In L2 acquisition studies, segmental errors found in L2 speech are often investigated, which can result in word confusion because the segmental errors sometime make two distinct words phonetically closer. In a previous work, word productions of native and non-native speakers were plotted in a word embedding space, and lexical density was estimated separately for each speaker group. In this work, a similar analysis is conducted for native and non-native syllable productions, where American English (AE) and Japanese English (JE) are compared. In Japanese, the number of vowels and consonants is smaller, and the syllable structure is by far simpler. Taking these facts into account, density and entropy of spoken syllables of AE and those of JE are estimated and compared based on word embedding techniques. Experiments show a good potential of syllable density analysis because the syllable density of a speaker has a reasonable correlation with his/her pronunciation proficiency.