A computer-implemented letter-based method of encoding a length-significant portion of natural language text to generate a letter-based fingerprint of the text portion, the method including detecting letter-based locations of occurrences of pre-determined single-letter and/or multi-letter pattern(s) within the length-significant portion, the detecting being carried out such that at least some occurrences are detected in a word-boundary independent manner that does not depend on locations of word-word boundaries, for a pattern occurrence letter-position signal which describes letter positions of the occurrences of the patterns within the text portion, computing frequency-dependent absolute or relative magnitudes of signal strength for a plurality of frequencies, the computed magnitudes representing letter-based frequencies of the pattern occurrences within the natural language text portion, and storing the computed signal strength magnitudes at the plurality of frequencies, the generated fingerprint comprising the stored signal strength magnitudes. Related apparatus and methods are also described.