Cognitive and Neural Mechanisms of Tonal Patterns and Semantic Processing in Poetry

School of Psychology, Nanjing Normal University, 210097
School of Chinese Language and Literature, Nanjing Normal University, 210097
Philosophy and Social Science Laboratory of Universities in Jiangsu Province
Laboratory of Adolescent Education and Intelligent Support, Nanjing Normal University, Nanjing

Abstract

Music is the lifeblood of poetry. In ancient Chinese poetry, the tonal pattern (pingze) serves as a quintessential expression of poetic musicality. This study selected five-character regulated verses (wuyan lüshi) as experimental stimuli to investigate the cognitive and neural mechanisms underlying tonal perception and poetic comprehension. By simultaneously manipulating the tonal patterns and semantic features of the poems and employing Event-Related Potential (ERP) technology, we examined these processes in 48 participants with extensive experience in classical poetry.

Behavioral results revealed a significant main effect of tonal patterns and a significant interaction between tonal patterns and semantics. Electrophysiological results demonstrated a distinct tonal effect during poetry reading, characterized by an early $P200$ effect, followed by an $N400$ effect and a late positive component ($LPC$) effect. Tonal perception and semantic comprehension were found to integrate during the mid-to-late stages of processing, as evidenced by significant interactions in the $N400$ and $LPC$ components. Specifically, irregular tonal patterns modulated phonological representation at early stages and constrained semantic integration at later stages. Furthermore, a neural network (Conv-LSTM) trained on the ERP data effectively classified the four types of poetic lines, confirming that tonal rules and semantic features are critical neural signatures in poetry processing. These findings suggest that readers maintain expectations for both phonological harmony and the coordination between sound and meaning, supporting the neurocognitive poetics model and interactive activation models of language processing.

Keywords: Poetry; Music; Tonal patterns (pingze); Semantics; ERP

1 Introduction

Music is the lifeblood of poetry \cite{朱光潜2012}. Across diverse cultures, poetry and music originated as a unified art form; poetry was meant to be sung, and song was invariably accompanied by music \cite{Aristotle1968, Nietzsche2003}. As these arts diverged, poetry evolved toward textual meaning, entering a period characterized by "lyrics without melodies." When lyrics and melodies coexisted, the musicality of poetry was found in the tune; once the lyrics were separated from the melody, the musicality had to be found within the words themselves. Taking ancient Chinese regulated verse (gelüshi) as an example, many meticulous rules regarding word count, tonal patterns, and rhyme schemes were established to manifest the inherent musicality of the language \cite{王力2012}.

As a tonal language, the rules of ping (level) and ze (oblique) tones are a unique feature of ancient Chinese regulated verse. Ancient Chinese possessed four tones: ping (level), shang (rising), qu (departing), and ru (entering). The term ping refers to the level tone (encompassing the modern first and second tones), while ze refers to the remaining three tones. The tonal rules for five-character and seven-character regulated verse dictate an alternation of ping and ze within a single line (e.g., "oblique oblique level level oblique"). Furthermore, there is a correspondence between the lines of a couplet and a "sticking" (zhan) rule between couplets to avoid monotonous tonal repetition \cite{王力2012}.

The systematic alternation of these two tonal categories creates a musical sensation, imbuing poetry with internal rhythm. For instance, in the couplet "In the world of men, April's blossoms have all faded; in the mountain temple, peach blossoms begin to bloom" (人间四月芳菲尽，山寺桃花始盛开), if the final word of the second line were changed to a different word for "bloom" (盛放), the meaning would remain largely unchanged, yet the line would feel awkward to the ear. Beyond rhythm, the four tones ensure coordination between sound and meaning. Sonorous and rapid sounds often describe galloping horses, while smooth sounds describe flowing water \cite{朱光潜2012}.

Cognitive poetics, emerging in the 1970s, focuses on how individuals perceive rhythmic features in poetry (Zhang, 2015; Tsur, 1992). Studies on Indo-European languages have found that reading poetry activates brain regions associated with rhythm and music, such as the superior temporal gyrus (Zeman et al., 2013). Rhythmic anomalies, such as irregular stress or rhyme violations, cause significant reading interference (Beck & Konieczny, 2020). In Chinese poetry, pauses, rhyming, and tonal patterns constitute the core prosodic features. Previous research has shown that rhyme violations induce larger $N400$ amplitudes (Chen et al., 2016) and that tonal processing is modulated by musical experience (Zhang, 2020). However, few studies have strictly isolated pingze rules from semantic plausibility.

Theoretical debates regarding the relationship between prosody and semantics revolve around modularity (Fodor, 1983), which suggests independent processing, versus interactive models (Friederici, 2002; McClelland & Rumelhart, 1981), which argue for real-time bidirectional interaction. This study uses ancient regulated poetry to investigate these mechanisms. We hypothesize that a pingze effect exists in early stages (e.g., $P200$), reflecting perceptual matching, and that an interaction with semantics occurs in the $N400$ and $P600/LPC$ stages, reflecting integration and reanalysis.

2 Methods

2.1 Participants

Forty-eight students specializing in ancient literature were recruited. Participants were required to score above 80% on an "Ancient Poetry Knowledge Assessment Questionnaire" \cite{2021}. All were right-handed native Chinese speakers with normal vision and no neurological history.

2.2 Materials and Design

A $2 \times 2$ within-subjects design was used: Semantic Consistency (Plausible vs. Implausible) and Tonal Consistency (Regular vs. Irregular). Couplets were selected from the Complete Tang Poems that strictly follow pingze rules and maintain consistent pronunciation from Middle Chinese to modern Mandarin \cite{1993}. To create the four conditions, the final character of the second line was modified. For example:
- S+T+: Semantically plausible, tonally regular (Original).
- S+T-: Semantically plausible, tonally irregular.
- S-T+: Semantically implausible, tonally regular.
- S-T-: Semantically implausible, tonally irregular.

Materials were controlled for character frequency and stroke count. Semantic plausibility was validated by an independent group ($p < 0.001$), ensuring that tonal changes did not confound semantic ratings.

2.3 Procedure

Stimuli were presented segmentally (600 ms display, 400 ms interval). Keywords were presented before punctuation to avoid wrap-up effects \cite{Just1980}. Participants performed a four-choice judgment task: "Metrically Correct & Semantically Plausible," etc. The spatial arrangement of options was counterbalanced.

2.4 Data Analysis

Behavioral data were analyzed using Logistic Linear Mixed Models (LLMM). EEG data were recorded at 1000 Hz, re-referenced to the average mastoids, and filtered (0.05–30 Hz). After ICA artifact rejection, epochs (-200 to 1000 ms) were analyzed using Linear Mixed Models (LMM) across four time windows: 80–120 ms, 150–250 ms ($P200$), 250–450 ms ($N400$), and 500–700 ms ($LPC$).

A decoding analysis using a Convolutional Long Short-Term Memory (Conv-LSTM) network was performed to classify the four verse types. Features included time-domain statistics and frequency-domain power spectral density. A $k$-fold cross-validation was used to evaluate classification accuracy.

3 Results

3.1 Behavioral Results

Accuracy was highest for S+T+ (92.25%) and lowest for S+T- (80.42%). LLMM revealed a significant main effect of tonal patterns ($\chi^2 = 7.01, p < 0.01$) and a significant interaction between tonal and semantic reasonableness ($\chi^2 = 52.45, p < 0.001$). When tonal patterns were regular, participants were more likely to misjudge semantically implausible verses as plausible.

3.2 ERP Results

• 150–250 ms ($P200$): A significant interaction between tonal regularity and region ($p = 0.024$) showed that irregular tonal patterns elicited larger positivity in the anterior region ($\beta = 0.51, p = 0.02$).
• 250–450 ms ($N400$): A significant main effect of semantics ($p < 0.001$) and a significant interaction between semantics and tonal regularity ($p = 0.01$) were found. Regular tonal patterns facilitated semantic processing, producing a larger $N400$ difference between congruent and incongruent conditions.
• 500–700 ms ($LPC$): A significant interaction ($p = 0.003$) revealed that when semantics were incongruent, tonally irregular verses elicited a larger positive response than regular ones ($\beta = 1.24, p < 0.001$).

3.3 Decoding Results

The Conv-LSTM model achieved high classification performance, with AUC values for all four categories ranging from 0.95 to 0.97. Permutation tests confirmed that classification precision, recall, and F1-scores were significantly above chance ($p < 0.001$), indicating that the EEG signal contains robust information regarding both tonal and semantic structures.

4 Discussion

This study demonstrates that tonal patterns (pingze) are a fundamental component of poetry processing. Behavioral data showed that regular tonal structures can "mask" semantic implausibility, leading to lower accuracy in detecting semantic errors. This suggests a "poetic function" where readers prioritize formal harmony.

The ERP data revealed a multi-stage processing model. The early $P200$ effect reflects the detection of tonal violations against internal templates. The $N400$ interaction suggests that regular prosody facilitates semantic integration, while irregular prosody consumes attentional resources, weakening the semantic effect. The late $LPC$ effect reflects the conscious reanalysis and integration of the poem's global structure. Unlike rhyme, which is a local phonological match, pingze is a global structural rule requiring more complex neural coordination.

The successful decoding of verse types using deep learning further confirms that the brain encodes tonal and semantic information in a distinct, accessible format. These findings support the Neurocognitive Poetics Model \cite{Jacobs2015}, showing a dynamic interaction between "foreground" prosodic features and "background" semantic content.

5 Conclusion

Tonal patterns in classical Chinese poetry influence lexical-phonological representation in early stages and interact with semantic integration in later stages. Readers possess a specialized sensitivity to the unity of sound and meaning. This research provides empirical evidence for the interactive nature of prosodic and semantic processing and highlights the unique cognitive mechanisms involved in literary appreciation.