Repetition and its Avoidance: The Case of Javanese

Download 166.87 Kb.
Date conversion05.05.2016
Size166.87 Kb.
  1   2   3
Repetition and its Avoidance: The Case of Javanese
Moira Yip
University of California, Irvine
It is argued that echo-words result from the tension between a requirement that penalizes a sequence of two identical stems, *Repeat(Stem), and one that requires two identical stems, Repeat(Stem). Based primarily on data from Javanese, I make three main points. First, at least some inputs to the Optimality Grammar must be abstract morphological specifications like Plural. They are phonologically incomplete outputs of the morpho-syntax. Second, morpheme realization results from an attempt to meet output targets in the form of constraints: Repeat, 2 =a; Pl=s, and so on. Such morphemes do not have underlying forms in the familiar sense (cf Hammond 1995, Russell 1995). Third, the target constraints may be out-ranked by phonological constraints of various kinds, particularly constraints against the repetition of elements, here called *Repeat. The elements may be phonological (feature, segment) or morphological (affix, stem). These findings support the view of Pierrehumbert (1993a) that identity has broad cognitive roots. The primary data comes from Javanese, but the paper also touches on English and Turkish.1 Section 1 gives some background on the handling of morphological data in OT. Section 2 discusses identity avoidance in morphology, sets out the basic proposal, and gives sketches of English and Turkish. Section 3 is an extended discussion of Javanese. Section 4 looks at secret languages, and section 5 sums up.
1. Blurring of morphology/phonology boundaries in Optimality Theory:

Optimality Theory (Prince and Smolensky 1993, McCarthy and Prince 1993a, and a host of others) as currently conceived is a theory of not only phonology but also many aspects of morphology. It includes mechanisms for controlling the size and content of reduplicative morphemes, constraints responsible for the precise placement of affixes, constraints that explain the choice among allomorphs, and constraints that pick the right member of a suppletive set. The essence of Optimality Theory is that it is an output-based grammar in which all possible outputs for some input are assessed by a universal set of ranked and violable output constraints. The optimal candidate is evaluated as follows. All outputs which violate the highest ranked constraint are thrown out, and those remaining are evaluated by the next highest ranked constraint. This procedure continues until only a single candidate survives. In the event of a tie at any point in the procedure, the tying candidates are passed on down to the next constraint, which decides matters. Let us see how a selection of morphological phenomena is handled in Optimality Theory.

1.1 Reduplication:

McCarthy and Prince 1993a, 1994 lay out an approach to reduplication in which an abstract input morpheme, RED, passes through GEN and results in a set of output candidates in which RED is realized as a full or partial copy of the base. The choice among these candidates is governed by a set of constraints that determine the size of the reduplicant (such as RED= ), and its content, controlled by a set of constraints that enforce identity between base and RED, and prefer maximal copying. If RED= is ranked above Max(imality), the reduplicant will be monosyllabic (e.g. Ilokano bas-basa, If Max is the higher-ranked of the two, reduplication will be total. (e.g. Yoruba agba-agba, oru-oru) These constraints can also interact with syllable structure constraints. For example, if NoCoda dominates Max, the reduplicant will be coda-less (e.g. Balangao: tagta-tagtag, tayna-taynan).
1.2 Affix placement:

A family of Alignment constraints (McCarthy and Prince 1993b) aligns the edges of prosodic and morphological categories with themselves and with each other. A purely phonological alignment phenomenon would be the placement of feet at the ends of prosodic words: Align-Left: (PrWd, Foot) (e.g. English (Táta)ma(góuchi), * Ta(táma)(góuchi)). A purely morphological example would be the placement of an affix at the beginning of a stem: Align-Left(Affix, Stem) ( Tagalog prefix ag-). A morphology/phonology interface example would place a foot at the end of a root, Align-Right (Root, Foot) as in Indonesian bi(cará)-kan (Cohn and McCarthy 1994)

Particularly striking results come in the treatment of affixes that vacillate between prefixation and infixation as a consequence of the interaction between these alignment constraints and syllable structure constraints. For example, in Tagalog the prefix um- of um-aral is placed in position by a constraint Align-Left (um-, stem), but this constraint is dominated by No-Coda. The result is to force infixation of um- before C-initial roots, such as gr-um-adwet, since the prefixed form *um-gradwet would have an extra coda.2 Zoll (1994) has used a similar approach to explain the behavior of morphemes that surface as floating features at varying positions in the root, like Chaha imperative palatalization; in some cases these moveable affixes may surface as independent segments as well, like Yawelmani suffixal glottalization.
1.3 Allomorphy:

Mester (1994) proposes that a prosodic selection process in the lexicon can pick one allomorph from a set of alternatives by looking at which would form the optimal output with respect to a set of constraints. He studies Latin perfect stems, which can be formed by attachment of either -u- or -s-. The default choice is attachment of -u, e.g. mon-u-i:, but in stems with final heavy syllables, -s is used instead: e.g. auk-s-i: *aug-u-i:. He suggests that this can be understood as the avoidance of an output in which a single light syllable, .u., cannot be incorporated into a foot because it is 'trapped' between two heavy syllables (one from the root, and one from the final suffix); here I mark foot boundaries with [ ]:

(1) *aug u i: cf auk si:

[] [] [] []

The prosodic selection process thus picks auk-si: from a set of alternative outputs {aug-u-i:, auk-s-i:} for the input /auk-PERF-i:/.
1.4 Suppletion:

Tranel (1994) examines French determiners, where the feminine 1-sg-poss ma is replaced by the masculine mon before vowel-initial feminine nouns, and the masculine ce 'this' is replaced by the feminine cet before vowel-initial masculine nouns.3 Standard accounts simply stipulate this distribution, but Tranel's insight is that both suppletions supply an onset for the following syllable, and are thus phonologically driven. He suggests that a suppletion set is judged against constraints that require gender agreement, and onset satisfaction, and that no one form is basic. Gender agreement can be over-ridden by the need for an onset: Onset >> Gender. The result will be, correctly, that the grammar will pick the C-final candidate before a V-initial noun, irrespective of gender.




a. monM abbéM

b. maF abbéM



c. monM armeF


d. maF armeF


e. ceM abbéM


f. cetF abbéM


g. ceM armeF



h. cetF armeF

These results make it hard to identify a clear dividing line between morphology and phonology. What is more, they go much further to blur the distinction than does the interleaving of phonology and morphology found in lexical phonology. In lexical phonology, each component has its own character: the entities are different, and the rules are different. In Optimality Theory, this is not necessarily the case. Alignment is the most striking example. Alignment appears to play a role in pure morphology, in pure phonology, and at the interface.

In this paper, I want to focus on another area in which phonology and morphology appear to overlap, the area of identity avoidance. It is a commonplace in phonology that sequences of adjacent identical elements are avoided, and this is enshrined as the Obligatory Contour Principle, or OCP (Leben 1973, McCarthy 1986, Yip 1988, Odden 1988, Myers 1993, Pierrehumbert 1993a, and others). What has received less attention in OT are superficially similar cases in morphology, although the generative literature includes many such cases. See for example Stemberger 1981, Menn and MacWhinney 1984, Hyman and Mchombo 1992.

2. Identity Avoidance in Morphology

Avoidance of identity in morphology takes several forms. I will divide them into four categories. () a. The same morpheme cannot appear twice in the same word

b. Different but homophonous morphemes cannot appear in the same word, or otherwise adjacent in the sentence

c. Homophonous morphemes cannot appear on adjacent words

d. The output of reduplication cannot be total identity
The first type is rare, perhaps non-existent, but it is not clear that the morpho-phonology underlies this: in general it seems likely that syntactic and morpho-syntactic principles will achieve this end without identity avoidance being involved at all.

The second type is quite common; the references cited above include numerous examples. A familiar and typical example is the English possessive plural: *cats's, cats'. Further examples include Mandarin perfective le and Currently Relevant State le (Chao 1968, Li and Thompson 1981), Classical Greek determiners (Golston 1994) and Mandarin third person pronoun ta (Yeh 1994). A common response in these cases is omission of one morpheme, with the remaining one carrying the semantics of both. This phenomenon is called haplology.

The third type involves identical morphemes attached to adjacent words, but where the morphemes themselves are not string adjacent. Since the presence of a morpheme on one word does not satisfy the requirements of the second word, omission of a morpheme is rarely the preferred strategy for resolving the situation; instead we are more likely to see syntactic movement, replacement by an alternative morpheme, or simple blocking. Cases of this kind include English -ing (see Ross 1972, Milsark 1988, and, for a different view, Pullum and Zwicky (1991)), and Hindi -ko Dative and Accusative markers (Mohanan 1992). In the Hindi example, sequences of two NP's, each marked by the suffix -ko, are avoided. For a discussion of cases of types two and three, see Yip (forthcoming).

The fourth type is the primary focus of this paper. I will address the phenomenon of so-called "echo words": reduplication accompanied by a small change such that the two halves are not quite identical. English table-shmable is an example of an echo-word. I will propose that these result from a tension between two constraints, one requiring repetition (reduplication) and one banning repetition (identity-avoidance). I will begin with an overview, then I will use English and Turkish as brief illustrations of aspects of my proposal. Finally, a complex example of identity avoidance in Javanese echo-words will be discussed at some length.

2.1 A Summary of the Proposal:

The central theme of this paper is the avoidance of complete identity. In phonology the OCP has been the usual way of addressing such issues, but the term OCP becomes less useful when one looks at morphology. Firstly, carrying the term over into morphology tends to imply that morphology is a sub-branch of phonology, but this is of course not true; it is more that both phonology and morphology are subject to a single general principle that avoids repetition. Secondly, talking of "contours" in the domain of morphemes is inappropriate. In the rest of the paper I will use constraints of a family I will call *Repeat, as defined sweepingly below.4

() *Repeat: Output must not contain two identical elements
Like many constraints, this is subject to adjacency effects, and it may also be judged gradiently at a featural level. The consequence is that violations will be more serious the nearer two things are, and the more similar they are. In most of this paper these subtleties will play no role.

The model I am proposing has two main parts. I outline the proposal below; further details will become clear during the body of the paper. First, there is a set of UG constraints:

() Repeat: Output must contain two identical elements

*Repeat: Output must not contain two identical elements

*Repeat(Input): Output must not contain elements identical to input

MorphDis: "Distinct instances of morphemes have distinct contents, tokenwise" (McCarthy and Prince 1995:67)
The Repeat constraint forces reduplication by self-compounding. Instead of supposing that there is an affix, RED, which must be filled, it assumes that the input has only a morphological annotation such as "plural", and the grammar includes a constraint RepeatPlural which must be satisfied for all inputs. This can be combined if necessary with constraints governing the size of the reduplicant: I will have nothing to say about this latter point.

The *Repeat constraint blocks complete repetition.5 If *Repeat >> Repeat, we derive the echo-word pattern: reduplication that falls just short of complete identity. This proposal explain why echo-words seem to be most common in the case of word reduplication, where the reduplication would otherwise be total: in the case of partial reduplication, *Repeat is satisfied anyway by the failure to copy the entire base. The ranking Repeat >> *Repeat will mask the effects of *Repeat completely, giving total reduplication.

The *Repeat(Input) constraint is a very particular kind of *Repeat constraint in that it compares input and output, not two portions of the output. It penalizes any output that fully realizes the input. As such, it selects highly opaque outputs. This constraint is necessary to explain language game data, and it is presumably low-ranked, perhaps even absent, from normal grammars. Like ordinary *Repeat it is held in check by lower ranked Repeat, which encourages reduplication, so we find that each half of the output is made minimally different from the input.

Summarizing, the tension between these first three constraints gives rise to the following partial typology:

() Repeat >> *Repeat true reduplication

*Repeat >> Repeat echo words: change in one half

*Repeat(Input) >> Repeat secret languages: change in both halves

The fourth constraint, MorphDis:"Distinct instances of morphemes have distinct contents, tokenwise", is drawn from McCarthy and Prince (1995). They find the need for a constraint that is violated any time a segment does double duty to fulfil more than one morphological role. If this constraint dominates *Repeat, sequences of homophonous morphemes will be acceptable. If the ranking is reversed, however, we will observe haplology: to avoid repetition, one set of segments is recruited to do the work of two morphemes:

() *Repeat >> MorphDis haplology

MorphDis >> *Repeat no haplology
Repeat bears obvious similarities to various constraints proposed in the OT literature on reduplication, particularly McCarthy and Prince (1993, 1994, 1995). It does the work of two constraints in their 1995 paper, Ident-BR, and Max-BR. For the purposes of this paper, it is sufficient to merge these two into the single Repeat.

The second part of the proposal is given below:

() a. Inputs consist of morphologically annotated roots, rather than roots with phonologically specified affixes: /kætPL /, not /kæt-s/, and /udanHAB-REP/ not /udan-RED/.

b. These are realized in order to satisfy specific output constraints
This is very similar to proposals of Hammond (1995) and Russell (1995). It is also what seems to be assumed by Mester (1994), although not in an OT framework. The primary advantage of this proposal in the present context is that it allows for the absence of an affix precisely when some other affix or the root itself is able to satisfy the output constraint in question. The discussion of English in the next section will illustrate this point, and it will play an important role in Javanese.
2.2 English 's:

The best known case of haplology comes from English. The plural /s/ and the possessive /s/ cannot co-occur, although adding possessive /s/ to an irregular plural is fine, and so is adding it to a singular ending in /s/, or even a singular ending in /s s/..

() Singular Plural Possessive Sg. Possessive Pl.

child children child's children's

mouse mice mouse's ?mice's

cat cats cat's cats' *cats's

Katz Katzes Katz's Katzes' *Katzes's

coreopsis coreopsis's

Compare especially Katz's vs *cats's; coreopsis's vs. *Katzes's. Two strategies are used to avoid /s-s/. One strategy is haplology: the omission of a morpheme, as in the possessive plural cats'. The other is insertion of a buffer vowel, as in the simple possessive Katz's, Kat[s z] (and between all stridents and suffixal -s).

Optimality Theory, as an output-based grammar, is well-suited to capturing Stemberger's (1981) insight that this and other cases of haplology do not appear to involve deletion so much as a failure to insert a superfluous morpheme if a homophonous morpheme is already in the right position. Thus if the plural /s/ is present, a plural possessive can satisfy the need to end all possessives in /s/ without adding a second /s/. This explanation, though, does not extend to the vowel-insertion between a root /s/ and a suffix /s/, and thus no unified explanation is possible. Within Optimality Theory, we can provide a single straightforward account.

I will now offer an explicit Optimality Theory analysis of the core aspects of identity avoidance, using this as my first example. Suppose, following Myers (1993) , that *Repeat (ie his OCP) is a constraint that can be ranked with respect to the other constraints of the grammar. Further suppose that *Repeat is a sort of meta-constraint (Pierrehumbert 1993b) which can be instantiated with different arguments, and includes at least the following family;
() *Repeat (feature) *Repeat (segment)

*Repeat (affix) *Repeat (stem)

Consider a case in which insertion is the preferred remediation strategy. *Repeat must then dominate some sort of constraint against epenthesis which, following Prince and Smolensky 1993, I will call Fill. Also high-ranked will be the output constraints that require some morphological category to be phonologically instantiated in a particular way:
() English 's:

a. Plural: Plurals must consist of a stem plus an -s affix.

b. Poss: Possessives must consist of a phrase plus an -s affix.

c. *Repeat (s): *Repeat (feature), where feature=[strident]

d. Fill: Don't insert
Plural, Poss, *Repeat(s) >> Fill (Epenthesis as last resort)
In the tableau below, the possessive plural cats' with only one s wins because the candidate with two s's violates *Repeat(s), and the candidate with epenthesis violates Fill. Crucially, the single s satisfies the Plural and Poss constraints.












In the possessive of Katz's, the affix must be retained to satisfy Poss. Fill is thus violated in order to satisfy the higher-ranked *Repeat(s).













These tableaux demonstrate that the omission of one affix after the possessive plural of cat versus the epenthesis into the simple possessive of Katz follow from the dominance of *Repeat(s), and of the output requirement that the plural and the possessive must end in an 's morpheme. This output requirement blocks deletion of a lone plural or possessive morpheme, and *Repeat(s) forces use of the fall-back strategy, epenthesis. Two 's affixes will never be optimal, because they will always violate either *Repeat(s), if adjacent, or Fill, if separated by an epenthetic vowel, and there is always available a candidate with only one affix that violates neither. This analysis thus allows us to link the morphological "haplology" of the plural and possessive morphemes with the phonological epenthesis of the English Plural Rule by assuming that *Repeat(s) plays a role in both "components".6

English demonstrates the advantages of assuming that affixes are not present underlyingly, but are a response to satisfying an output constraint. In the next section we will see the role of *Repeat when it interacts with reduplication.

  1   2   3

The database is protected by copyright © 2016
send message

    Main page