Pith. sign in

REVIEW 3 major objections 4 minor 27 references

Minding the Politeness Gap in Cross-cultural Communication

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read British and American English speakers interpret 'quite' differently because of both different literal thresholds and different utterance-cost weights, not politeness norms alone.

desk verdict New UK/US intensifier data worth having, but the model comparison claiming 'semantic plus pragmatic' is not robust once you look at the paper's own BIC column. read the letter →

arxiv 2506.15623 v1 pith:WPUTZP5W submitted 2025-06-18 cs.CL cs.CY

Yuka Machino , Matthias Hofer , Max Siegel , Joshua B. Tenenbaum , Robert D. Hawkins This is my paper · ORCID
classification cs.CLcs.CY
keywords cross-culturalpragmaticsintensifiersBritishEnglishAmericanpolitenesscomputationalcognitivemodelutterancecostsemanticvariation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish where cross-cultural differences in word interpretation come from: are British and American speakers literally assigning different meanings to words like 'quite' and 'very', or are they applying different social norms such as politeness and brevity? After three rating experiments with UK and US participants, the authors build a computational listener that reasons recursively about a speaker who balances informativity, politeness, and utterance cost, and they fit culture-specific parameters for literal thresholds and pragmatic weights. Their central claim is that both levels vary across cultures: the literal threshold for 'quite' differs, and the weight placed on utterance cost and informativity differs, while politeness ratings alone do not explain the interpretation gap. A sympathetic reader should care because this implies that successful cross-cultural communication requires recalibrating not just dictionary meanings but also expectations about why a speaker bothered to modify an utterance at all.

What carries the argument

The central object is a recursive listener model built on a speaker utility $U(w|s,\varphi_i,\varphi_s) = \varphi_i U_i(w|s) + \varphi_s U_s(w) - C(w)$, where the informativity term $U_i$ is the log-likelihood that a literal listener recovers the intended state, $U_s$ is the perceived social appropriateness of the utterance (taken from politeness ratings), and $C(w)$ penalizes adding a modifier. Culture enters through twelve double-threshold parameters fixing the literal denotation of each modifier and the unmodified baseline, plus pragmatic weights $\varphi_i,\varphi_s$ and cost $C(w)$. A pragmatic listener inverts the speaker's softmax production model via Bayes' rule, and the authors compare models allowing different subsets of these parameters to vary between UK and US to see which cultural variation is necessary and sufficient.

What would settle it

Fit the same model to simulated datasets generated from a known model with only threshold differences between two cultures and no cost differences; if the optimizer attributes the difference to cost or informativity weights instead, the semantic/pragmatic split is not identifiable from this design. A simpler version: compute profile likelihood or bootstrap confidence intervals for the threshold and cost parameters and check whether the culture differences survive the uncertainty.

Watch

Extended reading notes

Core claim

In the authors' terms, cross-cultural variation in modifier interpretation arises through both semantic variation and pragmatic variation: British and American participants differ in the literal threshold values they assign to intensifiers, most clearly for 'quite', and they differ in the perceived cost of producing a modifier and the weight placed on being informative. These conclusions come from comparing models in which different subsets of parameters are allowed to vary across cultures: the best-fitting model allows both semantic thresholds and pragmatic weights to differ, and constraining either the utterance cost or the informativity weight to be shared across cultures significantly degrades fit. The model's social-utility term contributed little to fit, and perceived politeness ratings did not linearly explain the cross-cultural interpretation differences, even though politeness ratings did predict interpretation overall. A robustness check shows the model generalizes reasonably to narrator-framed data when the social term is removed, with a moderate loss increase.

Load-bearing premise

The clean separation of semantic from pragmatic causes assumes that the twelve threshold parameters and the pragmatic weights are jointly identifiable from the pooled z-scored ratings, so a fitted threshold difference is not just absorbing what is really a cost difference (or vice versa).

Editorial extensions

If this is right

  • A purely semantic account is insufficient: models that let only literal thresholds vary miss the improvement gained from also letting cost and informativity weights vary.
  • A purely politeness-based account is insufficient: the 'quite' gap persists in narrator-framed utterances where politeness pressure is removed, and politeness ratings do not explain cross-cultural differences.
  • Culture-specific utterance cost means listeners infer different amounts of extra meaning from the same act of modification depending on the culture of the speaker.
  • Practical cross-cultural communication and machine translation should treat intensifiers as carrying both culture-specific denotations and culture-specific pragmatic weights.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the identified cost difference is real, one testable prediction is that UK and US listeners should differ in how much extra strength they infer from the mere presence of a modifier even for novel or invented modifiers with no established semantic difference.
  • The joint estimation of twelve thresholds and three pragmatic weights from z-scored ratings leaves a possible trade-off between threshold shifts and cost shifts; a synthetic recoverability analysis would tell whether the semantic/pragmatic split is an artifact of the optimizer.
  • The paper's own note that a culture-wide social utility parameter may average over heterogeneous subcommunities suggests a hierarchical extension with regional or dialect-level politeness norms could revive the role of politeness in the model.
  • The model's clustering by modifier regardless of predicate, and the outliers for marked expressions like 'extremely exhausted', point toward predicate-specific social meaning rather than a single modifier-level utility as the next modeling step.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. This paper reports three behavioral experiments comparing how British and American English speakers interpret intensifiers such as "quite" and "very," embedded in dialogue, narrator, and politeness-rating contexts. The experiments show robust cross-cultural differences, particularly for "quite" and "very," and show that politeness ratings alone do not explain the interpretation differences. The authors then develop an RSA-style computational model in which culture-specific semantic thresholds and culture-specific pragmatic weights (informativity, social utility, utterance cost) are fitted to the z-scored rating data. The central claim is that cross-cultural differences in interpretation arise from a combination of different literal meanings and different weights on utterance cost/informativity, rather than from politeness norms alone. The behavioral findings are well supported, but the modeling conclusion is weakened by model-selection ambiguity and the absence of identifiability checks.

Significance. The paper addresses a genuinely important question in cross-cultural pragmatics: whether UK-US differences in intensifier interpretation are semantic, pragmatic, or both. The behavioral experiments are a real contribution: they document specific, replicable differences for "quite" and "very," and the narrator manipulation is a sensible control for politeness-related pragmatics. The computational framework extends the RSA politeness model of Yoon et al. (2020) to a cross-cultural setting, and the authors provide code and materials. If the central modeling claim were established, the paper would push the field toward models that jointly incorporate culture-specific lexical semantics and culture-specific utterance costs. However, as it stands, the model-comparison evidence for the "both semantic and pragmatic" conclusion is not robust: the paper's own BIC values favor a simpler semantic-only model, and the parameter separation is not tested for identifiability. These issues are load-bearing for the paper's main claim, but they are addressable with additional analysis.

major comments (3)
  1. [Results / Model comparison (Table 2)] The paper states that the best-fitting model (M9) integrates cross-cultural differences in both semantic and pragmatic factors, but Table 2 does not uniformly support this. M9 has BIC 22631, while M6 (only "quite" threshold varies, no pragmatic parameters) has BIC 22590, and M8 (all thresholds vary, semantic only) has BIC 22656. BIC therefore favors M6 by 41 points over M9, while AIC favors M9 (22448 vs. 22485 for M8). The authors do not justify choosing AIC over BIC, nor do they report cross-validated log loss. Because the central conclusion depends on preferring M9, the paper needs to report a held-out predictive comparison or otherwise justify the model-selection criterion. Without that, the alternative reading under BIC—that only "quite" has a different literal threshold—is equally compatible with the reported numbers.
  2. [Computational Model and Appendix: Optimization Process] The separation of semantic from pragmatic causes assumes that the twelve semantic threshold parameters and the pragmatic weights (φ_i, φ_s, C(w)) are jointly identifiable from the pooled z-scored ratings. No parameter-recovery simulations, confidence intervals, or profile likelihoods are reported. If threshold shifts and cost shifts can trade off, a fitted UK-US difference in thresholds could be partially or wholly an artifact of the optimizer. The manuscript should include synthetic recoverability checks (e.g., simulate data from known parameters and confirm the fitting procedure recovers them) and some measure of parameter uncertainty before attributing differences cleanly to "different literal meanings" versus "different pragmatics." This issue is central to the paper's main claim.
  3. [Appendix: Robustness to dropping modifiers (Table 4)] The robustness check shows that dropping "extremely" changes the log loss from 11250 to 11287 on the original data, which is a larger change than dropping "impressive" or "difficult." The authors acknowledge this sensitivity, but it is not incorporated into the model-comparison conclusion. In addition, Figure 2's "model predictions" are in-sample fits from parameters optimized on the same data, so the visual agreement is not independent evidence. The paper would be substantially strengthened by an out-of-sample evaluation, such as cross-validation by predicate or by participant, to demonstrate that the M9 parameterization generalizes rather than overfitting.
minor comments (4)
  1. [Experiment 3: Politeness Ratings] There is a typo: "supprting" should be "supporting."
  2. [Table 2] The column labeled "Log Loss" is not defined in the text; it appears to be the negative log likelihood, and the relationship between this column and AIC/BIC should be stated explicitly.
  3. [Figure 1 caption] The caption "Experiment 1: Dialogue Experiment 2: Narrator Experiment 3: Politeness" lacks punctuation between panels and would be clearer as "Experiment 1: Dialogue; Experiment 2: Narrator; Experiment 3: Politeness."
  4. [Experiment 1: Dialogue Context] The valence analysis reports p=.053 for British participants' stronger interpretation of "very" with positive predicates and describes this as "most pronounced"; given the marginal p-value, the wording should be hedged.

Circularity Check

2 steps flagged · score 6.0 of 10

In-sample 'model predictions' and a z-scored Gaussian prior are the only circular elements; the central model comparison is not circular but is fragile.

  1. fitted input called prediction [Computational Model / Results, Figure 2 caption; Discussion]
    "The predictions generated are from the model parameters specifically optimized to fit each country's data separately. ... Figure 2 suggests that despite this limitation, the best-fitting models for each culture successfully capture the overall pattern of behavioral responses, with model predictions closely tracking empirical data for most modifier-predicate combinations."

    The parameters were optimized to minimize prediction loss on the same Experiment 1 ratings that Figure 2 plots, so 'model predictions closely tracking empirical data' is an in-sample fit guaranteed to improve with more parameters, not an independent test of the model. The figure is therefore presented as validation of the semantic/pragmatic decomposition while actually re-expressing the fitted values; the paper reports no held-out loss, cross-validation, or parameter-recovery check that would let the plot discriminate models.

  2. self definitional [Appendix, 'Modeling state prior as Gaussian']
    "We modeled the state prior as a normal distribution with mean 0 and variance 1. As seen in Figure 3, this assumption matches relatively well with the distribution of z-scored responses from participants suggesting that participants expect states to be distributed normally a priori."

    The fit is partly constructed: Experiment 1 z-scored all responses within each participant, so the empirical distribution has mean 0 and variance 1 by definition. A N(0,1) prior therefore matches the first two moments of the data regardless of the true state distribution; only the shape comparison is non-tautological, and no shape test is reported. The step is ancillary, but as stated it presents a preprocessing artifact as empirical support for the Gaussian prior.

full rationale

The main derivation—comparing RSA models with culture-specific semantic thresholds versus pragmatic weights via AIC/BIC—is not circular: the model parameters are fit to Experiment 1 data, and the model comparisons are genuine, if in-sample, statistical selections. The central claim that both semantic and pragmatic variation are needed rests on Table 2, not on the Figure 2 plot. However, two supporting steps do reduce to their inputs by construction: the Figure 2 'model predictions' are in-sample fits (parameters optimized on the very data plotted), and the N(0,1) prior 'match' is partially guaranteed by z-scoring. Also note that Table 2's BIC column favors the semantic-only M6 (BIC 22590) over the full M9 (BIC 22631), while the paper's choice of AIC over BIC is unargued; this is a robustness/correctness concern, not circularity. The paper transparently discloses that the predictions come from per-country fits and discusses outliers, which limits the severity. Net: partial circularity in the validation figures, but not in the formal model comparison.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The ledger shows that every component driving the central claim, the modifier thresholds, the informativity weight, the social utility weight, and the utterance cost, is a free parameter fitted to the same participant ratings. The Gaussian prior and double-threshold semantics are modeling assumptions needed to make the fits possible. No new physical or cognitive entities are postulated, but the decomposition between 'semantics' and 'pragmatics' is entirely a function of which fitted parameters are allowed to vary.

free parameters (4)
  • Semantic double thresholds for modifiers and baseline = not reported in paper
    Twelve parameters (two thresholds per modifier for five modifiers, plus two baseline thresholds) are optimized via CMA-ES on behavioral ratings; fitted numeric values are not listed.
  • Informativity weight phi_i per culture = not reported in paper
    Speaker utility weight on informativity; allowed to vary between UK and US in models M3, M5, M8, and M9; fitted values not reported.
  • Social utility weight phi_s per culture = not reported in paper
    Weight on politeness/kindness term, defined by Experiment 3 ratings; effectively zero for fit improvement; fitted values not reported.
  • Utterance cost C(w) per culture = not reported in paper
    Cost of producing a modifier; allowed to vary between cultures in models M4, M5, and M9; fitted values not reported.
assumptions (4)
  • domain assumption RSA recursive speaker-listener model is the correct process model for modifier interpretation
    Imported from Yoon et al. (2020), assumed rather than empirically validated within this paper (Computational Model, Modeling framework).
  • domain assumption State prior is Gaussian N(0,1)
    Appendix 'Modeling state prior as Gaussian'; the prior shapes the fitted thresholds and the paper only shows approximate correspondence to z-scored data.
  • ad hoc to paper Literal meaning of each modifier is a smooth double threshold
    Appendix 'Double thresholds for modifier semantics'; this form is chosen for gradient optimization and determines how much is attributed to semantics versus scalar implicature.
  • domain assumption Politeness ratings from Experiment 3 are a valid proxy for social utility
    The model defines U_soc via these ratings; the Discussion concedes this proxy may be poor, which weakens the politeness-related conclusions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Minding the Politeness Gap in Cross-cultural Communication." pith.science (2026). https://pith.science/paper/WPUTZP5W

@misc{pith2026250615623,
  author       = {Pith},
  title        = {Pith review of: Minding the Politeness Gap in Cross-cultural Communication},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WPUTZP5W}},
  note         = {Machine review of arXiv:2506.15623}
}
read the original abstract

Misunderstandings in cross-cultural communication often arise from subtle differences in interpretation, but it is unclear whether these differences arise from the literal meanings assigned to words or from more general pragmatic factors such as norms around politeness and brevity. In this paper, we report three experiments examining how speakers of British and American English interpret intensifiers like "quite" and "very." To better understand these cross-cultural differences, we developed a computational cognitive model where listeners recursively reason about speakers who balance informativity, politeness, and utterance cost. Our model comparisons suggested that cross-cultural differences in intensifier interpretation stem from a combination of (1) different literal meanings, (2) different weights on utterance cost. These findings challenge accounts based purely on semantic variation or politeness norms, demonstrating that cross-cultural differences in interpretation emerge from an intricate interplay between the two.

Figures

Figures reproduced from arXiv: 2506.15623 by the authors.

Figure 1
Figure 1. Results from three experiments reveal robust interpretation differences between UK and US participants. Experiments [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Comparison between model predictions and behav [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Gaussian state prior. Modeling state prior as Gaussian We modeled the state prior as a normal distribution with mean 0 and variance 1. As seen in [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 26 canonical work pages

  1. [1]

    Brown, P., & Levinson, S. C. (1987).Politeness: Some uni- versals in language usage. Cambridge University Press

    work page 1987

  2. [2]

    B., & Ragan-Kelley, J

    Chandra, K., Chen, T., Tenenbaum, J. B., & Ragan-Kelley, J. (2025). A domain-specific probabilistic programming language for reasoning about reasoning (or: a memo on memo).PsyArXiv preprint

    work page 2025

  3. [3]

    Culpeper, J., Haugh, M., & K´ad´ar, D. Z. (2017).The palgrave handbook of linguistic (im) politeness. Springer

    work page 2017

  4. [4]

    Desagulier, G. (2014). Visualizing distances in a set of near-synonyms: Rather, quite, fairly, and pretty. InCor- pus methods for semantics(pp. 145–178). John Benjamins Publishing Company

    work page 2014

  5. [5]

    Fowler, H. W. (2015).Fowler’s dictionary of modern english usage. Oxford University Press

    work page 2015

  6. [6]

    Goddard, C. (2012). ‘Early interactions’ in Australian En- glish, American English, and English English: Cultural dif- ferences and cultural scripts.Journal of Pragmatics,44, 1038-1050

    work page 2012

  7. [7]

    Goddard, C., & Wierzbicka, A. (2004). Cultural scripts: What are they and what are they good for?Intercultural Pragmatics,1(2), 153–166

    work page 2004

  8. [8]

    (2013).Words and mean- ings: Lexical semantics across domains, languages, and cultures

    Goddard, C., & Wierzbicka, A. (2013).Words and mean- ings: Lexical semantics across domains, languages, and cultures. Oxford University Press

    work page 2013

Show all 27 references
  1. [9]

    Grice, H. P. (1975). Logic and conversation.Syntax and Semantics,3, 41-58

    1975

  2. [10]

    (2023).The CMA Evolution Strategy: A Tutorial

    Hansen, N. (2023).The CMA Evolution Strategy: A Tutorial

    2023

  3. [11]

    Haugh, M., & Bousfield, D. (2012). Mock impoliteness, jocular mockery and jocular abuse in Australian and British English.Journal of Pragmatics,44(9), 1099-1114

    2012

  4. [12]

    Haugh, M., & Schneider, K. P. (2012). Im/politeness across englishes.Journal of Pragmatics,44(9), 1017-1021

    2012

  5. [13]

    Kecskes, I. (2023). Language variation and temporary norm development in intercultural interactions.Pragmatics & Cognition,30(2), 235–257

    2023

  6. [14]

    Lumer, E., & Buschmeier, H. (2022). Modeling social in- fluences on indirectness in a rational speech act approach to politeness. InProceedings of the Annual Meeting of the Cognitive Science Society

    2022

  7. [15]

    (2012).This is so cool! a comparative corpus study on intensifiers in British and American English.(Un- published manuscript, University of Tampere) Ruzait˙e, J

    Romero, S. (2012).This is so cool! a comparative corpus study on intensifiers in British and American English.(Un- published manuscript, University of Tampere) Ruzait˙e, J. (2007). Vague references to quantities as a face- saving strategy in teacher-student interaction.Lodz Pa...

    2012

  8. [16]

    (2017).Pragmatic aspects of scalar modifiers: The semantics-pragmatics interface(V ol

    Sawada, O. (2017).Pragmatic aspects of scalar modifiers: The semantics-pragmatics interface(V ol. 69). Oxford Uni- versity Press

    2017

  9. [17]

    Schneider, K. P. (2012). Appropriate behaviour across vari- eties of english.Journal of Pragmatics,44(9), 1022-1037

    2012

  10. [18]

    Schneider, K. P. (2024).Pragmatic variation within lan- guages(V ol. 232). Elsevier

    2024

  11. [19]

    Rieser, V

    Evans, G., . . . Rieser, V . (2025). Value profiles for encod- ing human variation.arXiv preprint arXiv:2503.15484

    2025

  12. [20]

    Spencer-Oatey, H., & K´ad´ar, D. Z. (2021).Intercultural po- liteness: Managing relations across cultures. Cambridge University Press

    2021

  13. [21]

    That’s proper cool

    Stratton, J. M. (2021). “That’s proper cool”: The emerging intensifier proper in British English.English Today,37(4), 206–213

    2021

  14. [22]

    Su, Y . (2016). Corpus-based comparative study of inten- sifiers: quite, pretty, rather and fairly.Journal of World Languages,3, 224-236

    2016

  15. [23]

    Thomas, J. (1983). Cross-cultural pragmatic failure.Applied linguistics,4(2), 91–112

    1983

  16. [24]

    Troutman, D. (2022). Sassy sasha?: The intersectionality of (im) politeness and sociolinguistics.Journal of Politeness Research,18(1), 121–149. van Dorst, I., Gillings, M., & Culpeper, J. (2024). Socioprag- matic variation in britain: A corpus-based study of polite- ness.Journ...

    2022

  17. [25]

    It’s rude to VP

    Waters, S. (2012). “It’s rude to VP”: The cultural semantics of rudeness.Journal of Pragmatics,44(9), 1051-1062

    2012

  18. [26]

    White, I., Pandey, S., & Pan, M. (2024). Communicate to play: Pragmatic reasoning for efficient cross-cultural com- munication. In Y . Al-Onaizan, M. Bansal, & Y.-N. Chen (Eds.),Findings of the Association for Computational Lin- guistics: EMNLP 2024(pp. 12201–12216)

    2024

  19. [27]

    J., Tessler, M

    Yoon, E. J., Tessler, M. H., Goodman, N. D., & Frank, M. C. (2020). Polite speech emerges from competing so- cial goals.Open Mind : Discoveries in Cognitive Science, 4, 71 - 87. Appendix Model ImplementationWe used the memo language (Chandra, Chen, Tenenbaum, & Ragan-Kelley, 2...

    2020

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.