Natural language modelled and printed in 3D: a multi-disciplinary approach

The paper asks how aspects of the high-dimensional form of natural language can be rendered in 3D. It motivates the question by contrasting familiar visualizations of biological and physical forms (e.g., DNA, viruses) with the lack of visual models for language’s form. Building on transcript-based representations of spoken discourse—organized into lines that reflect pauses and intonation—the authors conceptualize language beyond text on a page. Historically, grammar has been compared to geometry, suggesting that languages have a shape determined by structural principles. The study selects a single grammatical feature—evidentiality, which marks the source of information (seen, heard, inferred, reported, assumed)—as a third dimension often lost in translation. Using contemporary computational design tools, the authors aim to model evidentiality as a spatial dimension, producing four 3D prototypes from short samples in Kurdish, an Amazonian language (Tariana), Neo-Assyrian Akkadian, and American English. The significance lies in bridging linguistics and computational design to make aspects of linguistic structure perceptible and comparable as 3D forms.

The study draws on several strands of literature: (1) Linguistics and anthropology of language: classic and contemporary work on evidentiality and linguistic relativity (Boas, Jakobson, Hymes, Chafe, Aikhenvald), the analysis of narratives by pattern and texture (Hymes; Johnstone), and discourse transcription conventions. Evidentiality is grammar-encoded in many languages (especially Native American, Amazonian, Australian Aboriginal), with language- and culture-specific hierarchies of evidential value. (2) Design and computational modeling: history and techniques of spline-based and parametric design (Gaudí’s weighted string models and inversion; Hooke’s principle; Frei Otto’s form-finding and minimal surfaces; contemporary spline/parametric modeling). These inform the conceptualization of evidentials as weights exerting a gravitational pull on a digital weave. (3) Conceptual metaphors and materializations of language: the longstanding analogy between grammar and geometry; ‘text’ as woven fabric; visualization beyond 2D graphs to 3D geometric distortions that encode ‘levels of value’ on the z-axis. Together, these bodies of work justify mapping transcripts into Cartesian coordinates (timeline, syllabic structure, evidential weight) to generate a 3D ‘reality-effect’ where language strands dock into perceptual anchor points.

- Data selection and preparation: Short language samples were chosen from four sources (Kurdish narrative, Tariana narrative, Neo-Assyrian Akkadian letter, American English political debate). For spoken samples, transcripts reflect pauses and intonation; for Akkadian, lineation follows the cuneiform tablet lines. Evidentials in each sample were identified and assigned relative numerical weights based on expert analyses and language-specific hierarchies (e.g., Tariana categories: visual, non-visual, inferred, reported, assumed). Coding was manually formatted in the transcript (e.g., square brackets with numeric weights). - Data ingestion: The formatted transcript was entered into an Excel spreadsheet, with each transcript line as a row and positional data for evidential weights. A Grasshopper script (Rhino 3D) parsed the spreadsheet, computed syllable counts per line, and extracted evidential weight values and locations. - Coordinate mapping (Cartesian framework): The transcript was mapped onto 3D coordinates: x-axis = timeline (progression of speech across phrases/lines); y-axis = number of syllables per line; z-axis = evidential weight (relative strength/type of evidence at marked positions). The x–y grid forms a digital warp–weft weave of the transcript’s elementary structure; evidential weights translate into control points along splines that deflect the surface along z. - Wireframe generation: Control points (virtual evidential weights) were used to construct spline curves, generating a wireframe that encodes the transcript’s structure and evidential anchors. The model reflects parametric dependencies, so importing a different transcript regenerates the geometry without rebuilding from scratch. - Surface realization: The wireframe was virtually ‘woven’ into a continuous, undulating surface whose drape/fold characteristics are driven by the measured coordinates and parametric rules, turning the threadlike weave into a smooth, fabric-like 3D surface. - Inversion for perceptual proximity: Following historical parametric design (e.g., Gaudí’s inverted string models), the model was inverted so that segments with stronger evidential grounding protrude towards the viewer, mimicking perceptual closeness. - Visualization and 3D printing: Models were rendered digitally; prototypes were 3D-printed using Selective Laser Sintering (nylon, PA12) and a silver alloy; later, Multi Jet Fusion prints captured finer details. Blue highlights indicate reported speech in applicable models. - Software and tools: Rhino 3D with Grasshopper for visual programming; Excel for data structuring. The approach treats evidential markers and syllabic structure as adjustable parameters within a parametric pipeline.

- Feasibility of 3D language modeling: The study demonstrates that a single grammatical dimension—evidentiality—can be encoded as a third spatial dimension (z-axis) mapped onto transcript structure (x: timeline; y: syllables per line), producing meaningful 3D geometries that reflect evidential dynamics. - Four prototypes reveal language-specific evidential patterns: • Kurdish (Kurmancî): Regular undulations indicate sustained evidential presence; blue highlights mark direct reported speech (word-for-word quotations). The model’s protrusions correspond to explicit ties to perceptible reality. • Tariana (Amazonian): The sample comprises 120 evidentials over 33 lines (from the narrative “The Yanomami”). Crests appear across nearly every section, visualizing obligatory grammatical evidentials (visual, non-visual, reported). This yields a densely textured surface consistent with a language where evidentials are required line by line. • Neo-Assyrian Akkadian: A letter (“The Crimes of Guzana”) shows highly textured middle sections with reliance on reported information; blue highlights indicate reported speech, mapping how complaints are anchored in evidence. • American English (US presidential debate, 2004): A monochrome, gently undulating surface reflects the absence of direct reported speech in the sample and the non-obligatory status of evidentiality in English. - Physical prototyping: Successful 3D prints were achieved in nylon (Selective Laser Sintering) and a silver alloy; later nylon prints via Multi Jet Fusion captured fine details. The physical objects underscore the models’ textile-like qualities. - Visualization outcome: The models function as ‘reality-effects’—digitally woven fusions of utterance and perceptible reality—making tacit aspects of evidential structure perceptible, especially for non-speakers and for endangered languages where evidential nuance is often lost in translation.

The findings directly address the research question by showing that grammatical features—specifically evidentiality—can be mapped parametrically to produce 3D geometries that convey how language ties to perception. This operationalizes the longstanding analogy between grammar and geometry, turning abstract ‘form principles’ into viewable and tangible structures. The contrast between Tariana’s dense, crest-rich surface and English’s smoother, monochrome surface exemplifies linguistic relativity: languages differ in how they grammatically encode sources of knowledge. The Kurdish and Akkadian models illustrate genre- and context-specific uses of reported speech and evidential anchoring. Beyond linguistics, the work extends parametric design’s ‘form-finding’ tradition to an ephemeral medium—spoken language—by treating evidentials as weights shaping a digital fabric. The 3D ‘reality-effect’ reframes language not as text laid on a plane but as a woven, spatial phenomenon threaded into observable reality. This approach has implications for documenting endangered languages, enhancing understanding of evidential systems, and generating tacit knowledge through material models that conventional 2D graphs or transcripts cannot provide.

The study establishes a pipeline that converts transcripts into 3D, fabric-like geometries by mapping timeline (x), syllabic structure (y), and evidential weight (z), and demonstrates its viability through four prototypes across diverse languages. It contributes to computational design (by extending parametric form-finding to linguistic data) and to the study of evidentiality (by visualizing its deployment over utterances). Physical 3D prints in nylon and silver validate the materialization of these forms, though pushing current manufacturing limits. Future work includes: importing larger datasets; generating animated forms that evolve over time; modeling additional grammatical ‘placeholders’ (e.g., pronouns, deictics such as this/that); refining algorithms and materials to capture finer linguistic ‘fibers’; and exploring how such visualizations complement linguistic analysis and archiving, especially for endangered languages.

- Social context not modeled: While anthropological literature emphasizes social factors in evidential use, the models are based solely on recorded speech events and transcript structure; sociocultural variables were not incorporated. - Scope restricted to one grammatical dimension: Only evidentiality (plus syllabic structure) was modeled; other grammatical or discourse features (e.g., pronouns, deixis, prosody beyond syllable count) were not included. - Small, genre-specific samples: Each prototype derives from a brief excerpt within a particular genre (narrative, letter, political debate), limiting generalizability. - Static snapshots: Models are ‘frozen in time’; dynamic evolution during real-time speech was not visualized. - Printing and material constraints: Many models were rejected due to fine detail and thin structures; successful prints required iteration and still face limits on scale and resolution. Larger datasets may exceed current fabrication capabilities. - Comparative linguistics beyond scope: The work does not engage in systematic cross-linguistic comparison of evidential systems; expert coding was used per sample but broader typological claims were not pursued.