2016/12/01 by Thanapon Noraset, Liang Chen, Noraset, Thanapon +5 · 8 citations
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Natural Language Processing Techniques #Text Readability and Simplification #Topic Modeling
paper · pdf · doi:10.48550/arxiv.1612.00394
openalex publication_date 2016/12/01 · openalex created_date 2022/10/03 · openalex updated_date 2026/07/28
Distributed representations of words have been shown to capture lexical\nsemantics, as demonstrated by their effectiveness in word similarity and\nanalogical relation tasks. But, these tasks only evaluate lexical semantics\nindirectly. In this paper, we study whether it is possible to utilize\ndistributed representations to generate dictionary definitions of words, as a\nmore direct and transparent representation of the embeddings' semantics. We\nintroduce definition modeling, the task of generating a definition for a given\nword and its embedding. We present several definition model architectures based\non recurrent neural networks, and experiment with the models over multiple data\nsets. Our results show that a model that controls dependencies between the word\nbeing defined and the definition words performs significantly better, and that\na character-level convolution layer designed to leverage morphology can\ncomplement word-level embeddings. Finally, an error analysis suggests that the\nerrors made by a definition model may provide insight into the shortcomings of\nword embeddings.\n