Lameta Tutorial: Entering Session Metadata for ELAR Deposits

The Endangered Language Archive (ELAR) at SOAS, University of London, represents one of the world's most established language documentation repositories, funded by SOAS for decades with public access requirements. Global documentation networks include PARADISEC in Australia and institutions in Japan. BRIN, in collaboration with RMPI and other organizations, established LADIN (Language Documentation in Indonesia) as Indonesia's national repository, launched in 2023 during the first International Conference on Language and Literature Preservation. LADIN provides public access to language documentation from anywhere and anytime.

Managing endangered language archives requires balancing open research principles with community privacy needs, involving complex workflows from field data collection through standardized formatting, curation, and controlled access to ensure long-term preservation while respecting cultural sensitivities and legal requirements across different jurisdictions.

Archiving involves transferring copies of collected materials to a dedicated repository like the Endangered Languages Archive (ELAR), which is a digital archive where materials can be set to openly accessible. The archiving process is very involved and time-consuming, requiring all project metadata to be formatted for the archive and ensuring no errors or repeats in the data.

Endangered language documentation creates primary resources for languages at risk of disappearing. Key debates include determining language counts, identifying disappearing languages, and defining language loss versus shift. Researchers select fluent speakers with minimal loanwords as informants. The process involves recording, transcribing, segmenting, glossing, and translating to create corpora. Three basic elements include the corpus (recorded audio/video), lexicon (dictionary), and grammar. The corpus serves as the foundation for developing lexicon and grammar. ELAN software, developed at the Max Planck Institute, serves as the standard tool using XML format. Orthographies can represent underlying forms (abstract morphological structure) or surface forms (actual spoken sounds). The transcription bottleneck describes the challenge where researchers can record dozens of hours of audio but can only transcribe a fraction. ASR is increasingly explored as a solution to reduce human effort. Extrinsic goals focus on creating transcribed audio for archiving, while intrinsic goals focus on transcription quality. Character Error Rate measures accuracy but is not an extrinsic goal itself.

ELRA develops video-annotated resources for speaker identification in TV and news content, capturing both video and audio information. For meeting analysis, multiple cameras and microphones capture speaker and movement information. Bilingual corpora like English-French enable neural machine translation systems. Sentiment analysis resources involve linguist-annotated tweets with sentiment tags. The LRE Map initiative compiles data from conference papers to document existing resources and identify gaps. The Share Your Language Resources initiative encourages community sharing, including PhD student research. Resources are distributed to 60 non-ELRA members, with 54% available free and 45% distributed outside Europe. ELRA produces resources through international partnerships with local institutions, including Indian partners like CDAC. Partnerships span South America, Mediterranean, North Africa, Middle East, and Southeast Asia. The quality assurance model requires partners to produce specific country data with consistent criteria, validated by evaluation centers. Partners meeting quality standards gain access to all other languages free of charge.

Records can be written in different ways. The most effective way is the process recording. Other ways of recording are narrative recording, role recording, summary recording and abstract. Process and role recording form a continuum. Narrative recording is a beginner in case work should start with narrative recording. In narrative recording, everything that has happened is recorded as it has happened. The attempts are to reproduce the situation. It consists of all the statements, observations and comments of the worker. It is narration of the happenings, a detailed accounts of the events.

Successful field linguistics requires both physical equipment and standardized linguistic tools. Essential equipment includes tape recorders, cameras, video cameras, and laptops for data collection. Two fundamental reference lists guide preliminary fieldwork: the Basic Word List (BWL) containing approximately 200 stable words resistant to change across time and space, serving as a comparative foundation; and the Basic Sentence List with 160 sentences incorporating Indian linguistic typology. For Indian languages specifically, a noun-heavy 300-word list accounts for languages lacking strict noun-verb dichotomies. Five primary elicitation methods exist: observation (monitoring natural language use), interview (direct questioning of consultants/informants), questionnaire (written questions for literate informants), documentary sources (for extinct languages), and audio/video recording of unstructured events. While observation aids grammar development and question formulation, unstructured recording poses challenges due to overlapping conversation and conversation flow, making it suitable only for initial fieldwork stages. The interview method remains the most effective approach for systematic data collection. Field linguistics uniquely trains researchers to acquire first-hand knowledge of unstudied or under-researched languages while providing verification methods for previously documented languages.

A typical dialect fieldwork team consists of three roles: the dialect speaker (informant), the conversation leader, and the recording leader. The conversation leader should know the local dialect well and be comfortable speaking it, while the recording leader handles technical equipment. The goal is collecting spontaneous, natural language use to reveal authentic linguistic phenomena. To avoid artificial speech patterns, recordings should occur in familiar environments. Techniques include informal pre-recordings without equipment to build rapport, and meeting informants beforehand. For endangered languages, high-quality recordings are crucial as they may represent the last opportunity to document the language.

This section introduces the fundamental approach to linguistic fieldwork with unfamiliar languages. Key principles include: starting with simple data collection (asking for personal names to establish baseline vocabulary), using standardized presentation formats (sentence + gloss + translation), gathering minimal pairs to identify morphemes, generating and testing hypotheses about underlying rules, and recognizing that native speakers may not provide accurate theoretical explanations. The instructor emphasizes maintaining scientific objectivity, being prepared for unexpected responses, and understanding that successful fieldwork requires both systematic planning and flexibility.

Fieldwork methodology involves systematic data collection from native speakers to document language features. The video explains the procedures used in linguistic fieldwork, including recording, transcription, and analysis of spoken language data.

Image file formats include: BMP (bitmap, uncompressed, large files), GIF (supports animation, limited colors), JPEG (lossy compression, good for photos), PNG (lossless compression, supports transparency), and TIFF (high quality, large files). Audio file formats include: MP3 (compressed audio, widely used), MP4 (audio and video), MIDI (musical instrument digital interface), and WAV (uncompressed audio). Text file formats include: TXT (plain text), CSV (comma-separated values), and RTF (rich text format). PDF can contain both text and images, making it versatile. Understanding these formats helps in selecting appropriate formats for different purposes.

WAV is an audio file format standard that specifies how audio data should be encoded in a file. It uses the IFF (Interchange File Format) as a sub-branch. The standard defines exactly how to encode data so that any compliant player can read and play the file. A WAV file contains chunks: an ID chunk (RIFF) at the top, a chunk size field, a format chunk with encoding information (sample rate, bitrate, number of channels, compression format), and a data chunk containing the actual audio. PCM (Pulse Code Modulation) is an uncompressed audio format where each sample uses 2 bytes for 16-bit stereo.

Audio file formats include WAV (uncompressed, high quality, large files), MP3 (compressed, 10x smaller than WAV, acceptable quality), OGG (open-source, good compression), and WMA (Windows Media Audio, proprietary). Video file formats include AVI (older format supporting audio and video), MPEG (compressed for streaming), and MP4 (most efficient for online streaming). Video files are container formats that combine audio and video streams in a single file. Data compression reduces file size by removing redundant information, similar to packing items tightly in a suitcase. The main benefits are reduced storage space and faster download times. However, compression typically results in some quality loss as information is sacrificed for smaller file sizes. The goal is to compress files as much as possible while maintaining quality acceptable to human perception.

Digital file formats are standardized methods for encoding and storing data in digital form. Text formats include TXT (plain text), DOC (Microsoft Word), and PDF (Portable Document Format). Image formats include JPEG (lossy compression for photographs), PNG (lossless compression with transparency), and GIF (supports simple animations). Audio formats include MP3 (compressed for music), WAV (uncompressed high quality), and Real Audio (streaming). Video formats include MP4 (versatile for streaming and mobile), AVI (widely compatible), and MOV (Apple-developed). MPEG formats (MPEG-2, MPEG-4) are international standards for video compression, ensuring compatibility across devices and platforms for applications from television broadcasting to internet streaming.

Video formats include AVI (Windows standard, can contain uncompressed video), MPEG-2 (early digital TV and DVD standard), QuickTime/MOV (Apple format), and MP4 (current mainstream container supporting H.264/H.265 codecs). Audio formats include WAV (Windows standard, supports various configurations), AIFF (Mac equivalent), and MP3 (MPEG-1 Audio Layer 3, lossy compression reducing size to ~1/10). The video clarifies that MP3 is not MPEG-3, and WAV files have a 2GB size limit due to header constraints.