Uncovering divergent linguistic information in word embeddings with lessons for intrinsic and extrinsic evaluation

Abstract

Following the recent success of word embeddings, it has been argued that there is no such thing as an ideal representation for words, as different models tend to capture divergent and often mutually incompatible aspects like semantics/syntax and similarity/relatedness. In this paper, we show that each embedding model captures more information than directly apparent. A linear transformation that adjusts the similarity order of the model without any external resource can tailor it to achieve better results in those aspects, providing a new perspective on how embeddings encode divergent linguistic information. In addition, we explore the relation between intrinsic and extrinsic evaluation, as the effect of our transformations in downstream tasks is higher for unsupervised systems than for supervised ones.

CoNLL 2018

Cited by in corpus (6)

• Extracting Sentence Embeddings from Pretrained Transformer Models
• How Can BERT Help Lexical Semantics Tasks?
• Raw-to-End Name Entity Recognition in Social Media
• Meta-Embeddings Based On Self-Attention
• Exploring the Suitability of Semantic Spaces as Word Association Models for the Extraction of Semantic Relationships
• Analyzing the Surprising Variability in Word Embedding Stability Across Languages