Abstract
Traditional natural language generation often relies on hard-coded metaphorical substitutions. This paper introduces a framework for generating "Transparently-Motivated" (T-M) metaphors—subtle, everyday expressions grounded in physical experience. By utilizing an "is-a" abstraction hierarchy and goal-oriented filtering, the proposed system selects metaphors that align with specific communicative objectives, such as emphasizing effort or shifting focus.
Introduction
In the early 1990s, while natural language processing (NLP) had made significant progress in metaphor comprehension, the generation of metaphorical language remained largely "canned." Most systems relied on direct encodings of alternatives rather than a robust computational theory. Mark Alan Jones addresses this gap by focusing on Transparently-Motivated (T-M) metaphors. Unlike poetic flourishes, T-M metaphors are ubiquitous, subtle expressions—like "putting money into a stock"—that treat abstract concepts as physical entities (e.g., containers). The goal is to move from hard-coded lists to a system that understands the underlying logic of metaphorical choice.
Methodology
The proposed architecture for metaphor generation follows a structured multi-step process:
- Identification: The system identifies the literal "tenor" and the communicative goal (e.g., "emphasize difficulty").
- Abstraction: The system "climbs" an "is-a" hierarchy to find a high-level category. For instance, "leaving a party" is abstracted to a human process with a termination role.
- Cross-Domain Search: It searches for other domains sharing the same ancestor or role, such as "camping" or "electronics."
- Filtering: Potential metaphors are evaluated based on universality (user familiarity), goal achievement (e.g., "pulling up stakes" vs. "pulling the plug"), and lexical availability.
- Mapping: Conceptual mapping rules translate specific states into metaphorical equivalents.
Key Insights
The framework relies on several core principles:
- Bodily Grounding: Rooted in the work of Lakoff and Johnson, the system prioritizes metaphors based on physical experiences like containment and motion, making them intuitive and "transparent."
- Focus Manipulation: Metaphors can manipulate a reader's attention. Personifying an abstract concept can "push" it up a semantic hierarchy, making it feel more concrete and central to the reader's attention than a literal description would.
- Perspective Shifting: Metaphors serve as tools for perspective. Choosing "pulling up stakes" over "leaving" highlights the labor involved in the transition, a nuance a literal description lacks.
Results and Discussion
The paper demonstrates the system's utility through qualitative examples. In the "Progress as Motion" metaphor, the system maps various states of progress to directional or physical movement to achieve clarity and impact.
Beyond simple mappings, the author suggests that a detailed domain model (e.g., a vehicle) allows the system to generate novel expressions like "My career is running out of gas," extending beyond pre-encoded rules. This demonstrates the generative potential of a system that understands the underlying mechanics of the source domain.
Conclusion
While currently limited to verb-phrase meanings and reliant on pre-defined hierarchies, this research points toward a more general principle of metaphor generation. Future work aims to derive metaphors from general knowledge—such as the relationship between energy and motion—rather than specific mapping rules, moving closer to a truly autonomous creative language system.
Summary updated