A Study of the Initial System of the Yongle Nanzang 永乐南藏 Based on Phonological Correlations and Their Relationship with the Qishazang 磧砂藏

Abstract

This study investigates the initial consonant system of the Yongle Nanzang 永乐南藏, the second officially printed edition of the Chinese Buddhist canon of the Ming dynasty, and its relationship to the Qishazang 磧砂藏. Based on a digital and quantitative analysis of 27,832 phonetic entries—including Fanqie 反切 and Zhiyin 直音—we apply correlation coefficients (with significance tests), hierarchical clustering, and data mining methods. The results show the following: (1) In the labial series, contrasts between aspirated and unaspirated sounds are weak; Bang 幫 and Pang 滂 are occasionally confused; and Fei 非 and Fu 敷 are merged. Voiced and voiceless initials are sometimes mixed, but full devoicing has not occurred. (2) In the coronal series, Duan Zu 端組 shows high internal interchange and Zhi Zu 知組 tends to mix with Zhang Zu 章組 and Zhuang Zu 莊組. (3) In the sibilant and affricate series, Jing 精 and Zhuang 莊 are merged; Chuan 船 and Shan 禪 also merge, while Cong 從 and Xie 邪 only occasionally alternate. (4) In the velar and laryngeal series, fully voiced Qun 羣 is close to voiceless initials, while Yun 云 and Yi 以 are closely related. Nasal and lateral initials maintain independence. The findings suggest that Yongle Nanzang 永乐南藏 is not a direct reprint of the Qishazang 磧砂藏, but rather a revised edition adapted to editorial or regional considerations. This study demonstrates the value of digital phonological methods in historical phonology and the significance of the Ming–Qing Chinese Buddhist canon.

1. Introduction

The formation of the Chinese Buddhist canon was grounded in the extensive translation of scriptures originating from India and Central Asia. For over a millennium since the Han dynasty, the Chinese Buddhist community continuously engaged in the systematic compilation and classification of these translated texts (Wu 2015; Miyazaki 2019). Drawing on extensive international scholarship on the formation and classification of the Chinese Buddhist canon (e.g., Chikusa 2000; Li and He 2003; Fang 2006; Xu 2006; Wu 2015; Miyazaki 2019), the historical development of the Chinese Buddhist canon may be broadly divided into two main phases: the manuscript period and the printed period. Throughout both phases, Buddhist scriptures carved in stone served as a unique medium of textual transmission. The manuscript period is represented most prominently by the Dunhuang manuscripts. By the Song dynasty, the maturing and widespread use of woodblock printing technology brought about a transformative shift in the transmission of Buddhist texts from hand-copied manuscripts to block-printed editions. This technological and institutional transition was inaugurated by the state-commissioned printing of the Kaibaozang 開寶藏1 in the early Northern Song, marking the beginning of printed canons in China. Over the following centuries, the Chinese Buddhist canon evolved into a system comprising more than 20 distinct editions, each with unique characteristics.

This developmental trajectory has attracted significant scholarly attention. Researchers such as Wu (2015), Long (2015, 2022), Chia (2015), Stefano Zacchetti (2015), and Miyazaki (2019) have conducted systematic studies of various versions of the canon from multiple perspectives. These academic contributions not only provide essential support for the present study, but also offer a solid foundation for examining the interplay between textual transmission and linguistic characteristics in the canon’s philological evolution. Building upon this body of research, the present study undertakes a preliminary mapping of the genealogical relationships among major editions of the Chinese Buddhist canon from both diachronic and versional perspectives (see Figure 1).

As shown in Figure 1, the development of the Chinese block-printed Buddhist canon may be divided into three major stages. The first stage corresponds to the Northern Song, Liao, and Jin dynasties (10th–12th centuries), during which editions were primarily sponsored and organized by central governments. Scholars generally classify the Kaibaozang, Zhaocheng Jinzang 趙城金藏, and both the early and re-engraved editions of the Gaolizang 高麗藏 into the Central Plains Tradition (Zhongyuan Xitong 中原系統), while the Liaozang 遼藏 is typically considered part of the Northern Tradition (Beifang Xitong 北方系統). The second stage spans from the late Northern Song to the Yuan dynasty (12th–14th centuries), characterized by the increased role of private patronage and monastic initiatives in the production of the canon. The geographical focus of this activity shifted to the southern regions of China. Editions produced during this phase are generally referred to as the Southern Tradition (Nanfang Xitong 南方系統). Among them, the Chongningzang 崇寧藏 and Piluzang 毗盧藏, both engraved in Fuzhou, Fujian, are collectively known as the Fuzhouzang 福州藏. Similarly, the Yuanjuezang 圓覺藏 and Zifuzang 資福藏, engraved in Sixi, Huzhou, Zhejiang, are jointly referred to as the Sixizang 思溪藏. The third stage spans the Ming and Qing dynasties (14th–19th centuries), during which stable political conditions and economic prosperity enabled a renewed state-led effort to compile the canon. Major editions from this period include the Hongwu Nanzang 洪武南藏 and Yongle Beizang 永樂北藏 of the Ming dynasty and the Longzang 龍藏 of the Qing dynasty. The Jiaxingzang 嘉興藏, by contrast, was initiated during the Ming dynasty and continued into the Qing, with its compilation largely supported by private donations, as evidenced by early colophons from Wutaishan editions.

In the modern era, advancements in printing technology have given rise to movable-type and photomechanical editions of the canon in China, Japan, and elsewhere. These modern developments—though extensively studied (e.g., Wu 2015)—fall outside the scope of the present discussion and will not be elaborated upon here.

The Yongle Nanzang 永乐南藏, formally titled the Daming Sanzang Shengjiao Nanzang 大明三藏聖教南藏, was an imperially authorized and printed edition of the Chinese Buddhist canon during the Yongle reign of the Ming dynasty and was commissioned by Emperor Zhu Di 朱棣. This edition was engraved and preserved at the Da Bao’en Temple 大报恩寺 in Nanjing. Following the Kaibaozang of the Song dynasty and the Puningzang of the Yuan dynasty, it became another crucial Chinese Buddhist canon that had a profound influence on the transmission and development of Buddhism in China. Currently, no historical records precisely document the exact time of the Yongle Nanzang’s engraving. Based on works such as the Daming Gaoseng Zhuan 大明高僧傳, Gu Zunsu Yulu 古尊宿語錄, and the Jinling Fancha Zhi 金陵梵刹志, it is estimated that the plates were carved between the 11th and 18th years of the Yongle reign.

The engraving of the Yongle Nanzang directly traces back to the Hongwu Nanzang. Only two copies of the Hongwu Nanzang were ever produced, one of which was preserved at Tianxisi 天禧寺. In the sixth year of the Yongle reign, Tianxisi was destroyed by a large fire, and the Hongwu Nanzang manuscripts were consequently consumed by fire (Lü [1938] 2014). At that time, there was an urgent need for a new edition of the Chinese Buddhist canon, prompting Emperor Zhu Di to order the carving of the Yongle Nanzang. In terms of canonical transmission (Figure 1), scholars generally acknowledge that the Hongwu Nanzang belongs to the Fuzhouzang series and is effectively a reprint of the Song–Yuan-engraved Qishazang. Accordingly, the content and layout of the Qishazang significantly influenced both the Hongwu Nanzang and the Yongle Nanzang (He 2005; Li 2021). In modern times, research on the Ming and Qing dynasty Buddhist canons has been limited, not only due to the immense volume of these canons, but also because they are often regarded as mere reproductions or reprints of earlier editions.

Some scholars have noted that, while the phonetic annotations (shiyin 釋音) in the Yongle Nanzang generally continue the traditions of the Qishazang and Hongwu Nanzang, there are notable differences in aspects such as the usage of Fanqie 反切 for phonetic annotation practices, the quantity of characters and words, and the characters used in the phonetic annotations (He 2005). To further uncover the linguistic value of its phonetic materials, this research focuses on the Yongle Nanzang’s Suihan Yinyi 隨函音義 (phonetic and meaning annotations accompanying the Chinese Buddhist canon) and adopts methods including the custom encoding of unencoded characters, a combination of Optical Character Recognition (abbreviated as OCR) and manual proofreading, and a mixture of similarity-based analysis and hierarchical clustering. Through these digital techniques, this study analyzes the phonological characteristics of the Yongle Nanzang and compares them with the Qishazang, aiming to investigate the relationships between them from a phonological perspective and to provide linguistic evidence for studies of canonical transmission. This study seeks to establish an interdisciplinary research framework that integrates religious studies, philology, linguistics, data science, and computer science. Its scholarly significance unfolds along two primary dimensions. On the one hand, it applies contemporary digital humanities methodologies to analyze linguistic phenomena found in the Chinese Buddhist canon. On the other hand, it employs phonological approaches to explore new avenues for the study of textual transmission and versional genealogy within Chinese Buddhist scriptures. This bidirectional and mutually corroborative research paradigm not only broadens the scope of Buddhist philological inquiry, but also achieves a productive integration between the study of religious texts and linguistic analysis.

2. Materials and Methods

2.1. Research Materials and the Construction of the Yongle Nanzang Phonological Corpus

As a distinctive form of exegetical annotation in Buddhist scriptures, Suihan Yinyi refers to glosses appended at the end of scripture boxes or scrolls, which provide phonetic notations, graphic clarifications, and semantic explanations for obscure characters and expressions in the text. The phonological components of these annotations not only reflect the phonetic developments across different historical periods, but also encapsulate features of initial and final systems that serve as critical reference points for the periodization of Chinese phonological history. Additionally, they offer valuable linguistic evidence for the identification and authentication of textual versions.

From the perspective of canonical genealogy as outlined in Figure 1, the Fuzhouzang holds triple significance. First, it established the precedent for privately sponsored editions of the Chinese Buddhist canon. Second, it marked a historical shift in the center of canonical production from the Northern to the Southern Tradition. Third, it was the first to establish a fully developed system of Suihan Yinyi, which became the direct prototype for phonetic–semantic glosses found at the end of scripture scrolls in later canons produced during the Song, Yuan, Ming, and Qing dynasties. (It is worth noting that, although some Yinyi traces appear in first-stage editions like the Zhaocheng Jinzang and the Liaozang, their origins are still under scholarly debate.) The Suihan Yinyi model was further inherited and expanded by subsequent editions such as the Sixizang, Qishazang, Hongwu Nanzang (Li and He 2003). However, the identities of the compilers of these annotations remain undocumented in extant historical sources. It is plausible that they were either groups of engravers or monastic communities involved in the carving process. The phonological systems reflected in the glosses may have derived from sound glossing traditions circulating in Southern China or from officially regulated pronunciation standards. Regardless of whether their compilation was state-directed or privately organized, the emergence of this annotation format was closely related to the practical need of Buddhist practitioners to resolve difficulties in character recognition during scriptural recitation and study.

Based on this understanding, one may reasonably infer that Suihan Yinyi annotations from different periods should exhibit diachronic variation. If the phonological systems of glosses in different canonical editions are highly consistent, this would suggest a direct line of reprinting and textual transmission. Conversely, significant phonological differences would point to conscious revision in later editions. Japanese scholar Ding (2021) conducted a sample-based analysis of Suihan Yinyi in five scriptures from the Fuzhouzang and successfully identified salient features of late Middle Chinese initials, rhyme categories, and tonal distinctions. His findings strongly support the view that Suihan Yinyi materials can faithfully reflect the actual phonological state of the language at the time of canon production.

Building on these foundations, this study integrates the canonical genealogical framework (Figure 1) with phonological analysis to investigate the following central question: as a later edition of the Qishazang, in what ways does the Yongle Nanzang inherit or innovate upon the Suihan Yinyi system of its predecessor? This inquiry seeks to provide empirical evidence, at the level of micro-phonetic features, for assessing the transmission and editorial relationship between these two canons. This study uses the Yongle Nanzang preserved in the Shandong University Library as its primary source. Dedicated phonological treatises, such as Xuanying’s Yiqiejing Yinyi 玄應一切經音義, were excluded because, strictly speaking, such works are independent phonological books and not part of the Suihan Yinyi of the Yongle Nanzang. From the Yongle Nanzang, a total of 27,832 phonological gloss entries were extracted. Following the principles of textual fidelity and digital processability, a combination of OCR and manual proofreading was adopted to digitize the gloss materials. The process specifically included (1) verifying OCR outputs; (2) formulating rules for combining unencoded characters; (3) retaining character images for reference; and (4) establishing a character set of unencoded characters within the Unicode Private Use Area (PUA).

Through the systematic processing of the Yongle Nanzang phonetic materials, two databases were constructed: (1) The Yongle Nanzang Phonetic Gloss Corpus, which records Fanqie and Zhiyin 直音 (the Target Character and its phonetic marker are homophonous) to facilitate both the handling of annotation data and the traceability of phonetic gloss materials within the Chinese Buddhist canon, the data were organized following a uniform structure: Canonical Position ID, Target Character (Beizhuzi 被註字), First Fanqie Character (Fanqie Shangzi 反切上字), Second Fanqie Character (Fanqie Xiazi 反切下字), Zhiyin, and Sutra Title (a sample is provided in

Table 1. Sample entries from the Yongle Nanzang Phonetic Gloss Corpus.

ID	Target Character	First Fanqie Character	Second Fanqie Character	Zhiyin	Sutra Title
11111211_33-000032	Jiu 糺	Ji 吉	You 酉		Da Bore Boluomiduo Jing 大般若波羅蜜多經 (600 volumes) (001)
11111211_33-000032	Qi 賾	Qi 七	e 萼		Da Bore Boluomiduo Jing (600 volumes) (001)
11111211_33-000032	Chen 齓	Chu 初	Juan 䚈		Da Bore Boluomiduo Jing (600 volumes) (001)
11111211_33-000032	Zu 足	Zi 子	Yu 遇		Da Bore Boluomiduo Jing (600 volumes) (001)
11111211_33-000032	Luan 孿	Ban 闆	Yuan 員		Da Bore Boluomiduo Jing (600 volumes) (001)
11111211_33-000032	Pi 嫓	Pi 匹	Yi 詣		Da Bore Boluomiduo Jing (600 volumes) (001)

); and (2) The Yongle Nanzang Phonological Relation Corpus, which organizes the phonological properties and relationships among the Target Characters and their annotations.

For phonological annotation, we used the 37 initial system of Middle Chinese as the basic reference, while additionally distinguishing Fei Fu Feng Wei 非敷奉微 among labials and separating Yun 云 from Xia 匣, thus forming a system of 42 initials in total. Using the phonological correspondence and phonetic annotations from the Guangyun 廣韻 as the basis for comparison, phonological attribute information—including the initial consonant, initial category, voicing (voiceless/voiced), rhyme, rhyme group (Yunxi 韻係), rhyme division (She 攝), mouth openness and roundness (Hu 呼), division (Deng 等), and tone category—was sequentially annotated for the Target Character, the First Fanqie Character, the Second Fanqie Character, or Zhiyin listed in Table 1. Through a process of extraction, annotation, and integration, we built the Yongle Nanzang Phonological Relation Database, as shown in

Table 2. Sample entries from the Yongle Nanzang Phonological Relation Database.

Fanqie	Initial	Initial Type	Voicing	Rhyme	Rhyme Group	She	Hu	Deng	Tone
Jiu 糾 Ji 吉/You 酉	Jian 見 Jian 見/Yi 以	牙 牙/喉	全清 全清/次濁	You 黝 Zhi 質/You 有	You 幽 Zhi 質/You 尤	Liu 流 Zhen 臻/Liu 流	Kai 開 Kai 開/Kai 開	SanA三A SanC三C/SanA三A	Shang 上 Ru 入/Shang 上
Ze 賾 Shi 士/Ge 革	Chong 崇 Chong 崇/Jian 見	正齒 正齒/牙	全濁 全濁/全清	Mai 麥 Zhi 止/Mai 麥	Mai 麥 Zhi 之/Mai 麥	Geng 梗 Zhi 止/Geng 梗	Kai 開 Kai 開/Kai 開	er 二 SanA三A/er 二	Ru 入 Shang 上/Ru 入
Chen 齔 Chu 初 /Juan 䚈	Chu 初 Chu 初 /Jian 見	正齒 正齒/牙	次清 次清/全清	Chen 櫬 Yu 魚/Xian 線	Zhen 臻 Yu 魚/Xian 仙	Zhen 臻 Yu 遇/Shan 山	Kai 開 He 合/He 合	SanD三D SanA三A/SanC三C	Qu 去 Ping 平 /Qu 去
Zu 足 Zi 子/Yu 遇	Jing 精 Jing 精 /Yi 疑	齒頭 齒頭/牙	全清 全清/次濁	Yu 遇 Zhi 止/Yu 遇	Yu 虞 Zhi 之/Yu 虞	Yu 遇 Zhi 止/Yu 遇	He 合 Kai 开/He 合	SanA三A SanA三A/SanA三A	Qu 去 Shang 上 /Qu 去

Note: Yayin 牙音 (velars); Houyin 喉音 (laryngeals); Zhengchiyin 正齒音 (palatals); Chitouyin 齒頭音 (alveolars); Quanqing 全清 (fully voiceless); Ciqing 次清 (semi-voiceless); Cizhuo 次濁 (semi-voiced); Quanzhuo 全濁 (fully voiced).

. The specific implementation methods refer to the procedure for the ancient manuscript version of Yupian 玉篇 (Jiang and Huang 2024).

As demonstrated in Table 2, the phonological relation database documents the phonetic properties and internal relations between the Target Character, the First Fanqie Character, and the Second Fanqie Character for each entry. This structure lays a solid foundation for future research into the correspondences of initials, places of articulation, voicing, rhyme categories, and other phonological features across different Chinese Buddhist canons.

2.2. Quantifying Initial Relationships in the Yongle Nanzang’s Suihan Yinyi

To systematically analyze the phonological features of the Yongle Nanzang, it is necessary to comprehensively consider each phonetic gloss entry in the database, including cases where the Fanqie notation yields identical or different phonetic readings. By numerically representing the phonological relationships within each entry, the overall characteristics of the phonological system of the Yongle Nanzang can be revealed in a straightforward manner. First, for each gloss entry in the Yongle Nanzang Phonological Relation Database, the initial consonants of the Target Character and the First Fanqie Character were extracted in pairs, generating a matrix of co-occurrence frequencies between initials. Next, using the Pearson correlation coefficient formula, the co-occurrence frequency matrix was transformed into a matrix of correlation coefficients, thereby quantifying the degree of similarity between the initials of Target Characters and the initials of their corresponding First Fanqie Characters, as indicated in

Table 3. Sample co-occurrence and correlation coefficient matrices of initial consonants.

The Initial of the First Fanqie Character	The Initial of the Target Character				The Initial of the Target Character
	Bang 幫	Pang 滂	Bing 並	Ming 明	Bang 幫	Pang 滂	Bing 並	Ming 明
Bang 幫	782	27	42		1.000	0.071	0.049	−0.027
Pang 滂	13	334	16			1.000	0.044	−0.030
Bing 並	20	17	852				1.000	−0.028
Ming 明				604				1.000
Notes	Co-occurrence				Correlation Coefficients

.

In this table, positive correlation coefficients indicate a positive relationship between the initials of the Target Character and the First Fanqie Character; the larger the coefficient, the stronger the correlation. Negative values indicate a negative correlation; the larger the absolute value, the weaker the relationship. A correlation coefficient of 0 suggests no relationship between the two initials. As shown, some correlation coefficients are exactly 1, indicating that the Target Character’s initial and the First Fanqie Character’s initial are phonologically identical, which corresponds to cases where the same initial category is employed (Benniu Xiangqie 本纽相切, self-same initial Fanqie).

In this study, we primarily adopt the critical value r_0.01,40 = 0.393 to assess whether two initials are significantly correlated. Additionally, r_0.05,40 = 0.304 is used as a secondary significance threshold. If the correlation coefficient between two initials exceeds these critical values, it is interpreted as evidence of a close phonological relationship or potential merger.

2.3. Hierarchical Clustering of Initial Consonants in the Yongle Nanzang’s Suihan Yinyi

Although the correlation coefficients and their significance tests provide a rigorous quantitative standard for judging whether two initials are closely related, this approach alone does not allow for a systematic visualization of the overall clustering patterns and degrees of proximity among multiple initials within the phonological system of the Yongle Nanzang. Therefore, this study additionally employs hierarchical clustering analysis, supplemented by empirical consensus values from previous scholarly studies, to further explore the phonological features of the Suihan Yinyi materials and to facilitate comparative studies with other canonical editions. Specifically, based on the correlation coefficient matrix of initials extracted from the Yongle Nanzang data, Euclidean distances between all 42 initials were calculated. The average linkage method (i.e., the unweighted pair–group method using arithmetic averages, UPGMA) was applied to compute the shortest distances between clusters, thus generating a hierarchical clustering dendrogram representing the relative proximities among initials. In the resulting dendrogram, the shorter the distance between two initials, the higher their phonetic similarity within the Yongle Nanzang’s phonological system. By analyzing the clustering structure, we can identify which initials are phonologically close enough to suggest potential mergers, confusion, or systematic variation. Moreover, to provide a comparative baseline, this study refers to established scholarly research on the phonological structure of the Hongwu Zhengyun 洪武正韻 (the Authorized Rime Book of the Hongwu Reign), particularly regarding the classification and separation of initials. By combining the results of correlation coefficient analysis and hierarchical clustering visualization, this paper aims to reveal, from a systematic phonological perspective, the distinctive features of the Yongle Nanzang’s initial system and its relationship to the Qishazang.

3. Results

3.1. Results and Analysis of Correlation Coefficients

The correlation coefficients (see

Table 4. Correlation coefficients of labial initials.

	Bang 幫	Pang 滂	Bing 並	Ming 明	Fei 非	Fu 敷	Feng 奉	Wei 微
Bang 幫	1.000	0.071	0.049	−0.027	−0.005	−0.035	−0.019	−0.028
Pang 滂		1.000	0.044	−0.030	−0.030	−0.004	−0.031	−0.031
Bing 並			1.000	−0.028	−0.025	−0.034	0.003	−0.027
Ming 明				1.000	−0.029	−0.038	−0.031	0.078
Fei 非					1.000	0.534	0.214	−0.021
Fu 敷						1.000	0.156	−0.026
Feng 奉							1.000	−0.002
Wei 微								1.000

) show that, within the labial group and among the bilabial initials (Zhongchun 重唇), Bang 幫 and Pang 滂 (0.071), Bang and Bing 並 (0.049), and Pang and Bing (0.044) are positively correlated. Among the labiodental initials (Qingchun 輕唇), Fei 非 and Fu敷 (0.534), Fei and Feng 奉 (0.214), and Fu and Feng (0.156) show positive correlations. Between bilabial and labiodental initials, only Bing and Feng (0.003) and Ming 明 and Wei 微 (0.078) show positive correlations; other initial pairs are negatively correlated. These results indicate the following characteristics of the labial initials in the Yongle Nanzang: (1) Within the bilabial and labiodental groups, respectively, multiple positive correlations between voiceless and voiced initials are observed, such as among Bang, Pang, and Bing. (2) Between bilabials and labiodentals, only Bing and Feng and Ming and Wei show positive correlations. (3) Nasal initials (Ming and Wei) do not show strong positive correlations with other labials, indicating their phonological independence. Among these, the correlation coefficient between Fei and Fu (0.534) exceeds the critical value r_0.01,40 = 0.393, indicating that these two initials can be considered merged. Although other pairs show positive correlations, their coefficients are lower than r_0.01,40 = 0.393, meaning that the other initials remain independent.

In summary, within the Yongle Nanzang, the initials Fei and Fu are not distinguished, and this series does not contrast aspirated with unaspirated features. The bilabial initials and labiodental initials, fully voiced initials and voiceless initials, and the nasal labials remain independent. However, certain phonological mergers between voiced obstruents and voiceless initials do occur, as exemplified by the character Piao 驃, whose Fanqie spelling is Pi-miao Qie匹妙切. Here, Piao is categorized under the Bing initial (a voiced obstruent), while the First Fanqie Character Pi belongs to the Pang initial (a voiceless aspirated initial).

The correlation coefficients reveal that (

Table 5. Correlation coefficients of tongue and dorsal nasal fricative initials.

	Duan 端	Tou 透	Ding 定	Ni 泥	Zhi 知	Che 徹	Cheng 澄	Niang 孃	Ri 日
Duan 端	1.000	−0.001	0.135	−0.034	0.013	−0.023	0.001	−0.032	−0.031
Tou 透		1.000	0.046	−0.033	−0.024	−0.028	−0.026	−0.030	−0.025
Ding 定			1.000	−0.032	−0.026	−0.026	−0.008	−0.029	−0.025
Ni 泥				1.000	−0.037	−0.035	−0.031	0.181	−0.023
Zhi 知					1.000	0.014	−0.001	−0.032	−0.030
Che 徹						1.000	0.021	−0.029	−0.026
Cheng 澄							1.000	−0.030	−0.027
Niang 孃								1.000	−0.028
Ri 日									1.000

), among the tongue and dorsal nasal fricative initials, the following positive correlations are observed: between the alveolar stops Duan 端 and Ding 定 (0.135) and between Ding and Tou 透 (0.046); among the retroflex stops, between Zhi 知 and Che 徹 (0.014) and between Che and Cheng 澄 (0.021); between alveolar and retroflex stops, positive correlations exist between Duan and Zhi (0.013) and between Duan and Cheng (0.001); and among the coronal nasal and dorsal nasal fricative initial, positive correlations are found between Ni 泥 and Niang 孃 (0.181),and between Ni and Ri (0.120). All other pairs of initials exhibit negative correlations. These findings indicate the following characteristics regarding the tongue and dorsal nasal fricative initials: (1) Within the categories of alveolar stops and retroflex stops, multiple positive correlations exist between voiceless and voiced initials, such as the positive correlations between Duan and Ding and between Ding and Tou. (2) Between the alveolar and retroflex stops, positive correlations exist between voiceless initials Duan and Zhi, between voiceless and voiced initials Duan and Cheng, and between the coronal nasal initials Ni and Niang with the dorsal nasal fricative Ri. (3) The coronal nasal initials and Ri do not exhibit positive correlations with other coronal initials, indicating that the nasal initials maintain independence. Although these positive correlations are observed among the coronal stops, all correlation coefficients are lower than the critical value of r_0.01,40 = 0.393, and even when considering a 95% confidence level, they remain below the critical value of r_0.05,40 = 0.304.

In conclusion, the alveolar stops, retroflex stops, and the dorsal nasal fricative initial Ri 日 in the Yongle Nanzang are each independently maintained. Nonetheless, certain phenomena of voiceless and voiced convergence do occur, as exemplified by the case of the character Tiao 跳, whose Fanqie spelling is Ta-diao Qie 他弔切. In this instance, Tiao belongs to the Ding initial, a voiced initial, while the First Fanqie Character Ta belongs to the Tou initial, a voiceless initial.

Correlation coefficients (see

Table 6. Correlation coefficients of teeth initials.

	Jing 精	Qing 清	Cong 從	Xin 心	Xie 邪	Zhuang 莊	Chu 初	Chong 崇	Sheng 生	Si 俟	Zhang 章	Chang 昌	Chuan 船	Shu 書	Shan 禪
Jing 精	1.000	−0.004	0.044	−0.021	−0.027	0.261	−0.027	−0.027	−0.026	−0.024	−0.022	−0.028	−0.036	−0.026	−0.033
Qing 清		1.000	−0.013	−0.010	−0.028	−0.025	0.049	−0.027	−0.018	−0.024	−0.028	0.009	−0.021	−0.028	−0.033
Cong 從			1.000	−0.027	0.050	−0.008	−0.034	0.026	−0.030	−0.027	−0.031	−0.029	−0.038	−0.029	−0.034
Xin 心				1.000	−0.011	−0.036	−0.033	−0.029	0.065	−0.026	−0.024	−0.026	−0.036	−0.024	−0.032
Xie 邪					1.000	−0.038	−0.034	−0.022	−0.029	−0.021	−0.030	−0.028	−0.036	−0.028	−0.031
Zhuang 莊						1.000	−0.027	−0.009	−0.033	−0.007	0.117	−0.037	−0.050	−0.037	−0.043
Chu 初							1.000	0.008	−0.017	−0.026	−0.032	0.137	−0.038	−0.027	−0.025
Chong 崇								1.000	−0.030	0.998	−0.021	−0.029	−0.030	−0.029	−0.012
Sheng 生									1.000	−0.027	−0.030	−0.029	−0.037	−0.016	−0.034
Si 俟										1.000	−0.028	−0.026	−0.034	−0.026	−0.028
Zhang 章											1.000	−0.019	−0.034	−0.013	−0.014
Chang 昌												1.000	−0.031	−0.020	0.002
Chuan 船													1.000	−0.018	0.408
Shu 書														1.000	−0.020
Shan 禪															1.000

) reveal that, within the teeth group, positive correlations are observed between the alveolar affricate and fricative initials Jing 精 and Cong 從 (0.044) and between Cong and Xie 邪 (0.050); between the retroflex affricate and fricative initials Chu 初 and Chong 崇 (0.008) and between Chong and Si 俟 (0.998); and between the palatal affricate and fricative initials Chang 昌 and Chuan 船 (0.002) and between Chuan and Shan 禪 (0.408). Among the three categories, positive correlations are found between Qing 清 and Chu (0.049), Chu and Chang (0.137), Qing and Chang (0.009), Jing and Zhuang 莊 (0.261), Zhuang and Zhang 章 (0.117), Xin 心 and Sheng 生 (0.065), and Cong and Chong (0.026). All other pairs of initials exhibit negative correlations. These results indicate the following phonological characteristics: (1) Within the three categories, positive correlations are observed among voiced affricates and voiced fricatives (Cong-Xie, Chong-Si, Chuan-Shan), suggesting that, in syllable formation, the plosive component tends to weaken, leading to confusion between affricates and fricatives. Positive correlations also occur either between voiced affricates and voiceless affricates (Jing-Cong), or between voiced affricates and semi-voiced affricates (Chong-Chu, Chuan-Chang), but not simultaneously with both voiceless and semi-voiced affricates. For example, Cong is positively correlated only with the voiceless Jing and not with the semi-voiced Qing. (2) Among the three categories, positive correlations are found among the three voiceless initial groups (Jing, Zhuang, Zhang), among the semi-voiced initial groups (Qing, Chu, Chang), among the two voiceless fricatives (Xin, Sheng), and among the two voiced affricates (Cong, Chong). Among these, the correlation coefficients between Chong and Si (0.998) and between Chuan and Shan (0.408) exceed the critical value of r_0.01,40 = 0.393, indicating that Chong and Si, as well as Chuan and Shan, can be considered merged. Although other pairs show positive correlations, their coefficients are lower than both r_0.01,40 = 0.393 and r_0.05,40 = 0.304, thus indicating that these initials remain independent.

In conclusion, in the Yongle Nanzang, Chong and Si and Chuan and Shan are not distinguished, while each voiced initial remains distinct from the corresponding voiceless or semi-voiced initials. Distinctions among the Jing, Zhuang, and Zhang dental categories are preserved, although occasional phonological confusion between Jing and Zhuang and between Qing and Chu or Chang is observed.

Correlation coefficients also reveal that (

Table 7. Correlation coefficients of velar and guttural initials.

	Jian 見	Xi 溪	Qun 羣	Yi 疑	Ying 影	Xiao 曉	Xia 匣	Yun 云	Yi 以
Jian 見	1.000	0.009	0.013	−0.019	−0.020	−0.008	0.009	−0.028	−0.025
Xi 溪		1.000	−0.009	−0.021	−0.021	−0.017	−0.020	−0.029	−0.029
Qun 羣			1.000	−0.018	−0.026	−0.025	−0.022	−0.029	−0.027
Yi 疑				1.000	−0.021	−0.024	−0.021	−0.020	−0.027
Ying 影					1.000	−0.013	−0.016	−0.002	−0.023
Xiao 曉						1.000	0.004	−0.027	−0.025
Xia 匣							1.000	−0.009	−0.005
Yun 云								1.000	0.051
Yi 以									1.000

), within the velar initials, positive correlations are observed between Jian 見 and Xi 溪 (0.009) and between Jian and Qun 羣 (0.013); among the guttural initials, between Xiao 曉 and Xia 匣 (0.004) and between Yun 云 and Yi 以 (0.051); and between the velar and guttural groups of Jian and Xia (0.009). All other pairs of initials exhibit negative correlations. These findings indicate the following phonological characteristics of the velar and guttural initials in the Yongle Nanzang: (1) Within the Jian group and the Ying group, multiple positive correlations exist among voiceless and voiced initials, such as between Jian, Xi, Qun, and Xia. (2) Between the velar and guttural initials, only the voiced Jian and Xia show a positive correlation. (3) The nasal velar initial Yi 疑 does not show a positive correlation with other velar initials, indicating the independence of nasal initial. However, all these correlation coefficients are lower than the critical value of r_0.01,40 = 0.393 and r_0.05,40 = 0.304, meaning that each initial maintains independence.

In conclusion, among the velar and guttural initials of the Yongle Nanzang, each initial remains independent. The initials of the Jian and Ying groups are distinguished, as are the voiceless and voiced initials, although occasional phenomena of voiceless and voiced merging are observed.

3.2. Results and Analysis of Hierarchical Clustering

The correlation coefficients and their significance tests indicate the degree of affinity between pairs of initials, as well as their separation or merger. However, they do not visualize the overall distribution of initial groups (comprising two or more initials) across the phonological system. Through the method of hierarchical average linkage clustering, the relative affinity distances among the initials of Yongle Nanzang are normalized within the range of [0, 1.5] (see Figure 2). This enables an intuitive display of the clustering results and the degrees of affinity among the initials across the entire system. As shown in Figure 2 (the affinity dendrogram of initials in Yongle Nanzang), the pair of initials clustered at the first hierarchical level demonstrates the highest affinity, followed by those at the second level. The affinity between two initials is inversely proportional to their intervening hierarchical levels: fewer levels indicate a closer phonological relationship and vice versa. Within the same hierarchical level, shorter line segments between initials indicate higher affinity, whereas longer segments indicate greater distance. The clustering dendrogram, thus, not only reveals the relationships among initials within the same group, but also provides insight into the relationships between each initial and all other initials in the system.

In the labial group, Bang and Pang, Fei and Fu, and Ming and Wei each form independent clusters at the first hierarchical level. This indicates that, among initials sharing the same place of articulation but differing in manner of articulation, there exists a closer relationship and higher degree of phonetic similarity, consistent with the results suggested by the correlation coefficients. However, the varying lengths of the connecting lines within each group reflect differences in internal similarity: shorter lines indicate closer relationships and higher similarity, while longer lines reflect more distant relationships. In the automatic clustering process, Fei, Fu, and Feng cluster together at the second level, as do Bang, Pang, and Bing. Subsequently, at the third hierarchical level, Bang, Pang, and Bing further merge with Ming and Wei, forming a unified cluster that reflects their relatively close phonetic relationship. In contrast, Fei, Fu, and Feng remain at a greater hierarchical distance from the Bang-Pang-Bing-Ming-Wei group and form an independent cluster composed solely of the three initials. This reveals the following phonological phenomena: within the labial group, the difference between aspirated and unaspirated sounds with the same place of articulation is relatively small. The affinity between voiced and voiceless initials is greater than that between voiced and semi-voiced nasal initials, suggesting a tendency toward devoicing among voiced initials, albeit not a pronounced one. On the other hand, initials with the same manner of articulation but different places of articulation show larger phonetic distances, especially between Fei, Fu, and Feng and Bang, Pang, and Bing, indicating a lower degree of affinity. The nasal labials Ming and Wei are more closely related to the plosive labials Bang, Pang, and Bing, suggesting that, compared to labiodental stops, they may carry certain acoustic features typical of bilabial stops.

In the tongue group, Duan and Ding, Ni and Niang, and Cheng and Che each form independent clusters at the first hierarchical level. This pattern indicates a close relationship and high phonetic similarity among initials that share the same place of articulation but differ in their manner of articulation, consistent with the results derived from the correlation coefficients. However, variations in the connecting line lengths within each group reveal differences in internal similarity: the shortest line segment appears between Ni and Niang, which aligns with the highest observed correlation within the tongue group. In the automatic clustering process, Duan, Ding, and Tou merge into a single cluster at the second hierarchical level; Ni, Niang, and Ri also cluster together at the second level. Cheng and Che form an independent cluster, as does Lai. At the third hierarchical level, Duan, Ding, and Tou further merge with Cheng and Che to form a larger cluster. In contrast, Ni and Niang are at a much greater hierarchical distance from Duan, Ding, Tou, Cheng, and Che, and only at the fourth level do they finally merge with this larger group. Multiple hierarchical levels separate alveolar stops, retroflex stop, and alveolar lateral, a pattern similar to that observed in the labial group, where initials with the same manner of articulation but different places of articulation display lower phonetic affinity. This reveals the following phonological patterns: the phonetic affinity between voiced initials and voiceless initials is greater than that between voiced and semi-voiced initials, suggesting a tendency toward devoicing among voiced initials within the coronal group. Nasal initials show relatively low phonetic affinity with plosive initials, especially in the case of Ni and Niang, whose distance from Ding and Cheng indicates that the voiced nasals Ni and Niang likely do not share the acoustic features typically associated with voiced plosives. Notably, the retroflex initial Zhi and the palatal affricate and fricative initial Zhang merge at the first hierarchical level, indicating a phonological relationship between the third-division retroflex group Zhi and the third-division palatal affricate and fricative group Zhang. This may reflect the later historical convergence of the Zhi, Zhuang, and Zhang groups into retroflex initials.

In the teeth group, the pairs of Cong and Xie, Chong and Si, and Chuan and Shan form a cluster at the first hierarchical level. Jing and Zhuang and Chu and Chang, as well as Xin and Sheng, also form clusters at the first level, respectively. However, based on the length of connecting lines, the average line segments among the first three pairs are shorter than those of the latter three, indicating that the phonetic affinity within the first three pairs is higher than that within the latter three. In the automatic clustering process, Jing and Zhuang merge with Zhi and Zhang at the second hierarchical level; Chu and Chang cluster with Qing at the second level; and Xin and Sheng merge with Shu at the second level. Cong and Xie, Chuan and Shan, and Chong and Si each form independent clusters, while Ri merges with Ni and Niang at the second level. These results indicate the following phonological phenomena: among fully voiced initials in the dental group, those with the same place of articulation but different manners of articulation show closer phonetic affinity, whereas, among voiceless initials, those with the same manner of articulation but different places of articulation show greater affinity. The relatively large distances between Cong and Xie and Jing and Qing, between Chong and Si and Zhuang and Chu, and between Chuan and Shan and Zhang and Chang reflect the independence of fully voiced initials from voiceless ones. The clustering of Jing and Zhuang with Zhi and Zhang and of Chu and Chang with Qing, both at the second level, further illustrates the tendency for phonological merging among the Jing, Zhuang, and Zhang groups. Within the dental group, there is only one nasal fricative, Ri, which merges with the apical nasals Ni and Niang in the clustering process, reflecting the phonetic similarity among the three and the strong influence of manner of articulation on the phonetic affinity between initials.

In the velar and guttural group, Jian and Qun, Xia and Xiao, and Yun and Yi each form clusters at the first hierarchical level. Subsequently, Jian and Qun merge with Xi at the second level, while Yun and Yi cluster with Ying at the second level. At the third hierarchical level, Jian, Qun, and Xi further merge with Xia and Xiao to form a larger cluster, while Yi forms an independent cluster. In the automatic clustering process, there is no instance of initials within the Jian group or the Ying group merging together at the first level, and the connecting lines within each group are relatively long when clustering occurs. This indicates the following phonological phenomena: within the Jian and Ying groups, initials sharing the same place of articulation but differing in manner of articulation display closer phonetic affinity, while the phonetic distance between the Jian and Ying groups themselves remains large, suggesting a clear distinction between the two groups. Furthermore, multiple hierarchical separations are observed between Ying and Yi, and no significant phonetic similarity is found, implying that there is no merging tendency between them and that they remain independent, a finding consistent with the results of the correlation coefficient analysis. Although voiced initials occasionally cluster together with their corresponding voiceless initials, the longer connecting lines suggest that a clear phonological distinction between voiced and voiceless initials remains.

4. Discussion

4.1. Affinity Characteristics of the Phonological System in the Yongle Nanzang

The clustering dendrogram provides an intuitive visualization of the affinity relationships among all initials within the phonological system of the Yongle Nanzang. However, it does not directly indicate which initials should be considered merged or separated. Quantitative research must be grounded in qualitative findings. Therefore, in the hierarchical clustering analysis, no specific numerical threshold was applied for distinguishing mergers; rather, previous research was used as a reference to determine the points of comparison for mergers and separations among initials.

This study adopts the Hongwu Zhengyun as a phonological reference point for the comparative analysis of the phonological features observed in the Yongle Nanzang. The rationale for this choice is grounded in two primary considerations.

First is the historical context of language policy.

Following the establishment of the Ming dynasty and the relocation of the capital to Nanjing, the Hongwu Emperor Zhu Yuanzhang 朱元璋 implemented a series of measures aimed at reestablishing the orthodox order of Han Chinese culture, including the standardization of language and script. Recognizing that the rhyme books inherited from the Song and Yuan periods no longer reflected the linguistic realities of the time—and in particular seeking to eradicate the influence of Mongolian phonology during the Yuan dynasty—the court issued decrees to abolish earlier rhyme books such as the Zhongyuan yinyun 中原音韻 and Menggu ziyun 蒙古字韻. In their place, the so-called Zhongyuan Yayin 中原雅音 was adopted as the standard phonological norm for compiling official rhyme books2. The Hongwu Zhengyun, first compiled in 1375, was produced within this political and linguistic framework. The compilation team was primarily composed of scholars from Southern China, including the chief editor Yue Shaofeng 樂韶鳳 (from Chuzhou, Anhui) and Song Lian 宋濂 (from Jinhua, Zhejiang). Notably, by the 23rd year of the Hongwu reign (1390), the rhyme book had undergone several rounds of revision. Among the twenty identifiable contributors across these revisions, fifteen were of southern origin, while only two were from the north (Ning 2003, pp. 3–9). As a result, despite its prescriptive aims, the Hongwu Zhengyun inevitably absorbed certain phonological traits of Southern Chinese varieties. This continuous revision process underscores the Ming court’s determination to enforce a unified linguistic standard across the realm.

Second is the close temporal and spatial correspondence between the two sources.

As shown in Figure 1, the compilation and publication periods of the Hongwu Nanzang (ca. 1372–1402) and its re-engraved edition, the Yongle Nanzang (ca. 1413–1420), overlapped significantly with the compilation and revision timeline of the Hongwu Zhengyun. These projects also shared the same geographic locus—Nanjing—and were both government-sponsored cultural undertakings. It is, therefore, highly plausible that the canon’s production was influenced by contemporary phonological policies. This influence is reflected in at least three aspects: (1) To ensure accurate recitation among the monastic community, the engraving institutions would have had to adjust the Suihan Yinyi in accordance with the official pronunciation norms. (2) Textual source selection reflected political intent. The Hongwu Nanzang, of which the Yongle Nanzang is a re-engraving, was based on the Qishazang, with only minor adjustments to the sequence of texts according to the Puningzang (Li and He 2003). This deliberate avoidance of Yuan-era canon aligns with the broader policy objective of eliminating Mongol linguistic influence, as seen in the repudiation of the Menggu Ziyun. (3) There is empirical evidence of rhyme book influence. The Hongwu Zhengyun was officially designated as the phonological standard for use in the Yongle Dadian 永樂大典3. This provides direct evidence of the rhyme book’s authority and influence over other major state-sponsored publications.

According to existing studies by Wang (2001), Gao (2007), and Yang (2014), it is generally agreed that, in the Hongwu Zhengyun, mergers had already occurred between Fei and Fu, as well as between Ni and Niang. In the clustering dendrogram, the line segment connecting Ni and Niang is the longest among first-level clusters. Therefore, the length of the Ni-niang segment is taken as a reference point for distinguishing merged from unmerged initials (indicated by the dashed vertical line in Figure 2, representing the strict criterion for merger). Initials clustering to the right of this line are considered merged, while those to the left are considered separate. Additionally, based on previous consensus (Wang 2001; Dong 2004; Yang 2014), the Wei initial is recognized as independent. Accordingly, the line segment length of the Ming Wei cluster (solid vertical line in Figure 2) serves as a broader reference standard (looser criterion). Thus, initials whose first-level cluster segment is longer than that between Ming and Wei are considered independent, whereas those with shorter segments are tentatively treated as merged. Where the segment length between two initials falls between that of Ni-niang and Ming-wei, the decision is made by cross-referencing their correlation coefficients.

In the labial group, the pairs Bang and Fei, Pang and Fu, and Bing and Feng do not cluster at the first hierarchical level, and each pair is separated by multiple levels. This suggests that the labials are divided into bilabials and labiodentals. Among the bilabials, Bang and Pang cluster together at the first level, but their connecting line segment is longer than that Ming-wei; thus, they should be considered separate. Bang, Pang, and Bing cluster at the second level, suggesting phonetic proximity among them, with occasional confusion but no complete merger. Among the labiodentals, Fei and Fu cluster at the first level, and their connecting line segment is significantly shorter than that Ni-niang, indicating that Fei and Fu have essentially merged. Feng clusters with Fei and Fu at the second level. Although the connecting line is longer than that of Ni-niang but shorter than that of Ming-wei, correlation coefficients and significance tests (correlation between Feng and Fei is 0.214; correlation between Feng and Fu is 0.156; critical value r_0.05,40 = 0.304) indicate that Feng remains independent but exhibits occasional mixing with Fei and Fu. Although Ming and Wei cluster at the first level, the connecting line segment is longer than that of Ni-niang, and the correlation between Ming and Wei does not reach statistical significance (correlation 0.078, r_0.05,40 = 0.304), indicating that Ming and Wei are independent.

In summary, the phonological characteristics of labials in the Yongle Nanzang are as follows: The labials are divided into bilabials and labiodentals. The distinction between aspirated and unaspirated sounds is not prominent: Bang and Pang exhibit occasional confusion but are not merged, while Fei and Fu are merged. Although voiceless and voiced sounds occasionally mix, the full devoicing of voiced initials has not occurred. Bing occasionally mixes with Bang, and Feng occasionally mixes with Fei and Fu, and yet fully voiced initials remain independent. Ming and Wei are also independent.

In the tongue group, the pairs Tou and Che, Ding and Cheng, and Duan and Zhi do not cluster at the first hierarchical level, and each pair is separated by multiple hierarchical levels. This indicates that, within this phonological system, the alveolar sounds of Duan and the retroflex sounds of Zhi remain distinct, with clear differentiation between the Duan and Zhi groups. Although Ni and Niang cluster at the first hierarchical level, their correlation does not reach statistical significance (correlation coefficient 0.181, r_0.05,40 = 0.304). Additionally, Ni and Ri cluster together at the second level, reflecting a residual feature of the Old Chinese phonological system, where Ni and Ri were not fully differentiated. Nevertheless, neither Ni nor Ri cluster with any stop consonants. Within the alveolar sounds, Ding and Duan cluster together at the first level, with their connecting line segment longer than that of Ni-niang but shorter than that of Ming-wei. Furthermore, their correlation fails to reach significance (correlation 0.135, r_0.05,40 = 0.304), indicating that Ding and Duan are independent but exhibit occasional phonetic mixing. At the second hierarchical level, Ding, Duan, and Tou cluster together, indicating phonetic proximity among these initials, with more frequent voiceless and voiced confusion between Ding and Duan than between Ding and Tou. Within the retroflex group, Cheng and Che cluster at the first level. However, the connecting line segment is longer than that of Ming-wei, suggesting that the two should be considered independent. Moreover, Cheng and Zhi are separated by several hierarchical levels. Thus, among the coronals, the full devoicing trend is most evident for Ding, followed by Cheng, with Ding demonstrating a clearer tendency toward full devoicing, and Cheng showing a secondary tendency. The lateral initial Lai does not cluster with other initials, indicating that Lai remains independent.

In summary, the phonological characteristics of the tongues in the Yongle Nanzang are as follows: Tongues are divided into alveolar and retroflex groups, with the lateral initial Lai remaining independent. There is evidence of voiceless and voiced phoneme convergence, with mixing between Ding and Duan more frequent than between Cheng and Che. A full devoicing trend is most evident for Ding, followed by a partial devoicing trend for Cheng, although neither has fully completed devoicing. Ni exhibits greater phonetic proximity to Niang than to Ri, with occasional mixing between Ni and Niang. However, all three nasal initials—Ni, Niang, and Ri—remain independent.

In the teeth group, Jing and Zhuang cluster together at the first hierarchical level, and the length of their connecting line is shorter than that of Ni-niang; thus, Jing and Zhuang are considered merged. Zhang and Zhi, as well as Chu and Chang, also cluster at the first level. However, their line segments are longer than those of Ni-niang, and yet shorter than those of Ming-wei, and their correlations fail to pass the significance test (Zhang–Zhi: 0.157; Chu–Chang: 0.137; r_0.05,40 = 0.304). Therefore, Zhang and Zhi and Chu and Chang are phonologically independent but exhibit phonetic affinity. Qing clusters with Chu and Chang at the second level, indicating phonetic correlation among the three, though each remains independent. The phonetic distance between Xin and Sheng is greater than that between Ming and Wei, indicating that Xin and Sheng are independent. Aside from the above-mentioned clusters at the first level, no other initials within the Jing, Zhuang, and Zhang groups cluster at the first level, meaning that, apart from the merger between Jing and Zhuang, these three groups remain phonologically distinct. Among the voiced initials within the three groups, the pairs Cong and Xie, Chong and Si, and Chuan and Shan all cluster at the first level. Chong and Si and Chuan and Chan show closer phonetic proximity than Ni-niang, and thus are considered merged. Cong and Xie, although clustered at the first level, have a greater phonetic distance than Ming-wei, and are, therefore, considered independent. The fully voiced initials in the Jing, Zhuang, and Zhang groups do not cluster with their corresponding voiceless initials and are separated by multiple hierarchical levels, indicating that the voiced initials are phonologically independent.

In summary, the phonological characteristics of the dental initials in the Yongle Nanzang are as follows: Jing and Zhuang show relatively close phonological affinity with Jing and Zhuang merging and Chong and Si merging; while other initials may occasionally show confusion, they remain independent. The Zhang group is more distant from the Jing and Zhuang groups, but shows closer affinity with the Zhi group. Chuan and Shan are merged, while other initials within the Zhang group are independent. Voiced initials are phonologically independent.

In the velar and guttural group, the Jian group and Ying do not cluster at the first level, indicating that the Jian and Ying groups are phonologically separate and independent. Within the Jian group, Jian and Qun cluster at the first level, but their phonetic distance exceeds that of Ming-wei, and, thus, they are considered independent. Jian, Qun, and Xi cluster together at the second level, indicating phonetic affinity among the three. The voiced Qun shows greater similarity with the voiceless Jian than with the aspirated Xi. The nasal initial Yi does not cluster with any other initials in the group, and is, therefore, considered independent. Within the Ying group, Xiao and Xia, as well as Yun and Yi, each form clusters at the first level. However, in both cases, the phonetic distances are greater than those of Ming-wei, suggesting phonetic affinity within each pair but phonological independence of each initial. Ying clusters with Yun and Yi at the second level, while Ying remains phonologically independent. Yun does not cluster with Xia, indicating that third-division Yun shows greater phonetic similarity to third-division Yi than to Xia.

In summary, the phonological characteristics of the velar and guttural initials in the Yongle Nanzang are as follows: the Jian and Ying groups are distinct and independent; within both groups, individual initials remain phonologically independent. The nasal initial does not mix with either voiceless or voiced initials. The voiced Qun shows closer affinity with voiceless initials than with semi-voiced ones. Yun and Yi show greater phonetic proximity, while their relationship to Xia is more distant. Xiao and Xia occasionally mix, but not with Yun.

4.2. Examining the Relationship Between the Yongle Nanzang and the Qishazang

Among the various canons engraved in Southern China, the Yongle Nanzang was unique because, as Lucille Chia shows, it was largely sponsored by private funds and played a transitional role in continuing this Southern Tradition from the Song and Yuan to the Ming and Qing dynasties. According to Li Fuhua, the structural innovations observed in the Yongle Nanzang represented the fourth system, distinct from the three typologies previously proposed by Chikusa Masaaki and Fang Guangchang (Wu 2015). Some scholars have argued that the Yongle Nanzang is merely a re-engraved version of the Qishazang. This raises a critical question: Was it a direct facsimile, or were there substantial revisions?

Li and He (2003) conducted a preliminary comparative study of the Suihan Yinyi found in Song, Yuan, Ming, and Qing editions of the canon. Their findings indicate that the phonological glosses in the Song and Yuan canons were primarily derived from the Chongningzang and exhibit only minor internal variations, whereas they differ significantly from those found in Ming and Qing editions. One key issue emerging from this comparison is whether these differences manifest in the Shengmu Xitong 聲母系統 (initial consonant system), which could serve as a determining criterion for whether the Yongle Nanzang represents a distinct fourth system.

This line of inquiry may be analogized to the practice of annotating the Shijing 詩經 (Classic of Poetry, the earliest anthology of Chinese poetry) using different phonological systems. If the Shijing were transcribed in Minnan, Cantonese, and Modern Standard Chinese, one would generate, respectively, a Minnan version, a Cantonese version, and a Putonghua version of the Shijing. Likewise, if annotated using Old Chinese, Sui-Tang phonology, or Modern Mandarin, it would result in Old Chinese, Sui-Tang, and Modern Mandarin versions of the Shijing. Both synchronic variation across regional dialects and diachronic evolution across historical periods lead to diverse phonological renderings of the same text. These divergences in phonological systems—particularly in both initial and final systems—naturally give rise to distinct textual versions.

Based on the combined results of correlation coefficients and hierarchical clustering analysis, we have identified the phonological characteristics of the Yongle Nanzang. By comparing these characteristics with those of the Qishazang, we aim to explore the relationship between the two. Existing research has laid a solid foundation for this comparison. Shao (2012) offers a relatively comprehensive study of the phonological system in the Qishazang, while Tan (2015), Li and Lu (2021), and other scholars have discussed the phonetic features of the Song–Yuan period as reflected in the Qishazang, primarily using methods such as case-based analysis, percentage comparison, and fanqie comparison. By synthesizing these findings, the phonological system of the Qishazang has been established. A comparative table of the initials in the Yongle Nanzang and the Qishazang is presented below (

Table 8. Comparative table of the initials in the Yongle Nanzang and the Qishazang.

Place of Articulation	Yongle Nanzang Initial System	Qishazang Initial System
Labials	Bang 幫, Pang 滂, Bing 並, Ming 明	Bang 幫, Pang 滂 (Bing 並), Ming 明
	Fei 非 (Fu 敷), (Feng 奉), Wei 微	Fei 非, Fu 敷 (Feng 奉), Wei 微
Tongues/ Coronals	Duan 端, Tou 透, Ding 定, Ni 泥	Duan 端, Tou 透 (Ding 定), Ni 泥 (Niang 孃)
	Zhi 知, Che 徹, Cheng 澄, Niang 孃, Lai 來	Zhi 知, Che 徹 (Cheng 澄), Lai 來
Teethes/ Dentals	Jing 精 (Zhuang 莊), Qing 清, Cong 從, Xin 心, Xie 邪	Jing 精, Qing 清 (Cong 從), Xin 心 (Xie 邪)
	Chu 初, Chong 崇 (Si 俟), Sheng 生	Zhuang 莊, Chu 初, Chong 崇, Sheng 生/(Zhang 章, Chang 昌, Chuan 船, Shu 書, Chan 禪, Ri 日) (Si 俟)
	Zhang 章, Chang 昌, Chuan 船, Shan 禪, Shu 書, Ri 日
Velars	Jian 見, Xi 溪, Qun 羣, Yi 疑	Jian 見, Xi 溪 (Qun 羣), Yi 疑
Gutturals	Ying 影, Xiao 曉, Xia 匣, Yun 云, Yi 以	Ying 影, Xiao 曉 (Xia 匣), Yu 喻

). While previous research has concentrated primarily on the Qishazang, our study introduces a new dimension by applying quantitative methods to analyze the phonological features of the Yongle Nanzang, thereby offering fresh insights into the diachronic development of the Chinese Buddhist canon.

In the labial group, both systems exhibit a division into bilabials and labiodentals. However, differences are noted. In the Yongle Nanzang, the distinction between aspirated and unaspirated sounds is not prominent: Fei and Fu have merged, while Bang and Pang show occasional mixing without a complete merger. Distinctions between Bang and Pang remain evident. In the Qishazang, the boundary between bilabials and labiodentals is blurred. There is a trend toward mergers of initials with the same manner of articulation, and even interchange between bilabials and labiodentals is observed, such as Fei ↔ Wei, Feng ↔ Ming, Bang ↔ Feng, and Bing ↔ Fei.

In the coronal group, both systems show separation between alveolar and retroflex sounds, and mixing between Ni and Niang is observed. However, in the Yongle Nanzang, Ni and Niang show occasional confusion without a complete merger, and neither merges with Lai. In the Qishazang, Ni and Niang merge and further mix with Lai. Confusion among Duan and Zhi, as well as between Tou and Che and between Ding and Cheng, is more extensive.

In the dental group, in the Yongle Nanzang, Jing and Zhuang merge, while other initials of the Jing, Zhuang, and Zhang groups remain independent. There is a closer phonetic relationship between the Jing and Zhuang groups, possibly reflecting a residual phenomenon of “Zhao-er merging into Jing.” The Zhang group maintains greater distance from the Jing and Zhuang groups but is relatively closer to the Zhi group, suggesting a remnant of “Zhao-san merging into Zhi.” Occasional mergers like Chuan and Shan are observed. Voiced initials remain independent. In contrast, in the Qishazang, the Zhuang and Zhang groups merge and mix with the Zhi group. The Jing group mixes with the Jian group. Ri mixes with Shan and Yu.

For velar and guttural initials, in the Yongle Nanzang, the Jian and Ying groups are independent, and initials within each group also maintain independence. The nasal initials do not mix with voiceless or voiced sounds. Qun shows closer proximity to Jian than to Xi. Yun and Yi are more closely related than either is to Xia. Occasional confusion occurs between Xiao and Xia, but Yun does not mix with Xia. In contrast, in the Qishazang, the Ying group merges and mixes with the Jian and Jing groups. Xiao mixes with Xia and Yu. The third- and fourth-division glides (Yu) merge and mix with Ying.

Additionally, in the Yongle Nanzang, although voiceless and voiced initials occasionally mix, full devoicing has not been completed: Bing mixes occasionally with Bang and Pang; Feng mixes occasionally with Fei and Fu; and Ding mixes more frequently with Duan than Cheng does with Che. Voiced initials such as Chong and Chuan have merged, and Qun is closer to voiceless sounds than to aspirated ones. Xia occasionally mixes with Xiao but not with Yun. Nasal and lateral initials remain independent. By contrast, in the Qishazang, the full devoicing of voiced initials has already become systematic: some voiced initials transform into voiceless unaspirated initials, others into voiceless aspirated initials. Ding and Cheng transform into corresponding voiceless sounds; Chuan transforms into Zhang; and voiced fricatives become voiceless fricatives.

The comparative analysis of initial consonants in these two canons demonstrates that, while the Yongle Nanzang and the Qishazang both belong to the same textual lineage, notable divergences exist in their respective phonological systems and inter-consonantal relationships. These systematic differences in the initial consonant inventory lend further support to the view articulated in Wu (2015): “Li Fuhua believes that the new structure in Yongle Southern Canon represented the fourth system”. This suggests that the second and third phases of canon production (see Figure 1) encapsulate distinct phonological profiles corresponding to different historical stages of Chinese Buddhist textual transmission. From a linguistic periodization perspective, such variation signals the broader phonological shift from Middle Chinese to Early Modern Chinese.

The emergence of these distinctions reflects not only linguistic evolution, but also the ideological tensions underlying the compilation of Buddhist canons between state-sanctioned standardization and regional vernacular traditions. As a state-directed cultural project, the Yongle Nanzang was profoundly shaped by the phonological norms codified in the Hongwu Zhengyun, an official rhyme book compiled under imperial authority. For instance, although the Hongwu Zhengyun preserves the full series of voiced initials (such as Bing, Feng, Ding, and Cheng), these sounds had already disappeared from Nanjing dialect by the early Ming period. Nevertheless, the Suihan Yinyi of the Yongle Nanzang retains these voiced initials, illustrating how official phonological standards could override contemporary spoken forms in canonical texts. At the same time, the Yongle Nanzang also exhibits traces of the actual pronunciation habits of local Nanjing engravers or monastic editors. For example, phonological features such as the neutralization of labials (Qingchun merging Fei and Fu) and the independent treatment of the Wei—both characteristic of Nanjing speech—are reflected in the Yongle Nanzang system of initials. In contrast, the Qishazang, produced outside the reach of Ming government intervention, more faithfully records dialectal features from southern coastal regions such as Zhejiang and Fujian. These include the distinct realization of Fei and Fu, the devoicing of fully voiced initials, and the merger of the Zhuang and Zhang series. Consequently, the Yongle Nanzang may be viewed as a politicized product of Ming dynasty language policy, while the Qishazang better preserves the vernacular phonological characteristics of the Song–Yuan era. These distinctions underscore the stage-specific nature of the Dazangjing in Chinese.

Furthermore, the phonological profile of a newly discovered Buddhist canon can assist in determining its textual lineage. In addition to philological methods, if its system of initials or finals aligns with that of the Qishazang, it may be classified within the Qisha lineage; conversely, a system consistent with the Yongle Nanzang would place it within the Yongle line.

5. Conclusions

The core function of the Suihan Yinyi in Chinese Buddhist canons lies in its provision of phonological and semantic annotations for obscure characters, polyphonic forms, and polysemous terms in the scriptures. These annotations are designed to ensure that both monastics and lay followers can recite the sutras accurately, thereby facilitating the effective dissemination of Buddhist teachings. This practical aim necessitates that the Suihan Yinyi be grounded in the prevailing phonological norms of its time. At the same time, the regional backgrounds of the engravers or sponsoring institutions may have introduced local dialectal features into the annotations. As a result, Suihan Yinyi materials from different historical periods or geographic regions inevitably exhibit distinct diachronic and regional phonological characteristics. As illustrated in Figure 1, although both the Qishazang and the Yongle Nanzang were engraved in Southern China, they differ significantly in origin and institutional context. The former was produced during the Southern Song and Yuan periods by temple communities and local sponsors south of the Yangtze River, whereas the latter represents a centrally administered and officially revised edition of the canon from the early Ming dynasty. These contrasting circumstances resulted in fundamental differences between the two canons across several dimensions: (1) Historical context: The Qishazang was produced during the transitional period between the Song and Yuan dynasties, whereas the Yongle Nanzang emerged under the early Ming unification efforts. (2) Institutional sponsorship: The former reflects the decentralized temple-based initiatives of the Buddhist community, while the latter was orchestrated by the Ming central government. (3) Regional variation: The Qishazang was produced in Suzhou, while the Yongle Nanzang was engraved in Nanjing, two cities with distinct phonological traditions. (4) Impact of language policy: The Qishazang was developed largely in the absence of direct state intervention, whereas the Yongle Nanzang was shaped by official language standardization efforts.

This study adopts an interdisciplinary approach, incorporating digital processing and quantitative analysis to examine a dataset of 27,832 Suihan Yinyi entries from the Yongle Nanzang. By analyzing the linguistic features of these glosses from a phonological perspective, this study not only enhances our understanding of canon transmission and textual stratification, but also proposes a new paradigm for the interaction between religious literature and historical linguistics. The main findings are as follows: (1) Labials: The labials are divided into bilabials and labiodentals. The distinction between aspirated and unaspirated sounds is not prominent; Bang and Pang occasionally mix but have not merged, while Fei and Fu have merged. Although there is occasional mixing between voiceless and voiced sounds, complete devoicing has not occurred. Bing occasionally mixes with Bang and Pang, and Feng occasionally mixes with Fei and Fu. The initials Ming and Wei remain independent. (2) Coronals: The coronals are divided into alveolar and retroflex groups. Within the alveolar group, initials sharing the same place of articulation (Duan, Tou, and Ding) are more prone to mixing. The retroflex group (Zhi, Che, and Cheng) shows greater phonetic affinity with initials from the Zhang, Zhuang, or Jing groups. Although there is occasional mixing between voiceless and voiced sounds, full devoicing has not been completed; Ding exhibits more frequent mixing with Duan than Cheng with Che, with Ding showing a more pronounced trend toward complete devoicing. The nasal initials Ni and Niang show higher phonetic proximity than either does with Ri, and yet all three nasal initials remain independent and distinct from Lai. (3) Dentals: The dentals are broadly divided into Jing, Zhao, and Zhang groups. The Jing and Zhuang initials have merged. The Jing and Zhuang groups exhibit closer phonetic relationships, whereas the Zhang group is relatively closer to the Zhi group. The fully voiced initials Chong and Si are merged; Chuan and Chan are also merged, while Cong and Xie show occasional confusion but have not fully merged. Voiced initials remain independent. (4) Velars and gutturals: The velar and guttural initials are divided into the Jian group and the Ying group. The nasal initials do not mix with voiceless or voiced initials. The voiced initial Qun exhibits closer phonetic proximity to the voiceless unaspirated Jian than to the voiceless aspirated Xi. Yun and Yi show closer relationships to each other than either does to Xia, but all three remain independent. Occasional mixing occurs between Xiao and Xia, but Yun does not mix with Xia. Apart from occasional mixing between the semivowel Yu and Yun, the semi-voiced nasal and lateral initials remain independent.

The distinctive patterns of initial consonant mergers and splits observed in the Yongle Nanzang not only serve as the basis for reconstructing phonetic values, but also constitute the primary object of such reconstruction. For instance, for the non-distinction between the Fei and Fu categories, phonemic reconstruction demonstrates that they constitute a single phoneme. It is essential to emphasize that phonological systems from different historical periods exhibit distinct temporal features, and, thus, phonological data from various periods must be carefully distinguished in the process of sound value reconstruction. In reconstructing the phonological values reflected in the Yongle Nanzang, one must not only rely on general principles of historical phonology in Chinese, but also systematically integrate a wide array of synchronically and diachronically relevant sources. These include dialect records from the Nanjing region in the early Ming dynasty, contemporary Sino-Sanskrit transliterations, and phonetic materials from Chinese language primers compiled by missionaries during the same period. Only through this comprehensive and methodologically rigorous triangulation of evidence can the reconstructed phonetic values be brought as close as possible to the actual pronunciations at the time of the Yongle Nanzang’s engraving. This methodological principle is reinforced by a retrospective examination of Bernhard Karlgren (1889–1978), whose pioneering reconstructions of Middle Chinese were later critiqued for conflating phonological materials from disparate temporal and spatial contexts. Karlgren did not sufficiently differentiate between the Qieyun 切韻 (Sui dynasty) and the Guangyun (Song dynasty), and indiscriminately incorporated data from Japanese Go-on 吳音 (the older layer of Sino-Japanese pronunciation) and Kan-on 漢音 (the Tang dynasty standard pronunciation), Korean and Vietnamese loanwords, Yuntu 韻圖 (phonological charts) of the Kangxi Zidian 康熙字典 (Qing dynasty), modern dialect surveys, and dialect dictionaries. This lack of attention to the temporal and regional specificity of the data led to reconstructions that have since been called into question (He 1961; Pulleyblank 1984). The present study treats the reconstruction of the phonetic system in the Yongle Nanzang as a separate and ongoing project, involving the systematic collection and critical evaluation of relevant historical materials.

Through a comparative analysis of the initial consonant systems between the Yongle Nanzang and the Qishazang, several important observations emerge: (1) The differences in the initial systems are not isolated or incidental but reflect two distinct phonological standards. This indicates that the Yongle Nanzang was not a straightforward reprint of the Qishazang, but rather a key component of the Ming dynasty’s broader agenda of phonological standardization. (2) The phonetic and semantic features in the Yongle Nanzang reveal a hybrid nature; while adhering to the officially endorsed standard—exemplified by the Hongwu Zhengyun—they also retain dialectal imprints from the Nanjing engraving teams. (3) These findings affirm the Yongle Nanzang’s status as a distinct canon shaped by official language policy and justify its classification as a new type of Buddhist canon, different from its Song–Yuan editions.

Whether similar initial consonant systems appear in other Buddhist canons remains an open question. As illustrated in Figure 1, the canons most closely related to the Yongle Nanzang include the Hongwu Nanzang and the Yongle Beizang. Existing research suggests that, although the phonetic glossing system in the Yongle Nanzang is derived from that of the Hongwu Nanzang, it exhibits several systematic differences: (1) A significant reduction in the number of glossed entries: For example, in the first scroll of the Da Bore Boluomiduo Jing within the Tian character folder (Tianzi Han 天字函), the Hongwu Nanzang glosses 23 characters, while the Yongle Nanzang retains only 2. (2) Variations in fanqie spellings: The character Qī 𦣱, for instance, is glossed as Shi-ge Fan 士革反 in the Yongle Nanzang, whereas it appears as Shi-ze Fan 士責反 in the Hongwu Nanzang. Whether these differences reflect actual shifts in phonological systems remains to be verified through rigorous phonological analysis. Our research team is currently undertaking a comprehensive digital and phonological study of the Hongwu Nanzang to clarify its phonetic correspondences with the Yongle Nanzang. As for the Yongle Beizang, an imperial edition compiled during the same reign as the Yongle Nanzang, its phonetic glossing system exhibits the following features: (1) a balanced emphasis on both phonetic and semantic annotation, in contrast to the Yongle Nanzang, which primarily focuses on difficult pronunciations, and (2) its reprinting process followed a series of imperial directives issued by Emperor Taizong of the Ming, resulting in revisions not only in textual collation, but also in phonetic annotation. Based on current research, it is still premature to determine whether these canons exhibit phonological systems comparable to that of the Yongle Nanzang. Further systematic investigation into both the Hongwu Nanzang and Yongle Beizang is necessary. Given the period-specific characteristics of Suihan Yinyi, it is unlikely that canons from the earlier phases (Stages 1 and 2 in Figure 1) would contain features consistent with the initial consonant system observed in the Yongle Nanzang.