Five of the featured papers deal with the inference of finite-state machines. The papers may further be divided in three groups, depending whether the DTRNN used are trained to predict the next symbol of a word (Cleeremans et al., 1989), to classify a string (word) as belonging or not to a language (Manolios and Fanelli, 1994; Pollack, 1991; Giles et al., 1992), or to translate a string over the input alphabet into a string over the output alphabet (Tino and Sajda, 1995). The papers also present a wide variety of DTRNN architectures as well as of training and extraction schemes.