Some Hidden Markov Model generated surnames


Trained from list of real surnames, then sampled from learned model. 100 hidden states. Letter depends on current hidden state and previous letter.

Knoncerellint, Ffut, Soopen, Affladixeablecought, Exci, Foner, Entse, Tury, Pleducou, Inghare, Mrat, Anerabla, Foren, Jually, Whe, Totif, Shast, Suche, Woughutlan, Weakelart, Ouldre, Butimom, Juatimur, Woulin, Mard, Adengh, Juli, Ely, Las, Thath, Yotuns, Thet, Indedpto, Deaso, Kimed, Tanothe, Hof, Ise

2/6/07: Actually, those are probably little better than Markov modelled. My learning algorithm sucked. Here are some purely Hidden Markov Model names (no Markov component), 64 states:

Hadzs, Hook, Sey, Larler, Frests, Chingh, Shalthaurthe, Kon, Beuh, Gasin, Berl, Mussheder, Jild, Sabm, Lap, Lonsh, Cles, Shah, Gilry, Terninsice, Shicqy, Pauchngupndluntten, Papdurn, Wod, Malqer, Frinmarmon, La, Crer, Mundykthmsterll, Chrzy, Rolbs, Wargalle, Bauman, Ipne, Pefiammaston, Yan, Prang, Rurveszner, Hilsomes, Sak, Mrorness

The hidden states often group similar vowels or similar consonants together. Huzzah!