Terminology Silos and the Consilience Gap: A Quantitative Audit of Cross-Domain Vocabulary

In the paper's words

Interdisciplinary consilience is claimed everywhere and measured almost nowhere. This paper measures one of its basic preconditions: whether scientific domains share vocabulary. 97.2% of external technical vocabulary occurs in exactly one discipline; bridge terms are massively enriched in method-level vocabulary (Fisher exact p = 8.5e-7, odds ratio 70.2).

The abstract, verbatim, and its opening sentence is the claim. Every line below is quoted from the text.

Numbers it reports

1. 97.2% of the external technical vocabulary occurs in exactly one discipline; the shared core (terms in three or more disciplines) is empty; and the 2.8% of terms that do bridge disciplines are massively enriched in method-level vocabulary (machine learning, language models, upper bounds) rather than structural concepts (Fisher exact $p = 8.5e-7$, odds ratio 70).

   from Abstract

2. If the shared vocabulary of science is 3% of its technical terms, then keyword-based discovery is a 3%-recall instrument for the connections that interdisciplinary consilience depends on.

   from 1. Introduction

3. This paper measures the size of the tax (97% of technical vocabulary is domain-local), its structure (what little is shared is method, not substance), and what to do about it (build bridge vocabularies; do not wait for them to emerge).

   from 1. Introduction

4. Quantity Value Distinct technical terms 466 Domain-local terms (exactly one domain) 453 Partitionality index 0.9721 Bridge terms (two or more domains) 13 Bridge share 0.0279 Shared core (three or more domains) 0 The partitionality hypothesis (H-SILO-1: at least 90% of vocabulary domain-local) is confirmed: 97.2%.

   from 4. Results I: six disciplines, one vocabulary wall

5. The seed audit is itself the exception that tests the rule: it took a computational audit to discover that an organization's own vocabulary was 99.7% partitional.

   from 6. Why silos persist

6. With a 2.8% bridge share, keyword search for a concept outside the searcher's home domain has a recall ceiling of a few percent for cross-domain correspondences.

   from 7. What the silo costs

How it is done

1. 97.2% of the external technical vocabulary occurs in exactly one discipline; the shared core (terms in three or more disciplines) is empty; and the 2.8% of terms that do bridge disciplines are massively enriched in method-level vocabulary (machine learning, language models, upper bounds) rather than structural concepts (Fisher exact $p = 8.5e-7$, odds ratio 70).

   from Abstract

2. This paper measures the size of the tax (97% of technical vocabulary is domain-local), its structure (what little is shared is method, not substance), and what to do about it (build bridge vocabularies; do not wait for them to emerge).

   from 1. Introduction

3. A Fisher exact test of whether the four bridge families (valuation, hierarchy, distinction, bound) coincide with load-bearing vocabulary (keywords shared by three or more programs) gives $p = 1.0$: no enrichment whatsoever.

   from 2. The seed audit: a seven-domain organization's own vocabulary

4. Plus a structural probe: a one-sided Fisher exact test on the bridge x general-family contingency, where the general family is a curated list of structurally general method and pattern terms (machine learning, neural networks, Monte Carlo, phase transitions, upper bounds, and similar vocabulary that plausibly travels across fields).

   from 3. Definitions and instrument

What would falsify it

1. If the shared vocabulary of science is 3% of its technical terms, then keyword-based discovery is a 3%-recall instrument for the connections that interdisciplinary consilience depends on.

   from 1. Introduction

2. Hypothesis Prediction Result H-SILO-1 partitionality >= 0.90 Confirmed (0.9721 external; 0.9970 seed) H-SILO-2 bridge share < 0.10; Fisher $p \lt 0.05$ Confirmed (0.0279; $p = 8.5e-7$, OR 70.2) H-SILO-3 semantic links carry no lexical signature Supported (0.0 on non-authored bridges; within-program links visible)

   from 5. Results II: the QNFO corpus — semantic links and lexical silence

3. The Ultrametric Program: One Structural Object Across Seven Research Domains, and Its Falsifiable Tests. the QNFO archive, https://papers.qnfo.org/papers/keyword-taxonomy-consilience/.

   from References

Where the claim stops

1. Their vocabularies do not overlap; their results do.

   from 1. Introduction

2. This paper measures the size of the tax (97% of technical vocabulary is domain-local), its structure (what little is shared is method, not substance), and what to do about it (build bridge vocabularies; do not wait for them to emerge).

   from 1. Introduction

3. The graph links papers within programs; it has not yet built the cross-program network.

   from 5. Results II: the QNFO corpus — semantic links and lexical silence

Where to read

1. Abstract233 words · 1 min
2. 1. Introduction282 words · 1 min
3. 2. The seed audit: a seven-domain organization's own vocabulary216 words · 1 min
4. 3. Definitions and instrument283 words · 1 min
5. 4. Results I: six disciplines, one vocabulary wall274 words · 1 min
6. 5. Results II: the QNFO corpus — semantic links and lexical silence322 words · 1 min
7. 6. Why silos persist211 words · 1 min
8. 7. What the silo costs158 words · 1 min
9. 8. Consilience infrastructure: bridges are built, not emergent229 words · 1 min
10. 9. Where the premises end239 words · 1 min
11. 10. Limitations and next steps121 words · 1 min
12. 11. Reproducibility65 words · 1 min