Show Summary Details

Page of

PRINTED FROM the OXFORD RESEARCH ENCYCLOPEDIA, LINGUISTICS (linguistics.oxfordre.com). (c) Oxford University Press USA, 2016. All Rights Reserved. Personal use only; commercial use is strictly prohibited. Please see applicable Privacy Policy and Legal Notice (for details see Privacy Policy).

date: 26 July 2017

Rendaku or Sequential Voicing in Japanese Phonology

Summary and Keywords

The term rendaku, sometimes translated as sequential voicing, denotes a morphophonemic phenomenon in Japanese. In a prototypical case, an alternating morpheme appears with an initial voiceless obstruent as a word on its own or as the initial element (E1) in a compound but with an initial voiced obstruent as the second element (E2) in a two-element compound. For example, the simplex word /take/ ‘bamboo’ and the compound /take+yabu/ ‘bamboo grove’ (cf. /yabu/ ‘grove’) begin with voiceless /t/, but this morpheme meaning ‘bamboo’ begins with voiced /d/ in /sao+dake/ ‘bamboo (made into a) pole’ (cf. /sao/ ‘pole’). Rendaku was already firmly established in 8th-century Old Japanese (OJ), the earliest variety for which extensive written records exist, and subsequent sound changes have made the alternations phonetically heterogeneous. Many OJ compounds with eligible E2s did not undergo rendaku, and the phenomenon remains pervasively irregular in modern Japanese. There are, however, many factors that promote or inhibit rendaku, and some of these appear to influence native-speaker behavior on experimental tasks. The best known phonological factor is Lyman’s Law, according to which rendaku does not apply to E2s that contain a non-initial voiced obstruent. Many theoretical phonologists endorse the idea that Lyman’s Law is a sub-case of the Obligatory Contour Principle, which rules out identical or similar units if they would be adjacent in some domain. Other well-known factors involve vocabulary stratum (e.g., the resistance to rendaku of recently borrowed E2s) or the morphological/semantic relationship between E2 and E1 (e.g., the resistance to rendaku of coordinate compounds). Some morphemes are idiosyncratically immune to rendaku. Other morphemes alternate but undergo rendaku in some compounds while failing to undergo it in others, even though no known factor is relevant. In addition, many individual compounds vary between a form with rendaku and a form without, and this variability is often not reflected in dictionary entries. Despite its irregularity, rendaku is productive in the sense that it often applies to newly created compounds. Many compounds, of course, are stored (with or without rendaku) in a speaker’s lexicon, but fact that native speakers can apply rendaku not just to existing E2s in novel compounds but even to made-up E2s shows that rendaku as an active process is somehow incorporated into the grammar.

Keywords: compound, Japanese, morphophonemic alternation, Obligatory Contour Principle, voicing

1. Definition of Rendaku

The technical term rendaku (連濁‎‎), sometimes translated as sequential voicing (Martin, 1952, p. 48), denotes a morphophonemic phenomenon that, in the late 20th century, became familiar to phonologists all over the world. Described as a process, rendaku replaces a morpheme-initial voiceless obstruent with a voiced obstruent. The prototypical environment is immediately following the boundary between the elements in a two two-element (E1+E2) compound. For example, a morpheme meaning ‘strength’ appears with initial /č/ [tɕ] as a word on its own, /čikara/, and as E1 in /čikara+kurabe/ ‘strength competition’ (cf. /kurabe/ ‘comparing’), but it appears with initial /ǰ/ [dʑ] as E2 in /soko+ǰikara/ ‘latent strength’ (cf. /soko/ ‘bottom’).1

Two important characteristics of rendaku are often ignored or treated as uninteresting background details. First, the voiced and voiceless obstruents paired by rendaku differ in many cases by more than just the presence vs. absence of voicing (Vance, 2014a, pp. 139–141; 2015a, pp. 397–398). Second, rendaku is irregular to a significant degree, often failing to apply to an eligible E2 even when no known inhibiting factor is at work (Vance, 2015a, p. 408).

2. Historical Development

Rendaku was already well established in 8th-century Old Japanese (OJ), the earliest variety of Japanese for which substantial written records exist. This section briefly sketches a plausible account of how rendaku originated. (For details and references, see Frellesvig, 2010, pp. 40–43; Vance, 2015a, pp. 299–402.) The OJ obstruents corresponding to modern (Tokyo “standard” Japanese) voiced obstruents were prenasalized (Frellesvig, 2010, pp. 34–35; Takayama, 2015, p. 627), and those corresponding to modern voiceless obstruents were probably voiceless word-initially and allophonically voiced (but not prenasalized) word-medially (Frellesvig, 2010, p. 35). Since OJ syllable structure was (C)(G)V (Frellesvig, 2010, p. 39), a word-medial consonant was immediately preceded by a vowel and immediately followed by a glide or a vowel. The prenasalized obstruents did not occur word-initially in OJ (Frellesvig, 2010, p. 43; Takayama, 2015, pp. 627–629) except possibly in mimetics and in some early borrowings from Chinese (Okumura, 1972, p. 111; Martin, 1987, pp. 29–30).

The attested OJ vocabulary includes compounds without rendaku, such as OJ/miya+pito/ ‘palace retainer’ (cf. OJ/miya/ ‘palace’, OJ/pito/ ‘person’), compounds with rendaku, such as OJ/satwo+bito/ ‘village person’ (cf. OJ/satwo/ ‘village’), and phrases such as OJ/mukasi no pito/ ‘person of long ago’, with the genitive particle OJ/no/ connecting two nouns (cf. OJ/mukasi/ ‘long ago’). Assuming that prenasalization had not developed just recently in OJ, it is reasonable to suppose that compounds with rendaku developed from earlier phrases in which the grammatical element (i.e., the genitive particle) contracted (Murayama, 1954, p. 107; Unger, 1975, pp. 8–9; Vance, 1982, pp. 335–338). Using “T” to represent any non-prenasalized obstruent and “nD” to represent the corresponding prenasalized obstruent, the proposal is: pre-OJ/…noT…/ > OJ/…nD…/. This change is natural, especially if the phrase had become “frozen” as a conventional collocation. If this scenario is correct, it explains why rendaku was irregular in OJ, just as it is in modern Japanese. There is no reason to assume that all OJ compounds go back to earlier noun+genitive+noun phrases; surely, some were created simply by juxtaposing E1 and E2.

Some instances of rendaku in OJ cannot plausibly be reconstructed as earlier E1+NV+E2 phrases, which indicates that the appearance of OJ/nD/ in place of OJ/T/ had already been reinterpreted as an indicator of compounding that could be extended to new compounds by analogy (Frellesvig, 2010, pp. 40–41). Since OJ prenasalized obstruents did not occur word-initially in non-mimetic native words (as noted above), a prenasalized obstruent was usually a reliable signal that there was no immediately preceding word boundary (Komatsu, 1981, p. 104).

The phoneme pairings in OJ rendaku were phonetically uniform:

Rendaku or Sequential Voicing in Japanese Phonology

Subsequently, well-understood phonological changes led to the phonetically heterogeneous pairings that we find in modern Japanese (Vance, 1987, p. 134):

Rendaku or Sequential Voicing in Japanese Phonology

Some treatments of rendaku involve a sub-phonemic linking element (𝕽) that can be understood as a synchronic residue of an earlier NV syllable (e.g., Itô & Mester, 1986, pp. 56–57; 2003, pp. 83–84). Feature-size compound-marking elements appear to have developed in many other languages as well (Ito & Mester, 2003, pp. 84–85; Labrune, 2016). Describing Japanese 𝕽 as just [vce] is misleading, of course, since it docks onto an E2-initial voiceless obstruent and converts that obstruent into its rendaku partner, rather than just adding phonetic voicing (§1). The feature label [daku] is used here, reflecting the traditional Japanese name dakuon (濁音‎‎) for the class of consonants that are the outputs of rendaku. (The literal meanings of the two Sino-Japanese morphemes in daku-on are ‘muddy, turbid’ and ‘sound’.)

Rendaku or Sequential Voicing in Japanese Phonology

The details of how [daku] is associated and realized depend on other assumptions built into a particular model.

In OJ, both prenasalized and non-prenasalized obstruents occurred morph-medially, and minimal pairs, like OJ/kasu/ ‘dregs’ and OJ/kazu/ ‘number’ were unremarkable. The corresponding two series of consonants in modern Japanese also occur freely in morph-medial position, and the descendants of the two words just cited remain a minimal pair: /kasu/ vs. /kazu/. Thus, any attempt to attribute rendaku to a phonotactic restriction against intervocalic voiceless obstruents is clearly misguided (Vance, 2015a, p. 413).

On the other hand, the idea of analyzing the rendaku alternations as word-initial devoicing has some obvious appeal (Kuroda, 2002), although morphemes that appear in isolation with an initial voiced obstruent are obviously problematic. An example is /gara/ ‘pattern; nature, character’, which originated as an allomorph with rendaku (cf. OJ/kara/) but eventually displaced the form without rendaku as an independent word. Inconsistent E2s (§6.2) are even more challenging. Assuming the underlying form /gi/ for /ki/~/gi/ ‘wood’, a devoicing analysis accounts for E2 in /yose+gi/ ‘parquetry’ (cf. /yose/ ‘bringing together’) and for the independent word /ki/ ‘wood’, but the absence of rendaku is mysterious in /cumi+ki/ ‘toy blocks’ (cf. /cumi/ ‘stacking’).

3. Phonological Factors

3.1 Lyman’s Law

Lyman’s Law is named after Benjamin Smith Lyman, an American geologist who proposed a version of it in the late 19th century (Lyman, 1894; Vance, 2007a). Lyman’s Law is a constraint on rendaku, and the version accepted by phonologists today is that rendaku does not apply to an E2 that contains a non-initial voiced obstruent. The examples below illustrate.

Rendaku or Sequential Voicing in Japanese Phonology

Some accounts in the literature invite the inference that rendaku always applies as long as Lyman’s Law is not violated, but it will be clear from the discussion below (§§4–6) that Lyman’s Law is a conditional (voiced obstruent in E2 → no rendaku) and that the inverse (no voiced obstruent in E2 → rendaku) is not even close to true.

In OJ, it appears that rendaku was inhibited by a nearby prenasalized obstruent on either side of the target consonant (i.e., the initial consonant of E2). This “strong version” of Lyman’s Law (Ramsey & Unger, 1972, pp. 287–289) correctly predicts the absence of rendaku in examples like OJ/siba+kaki/ ‘brushwood hedge’. There were no long (i.e., two-mora) syllables in OJ, and most morphs were one or two syllables. Since prenasalized obstruents did not occur initially in non-mimetic native words (§2), in most examples to which the strong version of Lyman’s Law applied, the isolation form of E1, E2, or both was (C1)(G)VC2(G)V, where C2 was a prenasalized obstruent (like OJ/b/ [ᵐb] in OJ/siba/). Consequently, it is hard to tell whether the inhibiting effect was limited to a syllable immediately adjacent to the syllable containing the target consonant. Only a handful of relevant examples are attested phonographically.3

In OJ/swode+tuke+goromo/ ‘sleeved garment’ (cf. OJ/swode/ ‘sleeve’, OJ/tuke/ ‘attaching’, OJ/koromo/ ‘garment’), E1 is complex: OJ/{swode+tuke}+goromo/. This example has rendaku even though E1 contains OJ/d/, but OJ/d/ is not in the morph immediately preceding E2. The question that remains, then, is whether or not an inhibitor consonant in E1 had to be in the last syllable of E1. The one relevant example indicates that the answer to this question is affirmative. In OJ/madara+busuma/ ‘multicolored bedding’ (cf. OJ/madara/ ‘multicolor’, OJ/pusuma/ ‘bedding’), the voiced/prenasalized obstruent in E1 is not in the syllable immediately preceding E2, and the compound has rendaku.

As for E2, the absence of rendaku, in examples like OJ/moto+pototogisu/ ‘returned cuckoo’ and OJ/yama+tatibana/ ‘mountain tangerine,’ is consistent with the idea that an inhibitor consonant did not have to be the first consonant following the target consonant, but the picture is unclear. It could be that OJ/pototogisu/ and OJ/tatibana/ were monomorphemic for OJ speakers, but they are much longer than the norm for native morphs, and there are plausible compound etymologies for both (Martin, 1987, pp. 416, 543). If the right-branch condition (§5.1) held in pre-OJ, originally compound E2s would have resisted rendaku, and they could have remained immune, even when they were no longer transparently analyzable. Furthermore,OJ/poto/ in OJ/pototogisu/ was probably mimetic (Martin, 1987, p. 416) and resistant to rendaku for that reason (§4.3).

Understood as a ban on prenasalized obstruents in adjacent syllables, the strong version of Lyman’s Law appears to have held without exception in OJ (Vance, 2005b). There is also an independent reason for thinking that adjacency was relevant. Northern Tōhoku dialects of Japanese, which preserve prenasalized voiced obstruents, prohibit these marked consonants from occurring in adjacent syllables (Miyashita, Irwin, Wilson, & Vance, 2016, p. 189). There clearly is no analogous restriction on voiced obstruents in consecutive syllables in modern Japanese. Examples like /kaze+gusuri/ ‘cold medicine’ (cf. /kusuri/ ‘medicine’) abound.

Many phonologists have enthusiastically adopted the proposal by Itô and Mester (1986, pp. 71–72) that Lyman’s Law in modern Japanese is a manifestation of the Obligatory Contour Principle (OCP). Specifically, OCP (voice) is credited with preventing the occurrence of two voiced obstruents in a single native morpheme. Kawahara (2008, pp. 324–327), however, argues that while OCP (prenasalization) is a plausible universal con­straint, OCP (voice) is not. It appears that Lyman’s Law lost its phonetic grounding when prenasalization disappeared, and it subsequently mutated into the modern Japanese version, in which adjacency is irrelevant. As noted above, a voiced obstruent in a non-adjacent syllable of the same morpheme seems to prevent rendaku (as in /ko+hicuǰi/ ‘lamb’), and voiced obstruents in consecutive syllables occur freely in native words as long as they straddle a morpheme boundary (as in /nabe+zoko/ ‘pot bottom’). For conservative modern speakers who have word-medial [ŋ] (Hibiya, 1999), E2-medial [ŋ] blocks rendaku just like a voiced obstruent (Kawahara & Zamma, 2016, pp. 25–26), and this “opacity” is further evidence that Lyman’s Law is no longer phonetically grounded.

A very small number of exceptions to Lyman’s Law exist in modern Japanese (Kindaichi, 1976, p. 5; Martin, 1987, p. 115; Suzuki, 2005), the best known of which is /nawa+bašigo/ ‘rope ladder’ (cf. /nawa/ ‘rope’, /hašigo/ ‘ladder’). There is a plausible account of the development of this item (and other compounds of the form X+/bašigo/) involving the now obsolete word /haši/ ‘ladder, stairway’. When this /haši/ dropped out of use, three-element compounds like /{nawa+baši}+go/, with /ko/~/go/ ‘(lit.) child’ used as a diminutive (Martin, 1987, p. 115), were reinterpreted as two-element compounds containing a monomorphemic E2: /hašigo/~/bašigo/ (Suzuki, 2005, 2007, p. 288). This kind of reanalysis scenario cannot account, however, for all the Lyman’s Law violations in the existing vocabulary (Vance, 2015a, pp. 404–405).

3.2 Identity Avoidance

Many languages display phenomena that suggest a dispreference for identical elements that are adjacent or in close proximity. It has been suggested that rendaku is inhibited if it would create a sequence of two identical CV syllables, and promoted if it would eliminate such a sequence (Satō, 1989, p. 256). A systematic search of the existing vocabulary (Irwin, 2014a) did not find evidence for this “identity avoidance” effect, but in an experimental study that asked participants to decide whether or not to apply rendaku to test compounds with real E1s and made-up E2s (Kawahara & Sano, 2016), identity avoidance appeared to be a factor influencing responses. These experimental results are consistent with the notion that tendencies in speaker behavior do not necessarily reflect patterns in the existing vocabulary in some straightforward way (Kawahara & Sano, 2016, pp. 53–54). Speakers may internalize generalizations for which there appears to be no basis in the vocabulary, and they may fail to internalize generalizations for which the vocabulary provides ample evidence.

Kawahara and Sano (2016) also found that participants’ response choices were influenced by identity avoidance at the consonantal level. That is, when E1 ended in a CV syllable and E2 began with a CV syllable containing a different vowel, there was a preference for responses in which the consonants in these two syllables were not the same (i.e., /…C1V1+C2V2…/ was preferred to /…C1V1+C1V2…/). This aversion to consecutive CV syllables with different vowels but the same consonant was not as strong as the tendency to avoid consecutive CV syllables in which both the consonant and the vowel were identical.

3.3 Rosen’s Rule

Rosen (2001, p. 70; 2003, p. 6) proposes that, in a restricted sector of the vocabulary, rendaku is predictable when either element in a two-element compound is longer than two moras. This claim has been dubbed “Rosen’s Rule” (Irwin, 2016b; Vance, 2015a, p. 410, 2015b), and Rosen himself limited it to two-element compounds in which both elements are nouns belonging to the native vocabulary stratum (§4.2). Rosen (2001, p. 28) explicitly excludes coordinate compounds like /kami+hotoke/ ‘gods and Buddhas’ because they generally resist rendaku (§5.3). Compounds containing the bound native numeral /hito/ ‘one’ as E1 also resist rendaku (§5.6), so examples like /hito+fukuro/ ‘one bag’ should be excluded as well.

When E1, E2, or both are longer than two moras in a relevant compound, Rosen’s Rule predicts that if E2 begins with a voiceless obstruent as a word on its own and is not immune to rendaku, E1+E2 will have rendaku. Typical examples are /abura+gami/ ‘oil paper’ (cf. /kami/ ‘paper’), with a three-mora E1, and /šita+gokoro/ ‘ulterior motive’ (cf. /kokoro/ ‘heart, mind’), with a three-mora E2. Rosen’s Rule does not predict rendaku in examples such as /abura+kasu/ ‘oily dregs’ and /atama+kazu/ ‘number of persons’ because the E2s are immune, /kasu/ ‘dregs’ idiosyncratically (§4.2) and /kazu/ ‘number’ because of Lyman’s Law (§3.1).

There are exceptions to Rosen’s Rule, such as /hidari+te/ ‘left hand’, which has a three-mora E1 and a non-immune E2 (Kubozono, 2005, p. 16), but the tendency is very strong in existing noun+noun compound nouns, even if elements that are not monomorphemic and/or not native are taken into consideration (Vance, 2015a, pp. 410–412; 2015b). The motivation for Rosen’s Rule, however, remains something of a mystery (Vance, 2015b, pp. 212–213), and the results of a psycholinguistic experiment (Kawahara & Sano, 2014) suggest that native speakers may not internalize it.

3.4 Individual Segments

Vance and Asai (2016) survey claims that particular phonemes favor or disfavor rendaku, either as targets or as nearby segments in the environment. In existing compounds, E2s beginning with /s/ as independent words have a lower rendaku rate than E2s beginning with other voiceless obstruents (Toda, 1988), and some preliminary experimental results suggest that speakers internalize a preference reflecting this tendency (Ihara, Tamaoka, & Lim, 2011).

As for segments in the environment, the most common claim is that an E2-final moraic nasal /N/ promotes rendaku. This claim presupposes that the E1-initial voiced obstruent in examples like /peN+zara/ ‘pen tray’ (cf. /sara/ ‘plate’) should not be attributed to “postnasal voicing” (§4.5). When potential confounding factors are controlled to a reasonable extent, the hypothesized trend holds in certain small sectors of the vocabulary but not overall (Vance & Asai, 2016, pp. 123–129; Vance, 2017). Nonetheless, in one experiment with test items consisting of a real E1 and a made-up E2, the responses showed a higher rendaku rate for the E1s ending in /N/ than for E1s ending in a vowel (Tamaoka, Ihara, Murata, & Lim, 2009, pp. 28–31). The number of test items was small, and other factors may have been at work, but these results are suggestive, and it would be not at all surprising to see the same effect of E1-final /N/ emerge in other experiments.

4. Vocabulary Strata

4.1 Etymological Origin and Stratification

Japanese morphemes exhibit differences in phonological structure and behavior that correlate with etymological origin, and much work on Japanese phonology incorporates an explicit stratification of the lexicon (Irwin, 2011, pp. 4–14; Itô & Mester, 1995, p. 817; McCawley, 1968, pp. 62–75). Most researchers assume at least these four strata: native Japanese (sometimes labeled “Yamato”), mimetic, Sino-Japanese, and recent loans (sometimes labeled “foreign”).

It is often suggested that stratum membership should be determined solely by phonological criteria, without regard to etymology (Itô & Mester, 1999, p. 63), but this is easier said than done (Labrune, 2012, pp. 13–24; Vance, 2002a). A core–periphery approach (Itô & Mester, 1999) offers a way of modeling phonologically based strata that allows for fuzzy categories, but the problems are complex, and no attempt will be made to resolve them here. Instead, the stratum labels below denote categories that are essentially etymological.

4.2 Native Elements

As E2s, rendaku-eligible native morphemes typically show rendaku, as long as Lyman’s Law (§3.1) is not violated, but there are many exceptions. In contrast, mimetic elements (§4.3) consistently resist rendaku, and recent loan elements (§4.4) resist almost as consistently. Sino-Japanese elements (§4.5) present a complex picture, but resistance is clearly the norm (Vance, 1996). Many researchers simply say that rendaku is confined to native elements, but such assertions normally presuppose that etymologically non-native morphemes can be treated as phonologically native (§4.1).

A few native morphemes are idiosyncratically immune to rendaku (Irwin, 2009, pp. 192–193; Kuroda, 2002, p. 340), including these:

Rendaku or Sequential Voicing in Japanese Phonology

Reduplication, however, strongly favors rendaku and can trump immunity (§5.2).

4.3 Mimetic Elements

Japanese has a rich inventory of mimetic words (Hamano, 1998; Shibatani, 1990, pp. 153–157), and although mimetic elements are etymologically native, they are usually treated as a separate stratum in phonological analysis (Itô & Mester, 1999, p. 63; McCawley, 1968, pp. 64–65; Nasu, 2015). Many mimetic words are reduplicated, and such words consistently resist rendaku (Martin, 1952, p. 49; Okumura, 1955, p. 962), as in the examples below.

Rendaku or Sequential Voicing in Japanese Phonology

Rendaku in words like these would, of course, diminish the iconicity of reduplication as an indicator or repetition or continuation (Nasu, 2015, p. 283; Vance, 1987, p. 122). There are, however, a few non-reduplicated mimetic compounds (Hamano, 1998, pp. 47–50), and these also resist rendaku, as in /peča+kuča/ ‘chitter-chatter’ (cf. /peča+peča/ ‘chatter-chatter’, /kuča+kuča/ ‘chomp-chomp’).

It thus appears that mimetic elements resist rendaku whether or not they are reduplicated. Non-reduplicated mimetic compounds are arguably coordinate, however, and coordinate compounds generally do not have rendaku (§5.3). To demonstrate beyond doubt that mimetic elements are immune to rendaku, it would be necessary to show that rendaku does not occur in clearly non-coordinate compounds with mimetic second elements. Such compounds are rare, and the few that have been cited in the literature (Hamano, 1998, p. 55) do not have rendaku-eligible second elements. In sharp contrast to reduplicated mimetic words, rendaku is the norm in most other kinds of reduplicated words involving native Japanese bases (§5.2).

There also seems to be a consistent absence of rendaku in recently coined reduplicated words that could be called “quasi-mimetic.” Such words are highly colloquial, and their bases are not etymologically mimetic, but they resemble mimetic words in terms of semantics and grammatical behavior (Nishimura, 2013, pp. 84–85). An example is /šima+šima/ ‘stripey’ (Vance, 2014b, pp. 33–34). The base /šima/ ‘stripe’ is a native noun and is not immune to rendaku, as shown by compounds like /tate+ǰima/ ‘vertical stripe’. Historically, the boundary between the mimetic and non-mimetic sectors of the Japanese vocabulary has always been fuzzy, allowing morphemes to cross from one side to the other (Hamano, 1998, pp. 6–7).

There is a widely recognized phonesthetic association between voiced obstruents and mostly negative attributes, particularly in initial position in native words (Suzuki, 1962, pp. 23–24; Endō, 1977, pp. 222–228; Komatsu, 1981, pp. 87–88). This association is especially clear and relatively systematic in the mimetic vocabulary, in which root-initial voiced obstruents signal characteristics labeled with adjectives such as “big,” “coarse,” “heavy,” “ponderous,” “vulgar,” etc. (Hamano, 1998, pp. 83–85; Komatsu, 1981, p. 75). The examples below are typical.

Rendaku or Sequential Voicing in Japanese Phonology

As noted above, prenasalized voiced obstruents in OJ did not occur initially in non-mimetic native words (§2), and word-initial voiced obstruents in modern Japanese are intuitively “marked” (Martin, 1987, p. 130). A variety of changes have produced word-initial voiced obstruents in a small number of native words (Suzuki, 2009), but the examples of interest here appear to have originated as deliberate creations that exploited the phonesthetic association described above. In the pairs below, the word beginning with a voiced obstruent has a more colloquial flavor as well as a less favorable meaning.

Rendaku or Sequential Voicing in Japanese Phonology

The forms with initial voicing can be treated as derivatives of the forms without (Suzuki, 1962, pp. 26–27), although this analysis is not always appropriate synchronically, since native speakers do not see the connection in every pair. In a few cases, a voiced-initial form has replaced its voiceless-initial source rather than developing into a separate lexical item (Komatsu, 1981, pp. 96–97; Suzuki, 2009, p. 134). An example is /boke-ru/ ‘to become senile’. An earlier voiceless-initial form is attested, but there is no corresponding verb */hoke-ru/ in modern Japanese. This replacement was probably fostered by the fact that the original verb had an intrinsically negative meaning, making it compatible with an initial voiced obstruent.

Ladd (2014, pp. 103–105) uses the term “modulations” for sub-phonemic differences that carry “expressive” meanings like those conveyed by initial voiced obstruents in Japanese mimetics. A feature-size morpheme is one way of modeling this phonesthetic voicing, and in its extension to the non-mimetic vocabulary, it involves the same pairs of voiced and voiceless obstruents as rendaku, which means that the feature label [daku] (§2) is appropriate.4 Thus, it has the same phonological form as the rendaku linking element 𝕽 (§2), but its meaning requires it to be treated as a different morpheme.

4.4. Recent Loan Elements

Recent loan elements can be defined as morphemes adopted into Japanese after the mid-16th century (Irwin, 2011, p. 10), although the few borrowings from Chinese during this period are typically excluded. Recent loan elements are generally immune to rendaku. The examples below illustrate with a recently borrowed morpheme and a comparable native morpheme.

Rendaku or Sequential Voicing in Japanese Phonology

A few morphemes borrowed from languages other than Chinese are attested with rendaku, but most of these elements are relatively old (Irwin, 2011, pp.150–153; Takayama, 2005, pp. 178–181; Vance, 2015a, pp. 415–416). The examples below illustrate, with the date of the earliest citation for each borrowed element shown in parentheses.5

Rendaku or Sequential Voicing in Japanese Phonology

Rendaku sporadically affects elements that are obvious recent loans (Nakagawa, 1966, p. 308), as in ?/iNdo+gareH/ ‘Indian curry’ (cf. /kareH/ < English curry). Native speakers typically interpret such examples as attempts at humor rather than as viable candidates for inclusion in the lexicon.

The etymological stratum of E1 has no obvious effect on the likelihood of rendaku, but a recent study suggests that, when experimental subjects respond to test compounds with a nonce E2, a recently borrowed E1 reduces the likelihood of rendaku (Tamaoka et al., 2009, pp. 28–34). In any case, a recently borrowed E1 certainly does not prevent rendaku; there is nothing unusual about examples like /boHru+dama/ ‘pitch outside the strike zone’ (cf. /boHru/ ‘ball’, /tama/ ‘ball, pitch’).

4.5. Sino-Japanese Elements

A very small number of Japanese morphemes may be prehistoric loans from Chinese (Kamei, 1954), but for all practical purposes they are native. One such morpheme that is rendaku-eligible is /fumi/ ‘letter’ (Martin, 1987, p. 417), and it shows rendaku in examples such as /koi+bumi/ ‘love letter’.

Overall, Sino-Japanese elements show rendaku much less than native Japanese elements but much more than recently borrowed elements. A prototypical Sino-Japanese word is a binom, a word written with two Chinese characters (kanji), each character representing (at least arguably) a Sino-Japanese morpheme (Martin, 1975, p. 151). A typical example is /ka·soku/ (加速‎‎) ‘acceleration’, with the boundary between the two morphs indicated by a dot rather than a plus. Like the majority of Sino-Japanese morphs, /ka/ ‘adding’ and /soku/ ‘speed’ are both bound. Many binoms were coined by Japanese speakers; they are Sino-Japanese because the component morphemes are borrowings from Chinese.

Most Sino-Japanese binoms are immune to rendaku, but many are not (Vance, 1996), including those in the examples below.

Rendaku or Sequential Voicing in Japanese Phonology

Many Sino-Japanese binoms begin with a voiceless obstruent but contain a medial voiced obstruent at the beginning of the second morph. No such binom ever appears with rendaku as an E2 (Vance, 2015a, pp. 419–420; Vance & Asai, 2016, p. 121), presumably because of Lyman’s Law (§3.1).6 In the case of /hoH·ǰiN/ ‘corporation’ (cf. /hoH/ ‘law’, /ǰiN/ ‘person’), for example, there are no words like */ei·ri+boH·ǰiN/, as opposed to actually occurring /ei·ri+hoH·ǰiN/ ‘profit-making corporation’. Since the target consonant (/h/ in /hoH·ǰiN/) and the inhibitor consonant (/ǰ/ in /hoH·ǰiN/) are in different morphs in a Sino-Japanese binom, binom E2s necessitate careful consideration of the domain of Lyman’s Law in modern Japanese.

The behavior of individual Sino-Japanese morphemes raises some very awkward problems for the analysis of rendaku, but even a superficial explanation would require a long digression (for details and references, see Vance, 1996, 2011). One source of these problems is the fact that a process called “postnasal voicing” (PNV) seems to have applied without exception in Early Middle Japanese (c. 800–1200) (Frellesvig, 2010, pp. 307–308). PNV is no longer active in modern Japanese, and sound changes have rendered its effects opaque (Vance & Asai, 2016, pp. 127–129), but some binom-medial voiced obstruents are historically due to PNV.7 One example is the /ǰ/ in /kaN·ǰiN/ (勧進‎‎) ‘soliciting donations’. When the kanji 進‎‎ signifies the first morpheme in a binom, as in /šiN·teN/ (進展‎‎) ‘progress’, it always represents /šiN/, so it is easy to understand the temptation to treat /ǰiN/ in /kaN·ǰiN/ as an instance of rendaku. Nonetheless, it is better to assume that Sino-Japanese binoms are not ordinary compounds and that binom-internal voiced obstruents are not instances of rendaku (Vance & Asai, 2016, p. 124). Itô and Mester (2003, p. 80) treat Sino-Japanese binoms as instances of “root compounding” rather than “word compounding” and say that only the latter “is the locus of rendaku voicing.”

5. Morphological/Semantic Factors

5.1. The Right-Branch Condition

Otsu (1980, pp. 217–222) proposes that rendaku can only appear in a morph that is on a right-side branch in constituent structure. This right-branch condition predicts the absence of rendaku in the three-element compound /yama+šima+uma/ ‘mountain zebra’ (cf. /yama/ ‘mountain’, /šima/ ‘stripe’, /uma/ ‘horse’) because the constituent structure is /yama+{šima+uma}/, with /šima/ on a left branch. Since there is no voiced obstruent in /šima+uma/ ‘zebra’, the absence of rendaku in the three-element compound cannot be attributed to any version of Lyman’s Law (§3.1), and /šima/ is not im­mune to rendaku (§4.3).

There have been some interesting theoretical attempts to derive the right-branch condition from other principles (Itô & Mester, 2003, pp. 202–212; Kubozono, 2005, pp. 11–15), but the constraint itself is suspect because the existing vocabulary is rife with counterexamples (Vance, 2007b, pp. 224–226; 2015a, pp. 424–425). Experimental studies involving real vocabulary elements combined into three-element nonce compounds (Kozman, 1998; Kumagai, 2014) have cast doubt on the psychological status of the right-branch condition for ordinary speakers.

5.2. Non-Mimetic Reduplication

In contrast to reduplicated mimetic words (§4.3), rendaku is strongly favored in most other kinds of reduplicated words involving native Japanese bases (Ogura, 1910, pp. 21–22), as long as Lyman’s Law (§3.1) is not violated. The examples below are typical (Vance, 2015a, pp. 417–419).

Rendaku or Sequential Voicing in Japanese Phonology

The preference for rendaku is so strong that it can trump immunity (Nishimura, 2007, pp. 22–23). For example, /saki+zaki/ ‘destinations; future’ has rendaku despite the fact that /saki/ ‘tip; ahead’ is otherwise rendaku-immune (§4.2).

Many reduplicated words with rendaku function as adverbs and are semantically and grammatically very similar to reduplicated mimetic words (Martin, 1975, pp. 410–411, 799–800). The assertion that “quasi-mimetic” words consistently resist rendaku (§4.3) seems correct intuitively, but it is not at all clear that reduplicated words with non-mimetic bases can be reliably identified as “quasi-mimetic” on independent grounds (i.e., using criteria other than the absence of rendaku) (Vance, 2014b, pp. 34–36).

5.3. Coordinate Compounds

It has long been observed that coordinate (i.e., dvandva) compounds generally resist rendaku (Lyman, 1894, p. 9; Okumura, 1955, p. 962). None of the E2s in the examples below is immune to rendaku.

Rendaku or Sequential Voicing in Japanese Phonology

There are, however, a few compounds that are at least arguably coordinate but nonetheless have rendaku (Irwin, 2016a, p. 82; Vance, 2015a, pp. 425–426). One clear example is /mono+goto/ ‘things’ (cf. /mono/ ‘[concrete] thing’, /koto/ ‘[abstract] thing’). Dictionaries list /ya+dama/ with the coordinate definition ‘arrows & bullets’ (cf. /ya/ ‘arrow’, /tama/ ‘bullet’), but also with the non-coordinate definition ‘arrow’ (i.e., an arrow or arrow-shaped object used as a projectile).

There are also coordinate compounds that have rendaku but occur only as elements in longer compounds, not as words on their own. One example is /{aši+de}+matoi/ ‘hindrance’ (cf. /aši/ ‘foot’, /te/ ‘hand’,\ /mato-u/ ‘to wrap’) (Vance, 2015a, p. 426).

Dictionaries give coordinate definitions for the compound adjectives /ita+gayu-i/ ‘painful & itchy’ (cf. /ita-i/ ‘painful’, /kayu-i/ ‘itchy’) and /ama+zuQpa-i/ ‘sweet & sour’ (cf. /ama-i/ ‘sweet’, /suQpa-i/ ‘sour’), but it is not entirely implausible to interpret such words as involving a modifier+head relationship between the two roots (Vance, 2015a, p. 425).

5.4. Compounds Containing Verb and Adjective Elements

Japanese has many V+V=V compounds, that is, compound verbs in which both E1 and E2 are based on verbs (Shibatani, 1990, pp. 246–247; Tagashira & Hoff, 1986). A typical example is /kiri+hanas-u/ ‘to sever’ (cf. /kir-u/ ‘to cut’, /hanas-u/ ‘to separate’). E1 in a V+V=V compound is segmentally identical to the “adverbial form” (sometimes called the “stem” or the “infinitive”) of the base verb, which is used to mark a clause as non-sentence-final. In many cases, there is also a segmentally identical deverbal noun (cf. /kiri/ ‘end, limit’).8 Whether or not such a deverbal noun exists as an independent word, it is often unclear whether E1 in a V+V=V compound should be analyzed as verbal or nominal, and no attempt will be made here to resolve this problem. The inflectional endings on a V+V=V compound are the same as those that E2 would take as an independent verb.

Compound nouns based on two verb elements (V+V=N compounds) are also abundant in Japanese. A typical example is /kiki+tori/ ‘catching what someone says’ (cf. /kik-u/ ‘to listen’, /tor-u/ ‘to take’). The uncertainty about whether an element is verbal or nominal applies to both E1 and E2 in a V+V=N compound. In many cases, a V+V=N compound and a V+V=V compound based on the same two verbs both exist (cf. /kiki+tor-u/ ‘to catch what someone says’), but some V+V=N compounds have no corresponding compound verb, and some V+V=V compounds have no corresponding compound noun.

When a V+V=V compound and a paired V+V=N compound both exist, the most common pattern by far is for both to lack rendaku (Vance, 2005a, pp. 93–98), as in /kiki+tor-u/ and /kiki+tori/ above. A small minority of such pairs have rendaku in both words, as in /wari+bik-u/ ‘to discount’ and /wari+biki/ ‘discount’ (cf. /war-u/ ‘to divide’, /hik-u/ ‘to subtract’). Another small minority have rendaku in the noun but not in the verb, as in /toHri+kakar-u/ ‘to happen to pass by’ and /toHri+gakari/ ‘happening to pass by’ (cf. /toHr-u/ ‘to pass’, /kakar-u/ ‘to become involved’).

On the other hand, unpaired V+V=N compounds are much more likely to show rendaku than unpaired V+V=V compounds (Vance, 2005a, p. 99). For example, /okuri+kaes-u/ ‘to send back’ (cf. /okur-u/ ‘to send’, /kaes-u/ ‘to return’) is a V+V=N compound, but there is no corresponding V+V=N compound */okuri+kaeši/ or */okuri+gaeši/ (although the adverbial form /okuri+kaeš-i/ does exist, since all verbs have this inflectional category). In contrast, /oboe+gaki/ ‘memorandum’ (cf. /oboe-ru/ ‘to recall’, /kak-u/ ‘to write’) is a V+V=N compound, but there is no corresponding V+V=V compound */oboe+kak-u/ or */oboe+gak-u/. The absence of rendaku in verbs like /okuri+kaes-u/ is typical, and so is the presence of rendaku in nouns like /oboe+gaki/, as long as Lyman’s Law (§3.1) is not violated (Vance, 2015a, p. 428).

Japanese adjectives, like verbs, are inflected, but in contrast to V+V=V compounds, rendaku appears in nearly half of A+A=A compounds (i.e., compound adjectives in which both E1 and E2 are based on adjectives) (Vance, 2005a, p. 110). Examples are relatively scarce, but one with rendaku is /usu+gura-i/ ‘dimly lit’ (cf. /usu-i/ ‘pale’, /kura-i/ ‘dark’), and one without rendaku is /sema+kuruši-i/ ‘cramped’ (cf. /sema-i/ ‘narrow’, /kuruši-i/ ‘oppressive’). In other types of compounds that contain an adjective element, rendaku seems to be the norm (Kikuta, 1971; Toda, 1994; Vance, 2005a, pp. 98–99), as in A+V=V compounds like /čika+zuke-ru/ ‘to bring near’ (cf. /čika-i/ ‘near’, /cuke-ru/ ‘to attach’), in A+V=N compounds like /hoso+gaki/ ‘writing with fine lines’ (cf. /hoso-i/ ‘thin’, /kak-u/ ‘to write’), and in V+A=A compounds like /utagai+buka-i/ ‘suspicious’ (cf. /utaga-u/ ‘to doubt’, /fuka-i/ ‘deep’).

A+A=N and V+A=N compounds are too rare to allow any meaningful assessment regarding rendaku (Vance, 2005a, p. 428). N+A=A compounds like /kokoro+zuyo-i/ ‘reassuring’ (cf. /kokoro/ ‘heart, mind’, /cuyo-i/ ‘strong’) and N+A=N compounds like /ma+ǰika/ ‘close proximity’ (cf. /ma/ ‘space’, /čika-i/ ‘near’) are more common but have not been investigated thoroughly (Vance, 2015a, p. 429). N+V=V compounds like /na+zuke-ru/ ‘to name’ (cf. /na/ ‘name’, /cuke-ru/ ‘to attach’) are also common but are still awaiting systematic scrutiny. N+V=N compounds, on the other hand, have attracted a great deal of attention, and they are discussed immediately below in §5.5.

A+N=N compounds like /naga+gucu/ ‘boot’ (cf. /naga-i/ ‘long’, /kucu/ ‘shoe’) and V+N=N compounds like /nagare+boši/ ‘shooting star’ (cf. /nagare-ru/ ‘to flow’, /hoši/ ‘star’) are also common, but they do not seem to have been investigated separately from prototypical N+N=N compounds. It may well be that it makes no difference whether E1 is deverbal or deadjectival (as opposed to being an underived noun) when E2 is an underived noun, but the question is still open.

5.5. N+V=N Compounds

In an N+V=N compound, E1 is an underived noun and E2 is deverbal. It has long been claimed that the semantic relationship between E1 and E2 is a crucial factor in determining whether or not rendaku occurs in such compounds. Specifically, if E1 is semantically the direct object of the verb from which E2 is derived, rendaku is supposedly inhibited. The examples below illustrate the putative pattern with an E2 derived from the verb /kir-u/ ‘to cut’.

Rendaku or Sequential Voicing in Japanese Phonology

In the first example, since /kami/ is clearly the direct object of cutting, /kami+kiri/ is a DO+V=N compound. In the second example, the direct object is something that can be cut into round slices (e.g., a lemon), which makes /wa+giri/ a nonDO+V=N compound.

Early versions of the proposed contrast described the E1 in a nonDO+V=N compounds as an “adverbial modifier” of the verb underlying E2 (Okumura, 1955, p. 962; Sakurai, 1966, p. 41), presumably excluding the subject of that verb. N+V=N compounds involving a subject are very rare for transitive verbs, as in /muši+kui/ ‘worm damage’ (cf. /muši/ ‘worm’, /ku-u/ ‘eat’), but fairly common for intransitive verbs, as in /kata+kori/ ‘shoulder stiffness’ (cf. /kata/’ shoulder’, /kor-u/ ‘to become stiff’). There is no consensus about whether Subject+V=N compounds resist rendaku (Kindaichi, 1976, p. 12; Sugioka, 1986, p. 108). In any case, the systematic dictionary searches and experiments reported in the literature have focused on direct object vs. “adverbial” E1s and largely ignored subject E1s.

DO+V=N compounds with rendaku, like /kuǰi+biki/ ‘drawing lots’ (cf. /kuǰi/ ‘lot’, /hik-u/ ‘to draw’), are actually very common, whereas nonDO+V=N compounds without rendaku are in fact quite rare (Vance, 2014a, pp. 144–145), leaving aside cases in which rendaku would violate Lyman’s Law (§3.1). Thus, in the existing vocabulary, rendaku is more likely in nonDO+V=N compounds than in DO+V=N compounds (Yamaguchi, 2011, p. 124; Yamaguchi & Tanaka, 2013, pp. 160–161). It has been suggested that rendaku always occurs when possible in newly coined nonDO+V=N compounds but seldom occurs in newly coined DO+V=N compounds (Sugioka, 2005, pp. 217–218), but experimental results have been ambiguous. One study found no evidence for the psychological reality of the DO vs. non-DO contrast (Kozman, 1998), while another study (using a different experimental task) reports that the contrast was a significant factor influencing responses (Vance, 2014, pp. 143–149).

5.6. Bound E1s

Certain bound E1s clearly inhibit rendaku. The best-known instances are the honorific prefixes /o/ and /go/. Native /o/ attaches mainly but not only to native bases, while Sino-Japanese /go/ attaches almost exclusively to Sino-Japanese bases. The examples below illustrate. The examples in parentheses show that the E2s are not immune to rendaku.

Rendaku or Sequential Voicing in Japanese Phonology

The native numeral /hito/ ‘one’ also consistently inhibits rendaku as an E1 (Irwin, 2012, pp. 31–32; Nakagawa, 1966, p. 314), as in /hito+koe/ ‘one cry, one word’ (cf. /koe/ ‘voice’). Here again, the E2 is not immune to rendaku, as shown by /hito+goe/ ‘human voice’ (cf. /hito/ ‘person’).

5.7. E2 Polysemy

Almost all content morphemes are polysemous, and it is now well established that the rendaku behavior of an E2 can differ dramatically depending on the sense involved (Irwin, 2016a, pp. 104–105; Vance, 2015a, p. 433). For example, /te/~/de/ ‘hand; arm’ has a wide variety of figurative senses that are difficult to distinguish in any consistent way. Overall, among the many compounds with this E2, those without rendaku outnumber those with rendaku, but for most senses, there are at least some compounds with rendaku, as in the examples below.

Rendaku or Sequential Voicing in Japanese Phonology

In fact, /te/~/de/ in the sense of ‘wound’, as in the last example above, seems to appear consistently with rendaku, although the number of examples is very small. The two others are /asa+de/ ‘shallow wound’ (cf. /asa-i/ ‘shallow’) and /ita+de/ ‘serious wound’ (cf. /ita-i/ ‘painful’).

In contrast, there is another figurative sense of /te/~/de/ that is immune to rendaku. When this E2 combines with a verb E1 to form a noun denoting a human agent, as in /uri+te/ ‘seller’ (cf. /ur-u/ ‘to sell’), it never appear as /de/. This pattern is highly productive, and /te/ in this use is often described as a derivational suffix (Irwin, 2016a, p. 83), although it is no simple matter to decide what counts as an affix. In any event, examples with long E1s, such as /kasegi+te/ ‘breadwinner’ (cf. /kaseg-u/ ‘to earn’) would violate Rosen’s Rule (§3.3) if this sense of /te/ were lumped together with the other senses (Kubozono, 2005, p. 16).

6. Unpredictability

6.1. Variability in Existing Words

For some existing words, a form with rendaku and a form without rendaku co-exist, although a dictionary will often list one form to the exclusion of the other. For example, a compound meaning ‘dry cough’ can be pronounced either /kara+seki/ or /kara+zeki/ (cf. /kara/ ‘empty’, /seki/ ‘cough’). In many cases, an individual speaker will prefer one form to the other and dismiss the alternative as mistaken or dialectal, but surveys done for NHK (the Japanese public broadcasting service) have shown that the preference sometimes correlates with the speaker’s age or gender (Ōta, 2011; Shioda, 1999, 2001, 2011a, 2011b).

There is a universal aversion to homophony and to homosemy, sometimes called the one-form-one-meaning principle (Matthews, 1997, p. 255), and a natural reaction to variability in form is “(semantic) bifurcation” (Bolinger, 1968, p. 110; Hudson, 2000, pp. 262–263), that is, ascribing a semantic distinction, however slight, to the different forms. This preference for a one-to-one correspondence between forms and meanings influences all speakers, including linguists and lexicographers. Recently, a group of Japanese university students attending a lecture on rendaku reacted to /oku+fuka-i/~/oku+buka-i/ ‘deeply recessed’ (cf. /oku/ ‘interior’, /fuka-i/ ‘deep’) by insisting that the form without rendaku has the literal meaning and that the form with rendaku has the figurative meaning ‘profound’.9 It is always possible that a consensus or near-consensus will develop in a speech community, splitting two alternative pronunciations of a word into separate lexical items (Vance, 2002b). Despite this potential, however, there are very few firmly established pairs containing the same E1 and E2 in which the presence or absence of rendaku (sometimes with a concomitant difference in pitch-accent pattern) carries a semantic distinction (Vance, 2015a, pp. 434–435).

6.2. Inconsistent Behavior of Individual E2s

Many E2s behave inconsistently, appearing with rendaku in some words and without rendaku in others, even when no promoting or inhibiting factor is involved. The examples below illustrate, with the sense of E2 held constant in each pair to exclude the possibility that the difference might be due to polysemy (§5.7).

Rendaku or Sequential Voicing in Japanese Phonology

Inconsistent E2s vary widely in terms of their inclination to undergo rendaku, from almost always to almost never (Irwin, 2014b, 2016a, pp. 101–105).

6.3. Productivity

Most of the factors cataloged above (§§3–5) are not deterministic. They appear to influence the likelihood of rendaku but do not trigger or block rendaku consistently in every relevant lexical item. In short, the existing vocabulary displays many tendencies, some very strong and others rather weak. Despite this overall unpredictability, however, rendaku cannot simply be a matter of memorizing the words that have it, because native speakers often apply rendaku to neologisms that combine existing elements and even to experimental test items that contain a made-up E1 or E2. It seems safe to say that the rendaku behavior of ordinary speakers involves some combination of lexical retrieval and on-line assembly (Kawahara, 2015; Kawahara & Zamma, 2016, p. 32; Kubozono, 2005, pp. 5–7).

Decisions about whether or not rendaku appears in a novel compound are surely based in large part on analogy to existing vocabulary items (Ohno, 2000), and this means that different speakers may behave differently and that any given speaker may behave differently on different occasions. No two speakers have identical vocabularies, and many existing compounds vary from speaker to speaker with respect to the presence or absence of rendaku (§6.1). Also, ephemeral situational factors undoubtedly play a role in determining which existing vocabulary items are most salient on a particular occasion. It is difficult to say with any assurance what precisely the basis for analogy might be when an experimental participant chooses a form with or without rendaku for a nonce word. In many cases, existing E1+E2 combinations that are semantically and/or phonologically similar in some way provide likely models (Ohno, 2000, pp. 160–161).

A vague analogical account along these lines for nonce words does not predict unanimity, and this vagueness is a virtue. Even homogeneous groups of native speakers seldom respond unanimously to an experimental test item. This kind of variability is compatible with the spread of rendaku into borrowed E2s, both Sino-Japanese (§4.5) and more recent (§4.4). Furthermore, native speakers occasionally forget an existing word and re-create it on the fly, and they may re-evaluate the presence or absence of rendaku in a word they know when they hear it pronounced differently by someone else. Consequently, existing words can gain or lose rendaku over time (Vance, 2007a, p. 163).

Further Reading

Haraguchi, S. (2001). On rendaku. Phonological Studies, 4, 9–32.Find this resource:

Itô, J., & Mester, A. (2003). Japanese morphophonemics. Cambridge, MA: MIT Press.Find this resource:

Kindaichi, H. (1976). Rendaku no kai. Sophia Linguistica, 2, 1–22.Find this resource:

Kubozono, H. (2005). Rendaku: Its domain and linguistic conditions. In J. van der Weijer, K. Nanjo, & T. Nishihara (Eds.), Voicing in Japanese (pp. 5–24). Berlin: De Gruyter Mouton.Find this resource:

Ohno, K. (2000). The lexical nature of rendaku in Japanese. In M. Nakayama & C. J. Quinn, Jr. (Eds.), Japanese/Korean linguistics 9 (pp. 151–164). Stanford, CA: CSLI.Find this resource:

Rosen, E. (2003). Systematic irregularity in Japanese rendaku: How the grammar mediates patterned lexical exceptions. Canadian Journal of Linguistics, 48, 1–37.Find this resource:

Vance, T. J. (2007). Have we learned anything about rendaku that Lyman didn’t already know? In B. Frellesvig, M. Shibatani, & J. C. Smith (Eds.), Current issues in the history and structure of Japanese (pp. 153–170). Tokyo: Kurosio.Find this resource:

Vance, T. J. (2014). If rendaku isn’t a rule, what in the world is it? In K. Kabata & T. Ono (Eds.), Usage-based approaches to Japanese grammar (pp. 137–152). Amsterdam: John Benjamins.Find this resource:

Vance, T. J. (2015). Rendaku. In H. Kubozono (Ed.), The handbook of Japanese phonetics and phonology (pp. 397–441). Berlin: De Gruyter Mouton.Find this resource:

Vance, T. J., & Irwin, M. (Eds.). (2016). Sequential voicing in Japanese compounds: Papers from the NINJAL rendaku project. Amsterdam: John Benjamins.Find this resource:

References

Bolinger, D. (1968). Aspects of language. New York: Harcourt, Brace & World.Find this resource:

Endō, K. (1977). Dakuon genka ishiki: gotō no seidaku o koto ni suru nijū-go o taishō ni. Kokugo Kokubun, 46(4), 222–234.Find this resource:

Frellesvig, B. (2010). A history of the Japanese language. Cambridge, U.K.: Cambridge University Press.Find this resource:

Hamano, S. (1998). The sound-symbolic system of Japanese. Stanford, CA/Tokyo: CSLI/Kurosio.Find this resource:

Hibiya, J. (1999). Variationist sociolinguistics. In N. Tsujimura (Ed.), The handbook of Japanese linguistics (pp. 101–120). Oxford: Blackwell.Find this resource:

Hudson, G. (2000). Essential introductory linguistics. Oxford: Blackwell.Find this resource:

Ihara, M., Tamaoka, K., & Lim, H. (2011). Rendaku and markedness: Phonetic and phonological effects. Paper presented at the July meeting of the Tokyo Circle of Phonologists, Tokyo.Find this resource:

Irwin, M. (2009). Prosodic size and rendaku immunity. Journal of East Asian Linguistics, 18, 179–196.Find this resource:

Irwin, M. (2011). Loanwords in Japanese. Amsterdam: John Benjamins.Find this resource:

Irwin, M. (2012). Rendaku dampening and prefixes. NINJAL Research Papers, 4, 27–36.Find this resource:

Irwin, M. (2014a). Rendaku across duplicate moras. NINJAL Research Papers, 7, 93–109.Find this resource:

Irwin, M. (2014b). Rendaku lovers, rendaku haters and the logistic curve. In M. Giriko, N. Nagaya, A. Takemura, & T. J. Vance (Eds.), Japanese/Korean Linguistics (Vol. 22; pp. 37–51). Stanford, CA: CSLI.Find this resource:

Irwin, M. (2016a). The rendaku database. In T. J. Vance & M. Irwin (Eds.), Sequential voicing in Japanese compounds: Papers from the NINJAL rendaku project (pp. 79–106). Amsterdam: John Benjamins.Find this resource:

Irwin, M. (2016b). Rosen’s Rule. In T. J. Vance & M. Irwin (Eds.), Sequential voicing in Japanese compounds: Papers from the NINJAL rendaku project (pp. 107–117). Amsterdam: John Benjamins.Find this resource:

Itô, J., & Mester, R.-A. (1986). The phonology of voicing in Japanese. Linguistic Inquiry, 17, 49–73.Find this resource:

Itô, J., & Mester, A. (1995). Japanese phonology. In J. A. Goldsmith (Ed.), The handbook of phonological theory (pp. 817–838). Oxford: Blackwell.Find this resource:

Itô, J., & Mester, A. (1999). The phonological lexicon. In N. Tsujimura (Ed.), The handbook of Japanese linguistics (pp. 62–100). Oxford: Blackwell.Find this resource:

Itô, J., & Mester, A. (2003). Japanese morphophonemics. Cambridge, MA: MIT Press.Find this resource:

Kamei, T. (1954). Chinese borrowings in prehistoric Japanese. Tokyo: Yoshikawa Kōbunkan.Find this resource:

Kawahara, S. (2008). Phonetic naturalness and unnaturalness in Japanese loanword phonology. Journal of East Asian Linguistics, 17, 317–330.Find this resource:

Kawahara, S. (2015). Can we use rendaku for phonological argumentation? Linguistics Vanguard, 1(1), 3–14.Find this resource:

Kawahara, S., & Sano, S. (2014). Testing Rosen’s Rule and strong Lyman’s Law. NINJAL Research Papers, 7, 111–120.Find this resource:

Kawahara, S., & Sano, S. (2016). Rendaku and identity avoidance. In T. J. Vance & M. Irwin (Eds.), Sequential voicing in Japanese compounds: Papers from the NINJAL rendaku project (pp. 47–55). Amsterdam: John Benjamins.Find this resource:

Kawahara, S., & Zamma, H. (2016). Generative treatments of rendaku and related issues. In T. J. Vance & M. Irwin (Eds.), Sequential voicing in Japanese compounds: Papers from the NINJAL rendaku project (pp. 13–34). Amsterdam: John Benjamins.Find this resource:

Kikuta, T. (1971). Yōgen no rendaku no ichi-yōin. Kaishaku, 17(5), 24–29.Find this resource:

Kindaichi, H. (1976). Rendaku no kai. Sophia Linguistica, 2, 1–22.Find this resource:

Komatsu, H. (1981). Nihongo no on’in (Nihongo no sekai 7). Tokyo: Chūōkōronsha.Find this resource:

Kozman, T. (1998). The psychological status of syntactic constraints on rendaku. In D. Silva (Ed.), Japanese/Korean linguistics: Vol.8 (pp. 107–120). Stanford, CA: CSLI.Find this resource:

Kubozono, H. (2005). Rendaku: Its domain and linguistic conditions. In J. van der Weijer, K. Nanjo, & T. Nishihara (Eds.), Voicing in Japanese (pp. 5–24). Berlin: De Gruyter Mouton.Find this resource:

Kumagai, G. (2014). The psychological status of the right-branch condition on rendaku: An experiment with specific contexts. Studies in Language Sciences, 13, 124–145.Find this resource:

Kuroda, S.-Y. (2002). Rendaku. In N. M. Akatsuka & S. Strauss (Eds.), Japanese/Korean linguistics: Vol.10 (pp. 337–350). Stanford, CA: CSLI.Find this resource:

Labrune, L. (2012). The phonology of Japanese. Oxford: Oxford University Press.Find this resource:

Labrune, L. (2016). Rendaku in cross-linguistic perspective. In T. J. Vance & M. Irwin (Eds.), Sequential voicing in Japanese compounds: Papers from the NINJAL rendaku project (pp. 195–233). Amsterdam: John Benjamins.Find this resource:

Ladd, D. R. (2014). Simultaneous structure in phonology. Oxford: Oxford University Press.Find this resource:

Lyman, B. S. (1894). The change from surd to sonant in Japanese compounds. Philadelphia: Oriental Club of Philadelphia.Find this resource:

Martin, S. E. (1952). Morphophonemics of standard colloquial Japanese. (Language Dissertation No. 47). Baltimore: Waverly Press.Find this resource:

Martin, S. E. (1975). A reference grammar of Japanese. New Haven, CT: Yale University Press.Find this resource:

Martin, S. E. (1987). The Japanese language through time. New Haven, CT: Yale University Press.Find this resource:

Matthews, P. (1997). The concise Oxford dictionary of linguistics. Oxford: Oxford University Press.Find this resource:

McCawley, J. D. (1968). The phonological component of a grammar of Japanese. The Hague: Mouton.Find this resource:

Miyashita, M., Irwin, M., Wilson, I., & Vance, T. J. (2016). Rendaku in Tōhoku Japanese: The Kahoku-chō survey. In T. J. Vance & M. Irwin (Eds.), Sequential voicing in Japanese compounds: Papers from the NINJAL rendaku project (pp. 173–193). Amsterdam: John Benjamins.Find this resource:

Murayama, S. (1954). Rendaku ni tsuite. Gengo Kenkyū, 26/27, 106–110.Find this resource:

Nakagawa, Y. (1966). Rendaku, rensei (kashō) no keifu. Kokugo Kokubun, 35(6), 302–314.Find this resource:

Nasu, A. (2015). The phonological lexicon and mimetic phonology. In H. Kubozono (Ed.), The handbook of Japanese phonetics and phonology (pp. 253–288). Berlin: De Gruyter Mouton.Find this resource:

Nihon Kokugo Dai-jiten Dainihan HenshūIin-kai (Ed.). (2000–2002). Nihon kokugo dai-jiten (2d ed.). Tokyo: Shōgakukan.Find this resource:

Nishimura, K. (2007). Rendaku and morphological correspondence. Phonological Studies, 10, 21–30.Find this resource:

Nishimura, K. (2013). Morphophonology in Japanese compounding (Doctoral dissertation). University of Tokyo, Tokyo.Find this resource:

Ogura, S. (1910). Raiman-shi no rendaku-ron. Kokugakuin Zasshi, 16(7), 9–23; (8), 31–45.Find this resource:

Ohno, K. (2000). The lexical nature of rendaku in Japanese. In M. Nakayama & C. J. Quinn, Jr. (Eds.), Japanese/Korean linguistics: Vol.9 (151–164). Stanford, CA: CSLI.Find this resource:

Okumura, M. (1955). Rendaku. In Kokugo Gakkai (Ed.), Kokugo-gaku jiten (pp. 961–962). Tokyo: Tōkyōdō.Find this resource:

Okumura, M. (1972). Kodai no on’in. In N. Nakata (Ed.), Kōza Nihongo-shi 2: On’in-shi, moji-shi (pp. 63–171). Tokyo: Taishūkan.Find this resource:

Ōta, M. (2011). Onna wa otoko yori mo “tsumi-tsukuri”. Hōsō Kenkyū to Chōsa. November, 26–37.Find this resource:

Ramsey, S. R., & Unger, J. M. (1972). Evidence for a consonant shift in 7th century Japanese. Papers in Japanese Linguistics, 1, 278–295.Find this resource:

Otsu, Y. (1980). Some aspects of rendaku in Japanese and related problems. In Y. Otsu & A. Farmer (Eds.), MIT Working Papers in Linguistics 2: Theoretical issues in Japanese linguistics (pp. 207–227). Cambridge, MA: Department of Linguistics and Philosophy, MIT.Find this resource:

Rosen, E. R. (2001). Phonological processes interacting with the lexicon: Variable and non-regular effects in Japanese phonology (Doctoral dissertation). University of British Columbia, Vancouver, Canada.Find this resource:

Rosen, E. (2003). Systematic irregularity in Japanese rendaku: How the grammar mediates patterned lexical exceptions. Canadian Journal of Linguistics, 48, 1–37.Find this resource:

Sakurai, S. (1966). Kyōtsugo no hatsuon de chūi subeki kotogara. In Nihon Hōsō Kyōkai (Ed.), Nihongo hatsuon akusento jiten (pp. 31–43). Tokyo: Nihon Hōsō Shuppan Kyōkai.Find this resource:

Satō, H. (1989). Fukugō-go ni okeru akusento kisoku to rendaku kisoku. In M. Sugitō (Ed.), Kōza Nihongo to Nihongo kyōiku 2: Nihongo no onsei, on’in (jō) (pp. 233–265). Tokyo: Meiji Shoin.Find this resource:

Shibatani, M. (1990). The languages of Japan. Cambridge, U.K.: Cambridge University Press.Find this resource:

Shioda, T. (1999). Yōgo no kettei. Hōsō Kenkyū to Chōsa. April, 94–97.Find this resource:

Shioda, T. (2001). Yōgo no kettei. Hōsō Kenkyū to Chōsa. April, 96–105.Find this resource:

Shioda, T. (2011a). Yōgo no kettei. Hōsō Kenkyū to Chōsa. October, 98–105.Find this resource:

Shioda, T. (2011b). Yōgo no kettei. Hōsō Kenkyū to Chōsa. November, 86–97.Find this resource:

Sugioka, Y. (1986). Interaction of derivational morphology and syntax. New York: Garland.Find this resource:

Sugioka, Y. (2005). Multiple mechanisms underlying morphological productivity. In S. S. Mufwene, E. J. Francis, & R. S. Wheeler (Eds.), Polymorphous linguistics: Jim McCawley’s legacy (pp. 204–223). Cambridge, MA: MIT Press.Find this resource:

Suzuki, T. (1962). On’in kōtai to igi bunka no kankei ni tsuite: Iwayuru seidaku-on no tairitsu o chūshin to shite. Gengo Kenkyū, 42, 23–30.Find this resource:

Suzuki, Y. (2005). Raiman no hōsoku no reigai ni tsuite. Bunkyō Gakuin Daigaku Gaikoku-go Gakubu Bunkyō Gakuin Tanki Daigaku Kiyō, 4, 249–265.Find this resource:

Suzuki, Y. (2007). Raiman no Nihongo kenkyū. Bunkyō Gakuin Daigaku Gaikoku-go Gakubu Bunkyō Gakuin Tanki Daigaku Kiyō, 5, 225–239.Find this resource:

Suzuki, Y. (2009). Gotō dakuon-go “ba” no seiritsu katei ni tsuite. Bunkyō Gakuin Daigaku Gaikoku-go Gakubu Bunkyō Gakuin Tanki Daigaku Kiyō, 9, 133–149.Find this resource:

Tagashira, Y., & Hoff, J. (1986). Handbook of Japanese compound verbs. Tokyo: Hokuseido Press.Find this resource:

Takayama, T. (2005). A survey of rendaku in loanwords. In J. van de Weijer, K. Nanjo, & T. Nishihara (Eds.), Voicing in Japanese (pp. 177–190). Berlin: De Gruyter Mouton.Find this resource:

Takayama, T. (2015). Historical phonology. In H. Kubozono (Ed.), The handbook of Japanese phonetics and phonology (pp. 621–650). Berlin: De Gruyter Mouton.Find this resource:

Tamaoka, K., Ihara, M., Murata, T., & Lim, H. (2009). Effects of first-element phonological-length and etymological-type features on sequential voicing (rendaku) of second elements. Journal of Japanese Linguistics, 25, 17–38.Find this resource:

Toda, A. (1988). Wago no hi-rendaku kisoku to rendaku keikō. Dōshisha Kokubungaku, 31, 80–96.Find this resource:

Toda, A. (1994). Gendai-go ni okeru wago no rendaku (1): Fukugō keiyōshi no rendaku. Dōshisha Kokubungaku, 40, 159–172.Find this resource:

Unger, J. M. (1975). Studies in early Japanese morphophonemics (Doctoral dissertation). Yale University, New Haven, CT.Find this resource:

Vance, T. J. (1982). On the origin of voicing alteration in Japanese consonants. Journal of the American Oriental Society, 102, 333–341.Find this resource:

Vance, T. J. (1987). An introduction to Japanese phonology. Albany, NY: SUNY Press.Find this resource:

Vance, T. J. (1996). Sequential voicing and Sino-Japanese. Journal of the Association of Teachers of Japanese, 30, 22–43.Find this resource:

Vance, T. J. (2002a). Another look at vocabulary stratification. In F. Cavoto (Ed.), The linguist’s linguist: A collection of papers in honour of Alexis Manaster Ramer (pp. 433–436). Munich: Lincom Europa.Find this resource:

Vance, T. J. (2002b). Semantic bifurcation in Japanese compound verbs. In N. M. Akatsuka & S. Strauss (Eds.), Japanese/Korean linguistics: Vol.10 (pp. 365–377). Stanford, CA: CSLI.Find this resource:

Vance, T. J. (2005a). Rendaku in inflected words. In J. van de Weijer, K. Nanjo, & T. Nishihara (Eds.), Voicing in Japanese (pp. 89–103). Berlin: De Gruyter Mouton.Find this resource:

Vance, T. J. (2005b). Sequential voicing and Lyman’s Law in Old Japanese. In S. S. Mufwene, E. J. Francis, & R. S. Wheeler (Eds.), Polymorphous linguistics: Jim Mc­Cawley’s legacy (pp. 27–43). Cambridge, MA: MIT Press.Find this resource:

Vance, T. J. (2007a). Have we learned anything about rendaku that Lyman didn’t already know? In B. Frellesvig, M. Shibatani, & J. C. Smith (Eds.), Current issues in the history and structure of Japanese (pp. 153–170). Tokyo: Kurosio.Find this resource:

Vance, T. J. (2007b). Right branch or head: What difference does it make? In S. Kuno, S. Makino, & S. Strauss (Eds.), Aspects of Japanese linguistics: In honor of Noriko Akatsuka (pp. 221–240). Tokyo: Kurosio.Find this resource:

Vance, T. J. (2008). The sounds of Japanese. Cambridge, U.K.: Cambridge University Press.Find this resource:

Vance, T. J. (2011). Rendaku in Sino-Japanese: Reduplication and coordination. In H. Sohn, H. Cook, W. O’Grady, L. A. Serafim, & S. Y. Cheon (Eds.), Japanese/Korean linguistics 19 (pp. 465–482). Stanford, CA: CSLI.Find this resource:

Vance, T. J. (2014a). If rendaku isn’t a rule, what in the world is it? In K. Kabata & T. Ono (Eds.), Usage-based approaches to Japanese grammar (pp. 137–152). Amsterdam: John Benjamins.Find this resource:

Vance, T. J. (2014b). Rendaku to onomatope no jōgo. Kokugoken Purojekuto Rebyū, 5(1), 32–38.Find this resource:

Vance, T. J. (2015a). Rendaku. In H. Kubozono (Ed.), The handbook of Japanese phonetics and phonology (pp. 397–441). Berlin: De Gruyter Mouton.Find this resource:

Vance, T. J. (2015b). Rendaku no fu-kisoku-sei to Rōzen no hōsoku. Kokuritsu Kokugo Kenkyūjo Ronshū, 9, 207–214.Find this resource:

Vance, T. J. (2017). Rendaku following a moraic nasal. In S. Tanaka et al. (Eds.), Kubozono Haruo sensei kanreki kinen ronbun-shū (pp. 19–39). Tokyo: Kaitakusha.Find this resource:

Vance, T. J., & Asai, A. (2016). Rendaku and individual segments. In T. J. Vance & M. Irwin (Eds.), Sequential voicing in Japanese compounds: Papers from the NINJAL rendaku project (pp. 119–137). Amsterdam: John Benjamins.Find this resource:

Yamaguchi, K. (2011). Accentedness and rendaku in Japanese deverbal compounds. Gengo Kenkyū, 140, 117–134.Find this resource:

Yamaguchi, K., & Tanaka, S. (2013). Rendaku variation in deverbal compounds. In J. van de Weijer & T. Nishihara (Eds.), Current issues in Japanese phonology: Segmental variation in Japanese (pp. 149–166). Tokyo: Kaitakusha.Find this resource:

Notes:

(1.) Phonemic transcriptions of modern (Tokyo “standard”) Japanese follow the system described by Vance (2008), with /N/ representing a moraic nasal, /Q/ representing a moraic obstruent, and /H/ representing vowel length.

(3.) For an introduction to the OJ writing system, see Frellesvig (2010, pp. 11–20).

(4.) The phoneme pairings in mimetics are not exactly the same (Hamano, 1998, pp. 190–195).

(5.) The citation dates are from Nihon kokugo daijiten (Nihon Kokugo Dai-jiten Dainihan Henshū Iin-kai, 2000–2002), the Japanese counterpart to the Oxford English Dictionary.

(6.) This claim assumes that the name /saburoH/ is classified as something other than a Sino-Japanese binom, because /sabu/ is anomalous (Vance, 2015a, p. 420).

(7.) It is often claimed that PNV is active in modern Japanese as a phonotactic constraint within native morphemes and across inflectional boundaries. This claim is highly dubious, but this article is not the place to argue the point.

(8.) An adverbial form and a deverbal noun based on the same verb may differ in pitch-accent pattern (Martin, 1975, pp. 883–885).

(9.) Thanks to Mark Irwin for sharing this experience.