Statistical Analysis of Chinese Language and Language Modeling Based on Huge Text Corpora

  • Authors:

  • Hong Zhang;Bo Xu;Taiyi Huang

  • Affiliations:

  • -;-;-

  • Venue:
  • ICMI '00 Proceedings of the Third International Conference on Advances in Multimodal Interfaces
  • Year:
  • 2000

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper presents the statistical characteristics of Chinese language based on huge text corpora. From our investigation, we find that in writing Chinese it is more likely to use long words, while in other language styles the words are shorter. In large text corpora, the number of bigram and trigram can be estimated by the size of the corpus. In the recognition experiments, we find the correlation is weak between the perplexity to either the size of the training set or the recognition character error rate. However, in order to attain good performance, the large training set above tens of million words is necessary.