Dramaturge: The Art of Structural Coherence

• Dramaturge is a specialist who organizes dramatic structure by unifying text interpretation, rehearsal dynamics, and performance coherence.
• The role spans traditional staging to computational dramaturgy, leveraging machine learning pipelines and interactive narrative systems.
• This multidisciplinary function integrates formal analysis, space design, and algorithmic revision to enhance immersive dramatic experiences.

Searching arXiv for papers related to dramaturgy and dramaturge. A dramaturge is a practitioner concerned with the organization, interpretation, and coordination of dramatic structure across text, rehearsal, staging, and performance systems. In the materials surveyed here, the term appears both in its established theatrical sense—where a text requires “exegesis” and a synthetic organization of roles, scenes, and transformations—and in expanded computational senses, where dramaturgical functions are distributed across machine-learning pipelines, interactive environments, avatar systems, robotic rehearsal workflows, and multi-agent narrative architectures (0912.4879, Panagiotidis et al., 6 Feb 2026, Xie et al., 6 Oct 2025). Across these contexts, the dramaturge’s work centers on relations among action, character, space, pacing, scenic logic, and audience-facing coherence, whether the medium is dramatic text, mixed reality, responsive architecture, or LLM-mediated script refinement (Gagneré, 2024, Panagiotidis et al., 7 Jul 2026, Panagiotidis et al., 14 Nov 2025).

1. Historical and conceptual definitions

In the most explicitly theoretical account in this corpus, traditional staging begins from the proposition that interpretation “n'est pas immanente au texte” and that a text cannot exist on stage without the interpretive labor of the director; this labor is described as beginning with “l'établissement d'une ontologie synthétique de la dramaturgie” (0912.4879). In that formulation, dramaturgy consists in identifying character types, situations, scenic relations, and possible transformations, then organizing how those are instantiated in performance. This associates the dramaturge with classification, formal specification, and the management of “variations d'instanciation” across a play’s unfolding (0912.4879).

Several papers extend that conception rather than replacing it. One work on responsive architecture treats dramaturgy as the shaping of relations among action, space, atmosphere, intention, audience experience, and unfolding event structure, with architecture itself becoming “dramaturgical material” (Panagiotidis et al., 6 Feb 2026). Another rehearsal-oriented system for immersive theatre argues that sensing and actuation should be available as “dramaturgical and compositional material” during devising, not only as fixed infrastructure for finished productions (Panagiotidis et al., 7 Jul 2026). A study of robotic performance frames expressive behavior as a matter of rehearsal, interpretation, intention, and scenic meaning rather than as a purely technical animation problem (Panagiotidis et al., 5 Aug 2025). Taken together, these formulations suggest that the dramaturge’s domain is not limited to literary analysis; it includes the operational conditions under which performance becomes coherent, legible, and theatrically meaningful.

A further extension appears in computational narrative systems. In one multi-agent script-refinement framework titled “Dramaturge,” the central problem is how to revise long scripts by coordinating global structure and local detail without introducing inconsistencies (Xie et al., 6 Oct 2025). In another framework for interactive drama, a director agent and actor agents divide responsibilities between plot control and role fidelity (Han et al., 2024). These systems do not define the dramaturge as a human profession, but they instantiate recognizably dramaturgical functions: diagnosing structural weaknesses, maintaining coherence across scenes, shaping objectives and obstacles, and preserving the relation between high-level strategy and local execution (Xie et al., 6 Oct 2025, Han et al., 2024).

2. Core functions in text, authorship, and dramatic structure

One central dramaturgical function is the interpretation of dramatic text and its relation to authorship. In stylometric studies of drama, this appears as the effort to distinguish authorial style, character voice, and translational mediation. A study of Ibsen defines characterization as “the distinctiveness of textual contributions of characters” and operationalizes this through speaker-attributed dialogue, chunked by character and compared for internal coherence (Lynch et al., 2015). The dramaturgical relevance is direct: translated drama may preserve the play-level stylistic world while weakening the distinctiveness of individual roles (Lynch et al., 2015). A plausible implication is that the dramaturge’s concern with role differentiation can be examined computationally at the level of dialogue distribution.

Authorship attribution is addressed even more directly in the study of Juro Jánošík, where computational stylometry is used to clarify whether the play belongs within Milan Kundera’s dramatic oeuvre (Jungmannová et al., 2022). That paper matters for dramaturgy because the attribution alters how the play is read in relation to Kundera’s montage construction, divided space-time, simultaneous dialogue, political irony, and substitute authorship under censorship (Jungmannová et al., 2022). Here the dramaturge’s traditional tasks—establishing authorial poetics, historical conditions of theatrical circulation, and formal continuity across works—are supplemented by supervised machine learning, but not replaced by it.

Structural comparison across genres also bears on dramaturgical analysis. A network-based study of comedy and tragedy models plays through co-presence networks, speech distribution, characters-per-scene, clustering, density, diameter, maximum betweenness, and degree-based ratios (Botond et al., 2023). Its results indicate that comedies tend toward denser, more distributed ensemble structures, while tragedies tend toward more segmented, brokered, and unequal structures (Botond et al., 2023). This suggests that dramaturgical distinctions between genres can be described not only semantically but as formal organizations of stage society: who appears together, how speech is distributed, and whether one role mediates otherwise separated subgroups (Botond et al., 2023).

3. Rehearsal, devising, and the rehearsal room

Several papers reposition the dramaturge within rehearsal rather than textual interpretation alone. A system for drama-in-education, CO-OPERA, organizes playwriting into staged components—logline, characters, plot, scenes, and dialogue—using a conversational tutor for divergent thinking and drama-element agents for convergent thinking (Ma et al., 1 Jun 2025). Its value lies in helping users focus on “whole logical narrative development” rather than isolated lines, which closely resembles dramaturgical scaffolding (Ma et al., 1 Jun 2025). The paper does not formalize dramaturgy as theory, but its workflow embodies a dramaturgical sequence: premise refinement, character construction, plot causality, scene decomposition, and dialogue drafting (Ma et al., 1 Jun 2025).

A more explicitly rehearsal-centered model appears in “Theatre in the Loop,” where robotic behavior is shaped through a director-guided puppeteering workflow consisting of five parts: “Define Expressive Goal (Director),” “Embodied Improvisation (Puppeteer),” “Motion Capture (Roboticists),” “Collaborative Review (All),” and “Curation of Expressive Motion Dataset” (Panagiotidis et al., 5 Aug 2025). The paper repeatedly frames this as a rehearsal workflow rather than a data pipeline, and the dramaturgical content lies in objectives, obstacles, “inner monologue,” tension-release dynamics, hesitation, and sustained aggression (Panagiotidis et al., 5 Aug 2025). This suggests that the dramaturge’s role can be distributed across interdisciplinary teams but remains recognizable where scenic meaning is built through rehearsal, revision, and structured improvisation.

Rehearsal-oriented work on immersive environments makes the same point from the side of space and scenography. One paper reports six workshops with eight professional performance-makers using a modular no-code system that maps gesture, position, and speech to light and sound (Panagiotidis et al., 7 Jul 2026). Across the workshops, the process moved from separate ensemble and technical training to integrated dramaturgy, culminating in a single-spectator scratch performance (Panagiotidis et al., 7 Jul 2026). The paper identifies three ensemble-level strategies: rotating roles between operator, performer, and mediator; embracing controlled imperfection as a creative resource; and using technology-describing metaphors such as “another performer” or “co-designer” (Panagiotidis et al., 14 Nov 2025). A plausible implication is that the dramaturge, in such settings, becomes the organizer of conditions under which responsive environments can be rehearsed as part of the composition itself.

4. Space, scenography, and nonhuman performance partners

The dramaturge’s scope expands further in work on avatars and responsive architecture. In “The Shadow,” Georges Gagneré stages Andersen’s tale through five shadow avatars and one physical narrator, using a mixed arrangement in which virtual figures first appear as a shadow audience and then become scenic agents (Gagneré, 2024). The paper’s key concept is coevolution, derived from Edmond Couchot and used to describe how director, mocaptor, narrator, and avatars transform one another during rehearsal and performance (Gagneré, 2024). One practical dramaturgical innovation is the decomposition of movement into salient and idle segments, which allows cue-driven performance while preserving scenic availability and rhythmic flexibility (Gagneré, 2024). This is dramaturgy as score design: shaping intention, pause, transition, and the simulated autonomy of virtual partners.

AKN_Regie extends this logic by providing a cue-based authoring tool for directing avatars on a mixed stage (Gagneré et al., 2024, Gagneré, 2024). Its workflow builds digital scenery in Unreal Engine, places GoalAvatar, GoalProp, and GoalCamera markers, and organizes actions through a Regie_Cuesheet (Gagneré et al., 2024). The paper explicitly frames the tool as making avatar direction analogous to the direction of physical actors and usable by “non-programmer theatrical people” (Gagneré et al., 2024). Because cues, positions, props, fades, animations, and effects are organized through a theater-oriented interface, the dramaturgical function lies in coordinating heterogeneous scenic entities—actors, avatars, props, cameras, and effects—within a single mise-en-scène (Gagneré, 2024).

Responsive architecture pushes this still further. “Directing Space” proposes that interactive architecture can be rehearsed “as a performer,” with a system that senses spatial behavior and speech, interprets them through an LLM, and generates real-time environmental actions (Panagiotidis et al., 6 Feb 2026). The system’s architecture includes a dramaturgical layer, a directorial layer, a dramaturgical interpreter, dramaturgical memory, reasoning traces, and a final environment “score” (Panagiotidis et al., 6 Feb 2026). These traces are treated not merely as technical explanation but as rehearsal notes: they help artists inspect whether the system’s behavior aligns with the conceptual world, affective tone, and directorial rules of the performance (Panagiotidis et al., 6 Feb 2026). In this setting, the dramaturge’s role plausibly expands toward articulating the “inner world” of the space itself.

5. Computational dramaturgy and AI systems

A distinct line of work formalizes dramaturgical labor as algorithmic coordination. “Text-to-Stage” frames adaptation as a narrative-to-play task: given prose without explicit blocking, infer scene boundaries, speaker labels, a canonical $3 \times 3$ stage grid, movement, and room metadata (Hernandez et al., 18 Mar 2026). The system uses a deterministic evaluator with components such as structural validity, text grounding, spatial composition, movement coherence, and scene structure, and improves a Qwen3-8B backbone through rejection SFT and RL from verifiable rewards via GRPO (Hernandez et al., 18 Mar 2026). For dramaturgy, the significance is that blocking, proxemics, scene transition logic, and movement economy are made operationally measurable, though within a highly specific proscenium-style spatial logic (Hernandez et al., 18 Mar 2026).

Interactive drama systems distribute similar functions across agents. IBSEN defines a sequence of plot objectives $\langle G_1, G_2, \dots \rangle$, has a director generate an outline $S_G$ and a latent script $\langle \hat{T_1}, \hat{T_2}, \dots \rangle_G$, and lets actor agents produce actual lines conditioned by character profile, memory, and indirect instruction (Han et al., 2024). This architecture explicitly negotiates the tension between actor autonomy and plot control, which is a central dramaturgical problem (Han et al., 2024). OPEN-THEATRE generalizes this into a toolkit with one-for-all, director-actor, hybrid, and Director-Global Actor architectures, along with hierarchical memory stores for identity, events, summaries, and archives (Xu et al., 20 Sep 2025). The toolkit is scene-based, script-driven, and designed for experimentation with plotlines, character motivations, and architecture choice (Xu et al., 20 Sep 2025).

HAMLET develops these ideas for live embodied theatrics (Chen et al., 21 Jul 2025). It generates a narrative blueprint from a simple topic using actor designer, plot designer, reviewer, and director agents, then executes online performance through actors with autonomous minds, a PAD module choosing between FAST, SLOW, and SILENCE strategies, a narrator that adjudicates physical actions, and planner, transfer, and advancer agents that maintain progression (Chen et al., 21 Jul 2025). The paper defines a beat as “an effective interaction step in which an actor takes an effective action,” and the path between two points as a dynamically generated trajectory composed of beats (Chen et al., 21 Jul 2025). This is a fully operational dramaturgical architecture: acts, scenes, points, flags, results, private goals, plot milestones, embodied action, and runtime intervention are all made explicit.

The script-refinement system named “Dramaturge” is perhaps the clearest computational appropriation of the title itself (Xie et al., 6 Oct 2025). It is a plug-and-play post-processing framework with three stages—Global Review, Scene-level Review, and Hierarchical Coordinated Revision—organized as a coarse-to-fine iterative process until no further substantive improvements occur (Xie et al., 6 Oct 2025). It decomposes revision by task and narrative feature, using specialized evaluators for engagement, character, theme, and narrative flow at the global level, then dialogue, scene description, plot, and character inspection at the scene level (Xie et al., 6 Oct 2025). The core dramaturgical principle is top-down coordination: high-level strategies constrain local changes so that revisions improve structure without destroying coherence (Xie et al., 6 Oct 2025).

6. Misconceptions, tensions, and limits

A common misconception is that dramaturgy concerns only scripts and literary interpretation. The papers here consistently contradict that reduction. Dramaturgical work appears in authorship studies (Jungmannová et al., 2022), translation analysis (Lynch et al., 2015), rehearsal protocols (Panagiotidis et al., 5 Aug 2025), responsive environments (Panagiotidis et al., 6 Feb 2026), mixed-reality staging (Gagneré, 2024), robotic performance (Panagiotidis et al., 5 Aug 2025), and interactive narrative systems (Han et al., 2024, Xu et al., 20 Sep 2025, Chen et al., 21 Jul 2025). Another misconception is that computational systems eliminate dramaturgy; the surveyed work suggests the opposite. Quantitative methods often sharpen dramaturgical suspicion, make staging principles measurable, or distribute dramaturgical functions across new technical layers rather than abolishing them (Jungmannová et al., 2022, Hernandez et al., 18 Mar 2026, Xie et al., 6 Oct 2025).

At the same time, the literature is explicit about limits. Stylometric studies of drama face small corpora, short texts, and the difficulty of isolating authorial signals from editorial or theatrical conventions (Jungmannová et al., 2022, Lynch et al., 2015). Interactive and mixed-reality systems face latency, limited spatial reasoning, tracking errors, and the gap between simulation and embodied material presence (Panagiotidis et al., 6 Feb 2026, Panagiotidis et al., 7 Jul 2026, Panagiotidis et al., 14 Nov 2025). LLM-based drama systems still struggle with repetition, over-narration, softened conflict, and imperfect objective checking (Han et al., 2024). Text-to-stage models privilege one blocking grammar—downstage salience, triangular composition, sparse movement—that may suppress other theatrical traditions (Hernandez et al., 18 Mar 2026). This suggests that computational dramaturgy is most productive when treated as a bounded formalization of specific theatrical problems rather than as a general substitute for rehearsal, staging, or interpretation.

A further tension concerns authorship. When avatars, robots, environments, or LLM agents behave with partial autonomy, authorship becomes distributed across code, prompts, recordings, performers, and live timing (Gagneré, 2024, Panagiotidis et al., 6 Feb 2026, Panagiotidis et al., 5 Aug 2025, Chen et al., 21 Jul 2025). These systems do not resolve the question of who “writes” the performance; rather, they reorganize it. A plausible implication is that the dramaturge’s role increasingly includes mediating among heterogeneous agencies: textual, technical, embodied, and algorithmic.

7. Contemporary scope of the dramaturge

Across the works considered here, the dramaturge emerges as a figure concerned with structural coherence across multiple scales. At the smallest scale, this includes line, beat, gesture, cue, and scenic timing (Gagneré, 2024, Panagiotidis et al., 5 Aug 2025). At the meso scale, it includes scene composition, actor objectives, spatial arrangement, and relations among characters, props, and environment (Hernandez et al., 18 Mar 2026, Panagiotidis et al., 6 Feb 2026, Chen et al., 21 Jul 2025). At the largest scale, it includes authorship, oeuvre formation, genre, adaptation, and the iterative revision of long-form scripts (Jungmannová et al., 2022, Botond et al., 2023, Xie et al., 6 Oct 2025).

What unifies these roles is not a single medium or institutional job description, but a recurring set of operations: defining and testing objectives; articulating motivations and constraints; preserving continuity while permitting variation; coordinating local decisions with global form; and shaping how a performance world becomes legible to its audience. In classical terms, this may begin with “une ontologie synthétique de la dramaturgie” (0912.4879). In contemporary computational settings, it may take the form of scene-based memory systems, hierarchical review pipelines, reasoning traces, or beat-level runtime controllers (Xu et al., 20 Sep 2025, Xie et al., 6 Oct 2025, Chen et al., 21 Jul 2025). The continuity across these domains suggests that the dramaturge remains, above all, a specialist in the formal organization of dramatic experience.