Back to workshops

Workshop1: Models and methods in morphology


Organizer: James P. Blevins (University of  Cambridge)  –

Topic description:

The past decade has witnessed a substantial expansion of the role played by models in linguistic research and a significant shift in the relation between models and theories. Reflecting the influence of the generative tradition, earlier generations of models tended to serve as ‘implementations’ of antecedently-defined linguistic theories. These implementations were also largely interpreted as characterizing aspects of speaker ‘performance’ rather than underlying linguistic ‘competence’. One consequence was a fundamentally asymmetrical relation between theories and models: theoretical innovations could directly influence the design of models, but any results obtained by models could only indirectly bear on theory construction.

As modelling strategies have increased in scope and sophistication, models have outgrown this limited role and begun to encroach on areas formerly regarded as the domain of linguistic theory. It has become clear that highly successful models of language comprehension, production, acquisition and even language change can be formulated without reference to richly-articulated, domain-specific linguistic theories. Instead, it is theory that plays a circumscribed role in this family of approaches. For the most part, a theory just defines a hypothesis space for a class of models.

The overarching goal of this workshop would be to complement the more theoretical presentations of the general session with an overview of emerging modelling paradigms that address a number of common traditional analytical tasks from a radically different perspective. The workshop would provide a summary of the most influential morphological models, highlighting some of the central features that distinguish these models, including variation in architectures, empirical coverage, training data, and cognitive plausibility. The models covered would include ‘deep learning’ neural networks, ‘wide learning’ and other types of discriminative learning networks, neurally-motivated models of lexical and morphological knowledge, developmental and psycholinguistic models, and statistical diachronic models.