Incrustaciones contextualizadas de palabras con ELMO

Abstract

Natural language processing (NLP) is an essential and evolving field within machine learning, with applications such as machine translation, chatbots, sentiment analysis and plagiarism detection. Machine learning models for NLP seek efficient representations of words using different encodings, most notably word embeddings, which provide a simplified vector representation. However, these traditional models often omit the context of words. In this sense, ELMo (Embeddings from Language Models), a model that considers the context to generate dynamic vector representations, has emerged. ELMo employs a bidirectional language model (biLM), based on neural networks such as CNN , LSTM , and High-Way Network , allowing to capture context and solve polysemy problems. Introduced in 2018 by researchers at the Allen NLP Institute and the University of Washington, ELMo represents a significant advance in the field.