Statistical Techniques for Detecting and Validating Phonesthemes

In the lexicons of many of the world's languages, there seem to exist subword patterns of sound and meaning that cannot easily be analyzed as morphemes. English, for example, has a number of words that start with the consonant cluster gland share a meaning related to light or vision, including glimmer, glisten, glitter, gleam, glow, and glint. Firth (1930) coined the term PHONESTHEME to describe such patterns. 1 In this paper, I adopt the following definition of phonestheme from Bergen (2004): (1) [F]orm-meaning pairings that crucially are better attested in the lexicon of a language than would be predicted, all other things being equal. (2004: 293) Even while proposing over a hundred phonesthemes in English alone, linguists have long struggled with their status in theories of natural language: whether or not they qualify as morphemes, how they are related to sound symbolism, and how to decide if they are real rather than mere coincidences in the lexicon. Researchers including Hutchins (1998) and Bergen (2004) have conducted psycholinguistic experiments intended to demonstrate that phonesthemes have psychological reality. Such experiments hold out the promise of proving that phonesthemes form some part of the mental grammar of language users; however, they rely on the researcher being able to select strong candidate phonesthemes for their experiments. The psycholinguist, in other words, is faced with the necessity of somehow selecting phonesthemes before experiments requiring significant time and resources can be conducted to validate those phonesthemes. Furthermore, although there is a long history of proposed phonesthemes in English, other less-studied languages may not share this accumulated resource. In this paper, I propose and evaluate 1 In fact, he used the spelling phonaestheme, which is also sometimes spelled phonaestheme or phonestheme. The latter spelling is used here throughout except in quotations. 3 three statistical, language-independent methods for evaluating candidate phonesthemes that require only a dictionary of the target language in an electronic format and a computer running the necessary software. 2 Background Researchers studying phonesthemes have, broadly speaking, addressed them in three ways: first, simply proposing particular phonesthemes and their meanings; second, trying in various ways to formalize the theoretical treatment of phonesthemes; and third, attempting to determine if, or to what extent, they are real. Although Firth (1930: 184) coined the term phonestheme, he was not the first to notice these patterns in the English lexicon. Wallis (1653), in the section of his Grammatica Linguae Anglicanae …