Aikuma

Aikuma is a free mobile application for recording spoken language and attaching time-aligned respeaking, oral translation and written annotation, built to hand the core tasks of language documentation directly to the speech community that owns the recordings.「Aikuma」是一款免费的移动应用，用于录制口语并附加时间对齐的复说（「respeaking」）、口译与书面注释，其设计初衷是把语言记录的核心任务直接交给拥有这些录音的言语社团

Its name means a meeting place in Usarufa, a language of the Eastern Highlands of Papua New Guinea whose speakers were the first community to test the software in 2012, and the app grew out of the field-linguistics traditions associated with the DoBeS and ELDP documentation programmes.它的名字在「Usarufa」语中意为聚会的场所，这是一种巴布亚新几内亚东部高地语言，其使用者是 2012 年最早试用该软件的人群，而这款应用脱胎于与「DoBeS」和「ELDP」记录计划相联的田野语言学传统

Features

• Three linked tasks define the workflow, namely recording spontaneous speech, respeaking it clearly sentence by sentence, and orally translating it into a language of wider communication, which yields phrase-aligned bilingual audio that a later researcher can work with.三项相互衔接的任务定义了工作流，即录制自然口语、逐句清晰地复说，以及把它口译为更通用的语言，由此产出短语级双语对齐的音频，供后来的研究者使用^[1]
• The interface is deliberately text-free so that monolingual elders with no literacy can operate it, with recording started by pressing the red button or simply by raising the phone to the ear, and commentary produced by pressing and holding play to listen and record to speak.界面刻意不使用文字，不识字的单语长者也能操作，录音可以按红色按钮启动，也可以简单地把手机举到耳边启动，评注则通过按住播放键来听、按住录音键来说完成
• In voice-activated mode playback begins automatically when the phone touches the ear and pauses when it is lifted away, and as soon as the user speaks the app stops playback, records the utterance and aligns it against the source at that point.在语音激活模式下，手机贴上耳朵便自动放音、拿开即停，而用户一开口，应用就停止放音、录下这段话，并在当时位置上把它与源录音对齐
• A set of phones can be synchronised over a local network so that every team member can listen to and annotate recordings made by anyone else, a design which means only a single device has to survive the journey back from the village.一组手机可在本地网络上互相同步，团队的每个人都能收听并注释他人录下的内容，这一设计意味着从村里带回一台设备就够了
• Audio is captured at 16 kHz sampling rate and 16-bit depth in mono, roughly nine hours per gigabyte, and early versions could write only to internal storage because of an Android API limitation the developers hoped would be lifted.音频以 16「kHz」采样率、16 位深度、单声道采集，每「GB」约可录九小时，而早期版本受「Android」接口的一项限制只能写入内部存储，开发者曾希望这一限制日后能取消
• Every recording carries rich metadata, with language identifiers drawn from the ISO 639 registry, named speakers with photographs, GPS coordinates and timestamps, plus tags covering OLAC discourse types, free comments and short consent statements permitting sharing.每条录音都带有丰富的元数据，语言标识取自「ISO 639」注册表，另有带照片的说话人、GPS 坐标与时间戳，标签则覆盖「OLAC」语篇类型、自由评注以及授权分享的简短同意声明
• Literate users can segment a source recording and add written transcription and translation time-aligned to the same spans, and items may be rated by the linguist so that highly rated recordings are displayed more prominently and steer the documentation effort.识字的用户可以切分源录音，并在相同的时间段上添加书写的转写与翻译，条目还可以由语言学家评分，评分高的录音会更醒目地展示，从而引导记录工作的方向
• The app embeds a web server reachable through the phone's WiFi, Bluetooth or USB connection, letting a laptop browser open a transcription tool that shows the waveform together with all spoken annotations as clickable spans beneath it.应用内嵌一个「web」服务器，可通过手机的「WiFi」、蓝牙或「USB」连接访问，笔记本电脑的浏览器由此打开一个转写工具，把波形连同全部语音注释显示为下方可点击的时间段
• On disk the app keeps a transparent directory hierarchy with plain-text metadata, stores transcripts in the NIST HUB-4 format, exports aligned transcriptions in ELAN format for desktop analysis, and accepts imported WAV files for respeaking and translation.落盘时应用使用透明的目录层级与纯文本元数据，转写按「NIST HUB-4」格式存储，对齐的转写可导出为「ELAN」格式供桌面分析，也可以导入「WAV」文件再做复说与翻译
• From 2015 onward sessions could be backed up to a private Google Drive account and synced across devices, while companion cloud-library and cloud-server code added upload, accounts and a shared online library.自 2015 年起，会话内容可备份到私有的「Google Drive」并在多设备间同步，配套的云库与云服务器代码则补充了上传、账号与共享在线书库功能
• Playback interleaves the original recording with its stored commentaries, whose positions appear as speaker images along the slider, so an elder's story and its translations can be heard as one continuous stream.播放时会把源录音与其评注交替播出，评注的位置以说话人头像标在进度条上，长者的故事连同它的多种译文可以当作一条连续的声音流来听
• An import function brings single audio files or whole collections into the app with manually entered metadata, extending the respeaking and translation pipeline to recordings made on any other device.导入功能可以把单个音频文件或成批集合连同手工输入的元数据带进应用，使任何设备上的录音都能进入复说与翻译的流程
• The shortfalls were documented as plainly as the features, since video recording was never finished, no iOS build existed, tablets were not designed for, external-card storage was unavailable, and the project called itself beta software without formal support.短板与功能同样被直白地记录下来，因为视频录制始终没有完成，没有「iOS」版本，未针对平板设计，外接存储卡不可用，项目也自认是「beta」软件而无正式支持

History

• The project grew from Language Preservation 2.0, the effort Mark Liberman pursued at the University of Pennsylvania's Linguistic Data Consortium, and from the Human Language corpus programme that Abney and Bird had proposed in 2010.这一项目源自「Language Preservation 2.0」，即「Mark Liberman」在宾夕法尼亚大学「Linguistic Data Consortium」推进的工作，也上承「Abney」与「Bird」在 2010 年提出的人类语言语料计划
• Fieldwork in the Usarufa village of Imikori from 15 to 18 February 2012 trained ten residents to record storytelling after a communal mumu feast, building on an earlier Basic Oral Language Documentation effort for which Olympus had donated a hundred voice recorders.2012 年 2 月 15 日至 18 日，团队在「Usarufa」地区的「Imikori」村开展田野工作，在集体「mumu」餐宴之后训练十位村民录制故事，此前一项「Basic Oral Language Documentation」工作中「Olympus」曾捐赠一百台语音记录仪
• Development was coordinated by Steven Bird at the University of Melbourne with Florian Hanke as the leading engineer, opened on GitHub in October 2012, and drew in Oliver Adams, Haejoong Lee, Sangyeop Lee, Mat Bettinson, Lauren Gawne and other contributors through the project's credits.开发由墨尔本大学的「Steven Bird」统筹，「Florian Hanke」是核心工程师，代码于 2012 年 10 月在「GitHub」开源，项目名单里还有「Oliver Adams」、「Haejoong Lee」、「Sangyeop Lee」、「Mat Bettinson」、「Lauren Gawne」等贡献者
• The app reached Google Play in March 2013 under the package name org.lp20.aikuma, and the published tutorial was written against the 1.0.0-beta.17 line of builds.该应用于 2013 年 3 月进入「Google Play」，包名为「org.lp20.aikuma」，公开的教程以「1.0.0-beta.17」一系的构建为准
• In December 2013 Aikuma took the grand prize of the Open Source Software World Challenge, a win the Sydney Morning Herald reported under the headline of a digital Rosetta Stone for the world's vanishing languages.2013 年 12 月，「Aikuma」获得「Open Source Software World Challenge」最高奖，《「Sydney Morning Herald」》报道这一胜利时称它是为世界濒危语言打造的数字罗塞塔石碑^[2]
• Funding came from an August 2012 award of 101,500 dollars by the National Science Foundation and the National Endowment for the Humanities to the University of Pennsylvania, followed by a 2014-2017 NSF project with Notre Dame and the International Computer Science Institute, alongside support from the Australian Research Council and the Swiss National Science Foundation.经费方面，美国国家科学基金会与国家人文基金会 2012 年 8 月向宾夕法尼亚大学拨款 101,500 美元，2014 至 2017 年又与「Notre Dame」及国际计算机科学研究院立项，澳大利亚研究理事会与瑞士国家科学基金会亦提供资助
• The system was set out in the 2014 ComputEL workshop paper by Bird, Hanke, Adams and Lee, which reported the field trials and stated the design principles of text-free recording, phone-to-phone synchronisation and a transparent file format.整个系统在 2014 年「ComputEL」工作坊论文中由「Bird」、「Hanke」、「Adams」与「Lee」正式阐述，文中报告了田野试验，并阐明无文字录音、手机间同步与透明文件格式三条设计原则^[1]
• Permission to reuse the code was stated loosely in the earliest period, with project pages naming the MIT licence and later Apache, until a repository commit on 24 September 2015 switched the code to the GNU Affero GPL, the licence whose full text the archived repository still carries.早期代码的复用许可口径不一，项目页面先说「MIT」许可、后称「Apache」，直到 2015 年 9 月 24 日仓库提交明确改用「GNU Affero GPL」，如今归档仓库里仍附有该许可证全文
• Version numbers recorded for the app run from 0.1 up to 0.10.3 across 2013-2016, the final manifest in the repository reads 0.8.4 for the rewritten org.lp20.aikuma2 package, and APK mirrors preserve a 1.0.0-alpha build dated 8 October 2016.该应用记录的版本号在 2013 至 2016 年间从 0.1 一路排到 0.10.3，仓库最后的清单文件对重写的「org.lp20.aikuma2」包写着 0.8.4，而「APK」镜像站保留着一个标注 2016 年 10 月 8 日的「1.0.0-alpha」构建
• During 2015 the team built cloud synchronisation, bulk import and a tagging vocabulary into the development line, distributing dated beta APKs of 2 July 2015 and 30 November 2015 from the website.2015 年间，团队在开发线里加入云同步、批量导入与标签词表，并通过官网分发标注 2015 年 7 月 2 日与 2015 年 11 月 30 日两期的「beta」「APK」
• In 2016 the team declared Android development closed and bet on a web rewrite, shipping an Aikuma-NG Chrome app alpha billed as a speech annotation web app for regular folk, plus a lightweight client, none of which outlived the prototype stage.2016 年团队宣布「Android」端开发停止、转向「web」重写，发布了「Aikuma-NG」的「Chrome」应用「alpha」版，自称为面向普通人的语音注释网页应用，另有轻量版客户端，二者都停留在原型阶段
• Work on the successor produced three small MIT-licensed JavaScript libraries published in 2018, a microphone-capture package called webaudio, web components for language apps and a touch-gesture recorder named gestate, all since archived.为下一代产品衍生出三个 2018 年发布的「MIT」许可「JavaScript」小库，即麦克风采集库「webaudio」、语言应用「web」组件与触摸手势记录器「gestate」，如今均已归档
• From December 2015 the same team staged bilingual storytelling shows in Oakland, and after Bird moved to Australia's Top End in 2016 the Aikuma name passed to a community programme of Language Parties tried in six countries by the 2019 UN International Year of Indigenous Languages.2015 年 12 月起，同一团队在「Oakland」组织双语故事演出，2016 年「Bird」移居澳大利亚北领地后，「Aikuma」之名转入以语言聚会为形式的社区计划，到 2019 年「UN」国际原住民语言年时已试办于六个国家
• The GitHub organisation is now archived, the original Google Play listing has been withdrawn, and aikuma.org today presents the Aikuma Project, coordinated by Steven Bird and Nicole Curtin and aligned with the UN International Decade of Indigenous Languages 2022-2032 rather than with software releases.「GitHub」组织现已归档，最初的「Google Play」条目已经下架，而「aikuma.org」今天展示的是「Aikuma Project」，由「Steven Bird」与「Nicole Curtin」统筹，对标「UN」2022 至 2032 国际原住民语言十年，不再发布软件

Reception

• Press and community attention peaked around the award, when the Sydney Morning Herald account of the grand prize presented the app as a way of recording a digital Rosetta Stone of the world's oral languages.关注在获奖前后达到高点，《「Sydney Morning Herald」》对最高奖的报道把这款应用描述为记录世界口语数字罗塞塔石碑的途径^[2]
• Longer-form coverage in 2013 included Cyberlinguistics and Chronicle of Higher Education features such as Androids in Amazonia and Emerging Writers of the Amazon, which carried the Papua New Guinea and Brazilian fieldwork to wider audiences.2013 年的深度报道见于「Cyberlinguistics」与《 Chronicle of Higher Education》等媒体，其中《亚马逊地区的安卓机》与《亚马逊新一代书写者》把巴布亚新几内亚和巴西的田野工作介绍给更广泛的读者
• Lauren Gawne and Steven Bird kept the app in view with a 2016 public essay on computing tools for preserving disappearing languages, written at a point when active development of the Android client had already stopped.「Lauren Gawne」与「Steven Bird」在 2016 年发表面向公众的计算工具评述，谈保存濒危语言的技术手段，此时「Android」客户端的活跃开发其实已经停止
• The respeaking premise behind the design was tested academically by a University of Melbourne study which found that transcribers without knowledge of the language could produce phonetic transcriptions from respeaked audio that speakers of the language could understand.设计所依据的复说前提经墨尔本大学一项研究检验，结果表明不懂该语言的转写者也能从复说音频做出足以让本族语者听懂的语音转写
• The 2014 workshop paper remains the standard citation for the tool and was carried forward by later surveys of mobile language documentation, including the suite of applications described by Bettinson and Bird in 2017.2014 年工作坊论文至今仍是该工具的标准引用文献，并被后来的移动语言记录综述沿用，包括「Bettinson」与「Bird」在 2017 年描述的一组移动应用^[1]
• Adoption is evidenced chiefly by the documented deployments, since the field trials in Papua New Guinea, Brazil and Nepal gathered roughly ten hours of audio, equivalent to about 100k words, from speakers with no training in linguistics.其采用情况主要见于有据可查的部署，巴布亚新几内亚、巴西与尼泊尔的田野试验采集到约十小时音频、相当于约 100k 词，讲话人并未受过语言学训练^[1]
• In Central Africa the Lig-Aikuma variant built in Grenoble collected more than eighty hours of speech in Congo-Brazzaville for the ANR/DFG BULB project, among the largest corpora gathered with any Aikuma derivative.在中非，由格勒诺布尔方面打造的「Lig-Aikuma」变体在刚果布拉柴维尔为「ANR/DFG」的「BULB」项目采集了超过八十小时语音，是所有「Aikuma」衍生版本中规模较大的语料之一^[3]
• There is no substantial body of independent evaluation, and the project's own FAQ described the software as beta, the result of ongoing research without resources for support, so judgements of usability rest largely on the developers' field reports.独立的公开评测几乎阙如，项目自身的「FAQ」也把软件定位为「beta」，称其为持续研究的产物而无资源提供支持，因此可用性判断主要依赖开发者的田野报告

Community

• Development was open on GitHub under the aikuma organisation, with a public issue tracker, the aikuma-dev mailing list as the main channel between users and developers, and a repository wiki covering file storage, metadata and cloud-server internals.开发在「GitHub」的「aikuma」组织下公开进行，配有公开的问题追踪器，用户与开发者的主要交流渠道是「aikuma-dev」邮件列表，仓库「wiki」则记录文件存储、元数据与云服务器内部机制
• Installation ran through Google Play or through sideloaded APK files posted on the site, including the dated builds of 2015, and the package was never carried in the F-Droid repository.安装途径一是「Google Play」，二是官网发布的侧载「APK」文件，包括 2015 年的两期按日期命名的构建，该包从未进入「F-Droid」仓库
• Building from source required the Android SDK with the 4.2.2 platform, the Google Play Services library, JDK 6, Apache Ant and Gradle, and was driven by a single build script run against the checked-out branch.从源码构建需要含 4.2.2 平台的「Android SDK」、「Google Play Services」库、「JDK 6」、「Apache Ant」与「Gradle」，最后对一个检出的分支执行一个构建脚本即可完成
• Documentation was unusually field-oriented, pairing a long screenshot tutorial with an FAQ that covered informed consent, which phones to buy for remote villages, and synchronisation when no network exists.文档带有浓厚的田野取向，一边是有大量截图的长篇教程，另一边是「FAQ」，涉及知情同意、去偏远村庄该买什么手机、无网络时如何同步等话题
• The Laboratoire d'Informatique de Grenoble at Universite Grenoble Alpes, with Elodie Gauthier and Laurent Besacier leading, forked the app as Lig-Aikuma under the same copyleft licence and maintained releases through 2020, including v3.1.0 of February 2018 which added elicitation from text, images and video.格勒诺布尔-阿尔卑斯大学的格勒诺布尔信息实验室由「Elodie Gauthier」与「Laurent Besacier」牵头，在同样的「copyleft」许可下分叉出「Lig-Aikuma」并持续发布到 2020 年，其中 2018 年 2 月的 v3.1.0 新增了文本、图片与视频诱发模式^[3]
• Researchers built directly on that variant, from a Translators Without Borders app for recording written translations to Python re-implementations used in speech-recognition courses, while the French project also wrote a companion control panel for desktop management.研究者在该变体上继续开发，出现了记录书面翻译的「Translators Without Borders」应用，也有语音课程里的「Python」重实现，法国团队还另写了用于桌面管理的伴生控制台程序
• The repository wiki worked out a workflow exporting Aikuma sessions into SayMore, the SIL International desktop tool, and an independent HTML5 transcription tool was developed under the langtech project to open Aikuma recordings in ordinary browsers.仓库「wiki」设计过把「Aikuma」会话导出进「SayMore」的工作流，「SayMore」是「SIL International」的桌面工具，另有一个「HTML5」转写工具在「langtech」项目下开发，用普通浏览器即可打开「Aikuma」录音
• After the software wound down, the community channel became the Aikuma Project's Language Parties and Treasure Language storytelling events, with a language-learning section on the rebuilt site and a contact address kept open for users of the old app.软件停更之后，社区入口变成「Aikuma Project」的语言聚会与宝藏语言故事演出，重建后的官网设有语言学习板块，并为旧应用的用户保留了一个联系邮箱

Related software

• ELAN, the professional annotator of the Max Planck institute in Nijmegen, is the natural downstream stage, since Aikuma can export time-aligned transcripts in ELAN format so that village collection and laboratory-grade tier analysis fit into a single pipeline.「ELAN」是奈梅亨马克斯·普朗克研究所的专业标注器，也是自然的下游环节，因为「Aikuma」可把时间对齐的转写导出为「ELAN」格式，村庄采集与实验室级别的多层分析因此能接成一条流水线
• SayMore from SIL International shares the metadata-and-archive mission on the desktop, and was studied in the project wiki both as a destination for exported sessions and as a transcription workstation for material gathered in the field.「SIL International」的「SayMore」在桌面上承担同样的元数据与归档使命，项目「wiki」研究过把它用作会话导出的目的地，以及田野材料的转写工作台
• Both this app and its method descend from Basic Oral Language Documentation, the approach associated with Will Reiman and Bird of collecting large quantities of speech with minimal upfront processing, which Aikuma turned into a phone workflow.本应用及其方法都下承基本口语文档法，即与「Will Reiman」和「Bird」相关联、先大量采集口语而尽量少做前置处理的做法，「Aikuma」把这套做法变成了手机上的工作流
• Lingua Libre, run by Wikimedia France, attacks the same recording problem from the wiki side, letting speakers contribute word and phrase audio for endangered languages under free licences rather than assembling archived sessions with aligned translations.由法国维基媒体协会运营的「Lingua Libre」从「wiki」一侧处理同样的录音问题，它让说话人以自由许可贡献词语音频，而不是组装带对齐译文的归档会话
• For grammar and lexicon work after documentation, FieldWorks Language Explorer is the customary partner tool in the SIL ecosystem, where Aikuma deliberately stops at aligned audio and translation and leaves analysis to other software.记录完成后的语法与词典工作，在「SIL」生态中惯用的配套工具是「FieldWorks Language Explorer」，「Aikuma」刻意止步于对齐的音频与译文，把分析交给其他软件
• Among the forks, Lig-Aikuma is the most faithful continuation, adding elicitation, parallel-corpus mapping and smarter speaker metadata for experimental linguistics while keeping the original design of one-handed recording and respeaking.在各分叉中，「Lig-Aikuma」是最忠实的延续，它面向实验语言学增加了诱发任务、平行语料映射与更智能的说话人元数据，同时保留单手录音与复说的原始设计^[3]
• Against desktop recorders and ASR-based collection tools, the case for Aikuma always rested on collaboration without infrastructure, requiring no network at the recording site, no literacy of any participant, and no transcription at all for the field stage.与桌面录音器和基于语音识别的采集工具相比，选择「Aikuma」的理由始终是无需基础设施的协作，录音现场不需要网络，任何参与者不需要识字，田野阶段甚至可以完全不做转写