LINGUIST List 16.795
Wed Mar 16 2005
Review: Phonology/Morphology: Ito & Mester (2003)
Editor for this issue: Naomi Ogasawara <naomilinguistlist.org>
What follows is a review or discussion note contributed to our Book Discussion Forum. We expect discussions to be informal and interactive; and the author of the book discussed is cordially invited to join in. If you are interested in leading a book discussion, look for books announced on LINGUIST as "available for review." Then contact Sheila Collberg at collberglinguistlist.org.
Japanese Morphophonemics: Markedness and Word Structure
Message 1: Japanese Morphophonemics: Markedness and Word Structure
From: Michael Barrie <mike.barrieutoronto.ca>
Subject: Japanese Morphophonemics: Markedness and Word Structure
AUTHORS: Ito, Junko; Mester, Armin
TITLE: Japanese Morphophonemics
SUBTITLE: Markedness and Word Structure
SERIES: Linguistic Inquiry Monographs 42
PUBLISHER: MIT Press
Announced at http://linguistlist.org/issues/15/15-991.html
Michael Barrie, Manami Hirayama, Sara Mackenzie and Kenji Oda, University
Ito and Mester (2003) (hereafter I&M) provide a comprehensive analysis of
the phonology of voicing in Japanese. Their analysis focuses on the well-
known morphophonological process known as rendaku (or sequential voicing)
in which the initial segment of the second member of a compound is voiced:
a. /kaki/ 'persimmon' -> hosi-gaki 'dried persimmon'
b. /tana/ 'shelf' -> hon-dana 'book shelf'
There are two kinds of restrictions observed with regard to the
phenomenon: One is that the process cannot occur when the second member
already contains a voiced obstruent ---this restriction is widely known
as "Lyman's Law."
(2) /kaze/ 'wind' -> kami-kaze (but *kami-gaze) 'God's wind'
The other restriction is that the process applies only to the native
vocabularies and some of the vocabularies of old loanwords from Chinese.
Thus, the recent loans from Western languages do not show rendaku:
(3) kappu 'cup' -> koohii-kappu (but *koohii-gappu) 'coffee cup'
Both restrictions are taken up by I&M and analyzed using constraint
conjunction and differentiation of faithfulness constraints according to
particular strata of the Japanese lexicon.
Chapter 1 outlines the structure of the book, listing both the empirical
domain of investigation and the theoretical issues which are addressed
throughout. In terms of empirical phenomena, this monograph considers the
morpheme structure constraint that allows only a single voiced obstruent
to occur within native roots and rendaku and the interaction of these
phenomena as seen in Lyman's Law, illustrated in (2) above. Other areas
considered in the text include diachronic variation and lexical
stratification. Theoretical issues that will be considered throughout the
book, including self-conjunction of markedness constraints and the use of
faithfulness constraints which make reference to particular lexical strata
are introduced in this chapter.
Chapter 2 provides theoretical and empirical foundations for the rest of
the book. First, I&M question the validity of autosegmental Obligatory
Contour Principle (OCP) analyses of dissimilative effects. I&M propose
that dissimilative effects should instead be analyzed as a case of
crossing a markedness threshold; that is, a grammar finds it undesirable
to violate a markedness constraint twice in a certain domain while it
tolerates a single violation. In order to express the notion of threshold
while preserving the strict ranking system of OT, I&M employ the mechanism
of local constraint conjunction.
The second purpose of chapter 2 is to introduce a markedness threshold
analysis of the morpheme structure constraint barring multiple voiced
obstruents from occurring in Japanese roots. I&M provide data showing that
no Yamato (native) morpheme contains more than one voiced obstruent
although voicing itself is contrastive in the language. The observation
that voicing is contrastive in Japanese generally indicates that the
constraint prohibiting the occurrence of voiced obstruents, which I&M call
NO-D, is ranked below the IDENT constraint requiring faithfulness to input
values of [voice]. However, I&M take the ban on multiple voiced obstruents
in native morphemes as evidence that the self-conjoined version of NO-D
(NO-D^2) dominates IDENT, providing the ranking NO-D^2 >> IDENT >> NO-D.
Finally, I&M point out that multiple voicing restriction of Japanese is
not exceptionless but that the multiple voicing restriction is easily
violated by foreign and ideophonic items. I&M state that this does not
reflect a historical fact about Japanese phonology, but rather indicates
that certain synchronic phonological restrictions are active only in some
strata of the Japanese lexicon.
Chapter 3 extends the self-conjoined markedness analysis of OCP effects to
a wider range of phenomena. These include dissimilative degemination in
Latin, Amharic and Japanese as well as deaccentuation in Japanese
compounds. In all cases, I&M argue that the autosegmental OCP is ill-
equipped to handle these phenomena. In the degemination cases, an
autosegmental analysis is not available due to the common assumption that
the OCP is active only within the level of segmental features and does not
apply to prosodic properties such as length. The markedness threshold
approach advocated by I&M, on the other hand, can be extended to these
cases straightforwardly through the use of a self-conjoined markedness
constraint *Geminate^2. This chapter also includes a discussion of some
theoretical implications of constraint conjunction, both problematic and
promising discussing, in particular, Spaelti's (1997) Universal Conjoined
Constraint Ranking Hypothesis.
Chapter 4 discusses the morphological and phonological aspects of rendaku
upon which I&M formulate their analysis. After explaining how OCP effects
are captured by self-conjoined markedness constraints which outrank
relevant faithfulness constraints in the previous chapters, I&M apply this
approach to rendaku voicing. They argue first that rendaku voicing is the
result of a voicing morpheme that links the two sub-components of the
compound, and that this morpheme is preserved in rendaku compounds by a
constraint that requires this morpheme to be realized. They then discuss
various morphological considerations for rendaku, such as the difference
between word compounds and root compounds, and why rendaku is active only
in the former. With the morphological facts out of the way, the authors
implement their proposal described above (self-conjoined faithfulness
constraints and compound morphemes) into an OT analysis of rendaku,
accounting for the effects of Lyman's Law.
Chapter 5 discusses the issue of domains in the patterning of OCP effects
and voicing alternations in Japanese. I&M first address generally the
domains of conjoined and self-conjoined constraints. They introduce the
concept of Minimal Shared Domain as the local domain of a conjunction. The
domains of self-conjunction are discussed for the analysis of
dissimilation in rendaku, which involves the self-conjoined markedness
constraint on voiced obstruents, No-D^2. Looking at diachronic change, I&M
show that more than one domain is possible for dissimilation; prosodic,
specifically the Prosodic Word (PW), in Old Japanese and
grammatical/lexical, specifically the morpheme, in Modern Japanese.
Moreover, they analyze this variation as the result of diachronic
reranking of the PW-domain version of No-D^2 in the hierarchy when the
hierarchy also has the morpheme-domain version of No-D^2. Furthermore,
they consider a prediction made for the conceivable rendaku-type voicing
patterns. The possible patterns can all arise from the hierarchy developed
in their analysis, while the hierarchy successfully excludes an unattested
and impossible pattern. They argue that because their analysis can
necessarily prevent the impossible pattern in this way, their approach is
superior to others that put the base on linear precedence relations.
Chapter 6 asks for a mechanism that can account for the fact that certain
constraints are operative in certain vocabulary groups but not in others:
what does a grammar look like that can generate such lexical variation?
The key is the FAITH stratification model: an individual grammar fixes a
particular markedness hierarchy and the stratum-specific faithfulness
constraints insert themselves in the hierarchy in different places. I&M
start off with the general typological markedness law, or implicational
relationship, found in language inventories as a place where the proposed
model can explain the variation. They also illustrate the same model with
patterns stratified sociolinguistically by different registers. With
particular focus on Japanese, the stratum variation in voicing is
discussed. The authors demonstrate that the faithfulness model can explain
the lexicon-internal variation as a kind of faithfulness variation; here
the faithfulness constraints are from the family of IDENT, specified for
designated lexical strata, and inserted at specific points in the
The empirical foundation of chapter 7 concerns faithfulness in voicing as
it applies to rendaku voicing; however, I&M use the discussion here to
comment more deeply on the issue of faithfulness in OT in general. The
first section concerns the use of the faithfulness constraint IDENT [VOI]
which the authors take to apply only to obstruents. They then ask whether
IDENT [VOI] is asymmetric in the sense that voicing of a voiceless segment
is equally as punishable as devoicing of a voiced segment. I&M provide
data from Old Japanese to show that devoicing of a voiced segment
constitutes a more serious violation than the inverse scenario. They
explore the possibility that IDENT [VOI] can be split into IDENT [+VOI],
which outranks the more general IDENT [VOI], not the expected polar
opposite constraint IDENT [-VOI]. I&M conclude, however, that symmetric
IDENT is superior (i.e., referring to both positively and negatively
valued features). To capture the asymmetric faithfulness facts, they
propose that IDENT constraints must be conjoined with markedness
Chapter 8 addresses the pattern of rendaku found in complex compounds
(complex compounds defined as compounds which contain compounds as
members). I&M begin with the observation that in left-branching compounds
rendaku applies iteratively, voicing the initial segment of each member.
For example, /hosi/ 'dried', /kaki/ 'persimmon', /tukuri/ 'making', when
combined are realized as [hosigakizukuri] showing rendaku voicing on
both /kaki/ and /tukuri/. In right-branching compounds, however, rendaku
voicing does not apply at the juncture between the subcompound and the
additional constituent. For example, /hatu/ 'first', /kao/ 'face', /awase/
meeting', when joined in a compound are realized as [hatukaoawase] without
rendaku voicing on the initial /k/ of /kao/. I&M seek an account of these
facts that does not require reference to syntactic information such as
direction of branching.
First, they show that an analysis relying on output-output constraints,
while initially appealing, is unworkable. They propose an alternative
analysis which makes reference to prosodic structure. They note that the
compounds in which rendaku applies regularly differ from the compounds in
which rendaku is suppressed at the major constituent juncture in that the
former are always a single accentual domain whereas the latter may contain
two accents and are hence considered to be two prosodic words. They then
derive the pattern of rendaku from self-conjunction of the constraint
ANCHOR-L. This constraint penalizes multiple violations of a constraint
requiring the left edge of a grammatical word to correspond to the left
edge of a prosodic word. Appropriate ranking of this constraint requires
right-branching compounds to be parsed as two prosodic words whereas left-
branching compounds can be parsed as a single prosodic word without
violating this self-conjoined constraint. An additional markedness
constraint militating against voiced obstruents at the beginning of a
prosodic word is required to derive the pattern of rendaku. This analysis
is followed by some general consideration of the role of positional
faithfulness constraints and positional markedness constraints in OT. I&M
argue that, although the two types of constraints are equal in many
circumstances, the positional markedness approach is superior in this case.
This monograph is an important addition to the literature on rendaku in
Japanese. It offers a vast empirical coverage of the phenomenon with
abundant examples throughout the text and an extensive appendix. From a
theoretical perspective, this monograph makes important contributions to
the study of constraint conjunction and lexical stratification within OT.
As a whole, the authors offer a plausible analysis of rendaku, offering
several advantages over previous approaches.
The book offers a cohesive and comprehensive analysis of lexical
stratification, as described above. This analysis is particularly
attractive given its ability to relate stratal distinctions to
implicational hierarchies and its ability to account for some groups of
apparent exceptions to exceptions, as in the case of Common Sino-Japanese
words that undergo rendaku. Also, the difference in rendaku strategies
between modern Japanese and Old Japanese is accounted for by a simple
constraint re-ranking - a desirable solution.
Nevertheless, we have a few concerns with the analysis and its
presentation. First, the historical facts are explained by a constraint re-
ranking in which the markedness constraint No-D^2_w is demoted in Modern
Japanese (p. 111ff). This analysis requires further scrutiny from a
broader discussion of issues of diachronic change in OT, particularly in
light of claims that diachronic change tends to involve demotion of
faithfulness rather than markedness constraints (Hayes, 2004 and Uffmann,
Second, it is necessary to explain explicitly what limits the application
of constraint conjunction. I&M's analysis as a whole relies heavily on the
notion in various ways (e.g., self-conjoined markedness constraints for
the analysis of dissimilation and markedness thresholds, conjunction of
markedness and faithfulness constraints to account for asymmetric effects
of voicing and devoicing, and multiply conjoined constraints in their
analysis of degemination in Amharic). I&M acknowledge the potential
problems in learnability and restrictiveness that constraint conjunction
may lead to. They state that distinguishing reasonable from unreasonable
conjoined constraints 'cannot be relegated to the syntax of conjunction'
but rather is an is an issue of 'phonological substance and phonetic
groundedness' (p. 24). While we agree on intuitive grounds with such an
idea, we are left unsatisfied as little discussion of phonetic substance
or formal restrictions on the well-formedness of simple constraints is
provided. Furthermore, I&M mention (p. 265 endnote #23) that no convincing
evidence for No-D^3 has been found, although it is formally derivable, as
NO-D^2&NO-D. While the observation is generally agreeable, it is left
unanswered how such a restriction is naturally deduced from the theory.
Lastly, we would like to mention the learnability issue related to the
proposed faithfulness-based stratal model. In the proposed model, the
burden of the explanation for lexical-class-specific application of
certain rules, such as rendaku, is upon the positioning of class-specific
faithfulness constraints to the fixed hierarchy of antagonistic markedness
constraints. This implies that each lexical item has a 'tag' that
indicates the stratum it belongs to in order to be properly evaluated.
This is indeed true of any model of grammar that refers to lexical strata.
But this may cast a question with regard to learnability: how learnable is
this kind of grammar? The question about learnability is not addressed in
I&M's discussion. Of course this is not a problem particularly for the
analysis that I&M provide, and the organization of the lexicon is a
controversial topic (see Tateishi 2002, for example, for a summary).
In conclusion, this monograph is invaluable both empirically and
theoretically. Empirically I&M continue to set the standards of what must
be accounted for in the phonology of voicing in Japanese. Theoretically
the book is a major contribution to Optimality Theory. Thus, it is highly
recommended to a variety of people who wish to have a descriptive
foundation of Japanese morphophonology and wish to pursue developments in
phonological theory within the framework of OT.
Hayes, Bruce (2004) Phonological acquisition in optimality theory: the
early stages. Ms. University of California. [Also published in Kager,
Rene, Pater, Joe, and Zonneveld, Wim (eds.) Fixing priorities:
Constraints in phonological acquisition. Cambridge: Cambridge University
Spaelti, Philip (1997) Dimensions of reduplication in multi-pattern
reduplication. Doctoral dissertation, University of California, Santa Cruz.
Tateishi, Koichi (2002) Bunpoo no itibu to shiteno goisoo no zehi (Lexical
strata as a part of grammar), Onsei Kenkyu (Journal of the Phonetic
Society of Japan) 6, 34-43.
Uffmann, Christian (2003) Markedness, faithfulness, and creolization: The
retention of the unmarked. In Ingo Plag (ed.), The phonology and
morphology of creole languages. Tübingen: Niemeyer, 3-23.
ABOUT THE REVIEWERS
Michael Barrie, Manami Hirayama, Sara Mackenzie and Kenji Oda are all
graduate students at the University of Toronto. They formed a small
reading group where they discussed this monograph and they all have an
interest in the phonology of East Asian languages.
Respond to list|Read more issues|LINGUIST home page|Top of issue