Abstract
Introduction
LORELEI Bengali Representative Language Pack consists of Bengali monolingual text, Bengali-English parallel text, annotations, supplemental resources and related software tools developed by the Linguistic Data Consortium for the DARPA LORELEI program.
The LORELEI (Low Resource Languages for Emergent Incidents) program was concerned with building human language technology for low resource languages in the context of emergent situations like natural disasters or disease outbreaks. Linguistic resources for LORELEI include Representative Language Packs and Incident Language Packs for over two dozen low resource languages, comprising data, annotations, basic natural language processing tools, lexicons and grammatical resources. Representative languages were selected to provide broad typological coverage, while incident languages were selected to evaluate system performance on a language whose identity was disclosed at the start of the evaluation.
Data
Bengali is spoken mainly in Bangladesh and elsewhere in the Bengal region of South Asia. Data was collected in the following genres: news, social network, and weblogs. Both monolingual text collection and parallel text creation involved a combination of manual and automatic methods.
Data volumes are as follows:
- Over 144 million words of Bengali monolingual text, approximately 358,000 of which were translated into English
- 96,000 Bengali words translated from English data
- 2 million words of found Bengali-English parallel text
Approximately 86,000 words were annotated for named entities, and up to 25,000 words contain additional annotation, including situation frames (identifying entities, needs and issues) and entity linking and detection.
Lexical resources and software tools are also included in this release. The tools recreate original source data from the processed XML material, condition text data users download from Twitter, apply sentence segmentation to raw text, and support named entity tagging.
Monolingual and parallel text are presented in XML with associated dtds. Annotation data is presented as tab delimited files or XML. All text is UTF-8 encoded.
The knowledge base for entity linking annotation for this corpus and all LORELEI Representative Language and Incident Language Packs is available separately as LORELEI Entity Detection and Linking Knowledge Base (LDC2020T10).
(2022-10-17)