Understanding how neural models represent human-interpretable concepts is challenging.Prior work has explored linear concept subspaces from diverse perspectives, such as probing and concept erasure.We introduce a unified framework to study these subspaces along two axes: containment, which tests if a concept is fully represented in a subspace but not outside it, and disentanglement, which tests for isolation from other concepts.In experiments on both text and speech models, we first highlight that concept subspaces may not be uniquely determined, and discuss the implications for concept subspace analysis.Then, we compare properties of concept subspaces estimated using five estimators, proposed in different communities.We find that (1) the choice of estimator impacts the containment and disentanglement properties; (2) the state-of-theart concept erasure method, LEACE, performs well on both testing axes, but still struggles to generalize to unseen data; and (3) in HuBERT speech representations, phone information is both contained and disentangled from speaker information, while speaker information is hard to contain in a compact subspace, despite being disentangled from phones. 1 1 We release the source code at https://github.com/ burin-n/concept_space.

Code-Switching (CS) remains a challenge for Automatic Speech Recognition (ASR), especially character-based models. With the combined choice of characters from multiple languages, the out-come from character-based models suffers from phoneme duplication, resulting in language-inconsistent spellings. We propose Contextualized Connectionist Temporal Classification (CCTC) loss to encourage spelling consistencies of a character-based non-autoregressive ASR which allows for faster inference. The model trained by CCTC loss is aware of contexts since it learns to predict both center and surrounding letters in a multi-task manner. In contrast to existing CTC-based approaches, CCTC loss does not require frame-level alignments, since the context ground truth is obtained from the model’s estimated path. Compared to the same model trained with regular CTC loss, our method consistently improved the ASR performance on both CS and monolingual corpora.

Set prediction tasks require the matching between predicted set and ground truth set in order to propagate the gradient signal. Recent works have performed this matching in the original feature space thus requiring predefined distance functions. We propose a method for learning the distance function by performing the matching in the latent space learned from encoding networks. This method enables the use of teacher forcing which was not possible previously since matching in the feature space must be computed after the entire output sequence is generated. Nonetheless, a naive implementation of latent set prediction might not converge due to permutation instability. To address this problem, we provide sufficient conditions for permutation stability which begets an algorithm to improve the overall model convergence. Experiments on several set prediction tasks, including image captioning and object detection, demonstrate the effectiveness of our method.