Metaphor Comprehension - A Special Mode of Language Processing?

Jon M. Slack

Abstract

This paper investigates whether metaphor comprehension requires a unique cognitive "special mode" or if it leverages standard language processing mechanisms. By analyzing how humans interpret semantic mismatches—such as "Billboards are warts on the landscape"—we propose a model based on the interaction between Topic Knowledge Structures (TKS) and Generated Conceptual Structures (GCS). The findings suggest that metaphor comprehension is not a distinct process but a resource-intensive application of standard linguistic tools, requiring sophisticated control structures to manage complex mapping and feature pruning.

Introduction

How does a linguistic system—human or artificial—make sense of a sentence like "Billboards are warts on the landscape"? Literally, this is a category error; a billboard is a physical sign, while a wart is a biological growth. In Natural Language Processing (NLP), this is identified as a semantic mismatch. A long-standing debate in linguistics and AI asks whether the brain requires a "special mode" of processing to resolve these contradictions or if metaphor is simply an intensified version of literal language processing.

Jon M. Slack addresses this by proposing that metaphor comprehension utilizes standard cognitive mechanisms—such as fuzzy matching and spreading activation—but operates under a more complex control structure to manage the mapping and pruning of conceptual ideas.

Theoretical Framework: The Interaction Model

The proposed model relies on the interaction between two primary knowledge structures: the Topic Knowledge Structure (TKS) and the Vehicle Expansion.

1. Vehicle Expansion and the GCS

When a semantic mismatch is detected, the system initiates "Vehicle Expansion." For a metaphor like "The world is a chessboard," the "vehicle" (chessboard) is expanded into a Generated Conceptual Structure (GCS). This structure contains all salient features of the vehicle, such as "strategic," "competitive," "grid-based," and "game-oriented."

2. Topic Knowledge Structure (TKS)

The "topic" (the world) is represented by a TKS, which provides the context for the metaphor. The system identifies the vehicle by applying a "Minimal Match" rule: in isolated metaphors, the vehicle is typically the element with the least initial overlap with the rest of the sentence. This trigger tells the system which word needs to be expanded.

3. Integration and Pruning

The core of comprehension lies in mapping the GCS onto the TKS. This involves two critical steps:

  • Matching: Identifying low-saliency overlaps (e.g., both a chessboard and the world involve "complex interactions").
  • Pruning: Deleting features of the vehicle that directly contradict the topic. For instance, the literal "wooden board" aspect of a chessboard is pruned because it cannot apply to the "world."

Control Strategies and Resource Management

For more complex metaphors, the model employs advanced control strategies. Context Construction allows the system to search for missing background information, while Recursive Interpretation treats components of the metaphor as metaphors themselves.

The "specialness" of metaphor, therefore, does not lie in unique algorithms. Instead, it is a matter of control over process. Metaphor comprehension is functionally compatible with standard language understanding systems (such as those developed by Schank or Wilks) but forces the system to invoke a complex control structure to handle larger, richer knowledge domains. It is a more resource-intensive version of literal processing, governed by memory constraints and processing capacity.

Empirical Basis

The model was developed through an analysis of paraphrases produced by 20 human subjects for a collection of over 50 metaphors. These included various linguistic forms, such as copula metaphors ("The world is a chessboard") and verb-based metaphors ("His words were dried by the sun"). The qualitative data supported the idea that comprehension is a strategic management of conceptual features rather than a separate cognitive track.

Conclusion and Future Work

Metaphor comprehension is an extension of normal language processing, triggered by semantic restriction violations. While the underlying tools are standard, the management of these tools—specifically the pruning of contradictory features and the mapping of salient ones—requires significant cognitive resources.

Future research should focus on quantifying the processing capacity required for these operations and exploring how "extra-sentential context" (the broader conversation) influences the identification of topics and vehicles in real-world usage, as this adds another layer of complexity to the Topic Knowledge Structure.

Summary updated

BibTeX
@inproceedings{slack1980metaphor,
  author = {Jon M. Slack},
  title = {{Metaphor Comprehension - A Special Mode of Language Processing?}},
  booktitle = {Annual Meeting of the Association for Computational Linguistics},
  year = {1980},
  doi = {10.3115/981436.981442},
}
Figure