Prevalence of Defensive Speech: A Review of Six Traditions and a Unifying Hypothesis

The problem: why defensive speech has no general number

If one asks what percentage of everyday human communication is defensive — a question natural for a clinician, a communication instructor, or a researcher in social cognition — no answer exists in the literature. There are dozens of empirical works, each measuring something specific in a narrowly defined context and yielding an interpretable percentage for that context. But no single work can answer the general question, because the question itself has no unified referent in the existing conceptual infrastructure.

One cannot simply take Gottman's defensiveness, Blum-Kulka's indirectness, DSQ's immature defenses, Leary's self-presentation, Hyland's hedges, and Gibb's defensive climate, add them up, and call the total "defensive speech." These constructs partially overlap and fundamentally differ: they measure different units (episode, speech act, disposition, strategy, lexeme, climate), in different registers (conflict, request, introspection, public speaking, academic writing, workplace), with different goals (divorce prediction, cross-cultural comparison, psychodiagnostics, social interaction, academic rhetoric, organizational effectiveness).

Yet each tradition produces a numerical signal. Below is a verified compendium of these signals, against which the structure of the gap becomes visible.

Six lines of literature: verified numbers

1. Defensiveness in couple conflict (Gottman-type)

The most clinically well-known operationalization of defensiveness comes from marital conflict research initiated by J. Gottman and developed in the Specific Affect Coding System (SPAFF) and the Rapid Marital Interaction Coding System (RMICS). Defensiveness here is one component of the "high-intensity negative affect" cluster alongside belligerence and contempt, predicting divorce over 7–14 years (Gottman & Levenson, 1992, 2000; Gottman et al., 1998).

Crucially, defensiveness is measured not as a percentage of speech but as presence/intensity in 10–15-minute conflict discussions under laboratory conditions. Stamp's model (Stamp et al., 1992) supplements this line with a process model of defense: defense arises at the intersection of self-perceived flaw, sensitivity, partner attack, and perception of flaw by other. This model is important for Mindloom because it explicitly points to the role of predicted shame as the primary trigger of defensive operations.

Specific percentage estimates ("11.5–20.2% of conflict episodes," "28–34% of episodes") circulating in secondary sources were not found in the primary works of Gottman, Papp, or Heyman in the stated form and appear to be compilation artifacts.

Summary. Gottman's defensiveness is a clinical prognostic marker in a conflict context; using it as a prevalence estimate for defensive speech in everyday communication is methodologically unwarranted.

2. Cross-cultural speech act pragmatics (CCSARP)

The Cross-Cultural Speech Act Realization Project (Blum-Kulka, House & Kasper, 1989) is the most cited program for empirical measurement of indirect speech. Indirect speech here is not "defense" in the clinical sense but a pragmatic strategy for reducing face-threatening load (Brown & Levinson, 1987).

Key numbers. In a comparative study (Cenoz & Valencia, 1996): conventionally indirect — 85.2% of English requests and 72.9% of Spanish ones; hints — 4.8% of English, 1.72% of Spanish. In business English corpus studies (BELF, 415 authentic emails): direct — 71.1%, conventionally indirect — 26%. The distribution strongly depends on register: in spontaneous conversational speech, conventionally indirect dominates (80%+), while in institutional business correspondence, direct forms may prevail.

Summary. CCSARP provides the closest approximation to a "general number" for one speech act (request): in English everyday communication, about 85% of requests are conventionally indirect.

3. Ego defenses: DSQ, DMRS, Perry, Vaillant

The psychodynamic tradition has progressed from Anna Freud through Vaillant (1971, 1976), Bond et al. (1983) to modern DSQ-40/60 and DMRS-SR-30. Inventories of varying size (25–40+ defenses) are organized hierarchically: mature / neurotic / immature (Andrews, Singh & Bond, 1993).

The fundamental characteristic: defenses are measured as personality dispositions via self-report, not as frequencies of observed speech events. No existing instrument — DSQ, DMRS, DMI, Perrotta PDM-Q — answers the question of what percentage of natural speech at any given moment realizes a defensive operation.

Summary. Psychodynamic instruments offer a rich taxonomy of defenses but cannot count defenses in speech. The gap between disposition and speech event is structural.

4. Impression management and self-presentation

This line, launched by E. Goffman (1959) and formalized in the two-component model of Leary & Kowalski (1990), holds that self-presentation motives "underlie and pervade nearly every corner of interpersonal life" — that is, not merely frequent but structurally ubiquitous.

The key paradox: empirical quantification of self-presentation as a proportion of speech acts has never been performed. The literature is rich in taxonomies (ingratiation, self-promotion, intimidation, exemplification, supplication — Jones & Pittman, 1982), but no study has annotated a corpus of natural speech for presence/absence of impression operations.

Summary. Impression management is the most theoretically ambitious line and yet the least empirically quantified. It provides moral support for the >70% hypothesis but no numerical grounding.

5. Hedging and mitigation (corpus linguistics)

Since Lakoff (1972/1973), hedging is the only line with dense corpus measurements. Hyland (1996, 1998) found in a molecular biology corpus (75,000 words) a hedge frequency of approximately one marker per 50 word-forms (~2%), or 33–50% of sentences containing at least one hedge. Corpus studies report: sociology — 148.7 hedges per 10,000 words (Hyland & Jiang, 2016), psychology — 160.7.

Limitations: these numbers are for academic written register, where hedges serve an epistemic function, not necessarily a defensive one in the clinical sense. Hedging is only one family within the five vectors of the Mindloom ontology.

Summary. Hedging provides the only genuine numeric handles — 2% per word-form, 33–50% per sentence — but only for one subfamily of defensive operations in one register.

6. Organizational psychology: communicative climate (Gibb)

J. Gibb (1961) identified six binary pairs of behaviors forming either a defensive or a supportive communicative climate: Evaluation vs Description, Control vs Problem Orientation, Strategy vs Spontaneity, Neutrality vs Empathy, Superiority vs Equality, Certainty vs Provisionalism. The framework is based on an 8-year corpus of audio recordings.

Modern validation (Forward et al., 2011): empathy explains 68% of variance in relational satisfaction; the combination of strategy, neutrality, and problem orientation explains 56% of variance in job satisfaction. Scales show high internal consistency (α = .91 for empathy, .93 for spontaneity).

The key distinction from the previous five: defensiveness is treated not as an individual trait but as a property of the relational field between communicants. However, the Gibb framework also does not produce an estimate of the proportion of defensive speech in overall communication.

Numbers circulating in popular sources about consequences of defensive communication in organizations (40% reduction in information sharing, 37% increase in turnover) trace to a corporate blog without primary sources and are not used here.

Summary. Gibb + Forward provide a verified organizational-psychological operationalization but, like the five preceding lines, do not produce a prevalence estimate for defensive speech.

Diagnosis of fragmentation

Why do six lines yield six incommensurable pictures? Four structural reasons.

Different units of analysis. Defensiveness measures conflict episodes. CCSARP measures individual speech acts. DSQ measures personality dispositions. Impression management measures social self-representations. Hedging measures lexical units. Gibb measures organizational climate. There is no summation operation between them.

Different referents of defensiveness. Gottman: responding to reproach with denial. CCSARP: pre-modifying a proposition. DSQ: unconscious anxiety-reduction mechanism. Goffman: audience orientation. Hyland: lexical marker of caution. Gibb: relational field with perceived threat. The six lines diverge at the level of the object itself.

Different registers. Couple conflict, enacted requests in DCT, introspective questionnaire, laboratory experiment, academic article, workplace. None represents "everyday speech" as a whole.

Absence of a shared ontology. No tradition possesses an ontology in which the defensive units of other traditions would be recognized as special cases. There are partial overlaps (rationalization in DSQ and account in impression management; strategy in Gibb and manipulation in impression management), but no unified typology.

Therefore, the impossibility of giving a "general number" is not a data deficit but an ontological infrastructure deficit.

Mindloom precision-regime architecture as a unifying framework

The framework

Speech behavior is modeled as hierarchical active inference (Friston, 2010; Friston & Frith, 2015) with four precision regimes: DEFENCE (rigid defensive priors under threat), COGNITIVE (balanced data-sensitive updating), ILLUSION (systematic prediction bias with suppression of counter-evidence), META (inference about inference). At a finer observation level, ten behavioral regimes emerge — BUILD, SEEK, UNSEAL, LOCK, SEAL, DRAIN, FLOOD, EDGE, SHIFT, VOID — as attractors in the space of 28 core speech operations, composing into 86 pipeline operations and ~176 Speech Forms.

Defensive regimes are operationalized not as phrase templates but as configurational patterns of precision distribution over core operations.

Neurobiological support: bvFTD and indirect speech comprehension

Hewetson et al. (2020, Frontiers in Human Neuroscience) studied 21 patients with behavioral variant frontotemporal dementia (bvFTD) and 17 amnestic MCI patients as a brain-damaged control group. Participants were presented with dialogues requiring pragmatic bridging inference.

Healthy controls: direct 98.3%, indirect 97.5%. bvFTD patients: selective deficit on indirect (84.76%) with preserved direct (92.86%). 16 of 21 (76.20%) had a negative impairment score. MCI patients showed no such selective deficit.

Significance for Mindloom: indirect speech requires a dedicated computational resource on a specific neural substrate that can selectively degrade. This directly supports the precision-routing mechanism. The authors also state that indirect speech acts are "ubiquitous in daily conversation."

How the six lines map onto the ontology

• Gottman defensiveness → BLOCK-vector operations in LOCK/SEAL regime under high conflict-prior density
• CCSARP conventionally indirect → prep-modified functional speech codes, typically UNSEAL or SEEK with soft defensive modification
• DSQ/DMRS ego defenses → 57 classes of psychological defenses mapped onto five vectors
• Impression management → PULL-vector operations (ingratiation, self-promotion) and PUSH-vector operations (intimidation)
• Hedging → PROTECTIVE category operations (epistemic and deontic mitigation)
• Gibb defensive climate → not a separate operation but a property of the two-agent precision space: climate is "defensive" when both participants hold rigid defensive priors in LOCK/SEAL calibration

The mapping is not equivalence. The Mindloom ontology contains operations absent from all six lines (sacred binding, guilt transfer with imperative force), and conversely, some units of the six lines do not qualify as defensive in the strict sense. The framework refines rather than simply sums.

Hypothesis H1 and falsification protocol

Formulation

H1: In a natural corpus of spontaneous everyday speech by Russian speakers, the proportion of speech units classified by the Mindloom engine as belonging to the set of defensive regimes is no less than 70% of all classified units with a 95% confidence interval.

Cross-cultural moderator

H1a (low-context): In a low-context Anglo-Germanic-Scandinavian speech corpus ≥ 65% with 95% CI.

H1b (high-context): In a high-context Russian-Japanese-Chinese speech corpus ≥ 75% with 95% CI.

Threshold justification

The thresholds are not arbitrary. The general 70% threshold follows from the convergence of four independent estimates:

1. Metabolic argument — compensatory-defensive physiology operates for most of waking time
2. CCSARP convergence — ~85% of requests are conventionally indirect; defensive calibration extends to other speech acts
3. Leary axiom — self-presentation motives pervade nearly every corner of interpersonal life
4. Neuro-expert judgment — Hewetson et al.: indirect speech acts are "ubiquitous in daily conversation"

Their combined overlay yields a 70–85% interval. The 70% threshold is the lower bound, leaving room for conservative falsification.

What would refute H1

• Defensive regime frequency < 65% with properly calibrated engine — full rejection
• Frequency 65–69% — partial rejection, threshold or ontology revision needed
• In informal peer-to-peer below 50% — framework modification required
• High correlation with lexical-feature bias (val-OOD gap ≥ 42–47 pp) — questions the classifier
• H1a > H1b (low-context higher than high-context) — falsifies the cultural moderator

Measurement protocol

1. Corpus collection: ~10,000 utterances of spontaneous Russian speech from held-out sources. Private conversations with consent (anonymized), public interview transcripts, forum conversations. Balanced by gender and age.
2. Dual annotation: (a) Mindloom engine full-hybrid; (b) human annotator for 10% random subset in blind procedure.
3. Agreement metric: Cohen's κ ≥ 0.60 for regime; prevalence correction for inter-rater error.
4. Bootstrap CI (1,000 resamples); comparison against 70% threshold.
5. Pre-registration of hypothesis, threshold, protocol, and stopping rule before analysis (OSF or equivalent platform).

Key sources

Couple conflict: Gottman & Levenson (1992); Gottman et al. (1998); Stamp et al. (1992).

Pragmatics: Blum-Kulka, House & Kasper (1989); Cenoz & Valencia (1996); Brown & Levinson (1987).

Ego defenses: Vaillant, Bond & Vaillant (1986); Andrews, Singh & Bond (1993); Perry (1990).

Impression management: Goffman (1959); Leary & Kowalski (1990); Schlenker & Leary (1982).

Hedging: Lakoff (1973); Hyland (1996, 1998); Hyland & Jiang (2016).

Organizational climate: Gibb (1961); Forward, Czech & Lee (2011).

Neurobiology: Hewetson et al. (2020); Hooley (2007).

Active inference: Friston (2010); Friston & Frith (2015).