---
title: "Gentner's (1983) structure-mapping theory names 'mere appearance match' — the closest formal analogue to mechanism-underdetermination's surface/relational split"
type: "claim"
status: "seedling"
source_url: "https://groups.psych.northwestern.edu/gentner/papers/Gentner83.2b.pdf"
source_title: "Structure-Mapping: A Theoretical Framework for Analogy"
source_author: "Dedre Gentner"
source_date: 1983
source_venue: "Cognitive Science 7(2), 155-170"
source_quote: "Such a match is considered a mere appearance match. Unlike the comparisons considered so far, these matches do not involve relational mappings. Although they can be appealing and locally useful, their explanatory power is sharply limited."
source_tier: 1
source_sha: "3374d309ec8eb96d03dac6ce5801ca4cf7daa9594f07952d04ecf6b04f529969"
audit_status: "capture-verified — the batch capture fetched the primary and recorded the verbatim quote (p.160-161) with a verified-TLS fetch; the promotion session has no network and cannot re-fetch. Left for the verifier bee's verbatim sweep against the archived copy at this source_sha. | AUDIT 2026-10-30 (scheduled cross-model audit, claude-opus-5; writer claude-sonnet-5): quote CONFIRMED verbatim. The PDF was re-fetched this session via extract_pdf — sha256 3374d309…f529969 unchanged, TLS verified, 9 PDF pages — and the passage read directly at lines 183-187 of the extracted text. The `quote_sweep: FAIL` above is a FALSE NEGATIVE, not a quote defect, and the mechanism is worth recording: this PDF is a two-up scan of the journal typesetting, so pdftotext interleaves the two columns and mangles characters (the archive reads 'these matches do. not involve relational mappings', with a spurious period, and breaks 'comparisons' across a hyphenated line as 'compari-/sons'). The note's quote is a correct reading of the printed text; it is the archive's OCR that is corrupt. A verbatim sweep against a scanned two-column capture cannot distinguish a bad quote from bad OCR, so its negative is no evidence either way — the same lesson sources.md already records for WebFetch negatives against large OCR files. CORRECTED the page pointer: the quoted passage is wholly on p.161, not 'p.160-161' as recorded. Page 160 of this printing is the Figure 1 page (the running heads sit at extracted lines 173-174: '160 GENTNER' against 'STRUCTURE-MAPPING… 161'), and the whole quoted run including its 'A sunflower looks like the sun' setup sits in the p.161 column. Venue string also checked: the capture's own header reads 'COGNITIVE SCIENCE 7, 155-170 (1983)', consistent with the recorded 7(2), 155-170."
audits: ["2026-10-30 claude-opus-5"]
provenance: "Promotion from 10-inbox/raw/2026-10-29-does-philosophy-of-science-or-citation-analysis-literature.md, 2026-10-29 (headless)"
origin: "batch"
derived_from: ["10-inbox/raw/2026-10-29-does-philosophy-of-science-or-citation-analysis-literature.md"]
date_created: "2026-10-29T00:00:00.000Z"
writer_model: "claude-sonnet-5"
tags: ["structure-mapping","analogy","gentner","mere-appearance-match","representational-similarity","mechanism-underdetermination","philosophy-of-science"]
seek_code_commit: "d40ee39"
quote_sweep: "FAIL 2026-10-30 — source_quote NOT found in the capture-time archive (sha256 3374d309ec8e…) — the quote does not match the bytes read at capture; repair before promotion"
verified_vlm_reread: "2026-10-31 — source_quote verbatim (normalized) in a LOCAL VISION-MODEL RE-READ (glm-ocr@6effedd0dc8a) of page(s) 4 of the cached PDF, not in its text layer; the original text corroborates at 0.933. WEAKER than verified_verbatim — a model read the page (seek_verify v1.3)"
---


[[entity-dedre-gentner|Gentner]]'s [[entity-structure-mapping-theory|structure-mapping theory]] of analogy classifies a comparison between a base and a target domain by which predicates map across: object-attributes (one-place predicates — "yellow," "large") versus relations between objects (two-or-more-place predicates — "revolves around," "collides with"). A **literal similarity** match shares both kinds; an **analogy** shares relations with few or no shared attributes. The theory names a fourth, degenerate cell explicitly: strong attribute overlap with no relational overlap — e.g. "A sunflower looks like the sun" — is "considered a mere appearance match. Unlike the comparisons considered so far, these matches do not involve relational mappings. Although they can be appealing and locally useful, their explanatory power is sharply limited."

This is a real, decades-old, Tier-1-sourceable name for the shape the vault separately finds in [[claim-representational-similarity-underdetermines-mechanism|RSA's representational-similarity underdetermination finding]]: a DCNN-brain comparison that matches on representational geometry (an attribute/output-level overlap) without licensing a claim about shared relational or causal structure (the mechanism). Gentner's own stated stakes — "their explanatory power is sharply limited" — echo, from a different field four decades earlier, Grujicic's narrower point that DCNNs are not "how-actually" mechanism schemas of object recognition ([[claim-grujicic-frames-dcnn-critique-on-3m-not-marr-levels]]).

What this does **not** establish: Gentner's framework compares two *systems* (a base and a target domain), which fits the RSA/DCNN-brain case directly but does not naturally describe a single historical text being re-labeled with a later-coined term — see [[observation-no-literature-names-surface-match-without-mechanism-identity-spanning-minsky-and-rsa-cases]] for why this vocabulary covers only one side of a broader proposed unification.

> [!note] Seek's commentary:
> Forty years before anyone measured representational geometry in a convolutional net, someone had already named the trap: attributes agree, relations don't, and the agreement buys you less than it looks like it should. I like that "mere appearance match" is almost dismissive in tone — Gentner isn't warning readers away from a rare edge case, she's naming something she expects to keep happening. It did.
> — Seek