Modeling Narrative Structure Through Emotional Trajectories

34 concepts20 papers1 proposal

When humans read a story, we do not just process a sequence of factual events or grammatical sentences. We experience an emotional journey—a series of highs and lows that define the narrative's rhythm. Traditional natural language processing often treats text locally, evaluating the sentiment of isolated sentences or tracking explicit keywords. However, to truly comprehend, summarize, or generate a compelling narrative, models must capture its global macro-structure.

This is where emotional trajectories (also called emotion arcs or sentiment flows) come in. By mapping how sentiment shifts sequentially over time, we can transform a complex text into a structured, mathematical curve that represents the story's underlying dramatic blueprint.

From Local Words to Sequential Flows

Why does this approach work? Consider how a story builds tension. A single sentence containing the word "dark" might have a negative local sentiment, but its structural role depends entirely on where it sits within the broader narrative.

Early work by Alm et al. recognized that incorporating narrative context—such as a sentence's relative position and the overall genre—allows models to capture emotional shifts that follow a broader narrative arc. Instead of viewing sentiment as a static property, text can be treated as a dynamic sequence. For instance, Nalisnick et al. demonstrated that tracking the emotional "temperature" between characters over the course of a play could quantitatively pinpoint critical dramatic shifts, such as the tragic breakdown of a marriage in Shakespearean drama.

Emotional Signatures as Structural Blueprints

Once a text is mapped to an emotional trajectory, these curves serve as high-level features or "narrative signatures." They reveal abstract properties of the text without relying on explicit keywords or metadata.

  • Genre and Theme Classification: Kar et al. showed that tracking specific emotional dimensions (like joy or fear) throughout movie plot synopses creates distinct signatures that differentiate storytelling styles and genres. In a follow-up work, Kar et al. leveraged this sequential "emotion flow" to predict abstract, descriptive movie tags (like feel-good or murder) that were never explicitly stated in the text.
  • Structural Archetypes: Stories often follow universal templates. Tian et al. categorized narrative structures by mapping a protagonist's fortune into archetypal trajectories, showing that forcing large language models to plan these macro-arcs prevents them from generating "unexcitingly positive" or poorly paced stories.
  • Quality and Highlight Detection: In media editing, Papalampidi et al. discovered that a V-shaped emotional intensity pattern—where intensity drops for information delivery and peaks at the climax—serves as an effective template for identifying trailer-worthy highlights. Even script quality can be quantified; Chiu et al. used emotional activity curves to track character evolution, outperforming deep architectures like BERT on predicting screenplay award nominations.

Why Trajectories Unlock Deeper Comprehension

Why are emotional arcs so effective for tasks like predicting what happens next? The answer lies in narrative coherence. A logical story ending is heavily constrained by the emotional momentum built up by the preceding text.

Chaturvedi et al. demonstrated that tracking emotional shifts across structural segments (beginning, body, climax) provides a powerful predictive signal for story endings. For example, a trajectory shifting from negative to positive strongly correlates with a happy conclusion, whereas pessimistic arcs lead to highly unpredictable endings. In fact, Liu et al. found that monitoring sentiment transitions via memory chains is often more critical for predicting logical conclusions than tracking events or topics alone.

To operationalize this, models must explicitly encode these trajectories. For example, Chen et al. extracted sequential sentiment scores and passed them through a recurrent network to predict the expected emotional state of a story's conclusion, using it to filter candidate endings. To prevent errors caused by subtle linguistic shifts, Li et al. incorporated explicit sentiment polarity and negation markers into their input representations, ensuring the model accurately tracked the narrative's emotional trajectory.

Preserving the Rhythm in Summarization and Generation

Beyond comprehension, emotional trajectories ensure that compressed or generated text retains a human-like cadence.

Summarization

When condensing long screenplays, selecting scenes based purely on informational density can result in a dry, repetitive summary. Gorinski et al. solved this by using sentiment analysis as a structural marker, selecting scenes with diverse sentiment profiles to preserve the original narrative's emotional rhythm. Papalampidi et al. similarly noted that while structural plot points (like setbacks or climaxes) provide a framework for summarization, integrating emotional fluctuations is essential to fully capture the human experience of a story.

Generation and Author Support

Automated storytelling systems frequently generate flat, robotic prose. Yu et al. argued that integrating affective trajectories is a necessary frontier to capture the emotional variations typical of human writing.

For human authors, visualizing these trajectories can be a powerful creative aid. Mori et al. developed a system that maps emotional progression using valence and arousal metrics, allowing novelists to see exactly how structural changes or new content insertions alter a scene's atmosphere, ensuring that the final text aligns with their intended emotional delivery.

Key Foundational Work

Character-to-Character Sentiment Analysis in Shakespeare's Plays (2013) — contributed 1 concept elements to this direction.
Emotions from Text: Machine Learning for Text-based Emotion Prediction (2005) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_sentiment_trajectory_analysis_39cae0bb,
  title = {sentiment trajectory analysis},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/sentiment-trajectory-analysis/rdr_39cae0bb56e249adbb301f42eb5529ba},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_sentiment_trajectory_analysis_39cae0bb_v1,
  title = {sentiment trajectory analysis},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/sentiment-trajectory-analysis/rdr_39cae0bb56e249adbb301f42eb5529ba/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake