Colloquium: Janet Pierrehumbert

Capturing semantic and social factors in morphological derivation.

Janet B. Pierrehumbert (Department of Engineering Science, University of Oxford)

Abstract

In morphology, the factors predicting the productivity of inflectional patterns have been intensively studied. Both type frequency and phonological similarity are known to play important roles. Quantitative models have thus focussed on how these two factors interact. However, all of this work assumes that the syntax dictates the speaker’s choice from amongst a small and clearcut number of inflectional categories.

For derivational morphology, this assumption is not valid. Derivational morphology is not obligatory, and it covers a much greater range of meanings than inflectional morphology, The selection amongst meanings is driven by the speaker’s communicative goals, which are affected both by the meaning they wish to express, and the social context in which they wish to express it. For this reason, modelling the creation and interpretation of novel derived word forms is a far more challenging problem. Some linguists have viewed the problem as completely intractable.

In this talk, I will review some experimental and computational studies that tackle this challenge. I will describe some experimental results indicating that semantic and social factors are continuously active in processing novel forms. Then I will turn to some recent studies on quantifying these factors to predict the creation of new derived forms in Reddit, a social media platform. As I will show, tools from statistical natural language processing allow us to make surprising successful predictions about the production of forms such as “unnicknameable”, “trumpistan”, and “minecraftesque”.