Paper

Inflectional paradigms as interacting systems

---

Abstract

In the framework of Gradient Symbolic Computation, (Smolensky and Goldrick, 2016), we present a model that predicts correct forms in complex inflectional paradigms through a single underlying form for a lexeme along with underlying forms for certain morphosyntactic combinations. Output-Output Correspondence constraints (Burzio, 1996; Benua, 1997; Burzio, 1999) capture interdependencies between forms in different paradigm cells. Our model avoids complex sets of rules as well as the need to index lexemes to inflectional classes. Instead, the ways that an exponent can vary across lexemes derive from a lexeme\'s underlying representation, which can contain partially-activated blends of segments. This approach takes advantage of a blurring of the distinction between stems and affixes and evaluates MAX Faithfulness constraints across a whole paradigm rather than separately for each word form. We present a neural-based gradient ascent algorithm for learning weights and activations that correctly predict output forms, by optimizing the Harmony of a whole paradigm.

Keywords: inflectional paradigms, Gradient Symbolic Computation, word-based models

How to Cite:

Rosen, E. R., (2021) “Inflectional paradigms as interacting systems”, Society for Computation in Linguistics 4(1), 136-147. doi: https://doi.org/10.7275/0px7-da20