Molecules, Meaning and Post-Modernist Semantics

Wilks’ early English/French Machine Translation system was based on a notion called Preference Semantics. There were two key components of Preference Semantics. First was the notion of combining elementary meaning units of some kind (in Wilks’ case effectively surrogates for Roget thesaurus’ categories) in structures of arbitrary complexity and fineness of description. Second, was the notion of meaning selection: in this case choice of translation term; being one of preferential or balanced ranking rather than absolute selection. While Wilks’ system was driven by a dictionary hand crafted in much the manner of conventional lexicographic work, Wilks’ colleagues (and specifically Spärck Jones) were very interested in what would now be called supervised and unsupervised learning of these lexical structures. Such learning is probably needed to build a practical language processing system based on these ideas. The paper looks at these notions of molecular word meaning definitions and their acquisition in terms of modern developments in supervised and unsupervised learning. It will go on to look further at the notion of preference in the light of post-modernist notions of semantics developed by Zuidervaart amongst others, and then look briefly at how one would go about constructing a Wilks-like Machine Translation system using today’s state of knowledge.