Reconstruction of Ryukyuan tone classes of Middle Japanese Class 2.4 and 2.5 nouns

: It is generally accepted that of the ﬁ ve tone classes reconstructed for disyllabic nouns in Middle Japanese ( MJ ) , Classes 2.4 and 2.5 exhibit a split correspondence with proto - Ryukyuan ( pR ) tone Classes B and C. This split correspondence is of tremendous importance for the reconstruction of the proto - Japonic ( pJ ) tone system, because, in the absence of a conditioning factor accounting for the split, it has led to the reconstruction of additional tone classes at the pJ level. However, of the approximately 100 nouns belonging to these classes, the tone class of only half of them has been reconstructed for pR. Before embarking on the reconstruction of the pJ tone system, we must therefore ﬁ rst reconstruct the pR tone class of the nouns belonging to MJ Classes 2.4 and 2.5. This study provides a reconstruction of the tone classes in pR for 75 cognates using the latest comparative data of Northern and Southern Ryukyuan dialects. The results con ﬁ rmed the existence of a split correspondence, where Class 2.4 and 2.5 nouns are found to be roughly split in half between pR Classes B and C, demonstrating that it is not a merely sporadic irregularity.

that the split correspondence is genuine, and it reexamines the hypothesis that views it as a result of a conditioned merger. Section 5 concludes this study.
It is also clear from the discussion above that the previous literature has been based primarily on Northern Ryukyuan. In some respect, it was inevitable that Southern Ryukyuan was neglected. First, except for the Yonaguni dialect (Hirayama and Nakamoto 1964), dialects of Southern Ryukyuan distinguishing the three pR tone classes were not discovered until Matsumori's (2010) study of the Tarama dialect. Second, although many Southern Ryukyuan dialects retaining the three-way tonal contrast of pR were discovered one after another in the 2010s (Igarashi et al. 2011, Igarashi et al., 2012, Matsumori 2012, at the same time, the prosodic systems of these dialects turned out to be extremely complex. Indeed, because of extensive tonal neutralization processes at work in these dialects, the observation of words pronounced in isolation, or even embedded in short carrier sentences, proved completely insufficient for the correct identification of the tone classes to which each word belongs (Igarashi 2016a, Matsumori 2016. As a result, a tremendous amount of elicitation is necessary to determine accurately the tone class of each word in these dialects. Fortunately, extensive data on tone class membership have recently been made available for at least two Southern Ryukyuan dialects that preserve the three tone classes of pR; namely, the Ikema dialect of Miyako (Igarashi 2016b) and especially the Tarama dialect of Miyako (Tokuyama and Celik 2020). Although extensive data on tone classes membership in Yaeyama have also become available recently (e.g., Ishigaki [Miyagi 2003], Taketomi [Maeara et al. 2011], and Hatoma [Kajiku 2020]), pR Classes B and C have merged in most Yaeyama dialects, making it impossible to use data for the reconstruction of pR tone classes. This study therefore exploits data from Ikema, Tarama, and Yonaguni as primary data for Southern Ryukyuan to establish pR tone classes from the Northern and Southern Ryukyuan perspectives.

Data
We exploit two types of resources for reconstructing pR tone classes, primary and secondary data. The primary data are used for the reconstruction of pR tone classes. They consist of the information regarding the tone class membership of eleven dialects, which all preserve the distinction between pR Classes B and C (Table 3).  (2020), the author's field notes Ikema (Nishihara variety) The author's field notes Yonaguni Yonaguni Uwano (2010Uwano ( , 2013 The Ashikebu and Nakasato dialects merged Classes A and B. The Ie dialect merged Classes A and C in bimoraic nouns. The Ikema dialect preserves the three-way contrast in pR, but Class A has almost completely merged with Class B for bimoraic nouns (Igarashi 2016b). The other dialects far more faithfully preserve the three-way contrast in pR. The secondary data are used to examine whether a given cognate is widely observed in Ryukyuan dialects and to confirm the correspondences in vowels and consonants between dialects. They consist of seven dialects that merged pR Classes B and C or that do not have tonal contrasts at all (Table 4). MJ classes are based on Kindaichi (1974), which lists a total of 108 Class 2.4-2.5 nouns. The entries of his vocabulary are written in italics in the MJ forms (Akinaga et al. 1997).
Word forms are provided in broad phonetic transcription. The diacritic "′" indicates a glottalized consonant. Long vowels are represented by a series of identical vowels. "ï," "ɨ," and "ɿ" are uniformly written as "ɨ." The pairs "ʃ" and "ɕ" and "ʒ" and "ʑ" are uniformly represented as "ɕ" and "ʑ," respectively. Moreover, the diacritics " ┌ " and " ┐ " designate a pitch rise and fall, respectively. For example, "μ ┌ μ" ("μ" stands for mora) represents an LH pitch pattern ("L" and "H" stand for low and high pitch, respectively). "μ ┌ μ ┐ " is distinguished from "μ ┌ μ"; when produced in isolation, the two are (nearly) indistinguishable, but when immediately followed by another morpheme such as a case particle, the former is realized as LH-L, whereas the latter is realized as LH-H. In Tarama and Ikema, tonal neutralization occurs with nouns produced in isolation. Tone opposition can be observed, for example, by adding one or more bimoraic particles (one for Tarama and two for Ikema) after the noun. The -kara 'from' and -mai 'also' in the tables are bimoraic particles.
A superscript at the end of each word form indicates the correspondence with the pR tone class, where "A," "B," and "C" are the pitch patterns corresponding to the pR Class A, Class B, and Class C, respectively. Because of the conditioned merger that occurred in each dialect, one pitch pattern may correspond to more than one class. In that case, more than one letter is used to indicate the correspondence. For example, "AC" indicates that the pitch pattern corresponds to both pR Classes A and C, and "BC" indicates that the pitch pattern corresponds to both pR Classes B and C.
Because we use multiple sources for Tarama, there can be mismatches in the pitch pattern between the sources. In case of discrepancy, the pattern in the author's field note is adopted first, with that in Tokuyama and Celik (2020) in parentheses; for example, " (∼A) ."

Reconstruction
We postulate the phoneme inventory in Figure 1 for pR. We assume that an intervocalic *w occurred only before *a, an intervocalic *j was allowed only before a back vowel, and *d had already merged with *z before a high vowel. No obligatory glide insertion is assumed between vowels. Thus, the pR noun for "voice," for example, is reconstructed as *koe instead of *kowe or *koje. We leave aside whether *t, *d, *s, and *z were palatalized before a high vowel. We use a capital *U in cases where we could not decide whether *o or *u was to be reconstructed. Examining the segmental correspondence between dialects is  Miyanaga (1930) important for determining whether nouns that are similar in both meaning and form are cognate with each other. The determination of cognacy constitutes an indispensable part of the reconstruction of the pR tone class of the nouns. Furthermore, the correct reconstruction of word-final vowels in pR is crucial for testing the conditioned merger hypothesis. We discuss the issues concerning the segmental reconstruction and cognacy between nouns only if they require argumentation. When the tonal correspondence between the dialects is irregular, we adopt the principle of parsimony, according to which the reconstruction should require the fewest sound changes. The parsimony is assessed on the basis of the phylogenetic tree proposed by Pellard (2015) (Figure 2). For example, pR *poka 'outside' exhibits irregular tonal correspondence (Ashikebu ɸuka AB , Asama ɸukaa B , Wadomari ɸaa B , Yoron huka C , Ie ɸuk'aa B , Nakijin ɸuk'aa B , Shuri ɸuka B , Tarama puka C , Ikema huka C , Yonaguni huga B ). If we posit pR Class B, then the sporadic sound change B > C is reconstructed to occur two times, each for Yoron and proto-Miyako. By contrast, if we posit pR Class C, then the sporadic sound change C > B is reconstructed to occur five times, each for Ashikebu, Asama, Wadomari, proto-Okinawa, and Yonaguni. Thus, the principle of parsimony judges Class B as more plausible than Class C.

Confirmation of tone classes reconstructed by previous studies
The survey confirmed that the tone classes reconstructed by previous studies were valid for 52 of the 55 words ( Table 5). In the table, nouns that show an irregular tonal correspondence are shaded. In addition, segments that obviously show an irregular correspondence are underlined. The exceptional correspondence exhibited by quite a few dialects in nouns where cognates are found in most dialects can be regarded as the result of sporadic changes in individual dialects and does not affect the reconstruction of tone classes based on the principle of parsimony. Sporadic changes of this sort are not discussed in this section. Some nouns require further commentary, which is provided below.    Pellard (2015). The phylogenetic position of Hachijo is controversial, which is indicated by a dotted line. Ashikebu Tarama pa⸣ɭ C -mai pa⸣sɨ C -mai pi⸣ɾa C -mai fu⸣ni C -mai Ikema hai C -kaɾa⸣-mai hasɨ C -kaɾa⸣-mai hiɾa C -kaɾa⸣-mai funi C -kaɾa⸣-mai Yo. Yonaguni Tarama mu⸣ku C -mai na⸣bi C -mai ɕa⸣ɭ C -mai taa⸣bɨ C -mai Ikema muku C -kaɾa⸣-mai nabi C -kaɾa⸣-mai sai C -kaɾa⸣-mai tabi⸣ B -kaɾa-⸢mai Yo.
Yonaguni mu⸢ɡu⸣ C na⸢bi⸣ C sa⸢ɾu⸣ C ta⸢bi⸣ C Notes Reconstruction of Ryukyuan tone classes  241 Tarama 13. *sora C 'treetop': Matsumori (2009Matsumori ( , 2012 also reconstructs Class C for *sora 'treetop,' although she may not agree with the view that pR *sora C 'treetop' is related to MJ sora 2.4 'sky.' In the tenth century, Japanese sora referring to not only 'sky' but also 'the upper part of an object (such as roof, ceiling, and treetop)' is attested (NKDJ 2001, 507). This means that at the stage of MJ at the latest, Japanese sora also had a meaning equivalent to that of pR. This study does not examine the question of whether the pJ cognate had the meaning of 'the upper part of an object.' What is clear from the above discussion is that the semantic change between 'sky' and 'the upper part of an object' is a natural process, and that pR *sora 'treetop' is reasonably regarded as the cognate of MJ sora 2.4 'sky.' Note that pR *sora does not have a meaning 'sky,' which is indicated instead by a Chinese loan pR *ten A.
The (alveolo-)palatal consonant and front vowel in 'sour' indicate that the pR 'sour' has a segment with a feature [-back], such as *i and *j. In Yoron, although ɕii-'sour' has the same segmental makeup as ɕii B 'nest,' it is exceptional in that its adjective ending takes -ɕan rather than -san. The ending -ɕan is regarded to be attached to roots that end in *si (so-called shiku-katsuyō, *suju-'sour' are also reflected as tsɨv and sɨv-. Thus, the change pR *ju > v/*{s, t}u_ can be reconstructed in proto-Miyako.⁷ 26. *woke C 'bucket': The correspondence of the word-initial consonant is irregular in Nakijin, Tarama, and Yonaguni, suggesting that sporadic sound changes occurred in these dialects. These changes, though irregular, are phonetically motivated at least for Nakijin and Tarama; in Nakijin, w became ɸ by assimilating its voicing to that of the consonant in the following syllable, and in Tarama, *b (<*w through a regular sound change) became ɡ by assimilating its place of articulation to the consonant in the following syllable. The tonal correspondences are regular throughout all dialects. These facts indicate that these nouns are inherited from pR, allowing us to reconstruct pR *woke C 'bucket.' 32. *kama B 'sickle': Although Matsumori (2012) reconstructs Class B, only Amami exhibits pitch patterns corresponding to pR Class B according to our survey. The pitch patterns in Okinawa, in contrast, correspond to Class A. The cognates are perhaps absent in Southern Ryukyuan, where reflexes of pR *irana C 'sickle' are attested instead (Tarama ɨzara C , Ikema zzara C , Ishigaki ʔiɾana BC , Yonaguni irara C ). They are also attested in Okinawa (Nakijin ʔinaana C , Shuri ʔirana C ). Regardless of whether the pR Class of *kama is reconstructed as A or B, the number of reconstructed changes remains the same: the change A > B in proto-Amami or the change B > A in proto-Okinawa. Because Class B or C is a regular correspondence with MJ Class 2.4, it is more plausible to reconstruct pR Class B for *kama 'sickle' and a subsequent sporadic change B > A in proto-Okinawa.
36. *kinu B 'garment': The noun listed as a Class 2.4 noun in Kindaichi (1974) is not kinu 'garment' but kinu 'silk.' They both are also attested in OJ, the oldest attested stage of Japanese, and are generally recognized as cognates (Jodaigo 1967, Martin 1987. Their pitch patterns are both LH in MJ (Akinaga et al. 1997), a regular reflex of Class 2.4 in MJ. Additionally, there are modern Japanese dialects with kinu 'garment' also corresponding to Class 2.4 (Hirayama et al. eds. 1992-93). Reflexes of pR *kinu B 'garment' are observed throughout Ryukyu Islands. However, only a few dialects have nouns for 'silk,' in which the  e.g., pR *otorosi-'horrible' > Yoron uturu-ɕan 'id.') (Kiku and Takahashi 2005, 224); however, in ɕii-ɕan 'sour,' its root does not end in *si. The palatalization of the onset consonant of -san should not be the result of the progressive assimilation caused by the synchronic root-final vowel i because the root-final i derived from pR *u does not cause palatalization as in atɕi-san 'hot' (<pR *atu-'id.'). It follows that pR 'sour,' just as pR *otorosi-'horrible,' had [-back] in the root-final syllable in pR, which palatalized the onset consonant of an adjective ending. Therefore, the pR 'sour' cannot be *su-. In Japanese dialects, both suand forms with a feature [-back] such as suju-, sui-, and ɕiiare indeed observed as an adjective 'sour,' as expected from Ryukyuan cognates. A rough representation of the distribution of the forms for 'sour' is as follows: suju-, sui-, and ɕiiin Kyushu, suiin Chugoku and Shikoku Districts, suin Kansai and Chubu Districts, suppain Kanto District, and sukkain Tohoku Districts (LAJ 1966-1974. The three forms suju-, sui-, and ɕiican be regarded as reflexes of pJ *suju-'sour.' The fact that the three forms are geographically distributed in areas close to the Ryukyus also supports pR *suju-'sour.' 7 A reviewer suggests that Tarama tsɨv C 'dew' is the result of a contamination with pR *tubu C 'grain, droplet' rather than being a cognate with pR *tuju C 'dew.' As discussed in Section 2.4, however, there is good reason to assume that pR *tubu cannot be reflected as tsɨv in Tarama. The same reviewer also suggests that Tarama sɨv-'sour' is cognate with *sibu-'astringent' rather than *suju-'sour.' However, this hypothesis is not necessarily sufficient to explain the aspect of the adjectives for 'sour' and 'astringent' in Miyako. Of the four varieties of Miyako described by Hirayama et al. (1992-93), sɨvonly means both 'sour' and 'astringent' in Hirara. In other varieties, sɨv-(in the case of Ikema, suubecause of a further sound change *v > u) only signifies 'sour'; the meaning 'astringent' is signified by other adjectives such as futsɨgooin Tarama, ŋɡjain Ikema, and sɨtakupaɭin Irabu-Nagahama. The hypothesis that sɨv-'sour' is derived from *sibu-'astringent' presupposes four changes: the loss of the form *suju-, a semantic change in which *sibu-comes to mean 'sour' in addition to 'astringent,' the birth of a new word signifying 'astringent,' and the loss of the meaning 'sour' in *sibu-. This hypothesis fails to explain the motivation for the loss of the form *suju-. In contrast, the hypothesis of this study explains it as a result of a homonymic clash between *suju and *sibucaused by regular sound changes. We have postulated a regular sound change *suju-> sɨv-. This form is the same as sɨvderived from *sibu-through a regular sound change that spirantizes *bu into v(u) in many environments (see Section 4.2). The homonymic clash should have resulted in a polysemic adjective sɨv-'sour, astringent,' which remains as it is in Hirara. In other varieties, new words for 'astringent' such as Tarama futsɨgooare thought to have been coined so as to avoid the semantic merger between 'sour' and 'astringent.' Therefore, the hypothesis of this study can more parsimoniously explain the aspect of adjectives for 'sour' and 'astringent' in Miyako as the result of three well-motivated changes, namely, the homophonic clash because of regular sound changes, the birth of an adjective exclusively expressing 'astringent,' and the loss of the meaning of 'astringent' in sɨv-. tonal and segmental correspondence is irregular (cf. Yamatohama k'iɴ 'garment' vs k'iɴ 'garment'; Asama kiɴ BC 'garment' vs k'inuu A 'silk'; Yoron kiɴ B 'garment' vs kinu B 'silk'). It is suggested, therefore, that pR *kinu B lacked the meaning 'silk' and that apparently related Ryukyuan forms for 'silk' are Japanese loans. In any case, the fact that MJ and some Japanese dialects have kinu 'garment' belonging to Class 2.4 indicates that, regardless of whether 'garment' and 'silk' are cognates, the tone class of pR *kinu 'garment' is 2.4b.
38. *miso B 'bean paste': Matsumori (2012) reconstructs Class B. Tonal correspondence is regular between the dialects of Northern Ryukyuan but irregular between the dialects of Southern Ryukyuan. The former suggests that the pR Class of *miso is Class B, whereas the latter suggests it can be Class A, B, or C. The irregular tonal correspondence may suggest that borrowing is involved in this noun. However, the segmental correspondence is perfectly regular, which decreases the plausibility of the borrowing hypothesis. We, therefore, assume that the reflexes are inherited from pR, and the irregularity in the tonal correspondence is due to sporadic changes independently occurring in each dialect. Because Class B requires the minimal number of sound changes, we reconstruct Class B for pR *miso.
42. *soto B 'outside': Hattori (1979a,b) reconstructs Class B. In our primary data, reflexes of pR *soto exhibit a limited distribution. Hattori (1979a,b) reveals that its reflexes are observed in many dialects of Amami (in addition to Asama ɕutuu B , Kametsu ɕutu B , Shodon sɨt'uu BC , Naze sutʰu B , Aden sut'u B , and Onotsu sut'u BC ). Because the tonal correspondences between dialects are regular and the reflexes are observed both in Amami and Okinawa, we assume that they were inherited from pR *soto 'outside.' The lack of its attestation in Southern Ryukyuan may be because its reflexes were replaced by almost synonymous pR *poka B 'outside.' 49. *mado B 'spare time': Matsumori (2009Matsumori ( , 2012 also reconstructs Class B for *mado 'spare time,' although she may not agree with the view that pR *mado 'spare time' is related to MJ Class 2.4 MJ mado 'window.' We postulate a metaphoric semantic change 'window' > 'spare time,' that is, the change from 'special gap' to 'temporal gap.' This type of change is also observed in other languages such as Russian (cf. oкнo 'window,' oкнo мeждy лeкциями 'spare time between lectures') (Morkovkin et al. [eds.] 2016, 678). The nouns for 'window' per se are scarcely attested in Ryukyuan. Even if they are, they tend to be unrelated to MJ mado 2.5, as in Shuri haɕiɾu BC 'sliding window,' takabaɕiɾu BC 'high window,' and Ishigaki takaɸusaɾa BC 'high window.' It is true that Taketomi and Yonaguni have mədu 'window' and Yonaguni madu B 'window,' respectively, but they may be a Japanese loan. The Yonaguni madu B 'window' recorded by Uwano (2013) is especially controversial, because, as pointed out by the author, the same speaker produces amadu A 'window' instead of madu B , as described in an earlier study (Uwano 2010). Thus, a plausible scenario is that at the stage of pR, *mado lost its original meaning of 'window,' whereas the form *mado itself survived with its meaning metaphorically changed from 'window' to 'spare time.' 50. *majo B 'eyebrow': Although Martin (1987) regards MJ maju 2.5 'eyebrow' and maju 2.5 'cocoon' as cognates, they are listed as separate entries in the present study. Although the evidence for a distinction between pR *ju and *jo is in general scarce in modern Ryukyuan dialects (Pellard 2023), we reconstruct pR *majo, not *maju, based on the following evidence.⁸ The final syllable of pR *majo B 'eyebrow' is reflected differently from that of pR *kaju A 'porridge' in many Ryukyuan dialects (Ashikebu majo AB 'eyebrow' vs kai AB 'porridge'; Shuri maju B 'eyebrow' vs kee A 'porridge'; Hatoma maju 'eyebrow' vs kai 'porridge'; Ishigaki majoo BC 'eyebrow' vs kai A 'porridge'). Although it still seems possible to reconstruct pR *kai, not *kaju, for 'porridge,' this hypothesis is rejected because the second syllable of pR *kaju 'porridge' yields different forms from that of pR *mai A 'rice' in at least two Yaeyama dialects, namely, Hateruma (kee 'porridge' vs mëë 'rice') and Aragusuku (kai 'porridge' vs majɨ 'rice'). The correspondence with MJ kaju 2.1 'porridge' also provides an additional support for the reconstruction of pR *kaju A 'porridge.'  8 This discussion owes much to comments from Kenan Celik (personal communication).

Nouns for which the pR class was newly reconstructed (23 words)
We newly reconstructed the pR tone classes of 23 nouns ( Table 6). The majority of these nouns are attested in both Northern and Southern Ryukyuan, and their tonal correspondences are highly regular, but their pR tone classes have not been reconstructed in previous studies.  NKDJ's (2001NKDJ's ( , 1321 view. The Niigata dialect of Japanese also has kesa 'a short while ago' (Ohashi 2003, 90), suggesting that this type of semantic change is natural. Northern Ryukyuan forms exhibit an unexpected geminate, which requires further research.
71. *puna B 'gibel': As far as the data in Table 6 are concerned, the principle of parsimony cannot determine whether the tone class of pR *puna 'gibel' is B or C. The cognates of pR *puna are infrequent in Ryukyuan dialects. However, in Uwano (1998) study on many varieties in Okinoerabu Island, 24 of the 35 varieties have the cognates of pR *puna, and in all but one of the 24 varieties, the pitch pattern corresponds to Class B. Based on this, we reconstruct pR *puna B 'gibel.' The segmental correspondence in the first syllable in Yonaguni huna is irregular (compare pR *puna 'gibel' with pR *pune C 'boat' [Yonaguni nni C , Yoron puni C , Shuri ɸuɲi B ] and with pR *pugori A 'testicles' [Yonaguni ŋɡui A , Yoron pugui A , Shuri ɸuɡui A ]). It is therefore possible that borrowing is involved in this noun for Yonaguni. If Yonaguni huna C corresponding to Class C is a loanword, then the view that the tone class of pR *puna 'gibel' is B becomes more plausible.
73. *koto B 'zither': The segmental correspondences are irregular, suggesting that borrowing is involved. The correspondence between MJ and Southern Ryukyuan suggests that the first syllable in pR is *ko. Although pR *ko reflects as ɸu in Ie and Nakijin, the first syllable in 'zither' is kʰu in both dialects. The Shuri form has an unexpected long vowel in the second syllable. The Asama form has an unexpected vowel o, and the speaker is aware that this is a new word (Uwano 2017a, 152). However, the regularity of tonal correspondence (except Yonaguni) and the widespread attestation make us hesitant to deny the existence of this noun in pR.
74. *majo B 'cocoon': It is more difficult to determine whether the final syllable of 'eyebrow' is *ju or *jo than in the case of pR *majo B 'eyebrow.' Ashikebu distinguishes between majo AB 'eyebrow' and mai AB 'cocoon' and the final vowel of the latter agrees with that of kai AB 'porridge' (<pR *kaju A 'id.'); therefore, pR *maju 'eyebrow' is favored. However, Taketomi, which does not distinguish məju BC 'eyebrow' and məju BC  9 Pellard (2009a, 283) argues that the change *b > ɡ in *tubusi 'knee' is an irregular change that defines a subgroup called Common Miyako consisting of all the Miyako varieties but Tarama. In contrast, Celik (2020, 37) discovered that the change *b > ɡ is also observed in pR *tuburu C 'head, calabash' and suggests that this change may be a regular change with a very limited application environment. There are two other cognates that have a virtually identical environment: *tubus-'crushed' and *tubure-'get crushed' (Tarama tsɨbus-'crush,' tsɨburi-'get crushed'; Irabu tsɨbus-'crush,' tsɨburi-'get crushed'). The fact that *b does not change into ɡ in these verbs may suggest that the change *b > ɡ is sporadic, although there remains the possibility that *tubus-and *tubure-are loanwords given that their reflexes are scarcely found in Ryukyuan.
'cocoon,' has kəi A 'porridge' with a different vowel from that in 'eyebrow' and 'cocoon,' supporting pR *majo B 'cocoon.' Although the conflicting sound correspondences are difficult to interpret, we reconstruct pR *majo B 'cocoon' first, because both 'eyebrow' and 'cocoon' are majo 1 in OJ, and second, because there is a theory (Martin 1987, 474) that views these two nouns as cognates.

Class 2.4-2.5 nouns whose pR classes remain unsettled
The pR tone classes of the 17 nouns shown in Table 7 remain unsettled primarily because of irregular correspondences between Ryukyuan dialects or between pR and MJ.
76. *itu A? 'when,' 77. *sumi A? 'inside corner,' 78. *abo A? 'horsefly,' 79. *aki A? 'autumn,' 80. *pabu A? 'snake,' 81. *turo A? 'crane': The comparison of Ryukyuan dialects suggests that the tone class of these six nouns is Class A, which is an irregular reflex of MJ Class 2.4-2.5. At least for some of them, the irregularity may be explained by borrowing. However, the relatively regular tonal correspondences between dialects suggest that the borrowing, if any, occurred at the stage of pR. We leave open the issue regarding the irregularity and the reconstruction of tone classes of these nouns.
For *pabu A 'snake,' there is an additional issue about its cognacy. The irregularity of correspondence between MJ femi 2.5 'snake' and pR *pabu A occurs not only in tones but also in segments. Furthermore, the cognates of pR *pabu A usually refer to a venomous snake endemic to the Ryukyu Islands, rather than to snakes in general; thus, the meanings do not exactly correspond. Therefore, instead of the view that pR *pabu A is related to MJ femi 2.5 'snake' (e.g., Jarosz et al. [2022]), we should explore the possibility that pR *pabu A is related to MJ fami 'viper.' Cognates that may be related to MJ fami 'viper' are widely distributed in Kansai, Chugoku, and Shikoku Districts in the forms of hami, hame, hamu, habi, habe, habu, and so forth, all of which mean 'viper' (LAJ 1966-1974. Crucially, the pitch pattern of MJ fami 'viper' is HL, a regular reflex of Class 2.2, which in turn regularly corresponds to pR Class A. Meanwhile, Wadomari and Shuri have nouns referring to 'snake' in general, namely, çibu C and ɸiibu C , respectively. Although evidence is sparse, it may be possible to reconstruct pR *pebu C 'snake' along with pR *pabu A 'venomous snake.' If this holds true, then it is pR *pebu C 'snake' that is related to MJ femi 'snake,' and its tonal subclass is Class 2.5a. 82. *aka A? 'red,' 83. *ao B? 'blue,' 84. *kuro B? 'black,' 85. *siro B? 'white': These four words are color terms, which do not appear as free nominal morphemes in Ryukyuan but are adjectives that appear with a following component. In Table 7, the adjectives followed by a noun for 'color' (a reflex of pR *ero B 'color' [Pellard 2013]) are represented. *aka 'red' seems to belong to pR Class A, whereas the others seem to belong to pR Class B. However, these classes are those of adjectives, not nouns. Adjectives may have different tone systems than nouns, and the pR tone system of adjectives has been little investigated. We therefore do not reconstruct the tone class of these color terms.
86. *kazu B∼C 'number': Although de Boer (2010) reconstructs Class C, half of the dialects in our data exhibit pitch patterns corresponding with Class B.
However, if we follow Hattori's theory that 'oyster' and 'urchin' are related, it is necessary to compare the nouns for 'urchin' between dialects. They have various reflexes, such as Asama gatsɨɨtsɨ C , Ie ɡasisi AC , Nakijin ɡaɕiiɕi C , Shuri ɡatɕitɕaa BC , Tarama kadzɨtsɨ C , Ikema kazɨtsɨ C , Ishigaki kakidzɨ BC , and Taketomi  Table 7: Class 2.4-2.5 nouns for which pR tone classes remain unsettled. Ha79, Hattori (1979a,b), Ma09, Matsumori (2009), Bo10, de Boer (2010  Ashikebu kəkitɕi. Their tonal correspondences are perfectly regular, which allows us to reconstruct them as belonging to Class C. Irregular segmental correspondences make it difficult to reconstruct the proto-form. Assuming that variation in voicing across dialects is due to a sporadic voicing of original voiceless consonants, the proto-forms can be reconstructed as *kakisu, *kakiti, *kakesu, *kaketi, *kasesu, *kaseti, and so forth. If the proto-form for "urchin" were *kasesu or *kaseti, then it would be related not with MJ kaki 'oyster' but with MJ kase 'urchin.' By contrast, if the proto-form were *kakiti or *kakisu, then hypothetical relatedness between 'oyster' and 'urchin' becomes more plausible. In fact, Ashikebu has gak'isɨ B 'oyster' (not 'urchin'!), which corresponds to *kakisu. The Ashikebu form would provide a support for Hattori's (1979a,b) theory that the nouns for 'oyster' and 'urchin' are related. Noteworthy is that the pR tone class of 'urchin' is reconstructed as Class C based on the regular tonal correspondence between many dialects, whereas we reconstructed Class B for *gaki 'oyster' based on limited data of Northern Ryukyuan. If we focus on the limited distribution of *gaki, we can regard it as a loanword at later stage and reconstruct pR *kaki(-su) 'oyster,' whose tone class is C. The history of the pR noun for 'oyster' is complicated, which forces us to reserve judgment as to whether its tone class is B or C.
91. *koi?? 'carp': Only two dialects had a reflex of this noun in our primary data. Asama koi is irregular, as the dialect experienced a change from pR *ko to ku in word-initial position. In addition, it is a neologism according to the source. Irabu and Taketomi have kʊi and kui, respectively, although information about pitch patterns is not available. Nakijin and Shuri have polymorphemic forms, kʰuuʔjuu B and kuuʔiju BC , respectively. The second components must be a reflex of pR *io A 'fish,' although it is not clear whether the first component is related to MJ kofi 2.5 'carp.' It is especially doubtful for Nakijin, where pR *ko changes into ɸu in a word- Ha79 B 92. tune 2.5 93. asa 2.5 'longstanding' 'morning' pR *tune B~C *asa-?
Yonaguni --meanings coincide, they must be cognate. The former must have developed from the latter though an irregular sound change that drops the intervocalic *g. Notably, a verb igir-'to make a hole with an awl' is attested in dialects spoken in Kumamoto, Nagasaki, Saga, and Fukuoka Prefectures (NHDJ 1989, 155). This region completely covers the places where reflexes of *igiri are attested, suggesting that *igiri is a gerund of a verb *igir-. These observations lead us to assume that pR*iri is related to *iri, *igiri, and *igir-, but not to MJ kiri 'awl.' The irregular dropping of the intervocalic *g in *igiri is considered as a shared innovation between Ryukyuan and Japanese dialects in south Kyushu. 100. titi 2.4 'milk': Yoron tɕiitɕii C 'milk,' Nakijin tʑ'iitʑ'ii C , and Shuri tɕiitɕii BC seems related to titi 2.4 'milk.' However, they all are nursery words. The pR form for 'milk' is *tii B (cf. Ie tɕ'ii B , Nakijin tʑ'ii B , Shuri tɕii B , Tarama tsɨɨ B , Ikema tsɨɨ B ). The three nursery words must be a reduplication of pR *tii B 'milk' that developed independently in each dialect.
The other nouns listed in Table 8 have not been attested in our sources.

Discussion
Our survey reconstructed the pR tone classes of a total of 75 Class 2.4-2.5 nouns. They are roughly split in half between Class C (i.e., Class 2.4a/2.5a; 36 nouns) and Class B (i.e., Class 2.4b/2.5b; 39 nouns). The tonal correspondence between Ryukyuan dialects is generally regular, confirming that the split correspondence is genuine. Now that our survey has expanded the inventory of pR tone classes by 20 words, we can test the hypothesis that the split correspondence is due to a conditioned merger (Kindaichi 1960, Hirayama et al. 1966, Tokugawa 1990) based on a larger range of data. The conditioned merger hypothesis states that Class 2.4-2.5 nouns ending in a non-high vowel were tonally merged into Class 2.3 (corresponding to pR Class B), whereas those ending in a high vowel remained as a distinct class (i.e., pR Class C). Table 9 provides a contingency table that shows the cross tabulation of pR Class with word-final vowel (non-high vs high) in Class 2.4-2.5 nouns. Table 10 shows which nouns belong to each cell of Table 9, with a distinction between Classes 2.4 and 2.5. Nouns that contradict the predictions of the conditioned merger hypothesis are shaded.
Although there is a strong tendency for nouns with a word-final high vowel not to belong to 2.4b/2.5b (Class B), 10 of the 31 nouns go against this tendency. The tendency for nouns with a final non-high vowel not to belong to 2.4a/2.5a (Class C) is slightly weaker; 15 of the 44 nouns do not obey this tendency. Therefore, the conditioned merger hypothesis has also not been supported by this study. The absence of clear conditioning factors causing the split correspondence between MJ and pR requires reconstructing more contrasts for pJ. However, although the conditioned merger hypothesis is not tenable in its original formulation, it is also true that the skewed distribution in Table 9 strongly suggests a conditioned merger. Indeed, the association between the word-final vowel (non-high vs high) and the tone classes is statistically significant according to Pearson's chi-square test with Yates' continuity correction (χ 2 (1) = 6.9578, p < 0.01), although the strength of the association is moderate (Cramer's V = 0.3317 (95% CI [0.1132, 0.5195]). Observing the results of an analysis similar to that of this subsection (based on 42-43 nouns), de Boer (2010, 232) maintains that "[i]t is therefore not impossible that the presence of a close vowel in the second syllable played some role in preventing the merger of members of class 2.4/5 with class 2.3." On the one hand, it seems premature to reject the conditioned merger hypothesis completely, but on the other hand, to defend the hypothesis, we need to revise it so that it can account for exceptions. One solution would be to assume that another factor besides word-final high vowel prevented the merger. However, such assumption is merely speculative without further investigation. At the moment, the distinction between Class 2.4a/2.5a and Class 2.4b/2.5b must be seen as tracing back to pJ.