You are on page 1of 11

2016

Nasality and Rendaku


AN ANALYSIS OF VOICING IN JAPANESE, WITH APPLICATION
TOWARD UNDERSTANDING RENDAKU
NAKITA BARAKADYN
Barakadyn |1

Nasality and Rendaku: Merging Ideas

Nakita Barakadyn

Rendaku, or ‘sequential voicing’, is a well-known phenomenon in Japanese wherein when forming a


compound, if the second half begins with an unvoiced obstruent, it undergoes voicing: te ‘hand’ + kami
‘paper’ -> tegami ‘letter’. As often as scholars dip their toes into the topic, there is no clear answer as to
why rendaku occurs, or what specifically motivates it. There are too many circumstances in which this
voicing does not happen, the most famous and concrete of which is Lyman’s Law (Shibatani 1990), which
states that if the second element of the compound already contains a voiced obstruent, then rendaku
does not take place: jude ‘boiling’ + tamago ‘egg’ > judetamago (*judedamago). This sound law, however,
does not explain every instance in which rendaku fails and also says nothing about the process itself. Given
an historical phonological rule which states a morpheme cannot have more than one voiced element
(Labrune 2012), it could be asserted that Lyman’s Law is a superfluous description of a process already
present, having nothing to do with rendaku itself. By examining voicing patterns in Japanese, from an
historical perspective, I propose that rendaku is simply post-nasal voicing and that there is a nasal element
present in the UR in those compounds which undergo sequential voicing, such as the attributive no or the
instrumental nite (Classical Japanese).
Barakadyn |2

Introduction

This paper is an exploration of the internal voicing phenomenon in Japanese called rendaku or

‘sequential voicing’. It is a hotly contested phonological process, with a large number of exceptions and

blocking environments. There are many theories regarding the phonological motivation of this voicing,

but little evidence and even less agreement. Ultimately the entire affair is rather muddy1, but hopefully,

after laying out some currently accepted facts, a definitive statement can be made about when to

expect rendaku and why it might be happening. We will first talk about voicing in Japanese in general,

before talking about the particular circumstances surrounding rendaku. From there we will marry the

two ideas and see that perhaps rendaku isn’t very mysterious after all, being a result of underlying

connective particles with historically voicing elements.

Voicing as a Feature

Muddy Sounds

We will begin the discussion by overviewing voicing in Japanese from an historical perspective.

First, a note on voiced and voiceless counterparts. Some of the expected voiceless/voiced contrasts

exist: [s]/[z], [k]/[g], [t]/[d], but [b] is considered the voiced counterpart to [h] (historically [p]). Note that

[p] does exist in the modern language, having been reintroduced sometime around the Middle Japanese

period, and is called ‘half-voiced’ in the literature. Native vocabulary, termed Yamato, contain a low

instance of voiced obstruents- A survey of the Nara Japanese Dictionary indicates that in bimoraic2

nouns, /b/ /d/ /z/ and /g/ only occur word-initially a combined 8.1% of the time (Labrune 2012: 100).

1
I could not resist the pun! For those not in the know, it will become clear later.
2
Morae (singular: mora) refer to syllable-like units of measure. The distinction is time based, and a short vowel
(plus an optional consonant) constitute one mora, while long vowels are two morae. Geminated consonants also
count as a single mora, thus kuki = 2, kuuki = 3, and kuukki = 4 morae.
Barakadyn |3

Word-initial voiced consonants have a stigma in the language, with pairs existing where one word has a

positive or neutral connotation and the other a negative one. Examples include: sama ‘appearance’ but

zama ‘messy appearance or plight’; sakana ‘fish’ but zakana ‘rotten fish’; hareru ‘clear up’ but bareru

‘reveal (a secret)’ (Labrune 2012: 103). This concept seems to be understood, at least subconsciously, by

the Japanese, as can be seen in an essay denouncing all words that start with b- as having some

association with negativity (Hibbett & Itasaka 1965). Though it is a bit of an exaggeration, it works

towards proving the point. The term used for voicing in Japanese dakuon even translates to ‘muddy or

dirty sound’.

Where voiced obstruents did appear, it is considered a hard rule that there was never more than

one within a single morpheme, keeping in mind that nasals do not count in this restriction. That being

said, voicing is historically considered to be a bit unstable and is essentially non-distinctive. There exist

doublets with voiced and voiceless versions appearing indicating the same meaning at a single point in

history, as well as shifting surface forms diachronically but in either direction (voiced to voiceless, as well

as the reverse) (Labrune 2012). There is also the existence of a classical poetic device wherein puns and

ambiguities are created relying on homophony, but voicing was not an important consideration to the

creation of these puns, meaning that sounds could count as homophonous even if they had different

voicing specifications3. This sort of pun-like structure can still be seen in modern Japanese in both

number mnemonics and in music.

In writing, voicing is marked by diacritics, further driving home the point that voicing as a

feature is more of an afterthought, than a central tenant to the phonology of the language. There have

been periods in Japanese history where voicing was not marked at all (much like stress in many

languages) and these same diacritics were once used to mark accent in the language. It is important to

3
Phrasing borrowed from Dr. Peggy Renwick. Thank you for the clarity!
Barakadyn |4

remember not to conflate writing with language itself, but choices made in representation can provide

clues into historical processes and perspectives. It is considerations like these that lead Labrune and

others to believe that voicing in Japanese is autosegmental, much like tone. While it is perhaps not the

driving force of this paper, I agree that the evidence is very supportive of such a notion.

Post-Nasal Voicing

By the Classical period we begin to see intervocalic voicing. Most voiced consonants in the

modern language come from this, or from Chinese loan (Vovin 2003). This medial voicing is not a simple

voicing assimilation through contact with adjacent vowels, however- This medial voicing comes from

post-nasal assimilation. Voiced consonants are considered to be underlying /NC/, where C stands for any

voiceless obstruent and the suprascript N for any nasal. Most medial voiced obstruents can be traced

back to reductions of syllables with nasal onsets. For example, tabe- ‘eat’ < tamap- ‘bestow’ or touge

‘mountain pass’ < Old Japanese (OJ) tamukey (Martin 1987). The medial vowel elides, and the voiceless

consonant merges with the nasal forming a voiced stop. I say merged and not that the stop assimilated

the voicing element of the nasal before the nasal disappeared because we can still see something like

this happening today. I provide two examples below-

The first, geminated consonants, occurs in a similar environment in modern Japanese. If two

free morphemes are attached, or some process requires that a medial vowel is elided, the initial

consonant in the second morpheme will “double”, creating a geminated consonant. This likely has more

to do with preserving mora structure than anything else, as despite the fact that the syllables are now

technically fewer, the mora count remains the same. The trick is, however, that this only occurs with

voiceless initial consonants. If a voiced consonant is in a position to become geminated, there instead

appears an epenthetic generic nasal. Further, there are certain conjugation patterns that force a voiced

stop to come in contact with a voiceless one, such as the past tense of ‘fly’ tob- (Shibatani 1990). When
Barakadyn |5

the past tense suffix -ta is attached, something interesting happens. The coda /b/ geminates, but CV-

style syllable structure must be maintained4, and so the first /b/ loses its consonantal status5, which

turns it into a nasal. Further assimilation (of place) renders the final surface form tonda ‘flew’.

In addition to the previously mentioned medial voiced/voiceless doublets, we can also find

doublets with -m- and -b- (Martin 1987) like hibo : himo ‘cord’ and samishii : sabishii ‘lonely’. These exist

side-by-side, and the original forms of these words are a mixed bag- some are originally -m- and some -

b-, this is not a case of sound change in any particular direction. Shibatani also mentions a study

(Kindaichi 1942) which pointed out the decline of the velar nasal (previously in complimentary

distribution with [g]), the nasal being the medial, or intervocalic, allophone. The trend implied an

eventual total loss, but for a while, the two sounds were unpredictable in their distribution and even

today the distinction between a nasal and non-nasal [g] is a bit fuzzy, and not entirely dependent on

dialect.

Considering all of the above, it is reasonable to assume that voicing and nasality in Japanese are

inextricably linked, leading me to assert an UR for voiced consonants of something like [nasal + voiceless

stop]. The only exception to this would be those consonants which have no voiceless counterpart,

specifically the flap and the glides. The affricates and fricatives (apart from /s/) are of course allophonic

variants of the stop series6, so the UR being based on a voiceless stop specifically is not a problem. This

proposed UR falls in line with what we know about how voicing developed in the course of the history of

Japanese, as briefly mentioned above with post-nasal voicing.

4
Japanese does not allow a consonantal coda, apart from the nasal, which consists of 1 mora. This results in
acceptable syllable types V, CV, or CVN.
5
I am of course not claiming that nasals are not consonants, but there is some quality or feature of nasals which
allows for them to appear in the coda in Japanese. It could perhaps be the +sonorant feature, as the only other
non-vocalic sonorants in the language are the glides /w/ and /j/, but note that neither of them appear in the coda.
6
Again, the exception being /h/ which has no voiced counterpart, being the reflex of the original *p
Barakadyn |6

Rendaku

At its most basic, rendaku is a phenomenon that occurs when two morphemes form a

compound, and the unvoiced initial in the second half of the compound becomes voiced. For example:

te ‘hand’ + kami ‘paper’ -> tegami ‘letter’; na ‘name’ + takai ‘high’ -> nadakai ‘famous’. This should be an

easy process, presumably simply intervocalic voicing of an unvoiced element (something we see often in

many languages), but there are actually many exceptions to this rule. The most famous and well-

explained exception to rendaku is a blocking condition detailed in Lyman’s Law.

Lyman’s Law explains that should the second half of the compound already have a voiced

element within it, then rendaku is blocked. Yama ‘mountain’ + kado ‘gate/door’ -> yamakado

(*yamagado) ‘mountain gate’. Recall earlier, that in Yamato words it is extremely rare to find multiple

voiced obstruents within a single morpheme. This might be an effect of the Obligatory Contour

Principle7, but whatever the reason, this native aversion to ‘too many’ voiced sounds is prevalent and

far-reaching in its effects. Nasals do not count as a voiced obstruent8 in terms of blocking rendaku, with

the exception of some words with an internal [m]. These words correspond with those listed earlier

which have an alternation with -m- and -b-, presuming an underlying -b- or more likely -mb-. Lyman’s

Law, however, has its limits and does not tell the entire story of this sequential voicing phenomenon.

There are exceptions to every rule, but as a general statement rendaku can really only be said to

apply to Yamato words. Words of Chinese loan (Sino-Japanese words) are sometimes seen to undergo

sequential voicing, but typically only when they are considered so integral to language and society that

7
I hesitate over this because the prohibition is on more than just roots, and morphemes, but extends to
compounds as well, as we will see. This does not mean it ISN’T OCP, but that is not the direction of this paper.
8
This shouldn’t be unexpected, as nasals are considered sonorants, but they are also oral occlusives, therefore
behaving in some ways as obstruents.
Barakadyn |7

most native speakers think of them as native words. Even rarer still is when a properly foreign (non-

Asian) loan gairaigo undergoes this change. There are extremely limited examples, but they include

karuto ‘card’ (from Portuguese) in the compound irohagaruto ‘iroha (alphabet) card’. These words were

borrowed very long ago, and have completely assimilated into the Yamato class, thus allowing for

conformity to the conventions of rendaku (Labrune 2012). Another word class which typically resists

rendaku are mimetics, onomatopoeic phrases that are typically reduplicated forms, though this

reduplication shouldn’t directly interfere with the processes as reduplicated forms regularly undergo

voicing where the straight forward version of Lyman’s Law does not interfere- toki ‘time’ -> tokidoki

‘sometimes’, hito ‘person’ -> hitobito ‘people’.

In addition to word-class considerations, only certain types of compounds will bow to the whims

of rendaku. The most obvious (and potentially easiest to explain) are noun-noun (N+N) compounds.

Most experts agree it is likely that sequential voicing is the result of some intervening nasal. The reason

N+N compounds are seemingly so transparent is because the attributive or genitive marker no (used to

string nouns together) could be determined to be underlying in these compounds. There is a subset of

N+N which will not voice, but these compounds are combinative, not qualified. The most notorious

example is that of yama ‘mountain’ + kawa ‘river’ -> yamakawa ‘mountains and rivers’ but yamagawa

‘mountain river’. Additionally, Object-V(erb) compounds will not voice. The accusative particle wo has

no nasal and as such cannot effect voicing, but in Instrument+V compounds we do get rendaku

(Labrune). An example of one of these object/instrument pairs: e ‘picture’ + kaki ‘write/draw’ -> ekaki

‘painter (picture drawer)’, but hude ‘brush’ + kaki -> hudegaki ‘writing with a brush’. This is easy to

explain, knowing that the particle used to mark instrument also contains a nasal! Modern Japanese (MJ)

de < Classical Japanese (CJ) nite. The expected post-nasal voicing has occurred to give us the Modern

Japanese version (as seen above with the past tense-example): nite > n()te > nte > de. Additionally,

resultative compounds often (though not always) undergo rendaku, and ni can function as a resultative
Barakadyn |8

particle. V+V compounds are also resistant to rendaku, which follows the pattern we are illustrating as

verbs are connected without any intervening particle. Those V+V compounds which do undergo voicing

are considered to be backformation or analogy (Nishiyama 2008).

It is worthy to note that Labrune proposes evidence that length is a factor in determining

whether an element will undergo voicing or not- if either part of the compound is long (defined as more

than two morae), things that might otherwise have a habit of resisting the change will go ahead and

voice, even in the event that there might be a blocking element: nawa ‘rope’ + hashigo ‘ladder’ ->

nawabashigo9 ‘rope ladder’. He offers no explanation, though this could be another side effect of OCP if

that is in fact what is limiting the number of voiced stops present in a single word. Perhaps the required

distance between these voiced stop consonants is simply more than two morae.

Conclusion

If we take the position that rendaku is simply a native process of voicing effected by a stop

merging with a nasal, this can explain why foreign words (both Sino- and gairaigo) resist the process.

The analytic nature of Chinese compounds means that while morph(eme)s are often created by

combining two elements to create new-but-related concepts, there is no intervening element. Within

the Chinese languages exists the very same compound ‘rivers and mountains’, represented

(pictographically) the exact same way, and it too can mean ‘a mountain river’, depending on the

context. There is simply no intervening element to tell us if we are looking at an additive compound, or

not. Japanese, being so highly agglutinative, is no stranger to compounding, but it marks with particles

to tell us how content elements relate to each other. It is these markings, even when they disappear,

9
nawa is comprised of two morae, while hashigo equals three. In this example hashigo is the “long” element.
Barakadyn |9

which give us a clue to the processes examined here. Further analysis would be required, a proper

survey of words to test for underlying forms and historical developments, but it could just be that

rendaku is as simple as post-nasal voicing (a process well understood in historical Japanese) applied to

native (or thought to be native) words, and Lyman’s Law simply details an already established historical

phenomenon of dissimilation.

Potential problems with this analysis include the fact that one would expect phonological

processes to take effect in all environments- just because sei ‘sex’ is a Sino- reading, there should be no

reason that rendaku would not take effect and create *nanzei ‘what gender?’ < nan ‘what’ + sei, except

it doesn’t because sei does not belong to the Yamato class. This analysis also does not directly address

Labrune’s assertions regarding length of compound having an effect, although we perhaps have an

explanation for that using OCP.


B a r a k a d y n | 10

References

Hibbett, Howard., and Gen Itasaka. 1965. Modern Japanese: A Basic Reader. Harvard-Yenching Institute.

Labrune, Laurence. 2012. The Phonology of Japanese. Oxford: OUP Oxford.

Martin, Samuel Elmo. 1987. The Japanese Language Through Time. Boston: Yale University Press.

Nishiyama, Kunio. “V-V Compounds.” (2008): n. pag. www.oxfordhandbooks.com. Web. 8 Dec. 2016.

Shibatani, Masayoshi. 1990. The Languages of Japan. Cambridge University Press.

Shirane, Haruo. 2013. Classical Japanese: A Grammar. Columbia University Press.

Vovin, Alexander. 2003. A Reference Grammar of Classical Japanese Prose. Psychology Press.

You might also like