Paper

Modeling human-like morphological prediction

---

Abstract

We test a model of morphological prediction based on analogical deduction using phonemic similarity by applying it to German plural suffix prediction for a set of 24 nonce forms for which McCurdy et al. (2020) elicited human judgements, and which they found were poorly matched by productions of an encoder-decoder model of Kirov and Cotterell (2018). Their results raise the question of what kinds of models best mirror human judgements. We show that the predictions of the analogical models we tested mirror human judgements better than the encoder-decoder model.

Keywords: morphological prediction, analogical deduction, nonce forms, neural models, human judgements

How to Cite:

Rosen, E. R., (2022) “Modeling human-like morphological prediction”, Society for Computation in Linguistics 5(1), 133-142. doi: https://doi.org/10.7275/y721-s608