Approaching word length distribution via level spectra

Treating a text, after the removal of paragraphs and punctuations, as a spectrum of blanks, the distributions of the length of words of ten languages are analyzed. Using models from the statistical theory of spectra, it is found that the ten languages can be classified into two families: one with words that follow a Wigner-like distribution while the words of the other obey a Poisson-like distribution.