English

Segmentation-free Compositional $n$-gram Embedding

Computation and Language 2019-05-30 v2 Machine Learning

Abstract

We propose a new type of representation learning method that models words, phrases and sentences seamlessly. Our method does not depend on word segmentation and any human-annotated resources (e.g., word dictionaries), yet it is very effective for noisy corpora written in unsegmented languages such as Chinese and Japanese. The main idea of our method is to ignore word boundaries completely (i.e., segmentation-free), and construct representations for all character nn-grams in a raw corpus with embeddings of compositional sub-nn-grams. Although the idea is simple, our experiments on various benchmarks and real-world datasets show the efficacy of our proposal.

Keywords

Cite

@article{arxiv.1809.00918,
  title  = {Segmentation-free Compositional $n$-gram Embedding},
  author = {Geewook Kim and Kazuki Fukui and Hidetoshi Shimodaira},
  journal= {arXiv preprint arXiv:1809.00918},
  year   = {2019}
}

Comments

NAACL-HLT 2019

R2 v1 2026-06-23T03:53:35.500Z