Representing Sentences as Low-Rank Subspaces
This provides a generic, unsupervised sentence representation for natural language processing applications, but it is incremental as it builds on existing word vector methods.
The paper tackled the problem of representing sentences as low-rank subspaces based on the observation that word vectors of a sentence lie in a low-rank subspace, and it outperformed neural network models by 15% on average in semantic textual similarity tasks across 19 datasets.
Sentences are important semantic units of natural language. A generic, distributional representation of sentences that can capture the latent semantics is beneficial to multiple downstream applications. We observe a simple geometry of sentences -- the word representations of a given sentence (on average 10.23 words in all SemEval datasets with a standard deviation 4.84) roughly lie in a low-rank subspace (roughly, rank 4). Motivated by this observation, we represent a sentence by the low-rank subspace spanned by its word vectors. Such an unsupervised representation is empirically validated via semantic textual similarity tasks on 19 different datasets, where it outperforms the sophisticated neural network models, including skip-thought vectors, by 15% on average.