Reproduce “Learning inflection classes using Adaptive Resonance Theory” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Peter Dekker, Heikki Rasilo, Bart de Boer
A study from Vrije Universiteit Brussel investigated the unsupervised acquisition of verbal inflection classes using Adaptive Resonance Theory 1, a cognitively plausible neural network. It achieved an Adjusted Rand Index of 0.94 for Latin macroclasses, 0.27 for Portuguese microclasses, and 0.4 for Estonian microclasses, demonstrating robust generalization of morphological patterns from surface-level phonetic input.
Key insights
- Inflection classes are learnable from surface-level phonetic information using ART1, a cognitively plausible unsupervised model, supporting theories that humans infer linguistic categories from observed input.
- The interpretability of ART1's learned cluster templates provides direct insight into the specific features (e.g., theme vowels, stem vowels) the model utilizes for categorization, connecting computational learning to linguistic diagnostics.
- The varying performance across languages highlights how the nature of morphological complexity (e.g., consistent affixal patterns vs. distributed stem variation or opaque features) significantly impacts learnability from surface forms.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.