ABSTRACT
Machine Translation (MT) has seen fluctuating periods of growth and attention since the 1950s, with the primary leading paradigms shifting from rule-based (RBMT) to statistical (SMT), and now neural (NMT). This chapter aims to analyse the seeds of these paradigms through a 1949 memorandum by Warren Weaver that presents five hypothetical approaches to MT. They are: word-for-word translation; disambiguation using micro context (i.e. co-text); an approach based on formal logic; a cryptography approach to translation as decoding; and translation based on invariants (universal symbolic representation of meaning). This chapter shows how some developments external to the translation industry have enabled the realization of Weaver’s vision and discusses various connections among the seeds, the enabling developments and the three machine-translation paradigms. The chapter concludes with a discussion of possible directions for the future of MT and its connection with human translation.
