Why Are You Wrong? Counterfactual Explanations for Language Grounding with 3D Objects

2025-05-09

AI Summary

Model misclassifications in language-to-3D object referring expression grounding suffer from poor interpretability. Method: This work introduces counterfactual explanations to this task for the first time, proposing a gradient-guided text perturbation method that generates semantically similar and syntactically preserved counterfactual descriptions—enforced via BERTScore constraints and syntactic consistency regularization. These counterfactuals expose latent ambiguities in original utterances and reveal model biases in spatial relation reasoning. Results: Experiments on ShapeTalk across three state-of-the-art models demonstrate high-quality, diagnostically informative counterfactuals that precisely localize linguistic ambiguity and model cognitive blind spots, significantly improving human-in-the-loop debugging efficiency. The core contribution is the first controllable counterfactual explanation framework tailored for 3D referring grounding, jointly optimizing interpretability, semantic fidelity, and practical diagnostic utility.

Technology Category

• Natural Language Processing: Language Grounding & Multi-modal NLP
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Computer Vision: Interpretability, Explainability, and Transparency

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Combining natural language and geometric shapes is an emerging research area with multiple applications in robotics and language-assisted design. A crucial task in this domain is object referent identification, which involves selecting a 3D object given a textual description of the target. Variability in language descriptions and spatial relationships of 3D objects makes this a complex task, increasing the need to better understand the behavior of neural network models in this domain. However, limited research has been conducted in this area. Specifically, when a model makes an incorrect prediction despite being provided with a seemingly correct object description, practitioners are left wondering:"Why is the model wrong?". In this work, we present a method answering this question by generating counterfactual examples. Our method takes a misclassified sample, which includes two objects and a text description, and generates an alternative yet similar formulation that would have resulted in a correct prediction by the model. We have evaluated our approach with data from the ShapeTalk dataset along with three distinct models. Our counterfactual examples maintain the structure of the original description, are semantically similar and meaningful. They reveal weaknesses in the description, model bias and enhance the understanding of the models behavior. Theses insights help practitioners to better interact with systems as well as engineers to improve models.

Problem

• Understanding why models fail in 3D object referent identification
• Generating counterfactual explanations for model misclassifications
• Improving model behavior analysis in language-grounding tasks

Innovation

• Generates counterfactual examples for misclassified samples
• Maintains original description structure and semantic similarity
• Reveals model weaknesses and biases effectively