A Correspondence-theoretic Account of Fixed Segmentism Reduplication

In fixed segmentism reduplication (FSR), reduplication is accompanied by addition of an affix which partially overwrites the reduplicant (the FSR affix). Nevins (2005) claims that the correspondencetheoretic analysis of FSR proposed by Alderete et al. (1999) faces three serious problems: First, it predicts the existence of unattested FSR systems where the FSR affix is backcopied to the base. Second, it predicts unattested FSR systems where the realization of the FSR affix depends on its relative size with respect to the portion of the reduplicant it strives to overwrite. Third, it cannot account for cases where overwriting replaces parts of the reduplicant even though concatenation of the FSR affix and the reduplicant would result in a phonologically licit structure. In this paper, we show that the first problem is empirically flawed since FSR copying is attested, and argue that the second and third problem find a straightforward solution in the independently motivated parametrization of optimality-theoretic constraints. 1. Alderete et al. (1999) on Fixed Segmentism Reduplication Alderete et al. (1999) distinguish phonological and morphological FSR. In the former, a default segment is phonologically motivated whereas in the latter the fixed segmentism is a kind of affix. An example for phonological FSR is diminutive reduplication in Lushootseed. According to Urbanczyk (1996), the fixed segmentism ı́ appears in reduplicants whhich would otherwise contain marked structure such as stressed schwa (the reduplicative prefix is stress-attracting) or a long vowel (1-a,b), but not with other roots (1-c,d): 1-2-many, 303-333 Jochen Trommer & Andreas Opitz (eds.) Linguistische Arbeits Berichte 85, Universität Leipzig 2007 304 Eva Zimmermann & Jochen Trommer (1) Reduplication in Lushootseed (Urbanczyk, 1996) a. t@láw-il ‘run’ t́ı-t@law’-il ‘jog’ b. s-du:k ‘knife’ s-d́ı-du:k ‘small knife’ c. čál@s ‘go ahead’ čá-čal@s ‘go ahead a bit’ d. s-duk ‘bad’ s-dú-P-du:k ‘riff-raff’ These cases are analyzed by Alderete et al. (1999) as phonological Emergence of the Unmarked and are irrelevant for our argument. A notorious example for morphological FSR is English schm-reduplication which expresses roughly derision or irony. In schm-reduplication, the base is copied and schm is realized as the onset of the first syllable of the reduplicant, replacing the original onset of the base if necessary: (2) English Schm-reduplication a. table table-schmable b. plan plan-schman c. string string-schming d. apple apple-schmapple In contrast to cases of phonological FSR, the appearance of schm cannot be analyzed as a result of phonological optimization since this consonant combination is highly marked in English. In the correspondencetheoretic analysis proposed in Alderete et al. (1999), schm is simply taken to be an affix which is attached to the base concomitantly to reduplication. Hence the English case is parallel to the Bambara reduplication pattern in (3) where base and reduplicant are linked by the affixal element o : (3) Reduplication in Bambara (Culy, 1985; Dumestre, 2003) a. wulu ‘dog’ wulu-o-wulu ‘whichever dog’ b. malo ‘uncooked rice’ malo-o-malo ‘whatever uncooked rice’ c. muso ‘woman’ muso-o-muso ‘whatever woman’ Whereas affixation of o generally leads to phonologically wellformed structures in Bambara, only (2-d), based on the vowel-initial base apple, is phonotactically licit in English without further modification. Combining schm and consonant-initial bases (2-a-c) would lead to clusters such as *Smt which are excluded in English by high-ranked markedness constraints. Assuming that epenthesis is not possible, either schm or the onset of the reduplicant must be deleted, and hence compete for realization – a competition which is resolved by the two A Correspondence-theoretic Account of Fixed Segmentism Reduplication 305 faithfulness constraints Max-IO and Max-BR, where the former demands realization of all input material in the output, and the latter requires that all segments of the base also appear in the reduplicant. Reduplication is triggered by the abstract formant RED which consists of no phonological material of its own but whose “content [. . . ] is determined by the base” (Nelson, 2002:321). Thus the input for the OT-grammar consists of the root, the affix schm and RED. The correct English pattern is derived by ranking Max-IO over Max-BR as illustrated in table (4). (4) Analysis: Max-IO Max-BR t1a2b3l4e5-sch6m7-RED Max-IO Max-BR ☞ a. t1a2b3l4e5-sch6m7a2b3l4e5 * b. sch6m7a2b3l4e5-sch6m7a2b3l4e5 *! c. sch6m7a2b3l4e5-t1a2b3l4e5 *!* d. t1a2b3l4e5-t1a2b3l4e5 *!* Nevins’ critique of Alderete et al. (1999) does not directly attack this simple and conceptually attractive analysis of the English data, but identifies two problems with the typological predictions of the assumed constraint set and claims that the correspondence-theoretic analysis is inadequate for a similar FSR pattern in Hindi. We will address all three problems in the following sections. Section 2 and 4 discuss the typological problems with backcopying and size-dependent FSR. The Hindi case is addressed in section 5. Section 3 provides an analysis for related problems in the root-and-pattern morphology of Hebrew. We will show that all alleged complications for a correspondence-theoretic analysis are either empirically untenable or are obliterated by the independently motivated parametrization of optimality-theoretic constraints. In secStrictly speaking, there might be different RED morphemes in a single language resulting in different reduplication patterns (Urbanczyk, 1999). Each distinct RED, i.e. each morpheme which has the RED property, establishes a distinct correspondence relation and is governed by distinct BR faithfulness constraints defined with respect to this relation, which in turn trigger copying. Since we are only dealing with single reduplication processes for any given language, we will skip over these subtleties. Cf. Alderete et al. (1999:356). 306 Eva Zimmermann & Jochen Trommer tion 6 we discuss the alternative approach to FSR advocated by Nevins, and in section 7 we present our conclusions. 2. Morphological Backcopying Because correspondence-theoretic OT allows candidates exhibiting any conceivable modification to the input, one of the possible outcomes in (4) is (4-b), where the FSR affix “backcopies” from the reduplicant to the base. As Nevins correctly points out, this candidate becomes optimal if the ranking of Max-IO and Max-BR is reversed: (5) Analysis: Max-BR Max-IO t1a2b3l4e5-sch6m7-RED Max-BR Max-IO a. t1a2b3l4e5-sch6m7a2b3l4e5 *! ☞ b. sch6m7a2b3l4e5-sch6m7a2b3l4e5 * c. sch6m7a2b3l4e5-t1a2b3l4e5 *!* d. t1a2b3l4e5-t1a2b3l4e5 *!* Since it is one of the foundational tenets of Optimality Theory that – apart from systematic restrictions on possible rankings – constraints can be freely reranked, this combination of FSR and backcopying should be attested in some language. Hence we expect to find a language English" with the backcopying FSR construction in (5). Nevins (2005) classifies this pattern as typologically not attested and takes this alleged gap as evidence for a morphological approach to reduplication as in Raimy (2000) which he claims to be uncapable to derive fixed segmentism backcopying. We will discuss the differences between the correspondence-theoretic and the representational approach in section 6. Here, we show that backcopying of morphological material is indeed attested in the languages of the world and Nevins’ argument is empirically problematic. First, FSR involving backcopying is found in Siroi, a nonAustronesian language of Papua New Guinea (Wells, 1979; Inkelas and Zoll, 2005). In Siroi, adjectives are reduplicated to express plural formation. In addition to reduplication, the fixed segmentism g replaces the onset of the second syllable in disyllabic words (6-a,b) and is infixed A Correspondence-theoretic Account of Fixed Segmentism Reduplication 307 in monosyllabic words (6-c). Crucially, g does not only appear in the reduplicant but also in the base: (6) Reduplication in Siroi (Wells, 1979) a. maye mage-mage ‘good’ b. sungo sugo-sugo ‘big’ c. kuen kugen-kugen ‘tall’ A slightly different case of morphological backcopying can be observed in Seereer-Siin, an Atlantic language analyzed in detail by Mc Laughlin (2000). In Seerer, the first consonant of a noun stem undergoes mutation after specific noun class prefixes. Two patterns of mutation are found, voicing mutation (changing a voiced into a voiceless stop (7-a,b)) and continuancy mutation (changing a continuant into a stop, (7-c,d)). In (7), these mutation processes are triggered by the singular class prefix owhile the plural forms show the underlying root-initial consonant: (7) Consonant Mutation in Seerer-Siin (Mc Laughlin, 2000) Sg Pl a. o-cir éir ‘sick person’ b. o-kawul gawul ‘griot’ Voicing mutation c. o-paâ faâ ‘slave’ d. o-tew rew ‘woman’ Continuancy mutation Consonant mutation interacts with a second process, derivation of agent nouns through reduplication where the reduplicative prefix is truncated to a CV: template (8). The patterns of interest here are the ones in (8-d-g): In contrast to voicing mutation (8-a-c), continuancy mutation affects the initial consonant of the root and applies optionally also to the reduplicant: Interestingly Siroi also violates the claim of McCarthy and Prince (1999) that there is no backcopying of prosodic templates (the “Kager-Hamilton problem”). Independent evidence for prosodic backcopying is found in Guarijio (Caballero, 2006). 308 Eva Zimmermann & Jochen Trommer (8) Reduplication and Mutation in Seerer-Siin (Mc Laughlin, 2000) Voicing Mutation: No Featural Transfer a. bind ‘write’ o-pii-bind ‘writer’ b. dap ‘launder’ o-taa-dap ‘launderer’ c. gim ‘sing’ o-kii-gim ‘singer’ Continuancy Mutation: Optional Featural Transfer d. xoox ‘cultivate’ o-qoo-xoox o-qoo-qoox ‘farmer’ e. fec ‘dance’ o-pee-fec o-pee-pec ‘dancer’ f. war ‘kill’ o-baa-war o-baa-bar ‘killer’ g. riw ‘weave’ o-tii-riw o-tii-tiw ‘weaver’ Following Mc Laughlin (2000) we assume that mutation in Seerer is featural affixation of the features [–cont] and [–voice]. Under this analysis, backcopying in Seerer, although not FSR in the strict sense, is completely parallel to the situation in Siroi: A (featural) affix can only be realized by overwriting a feature specification of the reduplicant ([– cont] replaces [+cont] of the initial consonant) and this change is copied back to the base. Note that a derivational account of these patterns is problematic: One could assume that for the backcopying options in (8), mutation applies first to the base followed by reduplication. But morphologically mutation in these cases applies to nouns, not to verbs, hence the morphological structure of o-baa-bar is as in (9) which implies exactly the opposite ordering of phonological operations: o triggers mutation in the noun derived previously by reduplication: (9) [oClass [RedN [bar]V]N]Class Moreover, the fact that there is no featural transfer for the voicing mutation which can be straightforwardly derived in Correspondence theory by the different ranking of base-reduplicant faithfulness constraints for voicing and continuancy, appears to be a mistery under a derivational account. We conclude that morphological backcopying in FSR and more generally is empirically attested lending support to the correspondencetheoretic approach to FSR which naturally predicts this type of phenomena. Another possible example of morphological backcopying outside of FSR is found in Chumash: A Correspondence-theoretic Account of Fixed Segmentism Reduplication 309 3. Segment-counting Root-and-Pattern Morphology Nevins extends his attack against a correspondence-theoretic account of FSR to Semitic root-and-pattern morphology based on the analysis of Hebrew denominal verb formation in Ussishkin (1999). Since the Hebrew case offers some essential insights which are important for the analysis of FSR we develop in the following sections, we will discuss this case even though we are not primarily concerned with morphological formations outside of reduplication. In a significant subgroup of Hebrew denominal verb formation, base vowels are overwritten by the vowel melody i – e and extended to the size of a bisyllabic minimal word by doubling the second root consonant: (10) Hebrew Denominal Verb Formation (Ussishkin, 1999) a. dam ‘blood’ dimem ‘to bleed’ b. xam ‘hot’ ximem ‘to heat’ c. xad ‘sharp’ xided ‘to sharpen’ d. cad ‘side’ cided ‘to side with’ Intuitively Ussishkin captures this pattern by the assumption that affixal vowels have to be realized inside the base, but since the size of the resulting structure is restricted to bisyllabicity, not all vowels can be parsed. Preference for the realization of affixal vowels is implemented by two separate faithfulness constraints for stem and affix vowels, MaxVowel-Af and Max-Vowel-Stem, ranked in this order. MinWd abbreviates a set of constraints which jointly require that the prosodic word is a bisyllabic foot with a final consonant. Integrity penalizes the doubling of segments: (i) Reduplication in Chumash (Frampton, 2004) a. s-kitwon skit-kitwon ‘it is coming out’ b. s-ikuk sik-sikuk ‘he is chopping, hacking’ c. s-ǐs-expeč sisex-sexpeč ‘they two are singing’ McCarthy and Prince (1995) assign the morphological structure s-RED-Root to these forms (where sis an independent prefix), and argue that the segmental prefix is backcopied to the base with vowel-initial roots to satisfy the requirement that the reduplicant should be a heavy syllable (cf. sik.si.kuk vs. *si.ki.kuk without backcopying). However Inkelas and Zoll (2005) and Frampton (2004) argue against this analysis for Chumash. 310 Eva Zimmermann & Jochen Trommer (11) Denominal Verb from Biconsonantal Base (Ussishkin, 1999) d1a2m3 + i4 e5 MinWd Max-V-Af Max-V-Stem Integrity a. d1a2m3e5m3 *! * b. d1i4m3a2m3 *! * c. d1a2m3i4m3e5 *! * ☞ d. d1i4m3e5m3 * * For roots with a high vowel, there is an alternative which allows to maintain base and affix vowels. The base vowel can be employed as the featurally equivalent glide j in the onset position of the second syllable: (12) Denominal Verb from Glide-medial Base (Ussishkin, 1999) t1i2k3 + i4 e5 MinWd Max-V-Af Max-V-Stem Integrity a. t1i2i4e5k3 *! b. t1i4k3e5k3 *! * ☞ c. t1i4j2e5k3 According to Nevins, a fatal flaw of this move is that it predicts the wrong result for dam. The i of the affix melody could also be used as a glide resulting in dajem (in the following ☛ indicates candidates which are empirically correct, but do not become optimal under the given ranking): (13) Problematic Candidate with Biconsonantal Base (Nevins, 2005) d1a2m3 + i4 e5 MinWd Max-V-Af Max-V-Stem Integrity a. d1a2m3e5m3 *! * b. d1i4m3a2m3 *! * c. d1a2m3i4m3e5 *! * ☛ d. d1i4m3e5m3 *! * ☞ e. d1a2j4e5m3 Nevins attributes this problem to a fundamental problem with Correspondence Theory, namely the implementation of overwriting through constraint evaluation. However, we think that (13-d) is excluded by constraints and techniques which are fairly standard in OT. Note first that although i and j have the same distinctive features, they are not completely identical: i is dominated by a mora while j is not, hence A Correspondence-theoretic Account of Fixed Segmentism Reduplication 311 replacing the former by the latter violates faithfulness since it implies deletion of a mora penalized by the constraint Max-μ: (14) Max-μ: Input moras should have correspondent moras in the output. Moreover, we assume that faithfulness constraints are parametrized in a way which is standard in the optimality-theoretic literature, namely the parametrization of faithfulness constraints to the domains affix and stem, which goes back to the original formulation of Correspondence Theory in McCarthy and Prince (1995): “It must be, then, that correspondence constraints are tied not only to specific dimensions (B-R, I-O, [...]), but also, in some cases at least, to specific morphemes or morpheme classes. Thus, the full schema for a faithfulness constraint may include such specifics as these: [...] the morphological domain (root, affix, or even specific morpheme) to which the constraint is relevant”. (McCarthy and Prince, 1995:17) In Ussishkin’s analysis the parametrization of faithfulness constraints to stems and affixes is applied to the constraintMax-V giving two Max constraints which are ranked differently. We apply the same strategy to all faithfulness constraints, namely Max-μ resulting in the subconstraints Max-μAf(fix) and Max-μS(tem), again with different ranking potential, and in the same way to Integrity. Under the assumption that the vowel melody i – e contains true, i.e. moraic vowels, this gives straightforwardly the correct results. In (15), the stem vowel can be recycled as a glide since Max-μS is ranked below all other constraints, but in (16) Max-μAf which is ranked above IntegrityS blocks turning i into a glide by deleting its mora: (15) Glide-medial Base under Constraint Parametrization t1i2k3 + i4 e5 Max-VAf IntAf Max-μAf Max-VS IntS Max-μS a. t1i4e5k3 *! * b. t1i4k3e5k3 *! ☞ c. t1i4j2e5k3 * 312 Eva Zimmermann & Jochen Trommer (16) Biconsonantal Base under Constraint Parametrization d1a2m3 + i4 e5 Max-VAf IntAf Max-μAf Max-VS IntS Max-μS a. d1a2m3e5m3 *! * * b. d1i4m3a2m3 *! * * ☞ c. d1i4m3e5m3 * * * d. d1a2j4e5m3 *! There are two important points to note: First, this analysis systematically violates a metacondition McCarthy and Prince (1995) have proposed for morphologically parametrized faithfulness constraints, the Root-Affix Faithfulness Metaconstraint : (17) Root-Affix Faithfulness Metaconstraint, RAFM RootFaith AffixFaith (McCarthy and Prince, 1995) The RAFM is based on the observation that in many harmony processes affixes systematically take over harmonic features from roots, e.g. in root-controlled vowel harmony in Turkish or Finnish. The RAFM is also inspired by the observation that the distribution of marked phonological structure in roots and affixes seems to differ: Affixes generally tend to be less marked than roots and it is “not uncommon to find languages where affixes have no complex onsets, consonant clusters, long vowels, or geminates, even if such structures do appear in roots” (Ussishkin, 1999:72). This is consistent with the assumption that faithfulness constraints for affixes are systematically lower-ranked than the corresponding constraints for roots. However, there are a number of cases where the RAFM is systematically violated. Thus according to Krämer (2002), in Pulaar it is the suffix which controls vowel harmony for advanced tongue root, as can be seen in (18). The root appears in a [+ATR] and a [–ATR] version according to the [ATR] feature of the suffix (-du, -u and -ru are allomorphs of the singular class marker, -O is the diminutive singular class marker): Cf. also similar facts in Turkana (Noske, 2001). A Correspondence-theoretic Account of Fixed Segmentism Reduplication 313 (18) Affix-controlled Vowel Harmony in Pulaar (Paradis, 1992:87) [+ATR] Affix [–ATR] Affix a. ser-du sEr-On ‘rifle butt’ b. beel-u bEEl-On ‘shadow’ c. dog-oo-ru dOg-O-w-On ‘runner’ d. lot-oo-ru lOt-O-w-On ‘washer’ Also the observation that affixes are generally less marked than roots is problematic. For example, the English inflectional affixes of the shape -z and -d are probably unmarked with respect to their place of articulation (coronal), but marked in the sense that they are subminimal, i.e. form neither a minimal word nor a minimal syllable. Probably a better way to think about characteristic shapes of affixes is that they are systematically smaller and contain less phonological structure by avoiding branching in the form of complex onsets, codas, etc. See Downing (2006) for a recent discussion of evidence for the general tendency that the morphological status of linguistic expression correlates in a systematic way with phonological size. There is a second interesting point about the analysis of Hebrew sketched above: The Max constraints relativized to specific morphological domains seem to be ranked “in blocks”. All constraints relativized to affix material are ranked above the corresponding constraints relativized to stems. This is crucial for the constraints Max-V and Max-μ which suggests that the RAFM might be replaced by the metacondition on the ranking of faithfulness constraints formulated in (19): (19) Max-Dep Adjacency: Let α and β be different morphological domains (e.g root, affix, base-reduplicant), and {C1, . . . , Cn} the set of Max and Dep constraints, then either {C1α . . . Cnα} {C1β . . . Cnβ} or {C1β . . . Cnβ} {C1α . . . Cnα}. (19) licenses the ranking in (15) and (16) summarized schematically in (20-a), but also the ranking in (20-b), where the constraints relativized to stems and affixes are systematically flipped. What is systematically excluded are rankings as in (20-c,d), where stem and affix Max constraints alternate in their ranking: 314 Eva Zimmermann & Jochen Trommer