The first time we ran a Bollywood screenplay through a structural analysis model trained primarily on Hollywood screenplay corpora, it returned a series of structural gap alerts that amounted to: this story does not follow conventional three-act structure, the midpoint is misplaced, and the act breaks are irregular. Which was accurate in the sense that the description matched the data. And completely useless, because the screenplay was doing exactly what it was designed to do within a different tradition.
That failure mode is what we set out to correct when we built the story structure analysis in Mugafi. The question of whether an AI system can actually understand Indian narrative structure is not a simple yes or no. It depends on what you mean by understanding and which traditions you are talking about. India has multiple robust storytelling traditions with distinct structural logics, and they do not all look like variations on Syd Field.
What Western Three-Act Structure Gets Wrong About Indian Film
The three-act model, in its classic Hollywood formulation, assumes a relatively linear escalation of external conflict driving the protagonist toward a clear resolution. The protagonist wants something, obstacles appear, escalation occurs, climax, resolution. Plot mechanically drives character.
Indian mainstream cinema, particularly the commercial Hindi film tradition from which much current OTT writing derives its instincts, operates on a different logic. The dramatic arc is often more episodic. Character relationships and their social context carry as much structural weight as individual wants and obstacles. Collective identity, particularly family dynamics and community belonging, can function as a structural force rather than just a backdrop. And the emotional register that counts as satisfying resolution is different: catharsis in the Hollywood model tends to be individual; in much Indian cinema it is communal.
There is also the question of tonal range. A mainstream Hindi film from any decade you choose will likely contain sequences that shift between comedy, melodrama, action, and deeply earnest emotional directness within a single film, sometimes within a single scene. A structural analysis model that reads tonal inconsistency as a structural defect is going to misread this tradition systematically.
Regional cinema adds additional structural traditions. Malayalam cinema has developed a naturalistic storytelling mode that is quite different from the musical-melodrama tradition of Hindi commercial film. Tamil cinema has its own set of structural conventions around hero-entry and interval placement. Telugu commercial cinema has evolved genre conventions around mass entertainment that have their own structural logic. These are not variations on the same model; they are distinct traditions.
How We Approached the Training Data Problem
The fundamental challenge for any structural analysis model working on Indian film is data. Annotation requires expertise, and expertise in Indian narrative structure is not uniformly distributed. There is a large body of published work on Hollywood screenplay structure. There is significantly less that has been systematically codified about the structural conventions of Indian commercial cinema in a form that is usable for NLP training.
Our approach was to work directly with practicing screenwriters and script editors who had deep familiarity with specific traditions. We ran annotation sessions that were structured not around labeling structural elements according to a pre-existing taxonomy, but around asking annotators to explain their structural intuitions in prose. Where does this story feel like it should turn? What is the narrative promise of this opening? At what point does the central relationship reach the moment of maximum pressure?
Those prose explanations became the basis for building a taxonomy from the material rather than applying an external taxonomy to it. This is slower than using an existing structural framework as a labeling guide. It is also more accurate, because the taxonomy reflects actual structural intuitions rather than retroactive compliance with a borrowed framework.
We are not claiming we have solved this problem. The annotation coverage for some regional traditions is thin. We know our structural analysis is more reliable for Hindi mainstream material than for, say, Marathi arthouse or Kannada commercial cinema, and we are explicit about that in the product. But the approach of deriving the structural framework from the material rather than imposing one is the right direction, and it has produced a meaningfully different kind of feedback than what a generic Hollywood-trained model returns.
The Panchanga and Non-Linear Structures
Indian classical narrative theory, most codified in texts like the Natyashastra, describes dramatic structure in terms quite different from Western models. The Panchanga model, for instance, describes narrative in terms of five stages (Arambha, Prayatna, Pratyasha, Niyatapti, Phalagama) that map onto a different arc logic than beginning-middle-end. These are not obscure academic references; they inform the structural intuitions of writers who were educated in or deeply familiar with classical performing arts traditions.
We are not saying every contemporary Indian screenwriter is consciously working from classical theory. Most are not. But classical structures have seeped into regional performance traditions, into folk theater, into the conventions of specific regional cinema industries, in ways that show up as structural patterns even when the writer is not consciously thinking about them. A model that cannot recognize these patterns will misread a significant portion of Indian material.
We have incorporated awareness of these classical structural patterns into the beat engine, with the understanding that they function as reference points rather than prescriptions. The goal is not to tell a writer their screenplay should follow Panchanga structure. The goal is to have a system that does not generate false positives when it encounters structure that deviates from a Hollywood template but is following a different, equally coherent logic.
What Mugafi Can and Cannot Do
We should be direct about limits. The beat engine in Mugafi is better at identifying structural gaps within a tradition it has good coverage of than at analyzing work that is genuinely hybrid or experimental in its structural approach. A film that is consciously deconstructing its own structure will generate confusing annotations. A story that works emotionally through accumulation rather than escalation may not map well onto the structural analysis we return.
The tool is also not a replacement for human script editors who understand specific regional traditions deeply. A script editor who has worked extensively in Malayalam cinema will bring contextual knowledge about convention, audience expectation, and genre history that cannot be encoded in a model at our current stage. We are a working aid, not a substitute for that expertise.
What we can do is provide structural analysis that does not default to a Hollywood-trained baseline and return generic gap alerts every time a story follows Indian narrative logic. That was the floor we needed to clear before the tool could be genuinely useful for the writers we built it for. Whether the ceiling is as high as we want it to be is an ongoing engineering and annotation challenge. But clearing the floor matters, and most of the tools currently available have not done it.
Indian writers do not need a tool that tells them their stories are structurally wrong because they do not look like an American prestige drama. They need a tool that understands what they are trying to do and can help them do it more rigorously. That is the problem we are working on.