What is the term for a numerical representation that encapsulates the meaning of words or sentences?

The term that best describes a numerical representation that captures the meaning of words or sentences is text embedding. This concept involves transforming textual data into a high-dimensional vector space where similar meanings are represented by vectors that are close to one another. Text embeddings are crucial for various natural language processing tasks, as they allow models to understand and utilize semantic information within the text.

Text embedding techniques, such as Word2Vec, GloVe, and more advanced models like BERT and GPT, generate these vectors by taking into account context and the relationships between words or phrases in a way that reflects their meanings. This enables a more nuanced understanding of language as the embeddings can encapsulate various linguistic features, such as syntactic and semantic relations.

Other terms such as tokenization refer to the process of breaking down text into smaller parts (tokens), while word vectors specifically denote the numerical representations of individual words in vector space but do not imply the same encapsulation of entire sentences that text embeddings provide. Language modeling relates to predicting the probability of sequences of words and does not necessarily focus on the representation of meaning as encapsulated by embeddings.