Unlabeled text is natural-language data without human-provided categories, entities, or target values. It may be uncategorized, unstructured, and abundant, making it useful for learning without manual annotation. section 6.3 section 1.12
Word2Vec demonstrates the practical use: a model processes a large corpus and learns word meanings from surrounding words, without labels stating that Marie Curie is a scientist or that soccer is a sport. This contrasts with supervised learning, where people provide correct target values. section 6.3
Unlabeled text supports several NLP capabilities:
- Learning relationships among words from their neighboring contexts. section 6.3
- Building knowledge about the world without organizing or labeling every document by hand. section 6.3 section 1.12
- Supporting NLP systems that handle ambiguity, correct mistakes, or retrieve relevant responses. section 1.12
