English and Chinese language frequency time series (LFTS) were constructed based on an English and two Chinese novels. Methods of statistical hypothesis testing were adopted to test the nonlinear properties of the LFTS. Results suggest the series exhibited non-normal, auto-correlative, and stationary characteristics. Moreover, we found that LFTS follow the power law distributions, and thereby we investigated the fractal structure, long range correlation, and intermittency, which indicated the self-similarity features of LFTS, and also provided hints that human societies are likely to share some universal properties. , on the other hand, analyzed the Chinese character system, supposing that radicals comprised nodes and that two nodes were linked if they could form a character or part of one. Their results revealed that character networks displayed small-word properties and showed non-Poisson degree distributions. Liu [5] built a Chinese semantic network based on a treebank with semantic role annotation and then investigated its global statistical properties. Liu and Li [6] also explored 15 linguistic complex networks based on the dependency of the syntactic treebanks of 15 languages. Yu et al. [7] described a series of identification experiments and rating experiments on the influences of the distance, spectral shape, and relative amplitude of the first two formants of the phonetic quality of /γ/.In addition to the network point of view, however, time series analysis is also an important method for extracting information from signals related to real world complex systems. By analyzing such signals, we can better understand the underlying properties of complex systems.Thus, time series analysis methods have also been used to investigate written human language texts [8][9][10][11]. Currently, there are two ways to map a text into a time series. One counts the number of letters of each word, namely word length l, while time t refers to the position of the word in the document, i.e. the first word is considered to appear at time t=1, the second at time t=2, etc. By mapping word length to time in this way, length time series l(t) are constructed. The second way calculates the probability of appearances of each word in a text, namely the frequency f, while time t refers, again, to the position of the word; thus frequency time series f(t) are constructed.