This repository contains the official release of the model "BanglaBERT" and associated downstream finetuning code and datasets introduced in the paper titled "BanglaBERT: Language Model Pretraining and Benchmarks for Low-Resource Language Understanding Evaluation in Bangla" accpeted in Findings of the Annual Conference of the North American Chap…

Improving Word Translation via Two-Stage Contrastive Learning (ACL 2022). Keywords: Bilingual Lexicon Induction, Word Translation, Cross-Lingual Word Embeddings.

An observatory of anglicism usage in the Spanish press

A Large-scale Multilingual Benchmark Dataset for Automated Translation of Bangla Regional Dialects to Bangla Language

Official Repository for the paper titled "Meta-Learning for Effective Multi-task and Multilingual Modelling" accepted at EACL 2021

[EMNLP 2022] Discovering Language-neutral Sub-networks in Multilingual Language Models.

This study introduces MultiBanFakeDetect, a novel multimodal dataset for Bangla fake news detection, combining textual and visual information. It features TextFakeNet for text analysis and MultiFusionFake for integrating multimodal data.

Zero-shot and Translation Experiments on XQuAD, MLQA and TyDiQA

This research examines the performance of Large Language Models (GPT-3.5 Turbo and Gemini 1.5 Pro) in Bengali Natural Language Inference, comparing them with state-of-the-art models using the XNLI dataset. It explores zero-shot and few-shot scenarios to evaluate their efficacy in low-resource settings.

mBERT and XLM-R for encodeing of Scandinavian languages

ICEBERT: Interlingual-Clusters Enhanced BERT. A BERT-like model trained on clusters of monolingual subwords.

ZeroShot XNLI

Deployed model which can summarize Lithuanian language text by leveraging Artificial Neural Networks, Transformers, mBERT.

HASOC2021: Subtask 2 a) Codemix Challenge; Contains baselines and hierarchical approach that extracts the relevant context useful for classification of hostile tweets on English-Hindi code-mix data obtained from twitter.

Multilingual hate speech detection for German, Italian and Spanish Social Media Posts #machine learning #classifier

By using the hypothesis of historical linguistics, we found a way to improve the performance of multilingual transformers with limited amount of data

Online gender-based violence limits marginalized voices. Detection in Indic languages is hard due to limited data and linguistic complexity. This work builds better classifiers for improved abuse detection in such settings.

The Multilingual QA System is a Flask-based web app that allows users to ask questions in multiple languages (Tamil, English) and receive accurate answers. Using pretrained transformer models for efficient question answering.

This study presents a novel multimodal fusion technique for disaster identification in Bangla, combining text and image data using the "BanglaCalamityMMD" dataset. Employing DisasterTextNet, DisasterImageNet, and DisasterMultFusionNet, the approach addresses a key gap in Bangla disaster research.

Bengali Misogyny Identification with Deep Learning and LIME.