Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Repetition Without Exclusivity: Scale Sensitivity of Referential Mechanisms in Child-Scale Language Models

Jon-Paul Cacioli

Abstract

We present the first systematic evaluation of mutual exclusivity (ME) -- the bias to map novel words to novel referents -- in text-only language models trained on child-directed speech. We operationalise ME as referential suppression: when a familiar object is relabelled in a two-referent discourse context, ME predicts decreased probability of the labelled noun at a subsequent completion position. Three pilot findings motivate a pre-registered scale-sensitivity experiment: (1) a masked language model (BabyBERTa) is entirely insensitive to multi-sentence referential context; (2) autoregressive models show robust repetition priming -- the opposite of ME -- when familiar nouns are re-labelled; and (3) a novel context-dependence diagnostic reveals that apparent ME-like patterns with nonce tokens are fully explained by embedding similarity, not referential disambiguation. In the confirmatory experiment, we train 45 GPT-2-architecture models (2.9M, 8.9M, and 33.5M parameters; 5, 10, and 20 epochs on AO-CHILDES; 5 seeds each) and evaluate on a pre-registered ME battery. Anti-ME repetition priming is significant in all 9 cells (85-100% of items; all p < 2.4 x 10^-13). Priming attenuates with improved language modelling (Spearman rho = -0.533, p = 0.0002) but never crosses zero across a 3.8x perplexity range. The context-dependence diagnostic replicates in all 9 cells, and dose-response priming increases with repetitions in 8/9 cells (all trend p < 0.002). These findings indicate that distributional learning on child-directed speech produces repetition-based reference tracking rather than lexical exclusivity. We connect this to the grounded cognition literature and argue that referential grounding may be a necessary ingredient for ME -- an empirical claim about required input structure, not a nativist one.

Repetition Without Exclusivity: Scale Sensitivity of Referential Mechanisms in Child-Scale Language Models

Abstract

We present the first systematic evaluation of mutual exclusivity (ME) -- the bias to map novel words to novel referents -- in text-only language models trained on child-directed speech. We operationalise ME as referential suppression: when a familiar object is relabelled in a two-referent discourse context, ME predicts decreased probability of the labelled noun at a subsequent completion position. Three pilot findings motivate a pre-registered scale-sensitivity experiment: (1) a masked language model (BabyBERTa) is entirely insensitive to multi-sentence referential context; (2) autoregressive models show robust repetition priming -- the opposite of ME -- when familiar nouns are re-labelled; and (3) a novel context-dependence diagnostic reveals that apparent ME-like patterns with nonce tokens are fully explained by embedding similarity, not referential disambiguation. In the confirmatory experiment, we train 45 GPT-2-architecture models (2.9M, 8.9M, and 33.5M parameters; 5, 10, and 20 epochs on AO-CHILDES; 5 seeds each) and evaluate on a pre-registered ME battery. Anti-ME repetition priming is significant in all 9 cells (85-100% of items; all p < 2.4 x 10^-13). Priming attenuates with improved language modelling (Spearman rho = -0.533, p = 0.0002) but never crosses zero across a 3.8x perplexity range. The context-dependence diagnostic replicates in all 9 cells, and dose-response priming increases with repetitions in 8/9 cells (all trend p < 0.002). These findings indicate that distributional learning on child-directed speech produces repetition-based reference tracking rather than lexical exclusivity. We connect this to the grounded cognition literature and argue that referential grounding may be a necessary ingredient for ME -- an empirical claim about required input structure, not a nativist one.
Paper Structure (37 sections, 4 figures, 4 tables)

This paper contains 37 sections, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Background
  3. ME in Developmental Psychology
  4. ME in Computational Models
  5. Child-Scale Language Models
  6. Method
  7. Pre-Registration and Transparency
  8. Corpus
  9. Models
  10. Masked LM (pilot only).
  11. Autoregressive models (confirmatory).
  12. ME Evaluation Battery
  13. Familiar--Familiar Suppression Track (H1)
  14. Context-Dependence Diagnostic (H2)
  15. Dose--Response Track (H3)
  16. ...and 22 more sections

Figures (4)

  • Figure 1: Anti-ME rate (%) and mean priming (nats) across model scale and training duration. All cells show strong anti-ME; priming magnitude decreases toward the bottom-right (larger, more-trained models).
  • Figure 2: Context-dependence diagnostic (mean across 45 models). The nonce_only condition produces the highest apparent "ME," despite containing no familiar noun and no ME paradigm. The full ME design (full_context) scores below chance. Dashed line: chance (4/8).
  • Figure 3: Dose--response curves by model size. Priming advantage (labelled minus unlabelled log-probability) increases with repetition dose at all scales. Error bars: SEM across item--seed pairs. Concavity in the 20-epoch curves (dotted) reflects declining marginal returns on repetition.
  • Figure 4: PPL vs. mean priming across 45 models. Each point is one model; shapes indicate size class. Dashed line: linear regression. Dotted line at 0: ME threshold. Priming attenuates with lower perplexity but remains strictly negative (anti-ME).