Investigating Semantic Subspaces of Transformer Sentence Embeddings through Linear Structural Probing.
Dmitry NikolaevSebastian PadóPublished in: BlackboxNLP@EMNLP (2023)
Keyphrases
- hilbert space
- natural language
- low dimensional
- linear subspace
- high dimensional data
- natural language sentences
- dependency relations
- finite dimensional
- semantic content
- semantic information
- vector space
- structural information
- recognizing textual entailment
- infinite dimensional
- euclidean space
- grassmann manifold
- semantic roles
- von neumann
- semantic representation
- manifold learning
- semantic analysis
- semantic similarity
- fuzzy logic
- neural network
- semantic parser
- low dimensional spaces
- syntactic analysis
- structural features
- semantic network
- parameter space
- domain specific
- dimensionality reduction
- high level