Hidden prosody in philology: yìyŭ ‘transcriptions’ in the 15th century

This paper investigates how prosody is hidden behind transcriptions in historical resources. Three historical sources are used in the analysis. They are Chinese transcriptions from the 15th century in which Japanese, Korean and Ryukyuan phrases are recorded using Chinese characters. The argument concentrates on the prosodic patterns of disyllabic nouns in the three historical sources. The results of chi-square tests show that in the transcriptions Korean is significantly different from Japanese and Ryukyuan. In disyllabic nouns, the Chinese tonal category shăngshēng is favored in the first syllable of the Korean data to show changes from low to high tone. On the other hand, the transition is not salient in the Japanese and Ryukyuan data. In addition, the Chinese tonal cateogry yīnpíng is disfavored in the first syllable of the Korean data, whereas Chinese yīnpíng is not overtly excluded from the first syllable of Japanese and Ryukyuan data. This paper also discusses the projection of prosodic characteristics from Chinese onto the transcriptions: the second syllable in a disyllabic noun tends to be qùshēng.


Introduction
Reconstructing prosody is seldom an easy task in historical linguistics, and most attention in reconstruction has been paid to segments. Recovering a phonological inventory of any language can be done by applying the comparative method based on modern languages/dialects or by deciphering historical sources. The comparative method is usually applied prior to using historical sources that not only document the earlier stages of a language but also determine aspects of its history (Campbell 2004: 361).
Reconstructing prosody using modern languages/dialects has been done, for example, by Shimabukuro (2007) for Japanese and Ryukyuan pitch accent and Pittayaporn (2009) for tones of Proto-Tai. In addition to the use of modern languages/dialects, prosody can also be reconstructed through the use of historical sources, as we see, for instance, in the long tradition of studies in Old and Middle English stress (e.g. Sievers 1893, Minkova 2006Minkova , 2009Minkova , 2013 or in Ting's (1975) work on Middle Chinese tones. As Minkova (2013: 260) points out, as it lacks modern recording apparatus and proper analytical software, the foundation of prosodic reconstruction based on philological evidence should be secondary to using evidence from modern languages/dialects; nevertheless, the significance of philology in historical linguistics cannot be ignored.
When philological data are used to reconstruct prosody, different writing systems need different deciphering methods. When a writing tradition, that of early English for example, can preserve prosodic features, direct deciphering becomes more straightforward. Based on evidence from alliterative verses, for instance, scholars are able to posit that early English stress was root-initial (cf. Minkova 2006Minkova , 2009. Chinese, on the other hand, is written in logographic characters, and tones are not positionally assigned nor overtly expressed. External resources are necessary when a writing system fails to reflect the prosodic features. For instance, Chinese characters do not directly reflect tones, and transcriptions in other languages with overt-marking stress would be helpful. Ting (1975) explores Middle Chinese tones, which have four categories, by using Chinese transcriptions of Sanskrit. In comparing Sanskrit stress and Middle Chinese tones, Ting (1975) suggests that the four tones can be classified into one level tone and three non-level tones: high rising, mid falling and short checked. Ting's (1975) method gives an insight into how it becomes possible to decipher prosody by investigating transcriptions which preserve the phonological details of source and target languages and serve as a useful tool to reconstruct the phonological inventories of both. This paper investigates whether prosody is hidden behind transcriptions in historical sources and focuses on the realization of Chinese tones during the 15 th century. This paper follows Ting's (1975) method and uses Chinese transcriptions of foreign languages as key materials. Three historical sources were specifically selected, in which Japanese, Korean and Ryukyuan are recorded in the style of transcriptions, by using Chinese characters. Published during the Ming Dynasty (1368 -1644 AD), the three historical sources are Rìběnguănyìyŭ 'A wordlist of Chinese-Japanese phrases', Cháoxiānguănyìyŭ 'A wordlist of Chinese-Korean phrases', and Liúqiúguănyìyŭ 'A wordlist of Chinese-Ryukyuan phrases'. 1 The three wordlists follow the same format. Each entry is initially provided with a Chinese meaning, and the transcription is presented afterward. Take the word 'wind' as an example: in Rìběnguănyìyŭ, the meaning is provided in Chinese as 風 'wind', followed by the transcriptions in Chinese characters as 刊節 [k h an.tsiɛɁ] for Japanese. 2 Likewise, the word 'wind' is transcribed as 把論 [pa.lun] for Korean in Cháoxiānguănyìyŭ, and as 嗑集 [kaɁ.tsiɁ] for Ryukyuan in Liúqiúguănyìyŭ. The three wordlists have been thoroughly investigated; Ding (2008) and Fukushima (2008) have studied Rìběnguănyìyŭ, Moon (1972), Kang (1995) and Kwon (1998) have studied Cháoxiānguănyìyŭ, andDing (2008) has studied Liúqiúguănyìyŭ. Only Kwon (1998) touches upon prosody, discussing the correspondences between Chinese and Korean tones. 3 There is no detailed research that thoroughly utilizes Rìběnguănyìyŭ or Liúqiúguănyìyŭ to discuss Chinese tones or Japanese/Ryukyuan pitch accent during the 15 th century. Moreover, there is no cross-linguistic study to explore how prosody is preserved in transcriptions.
This paper deals with the correspondences between Chinese tones and three foreign languages during the 15 th century, considering Japanese and Ryukyuan pitch accent and Korean tones. I discuss two issues. The first is the realization of Chinese tones in the transcriptions in terms of prosodic differences between Japanese, Ryukyuan, and Korean, and the second is the projection of prosodic characteristics from Chinese onto the three foreign languages in the transcriptions. The data is disyllabic nouns from the three wordlists. To understand the prosodic systems of Chinese, Japanese, Korean and Ryukyuan, section 2 discusses Chinese and Korean tones and Japanese and Ryukyuan pitch accent. Section 3 deals with data selection criteria, transcriptions, and basic assumptions. Section 4 reports on the distributions of the correspondences, and section 5 focuses on different strategies that reflect the prosodic differences in the source and target languages. Section 6 concludes with a summary of major findings.

The prosody of Chinese, Japanese, Korean and Ryukyuan
This section describes the prosodic systems of Chinese, Japanese, Korean and Ryukyuan. Chinese tones are first explained in section 2.1, focusing on tonal categories in the 15 th century. Then, Japanese and Ryukyuan pitch accent, as recorded in historical sources, is introduced in sections 2.2 and 2.3, respectively. Korean tones are elaborated upon in section 2.4.

Chinese tones in the 15 th century
In Chinese historical phonology, there are four tonal categories: píng 'even ', shăng 'rising', qù 'departing' and rù 'entering'. 4 The four categories go hand in hand with the voicing of onsets, known as yīn 'voiceless' and yáng 'voiced', which cause the four tones to split and render eight tones. Table 1   Although there are eight tones in Table 1, Early Mandarin Chinese had less than eight tones. Yǜnluèyìtōng (1442 AD) (henceforth YLYT), a representative rime book from the 15 th century, has only five tones: yīnpíng, yángpíng, shăngshēng, qùshēng, and rùshēng. Píngshēng is divided into two tones according to the voicing of onsets. Although rùshēng gradually dispersed into the other three tones before the 15 th century, rùshēng is still a category in YLYT. 5 As the three historical sources were published in the 15 th century, this paper relies on YLYT as a reference point. According to Lin (2010: 149), the tones in YLYT resemble those in Zhōngyuányīnyùn (1324 AD). Zhang (2010) posits that yīnpíng is a non-high level tone 33, yángpíng a rising tone 35, shăngshēng a falling-rising tone 214, qùshēng a high falling tone 51. 6 On the basis of Zhang's (2010) reconstruction for Zhōngyuányīnyùn, 4 Rùshēng is not only a tone but also a syllable type, which refers to syllables ending in -p, -t, -k or -ʔ. 5 There was an important tonal change in the 14 th century. Rùshēng gradually merged into the other three tones. The process in the 14 th century has been captured in Zhōngyuányīnyùn (1324 AD), and therefore there are different interpretations of rùshēng (Ning 1985, Yang [From Lin 1992, Geng 2009 andZhang 2010). Ning (1985) reconstructs it as a high level tone, while the other three reconstructions are rising tones. Ning (1985) provides no reconstruction of rùshēng. Yang (from Lin 1992) suggests that rùshēng merged with the other four tones and become a high level tone, a low rising tone, and a high falling tone. Geng (2009) reconstructs rùshēng as a low falling tone. Zhang (2010) suggests a phonological environment for rùshēng: it is a low rising tone when the onset is voiceless. 6 In Chinese phonology, tones are usually presented by numbers, 1-5, a system developed by Chao (1930). The highest pitch is 5, and the lowest is 1. For example, a high level tone is 55, and a falling-rising tone is 214. the tones in YLYT are as follows: yīnpíng and shăngshēng are non-high tones; the former is a low level tone and the latter is a low falling-rising tone, yángpíng is a high rising tone, and qùshēng is a high falling tone. As for rùshēng, although it is a phonologically distinct category in YLYT, Zhang (1999: 40) points out that this tone had dispersed into the other tones. Following Zhang (1999), rùshēng is not an independent category in this paper. It is assigned to the other three tones.

Japanese pitch accent in historical sources
Unlike in modern Japanese phonology, Japanese prosody in historical sources is presented according to the conventions of Chinese phonology, as marked using dots for tonal categories. The dots are placed around a Japanese kana. 7 According to Akinaga, Ueno, Sakamoto, Sato & Suzuki (1998: 35) and Okimori (2010: 132), there are six tones in Middle Japanese: heisei or hyōshō 'even tone', jōsei or jōshō 'rising tone', kyosei or kyoshō 'departing tone', nyūsei or nisshō 'entering tone', heisei karu or hyosho karu 'light even tone', and nyūsei karu or nisshō karu 'light entering tone'. The six tones are surmised as a low level tone for heisei, a high level tone for jōsei, a rising tone for kyosei, a low entering tone for nyūsei, a falling tone for heisei karu and a high entering tone for nyūsei karu.
According to Mabuchi (1971: 157) and Kindaichi (2001: 306), Table 2 exemplifies the accent patterns in monosyllabic nouns. 8 The four accent patterns are high (1.1), falling (1.2), low (1.3) and rising (1.4). The patterns can be monosyllabic, H or L, or bimoraic, HL (=F) or LH (=R).   The five patterns are high + high (2.1), high + low (2.2), low + low (2.3), low + high (2.4) and low + falling (2.5). The first four patterns are the combinations of two level tones. The fifth pattern consists of a falling tone in the second syllable of a disyllabic noun. There is no rising tone for disyllabic nouns.

Reconstructed Ryukyuan pitch accent
No historical source explicitly records accent patterns in Ryukyuan. The most complete work concerned with early Ryukyuan pitch accent is Shimabukuro's (2007) reconstruction, which is based on modern languages. According to Shimabukuro (2007: 331-332), Table 4 shows the accent patterns in monosyllabic nouns.

Middle Korean tones
Middle Korean also adopts the tradition of Chinese phonology for tones: in this case consisting of pyeongseong 'even tone', sangseong 'rising tone', keoseong 'departing tone' and ipseong 'entering tone'. In Hunminjeongeum 'The Correct Sounds to Educate the People' (1446 AD), tones are presented by bangceom 'side dots', which are placed on the left side of the syllable. No side dot is used for pyeongseong. Keoseong is marked by a dot and sangseong by two dots. Ipseong is also treated as keoseong with one dot, but it is phonetically short and abrupt. Ipseong is usually not included in the discussion of Middle Korean tones because there is no clear pitch change. Table 6 illustrates the three tones in monosyllabic nouns.  According to Lee (1961: 134), Kim (1973), Oh (1988: 262), and Lee (2007), the tonal values of Korean tones are that pyeongseong is a low tone (L), keoseong is a high tone (H), and sangseong is a rising tone (R ~ LH), which is long in duration. There should be nine possible combinations for disyllabic nouns (3x3). Nevertheless, Lee (2007: 53) and Lee and Ramsey (2011: 163) suggest three major patterns, as shown in Table 7. 10 The three patterns are combinations of low and high tones: low + low, low + high and high + high (or low). Lee and Ramsey (2011: 168) suggest that the low-high pattern is the most pervasive in the distribution, more than three times as common as low-low, and more than five times as common as high-(high). There is no information about the high-low pattern. 11

Data selection criteria, transcriptions, and basic assumptions
This section introduces how data was sorted and transcribed (section 3.1), and then discusses the basic assumptions (section 3.2). The three historical sources used in this paper are Cháoxiānguănyìyŭ (henceforth CXGYY), Rìběnguănyìyŭ (henceforth RBGYY), and Liúqiúguănyìyŭ (henceforth LQGYY). In particular, I adopt Ding's (2008) versions of RBGYY and LQGYY, and Kang's (1995) version of CXGYY. In the three sources, there are 566 entries in RBGYY, 595 entries in LQGYY, and 596 entries in CXGYY. 12

Data selection criteria and transcriptions
I follow the data selection criteria in Lin (2013). As entries in the three historical sources are transcribed by Chinese characters, sorting instances is based on syllables rather than moras, even though the mora is a crucial prosodic feature in Japanese and Ryukyuan. In this paper, only disyllabic words that authentically contain two syllables are selected. Entries that have two moras in Japanese/Ryukyuan but correspond to 11 Lee (2007: 55) contends that the HL pattern is often found in the phonetic representation, derived by tone rules. 12 The three historical documents used in this paper are from Huáyíyìyŭ 'Transcriptions of Chinese and foreign languages', a collection including at least ten languages compiled during the 15 th century by the Ming Dynasty government. Although the three books did not provide the names of the compilers, it is assumed that the compilers were professionally trained, due to the fact that Huáyíyìyŭ was an official document which aims to provide transcriptions of foreign languages for Chinese diplomats and ambassadors.
one Chinese phonogram are excluded from the dataset. For example, the word 'three' in Japanese is saN with two moras, but it is transcribed by one Chinese phonogram 散 san in RBGYY. Japanese moraic nasal N in saN corresponds to Chinese coda nasal n in san. The word 'three' is not included in the corpus.
In CXGYY, one Korean syllable corresponds to one Chinese character. The Korean lateral coda -l, however, is represented by an independent character because Chinese lacks lateral coda -l. For instance, the word 'sky/heaven' in Middle Korean is [ha‧ nʌl]. This disyllabic word corresponds to three Chinese phonograms 哈嫩二 [ha.nun.ʐʅ]. The first two phonograms, 哈嫩 ha.nun, correspond to ha.nʌ (tones are omitted here). The lateral coda -l in this word is consistently transcribed by the third phonogram 二 [ʐʅ]. Although this paper explores disyllabic words, entries with lateral codas in Korean are also included in the dataset. The third phonogram for the lateral coda -l is not calculated. Therefore, the Korean word 'sky/heaven' in CXGYY (哈嫩二) is included in the dataset.
Not all of the collected disyllabic words in the three historical sources are used. Following Lin's (2013) second data selection criterion, I specifically analyzed disyllabic nouns. Unlike monosyllabic nouns, disyllabic nouns show more prosodic patterns. In addition, disyllabic nouns show more patterns of prosody than verbs, adverbs, and adjectives. Thus, using disyllabic nouns helps us understand the hidden prosodic patterns of Japanese, Korean and Ryukyuan behind the transcriptions in the three historical sources. Lacking sufficient trisyllabic nouns in the three historical sources makes it impossible to make significant comparisons.
According to the above criteria, 137 disyllabic nouns were collected from RBGYY, 115 from CXGYY and 101 from LQGYY. Table 8 illustrates the next step in the transcription of the nouns' prosody.  In the dataset, I list the original phonograms of each entry and the transcriptions based on YLYT and then give the meaning of each entry in the next row. In Table 8, there are original tones and converted tones. The original tones refer to the tonal categories of the phonograms in YLYT. For example, the word 'river' in Japanese is transcribed by 嗑哇 [kaʔ.wa], whose tones in YLYT are rùshēng + yīnpíng. Next, all the original tones are converted into high and low to accommodate Chinese tones with Japanese, Korean and Ryukyuan prosody. As discussed in section 2.1, Chinese yángpíng and qùshēng are converted into high tones, and yīnpíng and shăngshēng into low tones. The conversion of rùshēng depends on the tones they change into (Lin 2018

Basic assumptions
How Chinese tones are used to present other languages is seldom discussed in the literature. Table 9 shows a general comparison of the prosody in the three languages.  As Japanese and Ryukyuan are sister languages, the two are predicted to bear higher phonological affinity, and be prosodically distinct from Korean, which was a tone language in the 15 th century. If prosody is not neglected in the Chinese transcriptions, the corpus data should show that Japanese and Ryukyuan do not statistically differ from each other in prosody, while Korean should be statistically different from Japanese and Ryukyuan.
Furthermore, Lee and Ramsey (2011: 163) suggest that the first syllable in a disyllabic noun in Korean tends to behave like a rising tone. Nevertheless, the transition from the first to the second syllable in Japanese and Ryukyuan disyllabic nouns does not necessarily render a rising tone. Again, if prosody is hidden behind the Chinese transcriptions, Korean should show a significant difference from Japanese and Ryukyuan in the first syllable. Chinese characters with rising tones should be more frequently used to transcribe Korean than to transcribe Japanese and Ryukyuan. Although Chinese shăngshēng and yángpíng in YLYT are rising tones, this paper suggests that Chinese shăngshēng serves better than yángpíng, which is a high rising tone, for the transcriptions.

Cross-linguistic comparisons
In this section, cross-linguistic comparisons are made to show how Chinese tones are reflected in disyllabic nouns in the transcriptions. All of the sorted data,, including 137 entries in Japanese, 115 entries in Korean and 101 entries in Ryukyuan are used for cross-linguistic comparisons.   There are three impressionistic observations regarding the prosodic differences in the three historical sources. The transcriptions of Ryukyuan are more evenly distributed than those of Japanese and Korean in tones. 14 There are two peaks in Table 10: the set C+D in Korean (26 tokens), and the set D+D in Japanese (22 tokens). In Ryukyuan, no set of tones exceeds 20 tokens.
Next, the three most frequently-used sets of Chinese tones for the transcriptions in each language are analyzed. They are D+D (22 tokens), A+D (15) and B+D (14) for Japanese, C+D (26), B+D (15) and D+D (15) for Korean, and D+A (12), A+A (11) and A+D (11) for Ryukyuan. The top three sets in Japanese and Korean resemble each other in the second syllable which tends to be a falling tone (D = qùshēng), while this tendency is not observed in Ryukyuan. This phenomenon is discussed in detail below.
From the three most frequently-used sets, there is a prosodic difference between Korean and Japanese/Ryukyuan. The first syllable in the transcriptions for Korean tends to be Chinese shăngshēng, a low falling-rising tone. In Japanese and Ryukyuan, shăngshēng is not frequently used in the first syllable. This phenomenon accords with the marking of tone in the examples by Lee & Ramsey (2011: 163) which indicates that the first syllable of LH and HH (or HL) tends to behave phonetically, like a rising tone.
The observations based on Table 10 are insufficient to provide convincing details for the prosodic differences between the three languages in the transcriptions. A chi-square test was conducted to compare the three historical sources. In Table 10, due to low frequency (< 5 tokens), cells have to be merged. The merged cells are presented in Tables 11 and 12, focusing on different syllables.  Given that α = .01 and d.f. = 6, the result in Table 11 is significant (χ 2 (obtained) = 31.43 > χ 2 (critical) = 16.81, p < .01). In other words, there are significant differences in the transcriptions of the first syllables among the three historical sources. Table 12 shows the distribution in the three historical sources, focusing on the second syllable.  Under the same condition, where α = .01 and d.f. = 6, the result in Table 12 is also significant (χ 2 (obtained) = 39.46 > χ 2 (critical) = 16.81, p < .01). Again, there are significant differences in the transcriptions of the second syllables among the three historical sources. 16 The results in the above two tables indicate that the transcriptions are significantly different in the three historical sources. It becomes apparent that prosody is not completely ignored in transcriptions, so the prosodic differences are reflected by the chosen phonograms. Section 5 discusses different strategies for transcribing foreign languages by using Chinese characters in the three historical sources.

Different strategies for transcriptions
The results in Section 4 suggest that there are significant differences between the prosodic systems of Japanese, Korean, and Ryukyuan in the transcriptions; however, more details are needed to make clearer interpretations. Thus, post hoc tests were conducted.
Since the results in Tables 11 and 12 are significant, the tokens in Japanese, Korean and Ryukyuan were paired for post hoc tests (α = .01, d.f. = 3). Pairs that show significant differences in the post hoc tests are listed in Table 13.

Sets
C+x C+x A+x x+A x+A X+D There are six sets in Table 13: C+x in the Japanese-Korean pair, A+x, C+x in the Korean-Ryukyuan pair, x+A in the Japanese-Korean pair, and x+A, x+D in the Korean-Ryukyuan pair. In Table 13, all the significant differences are found between the pairs of Korean and Japanese and those of Korean and Ryukyuan. There is no significant difference between the pairs of Japanese and Ryukyuan. The results conform to the assumption that Japanese and Ryukyuan are not prosodically different from each other.
The six sets in Table 13 suggest three aspects of how Chinese tones are used in the three historical sources. First, Korean significantly differs from Japanese and Ryukyuan in the two sets, C+x, and x+A. Chinese shăngshēng is preferred to be in the first syllable of a disyllabic noun for Korean disyllabic nouns. On the other hand, in the second syllable, Chinese yīnpíng is more frequently used in Japanese and Ryukyuan than in Korean. The significant difference between Korean and Japanese/Ryukyuan in the set C+x is strong evidence for hidden prosody in transcriptions. As Chinese shăngshēng in YLYT is a low falling-rising tone, there must be a transition from low to high in the set C+x. The fact that the transcriptions favor C+x accords with the interpretation of the examples by Lee & Ramsey (2011: 163). The first syllable of LH tends to be like a rising tone. 17 It is evident that the compilers of CXGYY indeed perceived pitch changes from the first syllable to the second syllable and therefore translated the transition into Chinese shăngshēng, a low falling-rising tone, in the transcriptions. The significant difference in the set C+x also provides insight into the prosodic systems of Japanese and Ryukyuan. In Tables 11 and 12, the set C+x is the least favorable set for Japanese and Ryukyuan. As I have discussed in section 3, Japanese and Ryukyuan pitch accent does not render a perceptually salient contour from low to high as Chinese shăngshēng does. Thus,in RBGYY and LQGYY,17 It should be clarified that the contour can correspond to Chinese shăngshēng 214 or yángpíng 35. Although Chinese shăngshēng 214 is a falling-rising tone, it is assumed to be [+low]. Chinese yángpíng 35 is a rising tone, and it is often considered [+high]. The fact that Chinese shăngshēng 214 is lower than Chinese yángpíng 35 makes it more compatible with the first syllable of Korean nouns, which tends to be low.
Chinese shăngshēng is not frequently used in the first syllable of a disyllabic noun.
The other significant difference between Korean and Japanese/ Ryukyuan is in the set x+A. Chinese yīnpíng is disfavored in Korean more than in Japanese and Ryukyuan for the second syllable in a disyllabic noun. 18 In section 2, I have suggested that Chinese yīnpíng is a low level tone in YLYT. The low percentage of x+A in Korean (< 8%) is correlated with C+x. 19 As discussed above, the pitch change from the first syllable to the second syllable is pervasive in Korean, and a rising tone is often perceived. The pitch change from low to high not only leads to a rising tone in the first syllable but also results in saliency in the second syllable, which is perceptually higher than the first one. Therefore, the low pitch becomes contradictory in the second syllable. This is why the set x+A is not frequently utilized in the transcriptions. 20 In addition to the same differences between Korean and Japanese/ Ryukyuan in the sets C+x and x+A, there are two other significant differences in the sets A+x and x+D in the Korean-Ryukyuan pairs. In the set A+x, Chinese yīnpíng, a low level tone, is disfavored in the first syllable of a disyllabic noun in Korean (16% < 38%). 21 This could be due to the transition of pitch in a disyllabic noun in Korean. The low level tone in 18 Similar results are also reported by Sarashina (2005Sarashina ( , 2013. 19 The percentages of Tables 10 and 11 are shown in (i) and (ii) Tables 11 and 12 should also be shortly discussed, although the differences are not significant. Chinese qùshēng, a falling tone, outnumbers the other three tones in the second syllable in the three historical sources.  Given that α = .01 and d.f. = 3, the result in Table 14 is significant (χ 2 (obtained) = 19.4 > χ 2 (critical) = 11.35, p < .01). This tendency, however, is not attributed to Japanese, Korean and Ryukyuan prosody, but reflects Chinese prosody. When Japanese and Ryukyuan are transcribed, it is not problematic to use a falling tone in the second syllable, since the two languages allow patterns like LF. In Korean, the lack of a falling tone should result in the low frequency of Chinese qùshēng in the second syllable. Nevertheless, the falling tone still outnumbers the other three tones in Korean (68 tokens). A question immediately arises as to why Chinese qùshēng is largely used in the second syllable of a disyllabic noun. This is likely due to a particular prosodic characteristic of Chinese tones. As suggested by Wang (2011: 133-134) regarding the synchronic behavior of Mandarin Chinese and Ting (1969Ting ( , 1975 in terms of the historical development of Chinese tones, the second syllable of a disyllabic word tends to be a falling tone. Therefore, the tendency in which the second syllable is preferred to be a falling tone in the transcriptions in the three historical sources is motivated by the projection of the prosodic characteristic from Chinese onto Japanese, Korean and Ryukyuan.
To sum up these observations, different strategies of transcription are used in the three historical sources. The results of the chi-square tests seem to suggest that prosody is not omitted in the transcriptions. A distinction is clearly made between Korean and Japanese/Ryukyuan in the sets C+x and x+A. Chinese shăngshēng is used to present pitch change from low to high in Korean. This transition is not prominent in Japanese and Ryukyuan. In addition, Korean disfavors a low level tone in the second syllable of a disyllabic noun. Two minor differences have also been discussed; Korean, compared with Ryukyuan, disfavors a low level tone in the first syllable. The x+D set reflects a prosodic characteristic in Chinese. The second syllable of a disyllabic noun tends to be a falling tone.

Conclusion
This paper has explored prosody hidden in historical sources. Three historical sources in Japanese, Korean and Ryukyuan, transcribed using Chinese characters, were investigated, and specific attention was paid to the prosody of disyllabic nouns. The sorted data in the three historical sources were cross-linguistically compared. The results are promising. It was found that prosody is not completely omitted, but actually preserved in the three historical sources, indirectly hidden within the transcriptions.
The results of cross-linguistic comparisons also show significant differences. Korean in the 15 th century is a tone language, while Japanese and Ryukyuan were (and still are) pitch accent languages. The Chinese transcriptions reveal that Korean significantly differs from Japanese and Ryukyuan in prosody. To find more details, post hoc tests were conducted to make the interpretation easier. Korean differs prosodically from Japanese and Ryukyuan in the first syllable of a disyllabic noun. The pitch changes from low to high are salient in Korean, and the transition is often perceived and then translated into Chinese shăngshēng. In Japanese and Ryukyuan, the pitch change does not necessarily render a rising tone in the transcriptions. In addition to prosodic differences between Japanese, Korean and Ryukyuan, it is also crucial to point out that one prosodic characteristic of Chinese is projected onto the transcriptions, as shown in the second syllable of a disyllabic noun, in which a falling tone is obviously favored.
In observing the study in full, it is exciting to find that prosody is not completely neglected in philological data and that there are crosslinguistic prosodic differences in the textual record. In the use of historical sources, transcriptions are of great help not only for reconstructing segments for both source and target languages but also for unveiling hidden prosody. It is hoped that this paper will contribute to studies based on philological evidence in historical linguistics, despite the fact that, as Minkova (2013: 260) states, it is notoriously difficult to recover historical prosodic patterns.

Comments invited
PiHPh relies on post-publication review of the papers that it publishes. If you have any comments on this piece, please add them to its comments site. You are encouraged to consult this site after reading the paper, as there may be comments from other readers there, and replies from the author. This paper's site is here: