DBpedia is a community-driven project that extracts structured knowledge from Wikipedia via language-specific chapters. We present the first steps toward the Amharic DBpedia chapter by extending the DBpedia Extraction Framework (DEF) to support Amharic Wikipedia, including language-specific components such as Ethiopian date parsers, an Ethiopian–Gregorian calendar converter, an Arabic–Ge’ez number converter, and Amharic template mappings, together with automated extraction pipelines and the publication of the resulting knowledge graph through a live website, DBpedia Databus collection, and query endpoints. For mapping, we evaluate the zero-shot NLLB-200 translation model on Amharic infobox property names, achieving a BLEU score of 45.31. For ontology alignment, we link mapped properties to DBpedia ontology properties across 58 DBpedia classes and benchmark multilingual encoders with Amharic support, including Afro-XLM-R Base, XLM-R Base, and Amharic fine-tuned mBERT. The fine-tuned Afro-XLM-R model achieves 92.1% Top-10 accuracy and strong ranking performance, as measured by Mean Reciprocal Rank (MRR). We release all resources developed for the Amharic DBpedia chapter, including the Ethiopian date parser, Ethiopian–Gregorian calendar converter, Arabic–Geʽez numeral converter, Amharic template mappings, automated extraction workflows, and the resulting Amharic DBpedia knowledge graph with public access via the DBpedia Databus collection, Tentris query endpoint, and the live website at am.dbpedia.org.

Entity Linking involves detecting and linking entity mentions in natural language texts to a knowledge graph. Traditional methods use a two-step process with separate models for entity recognition and disambiguation, which can be computationally intensive and less effective. We propose a fine-tuned model that jointly integrates entity recognition and disambiguation in a unified framework. Furthermore, our approach leverages large language models to enrich the context of entity mentions, yielding better disambiguation. We evaluated our approach on benchmark datasets and compared with several baselines. The evaluation results show that our approach achieves state-of-the-art performance on out-of-domain datasets.

This paper presents LOLA, a massively multilingual large language model trained on more than 160 languages using a sparse Mixture-of-Experts Transformer architecture. Our architectural and implementation choices address the challenge of harnessing linguistic diversity while maintaining efficiency and avoiding the common pitfalls of multilinguality. Our analysis of the evaluation results shows competitive performance in natural language generation and understanding tasks. Additionally, we demonstrate how the learned expert-routing mechanism exploits implicit phylogenetic linguistic patterns to potentially alleviate the curse of multilinguality. We provide an in-depth look at the training process, an analysis of the datasets, and a balanced exploration of the model’s strengths and limitations. As an open-source model, LOLA promotes reproducibility and serves as a robust foundation for future research. Our findings enable the development of compute-efficient multilingual models with strong, scalable performance across languages.

Multilingual large language models (mLLMs) have significantly advanced machine translation, yet challenges remain for low-resource languages like Amharic. This study evaluates the performance of state-of-the-art mLLMs, specifically NLLB-200 (NLLB3.3, NLLB1.3 Distilled1.3, NLB600) and M2M (M2M1.2B, M2M418), in English-Amharic bidirectional translation using the Lesan AI dataset. We employed both automatic and human evaluation methods to analyze translation errors. Automatic evaluation used BLEU, METEOR, chrF, and TER metrics, while human evaluation assessed translation quality at both word and sentence levels. Sentence-level accuracy was rated by annotators on a scale from 0 to 5, and word-level quality was evaluated using Multidimensional Quality Metrics. Our findings indicate that the NLLB3.3B model consistently outperformed other mLLMs across all evaluation methods. Common error included mistranslation, omission, untranslated segments, and additions, with mistranslation being particularly common. Punctuation and spelling errors were rare in our experiment.