The Illusion of Control: Why Bare Classifier Inversion Silently Fails in Concept-Bottleneck Text Generation
2026-08-24 • Computation and Language
Computation and Language
AI summaryⓘ
The authors study how to control text generation by using a simple concept code that represents multiple attributes. They compare three ways to create this code at test time: reversing a classifier, encoding example texts, and using a learned prior distribution. Their experiments show that reversing the classifier often fails and produces poor codes, while a simple learned prior works better across various model sizes. They confirm these findings using real benchmarks and other evaluation methods.
concept bottlenecktext generationmulti-attribute controlclassifier inversionencodernormalising flowMahalanobis penaltycompositional generalisation
Authors
Qi Bing, Xiaowei Shao
Abstract
Concept-bottleneck controllable generation routes multi-attribute control through a low-dimensional concept code that, at deployment, must be synthesised from a target attribute configuration. We study this problem in concept-bottleneck text generation under multi-axis compositional generalisation, comparing three ways to obtain the inference-time code: classifier inversion against the encoder heads, reference-text encoding, and a post-hoc label-conditioned prior. Since a concept code admits no direct LM-fluency term, regularising inversion must instead constrain the code toward the encoder's training distribution. We therefore test bare inversion and three regularised variants: label-agnostic and label-conditioned Mahalanobis penalties, and a conditional normalising-flow density baseline. Every inversion variant we test underperforms a simple post-hoc prior fitted to per-combination encoder means on the same checkpoints, across three backbone families spanning $124$M to $8$B parameters. The bare form of classifier inversion also silently collapses to chance, traceable to a directly measured off-manifold code. We validate this diagnosis on real-world benchmarks and under external evaluators, enabling fair comparison with published baselines.