SLOG: A Structural Generalization Benchmark for Semantic Parsing
This work addresses the gap in benchmarks for structural generalization in semantic parsing, highlighting models' limited ability to handle unfamiliar syntactic structures, which is incremental as it builds on existing datasets like COGS.
The authors tackled the problem of evaluating models' structural generalization in semantic parsing by introducing SLOG, a benchmark extending COGS with 17 structural generalization cases, where Transformer models achieved only 40.6% accuracy and a structure-aware parser reached 70.8%, far below near-perfect scores on COGS.
The goal of compositional generalization benchmarks is to evaluate how well models generalize to new complex linguistic expressions. Existing benchmarks often focus on lexical generalization, the interpretation of novel lexical items in syntactic structures familiar from training; structural generalization tasks, where a model needs to interpret syntactic structures that are themselves unfamiliar from training, are often underrepresented, resulting in overly optimistic perceptions of how well models can generalize. We introduce SLOG, a semantic parsing dataset that extends COGS (Kim and Linzen, 2020) with 17 structural generalization cases. In our experiments, the generalization accuracy of Transformer models, including pretrained ones, only reaches 40.6%, while a structure-aware parser only achieves 70.8%. These results are far from the near-perfect accuracy existing models achieve on COGS, demonstrating the role of SLOG in foregrounding the large discrepancy between models' lexical and structural generalization capacities.