N-Gram Statistical Similarities and Differences between Chinese and English
First IEEE International Conference on Semantic Computing, IEEE (2007)
Pei Cao, Stewart Yang, Hongjun Zhu
In this paper, we present results of analyzing quantity and frequency of N-grams in 100 million randomly-sampled English web pages and 100 million randomly-sampled Chinese web papges. We found that 1-gram and 2-gram frequency distributions are very different between Chinese and English; this is understandable since one character in Chinese does not consistitute a word in English. However, we found that 3-gram and 4-grams frequency distributions are surprisingly similar between Chinese and English, leading us to conjecture that in both languages, frequent 3-grams and 4-grams represent a set of concepts that are similar. The distribution of unique numbers of n-grams is quite different between English and Chinese. However, the distribution appears to indicate that, on average, 1.5 Chinese characters corresponds to 1 English word.