Information Compression via Eliding Verb Phrase: A Dependency-Based Study

Zheyuan Dai, Chenliang Zhou, Haitao LiuPaclic 2023

Introduction

Verb Phrase Ellipsis (VPE) is a common linguistic phenomenon where a verb phrase is omitted because it has already been mentioned in the discourse. This study conceptualizes VPE as a "ZIP file" for human language—a mechanism for information compression. By applying dependency grammar and information theory, the researchers quantify how this shortcut simplifies sentence structure while maintaining the integrity of the message.

Problem Statement

While qualitative descriptions of VPE syntax and psycholinguistic processing speeds are well-documented, quantitative data regarding the efficiency of these structures has been lacking. The core question is whether VPE truly adheres to the "Principle of Least Effort": does it significantly reduce structural complexity, and if so, is there a corresponding loss of mathematical information (entropy)?

Methodology

The study utilized 343 VPE sentences from the Wall Street Journal (WSJ) Corpus, manually restoring them to their full forms for comparison. Two primary frameworks were employed:

  1. Dependency Grammar (DG): Researchers measured the Dependency Distance (DD)—the linear distance between a governor word and its dependent. The Dependency Distance Difference (DDD) was calculated to determine the degree of structural simplification.
  2. Information Theory: Shannon Entropy was used to calculate the average information content of both elided and full-form sentences to measure "meaning" stability.
Figure 2: Dependency structure of an elliptical sentence, illustrating the relationships between words in a VPE context.

Key Results and Discussion

Structural Efficiency

The analysis revealed a mean DDD of 11.503 (SD = 9.659). This positive and statistically significant value confirms that VPE fundamentally shrinks the dependency graph of a sentence, thereby reducing the cognitive memory burden required for syntactic processing.

Information Fidelity

Across the general dataset, no significant difference in entropy was found between elided and full sentences (p = .723). This suggests that VPE is a "fidelity-preserving" compression method, removing redundant form without sacrificing substance. However, interesting divergences appeared in specific sub-groups:

  • The "Be" Group: Entropy actually increased after ellipsis. The researchers suggest that eliding after "be" (e.g., "I am [happy]") flattens word frequency, creating higher uncertainty.
  • The "To" Group: Entropy decreased. In these cases, ellipsis allows speakers to avoid complex verb inflections required in full forms, resulting in a simpler information density.
Figure 6: Frequency distributions for 'have', 'modal', and 'to' auxiliary types, showing the differences between elided (el) and non-elided (nel) forms.

Conclusion

This research provides quantitative evidence that Verb Phrase Ellipsis is an efficient tool for human communication. It significantly simplifies the structural complexity of language while maintaining stable information content. While limited by a relatively small English-only sample size, this information-theoretic approach offers a promising path toward a unified theory of linguistic economy, balancing communicative accuracy with cognitive effort.

Summary updated

BibTeX
@inproceedings{dai2023information,
  author = {Zheyuan Dai and Chenliang Zhou and Haitao Liu},
  title = {{Information Compression via Eliding Verb Phrase: A Dependency-Based Study}},
  booktitle = {Paclic},
  year = {2023},
}
Figure