Demonstrative Reference and Semantic Space: A Large-Scale Demonstrative Choice Task Study

Roberta Rocca, Mikkel Wallentin

Abstract

Spatial demonstratives, such as this and that, are traditionally understood as tools for partitioning physical space into peripersonal (near) and extrapersonal (far) domains. However, recent research suggests these terms also navigate a complex multidimensional semantic space. This study utilizes the Demonstrative Choice Task (DCT) with a large sample (N = 2,197) to map demonstrative use across 506 words rated on over 70 semantic dimensions. We replicate findings that object manipulability influences choice and identify additional significant factors, including valence, arousal, loudness, and motion. Our results demonstrate that demonstrative choice is highly structured and anchored in a generalized "self," where this denotes entities proximal to the speaker's bodily and psychological experience, while that marks distance in a broader semantic field.

Introduction: Beyond Physical Distance

In linguistics, deictic expressions like this and that are context-dependent, relying on the speaker's perspective, or "origo." While traditionally defined by physical proximity, we frequently use these terms for abstract concepts or objects outside our immediate view. Previous small-scale studies indicated that "manipulability"—the ease with which an object can be handled—plays a crucial role in this categorization. This paper expands on that foundation, investigating whether a broader, multidimensional semantic space governs how we use these simple words to categorize the world.

Methodology: The Demonstrative Choice Task (DCT)

To explore these semantic underpinnings, researchers employed the Demonstrative Choice Task (DCT). Participants were presented with nouns (e.g., "apple," "tiger") in isolation and asked to pair them with either this or that.

To analyze the drivers of these choices, the team utilized a massive dataset of semantic ratings, integrating the Binder et al. (2016) cognitive-neural dimensions and the Lancaster Sensorimotor Norms. Factor Analysis was used to reduce 76 initial dimensions into 12 primary latent factors. The researchers then applied aggregate linear regression to predict word-level demonstrative preferences and a logistic regression classifier to predict individual trial choices based on semantic profiles.

Results: The Multidimensional Anchor of the Self

The study revealed that demonstrative choice is far from random. A split-sample comparison showed a remarkably high correlation ($r = 0.82$) between participant groups, suggesting a deeply shared cognitive structure.

Ten of the twelve identified semantic factors significantly influenced demonstrative choice. The strongest predictor for this was a factor labeled "Self," encompassing internal needs and pleasantness. Conversely, factors associated with high sensory intensity or external complexity favored the use of that.

Figure 3: Factor loadings for auditory-related dimensions. High loadings on features like 'Loud' and 'Sound' significantly predict the use of the distal demonstrative 'that'.

Specifically, sensory "push-back" was observed: stimuli characterized by high loudness or rapid motion were significantly more likely to be paired with that. For example, the factor capturing auditory intensity (Figure 3) showed a strong negative coefficient, indicating that as a concept becomes "louder," it is pushed further from the speaker's semantic center. Similarly, factors involving social complexity and biological motion (Figure 4) also predicted a preference for that.

Figure 4: Factor loadings for human and biological motion dimensions. Concepts associated with social entities and complex motion are more likely to be categorized as 'that'.

The predictive model achieved 58.5% accuracy in individual trials, nearing the theoretical ceiling of 59.5% established for this inherently ambiguous task. The resulting profiles are clear: this is reserved for the small, manipulable, quiet, and positive, while that is used for the large, loud, fast, and negative.

Discussion: The Self as an Extension of the Body

These findings suggest that the "self" serves as the primary anchor for demonstrative reference, extending the concept of the body beyond physical reach into a multidimensional semantic space. We "claim" positive and manageable entities with this, while distancing ourselves from the loud, the harmful, or the socially complex with that.

While the DCT effectively maps default semantic associations, it lacks the 3D physical context of real-world "exophoric" pointing. Future research should investigate individual differences—such as personal phobias or preferences—and cross-linguistic variations to determine if this semantic mapping is a universal feature of human cognition. Ultimately, this work demonstrates that our most basic linguistic tools are deeply rooted in how we perceive our place within the world.

Summary updated

Abstract

Spatial demonstratives (words like this and that) have been thought to primarily be used for carving up space into a peripersonal and extrapersonal domain. However, when given a noun out of context and asked to couple it with a demonstrative, speakers tend to choose this for words denoting manipulable objects (small, harmless, inanimate), while non-manipulable objects (large, harmful, animate) are more likely to be coupled with that. Here, we extend these findings using the Demonstrative Choice Task (DCT) procedure and map demonstrative use along a wide spectrum of semantic features. We conducted a large-scale (N = 2197) DCT experiment eliciting demonstratives for 506 words, rated across 65+11 perceptually and cognitively relevant semantic dimensions. We replicated the finding that demonstrative choice is influenced by object manipulability. Demonstrative choice was furthermore found to be related to a set of additional semantic factors, including valence, arousal, loudness, motion, time and more generally, the self. Importantly, demonstrative choices were highly structured across participants, as shown by a strong correlation detected in a split-sample comparison of by-word demonstrative choices. We argue that the DCT may be used to map a generalized semantic space anchored in the self of the speaker, the self being an extension of the body beyond physical space into a multidimensional semantic space.
BibTeX
@inproceedings{rocca2020demonstrative,
  author = {Roberta Rocca and Mikkel Wallentin},
  title = {{Demonstrative Reference and Semantic Space: A Large-Scale Demonstrative Choice Task Study}},
  booktitle = {Frontiers in Psychology},
  year = {2020},
  doi = {10.3389/fpsyg.2020.00629},
}
Figure