Chaining algorithms and historical adjective extension

Natural language relies on a finite lexicon to express a potentially infinite set of ideas. This tension often results in the innovative reuse of existing words to describe emerging ideas. In this chapter, we take a computational perspective to examine how English adjectives extend their range over time to modify nouns and form previously unattested adjective-noun pairs. We hypothesize that how novel adjective-noun pairings emerge is non-arbitrary and follows a process of chaining, whereby novel noun referents for an adjective link to existing nouns modified by the same adjective that are close in semantic space. We test this proposal by exploring a set of probabilistic models that predict adjective-noun pairs from a historical text corpus (Google Books) that spans the past 150 years. Our findings across three diverse sets of adjectives support a chaining mechanism sensitive to local semantic neighbourhood -- formulated as an exemplar model of categorization similar to the Generalized Context Model. These findings mirror existing work on chaining in the historical growth of grammatical categories. We discuss the limitations and implications of our approach toward a general theory of word meaning extension in natural language.