# Multilingual Named Entity Recognition Using Pretrained Embeddings, Attention Mechanism and NCRF

## TL;DR

This paper presents a multilingual NER approach leveraging pretrained BERT embeddings, attention, and NCRF, achieving effective recognition across Bulgarian, Czech, Polish, and Russian without fine-tuning.

## Contribution

The study introduces a multilingual NER model using BERT embeddings with attention and NCRF, applied without fine-tuning, across multiple languages.

## Key findings

- Effective NER performance on BSNLP dataset
- Model works across four different languages
- No fine-tuning of BERT was necessary

## Abstract

In this paper we tackle multilingual named entity recognition task. We use the BERT Language Model as embeddings with bidirectional recurrent network, attention, and NCRF on the top. We apply multilingual BERT only as embedder without any fine-tuning. We test out model on the dataset of the BSNLP shared task, which consists of texts in Bulgarian, Czech, Polish and Russian languages.