We present EROL, a multilingual neural machine translation system for four English–Indic language pairs: English↔Assamese, English↔Bodo, English↔Manipuri (Bengali script), and English↔Manipuri (Meitei Mayek). Our systems are built on the Transformer-Base architecture with 6 encoder and 6 decoder layers, a 512-dimensional model size, 2048-dimensional feed-forward networks, and 8 attention heads. We employ shared source and target embeddings, GELU activation, layer normalization, dropout regularization, and byte-pair encoding (BPE) tokenization. The models are trained using the standard sequence-to-sequence cross-entropy objective, providing a simple yet effective baseline for multilingual translation in low-resource settings. Experimental results demonstrate competitive performance across all translation directions, achieving BLEU scores of 25.10 for Assamese→English, 23.50 for Bodo→English, 22.47 for Manipuri (Bengali)→English, 24.71 for Manipuri (Meitei Mayek)→English, 15.31 for English→Assamese, 14.25 for English→Bodo, 10.21 for English→Manipuri (Bengali), and 3.96 for English→Manipuri (Meitei Mayek).

The distant supervision (DS) paradigm has been widely used for relation extraction (RE) to alleviate the need for expensive annotations. However, it suffers from noisy labels, which leads to worse performance than models trained on human-annotated data, even when trained using hundreds of times more data. We present a systematic study on the use of natural language inference (NLI) to improve distantly supervised document-level RE. We apply NLI in three scenarios: (i) as a filter for denoising DS labels, (ii) as a filter for model prediction, and (iii) as a standalone RE model. Our results show that NLI filtering consistently improves performance, reducing the performance gap with a model trained on human-annotated data by 2.3 F1.