Abstract
The enduring comparison between Li Bai (李白, 701–762) and Du Fu (杜甫, 712–770), two towering poets in Chinese literary history, has traditionally centred on semantic and thematic aspects, often overlooking the phonetic dimension or narrowly focusing on rhythm and rhyme while neglecting other sound features. This study applies sound symbolism principles to explore the emotional and perceptual effects of sounds in their poetry, focusing on a quantitative phonetic comparison of their works. Utilising character sound information extracted from the rhyme book Guangyun (廣韻 extended rhymes), their poems were transformed into sound vectors for authorship attribution by machine learning models. Several of these models achieved average F1 scores above 80%, effectively demonstrating the capability of sounds to distinguish between the two poets. Through statistical analyses, their preferred phonetic features were located and the sound-sentiment relationships in their poetry were identified, revealing that Li Bai favoured the more positive sound features while Du Fu showed an inclination towards the more negative ones. Moreover, the distinct sound preferences sculpted Li’s poetry with a vibrant, melodious, and unbound quality, while lending a subtle, sombre, and constrained undertone to Du’s verses. Focusing on the phonetic aspect, this study takes a novel digital humanities approach to the traditional Li-Du comparison topic, offering fresh insights into their differences and underscoring the impact of sounds on the aesthetic experience of poetry.
Introduction
Li Bai and Du Fu are the undisputed titans of Tang dynasty poetry. Traditionally, literary scholars have characterized their distinct styles through qualitative descriptors: Li Bai is often viewed as piaoyi (ethereal, bold, and unbound), while Du Fu is described as chenyu (profound, melancholic, and restrained). These comparisons typically rely on semantic analysis, thematic exploration, or basic rhyming schemes.
However, classical Chinese poetry was historically intended to be chanted or sung. This implies that the physical, acoustic properties of the words—such as voicing, aspiration, and tongue position—are integral to the artistic experience. This research addresses a gap in computational literary analysis by asking: Do these two poets actually sound different? By objectively quantifying "sound symbolism"—the non-arbitrary link between phonetic sounds and emotional effects—the study seeks to explain their divergent styles through the biomechanics of speech.
Methodology: From Verses to Vectors
The researchers framed this inquiry as a quantitative digital humanities problem, utilizing a pipeline that combines historical phonology with predictive modeling.
Feature Extraction and Vectorization
Using the Guangyun, a comprehensive Song dynasty rhyme book, the authors mapped every Chinese character in the poets' works to 470 distinct phonetic features, including initials, finals, tones, voicing, and place of articulation. To process this data, each poem was converted into a 470-dimensional vector representing the normalized frequency of these sound features.
Predictive Modeling and Statistical Analysis
To test if sound alone constitutes a reliable stylistic fingerprint, various classifiers—including Support Vector Machines (SVM), Logistic Regression, and Neural Networks—were trained to predict authorship based purely on these sound vectors. Additionally, the study utilized Chi-square tests to identify phonetic features significantly overused by each poet and correlated these with a sentiment-annotated corpus (FSPC) to map sounds to emotional valences.
Results: The Acoustic Fingerprint
The results demonstrate that phonetics alone are highly predictive of authorship, suggesting that poets possess distinct, perhaps subconscious, "sound signatures."
Model Performance
The Support Vector Machine (SVM) achieved the highest weighted F1 score of 83.7% for authorship attribution. Other models, such as Logistic Regression (83.2%) and Multilayer Perceptrons (82.1%), also performed exceptionally well, confirming that the phonetic profiles of Li Bai and Du Fu are statistically distinct.
Vowel Mapping and Phonetic Extremes
When comparing Li and Du against ten other major Tang poets, both were found to use more "back" vowels, which are acoustically associated with grandeur and depth. However, they sat at opposite extremes regarding vowel height: Du Fu utilized the highest vowels among the group (creating a focused, constrained sound), while Li Bai utilized some of the lowest (resulting in a sonorous, open sound).
Discussion: The Physics of Emotion
The study provides a quantitative basis for the "vibe" of these poets by linking phonetic mechanics to emotional resonance:
- Li Bai’s Unbound Resonance: Li Bai favors "positive" sound features, such as voiceless aspirated initials and throat sounds. Physically, these require significant airflow and an open mouth, creating a resonant and lively acoustic effect that mirrors his bold, Daoist themes.
- Du Fu’s Constrained Melancholy: Du Fu prefers "negative" sound features, including voiceless unaspirated initials and lip sounds. These require closed lips and restricted airflow, mirroring his introspective, solemn reflections on societal suffering and war.
Conclusion
This study successfully translates abstract literary concepts into tangible phonetic data, proving that the aesthetic experience of poetry is rooted in the biomechanics of speech. While the research is limited by the lack of historical polyphone disambiguation and the standardized nature of the Guangyun, it opens new avenues for digital humanities. Future work may integrate phonetic vectors with semantic models like BERT or utilize neuroimaging to measure how modern readers react to these ancient sound patterns.
Summary updated