FORMULEARN

How do we learn to organize a language in chunks and to use those chunks creatively? To address this question, FORMULEARN rethinks formulaicity and creativity through discriminative and integrative learning in a complex dynamic system, shifting the focus from the written to the spoken word.

Theories of chunking are based on abstract rules or on the storage of large numbers of exemplars, always with a view of linguistic knowledge as linear combinations of units, such as phonemes or morphemes. Recently, Baayen and collaborators have proposed that linguistic chunking is based on discriminative learning, which creates statistical expectations within the complex dynamic system of cues and outcomes underlying language, without discrete units being necessary.

The Parry-Lord theory of oral composition-in-performance argued that oral singers produce complex poems out of rehearsed improvisation through the mastery of a system of formulas, chunks that integrate phrasal, metrical, and semantic structures. Ubiquitous throughout the history of the species, oral traditional performance is the original form not only of poetry but also of public discourse in general. Its comparison with general speech and text can shed light on chunk formation, on how chunking is learned, and on how it relates to meaning and creativity.

FORMULEARN reconsiders formulaicity and creativity by contrasting these theories, seeking to design the first quantitative studies of formulaic creativity not based on morphosyntactic patterns, but on sequences of multimodal cues directly linked to semantic contrasts and cognitive states, using statistical and machine-learning techniques such as discriminative learning models and generalized additive models.