AI Subtitling: Will AI replace subtitle writers?

AI字幕:AI会取代字幕作者吗?

2024-04-23 01:25 CSOFT

本文共744个字,阅读需8分钟

阅读模式 切换至中文

We’ve previously written about how difficult it is to translate Japanese text using machine translation services – even Godzilla himself couldn’t overcome Google Translate – and we’ll be returning to Japan (or, at least, the collision between Japanese Media and American audiences) for our latest survey of how generative AI is impacting the world of media. Specifically, we’ll be looking at how this technology is challenging the Anime market – the country’s famous animation export, beloved by millions of fans worldwide. Some of our readers may be rolling their eyes, but it cannot be overstated how much of a media juggernaut this media sector is, being valued globally at an eye-watering $32 billion dollars in 2023, and with some projections saying the industry will hit $60 billion by the end of the decade. One of the largest players in this space is Crunchyroll, a streaming service devoted entirely to Anime content. It has 13 million paying subscribers and was acquired by Sony in 2022. Sony soon merged it with one of its major competitors, granting it an even larger market share. But in February, Crunchyroll President Rahul Purini set off a firestorm online by telling The Verge that the anime streaming service was experimenting with using AI language models to generate real-time subtitles for its Japanese-language content. This was, understandably, a tall order – Japanese is notoriously hard for machines to translate, and the dialogue in Anime programs isn’t as simple as one might think, with vast amounts of context being required to translate for a range of audiences across all knowledge-and-experiences properly. This wasn’t strictly a matter of cutting costs. Rather, Purini and his team wanted to combat piracy, as the auto-generated subtitles would make it a larger hassle for illegal downloaders to share copyrighted content. However, the company acknowledged that pirates would eventually use AI for the same purposes. Couple that with the medium’s history in the United States – where eager fans would learn Japanese to create unofficial subtitling tracks in the era after the establishment of the home video market but before the medium became a part of the balanced American media diet — and you have a recipe for a media company potentially embracing new technology with diminishing returns at the expense of its expert translators and subtitlers, possibly enabling pirates to claim that they have better versions of translated content available – and for free. It took a month, but the company walked back his statements, with an executive telling The Verge that “[q]uality is of paramount importance for us, and if we find out the technology is not to the level that it can create the best subtitles we want for our fans, then we will wait until the technology gets there. And if it never gets there, then it never gets there.” With that, fans and linguists breathed a sigh of relief. It’s not impossible to imagine a day in which AI-generated subtitling becomes the norm, and, indeed, it may become a valuable asset to those working in the media translation space rather than a replacement. But anyone who has turned on the auto-generated captions on a YouTube video understands that it is still in its imprecise infancy – videos are error-strewn if the speaker has so much as a hint of an accent, and the hearing-impaired, the very audience whose access to these captions are the reason that they exist in the first place. Smaller content creators may not think they have the resources to craft captions or provide translations for their content properly and will have to depend on this inaccurate technology to give their audience something resembling it. Still, those with dedicated followings and income streams should consider working with a qualified LSP for their captioning and subtitling needs. An LSP like CSOFT guarantees its clients a seamless and understandable end product that meets their customers’ standards, keeping the viewing experience as painless and immersive as possible. We can help you meet your audience’s high expectations (though no one is immune to criticism from a fanbase with the exacting standards that your average Otaku has) and bring in new viewers or readers to enjoy your content. So don’t wait another week to discover what happens on a brand-new episode of Translation Struggles Z – get in touch today and find out how our services can do what AI doesn’t. See you, Space Cowboys.
我们之前已经写过使用机器翻译服务翻译日语文本有多困难——甚至哥斯拉自己也无法克服谷歌翻译——我们将回到日本(或者至少是日本媒体和美国观众之间的碰撞)进行我们关于生成式人工智能如何影响媒体世界的最新调查。具体来说,我们将关注这项技术如何挑战动漫市场——该国著名的动画出口产品,受到全球数百万粉丝的喜爱。我们的一些读者可能会翻白眼,但这个媒体行业是一个多么强大的媒体巨头怎么强调都不为过,2023年全球估值为令人垂涎的320亿美元,一些预测称,到2020年,该行业将达到600亿美元。 这一领域最大的参与者之一是Crunchyroll,这是一家完全致力于动漫内容的流媒体服务公司。它拥有1300万付费用户,于2022年被索尼收购。索尼很快将其与其主要竞争对手之一合并,获得了更大的市场份额。但今年2月,Crunchyroll总裁拉胡尔·普里尼(Rahul Purini)在网上引发了一场轩然大波,他告诉The Verge,这家动漫流媒体服务公司正在尝试使用人工智能语言模型为其日语内容生成实时字幕。可以理解的是,这是一项艰巨的任务——众所周知,日语很难被机器翻译,动漫节目中的对话也不像人们想象的那么简单,需要大量的上下文才能为各种知识和经验的观众正确翻译。 严格来说,这不是削减成本的问题。相反,普里尼和他的团队希望打击盗版,因为自动生成的字幕会让非法下载者分享受版权保护的内容变得更麻烦。然而,该公司承认,盗版者最终会将人工智能用于同样的目的。再加上这种媒体在美国的历史——在家庭视频市场建立之后,但在这种媒体成为平衡的美国媒体饮食的一部分之前,热切的粉丝会学习日语来创作非官方的字幕曲目——你就有了一个媒体公司以牺牲其专业翻译和字幕员为代价,以回报递减的方式拥抱新技术的秘诀,这可能使盗版者能够声称他们有更好的翻译内容版本——而且是免费的。 这花了一个月的时间,但该公司收回了他的声明,一名高管告诉The Verge,“质量对我们来说至关重要,如果我们发现技术没有达到为我们的粉丝制作出我们想要的最佳字幕的水平,那么我们会等到技术出现。如果它永远不会出现,那它就永远不会出现。”说完,粉丝和语言学家都松了一口气。 不难想象有一天人工智能生成的字幕会成为常态,事实上,它可能会成为媒体翻译领域的宝贵资产,而不是替代品。但任何在YouTube视频上打开自动生成字幕的人都明白,它仍处于不精确的初级阶段——如果说话者有一点点口音,视频就会错误百出,而听力受损的观众,正是他们访问这些字幕的原因,才是它们存在的首要原因。较小的内容创作者可能认为他们没有资源来制作标题或为他们的内容提供适当的翻译,将不得不依赖这种不准确的技术来给他们的观众提供类似的东西。尽管如此,那些有忠实追随者和收入来源的人应该考虑与合格的LSP合作,以满足他们的字幕和字幕需求。 像CSOFT这样的LSP保证其客户的最终产品无缝且易于理解,符合客户的标准,尽可能保持无痛苦和身临其境的观看体验。我们可以帮助你满足观众的高期望(尽管没有人能像普通宅男一样免受粉丝群的批评),并带来新的观众或读者来欣赏你的内容。所以,不要再等一周来发现全新一集的《翻译斗争Z》会发生什么——今天就联系我们,了解我们的服务如何做人工智能做不到的事情。再见,太空牛仔。

以上中文文本为机器翻译,存在不同程度偏差和错误,请理解并参考英文原文阅读。

阅读原文