Ancient Sounds: mixing acoustic phonetics, statistics and comparative philology to bring speech back from the past

当凯撒大帝入侵不列颠时，他宣称“veni，vidi，维西”，他是怎么发音的？是不是像小学生的笑话里说的那样，“懦弱、软弱、软弱”？我们怎么知道--我们怎么知道--拉丁语或其他已经消亡的语言是如何发音的？19和20世纪的学术研究发展了重建过去单词发音的方法，并将其扩展到推断任何书面记录存在之前的单词形式。例如，从现代英语的“work”、古英语的“weorc”、古高地德语的“werc”、拉丁语的“orgia”、希腊语的“ergon”和亚美尼亚语的“gorc”等各种不同的词形中，语言学家推断出原始印欧语的词干uerg-，这是一个暗示发音类似于“werg”的公式。但它实际上听起来像什么？在这个项目中，我将应用并进一步开发一些新的软件方法，这些方法是我和我的合作者和项目伙伴，剑桥统计学教授约翰阿斯顿在过去几年中开发的，从现代语言中简单单词的当代音频记录中进行三角测量，以重新生成过去的可听口语单词。为了实现这一目标，我们使用从语音合成中借来的语音信号处理技术来分析现代录音，使用数学函数表示它们，使用计算统计技术推断可能的和可能的祖先形式，最后使用语音合成将推断的祖先形式转换回可听语音。在我们的第一个实验中，我们一直致力于模拟声音的变化，这些变化导致了法语，意大利语，西班牙语和葡萄牙语数字的当前发音，从早期的罗曼语单词最初追溯到拉丁语。在这个项目中，我将把这项工作扩展到一些日耳曼语言（英语，德语方言和荷兰语），以及现代希腊语，试图将音频重建的视野推进到印欧语系更深的过去。这一事业类似于虚拟重建，例如通过纪录片或电影中的恐龙动画进行可视化，或重建古代陶器或其他物品。如果我们只有一些碎片或骨头碎片，那么花瓶或头骨的重建将主要是虚拟重建，例如图纸或模型。音频重建同样是对我们过去建筑的虚拟补充。 在此之前，语言学家已经积累了相当多的和深刻的理解，在各种语言中发生了什么样的声音变化，在某种程度上，何时发生;如何或为什么仍然是猜测的领域。语音进化的声学模型使得有可能更好地理解什么，并投射光如何，如果只是通过揭示可能性。进化模型将受益于更精细和经验准确的语言变化模型;例如，通过发现语言的历史变化是否或如何与生物进化在细节上不同。除了我们具体的技术目标外，这类工作对科学和人文之间通常划分的分界线有更广泛的影响。这项工作跨越了这条线，将科学和计算带入了一个以前属于经典和语言学的工作领域。这种学科边界的重新划分，挑战并重新定义了艺术与科学“两种文化”的旧概念。在这一努力中，每一方都需要另一方：如果没有来自这些科学的定量计算技术，比较语言学家在很大程度上局限于书面记录和符号导向的方法;如果没有语言历史知识，统计言语科学将被迫追溯几个世纪以来的语言学工作，这是不合理的（也许是不成功的）。这项工作将科学的新方法带入文化研究领域，并部署文化知识来推进这些新的科学方法。它打破了“科学”与“文化”之间的壁垒。

When Julius Caesar invaded Britain, declaring "veni, vidi, vici", how did he pronounce it? Was it "weany, weedy, weaky", as the schoolboy joke has it? How do we know - how can we know - how Latin or other dead languages were pronounced? 19th and 20th century scholarship developed methods for reconstructing the pronunciation of words from the past, and extended this to inferring the forms of words from a time before any written records exist. For example, from word-forms as diverse as Modern English "work", Old English "weorc", Old High German "werc", Latin "orgia", Greek "ergon", and Armenian "gorc", philologists infer a Proto-Indo-European stem uerg-, a formula hinting at a pronunciation something like "werg". But what did it actually sound like? In this project, I shall apply and further develop some new software methods developed in the last few years by me and my collaborator and project partner, the Cambridge statistics Professor John Aston, to triangulate backwards from contemporary audio recordings of simple words in modern languages to regenerate audible spoken words from the past. To achieve this we use speech signal processing techniques borrowed from speech synthesis to analyse the modern recordings, represent them using mathematical functions, infer possible and probably ancestral forms using computational statistical techniques, and finally convert the inferred ancestral forms back into audible speech, using speech synthesis. In our first experiments, we have been working on modelling sound changes that led to the current pronunciations of French, Italian, Spanish and Portuguese digits, from earlier Romance words originally going back to Latin. In this project I shall extend this work to some Germanic languages (English, German dialects and Dutch), together with Modern Greek, to try to advance the horizon of audio reconstruction into the deeper past of the Indo-European language family. This enterprise is similar to the virtual reconstruction of e.g. the visualisation through animation of dinosaurs in documentaries or films, or the reconstruction of ancient pottery or other objects. If we only have a few sherds or bone-fragments to go on, the reconstruction of a vase or skull will be largely a virtual reconstruction, e.g. a drawing or model. Audio reconstruction is, equally, a virtual addition to our construction of the past. Previously, philologists have amassed a considerable and deep understanding of what sound changes have taken place in various languages, and to some extent, when; the how or why remain areas of speculation. Acoustic modelling of speech evolution makes possible an improved understanding of what, and casts light on how, if only by unveiling possibilities. And evolutionary modelling will benefit by more refined and empirically accurate models of language change; for instance by discovering whether or how historical change in language differs in detail from biological evolution. In addition to our specific technical aims, this type of work has wider ramifications for the dividing line that is normally drawn between science and humanities. This work storms across that line, bringing science and computation into an area of work that was previously firmly part of Classics and linguistics. Such a redrawing of disciplinary boundaries challenges and redefines old conceptions of the "two cultures" of arts vs. science. In this endeavour, each side needs the other: without the quantitative computational techniques from those sciences, comparative philologists are largely confined to written records and symbol-oriented methods; without linguistic-historical knowledge, statistical speech science would be forced to re-trace centuries of philological work, unncessarily (and perhaps unsuccessfully). This work brings new methods from science into an area of cultural studies, and deploys cultural knowledge to advance those new scientific methods. It demolishes a wall between "Science" and "Culture".