The aftermath of compounds: Investigating Compounds and their Semantic Representations

A recent study published on arXiv explores the alignment of computational embeddings with human semantic judgments in English compound words. By comparing static word vectors like GloVe and contextualized embeddings such as BERT against human ratings of meaning dominance and semantic transparency, the research sheds light on how well these models capture human language understanding. This investigation is significant as it can enhance natural language processing applications and improve the accuracy of language models.