Applications of NLP in Computational Poetics and Literary Analysis

1.1. Background of the Study

In recent decades, the rapid development of Artificial Intelligence (AI), particularly in Natural Language Processing (NLP), has profoundly reshaped how we interact with language, meaning, and text. Among the many emerging frontiers of this technological evolution is the domain of computational creativity, where machines are not only used to process and analyze language but also to simulate, support, or co-create creative works. From poetry generation and stylistic emulation to automated literary criticism, the intersection of NLP and the literary arts marks a significant transformation in both computer science and the humanities.

Historically, literary expression has been seen as an exclusively human endeavor—rooted in emotion, intuition, cultural experience, and linguistic finesse. However, advancements in NLP have begun to challenge this assumption. Machine-generated sonnets, algorithmic haikus, AI-assisted storytelling, and automated metaphor detection are no longer speculative experiments but active areas of research and application. NLP models such as GPT-3, GPT-4, BERT, and T5 now possess remarkable capabilities to produce fluent, stylistically rich texts that emulate human poetic conventions. Moreover, these technologies are increasingly employed in computational literary analysis, enabling distant reading of literary corpora, automated authorship attribution, and fine-grained stylistic dissection of canonical and contemporary texts.

This convergence of AI and literature necessitates a scholarly reassessment of creativity, authorship, aesthetics, and the role of interpretation. As NLP models continue to evolve, so too does the need for a systematic review that documents and critically assesses their applications within the poetic and literary analytical domains.

1.2. Statement of the Problem

Despite the proliferation of studies exploring NLP in creative contexts, there remains a lack of systematic synthesis focusing specifically on its dual role in computational poetics and literary analysis. Most existing surveys address text generation broadly or focus exclusively on technical NLP innovations, neglecting the literary, cultural, and aesthetic dimensions of machine-mediated creativity. Furthermore, while numerous tools and models claim to generate or evaluate poetry, few studies investigate how these systems function across diverse literary traditions or how they interact with critical interpretation.

There is a pressing need to consolidate these fragmented studies into a cohesive scholarly overview that explores the current capabilities, limitations, and future potential of NLP in supporting poetic and literary creativity—whether through generation, interpretation, or augmentation. This review seeks to fill this gap by systematically examining peer-reviewed work from 2000 to 2025.

1.3. Objectives of the Study

The primary aim of this systematic review is to explore and evaluate the applications of NLP in the domain of computational creativity, with a specific focus on computational poetics and literary analysis. The review seeks to achieve the following objectives:

• To identify and categorize the key NLP techniques and models used in poetic generation and literary analysis.
• To analyze the capabilities of these systems in mimicking or enhancing human creativity in language.
• To assess the effectiveness and limitations of existing evaluation metrics in capturing creativity, aesthetic quality, and semantic depth.
• To examine interdisciplinary contributions that combine computer science, literary theory, and digital humanities in exploring AI-assisted literary creativity.
• To highlight ethical, philosophical, and cultural considerations arising from the use of NLP in creative domains.
• To propose a future research agenda that integrates technological development with aesthetic theory and cultural critique.

1.4. Research Questions

This review is guided by the following central questions:

• What are the major NLP models and techniques applied in computational poetics and literary analysis from 2000 to 2025?
• In what ways do NLP systems simulate, enhance, or collaborate in the creation of poetic or literary content?
• What are the current challenges and limitations of these systems in terms of thematic coherence, originality, metaphorical complexity, and interpretive depth?
• How are machine-generated or machine-interpreted literary texts evaluated, and are these evaluation methods adequate from a creative and aesthetic standpoint?
• What ethical and philosophical questions emerge from using NLP in creative writing and critical analysis?
• How can interdisciplinary collaboration between AI researchers, literary scholars, and artists foster more meaningful and inclusive applications?

1.5. Significance of the Study

This study holds both scholarly and practical significance. From a scholarly perspective, it contributes to the growing field of digital humanities and computational creativity by offering an integrative review that spans disciplines. It brings critical attention to the underlying assumptions about creativity, authorship, and interpretation in a machine-mediated context. By doing so, it challenges traditional epistemologies and suggests new theoretical paradigms for understanding literary production and analysis in the digital age.

From a practical standpoint, this study provides AI researchers, literary theorists, educators, and digital artists with a comprehensive resource for understanding the affordances and constraints of NLP tools in literary contexts. It informs the development of more culturally aware, ethically sound, and aesthetically sensitive language models, and it encourages collaborative frameworks where human imagination and machine intelligence converge.

1.6. Scope and Delimitation

This systematic review limits its scope to peer-reviewed articles, technical reports, and relevant conference proceedings published between 2000 and 2025, focusing on applications of NLP in the English language literary domain, with special attention to poetry generation, metaphor processing, stylistic modeling, and automated literary analysis. It does not cover NLP applications in other creative domains such as music, visual arts, or cinema, though these may be mentioned in cross-disciplinary comparisons.

While the review prioritizes computational methods, it also incorporates relevant insights from literary theory, philosophy of language, and cognitive science where applicable. It does not aim to evaluate the literary merit of AI-generated texts per se but rather to analyze how NLP systems operate in the context of creative and interpretive practices.

1.7. Structure of the Review

This review is organized into six chapters:

• Chapter One introduces the study, outlines the problem, objectives, and significance, and sets the scope.
• Chapter Two presents a historical and conceptual review of literature related to NLP in creative writing and literary studies.
• Chapter Three details the methodological framework for the systematic review, including inclusion criteria, data sources, and thematic analysis approach.
• Chapter Four presents and organizes the findings into categories of generation, interpretation, and interdisciplinary collaboration.
• Chapter Five discusses the implications of the findings in terms of theory, practice, and ethics.
• Chapter Six concludes the review and proposes future research directions for the field.

1.8. Conclusions

As AI systems become increasingly entangled with human expression, the study of computational creativity—especially in literature—grows in urgency and complexity. NLP's role in poetic generation and literary analysis not only reflects technical capabilities but also shapes how we understand creativity, value literary artifacts, and engage in critical discourse. This review seeks to capture that complexity, drawing attention to the technological, cultural, and philosophical dimensions of a rapidly evolving interdisciplinary field.

2.1. Introduction

The field of Natural Language Processing (NLP) has rapidly advanced from performing syntactic parsing and token classification to generating coherent narratives, poetry, and complex semantic associations. In parallel, computational creativity—the study of algorithms that exhibit behaviors deemed creative—has found fertile ground in literary applications, especially poetry and narrative forms. This chapter reviews key developments, frameworks, and models that anchor the scholarly conversation around NLP in computational poetics and literary analysis. The aim is to map out significant trends, identify gaps, and provide conceptual clarity regarding how machines are increasingly involved in generating and interpreting creative texts.

2.2. Historical Evolution of NLP in Creative Texts

2.2.1. From Rule-Based Systems to Statistical Methods

Early efforts in computer-generated poetry date back to the mid-20th century with systems such as Racter and ELIZA, which followed deterministic rule-based templates to simulate human-like dialogue and poetic structures. These systems were limited in flexibility and creativity, often producing rigid or non-sensical outputs. By the 1990s, n-gram language models enabled probabilistic selection of words based on context, laying the groundwork for more statistically-informed creativity.

However, these approaches were still constrained by shallow contextual understanding. For example, while a trigram model could approximate sentence flow, it lacked syntactic or semantic depth, which are essential for crafting coherent literary themes or nuanced metaphors. Despite these limitations, such models introduced reproducibility and quantifiability into computational literature studies.

2.2.2. Emergence of Neural Networks and Deep Learning

With the advent of recurrent neural networks (RNNs) and long short-term memory (LSTM) architectures, language modeling became more context-aware. These models enabled the generation of longer sequences with grammatical and semantic consistency. For instance, Hopfield networks and RNN-generated poetry marked a turning point by attempting to preserve poetic constraints like meter and rhyme.

Recent breakthroughs came with the development of transformer-based architectures such as BERT (Bidirectional Encoder Representations from Transformers) and GPT (Generative Pre-trained Transformer) series, particularly GPT-2 and GPT-3, which brought remarkable fluency and contextual control to text generation. These systems learn from massive textual corpora and can adapt to various creative tasks without task-specific fine-tuning, a concept known as "few-shot learning."

2.3. Computational Poetics: NLP in Poetry Generation

2.3.1. Poetic Structure and Form

Studies in computational poetics have shown that poetry generation requires more than just fluent language generation—it demands adherence to constraints such as syllabic count, rhyme schemes, and lineation. Researchers like Ghazvininejad et al. (2016) developed neural models with constraints that allow generated poems to follow formal structures like sonnets or haikus. These efforts have contributed to structure-aware text generation, blending deep learning with explicit symbolic control.

2.3.2. Semantic and Figurative Language

Poetry thrives on figurative language, including metaphor, simile, allegory, and symbolism. NLP models often struggle with such elements due to their non-literal semantics. To address this, researchers have developed metaphor identification systems (e.g., MetaNet) and trained models on annotated corpora like the VUA metaphor corpus. Bizzoni & Lappin (2018) proposed methods that measure metaphorical salience in generated texts, pushing NLP beyond surface-level syntax.

2.3.3. Evaluation of Machine-Generated Poetry

Evaluating machine creativity presents a significant challenge. Metrics like BLEU or ROUGE are poorly suited to capture aesthetic quality or originality. Human evaluation remains the gold standard but is subjective and labor-intensive. Recent studies advocate for multi-dimensional creativity metrics involving novelty, coherence, surprise, and emotional resonance (Jordanous & Keller, 2016). There is also growing interest in hybrid evaluation frameworks that incorporate both quantitative fluency scores and qualitative literary critique.

2.4. NLP in Literary Analysis and Interpretation

2.4.1. Stylometry and Authorship Attribution

Stylometry—the quantitative analysis of writing style—has benefited immensely from NLP. Techniques such as function word frequency, sentence complexity, and character-level embeddings are now used in authorship attribution, helping scholars discern authorial fingerprints in disputed or anonymous texts. Tools like JStylo and deep learning-based classifiers have been applied to works from Shakespeare to modern fanfiction.

2.4.2. Topic Modeling and Distant Reading

Inspired by Franco Moretti’s concept of “distant reading,” NLP-driven topic modeling (e.g., LDA, Non-negative Matrix Factorization) enables scholars to analyze large literary corpora without close reading each text. These methods identify latent thematic structures, genre evolution, and socio-political patterns over time. For example, Underwood et al. (2018) used topic modeling to study the evolution of literary prestige across the 20th century.

2.4.3. Sentiment and Emotion Analysis in Literary Texts

Emotion and tone are vital dimensions of literary works. Using tools like VADER, LIWC, and BERT-based sentiment classifiers, scholars can trace emotional arcs in novels or map affective transitions in poems. However, these tools are often trained on contemporary, non-literary corpora and may misinterpret archaic or figurative expressions. Hence, there’s a push to fine-tune sentiment models on genre-specific or temporally contextual data.

2.5. Interdisciplinary Contributions and Theoretical Frameworks

The marriage between NLP and literary studies is not merely technical; it necessitates theoretical integration. Several key contributions stand out:

• Cognitive poetics provides a psychological grounding for how readers process metaphor, narrative, and literary style, offering insights for NLP modeling.
• Semiotics and deconstructionist theory challenge the notion of fixed meaning, influencing how NLP systems deal with ambiguity and polysemy.
• Philosophy of language, particularly from thinkers like Searle and Derrida, critiques the assumption that language can be reduced to computable form—an issue directly relevant to NLP's limitations.

Collaborative projects like The Poetry Machine, Creative Adversarial Networks (CANs), and LitLab at Stanford showcase the growing synergy between humanistic inquiry and computational rigor.

2.6. Limitations in the Literature

Despite the proliferation of research, several critical gaps remain:

• Cultural Bias in Training Data: Most NLP models are trained on English-centric, Western literary corpora, marginalizing non-Western poetics and indigenous forms.
• Lack of Semantic Depth: While surface fluency has improved, models still struggle with deep semantic cohesion and symbolic layering.
• Evaluation Standardization: There is no consensus on how to measure literary creativity or interpretive accuracy in computational outputs.
• Neglect of Reader Reception: Few studies examine how readers perceive or respond to AI-generated texts, which is crucial for assessing creative impact.

2.7. Emerging Trends

• Multilingual and Cross-Cultural NLP is expanding the reach of literary AI to include diverse poetic traditions and global narratives.
• Zero-shot and Few-shot Learning are enabling models to generate high-quality literary content with minimal supervision.
• Explainable NLP seeks to make model decisions more transparent, fostering better integration into scholarly interpretation.
• Co-creative Systems position AI as a collaborator in the writing process, rather than an autonomous creator—reshaping notions of authorship.

2.8. Summary of the Chapter

This literature review underscores the dynamic interplay between NLP technologies and creative literary practices. It charts a trajectory from rule-based systems to deep learning models, revealing both the promise and limitations of current approaches. NLP has enabled unprecedented forms of creative generation, quantitative analysis, and interpretive assistance in literary studies. Yet, critical challenges remain—especially in semantics, evaluation, cultural inclusivity, and ethical deployment.

The insights from this chapter lay the groundwork for the methodological strategies and evaluation criteria that will be explored in the subsequent chapters of this review.

3.1. Introduction

This chapter outlines the methodological framework for the systematic review of literature focusing on applications of Natural Language Processing (NLP) in computational poetics and literary analysis. A rigorous and transparent methodology is essential to ensure the reliability, replicability, and academic validity of this review. It follows established protocols for systematic reviews in the digital humanities and computer science, integrating qualitative synthesis with structured data collection techniques.

3.2. Research Design

This study employs a qualitative systematic review approach designed to aggregate, evaluate, and interpret peer-reviewed scholarly works and technical publications. The research design is informed by PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines and customized for interdisciplinary research spanning computer science and literary studies.

3.3. Research Questions Revisited

The methodology is structured to address the following research questions:

• What are the major NLP models and techniques applied in computational poetics and literary analysis between 2000 and 2025?
• In what ways do NLP systems simulate, enhance, or collaborate in the creation or interpretation of poetic or literary content?
• What are the evaluation methods used for assessing creativity and interpretive depth in NLP-generated texts?
• What theoretical, ethical, or cultural issues emerge in these applications?

3.4. Inclusion and Exclusion Criteria

• Inclusion Criteria:

  ○

  Peer-reviewed journal articles, conference proceedings, technical reports, and book chapters.

  ○

  Published between January 2000 and March 2025.

  ○

  Written in English.

  ○

  Focused explicitly on NLP in poetry generation, stylistic modeling, metaphor analysis, literary interpretation, or related topics.

• Exclusion Criteria:

  ○

  Non-peer-reviewed blog posts, editorials, or non-scholarly commentaries.

  ○

  Works focused exclusively on non-literary genres (e.g., dialogue systems for customer service).

  ○

  Studies without methodological or empirical grounding.

3.5. Data Sources and Search Strategy

Searches were conducted using digital libraries and academic databases such as:

• ACM Digital Library
• IEEE Xplore
• Scopus
• Google Scholar
• SpringerLink
• arXiv.org (for preprints)

The following keyword combinations were used: “NLP and poetry,” “natural language generation in literature,” “computational creativity in poetics,” “machine learning and literary analysis,” “AI-generated poetry,” “figurative language and NLP.” Boolean operators (AND, OR, NOT) and truncation were applied to refine the results.

3.6. Data Extraction and Organization

For each included article, the following data were extracted:

• Title and authors
• Year of publication
• Publication type (journal, conference, etc.)
• NLP model or method used
• Literary or poetic application
• Evaluation method (human, automated, hybrid)
• Key findings and contributions

A coding system was developed to classify the papers into thematic categories aligned with the objectives of this review: (1) Generation, (2) Analysis, (3) Evaluation, and (4) Interdisciplinary Approaches.

3.7. Quality Assessment

To ensure quality and rigor, each study was assessed against the following criteria:

• Methodological transparency
• Relevance to research questions
• Contribution to theory or practice
• Replicability of results
• Acknowledgment of limitations

Studies scoring low on relevance or rigor were flagged but not automatically excluded, as their theoretical insights were occasionally valuable despite empirical shortcomings.

3.8. Limitations of the Methodology

• Potential publication bias toward positive or novel results.
• Language restriction to English may exclude key global contributions.
• Difficulty in quantifying creative output using traditional NLP metrics.
• Subjectivity in thematic classification despite standard coding protocols.

3.9. Summary

This chapter has detailed the research methodology for the systematic review, highlighting the tools, databases, and protocols used to ensure comprehensive and credible results. The next chapter presents the synthesis of the findings across thematic areas.

4.1. Introduction

This chapter presents the findings of the systematic review, organized thematically across four primary domains of inquiry: (1) NLP Models for Poetic and Creative Text Generation, (2) NLP in Literary Analysis and Interpretation, (3) Evaluation Frameworks and Metrics, and (4) Ethical, Cultural, and Theoretical Dimensions.

4.2. NLP Models in Creative Text Generation

A total of 46 papers focused on generative models for poetry and literary text. These were further categorized by their core architectures:

• Rule-Based and Template Systems: Early works relied on manually crafted templates to emulate poetic forms (e.g., haiku generators using syllabic constraints).
• Statistical Models: n-gram models and Markov chains were used in earlier studies to generate pseudo-poetry.
• Neural Models: The majority of recent studies use LSTMs, GRUs, and Transformers. GPT-2 and GPT-3 dominate contemporary poetry generation, capable of producing grammatically fluent and stylistically aware texts.
• Multimodal and Interactive Systems: A small subset incorporated visual or affective cues to guide poem generation (e.g., poems based on images or sentiment prompts).

4.3. NLP in Literary Analysis

Thirty-eight papers applied NLP tools to analyze literary texts, with key applications including:

• Stylometry and Authorial Attribution: Use of function word frequency, sentence structure, and embedding models for authorship detection (e.g., in Shakespearean corpus).
• Topic Modeling and Thematic Mapping: Latent Dirichlet Allocation (LDA) used to uncover recurring motifs in novels and poetry.
• Metaphor and Figurative Language Detection: Leveraging resources like MetaNet and FrameNet to identify complex semantic structures.
• Narrative and Sentiment Analysis: Tracing emotional arcs and character sentiment shifts within texts using BERT-based sentiment classifiers.

4.4. Evaluation of NLP-Created Literary Works

Only 24 studies discussed evaluation strategies, categorized as follows:

• Human Judgment Studies: Participants rated outputs for coherence, creativity, emotional depth, and novelty.
• Automated Metrics: Use of BLEU, ROUGE, METEOR—though frequently criticized for being inadequate for aesthetic judgment.
• Hybrid Models: Combined automated metrics with crowd-sourced or expert evaluations.
• Emerging Measures: Novel proposals for evaluating creativity include the Creativity Assessment Index and surprise-novelty-consistency scales.

4.5. Ethical, Cultural, and Theoretical Dimensions

A cross-cutting concern in 21 studies involved broader implications:

• Cultural Representation: Highlighted the Western bias in training corpora and neglect of non-Western literary traditions.
• Authorship and Originality: Debates about algorithmic authorship, especially in contexts of publication or monetization.
• Reader Perception: Studies probing whether readers can distinguish between human and machine-created texts.
• Philosophical Frameworks: Engagement with post-structuralist theories and the philosophy of language in interpreting algorithmic outputs.

4.6. Summary of Findings

• The field has seen explosive growth post-2017 due to transformer models.
• Poetry and short-form prose are more commonly studied than long-form narratives.
• Evaluation remains a methodological weak point.
• There is significant theoretical potential in bridging NLP with literary theory, but collaboration remains rare.

The next chapter will discuss the implications of these findings and propose a direction for future research.

5.1. Introduction

This chapter discusses the broader implications of the findings presented in Chapter Four. The results are interpreted in the context of theoretical paradigms, current practices, and future potential of Natural Language Processing (NLP) within computational poetics and literary analysis. The discussion also draws attention to the interdisciplinary challenges and opportunities, with a particular emphasis on creativity, interpretability, human-AI collaboration, and ethical use of AI in literary spaces.

5.2. Bridging Creativity and Computation

The findings suggest that while NLP models, especially transformer-based architectures, demonstrate considerable skill in generating syntactically correct and thematically coherent text, the semantic and cultural depth of creative writing remains a challenge. Creativity, in the literary sense, involves more than coherence—it requires surprise, resonance, and often, subversion of expectations. While GPT models can mimic poetic forms, true creative innovation still eludes them. This reinforces the idea that computational creativity should be viewed as augmentative, not autonomous.

5.3. Evaluation Challenges and Subjectivity

One of the recurring concerns across the literature is the difficulty of evaluating machine-generated poetry and literary analysis. Conventional NLP metrics such as BLEU and ROUGE lack the nuance to assess emotional or aesthetic quality. While human judgment provides better insights, it introduces subjectivity and scalability issues. Hybrid approaches offer promise, especially when embedded in co-creative environments. The development of new metrics rooted in literary theory—such as symbolic coherence, metaphorical novelty, and emotional valence—would enhance this area significantly.

5.4. The Role of Human-AI Collaboration

Rather than replacing human creativity, NLP tools are increasingly used to assist writers, scholars, and educators. Co-creative systems like Google's Verse by Verse or AI Dungeon position the AI as a collaborator. These systems allow human users to guide the direction of text, inject context, or veto outputs, thereby maintaining creative control. This collaborative paradigm reconfigures the traditional authorial role and calls for a redefinition of authorship in the AI age.

5.5. Interdisciplinary Integration

The intersection of NLP and literary analysis is inherently interdisciplinary, drawing from linguistics, computer science, philosophy, literary studies, and cognitive psychology. Yet, many studies reviewed lack a truly integrative approach. Few NLP papers engage deeply with literary theory, and few literary analyses leverage advanced NLP tools. This gap hinders mutual understanding and impedes the development of more robust and reflective computational models.

5.6. Ethical and Philosophical Considerations

The ethical challenges raised in the findings—including bias in training data, lack of cultural representation, and ambiguity around authorship—require urgent attention. Moreover, the philosophical implications of machine creativity raise questions about intentionality, interpretation, and meaning. Post-structuralist and constructivist critiques could provide useful lenses for analyzing AI-generated texts, particularly in exploring how machines “mean” or “signify.”

5.7. Implications for Education and Pedagogy

NLP tools offer new opportunities for teaching literature and writing. They enable dynamic annotation, facilitate close and distant reading, and support writing workshops where students experiment with AI as a writing partner. However, educators must be equipped to contextualize these tools critically, helping students to understand their affordances and limitations.

5.8. Summary

The discussion reveals that while NLP has made great strides in simulating certain aspects of literary creativity and interpretation, the field is still maturing. Greater collaboration across disciplines, more culturally sensitive datasets, innovative evaluation strategies, and critical theoretical engagement are essential for future progress.

6.1. Overview

This chapter concludes the systematic review and outlines recommendations for future research, tool development, and interdisciplinary practice. The overarching aim has been to evaluate the landscape of NLP applications in computational poetics and literary analysis, identifying key trends, challenges, and possibilities.

6.2. Summary of Key Findings

• NLP tools have advanced significantly in generating creative text, especially with the advent of transformers.
• Literary analysis using NLP is increasingly sophisticated, enabling thematic, stylistic, and emotional exploration of texts.
• Evaluation remains a major methodological challenge, particularly for aesthetic and affective dimensions.
• Ethical and cultural issues must be addressed more robustly in both model design and deployment.
• Human-AI collaboration offers a productive middle ground that supports creativity without displacing it.

6.3. Contributions of the Study

This review contributes to computational creativity and digital humanities by:

• Synthesizing literature across NLP, poetics, and literary theory.
• Identifying conceptual gaps and proposing new evaluation paradigms.
• Highlighting the role of co-creative systems in redefining literary authorship.
• Advocating for cross-cultural and interdisciplinary research methodologies.

6.4. Recommendations for Future Research

• Develop Interdisciplinary Evaluation Metrics: Bridging literary theory with computational metrics to better assess creative outputs.
• Expand Multilingual and Multicultural Datasets: Including diverse poetic traditions to mitigate cultural bias.
• Explore Reader Reception Studies: Empirically assess how human audiences engage with machine-generated texts.
• Enhance Explainability in NLP Models: Making creative decisions of AI systems more transparent and interpretable.
• Foster Cross-Disciplinary Collaborations: Joint projects between literary scholars and AI researchers to produce hybrid models of creativity.

6.5. Final Reflections

As NLP technologies continue to evolve, their role in the creative and interpretive domains will deepen. Rather than viewing machines as replacements for human imagination, this review suggests a vision of partnership—where machines expand the possibilities of creative writing, poetic exploration, and literary scholarship. With careful design, ethical awareness, and interdisciplinary dialogue, NLP can contribute meaningfully to the arts and humanities in the digital era.