Mining Similes for Stereotypical Property Extraction

21 concepts14 papers1 proposal

Traditional lexical databases excel at capturing hierarchical, essential relationships. For instance, WordNet can confidently tell you that an elephant is a mammal or that a knife is a tool. However, these formal ontologies often omit the common-sense, experiential, and cultural attributes that dominate human thought. They rarely encode that an elephant is large, a snail is slow, or a knife is sharp.

As Palmira Marrafa et al. argue in Modeling Adjectives in Computational Relational Lexica, capturing these non-essential but culturally frequent properties requires directly mapping nouns to their characterizing adjectives. Without this descriptive layer, computational models remain rigid. For example, understanding when "chocolate" should be treated as a "drug" in an addiction context requires a fluid, context-sensitive understanding of its latent properties rather than a static taxonomic definition (Tony Veale et al. in Making Lexical Ontologies Functional and Context-Sensitive).

Similes as Semantic Anchors

How can a machine automatically discover these stereotypical traits without manual curation? The core insight of this research direction is that human language explicitly flags its most salient conceptual prototypes through a specific figurative structure: the simile.

When speakers want to convey the extreme intensity of a qualitative trait, they anchor it to a universally understood exemplar using comparative templates like "as [adjective] as a [noun]". In Ordering Concepts Based on Common Attribute Intensity, Tatsuya Iwanari et al. demonstrate that similes serve as high-signal indicators for the extreme ends of an attribute's intensity spectrum. Because these phrases highlight idealized states rather than literal descriptions, they filter out incidental modifiers and isolate core cultural stereotypes (Tony Veale et al. in Multilingual Harvesting of Cross-Cultural Stereotypes).

Furthermore, because similes explicitly pair a property with a noun via rigid syntactic patterns, they provide a high-precision signal for automated discovery that far exceeds the accuracy of mining implicit metaphors or general text co-occurrences (Hongsong Li et al. in Data-Driven Metaphor Recognition and Explanation).

Web-Scale Pattern Mining

To transform this linguistic phenomenon into structured knowledge, researchers employ template-driven wildcard queries across massive web corpora. By querying search engines or large n-gram databases with patterns like "as * as a [noun]" or "as [adjective] as a *", systems can harvest explicit property-noun pairings at scale (Tony Veale et al. in Comprehending and Generating Apt Metaphors: A Web-driven, Case-based Approach to Figurative Language).

For instance, querying autocomplete engines with templates like "as [property] as *" retrieves highly evocative nouns that represent specific attributes in natural language (Gözde Özbal et al. in A Computational Approach to the Automation of Creative Naming). Similarly, tracking patterns like "[verb]-ing like a [noun]" extracts common behavioral features associated with specific concepts (Tony Veale in A Context-sensitive, Multi-faceted Model of Lexico-Conceptual Affect).

A parallel approach by Shane Bergsma et al. in Using Conceptual Class Attributes to Characterize Social Media Users utilizes possessive pattern mining within n-gram corpora to extract class-specific attributes based on third-person linguistic structures, reinforcing how targeted syntactic patterns can surface latent common-sense knowledge.

Applications in Fluidity and Creativity

Once harvested, these web-derived similes act as a dense semantic layer that fuels downstream tasks requiring fluid reasoning and computational creativity.

Dynamic and Lateral Similarity

Instead of navigating rigid vertical hierarchies, models can perform lateral similarity judgments. In Creating Similarity: Lateral Thinking for Vertical Similarity Judgments, Tony Veale et al. show that validating adjective-noun pairs through web similes allows systems to explain why two distinct concepts share a semantic bond. This enables fluid knowledge representations where concepts adapt dynamically to their context (Tony Veale et al. in A Fluid Knowledge Representation for Understanding and Generating Creative Metaphors).

Taxonomy Bootstrapping

Because simile-derived properties capture highly discriminative stereotypical beliefs, they serve as exceptionally rich seeds for building fine-grained concept distinctions. Tony Veale et al. demonstrate in Growing Finely-Discriminating Taxonomies from Seeds of Varying Quality and Size that bootstrapping from these high-quality seeds yields superior taxonomies compared to using larger, noisier data sources.

Metaphor Generation and Creative Language

Similes provide an explicit bridge between literal attributes and figurative expressions. In Creative Language Retrieval: A Robust Hybrid of Information Retrieval and Linguistic Creativity, Tony Veale shows that mapping nouns to their salient properties via semantic wildcards enables powerful metaphor generation and noun categorization.

This underpins creative systems like the Jigsaw Bard, which reuses linguistic "readymades" from web data to generate novel, evocative similes (Tony Veale et al. in Exploiting Readymades in Linguistic Creativity: A System Demonstration of the Jigsaw Bard), as well as automated branding systems that leverage retrieved stereotypes to craft meaningful neologisms (Gözde Özbal et al. in A Computational Approach to the Automation of Creative Naming).

Key Foundational Work

Making Lexical Ontologies Functional and Context-Sensitive (2007) — contributed 2 concept elements to this direction.
Comprehending and Generating Apt Metaphors: A Web-driven, Case-based Approach to Figurative Language (2007) — contributed 2 concept elements to this direction.
Modeling Adjectives in Computational Relational Lexica (2006) — contributed 2 concept elements to this direction.
History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 14 papers
  2. Version 1 · 2026-10-04 · 21 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_simile_based_property_extraction_13d6a593,
  title = {simile-based property extraction},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/simile-based-property-extraction/rdr_13d6a5938f294de3bad94bbd45eb5431},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-22},
}

This exact version (version 2, frozen text):

@misc{lacuna_simile_based_property_extraction_13d6a593_v2,
  title = {simile-based property extraction},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/simile-based-property-extraction/rdr_13d6a5938f294de3bad94bbd45eb5431/v/2},
  note = {Version 2, frozen 2026-10-04. Accessed 2026-11-22},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake