Affective Tactility: Cross-Modal Translation from Natural Language to Procedural Height Maps

Affective Tactility: Cross-Modal Translation from Natural Language to Procedural Height Maps | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Affective Tactility: Cross-Modal Translation from Natural Language to Procedural Height Maps Yuusuke Harada This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8696852/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Text-to-image generation has rapidly matured, yet translating paralinguistic cues in text into physically perceivable tactility remains underexplored. We present Affective Tactility, a procedural framework that maps natural-language input to affect coordinates, valence (V) and arousal (A), and renders a manufacturable height map through an interpretable procedural texture generator. For end-to-end reproducibility, we describe a transparent lexicon-based baseline for estimating (V, A) and also validate the downstream mapping using controlled affect coordinates. The design is grounded in crossmodal correspondences (e.g., bouba–kiki) and is physiologically motivated by known tactile pathways. As a proof of concept, we demonstrate qualitative diversity of generated height maps and computational auditability: two simple descriptors, root-mean-square (RMS) roughness (Rq) and a spatial-frequency centroid, enable linear support vector machine (SVM) separation between affect regimes with 96.7% test accuracy, with ablations and a permuted control reported to address trivial separability. Beyond purely computational checks, we outline a no-participant physical verification protocol based on surface metrology and controlled scanning dynamics (profilometry, vibration, and friction measurements), positioning the method as a stimulus-generation foundation for future psychophysics and accessible haptic media. haptics affective computing procedural texture crossmodal correspondence natural language processing Figures Figure 1 Figure 2 Figure 3 Figure 4 1. Introduction Digital communication increasingly relies on text, yet much of the tactile and bodily nuance carried by voice, handwriting, or material artifacts is lost. While large language models can represent semantic context, their outputs are still experienced primarily as symbolic strings. In parallel, text-driven material synthesis has focused on visually realistic textures for physically based rendering, rather than on translating affective qualities into tactile structures. This work targets that gap: we treat text as a source of affective intent and seek to externalize it as a physical texture that can be touched. We build on two complementary lines of evidence. First, affective computing provides compact latent representations of emotion, such as the valence-arousal plane. Second, psychophysical studies of crossmodal correspondences show that auditory or linguistic sharpness tends to map onto spiky or angular shapes, while softness maps onto rounded forms. By using a procedural renderer with interpretable parameters, we can encode these correspondences into height maps that are manufacturable (e.g., 3D printing) or renderable on haptic displays. Contributions. (1) We propose a modular pipeline that maps natural language to valence-arousal coordinates and then to procedural texture parameters, producing a height map suitable for tactile output. (2) We formalize a parameter mapping that is interpretable and consistent with known crossmodal correspondences. (3) We provide a proof-of-concept evaluation including qualitative examples and a reversibility check demonstrating that affect classes are recoverable from height-map descriptors. Scope and validation strategy. This paper prioritizes a reproducible stimulus-generation foundation and objective auditability over immediate user studies. Specifically, we (i) fix all rendering parameters and physical scaling to enable independent replication, (ii) verify that affect regimes are structurally recoverable from height-map descriptors (Section 3.2 ), and (iii) provide a measurement-ready protocol for physical metrology and controlled scanning (Section 3.3 ). Human-subject psychophysics remains essential, but is best conducted once stimuli are controllable, safe, and comparable across fabrication and devices. 2. Method 2.1 System overview Figure 1 summarizes the proposed framework. Given input text, a semantic analysis module estimates affect coordinates (valence V and arousal A). A mapping function converts (V, A) into procedural parameters controlling spatial frequency, amplitude, and shape sharpness. A procedural renderer then generates a height map that can be interpreted as a tactile surface. 2.2 Affective semantic space We adopt the valence-arousal representation of affect, where valence captures pleasantness (negative to positive) and arousal captures activation (calm to excited). In practice, V and A can be estimated using (i) lexicon-based aggregation of token-level affect scores, (ii) regression from sentence embeddings trained on annotated corpora, or (iii) instruction-driven large language model (LLM) scoring. The present manuscript focuses on the downstream mapping and rendering, and uses controlled affect coordinates to validate the rendering behavior. 2.2.1 Baseline: Lexicon-based affect estimation (reproducible) To make the full pipeline reproducible without proprietary models, we provide a simple lexicon-based estimator for affect coordinates. For English, we look up token-level valence and arousal scores from normative lexicons and aggregate them over content words. Lexicon scores are first linearly mapped to [-1, 1] and then averaged: for tokens w_i with mapped scores (V_i, A_i), we compute V = clip((1/Z) Σ_i V_i, -1, 1) and A = clip((1/Z) Σ_i A_i, -1, 1), where Z is the number of matched tokens. We use the Warriner et al. norms as a default [ 19 ]; the National Research Council (NRC) Valence, Arousal, and Dominance (VAD) Lexicon provides a larger alternative with Best-Worst Scaling annotations [ 22 ]. For Japanese, an equivalent affect lexicon or cross-lingual projection can be used. In the experiments of this paper, we primarily use controlled (V, A) to isolate and validate the downstream mapping and rendering. 2.3 Mapping from (V, A) to procedural parameters Throughout this paper we assume V, A ∈ [-1, 1] (clipped if necessary). We parameterize the procedural texture generator with three interpretable variables: frequency f, amplitude s, and sharpness gamma. Frequency controls spatial periodicity and high-frequency content, amplitude controls global relief, and gamma controls peak sharpening through a nonlinear shaping transform. A simple mapping is: f = f_min + (A + 1)/2 * (f_max - f_min) s = s_min + (1 - (V + 1)/2) * (s_max - s_min) gamma = g_min + (0.65*(A + 1)/2 + 0.35*(1 - (V + 1)/2)) * (g_max - g_min) This mapping increases spatial frequency with arousal, and increases relief and sharpness with negative valence and/or high arousal. The convex weights (0.65 and 0.35) are an empirical initialization reflecting that sharpness is driven more strongly by arousal than valence in our current design. Importantly, these coefficients are not fixed laws but tunable parameters that can be calibrated against physical measurements or perceptual data while preserving interpretability. 2.4 Procedural rendering We render a height field H(x,y) using multi-octave sinusoidal fractional Brownian motion (fBm)-like noise with optional domain warping to produce organic variation. Each octave doubles spatial frequency and halves amplitude. A shaping transform controlled by gamma sharpens or smooths peaks; we use sign(H)*|H|^(1/gamma) as a simple peak-sharpening nonlinearity. Unless otherwise noted, we use the default configuration in Table 1 . For visualization and fabrication, the output is min-max normalized to [0,1] and can be exported as a grayscale height map or as a mesh; for quantitative descriptors (Section 3.2 ) we compute statistics on the centered, unnormalized field to retain amplitude information. A complete reproducibility package (scripts, configuration files, and figure sources) is provided as Online Resource 1 (ESM_1.zip). Algorithm 1 Procedural height-field rendering (summary). 1. Input: affect coordinates (V, A); grid resolution N. 2. Map (V, A) -> (f, s, gamma) using Eqs. (1)-(3). 3. Accumulate multi-octave sinusoidal noise with optional domain warping. 4. Apply shaping: H <- sign(H)*|H|^(1/gamma); scale by s. 5. Output: (i) centered H for analysis; (ii) normalized [0,1] height map for export. Table 1 Default configuration used in our experiments (reproducibility) Item Value Affect range V, A ∈ [-1, 1] (clipped) Parameter ranges f ∈ [3.0, 7.0], s ∈ [0.06, 0.30], gamma ∈ [0.9, 2.6] Grid resolution 128×128 Octaves 4 Warp strength 0.25 Random seed 42 Physical relief (example) H_max = 1.2 mm (scalable) 3. Experiments 3.1 Qualitative examples Figure 3 shows representative height maps generated under three affect settings. The calm condition produces low-frequency, smooth undulations; the anger condition produces dense, sharp peaks; and the mixed condition yields warped, organic structures. These are illustrative outputs intended to communicate the controllability of the parameterization. 3.2 Reversibility check via height-map descriptors To test whether affect is preserved as measurable surface structure rather than arbitrary appearance, we perform a reversibility check. We sample two affect classes (positive/calm vs negative/high-arousal), generate height maps with jittered (V, A), and compute two descriptors: RMS roughness Rq (root-mean-square height deviation) and a 2D fast Fourier transform (FFT)-based spatial-frequency centroid computed as the magnitude-weighted radial frequency of the spectrum. We train a linear SVM on these descriptors and report test accuracy. In this controlled-affect experiment (N = 120; 60 samples per class; 75/25 train-test split), the classifier achieved 96.7% test accuracy (Fig. 4 ). Table 2 reports repeated stratified 5-fold cross-validation (20 repeats) ablations: Rq-only 91.6 ± 4.9%, spatial-frequency centroid-only 74.2 ± 8.5%, and Rq+centroid 92.1 ± 4.8%. As a control, when class labels are randomly permuted, accuracy drops to 50.0 ± 9.5% (near chance), supporting that separability reflects structured regimes rather than classifier artifacts. Table 2 Ablations and control for the reversibility check (accuracy, mean ± standard deviation, SD) Setup Accuracy (mean ± SD) Rq only 91.6 ± 4.9% Spatial-frequency centroid only 74.2 ± 8.5% Rq + centroid 92.1 ± 4.8% Permuted labels 50.0 ± 9.5% 3.3 Physical metrology and interaction-level verification (no-user validation) Computational auditability (Section 3.2 ) confirms that affect regimes yield distinct structural signatures in the rendered height fields. However, fabrication and contact mechanics can distort surface statistics, and tactile perception depends on interaction dynamics. To strengthen validity without recruiting participants, we specify a measurement protocol that can be executed on fabricated tiles (e.g., 3D printed reliefs) or on haptic displays. 3.3.1 Geometry-level surface metrology We recommend measuring the realized surface using an optical profilometer, confocal microscope, or structured-light scanner to obtain a height field with known lateral sampling. From the measured field, compute standard areal surface parameters (e.g., Sa, Sq, Sz) and gradient-based roughness (Sdq) as defined in International Organization for Standardization (ISO) 25178-2 [ 26 ]. Frequency content can be summarized via a 2D power spectral density (PSD) and a radial spectral centroid, which are common in surface metrology [ 27 ]. High-resolution 3D-printed textures have also been used to isolate how spatial period and element geometry shape roughness perception, supporting the relevance of such measurable descriptors for fabrication-oriented studies [ 34 ]. These descriptors are directly comparable to those computed from the procedural output, enabling fabrication-aware reproducibility checks (e.g., deviation from the target PSD under different printers or materials). 3.3.2 Interaction-level scanning: vibration and friction During active touch, spatial structure is converted into temporal vibrations depending on scanning speed and contact mechanics. For a dominant spatial wavelength λ (mm) explored at speed v (mm/s), the corresponding temporal frequency scales approximately as f_t ≈ v/λ, motivating controlled-scanning measurements [ 31 ]. A practical setup uses a linear stage with controlled normal force and scan speed, while recording (i) tri-axial acceleration near the fingertip proxy (or device end-effector) and (ii) tangential force to characterize friction. Prior work links fine-texture perception to vibration codes and friction-induced vibrations, and reports systematic spectral signatures under speed manipulations [ 21 , 29 , 30 , 31 ]. Friction itself can shape perceived coarseness and should therefore be reported alongside vibration descriptors [ 28 ]. Therefore, we suggest reporting vibration PSD bands, spectral centroids, and friction coefficients as interaction-level correlates of the (V, A) regimes. 3.3.3 Analysis targets and acceptance criteria The above measurements enable objective, falsifiable checks: (i) monotonic trends of geometric descriptors with arousal (e.g., higher PSD centroid with higher A) and with valence (e.g., higher relief/gradient with more negative V), (ii) cross-fabrication reproducibility quantified by descriptor variance across replicas, and (iii) discriminability of affect regimes using physical descriptors, analogous to the SVM audit in Section 3.2 . Finally, metrology results can be tied to manufacturability and safety by bounding maximum slope and minimum feature size with respect to printer limits and tactile discrimination limits reported in prior psychophysics [ 33 ]. 4. Discussion 4.1 Physiological plausibility from a physical-therapy perspective Touch perception integrates multiple afferent channels, and the same surface can evoke different sensations depending on exploration dynamics. In active touch, spatial patterns are converted into temporal vibrations through scanning and frictional interactions, so scan speed and normal force act as hidden variables that should be controlled or reported [ 24 , 29 , 31 ]. Discriminative touch in glabrous skin is mediated by myelinated mechanoreceptors that encode indentation, edges, and vibration across frequency bands [ 4 , 24 , 25 ]. In our mapping, increasing arousal A increases spatial frequency, which at a fixed scan speed shifts energy toward higher temporal frequencies; decreasing valence V increases relief and slopes, potentially amplifying edge-related cues and friction modulation. Affective-touch hypotheses, in contrast, emphasize unmyelinated C-tactile (CT) afferents in hairy skin, tuned to gentle, slow stroking and projecting to affect-related cortical regions [ 7 , 14 , 15 , 20 ]. Because CT afferents are sparse in glabrous fingertips, CT-related interpretations are most directly applicable when stimuli are explored on hairy skin (e.g., forearm) or when devices emulate low-force stroking. Accordingly, we treat CT physiology as a motivation for parameter regimes (smooth, low-gradient surfaces) rather than as a claim of direct activation under all contact conditions. Finally, physiological plausibility also depends on physical scaling and safety. Although Table 1 provides an example relief range, any deployment should bound maximum slope and peak curvature to avoid nociceptive stimulation and to remain compatible with the target fabrication resolution [ 13 , 26 ]. 4.2 Interpretation drift and creative affordances Because the mapping is continuous, boundary regions between affect quadrants can produce ambiguous textures that invite interpretation. This may be valuable in digital art and therapy-oriented interaction design, where controlled ambiguity supports reflection rather than categorical labeling. Such boundary behaviors also resonate with accounts of unexpected creativity in generative systems [ 16 ]. 4.3 Validation without immediate user studies: physical auditability A common critique in human-computer interaction (HCI) and multimodal-interface venues is the lack of human-subject validation: do the generated textures actually induce the intended affect? We agree that psychophysics is ultimately necessary, but argue that it should be preceded by a stimulus-generation foundation: stimuli must be controllable, reproducible, and physically comparable across devices and fabrication. Accordingly, we complement the computational audit (Section 3.2 ) with a no-participant physical verification protocol (Section 3.3 ) grounded in surface metrology and interaction dynamics [ 26 , 32 ]. At the geometry level, fabricated surfaces can be measured via profilometry and summarized with ISO areal parameters (Sa, Sq, Sdq, Sal) and PSD descriptors [ 26 , 27 ]. Such descriptors are widely used to link precisely specified 3D-printed geometry to roughness perception, making them particularly suitable for fabrication-oriented validation [ 34 ]. This enables falsifiable checks that are independent of subjective judgment, such as: (i) whether the realized surface matches the target spectral regime implied by (V, A), and (ii) whether the mapping preserves separability under different printers/materials (cross-fabrication robustness). At the interaction level, controlled scanning connects spatial geometry to temporal vibration and friction signals that are known to support texture perception. Prior work reports that fine textures are strongly mediated by vibration codes and friction-induced vibrations, and that scanning speed systematically shifts vibration spectra [ 21 , 29 , 30 , 31 ]. Therefore, measuring vibration PSD bands and friction coefficients under standardized v and normal force provides an objective bridge between our spatial parameterization and the temporal signals available to mechanoreceptors. Importantly, these measurements also support safety-relevant bounds. Sharp peaks and high gradients can become uncomfortable or even noxious at larger physical scales; metrology can quantify maximum slope and curvature to keep stimuli within safe regimes before any user exposure [ 13 , 33 ]. 4.4 From heuristic mapping to calibratable models Our current mapping from (V, A) to (f, s, gamma) is intentionally simple and interpretable. It should be read as an empirical initialization rather than a universal law: the value is that the mapping is parameterized and thus calibratable. Calibration can be performed against either (i) physical measurements (e.g., targeting a desired PSD centroid or Sdq range) or (ii) psychophysical ratings once human studies are conducted [ 26 , 32 ]. Furthermore, tactile experience is multi-dimensional: beyond roughness, factors such as warmth/coldness, compliance, and friction contribute to material perception [ 32 ]. This suggests a natural extension of Affective Tactility toward multi-objective mappings and richer descriptors, while retaining the modular structure of text-to-affect estimation (Section 2.2.1 ) and procedural rendering. 5. Limitations and future work This manuscript reports a proof of concept: we validate the mapping using controlled affect coordinates and computational descriptors (Section 3.2 ), and we specify (but do not yet execute) a physical metrology and controlled-scanning protocol that enables no-participant verification (Section 3.3 ). End-to-end text-to-(V, A) estimation is provided as a reproducible lexicon baseline (Section 2.2.1 ), but more robust contextual estimators and multilingual coverage remain open problems. Future work will (i) collect physical measurements across fabrication methods and materials, (ii) calibrate mapping coefficients against metrology and/or psychophysical data, and (iii) evaluate perceived affect communication and comfort in controlled human-subject studies. When targeting clinical or assistive settings, ethics approval and safety constraints (e.g., avoiding nociceptive stimulation) will be required. 6. Conclusion We presented Affective Tactility, a modular framework for translating natural-language affect into procedural tactile textures. By mapping valence and arousal onto interpretable procedural parameters, the method produces manufacturable height maps that exhibit controllable qualitative differences. A reversibility check using surface descriptors supports that affect regimes correspond to distinct structural signatures. We hope this work enables accessible affect communication and new forms of embodied digital art. Declarations Funding: The author did not receive support from any organization for the submitted work. Competing interests: The author has no relevant financial or non-financial interests to disclose. Ethics approval: Not applicable. This study did not involve human participants or animals. Consent to participate: Not applicable. Consent for publication: Not applicable. Availability of data and materials: All data generated in this study are synthetic and can be regenerated using Online Resource. Code availability: Source code and configuration files used to generate the stimuli, figures, and tables are provided in Online Resource. Author contributions: Conceptualization, methodology, software, formal analysis and investigation, writing (original draft), and writing (review and editing): Yuusuke Harada. References Ramesh A et al (2021) Zero-shot text-to-image generation. Proceedings of ICML. https://arxiv.org/abs/2102.12092 Rombach R et al (2022) High-resolution image synthesis with latent diffusion models. CVPR. https://arxiv.org/abs/2112.10752 MacLean KE (2008) Haptic interaction design for everyday interfaces. Reviews Hum Factors Ergon 4(1):149–194. https://doi.org/10.1518/155723408X342826 Abraira VE, Ginty DD (2013) The sensory neurons of touch. Neuron 79(4):618–639. https://doi.org/10.1016/j.neuron.2013.07.051 Vaswani A et al (2017) Attention is all you need. NeurIPS. https://arxiv.org/abs/1706.03762 Dong Y et al (2023) Text-driven texture synthesis via diffusion models. ICCV. https://arxiv.org/abs/2303.11396 McGlone F, Wessberg J, Olausson H (2014) Discriminative and affective touch: sensing and feeling. Neuron 82(4):737–755. https://doi.org/10.1016/j.neuron.2014.05.020 Spence C (2011) Crossmodal correspondences: A tutorial review. Attention, Perception, & Psychophysics, 73(4), 971–995. https://doi.org/10.3758/s13414-010-0073-7 Russell JA (1980) A circumplex model of affect. J Personal Soc Psychol 39(6):1161–1178. https://doi.org/10.1037/h0077714 Perlin K (1985) An image synthesizer. ACM SIGGRAPH Comput Graphics 19(3):287–296. https://doi.org/10.1145/325165.325247 Ramachandran VS, Hubbard EM (2001) Synaesthesia: A window into perception, thought and language. J Conscious Stud 8(12):3–34 Cortes C, Vapnik V (1995) Support-vector networks. Mach Learn 20(3):273–297. https://doi.org/10.1007/BF00994018 Perl ER (2007) Ideas about pain, a historical view. Nat Rev Neurosci 8(1):71–80. https://doi.org/10.1038/nrn2042 Vallbo AB, Olausson H, Wessberg J (1999) Unmyelinated afferents constitute a second system coding tactile stimuli of the human hairy skin. J Neurophysiol 81(6):2753–2763. https://doi.org/10.1152/jn.1999.81.6.2753 Olausson H et al (2002) Unmyelinated tactile afferents signal touch and project to insular cortex. Nat Neurosci 5(9):900–904. https://doi.org/10.1038/nn896 Boden MA (2004) The Creative Mind: Myths and Mechanisms. Routledge Klatzky RL, Lederman SJ (2003) Touch. In Handbook of Psychology (Vol. 4), 147–176 Bradley MM, Lang PJ (1994) Measuring emotion: The Self-Assessment Manikin and the Semantic Differential. J Behav Ther Exp Psychiatry 25(1):49–59. https://doi.org/10.1016/0005-7916(94)90063-9 Warriner AB, Kuperman V, Brysbaert M (2013) Norms of valence, arousal, and dominance for 13,915 English lemmas. Behav Res Methods 45:1191–1207. https://doi.org/10.3758/s13428-012-0314-x Löken LS, Wessberg J, Morrison I, McGlone F, Olausson H (2009) Coding of pleasant touch by unmyelinated afferents in humans. Nat Neurosci 12(5):547–548. https://doi.org/10.1038/nn.2312 Weber AI, Saal HP, Lieber JD, Cheng J-W, Manfredi LR, Dammann JF (2013) and S. J. Bensmaia. Spatial and temporal codes mediate the tactile perception of natural textures. Proceedings of the National Academy of Sciences, 110(42):17107–17112. https://doi.org/10.1073/pnas.1305509110 S. M. Mohammad. Obtaining reliable human ratings of valence, arousal, and dominance for 20,000 English words. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, pages 174–184, (2018) https://doi.org/10.18653/v1/P18-1017 Hollins M, Risner SR (2000) Evidence for the duplex theory of tactile texture perception. Percept Psychophys 62(4):695–705. https://doi.org/10.3758/BF03211625 Johansson RS, Flanagan JR (2009) Coding and use of tactile signals from the fingertips in object manipulation tasks. Nat Rev Neurosci 10(5):345–359. https://doi.org/10.1038/nrn2621 K. O. Johnson. The roles and functions of cutaneous mechanoreceptors. Curr Opin Neurobiol, 11(4):455–461, (2001) https://doi.org/10.1016/S0959-4388(00)00234-8 ISO 25178-2:2012 (2012) Geometrical product specifications (GPS) - Surface texture: Areal - Part 2: Terms, definitions and surface texture parameters. International Organization for Standardization D. J. Whitehouse. Handbook of Surface Metrology. Institute of Physics Publishing, (1994) Skedung L, Danerlöv K, Olofsson U, Johannesson CM, Aikala M, Kettle J, Arvidsson M, Berglund B (2011) Finger friction, surface roughness and perceived coarseness of printed surfaces. Tribol Int 44:505–512. https://doi.org/10.1016/j.triboint.2010.04.010 Delhaye BP, Hayward V, Lefèvre P (2012) Thonnard. Texture-induced vibrations in the forearm during tactile exploration. Front Behav Neurosci 6:37. https://doi.org/10.3389/fnbeh.2012.00037 Bensmaia SJ, Hollins M (2005) Pacinian representations of fine surface texture. Percept Psychophys 67(5):842–854. https://doi.org/10.3758/BF03193537 Cascio CJ, Sathian K (2001) Temporal cues contribute to tactile perception of roughness. J Neurosci 21(14):5289–5296. https://doi.org/10.1523/JNEUROSCI.21-14-05289.2001 Okamoto S, Nagano H, Yamada Y (2013) Psychophysical dimensions of tactile perception of textures. IEEE Trans Haptics 6(1):81–93. https://doi.org/10.1109/TOH.2012.32 Skedung L, Arvidsson M, Chung JY, Johannesson CM, Olofsson U (2013) Berglund. Feeling small: Exploring the tactile perception limits. Sci Rep 3:2617. https://doi.org/10.1038/srep02617 Tymms C, Zorin D, Gardner EP (2018) Tactile perception of the roughness of 3D-printed textures. J Neurophysiol 119(3):862–876. https://doi.org/10.1152/jn.00564.2017 Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8696852","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":580295177,"identity":"d7b4e5cc-8c66-4e5d-960b-52fc2385513b","order_by":0,"name":"Yuusuke Harada","email":"data:image/png;base64,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","orcid":"","institution":"Hiroshima University","correspondingAuthor":true,"prefix":"","firstName":"Yuusuke","middleName":"","lastName":"Harada","suffix":""}],"badges":[],"createdAt":"2026-01-26 05:50:02","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-8696852/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8696852/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":101751654,"identity":"dd732493-d5bd-434a-aa47-5196e6523b20","added_by":"auto","created_at":"2026-02-03 10:22:04","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":328999,"visible":true,"origin":"","legend":"\u003cp\u003eSystem overview of Affective Tactility. Natural language is mapped to affect (V, A), then to procedural parameters, and finally rendered as a tactile height map\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8696852/v1/642e107a491b4479febac9ca.png"},{"id":101751663,"identity":"92740e32-4df9-4411-86ab-37376925b61b","added_by":"auto","created_at":"2026-02-03 10:22:06","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":367342,"visible":true,"origin":"","legend":"\u003cp\u003eConceptual mapping from affective semantic space to procedural parameters. Each quadrant corresponds to a characteristic parameter regime\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8696852/v1/01d2cef510bf507c450a1aba.png"},{"id":101435842,"identity":"11ddff9e-1d1c-4861-9064-6dde0ae4c640","added_by":"auto","created_at":"2026-01-29 16:19:42","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":752068,"visible":true,"origin":"","legend":"\u003cp\u003eQualitative examples of generated height maps shown as shaded renderings\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8696852/v1/b4aed93cb01614b153110e40.png"},{"id":101435839,"identity":"e9772804-efea-474d-b464-0eb34ed2c57b","added_by":"auto","created_at":"2026-01-29 16:19:42","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":353401,"visible":true,"origin":"","legend":"\u003cp\u003eReversibility check using a linear support vector machine (SVM) on two height-map descriptors under controlled affect coordinates. High separability indicates that affect regimes produce distinct structural signatures\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8696852/v1/41b748e16c737c58f79489c2.png"},{"id":101880686,"identity":"a382dc9c-86f1-40b7-a284-190b56c6f208","added_by":"auto","created_at":"2026-02-04 15:05:14","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2571265,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8696852/v1/5fffee55-4f62-4d2c-95a4-11dadb08ee77.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eAffective Tactility: Cross-Modal Translation from Natural Language to Procedural Height Maps\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eDigital communication increasingly relies on text, yet much of the tactile and bodily nuance carried by voice, handwriting, or material artifacts is lost. While large language models can represent semantic context, their outputs are still experienced primarily as symbolic strings. In parallel, text-driven material synthesis has focused on visually realistic textures for physically based rendering, rather than on translating affective qualities into tactile structures. This work targets that gap: we treat text as a source of affective intent and seek to externalize it as a physical texture that can be touched.\u003c/p\u003e \u003cp\u003eWe build on two complementary lines of evidence. First, affective computing provides compact latent representations of emotion, such as the valence-arousal plane. Second, psychophysical studies of crossmodal correspondences show that auditory or linguistic sharpness tends to map onto spiky or angular shapes, while softness maps onto rounded forms. By using a procedural renderer with interpretable parameters, we can encode these correspondences into height maps that are manufacturable (e.g., 3D printing) or renderable on haptic displays.\u003c/p\u003e \u003cp\u003eContributions. (1) We propose a modular pipeline that maps natural language to valence-arousal coordinates and then to procedural texture parameters, producing a height map suitable for tactile output. (2) We formalize a parameter mapping that is interpretable and consistent with known crossmodal correspondences. (3) We provide a proof-of-concept evaluation including qualitative examples and a reversibility check demonstrating that affect classes are recoverable from height-map descriptors.\u003c/p\u003e \u003cp\u003eScope and validation strategy. This paper prioritizes a reproducible stimulus-generation foundation and objective auditability over immediate user studies. Specifically, we (i) fix all rendering parameters and physical scaling to enable independent replication, (ii) verify that affect regimes are structurally recoverable from height-map descriptors (Section \u003cspan refid=\"Sec10\" class=\"InternalRef\"\u003e3.2\u003c/span\u003e), and (iii) provide a measurement-ready protocol for physical metrology and controlled scanning (Section \u003cspan refid=\"Sec11\" class=\"InternalRef\"\u003e3.3\u003c/span\u003e). Human-subject psychophysics remains essential, but is best conducted once stimuli are controllable, safe, and comparable across fabrication and devices.\u003c/p\u003e"},{"header":"2. Method","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 System overview\u003c/h2\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e summarizes the proposed framework. Given input text, a semantic analysis module estimates affect coordinates (valence V and arousal A). A mapping function converts (V, A) into procedural parameters controlling spatial frequency, amplitude, and shape sharpness. A procedural renderer then generates a height map that can be interpreted as a tactile surface.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Affective semantic space\u003c/h2\u003e \u003cp\u003eWe adopt the valence-arousal representation of affect, where valence captures pleasantness (negative to positive) and arousal captures activation (calm to excited). In practice, V and A can be estimated using (i) lexicon-based aggregation of token-level affect scores, (ii) regression from sentence embeddings trained on annotated corpora, or (iii) instruction-driven large language model (LLM) scoring. The present manuscript focuses on the downstream mapping and rendering, and uses controlled affect coordinates to validate the rendering behavior.\u003c/p\u003e \u003cdiv id=\"Sec5\" class=\"Section3\"\u003e \u003ch2\u003e2.2.1 Baseline: Lexicon-based affect estimation (reproducible)\u003c/h2\u003e \u003cp\u003eTo make the full pipeline reproducible without proprietary models, we provide a simple lexicon-based estimator for affect coordinates. For English, we look up token-level valence and arousal scores from normative lexicons and aggregate them over content words. Lexicon scores are first linearly mapped to [-1, 1] and then averaged: for tokens w_i with mapped scores (V_i, A_i), we compute V\u0026thinsp;=\u0026thinsp;clip((1/Z) Σ_i V_i, -1, 1) and A\u0026thinsp;=\u0026thinsp;clip((1/Z) Σ_i A_i, -1, 1), where Z is the number of matched tokens. We use the Warriner et al. norms as a default [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]; the National Research Council (NRC) Valence, Arousal, and Dominance (VAD) Lexicon provides a larger alternative with Best-Worst Scaling annotations [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. For Japanese, an equivalent affect lexicon or cross-lingual projection can be used. In the experiments of this paper, we primarily use controlled (V, A) to isolate and validate the downstream mapping and rendering.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Mapping from (V, A) to procedural parameters\u003c/h2\u003e \u003cp\u003eThroughout this paper we assume V, A \u0026isin; [-1, 1] (clipped if necessary).\u003c/p\u003e \u003cp\u003eWe parameterize the procedural texture generator with three interpretable variables: frequency f, amplitude s, and sharpness gamma. Frequency controls spatial periodicity and high-frequency content, amplitude controls global relief, and gamma controls peak sharpening through a nonlinear shaping transform. A simple mapping is:\u003c/p\u003e \u003cp\u003ef\u0026thinsp;=\u0026thinsp;f_min + (A\u0026thinsp;+\u0026thinsp;1)/2 * (f_max - f_min)\u003c/p\u003e \u003cp\u003es\u0026thinsp;=\u0026thinsp;s_min + (1 - (V\u0026thinsp;+\u0026thinsp;1)/2) * (s_max - s_min)\u003c/p\u003e \u003cp\u003egamma\u0026thinsp;=\u0026thinsp;g_min + (0.65*(A\u0026thinsp;+\u0026thinsp;1)/2\u0026thinsp;+\u0026thinsp;0.35*(1 - (V\u0026thinsp;+\u0026thinsp;1)/2)) * (g_max - g_min)\u003c/p\u003e \u003cp\u003eThis mapping increases spatial frequency with arousal, and increases relief and sharpness with negative valence and/or high arousal. The convex weights (0.65 and 0.35) are an empirical initialization reflecting that sharpness is driven more strongly by arousal than valence in our current design. Importantly, these coefficients are not fixed laws but tunable parameters that can be calibrated against physical measurements or perceptual data while preserving interpretability.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Procedural rendering\u003c/h2\u003e \u003cp\u003eWe render a height field H(x,y) using multi-octave sinusoidal fractional Brownian motion (fBm)-like noise with optional domain warping to produce organic variation. Each octave doubles spatial frequency and halves amplitude. A shaping transform controlled by gamma sharpens or smooths peaks; we use sign(H)*|H|^(1/gamma) as a simple peak-sharpening nonlinearity. Unless otherwise noted, we use the default configuration in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. For visualization and fabrication, the output is min-max normalized to [0,1] and can be exported as a grayscale height map or as a mesh; for quantitative descriptors (Section \u003cspan refid=\"Sec10\" class=\"InternalRef\"\u003e3.2\u003c/span\u003e) we compute statistics on the centered, unnormalized field to retain amplitude information. A complete reproducibility package (scripts, configuration files, and figure sources) is provided as Online Resource 1 (ESM_1.zip).\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eAlgorithm 1\u003c/strong\u003e \u003cp\u003eProcedural height-field rendering (summary).\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e1. Input: affect coordinates (V, A); grid resolution N.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e2. Map (V, A) -\u0026gt; (f, s, gamma) using Eqs.\u0026nbsp;(1)-(3).\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e3. Accumulate multi-octave sinusoidal noise with optional domain warping.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e4. Apply shaping: H \u0026lt;- sign(H)*|H|^(1/gamma); scale by s.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e5. Output: (i) centered H for analysis; (ii) normalized [0,1] height map for export.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDefault configuration used in our experiments (reproducibility)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eItem\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eValue\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAffect range\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eV, A \u0026isin; [-1, 1] (clipped)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParameter ranges\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ef \u0026isin; [3.0, 7.0], s \u0026isin; [0.06, 0.30], gamma \u0026isin; [0.9, 2.6]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGrid resolution\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e128\u0026times;128\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOctaves\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWarp strength\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.25\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom seed\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e42\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePhysical relief (example)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eH_max\u0026thinsp;=\u0026thinsp;1.2 mm (scalable)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"3. Experiments","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Qualitative examples\u003c/h2\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e shows representative height maps generated under three affect settings. The calm condition produces low-frequency, smooth undulations; the anger condition produces dense, sharp peaks; and the mixed condition yields warped, organic structures. These are illustrative outputs intended to communicate the controllability of the parameterization.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Reversibility check via height-map descriptors\u003c/h2\u003e \u003cp\u003eTo test whether affect is preserved as measurable surface structure rather than arbitrary appearance, we perform a reversibility check. We sample two affect classes (positive/calm vs negative/high-arousal), generate height maps with jittered (V, A), and compute two descriptors: RMS roughness Rq (root-mean-square height deviation) and a 2D fast Fourier transform (FFT)-based spatial-frequency centroid computed as the magnitude-weighted radial frequency of the spectrum. We train a linear SVM on these descriptors and report test accuracy.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn this controlled-affect experiment (N\u0026thinsp;=\u0026thinsp;120; 60 samples per class; 75/25 train-test split), the classifier achieved 96.7% test accuracy (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e reports repeated stratified 5-fold cross-validation (20 repeats) ablations: Rq-only 91.6\u0026thinsp;\u0026plusmn;\u0026thinsp;4.9%, spatial-frequency centroid-only 74.2\u0026thinsp;\u0026plusmn;\u0026thinsp;8.5%, and Rq+centroid 92.1\u0026thinsp;\u0026plusmn;\u0026thinsp;4.8%. As a control, when class labels are randomly permuted, accuracy drops to 50.0\u0026thinsp;\u0026plusmn;\u0026thinsp;9.5% (near chance), supporting that separability reflects structured regimes rather than classifier artifacts.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eAblations and control for the reversibility check (accuracy, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation, SD)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSetup\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy (mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRq only\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e91.6\u0026thinsp;\u0026plusmn;\u0026thinsp;4.9%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSpatial-frequency centroid only\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e74.2\u0026thinsp;\u0026plusmn;\u0026thinsp;8.5%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRq\u0026thinsp;+\u0026thinsp;centroid\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e92.1\u0026thinsp;\u0026plusmn;\u0026thinsp;4.8%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePermuted labels\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e50.0\u0026thinsp;\u0026plusmn;\u0026thinsp;9.5%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Physical metrology and interaction-level verification (no-user validation)\u003c/h2\u003e \u003cp\u003eComputational auditability (Section \u003cspan refid=\"Sec10\" class=\"InternalRef\"\u003e3.2\u003c/span\u003e) confirms that affect regimes yield distinct structural signatures in the rendered height fields. However, fabrication and contact mechanics can distort surface statistics, and tactile perception depends on interaction dynamics. To strengthen validity without recruiting participants, we specify a measurement protocol that can be executed on fabricated tiles (e.g., 3D printed reliefs) or on haptic displays.\u003c/p\u003e \u003cdiv id=\"Sec12\" class=\"Section3\"\u003e \u003ch2\u003e3.3.1 Geometry-level surface metrology\u003c/h2\u003e \u003cp\u003eWe recommend measuring the realized surface using an optical profilometer, confocal microscope, or structured-light scanner to obtain a height field with known lateral sampling. From the measured field, compute standard areal surface parameters (e.g., Sa, Sq, Sz) and gradient-based roughness (Sdq) as defined in International Organization for Standardization (ISO) 25178-2 [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. Frequency content can be summarized via a 2D power spectral density (PSD) and a radial spectral centroid, which are common in surface metrology [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. High-resolution 3D-printed textures have also been used to isolate how spatial period and element geometry shape roughness perception, supporting the relevance of such measurable descriptors for fabrication-oriented studies [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. These descriptors are directly comparable to those computed from the procedural output, enabling fabrication-aware reproducibility checks (e.g., deviation from the target PSD under different printers or materials).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section3\"\u003e \u003ch2\u003e3.3.2 Interaction-level scanning: vibration and friction\u003c/h2\u003e \u003cp\u003eDuring active touch, spatial structure is converted into temporal vibrations depending on scanning speed and contact mechanics. For a dominant spatial wavelength λ (mm) explored at speed v (mm/s), the corresponding temporal frequency scales approximately as f_t\u0026thinsp;\u0026asymp;\u0026thinsp;v/λ, motivating controlled-scanning measurements [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. A practical setup uses a linear stage with controlled normal force and scan speed, while recording (i) tri-axial acceleration near the fingertip proxy (or device end-effector) and (ii) tangential force to characterize friction. Prior work links fine-texture perception to vibration codes and friction-induced vibrations, and reports systematic spectral signatures under speed manipulations [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. Friction itself can shape perceived coarseness and should therefore be reported alongside vibration descriptors [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. Therefore, we suggest reporting vibration PSD bands, spectral centroids, and friction coefficients as interaction-level correlates of the (V, A) regimes.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section3\"\u003e \u003ch2\u003e3.3.3 Analysis targets and acceptance criteria\u003c/h2\u003e \u003cp\u003eThe above measurements enable objective, falsifiable checks: (i) monotonic trends of geometric descriptors with arousal (e.g., higher PSD centroid with higher A) and with valence (e.g., higher relief/gradient with more negative V), (ii) cross-fabrication reproducibility quantified by descriptor variance across replicas, and (iii) discriminability of affect regimes using physical descriptors, analogous to the SVM audit in Section \u003cspan refid=\"Sec10\" class=\"InternalRef\"\u003e3.2\u003c/span\u003e. Finally, metrology results can be tied to manufacturability and safety by bounding maximum slope and minimum feature size with respect to printer limits and tactile discrimination limits reported in prior psychophysics [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Physiological plausibility from a physical-therapy perspective\u003c/h2\u003e \u003cp\u003eTouch perception integrates multiple afferent channels, and the same surface can evoke different sensations depending on exploration dynamics. In active touch, spatial patterns are converted into temporal vibrations through scanning and frictional interactions, so scan speed and normal force act as hidden variables that should be controlled or reported [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eDiscriminative touch in glabrous skin is mediated by myelinated mechanoreceptors that encode indentation, edges, and vibration across frequency bands [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. In our mapping, increasing arousal A increases spatial frequency, which at a fixed scan speed shifts energy toward higher temporal frequencies; decreasing valence V increases relief and slopes, potentially amplifying edge-related cues and friction modulation.\u003c/p\u003e \u003cp\u003eAffective-touch hypotheses, in contrast, emphasize unmyelinated C-tactile (CT) afferents in hairy skin, tuned to gentle, slow stroking and projecting to affect-related cortical regions [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Because CT afferents are sparse in glabrous fingertips, CT-related interpretations are most directly applicable when stimuli are explored on hairy skin (e.g., forearm) or when devices emulate low-force stroking. Accordingly, we treat CT physiology as a motivation for parameter regimes (smooth, low-gradient surfaces) rather than as a claim of direct activation under all contact conditions.\u003c/p\u003e \u003cp\u003eFinally, physiological plausibility also depends on physical scaling and safety. Although Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e provides an example relief range, any deployment should bound maximum slope and peak curvature to avoid nociceptive stimulation and to remain compatible with the target fabrication resolution [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Interpretation drift and creative affordances\u003c/h2\u003e \u003cp\u003eBecause the mapping is continuous, boundary regions between affect quadrants can produce ambiguous textures that invite interpretation. This may be valuable in digital art and therapy-oriented interaction design, where controlled ambiguity supports reflection rather than categorical labeling. Such boundary behaviors also resonate with accounts of unexpected creativity in generative systems [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e4.3 Validation without immediate user studies: physical auditability\u003c/h2\u003e \u003cp\u003eA common critique in human-computer interaction (HCI) and multimodal-interface venues is the lack of human-subject validation: do the generated textures actually induce the intended affect? We agree that psychophysics is ultimately necessary, but argue that it should be preceded by a stimulus-generation foundation: stimuli must be controllable, reproducible, and physically comparable across devices and fabrication. Accordingly, we complement the computational audit (Section \u003cspan refid=\"Sec10\" class=\"InternalRef\"\u003e3.2\u003c/span\u003e) with a no-participant physical verification protocol (Section \u003cspan refid=\"Sec11\" class=\"InternalRef\"\u003e3.3\u003c/span\u003e) grounded in surface metrology and interaction dynamics [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAt the geometry level, fabricated surfaces can be measured via profilometry and summarized with ISO areal parameters (Sa, Sq, Sdq, Sal) and PSD descriptors [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. Such descriptors are widely used to link precisely specified 3D-printed geometry to roughness perception, making them particularly suitable for fabrication-oriented validation [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. This enables falsifiable checks that are independent of subjective judgment, such as: (i) whether the realized surface matches the target spectral regime implied by (V, A), and (ii) whether the mapping preserves separability under different printers/materials (cross-fabrication robustness).\u003c/p\u003e \u003cp\u003eAt the interaction level, controlled scanning connects spatial geometry to temporal vibration and friction signals that are known to support texture perception. Prior work reports that fine textures are strongly mediated by vibration codes and friction-induced vibrations, and that scanning speed systematically shifts vibration spectra [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. Therefore, measuring vibration PSD bands and friction coefficients under standardized v and normal force provides an objective bridge between our spatial parameterization and the temporal signals available to mechanoreceptors.\u003c/p\u003e \u003cp\u003eImportantly, these measurements also support safety-relevant bounds. Sharp peaks and high gradients can become uncomfortable or even noxious at larger physical scales; metrology can quantify maximum slope and curvature to keep stimuli within safe regimes before any user exposure [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e4.4 From heuristic mapping to calibratable models\u003c/h2\u003e \u003cp\u003eOur current mapping from (V, A) to (f, s, gamma) is intentionally simple and interpretable. It should be read as an empirical initialization rather than a universal law: the value is that the mapping is parameterized and thus calibratable. Calibration can be performed against either (i) physical measurements (e.g., targeting a desired PSD centroid or Sdq range) or (ii) psychophysical ratings once human studies are conducted [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eFurthermore, tactile experience is multi-dimensional: beyond roughness, factors such as warmth/coldness, compliance, and friction contribute to material perception [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. This suggests a natural extension of Affective Tactility toward multi-objective mappings and richer descriptors, while retaining the modular structure of text-to-affect estimation (Section \u003cspan refid=\"Sec5\" class=\"InternalRef\"\u003e2.2.1\u003c/span\u003e) and procedural rendering.\u003c/p\u003e \u003c/div\u003e"},{"header":"5. Limitations and future work","content":"\u003cp\u003eThis manuscript reports a proof of concept: we validate the mapping using controlled affect coordinates and computational descriptors (Section \u003cspan refid=\"Sec10\" class=\"InternalRef\"\u003e3.2\u003c/span\u003e), and we specify (but do not yet execute) a physical metrology and controlled-scanning protocol that enables no-participant verification (Section \u003cspan refid=\"Sec11\" class=\"InternalRef\"\u003e3.3\u003c/span\u003e). End-to-end text-to-(V, A) estimation is provided as a reproducible lexicon baseline (Section \u003cspan refid=\"Sec5\" class=\"InternalRef\"\u003e2.2.1\u003c/span\u003e), but more robust contextual estimators and multilingual coverage remain open problems. Future work will (i) collect physical measurements across fabrication methods and materials, (ii) calibrate mapping coefficients against metrology and/or psychophysical data, and (iii) evaluate perceived affect communication and comfort in controlled human-subject studies. When targeting clinical or assistive settings, ethics approval and safety constraints (e.g., avoiding nociceptive stimulation) will be required.\u003c/p\u003e"},{"header":"6. Conclusion","content":"\u003cp\u003eWe presented Affective Tactility, a modular framework for translating natural-language affect into procedural tactile textures. By mapping valence and arousal onto interpretable procedural parameters, the method produces manufacturable height maps that exhibit controllable qualitative differences. A reversibility check using surface descriptors supports that affect regimes correspond to distinct structural signatures. We hope this work enables accessible affect communication and new forms of embodied digital art.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eFunding:\u003c/h2\u003e \u003cp\u003eThe author did not receive support from any organization for the submitted work.\u003c/p\u003e \u003cp\u003eCompeting interests: The author has no relevant financial or non-financial interests to disclose.\u003c/p\u003e \u003cp\u003eEthics approval: Not applicable. This study did not involve human participants or animals.\u003c/p\u003e \u003cp\u003eConsent to participate: Not applicable.\u003c/p\u003e \u003cp\u003eConsent for publication: Not applicable.\u003c/p\u003e \u003cp\u003eAvailability of data and materials: All data generated in this study are synthetic and can be regenerated using Online Resource.\u003c/p\u003e \u003cp\u003eCode availability: Source code and configuration files used to generate the stimuli, figures, and tables are provided in Online Resource.\u003c/p\u003e\u003ch2\u003eAuthor contributions:\u003c/h2\u003e \u003cp\u003eConceptualization, methodology, software, formal analysis and investigation, writing (original draft), and writing (review and editing): Yuusuke Harada.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eRamesh A et al (2021) Zero-shot text-to-image generation. Proceedings of ICML. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://arxiv.org/abs/2102.12092\u003c/span\u003e\u003cspan address=\"https://arxiv.org/abs/2102.12092\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRombach R et al (2022) High-resolution image synthesis with latent diffusion models. CVPR. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://arxiv.org/abs/2112.10752\u003c/span\u003e\u003cspan address=\"https://arxiv.org/abs/2112.10752\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMacLean KE (2008) Haptic interaction design for everyday interfaces. Reviews Hum Factors Ergon 4(1):149\u0026ndash;194. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1518/155723408X342826\u003c/span\u003e\u003cspan address=\"10.1518/155723408X342826\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbraira VE, Ginty DD (2013) The sensory neurons of touch. Neuron 79(4):618\u0026ndash;639. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.neuron.2013.07.051\u003c/span\u003e\u003cspan address=\"10.1016/j.neuron.2013.07.051\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVaswani A et al (2017) Attention is all you need. NeurIPS. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://arxiv.org/abs/1706.03762\u003c/span\u003e\u003cspan address=\"https://arxiv.org/abs/1706.03762\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDong Y et al (2023) Text-driven texture synthesis via diffusion models. ICCV. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://arxiv.org/abs/2303.11396\u003c/span\u003e\u003cspan address=\"https://arxiv.org/abs/2303.11396\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcGlone F, Wessberg J, Olausson H (2014) Discriminative and affective touch: sensing and feeling. Neuron 82(4):737\u0026ndash;755. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.neuron.2014.05.020\u003c/span\u003e\u003cspan address=\"10.1016/j.neuron.2014.05.020\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSpence C (2011) Crossmodal correspondences: A tutorial review. Attention, Perception, \u0026amp; Psychophysics, 73(4), 971\u0026ndash;995. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3758/s13414-010-0073-7\u003c/span\u003e\u003cspan address=\"10.3758/s13414-010-0073-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRussell JA (1980) A circumplex model of affect. J Personal Soc Psychol 39(6):1161\u0026ndash;1178. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1037/h0077714\u003c/span\u003e\u003cspan address=\"10.1037/h0077714\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePerlin K (1985) An image synthesizer. ACM SIGGRAPH Comput Graphics 19(3):287\u0026ndash;296. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/325165.325247\u003c/span\u003e\u003cspan address=\"10.1145/325165.325247\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRamachandran VS, Hubbard EM (2001) Synaesthesia: A window into perception, thought and language. J Conscious Stud 8(12):3\u0026ndash;34\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCortes C, Vapnik V (1995) Support-vector networks. Mach Learn 20(3):273\u0026ndash;297. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/BF00994018\u003c/span\u003e\u003cspan address=\"10.1007/BF00994018\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePerl ER (2007) Ideas about pain, a historical view. Nat Rev Neurosci 8(1):71\u0026ndash;80. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/nrn2042\u003c/span\u003e\u003cspan address=\"10.1038/nrn2042\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVallbo AB, Olausson H, Wessberg J (1999) Unmyelinated afferents constitute a second system coding tactile stimuli of the human hairy skin. J Neurophysiol 81(6):2753\u0026ndash;2763. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1152/jn.1999.81.6.2753\u003c/span\u003e\u003cspan address=\"10.1152/jn.1999.81.6.2753\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOlausson H et al (2002) Unmyelinated tactile afferents signal touch and project to insular cortex. Nat Neurosci 5(9):900\u0026ndash;904. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/nn896\u003c/span\u003e\u003cspan address=\"10.1038/nn896\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBoden MA (2004) The Creative Mind: Myths and Mechanisms. Routledge\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKlatzky RL, Lederman SJ (2003) Touch. In Handbook of Psychology (Vol. 4), 147\u0026ndash;176\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBradley MM, Lang PJ (1994) Measuring emotion: The Self-Assessment Manikin and the Semantic Differential. J Behav Ther Exp Psychiatry 25(1):49\u0026ndash;59. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/0005-7916(94)90063-9\u003c/span\u003e\u003cspan address=\"10.1016/0005-7916(94)90063-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWarriner AB, Kuperman V, Brysbaert M (2013) Norms of valence, arousal, and dominance for 13,915 English lemmas. Behav Res Methods 45:1191\u0026ndash;1207. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3758/s13428-012-0314-x\u003c/span\u003e\u003cspan address=\"10.3758/s13428-012-0314-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eL\u0026ouml;ken LS, Wessberg J, Morrison I, McGlone F, Olausson H (2009) Coding of pleasant touch by unmyelinated afferents in humans. Nat Neurosci 12(5):547\u0026ndash;548. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/nn.2312\u003c/span\u003e\u003cspan address=\"10.1038/nn.2312\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWeber AI, Saal HP, Lieber JD, Cheng J-W, Manfredi LR, Dammann JF (2013) and S. J. Bensmaia. Spatial and temporal codes mediate the tactile perception of natural textures. Proceedings of the National Academy of Sciences, 110(42):17107\u0026ndash;17112. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1073/pnas.1305509110\u003c/span\u003e\u003cspan address=\"10.1073/pnas.1305509110\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eS. M. Mohammad. Obtaining reliable human ratings of valence, arousal, and dominance for 20,000 English words. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, pages 174\u0026ndash;184, (2018) \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.18653/v1/P18-1017\u003c/span\u003e\u003cspan address=\"10.18653/v1/P18-1017\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHollins M, Risner SR (2000) Evidence for the duplex theory of tactile texture perception. Percept Psychophys 62(4):695\u0026ndash;705. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3758/BF03211625\u003c/span\u003e\u003cspan address=\"10.3758/BF03211625\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJohansson RS, Flanagan JR (2009) Coding and use of tactile signals from the fingertips in object manipulation tasks. Nat Rev Neurosci 10(5):345\u0026ndash;359. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/nrn2621\u003c/span\u003e\u003cspan address=\"10.1038/nrn2621\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eK. O. Johnson. The roles and functions of cutaneous mechanoreceptors. Curr Opin Neurobiol, 11(4):455\u0026ndash;461, (2001) \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S0959-4388(00)00234-8\u003c/span\u003e\u003cspan address=\"10.1016/S0959-4388(00)00234-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eISO 25178-2:2012 (2012) Geometrical product specifications (GPS) - Surface texture: Areal - Part 2: Terms, definitions and surface texture parameters. International Organization for Standardization\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eD. J. Whitehouse. Handbook of Surface Metrology. Institute of Physics Publishing, (1994)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSkedung L, Danerl\u0026ouml;v K, Olofsson U, Johannesson CM, Aikala M, Kettle J, Arvidsson M, Berglund B (2011) Finger friction, surface roughness and perceived coarseness of printed surfaces. Tribol Int 44:505\u0026ndash;512. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.triboint.2010.04.010\u003c/span\u003e\u003cspan address=\"10.1016/j.triboint.2010.04.010\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDelhaye BP, Hayward V, Lef\u0026egrave;vre P (2012) Thonnard. Texture-induced vibrations in the forearm during tactile exploration. Front Behav Neurosci 6:37. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fnbeh.2012.00037\u003c/span\u003e\u003cspan address=\"10.3389/fnbeh.2012.00037\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBensmaia SJ, Hollins M (2005) Pacinian representations of fine surface texture. Percept Psychophys 67(5):842\u0026ndash;854. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3758/BF03193537\u003c/span\u003e\u003cspan address=\"10.3758/BF03193537\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCascio CJ, Sathian K (2001) Temporal cues contribute to tactile perception of roughness. J Neurosci 21(14):5289\u0026ndash;5296. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1523/JNEUROSCI.21-14-05289.2001\u003c/span\u003e\u003cspan address=\"10.1523/JNEUROSCI.21-14-05289.2001\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOkamoto S, Nagano H, Yamada Y (2013) Psychophysical dimensions of tactile perception of textures. IEEE Trans Haptics 6(1):81\u0026ndash;93. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/TOH.2012.32\u003c/span\u003e\u003cspan address=\"10.1109/TOH.2012.32\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSkedung L, Arvidsson M, Chung JY, Johannesson CM, Olofsson U (2013) Berglund. Feeling small: Exploring the tactile perception limits. Sci Rep 3:2617. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/srep02617\u003c/span\u003e\u003cspan address=\"10.1038/srep02617\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTymms C, Zorin D, Gardner EP (2018) Tactile perception of the roughness of 3D-printed textures. J Neurophysiol 119(3):862\u0026ndash;876. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1152/jn.00564.2017\u003c/span\u003e\u003cspan address=\"10.1152/jn.00564.2017\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"Hiroshima University","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"haptics, affective computing, procedural texture, crossmodal correspondence, natural language processing","lastPublishedDoi":"10.21203/rs.3.rs-8696852/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8696852/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eText-to-image generation has rapidly matured, yet translating paralinguistic cues in text into physically perceivable tactility remains underexplored. We present Affective Tactility, a procedural framework that maps natural-language input to affect coordinates, valence (V) and arousal (A), and renders a manufacturable height map through an interpretable procedural texture generator. For end-to-end reproducibility, we describe a transparent lexicon-based baseline for estimating (V, A) and also validate the downstream mapping using controlled affect coordinates. The design is grounded in crossmodal correspondences (e.g., bouba\u0026ndash;kiki) and is physiologically motivated by known tactile pathways. As a proof of concept, we demonstrate qualitative diversity of generated height maps and computational auditability: two simple descriptors, root-mean-square (RMS) roughness (Rq) and a spatial-frequency centroid, enable linear support vector machine (SVM) separation between affect regimes with 96.7% test accuracy, with ablations and a permuted control reported to address trivial separability. Beyond purely computational checks, we outline a no-participant physical verification protocol based on surface metrology and controlled scanning dynamics (profilometry, vibration, and friction measurements), positioning the method as a stimulus-generation foundation for future psychophysics and accessible haptic media.\u003c/p\u003e","manuscriptTitle":"Affective Tactility: Cross-Modal Translation from Natural Language to Procedural Height Maps","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-01-29 16:19:37","doi":"10.21203/rs.3.rs-8696852/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"2e6529a0-6895-4f8e-ab75-049dbebd29ac","owner":[],"postedDate":"January 29th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-01-29T16:19:37+00:00","versionOfRecord":[],"versionCreatedAt":"2026-01-29 16:19:37","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8696852","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8696852","identity":"rs-8696852","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}