Can Constructions "SCAN" Compositionality ?

2025-09-24

AI Summary

Sequence-to-sequence models exhibit significant deficiencies in compositional and systematic generalization—particularly under out-of-distribution conditions. To address this, we propose an unsupervised pseudo-construction mining method that requires no architectural modifications or additional annotations; instead, it automatically extracts variable-slot templates from training data to explicitly model form-meaning pairings and enhance structural recombination capability. Our approach integrates pseudo-constructions into the preprocessing pipeline of the SCAN dataset, thereby improving data efficiency and generalization robustness. Experiments demonstrate state-of-the-art performance on the highly challenging ADD JUMP and AROUND RIGHT splits, achieving accuracies of 47.8% and 20.3%, respectively—surpassing most supervised baselines using only 40% of the training data. This work constitutes the first fully unsupervised, construction-level representation mining framework, establishing a novel paradigm for systematic generalization in low-resource settings.

Technology Category

Machine Learning: Feature Construction/ReformulationNatural Language Processing: GenerationSearch and Optimization: Learning to Search

Application Category

Web Mining and Content Analysis: Large pretrained models with web dataSearch and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for rankingSemantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Sequence to Sequence models struggle at compositionality and systematic generalisation even while they excel at many other tasks. We attribute this limitation to their failure to internalise constructions conventionalised form meaning pairings that license productive recombination. Building on these insights, we introduce an unsupervised procedure for mining pseudo-constructions: variable-slot templates automatically extracted from training data. When applied to the SCAN dataset, our method yields large gains out-of-distribution splits: accuracy rises to 47.8 %on ADD JUMP and to 20.3% on AROUND RIGHT without any architectural changes or additional supervision. The model also attains competitive performance with? 40% of the original training data, demonstrating strong data efAciency. Our findings highlight the promise of construction-aware preprocessing as an alternative to heavy architectural or training-regime interventions.

Problem

• Sequence to Sequence models fail at compositionality and systematic generalization
• Models lack internalization of form-meaning pairings for productive recombination
• Unsupervised method mines pseudo-constructions to improve out-of-distribution performance

Innovation

• Unsupervised mining of pseudo-constructions from training data
• Variable-slot templates extracted automatically without supervision
• Construction-aware preprocessing as alternative to architectural changes