Empirical Study of Sentence Embeddings for English Sentences Quality Assessment

摘要

Novel deep learning and machine translation techniques have greatly advanced the field of computational linguistics enabling us to find meaningful latent spaces for text analysis. While several embedding techniques exist for words, sentences, and entire documents, the potential applications are still being explored. In this paper we present the impact of top-performing sentence embedding methodologies on the accuracy of a neural model trained to assess the quality of English sentences. We focus our efforts in the methodologies called Language Agnostic SEntence Representation (LASER), Sentence to Vector (S2V), and Universal Sentence Encoder (USE) to observe their ability to capture information related to sentence quality. Our study suggests that these state-of-the-art sentence embeddings are unable to capture sufficient information regarding sentence correctness and quality in the English language.