Python analysis of cloud motifs in Paustovsky

Analyzing Cloud Imagery in Paustovsky's Prose with Python and NLP

The analysis of Paustovsky's lexical techniques began with the complete collected works from 1981. Files from 9 volumes were downloaded and merged into a single .txt file containing 8,867,522 characters. Volumes 8 and 9, containing plays and letters, were excluded to focus on his fictional prose.

The texts were split into sentences, filtering for length (>20 characters):

sentences = re.split(r'[.!?…]+', text)
sentences = [s.strip() for s in sentences if len(s.strip()) > 20]

This yielded 108,090 sentences. A search was then conducted for the lemmas 'cloud', 'sky', and 'storm cloud' using pymorphy3:

target_keywords = ['cloud', 'sky', 'storm cloud', 'small cloud', 'clouds', 'sky', 'storm clouds']
target_forms = set(target_keywords)
morph = pymorphy3.MorphAnalyzer()
for lemma in ['cloud', 'sky', 'storm cloud']:
    parsed = morph.parse(lemma)[0]
    for form in parsed.lexeme:
        target_forms.add(form.word.lower())

2,282 sentences were found (2.1% of the total).

Classifying Cloud Descriptions

The keywords were classified by type:

• Meteorological terms (cloud types: cumulus, stratus, cirrus, etc.).
• Artistic metaphors (reamorphism: cities, ships, cotton; natmorphism: smoke, steam).

Distribution:

• metaphorical: 1,096 (48.0%)
• neutral: 963 (42.2%)
• mixed: 156 (6.8%)
• meteorological: 67 (2.9%)

Purely meteorological descriptions are minimal, indicating a figurative usage.

Intermediality: Connection to Painting

Co-occurrence with artists, paintings, terms ('brush', 'easel', 'landscape') and light/color verbs ('flare up', 'glow') was checked.

Statistics:

• Artist mentions: 2,017
• Paintings: 770
• Painting terms: 11,299
• Light verbs: 1,263

Of the 2,282 cloud sentences, 688 (30.1%) contain painting markers.

Visual element density (per 1,000 characters):

| Category | With Clouds | Without Clouds | Difference |

|--------------------|------------|-------------|---------|

| Light verbs | 0.43 | 0.14 | +0.29 |

| Color adjectives | 1.10 | 0.42 | +0.67 |

| Painting terms | 1.85 | 1.19 | +0.67 |

| Total density | 3.38 | 1.74 | +1.63 |

Top artists: Dalí (possibly an artifact of 'dali'), Manet, Levitan, Kiprensky. 10% of the corpus consists of sentences with painting terms.

Sentiment Analysis

Sentiment was analyzed using rubert-base-cased-sentiment and a lexicon (positive: happiness, peace; negative: anxiety, death):

sentiment_results = []
for sentence in target_sentences:
    sentiment = analyze_sentiment_lexicon(sentence)
    sentiment_results.append({
        'sentence': sentence[:200],
        'sentiment': sentiment
    })
all_sentiments = [r['sentiment'] for r in sentiment_results]

Distribution:

• Positive: 124 (5.4%)
• Negative: 67 (2.9%)
• Neutral: 2,087 (91.5%)
• Mixed: 4 (0.2%)

Negativity is linked to wartime prose (TASS correspondent in 1941). Positivity outweighs negativity.

Key Findings

• Clouds in Paustovsky: 48% metaphors, 2.9% purely meteorological.
• 30.1% of cloud descriptions are intermedial (connected to painting).
• Neutral sentiment predominates (91.5%), positivity > negativity.
• Visual marker density is 1.9 times higher in paragraphs with clouds.
• Corpus analysis confirms his status as a 'painter in prose'.

Clouds serve not as background, but as carriers of mood, symbols, and references to visual art. Selective descriptions enhance expressiveness without being merely decorative.

— Editorial Team