In Search of Dolphin Language

该项目将通过利用最先进的计算神经科学和统计物理方法积极解决归因问题等关键技术问题，解决海豚交流研究中的核心问题。该项目将把目前局限于实验台上的定量高通量研究转移到野生动物的实地研究中。这项跨学科的研究是由动物认知心理学家Diana Reiss和物理学家及计算神经科学家Marcelo Magnasco的实验室合作进行的。该项目的一个更广泛的影响是，水族馆研究特别有利于推广工作，因为与海豚有关的活动是在一个开放的圆形剧场环境中进行的，展览已经从海豚表演变成了更多关于动物护理、行为和保护的教育示范。研究活动将与公众参观和学生实地考察同时进行，因此，研究工作将实时向公众通报。实际上，只要付出适度的努力和成本，所有提出的活动都可以在巴尔的摩的国家水族馆举行公开展览，指导来访的学童了解研究过程和跨科学界限的合作研究。对研究地点的实地考察是与亨特学院本科生和研究生实地研究课程共同组织的，是向这些学生展示和参与实际前沿研究的绝佳机会。海豚的认知和沟通能力是传说中的。将这些传说整理成严谨的科学体系是一个缓慢的过程，充满了许多陷阱；过去二十年的严谨工作描绘了一幅比传说更微妙的画面。虽然海豚确实有惊人的认知能力，但它们在野外的自然同域交流仍然是一个谜，关于它们声音曲目的许多元素的性质以及它们整体交流的复杂性，人们一直在争论。关键的技术问题是归因问题的费力性，哪只动物发出哪一种叫声的分配，缺乏自动化的行为分析方法，最后是将归因的叫声与特定行为联系起来。因为我们目前不能自动三角定位和跟踪它们在水下的声音，所以没有现存的海豚发声语料库：所研究的大多是孤立的叫声，而不是“对话”。在过去的十年中，对鸣禽和啮齿动物的研究已经证明了通过自动化手段研究大量动物的高通量设置的力量。该项目的首要目标是将这些高通量设置扩展到圈养和野生海豚的交流研究中，并为这个更为困难的案例开发相关的自动音频和视频分析工具，以处理产生的大量数据。有两个具体的活动：第一，pi将研究学习和自由使用一种合成语言；第二，从多个有利位置记录，然后计算分析野生海豚的自然行为，包括在豆荚中的自然行为，以及在高度限制的特定行为，如弓乘。产生的声音语料库将被计算分析，以确定野生海豚是否确实有足够丰富的“语言”，而不是一套叫声。

This project will address central questions in dolphin communication research by aggressively attacking the key technical issues such as the attribution problem with state-of-the-art computational neuroscience and statistical physics methods. The project will move quantitative high-throughput studies, currently tied to the bench, into field research of wild animals. This interdisciplinary study is collaboration between the labs of animal cognitive psychologist Diana Reiss and physicist and computational neuroscientist Marcelo Magnasco. One of the broader impacts of the project, the aquarium studies are particularly conducive to outreach efforts since the dolphin related activities take place in an open amphitheater environment and the exhibit has changed from dolphin shows to more educational demonstrations about animal care, behavior, and conservation. The research activities will take place concurrently with visits of the public and field trips of schoolchildren and therefore, the research endeavor is being communicated to the public in real time. Virtually all the activities proposed can, with modest effort and cost, be made into an open exhibit at the National Aquarium in Baltimore, instructing visiting schoolchildren on the research process and on collaborative research across scientific boundaries. The field trips to the research sites are organized jointly with undergraduate and graduate courses in field research for Hunter College students and are excellent opportunities to showcase and engage actual cutting edge research to these students.The cognitive and communication capabilities of dolphins are legendary. Untangling the legends into a body of rigorous science has been a slow process fraught with many pitfalls; rigorous work in the last two decades has painted a rather more nuanced picture than the legends. While dolphins do indeed have amazing cognitive abilities, their natural conspecific communication in the wild remains enigmatic, with ongoing debates as to the nature of many elements of their vocal repertoire and the complexity of their communication overall. The key technical problems have been the laboriousness of the attribution problem, the assignment of which animal emitted which vocalization, the lack of automated behavior analysis methods, and finally the linking of attributed vocalizations to specific behaviors. Because we currently cannot automatically triangulate and track their voices underwater there are no extant corpora of dolphin vocalizations: what have been studied are mostly isolated calls, as opposed to "conversations". In the last decade, studies in songbirds and rodents have demonstrated the power of high-throughput setups in which a large number of animals are studied through automated means. The overarching aim of this project is to extend these high-throughput settings to dolphin communication studies, in captivity and in the wild, and develop, for this far more difficult case, the relevant automated audio and video analysis tools to handle the huge amount of data generated. There are two specific activities: first, the PIs will study learning and free use of a synthetic language and second record from multiple vantage points and then computationally analyze natural dolphin behavior in the wild, both during natural behavior in pods as well as during highly circumscribed, specific behaviors such as bowriding. The corpora of vocalizations generated will be analyzed computationally to establish whether wild dolphins do have a rich-enough "language", as opposed to a repertoire of calls.