Contextual Representation
A word representation generated dynamically by a model according to the current sentence, in which the same word receives different vectors in different contexts, as distinguished from static word embeddings that assign a unique vector to each word.
Timeline (last 90 days)
The model takes a weighted average of the context vectors (via the content projection W^C) to obtain a new context-dependent representation vector hᵢ.
Context-dependent representations are dynamically generated by the model according to the current sentence, so the same word gets different vectors in different contexts.
All Facts (2)
The model takes a weighted average of the context vectors (via the content projection W^C) to obtain a new context-dependent representation vector hᵢ.
50%UnverifiedContext-dependent representations are dynamically generated by the model according to the current sentence, so the same word gets different vectors in different contexts.
50%