# Word Embeddings **Domain:** Machine Learning / Natural Language Processing / Representation Learning **Doc Type:** Canonical Concept Node **Maturity:** Developed **Related:** [[wiki/Distributional Semantics|Distributional Semantics]], [[wiki/Machine Learning|Machine Learning]], [[wiki/Vector Retrieval|Vector Retrieval]], [[wiki/Language Model|Language Model]], [[wiki/Knowledge Representation|Knowledge Representation]] --- ## Definition **Word Embeddings are learned vector representations in which words with related distributional behavior occupy nearby regions of a numerical space.** They compress patterns of context into geometry that supports similarity, analogy and retrieval. Embeddings differ from explicit semantic networks. Proximity can be useful without naming whether the underlying relation is synonymy, topic, association, style or shared context. Their strength is graded generalization; their weakness is limited intrinsic explanation. ## Symbolic Language Engine Bridge The original [[wiki/Symbolic Language Engine|Symbolic Language Engine]] built explicit phonological, taxonomic and trigger-based structures. A rebuilt system could use embeddings to expand candidate neighborhoods and recover relations absent from hand-built resources. Those suggestions should remain distinguishable from corpus-attested lines, WordNet relations and phonological matches. [[wiki/Provenance-Sensitive Multiplicity|Provenance-Sensitive Multiplicity]] provides the labeling discipline needed to combine them. ## Key Insight **An embedding supplies useful nearness without guaranteeing an explanation of the relation.** ## See Also [[wiki/Distributional Semantics|Distributional Semantics]], [[wiki/Vector Retrieval|Vector Retrieval]], [[wiki/Language Model|Language Model]], [[wiki/Neuro-Symbolic AI|Neuro-Symbolic AI]], [[wiki/Semantic Network|Semantic Network]], [[wiki/Provenance-Sensitive Multiplicity|Provenance-Sensitive Multiplicity]] ## Sources / Provenance - Primary project account: [[projects/Ten Years Building a Symbolic Language Engine|Ten Years Building a Symbolic Language Engine]]. - Tomas Mikolov and colleagues, word2vec research (2013). - Jeffrey Pennington, Richard Socher and Christopher Manning, GloVe (2014).