Embeddings in Natural Language Processing: Theory and Advances in Vector Representations of Meaning

How can a computer represent meaning in a form that machine-learning models can work with? This book examines one influential answer: embeddings, which encode words and other linguistic structures as vectors. Mohammad Taher Pilehvar and Jose Camacho-Collados offer a focused survey of how these representations have developed and expanded across natural language processing.

From word vectors to richer representations

The discussion starts with vector-space models and word embeddings, including familiar approaches such as Word2Vec and GloVe. From there, it widens the lens to consider how embeddings can represent word senses, graphs, sentences, and documents. That progression helps place individual techniques within a larger picture of how language and meaning can be encoded.

Context matters in language models

Contextualized representations, including ELMo and BERT, receive attention alongside the earlier methods. Their inclusion gives readers a way to understand how embedding approaches have moved beyond assigning a single fixed vector to a word and toward representations shaped by context.

A guided survey of the embedding landscape

Rather than concentrating on a single model or application, the book synthesizes a range of techniques and research developments. It combines foundational material for readers building their understanding with a broad view of influential approaches—a useful structure for connecting core concepts to the wider NLP literature.

For readers of NLP and computational linguistics

This volume will speak to students, researchers, and practitioners interested in natural language processing, computational linguistics, semantics, or machine-learning representations. It is especially relevant to readers who want a structured orientation to word, sense, graph, sentence, document, and contextualized embeddings in one publication.