Filtering Input for Learning Constrained Grammatical Variability: The Case of Spanish Word Order
- Shalinee Maitra (University of California, Los Angeles)
- Laurel Perkins (University of California, Los Angeles)
Abstract
Children learn basic word order from data in which both subjects and objects can appear in variable positions. Spanish learners acquire a word order that deterministically places objects after verbs, and allows variation only in subject position. We present a model for acquiring this type of constrained variability from messy data. Our model expects that (1) its data contain a mixture of signal and noise for canonical word order, and (2) subjects control agreement on verbs. We find that this model can learn to filter noise from its data to identify the canonical word order for Spanish; a model that does not track subject-verb agreement cannot. These results suggest that having expectations about the types of regularities that the data will contain can help learners identify variability that is constrained along certain dimensions.
Keywords: language acquisition, probabilistic grammar, regularization
How to Cite:
Maitra, S. & Perkins, L., (2023) “Filtering Input for Learning Constrained Grammatical Variability: The Case of Spanish Word Order”, Society for Computation in Linguistics 6(1), 108-120. doi: https://doi.org/10.7275/5cc7-xb04
Downloads:
Download PDF