Pragmatics-utilizing distributional learner (PUDL) without deterministic hypothesis space
Abstract
We introduce Pragmatics-Utilizing Distributional Learner (PUDL) to simulate verb transitivity learning in 15-month-old English learners. The model incorporates pragmatic reasoning about question-answer relations in neutral wh-questions. Our proposal outlines a developmental trajectory that features a temporary overregularization stage where learners generalize all verbs into one category, due to difficulty in distinguishing Prepositional Phrases from Noun Phrase objects. The results demonstrate the effectiveness of a pure distributional model enhanced by pragmatic knowledge in addressing learning challenges posed by noisy input.
Keywords: verb transitivity learning, noisy input, pragmatic reasoning, distributional learning, empirical Bayes
How to Cite:
Kim, B. & Kang, J., (2024) “Pragmatics-utilizing distributional learner (PUDL) without deterministic hypothesis space”, Society for Computation in Linguistics 7(1), 86–95. doi: https://doi.org/10.7275/scil.2133
Downloads:
Download PDF