Discourse Coherence: Modeling Text Structure

Module: Natural Language Processing | Difficulty: Advanced

Entity-Based Coherence

Centering Theory

Evaluation

import numpy as np

def entity_coherence(text, entities):
    scores = []
    for i in range(len(text) - 1):
        overlap = len(set(entities[i]) & set(entities[i+1]))
        total = max(len(set(entities[i]) | set(entities[i+1])), 1)
        scores.append(overlap / total)
    return np.mean(scores)

Research Insight: Discourse coherence is important for text quality assessment and generation. Neural models that track entity mentions across sentences achieve state-of-the-art coherence evaluation. The key insight is that coherent texts maintain entity continuity.