锵铿锵”。该组织语言学家加斯珀·贝古什推测,这种异于人类的声码系统近乎摩斯密码,鲸群或许正以“类似人类元音传意”的方式运用这些声响。

但苏格兰圣安德鲁斯大学的鲸类研究专家卢克·伦德尔直斥元音类比“完全无稽”:“毫无证据表明鲸群对此类模式存在感知。”他指出单个鲸鱼咔嗒声实为连续音簇,录音过程中产生的声波涟漪极易被误判为有效信号。

对此贝古什回应,团队已通过不同实验室设备交叉验证该规律,相关数据尚待发布。研究海豚沟通四十载的丹妮丝·赫尔辛格则警示,“元音”表述易引发过度联想,上世纪因夸大海豚能力导致的学科寒冬犹在眼前。

尽管存在争议,赫尔辛格承认这项采用生成对抗网络的新研究方法“值得深入探索”。研究团队通过消除鲸语录音中的停顿间隙,使人类听觉得以分辨“咔嗒”与“铿锵”变体,再以语言学工具进行解析。

布里斯托大学海洋生物学家斯蒂芬妮·金提出另一种可能:该声学特征或反映鲸群兴奋程度。杰罗对此不否认,但其团队正持续采集鲸群发声时的方位与行为数据,试图揭开这些密码更深层的秘密。

英文来源:

AI eavesdropped on whale chatter. It may have helped find something new
Some sperm whale “clicks” could be “clacks,” but what that could mean is controversial
Dolphins whistle, humpback whales sing and sperm whales click. Now, a new analysis of sperm whale codas — a unique series of clicks — suggests a previously unrecognized acoustic pattern. The finding, reported November 12 in Open Mind, implies that the whales’ clicking communications might be more complex — and meaningful — than previously realized.
But the study faces sharp criticism from marine biologists who argue that these patterns are more likely to be recording artifacts or by-products of alertness rather than language-like signals.
For decades, biologists have known that both the number and timing of clicks in a coda matter and can even identify the clan of a sperm whale (Physeter macrocephalus). Sperm whales in the eastern Caribbean Sea off the coast of Dominica, for example, often use a series of two slow and three quick sounds: “click…click… click-click-click.”
Relying on artificial intelligence and linguistics analysis, the new study finds that sometimes this series sounds more like “clack…clack… clack-clack-clack,” says Shane Gero, a marine biologist at Project CETI, a Dominica-based nonprofit studying sperm whale communication.
Project CETI linguist Gašper Beguš wonders about the meanings a coda might convey. “It sounds really alien,” almost like Morse code, says Beguš, of the University of California, Berkeley. Based on his team’s result, he now speculates that sperm whales might use clicks or clacks “in a similar way as we use our vowels to transmit meaning.”
Not everyone agrees with that assessment.
The comparison to vowels is “completely nonsense,” says Luke Rendell, a marine biologist at the University of St. Andrews in Scotland who has studied sperm whales for more than 30 years. “There’s no evidence that the animals are responding in any way to this [new pattern].”
He notes that each sperm whale click isn’t just one tone but several in a row, and this can introduce ripples into a recording that aren’t present in the original. These ripples can look a lot like the pattern the CETI team found. He thinks the researchers didn’t do enough to rule out the possibility of recording artifacts.
“I was always worried that this is some sort of artifact,” Beguš says. “But we were very careful.” The team found the same pattern in codas recorded by other labs with different equipment, but that work hasn’t been published yet.
Marine biologist Denise Herzing, who has studied dolphin communication for over 40 years, also objects to the word “vowel.” People who read that might jump to the conclusion that the animals are using “something like human language,” says Herzing, of Florida Atlantic University in Boca Raton. Unfounded claims about dolphin abilities in the 1960s and ’70s, she says, killed communication research in her field for a long time.
Still, the new pattern is “well worth exploring,” Herzing says. This study takes “a novel look at sperm whale communication using a technique that hasn’t been used before.”
The CETI team initially used an AI system called a generative adversarial network, or GAN, to look for aspects of sperm whale codas that might carry meaning. Half of this system learned to recognize real sperm whale codas from data. The other half learned to create its own invented codas that could carry information. And it tried to trick the first half into thinking these were real. In the invented codas, manipulating frequency proved to be important.
So Beguš decided to study the frequencies of real codas. To help with this, he removed the spaces between clicks in real whale recordings so they all ran together. This made it possible for human ears to hear differences between the “click” and “clack” types of codas. He studied these sounds using tools that linguists use to study human words.
Techno tunes
An AI and linguistic analysis of sperm whale communication reveals that sometimes the whales clack instead of click, a new study suggests. Each set of clicking sounds, or coda, “has two different versions,” says marine biologist Shane Gero. Gero and his colleagues discovered the difference after removing the spaces between clicks. Here are before and after versions, courtesy of Project CETI.
Herzing says the idea to remove spaces is interesting: “It’s a way for humans to sort of listen differently.” But it’s unknown, she says, whether the technique reveals how whales experience these sounds.
Stephanie King, a marine biologist at the University of Bristol in England, is also skeptical. She’s not convinced that the pattern CETI found is something the whales notice or produce on purpose. It “might be more likely related to arousal,” she says, as similar patterns across the animal kingdom are often related to how alert or relaxed an animal is.
Project CETI’s Gero agrees that the new pattern might “encode for emotional state.” But he thinks it’s worth exploring other possibilities. His team is currently gathering data on the whales’ locations and activities when they make these and other types of codas.

AI科学News

文章目录


    扫描二维码,在手机上阅读