Natural Language Processing Pipeline to Annotate Bulgarian Legislative Documents

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 6988–6994 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 6988 Natural Language Processing Pipeline to Annotate Bulgarian Legislative Data Svetla Koeva, Nikola Obreshkov, Martin Yalamov Institute for Bulgarian Language "Prof. Lyubomir Andreychin" Bulgarian Academy of Sciences 52 Shipchenski prohod Blvd., Bldg. 17, Sofia 1113 {svetla, nikola, martin}@dcl.bas.bg Abstract The paper presents the Bulgarian MARCELL corpus, part of a recently developed multilingual corpus representing the national legislation in seven European countries and the NLP pipeline that turns the web crawled data into structured, linguistically annotated dataset. The Bulgarian data is web crawled, extracted from the original HTML format, filtered by document type, tokenised, sentence split, tagged and lemmatised with a fine-grained version of the Bulgarian Language Processing Chain, dependency parsed with NLP- Cube, annotated with named entities (persons, locations, organisations and others), noun phrases, IATE terms and EuroVoc descriptors. An orchestrator process has been developed to control the NLP pipeline performing an end-to-end data processing and annotation starting from the documents identification and ending in the generation of statistical reports. The Bulgarian MARCELL corpus consists of 25,283 documents (at the beginning of November 2019), which are classified into eleven types. Keywords: NLP pipeline, legislative corpus, Bulgarian language 1. Introduction The paper presents the Bulgarian MARCELL corpus, part of a recently developed multilingual corpus representing the national legislation in seven European countries and the NLP pipeline that turns the web crawled data into structured, linguistically annotated dataset. The presented work is an outcome of the CEF Telecom project Multilingual Resources for CEF.AT in the Legal Domain1 (MARCELL) aiming to enhance the eTranslation system of the European Commission by supplying large-scale domain specific data in seven languages (Bulgarian, Croatian, Hungarian, Polish, Romanian, Slovak and Slovenian). The NLP Pipeline for Bulgarian, developed specifically to answer the requirements of MARCEL project for autonomy and sustainability, is continuously feeding the Bulgarian MARCELL corpus with newly issued legislative documents. Data is extracted from a single web source and further transformed. The transformation phase makes changes to the data format, organises data in structures, accumulates data with linguistic information, analyses data and provides explicit links between different data segments. In particular, the Bulgarian data is web crawled, extracted from the original HTML format, filtered by document type, tokenised, sentence split, tagged and lemmatised with a fine-grained version of the Bulgarian Language Processing Chain2 and dependency parsed with NLP-Cube3. Further, named entities (persons, locations, organisations and other), noun phrases, IATE terms4 and EuroVoc descriptors5 are recognised and annotated within the corpus. The paper is focused on the NLP pipeline for Bulgarian to reveal some of its features: ● Standardisation: CoNLL-U Plus formatted output at any level of annotation. 1 https://marcell-project.eu 2 https://dcl.bas.bg/en/webservices/ 3 https://opensource.adobe.com/NLP-Cube/index.html 4 https://iate.europa.eu/home 5 https://marcell-project.eu/links.html ● Domain-specific adaptation: the NLP pipeline modules are tuned to annotate legal domain documents. ● Multiword coverage: Multiwords are handled at the levels of annotation. ● Self-driving solutions: All modules are synchronised to …

Content truncated to 500 words