Can Language Models Encode Perceptual Structure Without Grounding? A Case Study in Color

By Mostafa Abdou, Artur Kulmizev, Daniel Hershcovich, Stella Frank, Ellie Pavlick, Anders Søgaard. 2021

178 citations recorded by the source corpus.

Pretrained language models have been shown to encode relational information, such as the relations between entities or concepts in knowledge-bases — (Paris, Capital, France). However, simple relations of this type can often be recovered heuristically and the extent to which models implicitly reflect topological structure that is grounded in world, such as perceptual structure, is unknown. To explore this question, we conduct a thorough case study on color. Namely, we employ a dataset of monolexemic color terms and color chips represented in CIELAB, a color space with a perceptually meaningful distance metric. Using two methods of evaluating the structural alignment of colors in this space with text-derived color term representations, we find significant correspondence. Analyzing the differences in alignment across the color spectrum, we find that warmer colors are, on average, better aligned to the perceptual color space than cooler ones, suggesting an intriguing connection to findings from recent work on efficient communication in color naming. Further analysis suggests that differences in alignment are, in part, mediated by collocationality and differences in syntactic usage, posing questions as to the relationship between color perception and usage and context.

Subfields: Experimental and Cognitive Psychology, Natural Language Processing, Representation & Self-Supervised Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Experimental and Cognitive Psychology

In Natural Language Processing

In Representation & Self-Supervised Learning

Citations in the corpus

Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Them (2022, background): “, 2022), and answering questions about color, spatial, and cardinal concepts (Abdou et al., 2021; Patel and Pavlick, 2022).”
Pre-Trained Language Models for Interactive Decision-Making (2022, background): “, 2020) and the structure of color spaces (Abdou et al., 2021).”
Mapping Language Models to Grounded Conceptual Spaces (2022, methodology): “We discuss the limitations of focusing on text-only inputs to teach models grounded concepts, as well as the implications of such work for allowing large language models to be mapped, in a data-efﬁcient way, to the grounded world.”
Grounding Large Language Models in Interactive Environments with Online Reinforcement Learning (2023, unknown): “about space Patel and Pavlick [2022], colors Abdou et al. [2021] or even affordances between bodies and objects Ahn et al.”
The Platonic Representation Hypothesis (2024, unknown): “– Unlike Abdou et al. (2021), we did not perform linear regression from language embedding to CIELAB space, which distorts distances and easily overfits with only 20 samples.”
Reasoning or Reciting? Exploring the Capabilities and Limitations of Language Models Through Counterfactual Tasks (2023, background): “…being trained on only textual data, LMs have been shown to be able to structure their representations of perceptual concepts such as size and color (Abdou et al., 2021; Patel and Pavlick, 2022; Zhang et al., 2020; Ilharco et al., 2021; i.a. ) in a way that credibly mirrors the physical world.”
Language Models Represent Space and Time (2023, background): “Abdou et al. (2021) and Patel & Pavlick (2021) show LLMs have representations that reflect perceptual and conceptual structure in color and spatial domains.”
The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets (2023, background): “Substantial previous work has studied whether LLMs encode world models in their representations (Li et al., 2023a; 2021; Abdou et al., 2021; Patel & Pavlick, 2022).”
Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task (2022, background): “Recent evidence suggests language models can develop internal representations for very simple concepts, such as color, direction Abdou et al. (2021); Patel & Pavlick (2022), or tracking boolean states during synthetic tasks (Li et al., 2021) (see Related Work (section 6) for more detail).”