A Model of Poetic Comprehension

AAAI 1996

TL;DR

This paper proposes a computational model for aesthetic comprehension, viewing poetry as an "analogy teaching machine" that reshapes memory and influences future understanding. The authors implement a system that uses formal poetic cues (rhyme, rhythm) to guide analogical interpretation of Wordsworth's "Lines Written in Early Spring," demonstrating how these cues can reveal analogies not apparent from semantic analysis alone.

What Problem Does This Paper Address?

Understanding aesthetic experiences, particularly poetic comprehension, remains a significant challenge for computational models. Traditional approaches often focus on literal meaning or canonical representations, failing to capture the richness of metaphor, analogy, and the subjective impact of art. This paper tackles the question of how formal elements in poetry (like rhyme and meter) can cue novel analogical interpretations and how these experiences might fundamentally alter a system's (or a human's) future cognitive processes, essentially "rewiring" memory.

Methods

The core of the paper is an analogical comprehension framework that moves beyond reducing texts to canonical forms. Instead, it focuses on determining dynamic analogies between descriptions.

  • Analogical Comprehension Model: This model uses a semantic network where descriptions are nodes connected by "micro-relations" (grammatical structure) and "associations" (meaning and significance). Matching occurs at two levels:
  • Cognate Matches: Pairings based on unique common associations between nodes.
  • Structural Matches: Extensions of cognate matches that incorporate micro-relational structure.
  • Poetry Representation: Extends the prose representation by adding associations for rhyme (final phoneme) and metrical position. Word meaning is enriched using WordNet senses and hypernyms. This representation preserves ambiguity, allowing for metaphorical interpretations.
  • Implemented Miniature: A system that parses poetry, annotates it with formal cues (rhyme, meter), and then uses the analogical matcher to find correspondences between poetic lines or couplets. The system selectively disables certain annotation types to analyze the impact of specific cues.

Datasets

  • WordNet: Used for semantic associations and word meaning.
  • William Wordsworth's "Lines Written in Early Spring": The primary text used for analysis and demonstration. The poem was manually annotated with rhyme and metrical information.
  • News Summaries Corpus (approx. 10 million words): Mentioned as a prior application of the analogical comprehension model, but not directly used in this paper's experiment.

Key Insights and Ideas

  • Poetry as an Analogy Teaching Machine: The central thesis is that poems, through their formal structure, act as sophisticated analogy generators, teaching us new ways to see connections and influencing future cognitive processes.
  • Formal Cues as Analogy Catalysts: Rhyme and meter are not just decorative; they actively guide the analogical matcher to consider pairings that might be missed by purely semantic analysis.
  • Aesthetic Experience as Memory Reorganization: The paper posits that aesthetic experiences are those that fundamentally change how our memories are indexed and matched, leading to a "new perspective" on the world.
  • Defeasible Cognate Relations: Unlike some other analogical matchers, the "cognate relation" in this model is contextually sensitive and can be added or removed, making it more flexible.
  • Ambiguity as a Feature, Not a Bug: The representation explicitly preserves ambiguity, which is crucial for allowing metaphorical interpretations to emerge in poetry.

Key Results

The implemented miniature successfully identified several plausible analogical mappings in Wordsworth's poem, particularly in stanzas 1, 2, and 6.

  • Stanza 1: Matches like "heard" → "bring" and "notes" → "thoughts" were found, reflecting the poem's juxtaposition of nature and thought.
  • Stanza 2: Rhyme and meter cued analogies between "link" → "think" and "ran" → "made."
  • Stanza 6: The system found mappings between "belief" → "reason" and "plan" → "Man," aligning with the poem's themes of lost potential and intentional loss.
  • Impact of Formal Cues: By selectively disabling rhyme and meter, the authors could observe how these cues influenced the generated matches, highlighting their role in guiding interpretation.
  • Limitations in Complex Stanzas: The system struggled with stanzas 3, 4, and 5, which lacked clear inter-couplet structure or mixed natural and thought descriptions, suggesting the need for more sophisticated handling of complex poetic structures.

Limitations and Future Work

  • Manual Annotation: The rhyme and meter annotations were done by hand, a significant bottleneck. Automating this process is a key future step.
  • Limited Corpus: The experiment was conducted on a single poem. Future work involves applying the mechanisms to more pieces, larger corpora, and across different poets or styles.
  • Generalization of "New Perspective": The paper acknowledges that demonstrating how a single aesthetic experience transforms future comprehension is difficult and requires testing on larger, more integrated datasets.
  • Human Experience vs. Computational Model: The authors recognize that their model lacks the rich, lived experience that humans bring to poetry, which influences interpretation. WordNet is a poor substitute for this.
  • Defining Aesthetic Experience: The paper grapples with defining aesthetic experience as memory reorganization, acknowledging potential objections and refining the definition to distinguish it from mere memory changes or artistic intent.
  • Future Directions:
    • Automating poetic annotation.
  • Applying the model to larger poetic corpora and different artistic forms (music, visual arts).
  • Integrating poetry and prose databases to study how aesthetic mappings influence everyday comprehension.
  • Exploring the implications for critical theory and designing new interactive media.

Summary updated

BibTeX
@inproceedings{1996model,
  title = {{A Model of Poetic Comprehension}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {1996},
}

Loading related content...

Figure