The textural language machine: a system of associative electronic networks for efficient processing of neural language texts

A novel digital electronic system is presented, which is able to handle the human language very well. In contrast to conventional computers this can be obtained by the following measures: by concatenation of well-known associative memories with feedback, by the introduction of conditional association operations, and by a strategy for the stepwise abstraction of texts (in the opposite direction by a process of concretization). The resulting structure is a system of networks which matches very well with the process network of human language, which could be found by analyzing large amounts of texts. The efficiency of the language machine can be increased by using an appropriate learning strategy. Experimental results showed that it is possible to master large stocks of words right from the beginning. It can be seen that the process of abstraction yields high data compression values hitherto unthought of. Possible applications are the storage of large amounts of texts, the transmission of texts over great distances, the translation of texts into another language and the encoding and ciphering of texts