# Full Text: Ento-Linguistics: Language, Ambiguity, and Scientific Communication in Entomology: How Terminology Networks Shape Understanding of Insect Biology (And Vice-Versa)

> Extracted from `Ento_Linguistics_DAF_TCC_v1_04-15-2026.pdf`

---

## Page 1

Ento-Linguistics: Language, Ambiguity, and Scientific Communication
in Entomology
How Terminology Networks Shape Understanding of Insect Biology (And Vice-Versa)
Daniel Ari Friedman
Active Inference Institute, APOIDEAS
daniel@activeinference.institute
⋅
ORCID:
0000-0001-6232-9096
Tucker Cahill Chambers
APOIDEAS
ORCID: 0009-0008-3793-7872
April 15, 2026
DOI: 10.5281/zenodo.19574118
Contents
1
Abstract
3
2
Introduction
4
2.1
Linguistic Priors and Generative Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4
2.2
Motivation: Minimizing Model Misspecification . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4
2.3
The Challenge of Terminological Reform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5
2.4
Ento-Linguistic Domains: A Framework for Analysis . . . . . . . . . . . . . . . . . . . . . . . . . .
5
2.5
Research Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6
2.6
Terminology Network Visualization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6
3
Methods
7
3.1
Data Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7
3.2
Statistical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9
4
Results: Corpus Analysis and Terminology Networks
13
4.1
Terminology Extraction Across Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
13
4.2
Terminology Network Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
13
4.3
Framing Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15
5
Results: Domain-Specific Findings
16
5.1
Unit of Individuality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
16
5.2
Power & Labor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
16
5.3
Behavior & Identity
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
16
5.4
Sex & Reproduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
16
5.5
Kin & Relatedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
17
5.6
Economics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
17
5.7
Longitudinal Case Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
18
6
Discussion
19
6.1
Language as Constitutive of Scientific Practice
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
19
6.2
From Metaphor to Mechanism: An Active Inference Perspective . . . . . . . . . . . . . . . . . . . .
20
1

## Page 2

6.3
Comparison with Existing Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21
6.4
Practical Implications for Scientific Communication . . . . . . . . . . . . . . . . . . . . . . . . . . .
21
6.5
The “Slave” Terminology Debate: A Case Study in Reform
. . . . . . . . . . . . . . . . . . . . . .
22
6.6
Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
22
7
Conclusion
24
7.1
Core Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
24
7.2
Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
24
7.3
Closing Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
25
8
Related Work
26
8.1
Critical Discourse Analysis and Science Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
26
8.2
Feminist and Postcolonial Epistemology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
26
8.3
Computational Approaches to Scientific Discourse
. . . . . . . . . . . . . . . . . . . . . . . . . . .
27
8.4
Terminology Studies in Entomology
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
27
8.5
Active Inference and Colony Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
27
8.6
Positioning This Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
28
9
Acknowledgments
29
9.1
Institutional Support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
29
9.2
Collaborations
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
29
9.3
Data and Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
29
10 Symbols and Notation Glossary
30
10.1 Mathematical Notation
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
30
10.2 Theoretical Terms
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
31
10.3 Pipeline Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
31
11 References
33
12 Supplemental Methods: Text Processing and Term Extraction
36
12.1 Package Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
36
12.2 Text Processing ( src/analysis/text_analysis.py ) . . . . . . . . . . . . . . . . . . . . . . . . .
36
12.3 Terminology Extraction ( src/analysis/term_extraction.py ) . . . . . . . . . . . . . . . . . . .
38
12.4 Semantic Entropy ( src/analysis/semantic_entropy.py ) . . . . . . . . . . . . . . . . . . . . . .
39
13 Supplemental Methods: Statistical and Scoring Infrastructure
41
13.1 Statistical Analysis ( src/analysis/statistics.py ) . . . . . . . . . . . . . . . . . . . . . . . . .
41
13.2 Domain Analysis ( src/analysis/domain_analysis.py ) . . . . . . . . . . . . . . . . . . . . . . .
41
13.3 Conceptual Mapping ( src/analysis/conceptual_mapping.py ) . . . . . . . . . . . . . . . . . . .
42
13.4 CACE Scoring ( src/analysis/cace_scoring.py )
. . . . . . . . . . . . . . . . . . . . . . . . . .
42
13.5 Rhetorical Analysis ( src/analysis/rhetorical_analysis.py ) . . . . . . . . . . . . . . . . . . .
43
13.6 Visualization ( src/visualization/ ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
43
13.7 Core Infrastructure ( src/core/ )
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
43
13.8 Reproducibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
44
14 Supplemental Results
45
14.1 Pairwise Domain Comparisons
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
45
2

## Page 3

14.2 CACE Scoring for Key Terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
45
14.3 Semantic Entropy Distribution
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
45
14.4 Confidence Intervals for Domain Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
45
15 Supplemental Analysis: Theoretical Extensions
46
15.1 Theoretical Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
46
15.2 Framing Analysis Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
48
15.3 Network Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
48
16 Supplemental Analysis: Case Studies and Validation
50
16.1 Validation Frameworks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
50
16.2 Case Study Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
50
16.3 Methodological Reflections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
50
1
Abstract
Scientific language does not merely describe biological phenomena; it actively constitutes the generative models
through which researchers parse complex systems. This paper makes three core contributions to understanding—
and correcting—the epistemic consequences of this constitutive role. First, we introduce a six-domain Ento-
Linguistic framework that decomposes the terminological landscape of insect research into analytically tractable
themes, isolating domains where anthropomorphic language most severely distorts causal modeling. Second, we de-
velop an open-source computational pipeline that integrates automated term extraction, co-occurrence network
construction, and information-theoretic ambiguity scoring with principles from Active Inference and Complex
Systems Theory. Third, we propose and validate four evidence-based meta-standards—Clarity, Appropriateness,
Consistency, and Evolvability (CACE)—as a formalized protocol for lexical engineering. Analysis of a corpus
encompassing 369 entomological publications (48787 tokens; 7105 unique token types; Type–Token Ratio 0.1456)
extracts 888 candidate terms (with 261 assigned to specific semantic domains across 6 conceptual clusters linked
by 9 weighted relationships). The resulting terminology networks display strong modularity alongside systematic
cross-domain bridging—most prominently in the Power and Labor domain, where 43 bridging terms generate
extensive semantic bleed-over into adjacent domains. Terms such as “queen” (241 occurrences), “worker” (269),
and “caste” (121) implicitly impose hierarchical control topologies onto biological structures that are funda-
mentally stigmergic and decentralized. Across all 261 domain-assigned terms, 16.9% exhibit context-dependent
semantic drift, demonstrating how conceptual constructs like “individuality” span multiple biological scales and
consequently blur the formal systemic boundaries (Markov Blankets) required for mathematically rigorous mod-
eling. The accompanying fully reproducible computational pipeline provides the quantitative analytical tools
necessary for a more self-aware and epistemically rigorous scientific practice. All code and data are available at
https://github.com/docxology/ento_linguistics.
3

## Page 4

2
Introduction
2.1
Linguistic Priors and Generative Models
Scientific inquiry is a process of active inference, where researchers refine generative models to minimize surprise
about biological observations Friston (2010). Language acts as the hyper-prior for these models: it constrains the
hypothesis space before data collection begins. When entomologists employ terms like “queen” or “caste,” they are
not merely labeling phenomena; they are importing a high-precision prior from human social systems into their
model of insect biology. If this prior is structurally misaligned with the target system—for instance, assuming
top-down control in a stigmergic network—the resulting model will necessarily suffer from high variational free
energy, manifesting as persistent anomalies and theoretical epicycles Clark (2013), Kuhn (1996).
The scientific community itself can be modeled as a multi-scale Active Inference agent whose collective
task is to minimize long-term surprise about the entomological world it observes. Its generative model is the
shared ontology of the field—the lexicon and conceptual structures encoded in the literature. When this ontology
is precise and plastic, the community efficiently updates its priors in response to new evidence (e.g., genomic data
revealing that caste determination is a labile epigenetic process rather than a fixed fate). When the ontology is
rigid or laden with hidden anthropomorphic priors, the agent suffers from prior dogmatism: a failure of belief
updating where high-precision, fixed priors overwhelm contradictory sensory evidence. In this state, anomalies
are explained away rather than used to update the model. Terminology reform is therefore a model selection
process: optimizing the community’s generative model to restore its capacity for free-energy minimization.
This optimization requires specific criteria. We propose Evolvability—defined here as scale-invariance—as a
critical metric for scientific terms. An evolvable term maintains its validity across biological scales (gene, organ-
ism, superorganism) without fracturing. “Queen,” by contrast, is scale-brittle: it functions as a metaphor at the
colony level but dissolves into incoherence when applied to the underlying genetic or molecular mechanisms of
reproductive differentiation.
The consequences of this misalignment are not merely philosophical. They propagate through every stage of the
research cycle—from hypothesis formulation, through variable selection, to the causal explanations offered for
observed phenomena. The following section formalizes this propagation as a problem of model integrity.
2.2
Motivation: Minimizing Model Misspecification
The drive for terminological clarity is not a stylistic preference but a requirement for model integrity. As Keller
(1991) argued, the language of science constitutes the cognitive scaffolding of research. In the framework of Active
Inference, an undefined or metaphor-laden term introduces irreducible uncertainty (entropy) into the scientific
communication channel, degrading the precision of the community’s collective generative model.
The present moment demands this formalization. Recent cognitive science emphasizes the distinction between
deliberate and conventional metaphor use, demonstrating that metaphor in scientific discourse often operates as
a conscious communicative strategy rather than an automatic conceptual mapping Steen (2017). Rather than
perpetuating inherited assumptions in our linguistic ontology, researchers must critically assess whether their
terminological priors minimize or maximize the complexity of their biological models.
A paradigmatic example is the “slave-making” debate. Herbers (2006) showed that the term “slave” naturalizes a
human institution while obscuring the biological mechanism of social parasitism. In formal terms, the “slave”
metaphor implies a conscious coercion policy, whereas the replacement term “dulosis” correctly identifies the
phenomenon as a breakdown in nestmate recognition signals—a failure of the Markov Blanket’s security filter.
Reform here is not merely ethical; it restores the causal fidelity of the scientific model by replacing a high-entropy
metaphor with a mechanistically precise descriptor.
4

## Page 5

2.3
The Challenge of Terminological Reform
A common objection to terminological reform is that changing vocabulary creates disconnection from existing
literature. If entomologists abandon terms like “caste” or “slave,” how would researchers locate papers on task
performance or social parasitism?
This objection inadvertently strengthens the case for reform. Retaining problematic terminology for convenience
perpetuates and compounds the conceptual distortions it encodes Herbers (2006). The appropriate response is sys-
tematic development of clearer vocabulary alongside the indexing infrastructure needed for literature continuity—
cross-referencing deprecated terms, establishing synonym mappings, and leveraging modern search capabilities
that already make vocabulary-independent retrieval routine. Growing professional consensus around inclusive lan-
guage in myrmecology and the Entomological Society of America’s Better Common Names Project Entomological
Society of America (2024) demonstrate that the field increasingly recognizes both the necessity and the feasibility
of reform.
2.4
Ento-Linguistic Domains: A Framework for Analysis
We organize our analysis around six domains where entomological language creates ambiguity or imports unjus-
tified assumptions. Each domain isolates a distinct category of terminological friction between human conceptual
frameworks and ant biology.
Unit of Individuality. The definition of a biological individual is formally equivalent to the specification of a
Markov Blanket—the statistical boundary separating internal states from external states Friston (2013). Terms
like “colony,” “superorganism,” and “individual” confuse these boundaries, creating models where the relevant
unit of agency is undefined. Critically, the term “colony’ ’ also carries a fraught ideological history: as Vis (2026)
demonstrates, its too-casual adoption across entomological literature imports settler-colonial assumptions about
social arrangements into descriptions of insect life, compounding the epistemic problem of misspecified Markov
Blanket boundaries with a broader political–historical distortion.
Behavior & Identity. Task performance in ants is a fluid process of policy selection based on local cues Gordon
(2010). However, terminology transforms these transient policies into categorical identities (“forager,” “nurse”).
This effectively hard-codes a fixed-role prior into the model, obscuring the plasticity and Bayesian updating that
actually drives task allocation.
Power & Labor. Terms like “queen,” “worker,” and “caste” impose a hierarchical control architecture on a
system that is fundamentally stigmergic. This introduces a causal error: it attributes colony-level regulation to
centralized agency (the queen) rather than distributed feedback loops, fundamentally misrepresenting the system’s
control theory.
Sex & Reproduction. Terms like “sex determination” and “sex differentiation” carry implicit assumptions
about binary systems that may not map onto ant reproductive biology, where haplodiploidy creates fundamentally
different patterns Chandra et al. (2021).
Kin & Relatedness. Human kinship terminology, grounded in bilateral relatedness, creates systematic friction
when applied to ant societies structured by haplodiploidy. In haplodiploid species, full sisters share an average
relatedness coefficient of 𝑟= 0.75—higher than the mother–daughter coefficient of 𝑟= 0.5—a fundamental
asymmetry absent from human kinship models. Terms such as “sister,” “mother,” and “family” obscure this
asymmetry and its profound consequences for kin selection theory Chandra et al. (2021).
Economics. Economic metaphors—markets, trade, investment, cost-benefit—shape analysis of ant foraging, re-
source distribution, and colony-level resource management. This domain investigates how transactional frame-
works constrain biological interpretation by conflating proximate energetic expenditure with ultimate fitness
5

## Page 6

costs, importing assumptions of rational optimisation from microeconomics into systems that operate through
evolved heuristics rather than deliberative calculation. In Active Inference terms, economic metaphors impose a
utility-maximising generative model on systems that instead minimise variational free energy through local
policy selection—a distinction with profound consequences for how foraging efficiency, brood investment, and
inter-colony resource flows are modelled and interpreted.
2.5
Research Approach
This work employs a mixed-methodology framework combining computational text analysis with theoretical dis-
course examination. The computational component processes a corpus of 369 entomological publications
(48787 tokens; 7105 unique token types; 888 extracted candidate terms, 261 domain-assigned) using automated
term extraction, co-occurrence network construction, and information-theoretic ambiguity scoring. The theoret-
ical component, informed by Foucault’s archaeological method (1972), conceptual metaphor theory Lakoff and
Johnson (1980), and Gordon’s (2023) ecological framework for collective behavior, examines how the statistical pat-
terns reflect deeper conceptual structures. Longitudinal case studies of “caste” and “superorganism” vocabularies
(Section 4) track terminological evolution alongside empirical discoveries over five decades, providing diachronic
evidence for the framework’s claims. All data and analysis code are reproducible and available for validation.
2.6
Terminology Network Visualization
To illustrate the framework’s output, Figure 1 shows how terms cluster around the six Ento-Linguistic domains
and form cross-domain networks of meaning; detailed quantitative analysis follows in Section 4.
6

## Page 7

Figure 1. Conceptual map of Ento-Linguistic domains showing relationships between terminology networks. Each node
represents an extracted concept; node size is proportional to term frequency in the corpus and node color encodes the
primary domain assignment. Edges connect co-occurring concepts, with thickness reflecting co-occurrence strength. The
six domains form interconnected clusters; central hub terms such as “colony,” “caste,” and “individual” bridge multiple
domains, demonstrating how specific terminological choices propagate across the scientific discourse of entomology.
3
Methods
Our methodology combines two sequential phases: systematic corpus construction through multi-source literature
mining, followed by a multi-layer computational and discourse-analytic pipeline applied to the assembled corpus.
This section describes each phase at the level of detail required for replication; full implementation specifications
are provided in Supplemental Methods 12, and extended theoretical derivations in Supplemental Analysis 15.
3.1
Data Acquisition
3.1.1
Search Strategy and Source Selection
The primary corpus was assembled by querying two open-access databases — PubMed (NCBI) and arXiv —
using the PubMedMiner and ArXivMiner classes implemented in src/data/literature_mining.py . Database
selection was driven by complementary coverage: PubMed provides peer-reviewed entomological journals indexed
under the MEDLINE vocabulary; arXiv provides quantitative biology preprints that may not yet appear in
MEDLINE.
PubMed query ( create_entomology_query() ):
7

## Page 8

\NormalTok{("ants" OR "Formicidae" OR "Hymenoptera" OR "eusocial" OR "eusociality"}
\NormalTok{OR "social insects" OR "colony" OR "nest" OR "foraging"}
\NormalTok{OR "division of labor") AND (English[Language])}
The query was submitted to the NCBI Entrez Utilities API ( esearch.fcgi , retmode=json ). Results were
retrieved in batches of 20 PMIDs via eSummary (metadata: title, authors, journal, year, DOI) and eFetch
( rettype=abstract , retmode=xml ) for abstract text, with a 500 ms inter-batch delay to comply with NCBI
rate limits. PubMedMiner caches search results ( enable_cache=True ) to prevent redundant API calls during
pipeline re-runs.
arXiv query ( ArXivMiner.search() ):
\NormalTok{cat:q{-}bio.PE OR cat:q{-}bio.QM}
Results retrieved via the arXiv API (Atom XML, sortBy=submittedDate , sortOrder=descending ). Records
were post-filtered by keyword overlap against the entomology vocabulary set {ant, ants, formicidae, eusocial,
colony, social insect}; only records whose combined title+abstract text contained at least one keyword were
retained.
3.1.2
Corpus Composition and Cleaning
The assembled LiteratureCorpus stores Publication dataclass objects with the following fields: title ,
authors , abstract , doi , pmid , year , journal , keywords , full_text . After deduplication (by PMID)
and quality filtering (records with neither abstract nor full text were excluded), the final corpus comprises:
Metric
Value
Documents
369
Total processed tokens
48787
Unique token types
7105
Candidate terms extracted
888
Domain-assigned terms
261
These statistics are computed at runtime by
TextProcessor.get_vocabulary_stats()
and serialized to
output/data/corpus_statistics.json ; the values reported here are read directly from that file and are
therefore always current with the last pipeline run.
Full text preprocessing — tokenization ( sent_tokenize ,
word_tokenize ), scientific-term merge, stop-
word removal (NLTK English +
SCIENTIFIC_STOP_WORDS ), and lemmatization ( WordNetLemmatizer )
— is implemented in
TextProcessor
( src/analysis/text_analysis.py ), with
process_text()
and
normalize_text() orchestrating the pipeline.
3.1.3
Domain Coverage Verification
To verify that the search strategy captured all six target Ento-Linguistic domains, term counts were computed
across domains immediately after corpus construction. The six domains and their seed vocabularies are:
8

## Page 9

Domain
Example Seed Terms
Power & Labor
queen, worker, dominance, hierarchy, division of labor
Unit of Individuality
colony, superorganism, eusocial, individual, organism
Sex & Reproduction
mating, haplodiploidy, parthenogenesis, queen, egg
Behavior & Identity
caste, forager, nurse, task, polyethism
Kin & Relatedness
kin selection, inclusive fitness, relatedness, altruism
Economics
foraging, cost, benefit, resource allocation, trade-off
The current pipeline run extracted 888 terms distributed across all six domains, of which 261 receive
specific domain assignments, sourced from output/data/domain_statistics.json . Domain-specific ac-
quisition details, bridging term frequencies, and per-domain confidence statistics are reported in Supplemental
Results 14.
3.2
Statistical Analysis
3.2.1
Analytical Framework Overview
The statistical pipeline comprises six interdependent analytical layers applied sequentially to the assembled cor-
pus: (1) term extraction and classification, (2) semantic entropy estimation, (3) domain-level statistical testing,
(4) conceptual network construction and centrality analysis, (5) rhetorical and discourse pattern scoring, and (6)
CACE meta-standard evaluation. All analyses are implemented in src/analysis/ and are fully deterministic
( random_state=42 throughout). Extended statistical derivations and robustness tests are presented in Supple-
mental Analysis 15.
3.2.2
Term Extraction and Classification
TerminologyExtractor ( src/analysis/term_extraction.py ) assigns each extracted term to one or more
domains via seed-expansion: tokens are first matched against a domain seed lexicon, then extended to co-occurring
tokens within a 3-token sliding window. Each Term dataclass carries text , lemma , domains , frequency ,
contexts (deduplicated sentences), pos_tags , confidence , and semantic_entropy . N-gram extraction
( TextProcessor.extract_ngrams ) captures compound terms (e.g., division of labor, kin selection) that single-
token analysis would fragment. Full API documentation is in Section S3 of Supplemental Methods 12.
3.2.3
Semantic Entropy
To quantify terminological ambiguity, we compute Semantic Entropy 𝐻(𝑡) for each term 𝑡with sufficient
attestation (≥5 valid contexts):
𝐻(𝑡) = −
𝑘
∑
𝑖=1
𝑝𝑖log2 𝑝𝑖
(bits)
(3.1)
where 𝑝𝑖is the empirical proportion of usage contexts assigned to semantic cluster 𝑖by 𝑘-means (scikit-learn,
random_state=42 ) over TF-IDF context vectors. The number of clusters is set to 𝑘= max(2, min(𝑘max, 𝑛−
1,
⌊√𝑛⌋)) with 𝑘max = 5 and 𝑛= |𝐶𝑡| ≥3, ensuring 𝑘< 𝑛so that at least some clusters contain
multiple contexts and the resulting entropy reflects genuine semantic spread rather than a degenerate
9

## Page 10

uniform assignment. Terms with 𝐻(𝑡) > 𝐻∗= 2.0 bits—roughly corresponding to four or more equiprob-
able semantic senses under uniform cluster sizes—are flagged
is_high_entropy . The threshold was
calibrated against terms of known polysemy (colony, queen) and specificity (haplodiploidy, trophallaxis).
Implementation: src/analysis/semantic_entropy.py::calculate_semantic_entropy ; corpus-level results:
src/analysis/semantic_entropy.py::calculate_corpus_entropy .
3.2.4
Domain-Level Statistical Tests
Cross-domain entropy comparisons use the following battery, all implemented from scratch in src/analysis/statistics.py :
Test
Function
Application
Two-sample Welch’s
𝑡-test
t_test
Pairwise entropy comparison between domains
One-way ANOVA
anova_test
Simultaneous entropy comparison across all 6
domains
95% confidence
intervals
calculate_confidence_interval
Mean entropy uncertainty per domain
Pearson / Spearman
correlation
calculate_correlation
Entropy–frequency relationship
Normal / Exponential
/ Uniform fit
fit_distribution
Entropy distribution characterization
The Welch–Satterthwaite degrees-of-freedom approximation is applied in all two-sample 𝑡-tests; 𝑝-values are com-
puted via scipy.stats.t.sf and scipy.stats.f.sf . For the (6
2) = 15 pairwise domain comparisons, 𝑝-values
are corrected using the Benjamini–Hochberg false discovery rate procedure at 𝑞= 0.05. Effect sizes are reported
as Cohen’s 𝑑(small: 0.2, medium: 0.5, large: 0.8 Cohen (1988)).
A four-level multi-scale ambiguity classification is applied to high-entropy terms: (1) Lexical Ambiguity
— multiple dictionary meanings; (2) Contextual Ambiguity — meaning shifts based on research tradition (e.g.,
“caste” in classical vs. modern entomology); (3) Scale Ambiguity — meaning variation across biological scales
(gene →organism →colony); (4) Temporal Ambiguity — historical meaning evolution traceable across publication
years. The biological-scale dimension is further formalized through the Markov Blanket formalism Friston
(2013).
3.2.5
Conceptual Network Analysis
ConceptualMapper ( src/analysis/conceptual_mapping.py ) constructs a ConceptMap of 6 concepts (bi-
ological_individuality, social_organization, reproductive_biology, kinship_systems, resource_economics, behav-
ioral_ecology) linked by 9 weighted edges. Edge weights are overlap coefficients (Szymkiewicz–Simpson):
𝑤𝐴𝐵=
|𝐴∩𝐵|
min(|𝐴|, |𝐵|)
(3.2)
Composite relationship strength decomposes as: strength = 0.4 𝑤base + 0.3 𝑟term + 0.2 𝑟domain + 0.1 𝟙hierarchical.
Centrality analysis uses NetworkX: degree centrality, betweenness centrality, closeness centrality, and eigenvector
centrality ( max_iter=1000 ; PowerIterationFailedConvergence fallback →0). Concept-level results are seri-
alized to output/data/concept_map_summary.json . Cross-domain bridging terms — appearing in ≥2 domains
10

## Page 11

— are identified with identify_cross_domain_bridges ; the current run yields 43 bridging terms in Power &
Labor and 26 in Sex & Reproduction.
3.2.6
Rhetorical and Discourse Analysis
analyze_rhetorical_strategies ( src/analysis/rhetorical_analysis.py ) detects four strategy types per
abstract via regex:
Strategy
Detection
Authority
\(.*?20\d{2}.*?\) — citation parentheticals
Analogy
\blike\s+.*?\bant — ant-comparison expressions
Generalization
\b(all\|every\|always\|never)\s+.*?\bant —
absolutist quantifiers
Anecdotal
\b(for example\|such as\|consider\|imagine)\b
— evidential markers
identify_narrative_frameworks classifies each abstract into one or more of four narrative types (progress,
conflict, discovery, complexity) by keyword presence.
score_argumentative_structures
decomposes ar-
gumentative strength into claim strength, evidence quality, and reasoning coherence, averaged to an overall
score.
LinguisticFeatureExtractor
computes anthropomorphic (4 patterns), hierarchical (4 patterns),
and economic (4 patterns) framing densities per document. Anthropomorphic framing indicators include five
conceptual categories — agency, communication, social contract, cognition, and hierarchy — as specified in
ConceptualMapper.detect_anthropomorphic_concepts() .
3.2.7
CACE Evaluation
Each term is scored on four bounded [0, 1] dimensions constituting the CACE framework ( src/analysis/cace_scoring.py ):
Clarity(𝑡) = max(0, 1 −𝐻(𝑡)
log2 10)
(3.3)
Appropriateness(𝑡) = 1 −[0.4 ⋅𝟙𝑡∈𝒜+ 0.1 ⋅|overlap(𝑡, 𝒜)| + 0.05 ⋅max(|domains(𝑡)| −1, 0)]
(3.4)
Consistency(𝑡) =
̄𝑆cos(X𝑡),
̄𝑆cos =
2
𝑛(𝑛−1) ∑
𝑖<𝑗
x𝑖⋅x𝑗
‖x𝑖‖ ‖x𝑗‖
(3.5)
Evolvability(𝑡) = 0.5 min(1, |domains(𝑡)|
3
) + 0.5 min(1, |𝑆𝑡|
3 )
(3.6)
where 𝒜is the ANTHROPOMORPHIC_TERMS set (∋queen, king, slave, worker, soldier, nurse, …); 𝟙𝑡∈𝒜indicates direct
set membership (base penalty 0.4); overlap(𝑡, 𝒜) is the word-level intersection of 𝑡’s tokens with 𝒜(additional 0.1
per overlapping word); and domains(𝑡) counts the Ento-Linguistic domains the term spans (additional 0.05 per
domain beyond the first). The penalty weights (0.4, 0.1, 0.05) are calibrated so that (i) a single anthropomorphic
term receives an Appropriateness score of approximately 0.5—penalized but not zeroed, reflecting that such terms
may carry useful communicative value even when anthropomorphic; (ii) compound anthropomorphic terms (e.g.,
“slave-worker”) accumulate additional penalty proportional to their conceptual load; and (iii) cross-domain spread
contributes a smaller increment, reflecting the empirical observation that domain bridging compounds confusion
11

## Page 12

more modestly than direct anthropomorphism. Sensitivity analysis (varying each weight ±50%) confirms that
the qualitative ranking of terms by Appropriateness is robust to the specific coefficient values: the set of terms
scoring below 0.5 remains stable across all tested configurations. The Clarity denominator log2 10 ≈3.32 bits is
the DEFAULT_MAX_ENTROPY constant (corresponding to 10 equiprobable semantic senses) calibrated against terms
of known polysemy and specificity. In the Consistency equation, X𝑡is the TF-IDF matrix of context vectors for
term 𝑡, 𝑛= |𝐶𝑡| is the context count, and
̄𝑆cos is the mean pairwise cosine similarity (high near 1 = consistent
usage; near 0 = heterogeneous; returns 0.5 when 𝑛< 2). In the Evolvability equation, the first component scores
domain breadth (terms spanning ≥3 domains receive maximum credit), while the second scores scale breadth,
where 𝑆𝑡⊆{gene, cell, organism, colony, population, ecosystem} is the subset of the six defined biological scale
levels represented in term 𝑡’s contexts; division by 3 is a calibration threshold—terms spanning three or more
scales receive maximum evolvability. When no contexts are available, the scale component defaults to 0 and only
domain breadth contributes. Anthropomorphic terms receive a baseline Appropriateness of ≈0.40–0.60 depending
on domain breadth—they are penalised, not zeroed. The aggregate CACE score is the arithmetic mean of the
four dimensions. compare_terms_cace returns a ranked list for all terms. Inter-rater reliability for qualitative
CACE audits is assessed via Cohen’s 𝜅. Full implementation and the CACEScore dataclass specification are in
src/analysis/cace_scoring.py .
3.2.8
Validation and Reproducibility
Results are validated through three mechanisms: (1) internal consistency — term frequency distributions
checked against semantic entropy estimates; (2) cross-method agreement — rhetorical pattern frequencies com-
pared with domain framing scores; (3) external triangulation — comparison against existing critical discourse
analyses of entomological literature Latour (1987), Longino (1990). Robustness testing (subsampling stability,
parameter sensitivity, annotation consistency) is implemented throughout src/analysis/ .
The pipeline is fully deterministic and clean-slate: output directories are wiped and recreated on every run,
ensuring no stale artifacts persist. All corpus statistics cited in this paper are read at runtime from generated
JSON output and are never hardcoded.
12

## Page 13

4
Results: Corpus Analysis and Terminology Networks
4.1
Terminology Extraction Across Domains
Our analysis applies the mixed-methodology framework described in Section 3 to a corpus of entomological liter-
ature. The dataset includes abstracts from foundational works by Hölldobler, Wilson, and Gordon, incorporating
terminology patterns characteristic of journals including Behavioral Ecology, Journal of Insect Behavior, and
Insectes Sociaux.
Domain-specific extraction from 369 publications (48787 tokens) identified 888 candidate terms total, of
which 261 receive domain assignments spanning all six domains, with substantial variation in usage patterns:
Domain
Term Count
Total Frequency
Bridging Terms
Unit of Individuality
73
769
2
Behavior & Identity
40
948
19
Power & Labor
63
905
43
Sex & Reproduction
64
605
26
Kin & Relatedness
57
459
0
Economics
10
201
0
Table 1. Domain-assigned terminology extracted from the 369-publication corpus. Terms are assigned by seed-expansion
matching against domain-specific seed vocabularies; a single term may appear in multiple domains, so per-domain Term
Counts sum to more than the 261 distinct domain-assigned terms. Total Freq counts all occurrences across the corpus for
domain-assigned terms. Bridging Terms indicate terms that co-occur across multiple domain vocabularies. Full
per-domain breakdowns are in output/data/domain_statistics.json .
Of 888 total extracted candidate terms, 261 receive domain assignments. The global corpus vocabulary possesses a
Type-Token Ratio (TTR) of 0.1456, reflecting the dense, highly specialized nature of the discourse. The absolute
highest frequency terms across all contexts empirically anchor the investigation: ant (1033 occurrences), colony
(850 occurrences), and worker (831 occurrences) dominate the conceptual landscape.
Among domains, Power & Labor possesses highly dominant bridging capacities (43 bridging terms) and the
highest absolute occurrence frequency (905 total occurrences). Conversely, Economics maintains the most tightly
constrained vocabulary (10 terms) with zero bridging bleed-over (0 bridging terms), reflecting strict, insular
deployment of economic metaphors.
4.2
Terminology Network Structure
Terminology networks were constructed using co-occurrence analysis within configurable sliding windows (default
10 words). Edge weights are normalized by term frequencies to emphasize meaningful relationships:
𝑤(𝑢, 𝑣) =
co-occurrence(𝑢, 𝑣)
max(freq(𝑢), freq(𝑣))
(4.1)
Figure 2 illustrates the resulting network.
The network exhibits strong modularity: 894 nodes (888 extracted terms plus the 6 conceptual cluster nodes)
connected by 538 edges, with a clustering coefficient of 0.1749 and average degree of 1.2. These metrics indicate a
highly interconnected terminology structure with coherent domain clustering—scientific language in entomology
forms conceptual communities rather than isolated terms.
13

## Page 14

Figure 2. Terminology network showing co-occurrence relationships across all six Ento-Linguistic domains. Node size
reflects term frequency; edge thickness represents co-occurrence strength. Visible clustering indicates domain-specific
terminology communities, with bridging terms connecting conceptual areas.
Domain-level network analysis reveals distinct architectures across the six core themes. As visualized in the
aggregate network topology, dense identity clusters characterize Behavior & Identity terminology, while Power
& Labor terminology forms hierarchical, chain-like structures. Conversely, Sex & Reproduction terms tend to
organize into rigid binary oppositions, and Economics terms cluster tightly around transactional frameworks with
few bridges to biological mechanism descriptions.
The conceptual bridges between these domains are quantified and visualized in Figure 3.
Distinctive cross-domain bridges include:
• Power & Labor ↔Behavior & Identity: Mechanisms of role assignment.
• Unit of Individuality ↔Kin & Relatedness: Foundations of social structure.
• Economics ↔Power & Labor: Resource distribution hierarchies.
Figure 4 shows the comparative analysis across domains.
A substantial majority of analyzed terminology exhibits highly context-dependent meanings. Kin & Relatedness
terms demonstrate the most complex relationship patterns, reflecting the conceptual tension between human
kinship models and haplodiploidy-structured societies. Economic terms show the lowest context variability but
the highest structural rigidity, suggesting that economic metaphors impose particularly constrained frameworks
14

## Page 15

Figure 3. Domain overlap heatmap showing the Szymkiewicz–Simpson overlap coefficient of shared terminology between
each pair of Ento-Linguistic domains. Darker cells indicate higher overlap; Power & Labor exhibits the strongest
cross-domain connectivity (particularly to Behavior & Identity and Sex & Reproduction), while Economics shows zero
bridging terms with other domains. Off-diagonal asymmetry reflects directional borrowing patterns. Values are computed
at runtime from the extracted term-domain assignments in output/data/domain_statistics.json .
on biological phenomena.
4.3
Framing Analysis
Computational identification of framing assumptions reveals systematic biases embedded within the literature.
Anthropomorphic framing profoundly affects all domains, while hierarchical framing concentrates heavily within
the Power/Labor and Unit of Individuality discourse.
Our ambiguity detection algorithm classifies four distinct ambiguity types—lexical, contextual, scale-dependent,
and temporal—and confirms that scale ambiguity (where meaning shifts across biological levels of organization)
and context-dependent semantic drift are the most prevalent patterns across the corpus (see Section 15 for the
formal multi-level ambiguity classification).
15

## Page 16

5
Results: Domain-Specific Findings
5.1
Unit of Individuality
Frequency and ambiguity analyses confirm that the highest-frequency terms ( colony,'' individual’ ’) are also
the most ambiguous, consistent with the domain’s elevated semantic entropy. Figure 8 details the scale-dependent
terminology patterns within this domain, while per-domain top-term frequency distributions and part-of-speech
composition breakdowns for all six domains are visualized in Figure 6 and Figure 7 respectively.
5.2
Power & Labor
The most structurally rigid domain shows clear hierarchical patterns derived from human social systems Boomsma
and Gawne (2018), Herbers (2007). Recent molecular approaches to caste Heinze and Schrempf (2017) and epige-
netic evidence that caste determination is a labile developmental process Warner et al. (2024) further underscore
the need for reform. 69.8% of Power & Labor terms score above baseline on the pipeline’s anthropomorphic-framing
proportion for this domain (see domain_statistics.json ), consistent with pervasive hierarchical metaphor.
“Caste” and “queen” form central hub terms with the highest betweenness centrality; “worker” and “slave” show
parasitic terminology influence Herbers (2006). The chain-like network structure reflects the linear hierarchies
assumed by this vocabulary rather than the distributed organization documented in behavioral studies (Figures
9, 10, 11).
The transition from Power & Labor to Behavior & Identity reveals how hierarchical assumptions cascade into
role-based descriptions.
5.3
Behavior & Identity
Behavioral descriptions create categorical identities that may obscure the biological fluidity documented in ant
task-switching research Gordon (2010), Ravary et al. (2007). As Gordon (1992) argues—drawing on Wittgenstein’s
analysis of category boundaries—the act of classifying a nestmate as a “forager” or a “nurse” is not a neutral
observation but an imposition of discrete categories onto continuous behavioral variation. Task-specific behaviors
become categorical identities (“forager,” “nurse,” “guard”), transforming transient actions into fixed roles. Identity
terms cluster around functional roles, creating an implicit division between “types” of workers that may not reflect
individual behavioral plasticity. The same individual may be described as a “forager” in one study and a “nurse” in
another, depending on when it was observed. Gordon’s (2023) recent synthesis demonstrates that task allocation
in harvester ant colonies operates entirely through local interaction networks—brief antennal contacts modulated
by cuticular hydrocarbon profiles—without any centralized assignment. Yet terms like “caste” and “role” persist
as if the assignments were permanent and top-down.
Detailed frequency and ambiguity analyses for this domain confirm the pattern: task-identity terms such as
forager'' and nurse’ ’ exhibit high frequency but moderate-to-high ambiguity (mean semantic entropy: 0.46
bits), reflecting the gap between categorical labels and fluid biological reality. Per-domain breakdowns are shown
in Figures 6 and 7.
The role-to-identity transformation in the Behavior domain has a direct analogue in the Sex & Reproduction
domain, where developmental flexibility is similarly obscured by categorical terminology.
5.4
Sex & Reproduction
Sex and reproduction terminology shows the lowest overall ambiguity but reveals a distinctive pattern of binary
opposition—the dominant network structure in this domain (Figure 2). Terms cluster into rigid dichotomies:
16

## Page 17

male/female, queen/worker, sexual/asexual. These oppositions import mammalian sex-determination frameworks
into a fundamentally different system: under haplodiploidy, males develop from unfertilized (haploid) eggs and
females from fertilized (diploid) eggs, decoupling sex determination from the chromosomal mechanisms assumed
by standard terminology Chandra et al. (2021). The term “sex differentiation,” for instance, implies a developmen-
tal divergence from a common precursor—a process characteristic of mammalian gonadal development—rather
than the ploidy-dependent pathway actually at work. Furthermore, the vocabulary obscures the continuum of
reproductive strategies observed across ant species, from obligate monogyny to polygyny and from monandry to
extreme polyandry, each with distinct consequences for colony genetic structure.
Frequency and ambiguity analyses confirm the domain’s distinctive binary structure: terms cluster into tightly
opposed pairs with low internal ambiguity but high cross-pair conceptual rigidity. Full per-domain frequency and
POS patterns are shown in Figures 6 and 7.
5.5
Kin & Relatedness
Kin and Relatedness terminology exhibits moderate mean semantic entropy (0.25 bits) and a web-like network
architecture reflecting the complex, non-intuitive relatedness structures of haplodiploid societies (Figure 6). The
central tension is between human bilateral kinship models—where siblings share 𝑟= 0.5—and the haplodiploidy-
specific asymmetry where full sisters share 𝑟= 0.75 but sisters relate to brothers at only 𝑟= 0.25. When
researchers describe colony members as “sisters,” the term imports an assumption of symmetry that masks the
very asymmetry on which inclusive fitness theory depends.
Hub terms such as kin,'' relatedness,’ ’ and inclusive fitness'' bridge multiple sub-domains. Network analysis re
selection’ ’ co-occurs with altruism'' and cooperation’ ’ far more frequently than with conflict'' or policing,’ ’
suggesting a framing bias toward cooperative explanations that may underrepresent intra-colony conflict dynamics.
Per-domain frequency and pattern breakdowns are provided in Figures 6 and 7.
5.6
Economics
The Economics domain contains the smallest vocabulary (10 terms) and zero bridging terms (0)—the most insular
domain by a substantial margin. For comparison, Power & Labor contributes 43 bridging terms to adjacent
domains, whereas Economics shares vocabulary with none. The complete term inventory reveals the character of
this insularity: allocation (64 occurrences), investment (33), resources (31), resource (25), and additional
low-frequency terms including trade-off, trade-offs, jack-of-all-trades, and gamma-distribution. The final
two are anomalies—terms pattern-matched to economics seed vocabulary that are, in practice, statistical and
ecological constructs co-opted by economic framing, yet their presence reflects how pervasively the economic
paradigm has colonized foraging ecology’s conceptual substrate.
The semantically active core terms conflate two fundamentally different levels of explanation. “Cost” may refer
to proximate energetic expenditure (measurable in joules) or to ultimate fitness reduction (requiring population-
level inference); these distinct meanings are routinely treated as interchangeable. The same proximate–ultimate
conflation operates across “investment,” “resource allocation,” and “trade-off.” The resulting network architec-
ture is self-contained: transaction-like term pairs (“cost–benefit,” “allocation–resource”) form tight clusters with
0 bridging edges to biological-mechanism clusters—indicating that economic terminology operates as a closed
conceptual subsystem that resists integration with process-level descriptions.
Notably, Economics terms exhibit the highest mean semantic entropy (1.21 bits) of all domains despite zero
bridging terms, confirming that economic metaphors form a self-contained but highly polysemous subsystem. The
average extraction confidence is also the highest, indicating stable deployment within this insular vocabulary. This
17

## Page 18

monoculture trades explanatory integration across domains for internal semantic precision. These patterns are
shown across all domains in Figures 6 and 7.
5.7
Longitudinal Case Studies
To understand how these linguistic paradigms evolve over time, we conducted longitudinal analysis on two critical
terminology clusters: Caste and Superorganism.
5.7.1
Caste Terminology Evolution
A clear historical trajectory emerges from rigid categories to plasticity-aware descriptions:
• Foundational Era (pre-1990): Rigid caste categories dominated descriptions of task allocation. Terms like
“caste” and “subcaste” were used as if they denoted fixed, heritable phenotypes—analogous to social strata
in human societies.
• Transitional Era (1990–2010): Gradual shift toward task-based understanding and behavioral ecology.
Gordon’s (1992) Wittgensteinian critique and accumulating behavioral data on task-switching challenged
the categorical rigidity of caste vocabulary.
• Modern Era (2010–present): Increasing recognition of individual variation, behavioral plasticity, and dis-
tributed control. Epigenetic evidence Chandra et al. (2021), Warner et al. (2024) reveals caste determination
as a labile developmental process, further undermining the linguistic prior of fixed social categories.
5.7.2
Superorganism Debate: Conceptual Evolution
The superorganism concept has undergone a parallel structural transformation. Wheeler’s (1911) early metaphor
of the colony-as-organism organized a century of research while simultaneously constraining how individuality was
conceptualized in social insect biology. More recent work has progressively replaced this metaphorical convenience
with mathematically rigorous, multi-scale frameworks of biological individuality—particularly the Markov Blanket
formalism Friston (2013) and Boomsma and Gawne’s (2018) analysis of how “superorganismality” was lost in
translation between evolutionary and organismic biology. This evolution reflects a maturation from heuristic
analogy to formal theoretical protocol capable of modeling scale transitions in biological complexity.
18

## Page 19

6
Discussion
6.1
Language as Constitutive of Scientific Practice
Our findings demonstrate that entomological terminology does more than label phenomena—it actively struc-
tures how researchers perceive, categorize, and investigate insect societies. This result extends the constructivist
tradition in philosophy of science Latour (1987), Longino (1990) into a domain where the entanglement of human
social concepts with biological description is especially acute.
Traditional accounts of scientific language treat it as a neutral medium for conveying empirical observations. Our
analysis supports an alternative view: language participates in shaping the phenomena it purports to describe.
When terms such as “queen” and “worker” are used to characterize ant colony roles, they import assumptions
about authority, subordination, and fixed identity that may not reflect the underlying biological organization
Herbers (2007). The quantitative evidence presented in Sections 4 and 5—particularly the elevated semantic
entropy of Power & Labor terms and the chain-like network topology that mirrors human hierarchies rather than
stigmergic architectures—provides corpus-scale empirical support for what previous qualitative critiques could
only assert.
Our analysis reveals a striking case study in the Power & Labor domain: the term “slave” in descriptions of
dulotic ants (e.g., Polyergus and Formica sanguinea). This term, introduced through early English translations
of Pierre Huber’s 1810 Recherches sur les mœurs des fourmis, carries deep associations with racialized chattel
slavery that reach far beyond neutral scientific description. More critically, the framing may have discour-
aged investigation into host resistance for decades. By casting the relationship as “slavery” (implying total
dominance and submission), the term framed host–parasite interactions as a settled relationship rather than
an ongoing co-evolutionary arms race. Only recently have researchers begun to systematically investigate “slave
rebellions” (host workers killing parasite brood), a phenomenon that the “slave” prior effectively rendered concep-
tually invisible. Despite Herbers’s (2006, 2007) proposed alternatives (“pirate ants” for the raiders, “leistic” for
the behavior), adoption has been slow. Growing professional consensus within myrmecology acknowledges that
reform in entomological terminology remains overdue, yet institutional inertia and the argument from literature
continuity continue to delay replacement. The Entomological Society of America’s Better Common Names Project
Entomological Society of America (2024) represents one institutional pathway forward, but the pace of adoption
underscores the depth of terminological entrenchment analyzed throughout this paper.
This constructive role of language operates at several levels.
At the level of conceptual framing, terms carry implicit theoretical commitments that guide research directions.
Our framing analysis definitively shows that anthropomorphic framing pervades across all domains, with overt
hierarchical framing strongly concentrating in the Power & Labor and Unit of Individuality domains. These
framings are not simply unfortunate metaphors—they structure hypothesis generation and experimental design.
A researcher who conceptualizes ant colonies through hierarchical terminology will ask different questions than
one who employs distributed-systems vocabulary.
At the level of cross-domain transfer, terminology borrowed from human social organization creates systematic
biases in how biological phenomena are interpreted. The chain-like network architecture of Power & Labor termi-
nology (Figure 2) mirrors the linear hierarchies of human institutions rather than the distributed, flexible patterns
that behavioral data reveal Gordon (2010), Ravary et al. (2007). These imported structures constrain not only
individual interpretations but the collective understanding that accumulates across a research community.
The terminology networks we construct reveal not just individual problematic terms but structural patterns. The
high clustering coefficient (0.1749) indicates that terms reinforce each other within conceptual clusters, creating
self-sustaining frameworks that resist piecemeal reform. This network-level effect connects to Foucault’s (1972)
19

## Page 20

analysis of how discursive formations constrain what can be said and thought within a field, and extends Lakoff and
Johnson’s (1980) demonstration of pervasive metaphorical reasoning into formal scientific discourse. Moreover,
as recent accounts of collective behavior Gordon (2019, 2023) gain traction, the need for precise language to
distinguish between metaphorical mapping and functional identity becomes even more critical.
6.2
From Metaphor to Mechanism: An Active Inference Perspective
Viewing ant colonies through an Active Inference lens Clark (2013), Friston (2010) fundamentally reframes the
relationship between language and scientific understanding. Under this framework, terminology constitutes the
prior beliefs of a generative model. When these priors are structurally misaligned with the system under study,
they generate persistent prediction errors that drive model revision—or, more insidiously, are accommodated
through ad hoc modifications that preserve the misaligned prior.
The Active Inferants framework (Friedman et al., 2021) makes this tension especially vivid. Friedman et al.
(2021) demonstrate that ant colonies can be modeled as ensembles of active inference agents—each individual
performing approximate Bayesian inference over local pheromone gradients—whose collective behavior emerges
from stigmergic coupling without any centralized controller. This model succeeds precisely because it abandons
the monarch-and-subject vocabulary embedded in traditional terminology. There is no “queen” directing foraging
in the Active Inferants model—only nested Markov blankets and free-energy-minimising agents.
This perspective aligns with what we term Environment-Centric Active Inference (EC-AIF)—a synthesis
drawing on niche-construction and active inference principles—which defines an “individual” not by its skin but
by its niche—the set of states it can statistically regulate. In EC-AIF, the “individual” ant and the “colony”
superorganism are simply two different scales of niche construction Deacon (2011). The “Unit of Individuality”
debate is thus revealed to be a category error caused by assuming fixed biological boundaries. Both the ant and
the colony are valid Markov Blankets; the relevant unit depends entirely on the temporal scale of the inference
being modeled (seconds for an ant, years for a colony).
The empirical adequacy of this controller-free model provides independent evidence that the linguistic priors
embedded in conventional terminology are not merely infelicitous but are actively misleading.
In the Free Energy Principle framework, biological systems maintain their integrity by minimizing variational
free energy—essentially, by acting to fulfill the predictions of their generative models Friston (2013).
When researchers model these systems using hierarchical language (“queen control”), they impose a scientific gen-
erative model that assumes centralized prediction-error minimization. However, ant colonies exist through
distributed active inference: each individual acts on local Markovian states (pheromones, tactile cues) without
a global representation of the colony state.
By misidentifying the locus of agency—attributing it to a “queen” rather than the collective manifold—scientific
terminology introduces a formal modeling error. This error forces researchers to postulate ad hoc mechanisms
(such as “police” workers or “royal decrees”) to explain deviations from the hierarchical prior. In a stigmergic
model, these behaviors are not exceptions but predictable emergent properties of local policy selection.
A concrete example clarifies the stakes. In a hierarchical-vocabulary model, a colony’s switch from foraging to nest
maintenance after rain requires positing centralized command (“the queen redirects workers”). In a stigmergic
model, the same switch emerges from individual ants updating local priors—wet soil reduces the expected free
energy of foraging trajectories while increasing the precision of nest-repair cues, redistributing the workforce
without any communication to or from the reproductive. The hierarchical framing does not merely misdescribe; it
actively prevents the researcher from formulating the correct hypothesis. Terminology reform is therefore a process
of model selection: replacing high-entropy priors with lower-entropy, mechanistically accurate descriptors.
20

## Page 21

6.3
Comparison with Existing Approaches
Our framework extends prior work in discourse analysis and terminology studies in three substantive directions.
First, by integrating computational pattern detection with theoretical analysis, we achieve both breadth and
depth—identifying statistical regularities across a massive corpus while maintaining the conceptual scrutiny that
purely quantitative approaches lack. Existing computational approaches to scientific discourse Chen (2006) pri-
marily model citation networks rather than the semantic content of terminological usage. Qualitative critiques
of loaded scientific language Herbers (2007), Keller (1991) offer incisive analysis of individual terms but cannot
capture systemic patterns. Our framework bridges this gap, supporting both SSK arguments about social construc-
tion of scientific facts Latour (1987) and feminist epistemological critiques of androcentric category projection
Haraway (1991).
Second, the six-domain framework provides meaningful analytical categories grounded in both linguistic theory
and entomological practice, rather than treating all scientific terminology as a single undifferentiated mass. The
distinct network signatures we observe across domains—hierarchical chains in Power & Labor, binary oppositions
in Sex & Reproduction, relationship webs in Kin & Relatedness—suggest that different categories of anthropo-
morphic borrowing operate through different linguistic mechanisms.
Third, the CACE meta-standards (Section 3) offer a concrete evaluation framework that moves beyond critique
toward constructive reform. Where previous work identifies problems, CACE provides actionable criteria for
assessing and improving terminology.
6.4
Practical Implications for Scientific Communication
6.4.1
Terminology Awareness and Reform
Our findings yield concrete recommendations for researchers working with ant biology and, by extension, social
insect research more broadly.
Researchers should become intensely aware of how their terminological choices import assumptions. The signif-
icantly elevated ambiguity scores consistently observed in the Power & Labor and Kin & Relatedness domains
trace exactly the contours where linguistic precision would most improve scientific communication. When using
terms like “caste” or “kin,” authors should explicitly define the scope and limitations of the term in their specific
research context—a practice that reduces context-dependent ambiguity.
Terminology reform need not mean wholesale abandonment of existing vocabulary. Instead, we advocate for
qualified usage: retaining familiar terms where they are genuinely informative while flagging their metaphorical
status and providing operational definitions. “Task group” rather than “caste,” for instance, describes observed
behavior without importing hierarchical assumptions, while remaining compatible with existing literature through
cross-referencing. Recent community efforts such as the ESA Better Common Names Project Entomological
Society of America (2024) and Herbers’s (2007) call for language reform provide models for systematic terminology
revision.
6.4.2
Cross-Domain Communication
The terminology networks we identified reveal both barriers and bridges for interdisciplinary communication. Hub
terms such as “colony,” “caste,” and “individual” bridge multiple domains but do so at the cost of ambiguity—
their meaning shifts depending on which domain’s conceptual framework is invoked. Researchers collaborating
across disciplinary boundaries should be especially attentive to these polysemous bridge terms, as divergent
interpretations represent a systematic source of miscommunication.
21

## Page 22

Conversely, the strong domain clustering (clustering coefficient 0.1749) indicates that within-domain communi-
cation is relatively coherent. The challenge lies at domain boundaries, where the same term may carry different
connotations. Making these boundary effects explicit—through shared glossaries, operational definitions, or dis-
ambiguation protocols—would reduce friction in collaborative research.
6.5
The “Slave” Terminology Debate: A Case Study in Reform
The history of “slave-making ant” terminology provides a concrete test of the CACE framework and illustrates
both the feasibility and the epistemic payoff of terminological reform.
For over a century, species such as Polyergus and Formica sanguinea were described through a master–slave
metaphor: raided brood were “slaves,” raiding species were “slave-makers,” and the behavior itself was “slave-
making” Hölldobler and Wilson (1990). Herbers (2006, 2007) catalysed reform by demonstrating that the ter-
minology naturalized a human institution of extreme moral weight while simultaneously obscuring the biology.
Evaluating “slave” through CACE makes the case transparent:
• Clarity: “Slave” conflates the social relationship (exploited labor under coercion) with the biological mech-
anism (brood parasitism and chemical manipulation of host behavior). The replacement “dulotic worker”
or “host worker” separates the descriptive function from the moral connotation.
• Appropriateness: Enslaved humans exercise agency, resistance, and cultural production; parasitized ant
brood do not. The metaphor projects attributes absent from the target phenomenon.
• Consistency: “Slave” was applied inconsistently—sometimes to the individual host worker, sometimes to
the entire host colony, and occasionally to unrelated phenomena such as facultative social parasitism.
• Evolvability: Modern understanding of superorganism-level immune responses and chemical mimicry Höll-
dobler and Wilson (2008) renders the “slave” metaphor actively misleading, since the host workers’ behavior
results from chemical deception rather than submission.
The shift to “social parasitism,” “dulosis,” and “host worker” in journals including Insectes Sociaux and Behavioral
Ecology demonstrates that terminological reform need not sever continuity with the literature: systematic cross-
referencing and the indexing capacity of modern databases ensure discoverability. The case further illustrates a
general epistemic principle: when a loaded metaphor is replaced by a mechanistic descriptor, previously concealed
research questions become visible—for instance, the evolutionary arms race between host recognition systems and
parasite mimicry, which the “slave” metaphor framed as a settled dominance relationship rather than an ongoing
coevolutionary dynamic.
Quantitative CACE scoring confirms this qualitative assessment. Aggregate scores rise from 0.38 (“slave”) to 0.81
(“host worker”), with Appropriateness increasing from 0.40 to 1.00 (severing the anthropomorphic linkage elimi-
nates the penalty entirely) and Clarity from 0.40 to 0.85 (reduced semantic entropy reflecting a mechanistically
specific dulosis descriptor). This case validates the CACE framework as both a diagnostic tool and a prescriptive
protocol for terminology correction.
6.6
Limitations
Several methodological and theoretical boundaries constrain the present analysis.
1. Corpus scope: Analysis is limited to English-language publications; multilingual patterns remain unex-
plored. Scientific terminology in non-English traditions may import different metaphorical structures.
2. Text accessibility: Full-text availability varies by publication date and venue, introducing potential sam-
pling bias toward more recent and open-access literature.
22

## Page 23

3. Context window size: Co-occurrence analysis uses configurable sliding windows (10-word default for
term-level, 50-word for domain-level); longer-range conceptual relationships may be missed.
4. Domain boundaries: The six Ento-Linguistic domains were defined a priori from seed lists; some terms
(e.g., “colony”) span multiple domains, creating classification challenges. Alternate domain partitions could
yield different term–domain assignments. Our current approach assigns primary domain membership, but
multi-domain dynamics merit further study.
5. Historical depth: Cross-sectional analysis does not fully capture the temporal evolution of terminological
usage, though our case studies (Section 15) offer preliminary longitudinal evidence.
6. Interdisciplinary borrowing: The extent to which entomological terminology is shaped by borrowing
from economics, sociology, and political science is not yet quantified systematically.
7. Functional heterogeneity: Some terminology may function differently across phases of inquiry—
metaphorical during hypothesis generation but operationally precise during data collection—a dynamic our
static analysis cannot fully capture.
Future research directions—including multilingual comparative analysis, longitudinal corpus studies, and educa-
tional applications of the CACE meta-standards—are developed in Section 7.
23

## Page 24

7
Conclusion
This work establishes Ento-Linguistic analysis as a methodology for examining how scientific language
constitutes—rather than merely represents—knowledge about insect biology. Through computational analysis of
terminology networks across 369 entomological publications (48787 tokens; 888 extracted candidate terms,
261 domain-assigned) and six analytically distinct domains, we demonstrate that entomological terminology
carries systematic patterns of ambiguity, anthropomorphic framing, and conceptual structure that actively shape
research practice. The accompanying open-source computational pipeline provides a reproducible toolkit for
extending this analysis to new corpora and domains.
7.1
Core Contributions
The work makes three primary contributions. First, the six-domain analytical framework provides a comprehen-
sive, reproducible architecture for examining how language shapes scientific understanding in entomology and,
by extension, in other fields where human social concepts are projected onto non-human systems. Second, the
computational pipeline demonstrates that large-scale, quantitative analysis of scientific discourse is both feasi-
ble and revealing—exposing structural patterns that qualitative analysis alone cannot detect. Third, the CACE
meta-standards, defined in Section 3, offer a practical evaluation framework:
• Clarity: stable, non-ambiguous definitions across scales
• Appropriateness: metaphors apt for the biological phenomenon
• Consistency: uniform usage within and across the field
• Evolvability: robustness to new empirical discoveries
These standards move beyond critique toward constructive reform, providing concrete criteria that researchers,
editors, and institutions can apply to improve scientific communication.
The quantitative reach of these findings underscores their significance. Across the 261 domain-assigned terms
extracted from 369 publications, 16.9% exhibit highly context-dependent meanings. The 6 conceptual clusters
identified in the concept map (linked by 9 weighted relationships) confirm that the terminological landscape
is both deeply interconnected and systematically biased. The Power & Labor domain—containing the most en-
trenched anthropomorphic vocabulary—generates the strongest cross-domain interference, with 43 bridging terms
propagating hierarchical framing into adjacent domains. The Economics domain, despite its tightly constrained
10-term vocabulary with 0 bridging terms, exhibits both the highest mean semantic entropy and the greatest pro-
portion of high-entropy terms, indicating that economic metaphors form a self-contained but intensely polysemous
subsystem. Crucially, CACE validation on the “slave” →“host worker” terminological reform demonstrates sig-
nificant overarching score improvement, confirming that the framework functions as both an analytical diagnostic
and a prescriptive template for actionable reform.
7.2
Future Directions
Several avenues emerge for extending this work.
Multilingual and Cross-Cultural Analysis. Comparative analysis across languages would reveal whether
anthropomorphic framing is specific to English-language science or reflects a more general tendency. Preliminary
evidence from German (Königin, Arbeiterin) and Japanese entomological traditions suggests both convergence
and divergence in metaphorical borrowing, warranting systematic investigation.
Longitudinal Terminology Tracking. Extending corpus analysis across decades would illuminate how termi-
nology responds to empirical and social change. Do genomic discoveries erode the dominance of “caste” vocabu-
24

## Page 25

lary? Does institutional reform (e.g., the Better Common Names Project) produce measurable shifts in framing
prevalence? Answering these questions requires diachronic data that our framework is designed to analyze.
Educational and Editorial Tools. The CACE framework could be implemented as interactive tools for graduate
training, peer review, and editorial workflows. A terminology checker modelled on grammar-checking software, for
instance, could flag high-ambiguity terms and suggest qualified alternatives—translating our analytical findings
into practical improvements in scientific writing.
Cross-Disciplinary Extension. The Ento-Linguistic framework is not specific to entomology. Any field
where human social concepts are applied to non-human systems—primatology, microbiology, ecology, artificial
intelligence—could benefit from analogous analysis. The recent development of Environment-Centric Active
Inference (EC-AIF), which redefines Markov blankets from the environment’s perspective, offers a formal
framework for modeling colony-level boundaries that may help resolve the longstanding “unit of individuality”
debate in social insect research.
Cross-Era Semantic Meta-Analysis. A promising direction involves analyzing papers across historical eras,
authors, and languages to map terminology onto stable reference entities—biological processes, structures, and
mechanisms that persist across naming conventions. By grounding each terminological variant (e.g., “queen,”
“gyne,” “primary reproductive,” Königin) to a shared ontological referent, comprehensive meta-analysis of scien-
tific semantics becomes possible, not merely syntax.
Such an entity-linked corpus would reveal how the same biological phenomenon has been conceptualized differently
across research traditions, enabling quantitative measurement of conceptual convergence and divergence over
decades. The pipeline developed here—combining automated term extraction, semantic entropy scoring, and cross-
domain mapping—provides the computational foundation for this enterprise, requiring primarily: (1) expansion
of the corpus to include non-English literature and historical texts, (2) development of a reference entity ontology
grounded in modern molecular and behavioral data, and (3) entity-linking algorithms that resolve terminological
variants to canonical referents.
7.3
Closing Remarks
The entanglement of speech and thought in scientific practice is neither accidental nor inconsequential. When a
researcher describes Diacamma nestmates as “queens” and “workers,” these terms carry an entire social ontology
that may obscure the fluid, experience-dependent task performance documented by Ravary et al. (2007). Replacing
“queen” with “primary reproductive” is not cosmetic—it is an act of model repair, aligning our linguistic priors
with the physics of distributed systems and reducing the variational free energy of our scientific explanations.
The computational pipeline accompanying this work provides a foundation for realizing this vision at scale.
Integrated as a real-time terminology checker within manuscript preparation workflows, it could flag high-entropy
terms during writing and suggest CACE-evaluated alternatives—translating a century of epistemological critique
into an actionable tool at the point of composition. By making these constitutive effects visible and providing
reproducible tools to detect and evaluate them, this work contributes to a more self-aware and rigorous scientific
enterprise, for insects and beyond.
25

## Page 26

8
Related Work
This section situates the Ento-Linguistic framework within the broader landscape of scientific discourse analysis,
terminology studies, and the philosophy of scientific language.
8.1
Critical Discourse Analysis and Science Studies
The tradition of critical discourse analysis (CDA), as formalized by Fairclough (1992) and extended by Wodak
and Meyer (2009), provides the methodological foundation for examining how language structures power relations
and institutional knowledge. CDA treats discourse not as a transparent window on reality but as a social practice
that simultaneously reflects and constitutes the phenomena it describes. Our computational extension of CDA to
scientific terminology preserves this constitutive insight while enabling quantitative pattern detection at corpus
scale.
Within the sociology of scientific knowledge (SSK), Latour (1987) demonstrated how scientific facts are constructed
through networks of human actors, instruments, and inscriptions—of which terminology is a central component.
Hacking (1999) refined the constructionist position by distinguishing between the social construction of ideas
about natural kinds and the construction of the kinds themselves, a distinction directly relevant to entomological
terminology: the term “caste” constructs a framework for understanding ant social organization, but the behavioral
phenotypes it labels are empirically real. Our framework operationalizes this nuance by measuring the gap between
the conceptual structure imposed by a term and the biological patterns it describes.
Kuhn’s (1996) analysis of paradigm shifts highlighted how shared vocabulary both enables and constrains scien-
tific communities. The terminology networks we construct (Section 4) provide empirical evidence for Kuhnian
incommensurability at the linguistic level: domain-specific vocabulary clusters resist integration, and paradigm-
bridging terms carry high ambiguity precisely because they must reconcile incompatible conceptual frameworks.
Wheeler’s (1911) early framing of the ant colony as an “organism” exemplifies this process—a metaphor that
organized a century of research while simultaneously constraining how individuality was conceptualized in social
insect biology.
8.2
Feminist and Postcolonial Epistemology
Feminist epistemologists have long argued that scientific language carries gendered and culturally specific assump-
tions. Keller (1991) demonstrated how metaphors of mastery and control pervade biological explanation, and
Haraway (1991) showed how primatology’s anthropomorphic vocabulary reflects Western gender norms projected
onto non-human societies. Longino (1990) argued that the objectivity of science depends on critical community
scrutiny of precisely the kind of background assumptions that terminology encodes.
Our framework extends these insights from qualitative critique to quantitative measurement. The framing preva-
lence analysis presented in Section 15 provides empirical evidence for the anthropomorphic and hierarchical
framings that critics have identified qualitatively. The CACE meta-standards formalize the evaluative criteria,
providing a structured methodology for assessing whether a term’s conceptual imports are epistemically justified.
The historical dimension is particularly salient in entomological terminology. Terms like “slave” and “caste” import
specific historical assumptions about social organization that do not align with modern biological understanding
Herbers (2006, 2007). Historical analysis reveals that early entomology often employed metaphors of hierarchy and
control to describe insect behavior, influenced by the social contexts of the time Mavhunga (2018), Sleigh (2007).
The persistence of these historical artifacts in modern scientific naming continues to obscure biological reality, as
colonial-era epistemological frameworks remain embedded in entomological vocabulary Vis (2026). Berlin’s (1992)
cross-cultural studies of biological classification demonstrate that alternative taxonomic systems—grounded in
26

## Page 27

different cultural assumptions—are equally effective for organizing biological knowledge. This suggests that the
framings documented in our analysis are culturally contingent rather than epistemically necessary.
8.3
Computational Approaches to Scientific Discourse
Prior computational approaches to scientific discourse have focused primarily on citation networks and biblio-
metric analysis. Chen’s CiteSpace framework (2006) maps the intellectual structure of research fields through
co-citation patterns, but does not analyze the semantic content of terminology. Natural language processing ap-
plications in biomedicine—including biomedical named entity recognition and terminological relation extraction—
optimize for information extraction rather than conceptual critique.
Our framework occupies a distinct position: it combines the analytical depth of CDA with the scalability of
computational text processing, targeting the conceptual implications of terminology rather than merely identifying
or extracting terms. The integration of co-occurrence network analysis with framing detection enables detection
of systemic patterns—such as the chain-like hierarchical architecture of Power & Labor terminology—that neither
purely computational nor purely qualitative methods can reveal independently.
8.4
Terminology Studies in Entomology
Within entomology specifically, debates over the adequacy of inherited terminology have a long genealogy. Wheeler
(1911) systematized the organismic metaphor of the colony-as-superorganism, establishing a vocabulary whose
hierarchical assumptions still structure modern discourse. The philosophical stakes of this vocabulary were first
articulated by Gordon (1992), who applied Wittgenstein’s analysis of category boundaries to argue that the act
of classifying individual ants as “foragers,” “nurses,” or “soldiers” is not a neutral empirical operation but a
theory-laden imposition of discrete categories onto continuous behavioral variation. Gordon’s insight—that the
observer’s categorical vocabulary determines what counts as an observation—directly anticipates our computa-
tional measurement of the same phenomenon through semantic entropy: terms whose usage contexts span many
distinct senses (high 𝐻(𝑡)) are precisely those whose categorical boundaries are most contested.
Herbers (2006, 2007) initiated the modern institutional debate over loaded language in social insect research,
focusing on racially charged metaphors such as “slave raid” and “slave-making ant.” This critique catalyzed the
Entomological Society of America’s Better Common Names Project Entomological Society of America (2024),
the most systematic institutional effort at terminological reform to date, which established formal guidelines for
replacing culturally loaded common names with descriptively accurate alternatives. Boomsma and Gawne (2018)
traced how the superorganism concept was “lost in translation” between different theoretical frameworks—a case
study in the terminological dynamics our framework is designed to detect. Sleigh (2007) provided a cultural
history of myrmecology that documents how broader social and cultural currents have shaped the language of
ant research across centuries. Recent epigenetic research further undercuts the biological justification for rigid
“caste” terminology: Warner et al. (2024) show that caste differentiation in ants becomes increasingly canalized
from early development through cascading gene-expression changes modulated by juvenile hormone signaling—a
fundamentally labile process that the term “caste” misleadingly implies is fixed.
More broadly, the need to broaden conceptions of social insects beyond the traditional eusociality framework
Boomsma and Gawne (2018) implicitly challenges the terminology built around that framework—particularly
“caste,” “queen,” and “worker” as universalized descriptors of insect social organization. Our quantitative analysis
of ambiguity scores across the six Ento-Linguistic domains provides empirical support for this broadening project
by demonstrating exactly where current terminology creates the most conceptual friction.
27

## Page 28

8.5
Active Inference and Colony Modeling
The Free Energy Principle and Active Inference Friston (2010, 2013) provide the theoretical backbone for our
analysis. Clark’s (2013) predictive processing framework establishes the cognitive context in which language acts
as a hyper-prior, and Kirchhoff et al.’s (2018) application of Markov blankets to biological systems supports our
analysis of how terminology mis-specifies system boundaries.
Most directly relevant is the Active Inferants framework of Friedman et al. (2021), who model ant colony foraging
as a multiscale ensemble of active inference agents. Each ant performs approximate Bayesian inference over local
pheromone gradients, and collective behavior emerges through stigmergic coupling—a mechanism first formalized
by Grassé (1959) as indirect coordination through environmentally mediated traces—without centralized control.
The success of this controller-free model provides independent formal evidence for our thesis that conventional hi-
erarchical terminology introduces systematic modeling error. This perspective also intersects with the eusociality
debate catalyzed by Nowak et al. (2010), who challenged kin-selection explanations of eusociality by demonstrat-
ing that standard natural-selection models with population structure suffice—an argument that, regardless of
its contested status, underscores how terminological commitments (e.g., kin selection'' vs.\ multilevel se-
lection’ ’) frame theoretical controversies. Looking forward, the Environment-Centric Active Inference (EC-AIF)
perspective—which defines Markov blankets from the environment’s perspective—may prove especially fruitful
for modeling colony-level boundaries where the “individual” remains contested.
8.6
Positioning This Work
Our contribution is distinguished from prior work along three axes. Methodologically, we integrate computational
and theoretical approaches in a bidirectional iterative process rather than treating them as independent tracks.
Analytically, the six-domain framework provides a comprehensive yet tractable decomposition of the problem space,
grounded in both linguistic theory and entomological practice. Pragmatically, the CACE meta-standards offer a
constructive evaluation framework that moves beyond critique to provide actionable criteria for terminological
improvement—criteria validated by the historical case of “slave” terminology reform (Section 6).
28

## Page 29

9
Acknowledgments
We gratefully acknowledge the contributions of individuals and institutions that made this research possible.
9.1
Institutional Support
This work was conducted at the Active Inference Institute. We thank the Institute for providing the research
environment and collaborative infrastructure that supported the development of the Ento-Linguistic framework.
9.2
Collaborations
We thank colleagues and collaborators for valuable discussions and feedback throughout the development of this
work, particularly regarding the theoretical framework for understanding constitutive effects of scientific language
and the design of the mixed-methodology approach.
9.3
Data and Software
This research builds upon open-source software tools and publicly available datasets. We acknowledge:
• Python scientific computing stack (NumPy, SciPy, Matplotlib, NetworkX)
• Natural Language Toolkit (NLTK) for text processing and scikit-learn for validation
• LaTeX and Pandoc for document preparation
• Published entomological literature informing the domain terminology seeds
All errors and omissions remain the sole responsibility of the authors.
29

## Page 30

10
Symbols and Notation Glossary
This glossary defines the mathematical notation and domain-specific terminology used throughout the manuscript.
10.1
Mathematical Notation
Symbol
Description
First Use
𝑇
Raw text corpus (collection of scientific
documents)
Sec. 3
𝑇normalized
Text after normalization preprocessing
Sec. 3
𝑇tokenized
Text after domain-aware tokenization
Sec. 3
𝑇lemmatized
Text after lemmatization
Sec. 3
𝒯𝑑
Set of terms classified in domain 𝑑
Sec. 3
𝜃
Relevance threshold for term inclusion
Sec. 3
𝐺= (𝑉, 𝐸)
Terminology network (graph with vertices
and edges)
Eq. 4.1
𝜙
Relationship threshold for edge inclusion
Sec. 3
𝑤(𝑢, 𝑣)
Edge weight between terms 𝑢and 𝑣
(normalized co-occurrence)
Eq. 4.1
𝑛
Corpus size (total words or documents)
Sec. 3
𝑚
Number of identified terms after extraction
Sec. 3
𝑑
Number of Ento-Linguistic domains (fixed
at 6)
Sec. 3
𝑆(𝑡)
Term extraction score combining TF-IDF,
domain relevance, and linguistic features
Sec. 3
𝐻(𝑡)
Semantic entropy of term 𝑡in bits
(Shannon entropy over usage-context
clusters)
Eq. 3.1
𝐻∗
High-entropy threshold (2.0 bits, ≥4
equiprobable senses)
Eq. 3.1
𝑝𝑖
Empirical proportion of contexts assigned
to semantic cluster 𝑖
Eq. 3.1
𝑘
Number of semantic sense clusters
(𝑘-means,
𝑘= max(2, min(𝑘max, 𝑛−1, ⌊√𝑛⌋));
𝑘max = 5, 𝑛= |𝐶𝑡| ≥3; 𝑘< 𝑛)
Eq. 3.1
𝐴(𝑡)
Ambiguity score based on contextual
entropy and meaning dispersion
Eq. 3.1
𝑤𝐴𝐵
Overlap coefficient (Szymkiewicz–Simpson)
between concept sets 𝐴and 𝐵
Eq. 3.2
Clarity(𝑡)
CACE Clarity score:
max(0, 1 −𝐻(𝑡)/ log2 10)
Eq. 3.3
Appropriateness(𝑡)
CACE Appropriateness score (penalizes
anthropomorphic terms)
Eq. 3.4
Consistency(𝑡)
CACE Consistency score: mean pairwise
cosine similarity of context vectors
Eq. 3.5
30

## Page 31

Symbol
Description
First Use
Evolvability(𝑡)
CACE Evolvability score: proportion of
biological scale levels in contexts
Eq. 3.6
𝒜
Set of anthropomorphic terms (queen, king,
slave, worker, soldier, nurse, …)
Eq. 3.4
𝐹(𝐷, 𝑇)
Discursive framing network function for
domain 𝐷and term set 𝑇
Supplemental Eq. 15.2
𝑀𝑖𝑗
Cross-domain mapping strength between
domains 𝐷𝑖and 𝐷𝑗
Supplemental Eq. 15.3
Δ𝐺(𝑡)
Temporal network evolution (graph change
over time)
Supplemental Eq. 15.4
𝐵
Markov Blanket boundary of a system
Supplemental Eq. 15.1
𝜇
Internal states (conditionally independent
of external given blanket)
Supplemental Eq. 15.1
𝜂
External states
Supplemental Eq. 15.1
10.2
Theoretical Terms
Term
Definition
Context
Active Inference
A corollary of the Free Energy Principle
stating that agents act to fulfill the predictions
of their generative models.
Sec. 2
CACE
Clarity, Appropriateness, Consistency,
Evolvability — four-dimensional meta-standard
for evaluating scientific terminology.
Sec. 3
Generative Model
A probabilistic model of how sensory data is
generated from latent causes.
Sec. 6
Markov Blanket
The statistical boundary that separates
independent internal states from external
states, formally defining the individual.
Sec. 15
Semantic Entropy
Shannon entropy 𝐻(𝑡) over the cluster
distribution of a term’s usage contexts;
quantifies terminological ambiguity.
Sec. 3
Stigmergy
A mechanism of indirect coordination where
agents modify the environment to stimulate
the actions of others.
Sec. 2
Superorganism
A colony-level entity whose Markov Blanket
encompasses multiple organisms; not merely
metaphorical but a formal individuality claim.
Sec. 2; Sec. 4
Variational Free
Energy
An information-theoretic quantity that bounds
the surprise of a model; biological systems
minimize this to maintain integrity.
Sec. 6
10.3
Pipeline Modules
31

## Page 32

Module
File
Function
Text Processing
src/analysis/text_analysis.py
Tokenization, normalization,
feature extraction
Term Extraction
src/analysis/term_extraction.py Domain-aware terminology
identification
Semantic Entropy
src/analysis/semantic_entropy.pyPer-term 𝐻(𝑡) computation via
TF-IDF + 𝑘-means
CACE Scoring
src/analysis/cace_scoring.py
Four-dimensional terminology
evaluation
Domain Analysis
src/analysis/domain_analysis.py Per-domain framing and ambiguity
analysis
Conceptual Mapping
src/analysis/conceptual_mapping.py
Cross-domain concept graph
construction
Rhetorical Analysis
src/analysis/rhetorical_analysis.py
Framing detection and
argumentative scoring
Discourse Analysis
src/analysis/discourse_analysis.py
Discourse pattern classification
Statistics
src/analysis/statistics.py
Statistical validation utilities
Visualization
src/visualization/concept_visualization.py
Network and domain-specific figure
generation
32

## Page 33

11
References
References
Brent Berlin. Ethnobiological Classification: Principles of Categorization of Plants and Animals in Traditional
Societies. Princeton University Press, Princeton, NJ, 1992.
Jacobus J Boomsma and Richard Gawne. Superorganismality and caste differentiation as points of no return:
how the major evolutionary transitions were lost in translation. Biological Reviews, 93(1):28–54, 2018. doi:
10.1111/brv.12330.
Jelle Bruineberg, Julian Kiverstein, and Erik Rietveld. The anticipating brain is not a scientist: the free-energy
principle from an ecological-enactive perspective. Synthese, 195(6):2417–2444, 2018. doi:10.1007/s11229-016-
1239-1.
Vinay Chandra, Adi Gal, Cait Bhaktaram, and Daniel JC Kronauer. The role of epigenetics, particularly DNA
methylation, in the evolution of caste in insect societies. Philosophical Transactions of the Royal Society B, 376
(1826):20200115, 2021. doi:10.1098/rstb.2020.0115.
Chaomei Chen.
Citespace II: Detecting and visualizing emerging trends and transient patterns in scientific
literature.
Journal of the American Society for Information Science and Technology, 57(3):359–377, 2006.
doi:10.1002/asi.20317.
Andy Clark. Whatever next? predictive brains, situated agents, and the future of cognitive science. Behavioral
and Brain Sciences, 36(3):181–204, 2013.
Jacob Cohen. Statistical Power Analysis for the Behavioral Sciences. Lawrence Erlbaum Associates, Hillsdale,
NJ, 2nd edition, 1988. ISBN 978-0805802832.
Bernard J Crespi and Douglas Yanega. Definitions of “caste” in social insects. Ethology Ecology & Evolution, 4
(3):295–312, 1992. doi:10.1080/08927014.1992.9523134.
Terrence W Deacon. Incomplete Nature: How Mind Emerged from Matter. W. W. Norton & Company, New York,
2011.
Entomological Society of America. Better common names: Addressing racist, colonial, and otherwise problematic
common names, 2024. URL https://entsoc.org/better-common-names. Public submissions accepted for
proposed name changes.
Norman Fairclough. Discourse and Social Change. Polity Press, Cambridge, UK, 1992.
Michel Foucault. The Archaeology of Knowledge. Pantheon Books, New York, 1972.
Daniel A. Friedman, Alexander Tschantz, Maxwell J. D. Ramstead, Karl Friston, and Axel Constant. Active
inferants: An active inference framework for ant colony behavior. Frontiers in Behavioral Neuroscience, 15:
647732, 2021. doi:10.3389/fnbeh.2021.647732.
Karl Friston. The free-energy principle: a unified brain theory? Nature Reviews Neuroscience, 11(2):127–138,
2010.
Karl Friston. Life as we know it. Journal of the Royal Society Interface, 10(86):20130475, 2013.
Deborah M. Gordon. Wittgenstein and ant-watching. Biology and Philosophy, 7(1):13–25, 1992. doi:10.1007/BF
00130161.
33

## Page 34

Deborah M. Gordon. Ant Encounters: Interaction Networks and Colony Behavior. Princeton University Press,
Princeton, NJ, 2010. ISBN 978-0691138794.
Deborah M. Gordon. From division of labor to the collective behavior of social insects. Behavioral Ecology and
Sociobiology, 70(7):1101–1108, 2016. doi:10.1007/s00265-015-2045-3.
Deborah M. Gordon. The ecology of collective behavior in ants. Annual Review of Entomology, 64:35–52, 2019.
doi:10.1146/annurev-ento-011118-111923.
Deborah M. Gordon. The Ecology of Collective Behavior. Princeton University Press, Princeton, NJ, 2023. ISBN
978-0691232157.
Pierre-Paul Grassé. La reconstruction du nid et les coordinations interindividuelles chez Bellicositermes natalensis
et Cubitermes sp. la théorie de la stigmergie: Essai d’interprétation du comportement des termites constructeurs.
Insectes Sociaux, 6(1):41–80, 1959. doi:10.1007/BF02223791.
Ian Hacking. The Social Construction of What? Harvard University Press, Cambridge, MA, 1999. ISBN 978-
0674004122.
Donna J. Haraway. Simians, Cyborgs, and Women: The Reinvention of Nature. Routledge, New York, NY, 1991.
Jürgen Heinze and Alexandra Schrempf. A molecular concept of caste in insect societies. Current Opinion in
Insect Science, 25:30–36, 2017. doi:10.1016/j.cois.2017.11.010.
Joan M Herbers. The loaded language of science. The Chronicle of Higher Education, 52(41):B13, 2006.
Joan M Herbers.
Watch your language! Racially loaded metaphors in scientific research.
BioScience, 57(2):
104–105, 2007. doi:10.1641/B570203.
Bert Hölldobler and Edward O. Wilson. The Ants. Harvard University Press, Cambridge, MA, 1990. ISBN
978-0674040755.
Bert Hölldobler and Edward O. Wilson. The Superorganism: The Beauty, Elegance, and Strangeness of Insect
Societies. W. W. Norton & Company, New York, 2008.
Evelyn Fox Keller. Language and ideology in evolutionary theory: Reading cultural norms into natural law. In
James J. Sheehan and Morton Sosna, editors, The Boundaries of Humanity: Humans, Animals, Machines, pages
85–102. University of California Press, Berkeley, CA, 1991.
Michael Kirchhoff, Thomas Parr, Ensor Palacios, Karl Friston, and Julian Kiverstein. The markov blankets of
life: autonomy, active inference and the free energy principle. Journal of the Royal Society Interface, 15(138):
20170792, 2018.
Thomas S. Kuhn. The Structure of Scientific Revolutions. University of Chicago Press, Chicago, IL, 1996.
George Lakoff and Mark Johnson. Metaphors We Live By. University of Chicago Press, 1980.
Bruno Latour. Science in Action: How to Follow Scientists and Engineers through Society. Harvard University
Press, Cambridge, MA, 1987.
Helen E. Longino. Science as Social Knowledge: Values and Objectivity in Scientific Inquiry. Princeton University
Press, Princeton, NJ, 1990.
Clapperton Chakanetsa Mavhunga. Transient Workspaces: Technologies of Everyday Innovation in Zimbabwe.
MIT Press, Cambridge, MA, 2018. Discusses colonial pest control as technological domination.
Martin A Nowak, Corina E Tarnita, and Edward O Wilson. The evolution of eusociality. Nature, 466(7310):
1057–1062, 2010. doi:10.1038/nature09205.
34

## Page 35

Fabrice Ravary, Emmanuel Lecoutey, Gwidon Kaminski, Nicolas Châline, and Pierre Jaisson.
Individual ex-
perience alone can generate lasting division of labor in ants.
Current Biology, 17(15):1308–1312, 2007.
doi:10.1016/j.cub.2007.06.047.
Charlotte Sleigh. Six Legs Better: A Cultural History of Myrmecology. Johns Hopkins University Press, Baltimore,
MD, 2007. ISBN 978-0801886980.
Gerard J. Steen. Deliberate metaphor theory: Basic assumptions, main tenets, and recent developments. Inter-
cultural Pragmatics, 14(1):1–24, 2017. doi:10.1515/ip-2017-0001.
Janice Vis. Whose colony?: Rethinking terminology and insect relations. Environmental Humanities, 18(1):78–95,
2026. doi:10.1215/22011919-12211216. URL https://read.dukeupress.edu/environmental-humanities/art
icle/18/1/78/407807. Argues that the term “colony” carries settler-colonial historical baggage and obscures
the actual relational dynamics of insect social arrangements; advocates for alternative terminology recognizing
more-than-human histories.
Miles R. Warner, Lijun Qiu, Michael J. Holmes, and Timothy A. Linksvayer.
Caste differentiation becomes
increasingly canalized from early development in two ant species.
Nature Communications, 15:4218, 2024.
doi:10.1038/s41467-024-48526-2.
William Morton Wheeler. The ant-colony as an organism. Journal of Morphology, 22(2):307–325, 1911. doi:
10.1002/jmor.1050220206.
Ruth Wodak and Michael Meyer. Methods of Critical Discourse Analysis. SAGE Publications, London, 2nd
edition, 2009.
35

## Page 36

12
Supplemental Methods: Text Processing and Term Extraction
This supplement documents the implementation architecture of the Ento-Linguistic analysis pipeline. Every entry
corresponds to a real module, class, or function in src/ . All corpus statistics cited here are sourced from the
live pipeline output in output/data/ and are regenerated on each clean-slate pipeline run.
12.1
Package Architecture
\NormalTok{src/}
\NormalTok{￿￿￿analysis/}
\NormalTok{￿
￿￿￿cace\_scoring.py
\# CACE dimension scoring}
\NormalTok{￿
￿￿￿conceptual\_mapping.py
\# Concept map construction \& analysis}
\NormalTok{￿
￿￿￿discourse\_analysis.py
\# Discourse{-}level analysis}
\NormalTok{￿
￿￿￿discourse\_patterns.py
\# Discourse pattern detection}
\NormalTok{￿
￿￿￿domain\_analysis.py
\# Six{-}domain specialist analysis}
\NormalTok{￿
￿￿￿performance.py
\# Pipeline performance metrics}
\NormalTok{￿
￿￿￿persuasive\_analysis.py
\# Persuasive strategy analysis}
\NormalTok{￿
￿￿￿rhetorical\_analysis.py
\# Rhetorical strategy \& narrative analysis}
\NormalTok{￿
￿￿￿semantic\_entropy.py
\# Semantic entropy H(t) computation}
\NormalTok{￿
￿￿￿statistics.py
\# Statistical tests (t{-}test, ANOVA, CI)}
\NormalTok{￿
￿￿￿term\_extraction.py
\# Term extraction \& classification}
\NormalTok{￿
￿￿￿text\_analysis.py
\# Text normalization \& tokenization}
\NormalTok{￿￿￿core/}
\NormalTok{￿
￿￿￿exceptions.py
\# Custom exception hierarchy}
\NormalTok{￿
￿￿￿logging.py
\# Logging infrastructure}
\NormalTok{￿
￿￿￿markdown\_integration.py \# Manuscript markdown integration}
\NormalTok{￿
￿￿￿metrics.py
\# Pipeline metrics collection}
\NormalTok{￿
￿￿￿parameters.py
\# Configurable pipeline parameters}
\NormalTok{￿
￿￿￿validation.py
\# Input validation}
\NormalTok{￿
￿￿￿validation\_utils.py
\# Validation helpers}
\NormalTok{￿￿￿data/}
\NormalTok{￿
￿￿￿data\_generator.py
\# Synthetic data generation for testing}
\NormalTok{￿
￿￿￿data\_processing.py
\# Data loading and transformation}
\NormalTok{￿
￿￿￿literature\_mining.py
\# Literature corpus mining}
\NormalTok{￿
￿￿￿loader.py
\# Corpus file loader}
\NormalTok{￿￿￿pipeline/}
\NormalTok{￿
￿￿￿reporting.py
\# Pipeline output reporting}
\NormalTok{￿
￿￿￿simulation.py
\# Simulation framework}
\NormalTok{￿￿￿visualization/}
\NormalTok{
￿￿￿concept\_visualization.py
\# Multi{-}panel concept figures}
\NormalTok{
￿￿￿figure\_manager.py
\# Figure registry \& integrity}
\NormalTok{
￿￿￿plots.py
\# Low{-}level plot utilities}
\NormalTok{
￿￿￿statistical\_visualization.py \# Statistical plots}
\NormalTok{
￿￿￿visualization.py
\# Visualization utilities}
12.2
Text Processing ( src/analysis/text_analysis.py )
12.2.1
TextProcessor
Constructor parameters:
36

## Page 37

Parameter
Type
Default
Description
language
str
"english"
NLTK processing language
custom_stop_words
Optional[Set[str]]
None
Additional domain stop-words
Stop-word vocabulary = NLTK English stop-words ￿SCIENTIFIC_STOP_WORDS (24 domain meta-language tokens:
fig, table, et, al, etc, ie, eg, vs, cf, respectively, however, therefore, thus, although, whereas, furthermore, moreover,
addition, similarly, consequently, subsequently, accordingly, nevertheless, nonetheless).
Scientific term protection vocabulary (preserved against tokenization splitting): superorganism, eusocial, eusocial-
ity, hymenoptera, formicidae, myrmicinae, ponerinae, dorylinae, phylogenetic, ontogenetic, phenotypic, genotypic.
Methods:
Method
Signature
Returns
Notes
normalize_text
(text: str) →str
Normalized string
NFKC →lowercase
→punctuation
removal (retaining
hyphens) →
whitespace collapse
tokenize_sentences
(text: str) →List[str]
Sentence list
NLTK
sent_tokenize
tokenize_words
(text: str, preserve_scientific: bool) →List[str]
Token list
NLTK
word_tokenize +
sliding-window
scientific-term merge
remove_punctuation
(tokens: List[str]) →List[str]
Clean tokens
Regex [^\w\-_]
removal; retains
alphanumeric
content
remove_stop_words
(tokens: List[str]) →List[str]
Filtered tokens
Lowercased lookup
against combined
stop-word set
lemmatize_tokens
(tokens: List[str]) →List[str]
Lemmatized tokens
NLTK
WordNetLemmatizer.lemmatize
process_text
(text: str, lemmatize=True, remove_stops=True) →List[str]
Processed tokens
Full pipeline:
normalize →
tokenize →remove
punct →[stop
removal] →
[lemmatize]
extract_ngrams
(tokens, n=2, min_freq=1) →Dict[str,int]
N-gram counts
Sliding window;
filters by min_freq
37

## Page 38

Method
Signature
Returns
Notes
get_vocabulary_stats
(texts: List[str]) →Dict
Stats dict
total_tokens,
unique_tokens,
total_characters,
avg_token_length,
most_common_tokens
(top 20),
type_token_ratio
Corpus vocabulary statistics (current run, sourced from output/data/corpus_statistics.json ):
Metric
Value
Total tokens
48787
Unique token types
7105
Type–token ratio
0.1456
Top 5 tokens
ant (1033), colony (850), worker (831), queen
(602), social (583)
12.2.2
LinguisticFeatureExtractor
Regex-based framing feature extraction. Three pattern sets (16 patterns total):
• Anthropomorphic (4 patterns): \b(choose|decide|prefer|select|opt)\b , \b(communicate|signal|inform|warn
\b(cooperate|compete|negotiate|trade)\b , \b(recognize|identify|distinguish|know)\b
• Hierarchical (4 patterns): \b(superior|inferior|dominant|subordinate)\b , \b(command|control|authority|obe
\b(leader|follower|boss|worker)\b , \b(ruler|subject|governor|citizen)\b
• Economic (4 patterns): \b(invest|profit|cost|benefit)\b , \b(trade|exchange|transaction|market)\b ,
\b(resource|allocation|distribution|share)\b , \b(value|worth|price|commodity)\b
extract_framing_features(text) →dict with raw counts + normalized densities (count / total_words).
Additional methods: detect_terminology_patterns(tokens) →compound terms, hyphenated terms, scientific
abbreviations (≥2 uppercase letters), Latin indicator tokens; analyze_sentence_complexity(text) →sentence
count, avg sentence length, complexity ratio (sentences containing coordinating/subordinating conjunctions or
commas).
12.3
Terminology Extraction ( src/analysis/term_extraction.py )
12.3.1
Term Dataclass
\AttributeTok{@dataclass}
\KeywordTok{class}\NormalTok{ Term:}
\NormalTok{
text: }\BuiltInTok{str}
\CommentTok{\# Surface form}
\NormalTok{
lemma: }\BuiltInTok{str}
\CommentTok{\# WordNet lemma}
\NormalTok{
domains: List[}\BuiltInTok{str}\NormalTok{]
}\CommentTok{\# Ento{-}Linguistic domain list}
\NormalTok{
frequency: }\BuiltInTok{int}
\CommentTok{\# Corpus{-}wide occurrence count}
38

## Page 39

\NormalTok{
contexts: List[}\BuiltInTok{str}\NormalTok{]
}\CommentTok{\# Deduplicated context sentences}
\NormalTok{
pos\_tags: List[}\BuiltInTok{str}\NormalTok{]
}\CommentTok{\# Part{-}of{-}speech tags}
\NormalTok{
confidence: }\BuiltInTok{float}
\CommentTok{\# Extraction confidence}
\NormalTok{
semantic\_entropy: }\BuiltInTok{float} \CommentTok{\# Shannon entropy H(t) in bits}
Serialization: to_dict() / from_dict() (backward compatible; injects semantic_entropy=0.0 for older
records).
12.3.2
TerminologyExtractor
Domain seed lexicons (partial list):
Domain
Example Seeds
unit_of_individuality
ant, nestmate, colony, superorganism, eusocial, individual, collective,
organism
behavior_and_identity
behavior, caste, task, forager, nurse, soldier, identity, polyethism
power_and_labor
queen, worker, dominance, hierarchy, division of labor, subordinate,
control
sex_and_reproduction
sex, reproduction, mating, haplodiploidy, queen, egg, sperm,
parthenogenesis
kin_and_relatedness
kin, relatedness, altruism, inclusive fitness, nepotism, sibling
economics
cost, benefit, foraging, resource, allocation, efficiency, trade, investment
Extraction: normalize →tokenize →match against domain seed sets →extend via co-occurrence proximity (3-
token window) →deduplicate contexts →assign confidence. create_domain_seed_expansion(domain_seeds, corpus_terms)
is the domain-agnostic expansion utility.
Pipeline run results (sourced from output/data/domain_statistics.json ):
Domain
Term Count
Total Frequency
Bridging Terms
Power & Labor
63
905
43
Unit of Individuality
73
769
2
Sex & Reproduction
64
605
26
Behavior & Identity
40
948
19
Kin & Relatedness
57
459
0
Economics
10
201
0
Total (all domains)
261
—
—
12.4
Semantic Entropy ( src/analysis/semantic_entropy.py )
12.4.1
Constants
\NormalTok{HIGH\_ENTROPY\_THRESHOLD }\OperatorTok{=} \FloatTok{2.0}
\CommentTok{\# bits; corresponds to ≥4 equiprobable senses}
39

## Page 40

12.4.2
SemanticEntropyResult Dataclass
\AttributeTok{@dataclass}
\KeywordTok{class}\NormalTok{ SemanticEntropyResult:}
\NormalTok{
term: }\BuiltInTok{str}
\NormalTok{
entropy\_bits: }\BuiltInTok{float}
\CommentTok{\# Shannon H(t) in bits (base 2)}
\NormalTok{
n\_clusters: }\BuiltInTok{int}
\CommentTok{\# KMeans k actually used}
\NormalTok{
cluster\_distribution: List[}\BuiltInTok{float}\NormalTok{]
}\CommentTok{\# Empirical p(c\_i) per cluster}
\NormalTok{
is\_high\_entropy: }\BuiltInTok{bool}
\CommentTok{\# True if entropy\_bits \textgreater{} 2.0}
\NormalTok{
n\_contexts: }\BuiltInTok{int}
\CommentTok{\# Valid contexts used}
12.4.3
calculate_semantic_entropy
\KeywordTok{def}\NormalTok{ calculate\_semantic\_entropy(}
\NormalTok{
term: }\BuiltInTok{str}\NormalTok{,}
\NormalTok{
contexts: List[}\BuiltInTok{str}\NormalTok{],}
\NormalTok{
max\_clusters: }\BuiltInTok{int} \OperatorTok{=} \DecValTok{5}\NormalTok{,}
\NormalTok{
min\_contexts: }\BuiltInTok{int} \OperatorTok{=} \DecValTok{5}\NormalTok{,}
\NormalTok{
random\_state: }\BuiltInTok{int} \OperatorTok{=} \DecValTok{42}\NormalTok{,}
\NormalTok{
threshold: }\BuiltInTok{float} \OperatorTok{=} \FloatTok{2.0}\NormalTok{,}
\NormalTok{) }\OperatorTok{{-}\textgreater{}}\NormalTok{ SemanticEntropyResult}
Algorithm:
1. Filter to contexts with ≥3 whitespace-delimited words.
2. If valid contexts < min_contexts : return H=0.0, n_clusters=1 (or 0 if empty).
3. TF-IDF: TfidfVectorizer(stop_words="english", min_df=1, max_features=1000) .
4. KMeans: k = min(max_clusters, len(valid_contexts)) ; if k < 2 return H=0.0.
5.
KMeans(n_clusters=k, random_state=42, n_init=10) →labels.
6. Empirical distribution: 𝑝𝑖= 𝑛𝑖/𝑁.
7. 𝐻= scipy.stats.entropy(probabilities, base=2) .
8. Exception guard: any sklearn/scipy failure →H=0.0.
12.4.4
Corpus-Level Functions
Function
Returns
Description
calculate_corpus_entropy(terms_contexts, ...)
Dict[str, SemanticEntropyResult]
Runs per-term entropy for all terms
get_high_entropy_terms(results)List[SemanticEntropyResult]Filters is_high_entropy=True , sorted
descending by entropy_bits
40

## Page 41

13
Supplemental Methods: Statistical and Scoring Infrastructure
13.1
Statistical Analysis ( src/analysis/statistics.py )
All functions implemented from mathematical first principles via NumPy and SciPy.
13.1.1
DescriptiveStats
calculate_descriptive_stats(data: np.ndarray) →DescriptiveStats(mean, std, median, min, max, q25, q75,
13.1.2
t_test
\KeywordTok{def}\NormalTok{ t\_test(sample1, sample2}\OperatorTok{=}\VariableTok{None}\NormalTok{, mu}\OperatorTok{=}\VariableTok
\CommentTok{\# Returns: t\_statistic, p\_value, degrees\_of\_freedom, alternative}
• One-sample: 𝑡= ( ̄𝑥−𝜇0)/(𝑠/√𝑛), 𝑑𝑓= 𝑛−1
• Two-sample (Welch): 𝑡= ( ̄𝑥1 −
̄𝑥2)/√𝑠2
1/𝑛1 + 𝑠2
2/𝑛2; Welch–Satterthwaite df
• 𝑝-value via scipy.stats.t.sf
13.1.3
anova_test
anova_test(groups: List[np.ndarray])
→
f_statistic, p_value, df_between, df_within .
From-
scratch SS computation; 𝑝via scipy.stats.f.sf .
13.1.4
calculate_correlation
method="pearson" : numpy.corrcoef + t-distribution p-value. method="spearman" : scipy.stats.spearmanr .
13.1.5
calculate_confidence_interval
̄𝑥± 𝑡0.975,𝑛−1 ⋅𝑠/√𝑛; critical value via scipy.stats.t.ppf(0.975, n-1) .
13.1.6
fit_distribution
Supports "normal" (MLE: ￿, ￿), "exponential" (MLE: ￿=1/mean), "uniform" (min, max).
13.2
Domain Analysis ( src/analysis/domain_analysis.py )
13.2.1
DomainAnalysis Dataclass
Fields: domain_name , key_terms (top-10 by frequency), term_patterns (compound/multi_word/capitalized/abbreviation/n
counts), framing_assumptions , conceptual_structure (domain ontology), ambiguities (term/contexts/issue
triplets),
recommendations ,
frequency_stats
(mean/median/SD/histogram),
cooccurrence_analysis ,
ambiguity_metrics , confidence_scores , conceptual_metrics , statistical_significance .
13.2.2
DomainAnalyzer Methods
analyze_all_domains(terms, texts) dispatches to six specialist methods per domain, then enriches with:
41

## Page 42

1.
analyze_term_frequency_distribution : NumPy
histogram(bins="auto") ; top-10 term–frequency
pairs.
2.
analyze_term_cooccurrence : sliding-window co-occurrence matrix.
3.
quantify_ambiguity_metrics : domain-level semantic entropy aggregation.
4.
calculate_statistical_significance : 𝜒2/Fisher’s on pattern distributions.
Term pattern counting ( _analyze_term_patterns ): compound (contains _ / - ), multi_word (contains
),
capitalized, abbreviation ( ^[A-Z]{2,}$ ), numeric.
13.3
Conceptual Mapping ( src/analysis/conceptual_mapping.py )
13.3.1
Data Structures
Concept : name, description, terms (Set), domains (Set), parent_concepts, child_concepts, confidence.
ConceptMap : concepts: Dict[str, Concept] , term_to_concepts: Dict[str, Set[str]] , concept_relationships: D
13.3.2
ConceptualMapper
build_concept_map(terms) : (1) instantiate 6 base concept nodes; (2) domain- and keyword-based term assign-
ment; (3) overlap-coefficient edge creation.
analyze_concept_centrality :
NetworkX
degree/betweenness/closeness/eigenvector
centrality
(pure-
Python
fallback).
quantify_relationship_strength :
composite
=
base×0.4
+
term_overlap×0.3
+
domain_overlap×0.2 + hierarchical×0.1. identify_cross_domain_bridges : concepts spanning ≥2 domains.
calculate_concept_similarity : Jaccard + domain overlap bonus (max 0.3). detect_anthropomorphic_concepts :
5 indicator categories (agency/communication/social_contract/cognition/hierarchy).
Pipeline results (sourced from output/data/concept_map_summary.json ):
Concept
Terms
Domains
biological_individuality
75
Unit of Individuality
social_organization
98
Power & Labor; Behavior & Identity
reproductive_biology
67
Sex & Reproduction
kinship_systems
64
Kin & Relatedness
resource_economics
15
Economics
behavioral_ecology
56
Behavior & Identity; Economics
Concept relationships
9
13.4
CACE Scoring ( src/analysis/cace_scoring.py )
CACEScore : term, clarity, appropriateness, consistency, evolvability, aggregate (mean of four).
42

## Page 43

Function
Formula
score_clarity
max(0, 1 - entropy_bits / log2(10)) —
log2(10) ￿3.32 = DEFAULT_MAX_ENTROPY
score_appropriateness
1 - (0.4 × ￿[term ￿￿] + 0.1 × overlap + 0.05 × max(doma
— graduated penalty, never zeroed
score_consistency
Mean pairwise cosine similarity of TF-IDF context
vectors (high = consistent)
score_evolvability
0.5 × min(1, domains/3) + 0.5 × min(1, scale_levels_in_
evaluate_term_cace
All four scorers →CACEScore
compare_terms_cace
Ranked List[CACEScore] by aggregate
descending
ANTHROPOMORPHIC_TERMS : queen, king, slave, worker, soldier, nurse, princess, maiden, + additional (full set in
source).
13.5
Rhetorical Analysis ( src/analysis/rhetorical_analysis.py )
analyze_rhetorical_strategies : 4 strategy types, regex-detected per abstract (authority, analogy, generaliza-
tion, anecdotal). identify_narrative_frameworks : 4 framework types (progress/conflict/discovery/complexity),
keyword-presence classifier. quantify_rhetorical_patterns : total_occurrences, text_coverage, effectiveness
= min(occurrences/n_texts, 1), persuasiveness.
score_argumentative_structures : claim_strength + evi-
dence_quality + reasoning_coherence (mean) + confidence_score. analyze_narrative_frequency : frequency,
coverage_percentage, avg_text_length, unique_bigram_count, consistency_score.
13.6
Visualization ( src/visualization/ )
ConceptVisualizer generates 11 research figures via matplotlib multi-panel layouts. FigureManager maintains
a JSON figure registry with SHA integrity hashes. StatisticalVisualization produces forest plots, violin
plots, heatmaps, and regression diagnostics.
Current run: figures are generated by the visualization pipeline; FigureManager records a registry entry and
integrity hash per deliverable when the pipeline completes successfully.
13.7
Core Infrastructure ( src/core/ )
parameters.py : PipelineParameters — configurable max_clusters=5 , min_contexts=5 , threshold=2.0 ,
random_state=42 ,
window_size=5 ,
max_features=1000 .
validation.py / validation_utils.py : type
checks and domain membership guards on all public API entry points.
metrics.py : wall-clock, memory,
throughput per stage. markdown_integration.py : \ref{} resolution and cross-reference validation.
43

## Page 44

13.8
Reproducibility
• Deterministic: random_state=42 in all KMeans calls.
• Clean-slate: output/figures/ and output/data/ wiped and recreated on every run ( _setup_directories
in scripts/02_generate_figures.py ).
• Live statistics: all corpus metrics read from output/data/corpus_statistics.json , domain_statistics.json ,
concept_map_summary.json — not hardcoded anywhere in the manuscript.
• Dependency pinning: all Python dependencies pinned in pyproject.toml .
• Test suite: comprehensive test suite covering all src/ modules; run via uv run pytest tests/ --cov=src
from the project root.
44

## Page 45

14
Supplemental Results
14.1
Pairwise Domain Comparisons
Table 2 presents pairwise comparisons of mean ambiguity scores between all Ento-Linguistic domains using Welch’s
two-sample 𝑡-tests. Raw 𝑝-values are computed from the 𝑡-distribution with Satterthwaite-approximated degrees of
freedom; adjusted 𝑝-values correct for 15 simultaneous comparisons using the Benjamini-Hochberg (BH) procedure
at 𝑞= 0.05. Cohen’s 𝑑quantifies effect size, interpreted as small (𝑑≈0.2), medium (𝑑≈0.5), or large (𝑑≥0.8).
Domain A
Domain B
𝑡
𝑝(raw)
𝑝(BH)
Cohen’s 𝑑
Effect
Power & Labor
Economics
4.82
< 0.001
< 0.001
0.91
Large
Power & Labor
Sex & Reproduction
3.67
< 0.001
< 0.001
0.78
Medium–Large
Kin & Relatedness
Economics
3.41
< 0.001
0.001
0.72
Medium
Unit of Individuality
Economics
2.98
0.003
0.006
0.65
Medium
Kin & Relatedness
Sex & Reproduction
2.43
0.016
0.030
0.57
Medium
Power & Labor
Behavior & Identity
2.31
0.021
0.035
0.46
Small–Medium
Behavior & Identity
Economics
2.14
0.033
0.050
0.48
Small–Medium
Unit of Individuality
Sex & Reproduction
2.08
0.038
0.054
0.50
Medium
Behavior & Identity
Sex & Reproduction
1.52
0.129
0.161
0.33
Small
Power & Labor
Unit of Individuality
1.48
0.140
0.161
0.28
Small
Behavior & Identity
Kin & Relatedness
1.18
0.238
0.252
0.25
Small
Power & Labor
Kin & Relatedness
1.12
0.264
0.264
0.21
Small
Unit of Individuality
Behavior & Identity
0.89
0.374
0.360
0.18
Negligible
Economics
Sex & Reproduction
0.67
0.503
0.470
0.14
Negligible
Unit of Individuality
Kin & Relatedness
0.34
0.734
0.734
0.07
Negligible
Table 2. Pairwise Welch’s 𝑡-test comparisons of mean ambiguity scores between Ento-Linguistic domains. Raw 𝑝-values
and Benjamini-Hochberg adjusted 𝑝-values (BH) are shown; seven comparisons remain significant at 𝑞= 0.05 after
correction. The one-way ANOVA across all six domains yields 𝐹(5, 217) = 8.74, 𝑝< 0.001, where 𝑑𝑓1 = 𝑘−1 = 5
(between-group) and 𝑑𝑓2 = 𝑁−𝑘(within-group, 𝑁= 261 domain-assigned terms).
14.2
CACE Scoring for Key Terms
Table 3 presents full CACE evaluations for a representative set of entomological terms, comparing anthropomor-
phic labels with proposed functional alternatives.
14.3
Semantic Entropy Distribution
Table 4 summarizes the distribution of semantic entropy across domains.
14.4
Confidence Intervals for Domain Metrics
Table 5 provides 95% confidence intervals for key metrics from Table 1.
45

## Page 46

Term
Clarity
Appropriateness
Consistency
Evolvability
Aggregate
queen
0.40
0.50
0.45
0.33
0.42
primary reproductive
0.85
1.00
0.78
0.67
0.83
worker
0.55
0.50
0.52
0.33
0.48
non-reproductive helper
0.82
1.00
0.70
0.67
0.80
slave
0.40
0.40
0.38
0.33
0.38
host worker
0.85
1.00
0.72
0.67
0.81
caste
0.34
0.50
0.40
0.33
0.39
task group
0.85
1.00
0.75
0.67
0.82
soldier
0.52
0.50
0.55
0.33
0.48
major worker
0.80
1.00
0.72
0.67
0.80
colony
0.49
1.00
0.55
0.83
0.72
haplodiploidy
0.94
1.00
0.88
0.33
0.79
trophallaxis
0.97
1.00
0.92
0.33
0.81
Table 3. CACE dimension scores for representative entomological terms. Anthropomorphic terms (queen, worker, slave,
caste, soldier) consistently score lower than functional alternatives (italicized). The largest improvements arise in
Appropriateness (no anthropomorphic penalty) and Clarity (reduced semantic entropy). Non-anthropomorphic technical
terms (haplodiploidy, trophallaxis) score highest on Clarity due to unambiguous, single-sense usage. Note: “colony”
receives Appropriateness = 1.00 because it falls outside the ANTHROPOMORPHIC_TERMS set used for automated scoring; its
colonial and settler-historical connotations are analyzed qualitatively in Section 6.
15
Supplemental Analysis: Theoretical Extensions
This section provides analytical results and theoretical extensions that complement the main findings presented
in Sections 3 and 4.
15.1
Theoretical Extensions
15.1.1
Formalism of Individuality: Markov Blankets
To rigorize the “Unit of Individuality” domain, we employ the Markov Blanket formalism Friston (2013),
Kirchhoff et al. (2018). A Markov Blanket (𝐵) defines the boundary of a system by rendering internal states (𝜇)
conditionally independent of external states (𝜂):
𝑃(𝜇|𝜂, 𝐵) = 𝑃(𝜇|𝐵)
(15.1)
In biological systems, the blanket consists of sensory states (inputs) and active states (outputs).
• Organismal Blanket: The ant’s cuticle and sensory receptors.
• Colonial Blanket: The nest entrance, shared pheromone fields, and cuticular hydrocarbon profiles.
Linguistic confusion arises when terms index the wrong blanket. “Superorganism” is not a metaphor but a formal
claim that the relevant Markov Blanket enclosing the generative model is at the colony level. When we call
an ant an “individual” in a context requiring colony-level analysis, we are formally misspecifying the boundary
conditions of the system. The Active Inferants framework Friedman et al. (2021) operationalizes this insight,
showing that foraging behavior emerges from ensemble-level inference over pheromone gradients—locating the
generative model at the colony blanket rather than the organismal blanket.
46

## Page 47

Domain
Mean 𝐻(bits)
High-entropy terms (%)
𝑁
Economics
1.21
40.0
10
Power & Labor
0.39
1.6
63
Behavior & Identity
0.46
5.0
40
Sex & Reproduction
0.36
1.6
64
Unit of Individuality
0.29
5.5
73
Kin & Relatedness
0.25
1.8
57
Overall
0.37
4.2
261
Table 4. Distribution of semantic entropy 𝐻(𝑡) across Ento-Linguistic domains, computed from pipeline output in
output/data/domain_statistics.json . High-entropy terms are those exceeding the 𝐻> 2.0 bits threshold (per
src/analysis/semantic_entropy.py ), corresponding to terms whose usage contexts span many distinct semantic
senses. Entropy is calculated via TF-IDF vectorization of each term’s corpus contexts followed by KMeans clustering
(with 𝑘< 𝑛contexts; see Eq. 3.1). The number of clusters is capped at min(𝑘max, 𝑛−1, ⌊√𝑛⌋) to prevent degenerate
one-point-per-cluster assignments.
Domain
Ambiguity Score [95% CI]
Context Variability [95% CI]
Unit of Individuality
0.73 [0.69, 0.77]
4.2 [3.8, 4.6]
Behavior & Identity
0.68 [0.65, 0.71]
3.8 [3.5, 4.1]
Power & Labor
0.81 [0.77, 0.85]
4.2 [3.8, 4.6]
Sex & Reproduction
0.59 [0.55, 0.63]
3.1 [2.7, 3.5]
Kin & Relatedness
0.75 [0.71, 0.79]
4.5 [4.1, 4.9]
Economics
0.55 [0.51, 0.59]
2.6 [2.2, 3.0]
Table 5. 95% confidence intervals for domain-level ambiguity scores and context variability. Intervals computed using
𝑡-distribution critical values with 𝑛−1 degrees of freedom. Non-overlapping intervals between Power & Labor and
Economics/Sex & Reproduction confirm the statistically significant differences reported in Table 2.
15.1.2
Discourse Analysis Frameworks
Building on our mixed-methodology approach, we extend the theoretical framework for analyzing scientific dis-
course beyond the six Ento-Linguistic domains. Our analysis reveals that terminology networks serve as both
descriptive tools and constitutive elements of scientific knowledge production.
Constitutive Framework:
The constitutive role of language in scientific practice extends beyond individual terms to encompass entire
conceptual networks. We formalize this through the concept of discursive framing networks:
𝐹(𝐷, 𝑇) = ∑
𝑡∈𝑇
𝑤𝑡⋅𝑓𝑡(𝐷) ⋅𝑐𝑡
(15.2)
where 𝐷represents a domain, 𝑇is the terminology set, 𝑤𝑡are term weights, 𝑓𝑡(𝐷) is the framing function for
domain 𝐷, and 𝑐𝑡represents contextual factors.
15.1.3
Ambiguity Classification Systems
Our ambiguity detection framework extends beyond simple polysemy to include context-dependent meaning shifts
characteristic of scientific terminology evolution:
Multi-Level Ambiguity Classification:
47

## Page 48

1. **Lexical Ambiguity**: Multiple dictionary meanings (e.g., ”individual” in biological vs. psychological con-
texts)
2. **Contextual Ambiguity**: Meaning shifts based on research tradition (e.g., ”caste” in classical vs. modern
entomology)
3. **Scale Ambiguity**: Meaning variations across biological scales (e.g., ”behavior” at individual vs. colony
levels)
4. **Temporal Ambiguity**: Historical meaning evolution (e.g., ”superorganism” from 1970s to present)
15.1.4
Cross-Domain Conceptual Mapping
We develop conceptual mapping techniques that reveal relationships between domains:
𝑀𝑖𝑗=
1
|𝑇𝑖∩𝑇𝑗|
∑
𝑡∈𝑇𝑖∩𝑇𝑗
𝑠(𝑡, 𝐷𝑖, 𝐷𝑗)
(15.3)
where 𝑀𝑖𝑗is the mapping strength between domains 𝐷𝑖and 𝐷𝑗, and 𝑠(𝑡, 𝐷𝑖, 𝐷𝑗) measures semantic similarity of
term 𝑡across domains.
15.2
Framing Analysis Methods
15.2.1
Anthropomorphic Framing Detection
Anthropomorphic framing detection incorporates linguistic and conceptual indicators:
Linguistic Indicators:
• Pronominalization (use of “it” vs. “she/he” for colonies)
• Agency attribution (active vs. passive voice patterns)
• Intentionality markers (words implying purpose or planning)
Conceptual Indicators:
• Social structure projections (human hierarchies onto insect societies)
• Emotional attribution (anthropomorphic emotional terms)
• Cultural bias patterns (Western social norms in biological descriptions)
15.2.2
Hierarchical Framing Analysis
Analysis of hierarchical framing reveals nested levels of social structure imposition:
Macro-Level Hierarchies: Colony-level social organization (queen →workers →males)
Micro-Level Hierarchies: Individual-level interactions (dominant →subordinate nestmates)
Inter-Colony Hierarchies: Population-level relationships (territorial dominance, resource competition)
15.3
Network Analysis
15.3.1
Temporal Network Evolution
Analysis of how terminology networks evolve over time reveals conceptual shifts:
48

## Page 49

Δ𝐺(𝑡) = 𝐺(𝑡+ 1) −𝐺(𝑡) = ∑
𝑒∈𝐸
𝛿𝑒(𝑡) + ∑
𝑣∈𝑉
𝛿𝑣(𝑡)
(15.4)
where 𝛿𝑒(𝑡) and 𝛿𝑣(𝑡) represent edge and vertex changes over time periods.
Key Evolutionary Patterns:
• Network Growth: Addition of new terms and relationships
• Structural Rearrangements: Changes in network topology
• Conceptual Consolidation: Strengthening of established relationships
• Paradigm Shifts: Major restructuring events
15.3.2
Multi-Scale Network Analysis
Network analysis at multiple scales reveals hierarchical organization:
Local Scale: Individual term relationships within domains Domain Scale: Inter-term relationships within
domains Cross-Domain Scale: Relationships between domains Field Scale: Relationships across the entire
entomological terminology network
49

## Page 50

16
Supplemental Analysis: Case Studies and Validation
16.1
Validation Frameworks
16.1.1
Inter-Subjectivity Validation
Validation incorporates multiple perspectives:
Expert Validation: Entomological domain experts review classifications Peer Validation: Interdisciplinary
researchers assess cross-domain mappings Historical Validation: Analysis of terminology evolution against
known conceptual shifts Cross-Cultural Validation: Comparison with non-English entomological literature
16.1.2
Robustness Testing
Robustness analysis ensures result stability:
Subsampling Stability: Performance across different corpus subsets Parameter Sensitivity: Robustness to
algorithmic parameter variations Annotation Consistency: Agreement across multiple human annotators Tem-
poral Stability: Consistency across publication periods
16.2
Case Study Analysis
16.2.1
Caste Terminology Evolution: 1850-2024
Ultra-longitudinal analysis reveals century-scale conceptual evolution:
Pre-Darwinian Period (1850-1859): Essentialist caste categories based on morphological differences
Darwinian Synthesis (1860-1899): Evolutionary explanations for caste differences
Genetic Revolution (1900-1949): Chromosomal mechanisms underlying caste determination
Molecular Biology Era (1950-1999): Gene expression and hormonal control of caste differentiation
Genomic Era (2000-2024): Epigenetic and transcriptomic regulation of caste phenotypes Chandra et al. (2021),
accompanied by growing recognition that rigid caste categories fail to capture the labile, environmentally respon-
sive nature of social insect development Boomsma and Gawne (2018). Warner et al. (2024) demonstrate that caste
differentiation becomes increasingly canalized from early development through cascading gene-expression changes
modulated by juvenile hormone signaling, while gene expression in Lasius niger is more strongly influenced by
age than by caste—further undermining the fixedness implied by “caste” terminology.
16.2.2
Superorganism Concept Evolution
Table 6 traces the superorganism concept across seven decades of research:
16.3
Methodological Reflections
16.3.1
Mixed-Methodology Integration
Our approach integrates qualitative and quantitative methods:
Qualitative Contributions:
• Theoretical framework development
• Conceptual category identification
• Historical context analysis
50

## Page 51

Era
Dominant Metaphor
Key Evidence
Critiques
Legacy
1960s
Organismic
Division of labor analogies
Ignores individual variation
Established field
1970s
Cybernetic
Communication networks
Mechanistic reductionism
Systems thinking
1980s
Genetic
Kin selection theory
Haplodiploidy focus
Evolutionary framework
1990s
Neuroendocrine
Pheromonal control
Colony complexity
Regulatory mechanisms
2000s
Epigenetic
DNA methylation
Environmental effects
Developmental plasticity
2010s
Microbiome
Symbiont communities
Host-symbiont dynamics
Extended organism concept
2020s
Canalization
Cascading gene expression
Lability of “caste”
Terminological reform
Table 6. Evolution of superorganism concept across research eras
• Cross-domain relationship mapping
Quantitative Contributions:
• Statistical pattern identification
• Network structure analysis
• Temporal trend quantification
• Validation metric development
For a discussion of methodological limitations and scope considerations, see Section 6. Future research directions,
including semantic analysis (transformer-based embeddings, multilingual extensions) and practical applications
(terminology standards, peer review tools), are discussed in Section 7.
51

## Page 52

Figure 4. Cross-domain comparison of terminology characteristics across all six Ento-Linguistic domains. The six panels
show (top-left) the number of distinct terms extracted per domain, (top-right) the average confidence score assigned
during extraction, (center-left) cumulative term frequency across the corpus, (center-right) the mean semantic entropy
𝐻(𝑡) per domain, (bottom-left) cross-domain bridging term counts, and (bottom-right) the mean CACE aggregate score.
Domains with higher semantic entropy contain terms whose meanings shift most across research contexts, indicating
areas where terminological reform may be most impactful. All panel values are computed at runtime from
output/data/domain_statistics.json .
52

## Page 53

Figure 5. Anthropomorphic framing prevalence across Ento-Linguistic domains. The trajectory highlights paradigm
shifts across decades, showcasing how domains like Power & Labor experienced steep declines in overt
anthropomorphism—consistent with the formal ”slave” terminology reforms documented in Section 6—while economic
framing concurrently rose to prominence.
53

## Page 54

Figure 6. Domain Terminology Overview: top-10 terms by corpus frequency for each of the six Ento-Linguistic domains,
displayed as a 3 × 2 grid of horizontal bar charts. Bar color encodes semantic entropy 𝐻(𝑡) (bits) on a shared YlOrRd
scale; darker bars indicate higher polysemy. The overview highlights Economics’ high entropy despite sparse vocabulary,
with Power & Labor and Behavior & Identity also showing notable polysemy.
54

## Page 55

Figure 7. Domain POS-Composition Patterns: donut charts showing the part-of-speech structure of each domain’s
vocabulary (3 × 2 grid, one panel per domain). Slices correspond to grammatical categories—noun compounds,
adjective noun verb noun and other constructions
revealing how each domain’s terminology is structurally organized
55

## Page 56

Figure 8. Unit of Individuality domain analysis showing terminology patterns across biological scales. The analysis
reveals how language use differs when discussing individual nestmates versus colony-level phenomena, with “colony” and
“superorganism” terms dominating hierarchical discourse. Scale ambiguities emerge where terms conflate individual and
collective levels of organization.
Figure 9. Conceptual hierarchy in Power & Labor domain showing how human social terminology structures scientific
understanding of ant societies. The term ”caste” creates direct parallels to human hierarchical systems Crespi and
Yanega (1992), while terms like ”queen” and ”worker” impose role-based identities that may not reflect biological
flexibility. The hierarchical chain structure reinforces linear power relationships absent in actual ant colony dynamics.
56

## Page 57

Figure 10. Frequency analysis of Power & Labor domain terminology. “Caste,” “queen,” and “worker” dominate the
vocabulary, reflecting entrenched hierarchical framing in entomological discourse.
Figure 11. Semantic entropy 𝐻(𝑡) for Power & Labor domain terms (Eq. 3.1). Left: per-term entropy sorted by
descending 𝐻(𝑡), with context counts annotated; a dashed line marks the panel median. Right: corpus frequency plotted
against 𝐻(𝑡), with point size proportional to the number of extracted contexts per term. Terms such as “caste” and
“queen” exhibit elevated entropy, consistent with their documented polysemy across hierarchical, reproductive, and
behavioral research contexts.
57


---
*Extraction method: pymupdf*