Geminates and Picard Pronominal Clitic Allomorphy

Abstract In this paper, we examine five pronominal clitics in Vimeu Picard with a geminate-consonant allomorph. Assuming a doubly-linked (non-moraic) representation of geminates, we attribute the four different patterns observed to four different underlying structures serving as inputs to an OT constraint ranking. The 1sg and 2sg pronouns are singletons which are subject to gemination in one specific position, syllable structure permitting. The 3sg pronoun is a geminate which is subject to variable degemination or vowel epenthesis where there are insufficient syllable slots to accommodate it. The partitive/genitive is somewhere in between a singleton and a geminate (its UR is /nn/, where the superscript «n» represents a floating nasal; it may surface as [nn] or as [nε˜], depending upon where its floating nasal docks). Finally, the 3pl is neither a singleton nor a geminate underlyingly, but becomes a geminate by its first segment assimilating to its second. Key words: appendix, cross-syllabification, degemination, positional alignment, stressed lengthening, stochastic Optimality Theory, Sympathy Theory, variation, vowel epenthesis; Picard. * This research is supported by NSF grant number BCS-0091687 to the second author. We are grateful to the audiences at the 12th Annual Indiana University French & Italian Graduate Student Colloquium and the 33rd Linguistic Symposium on Romance Languages (LSRL 33), where earlier versions of this analysis were presented, and to an anonymous reviewer for their comments. The analysis has further benefited from discussions with Clancy Clements, Marie-Hélène Côté, Stuart Davis, Caitlin Dillon, Daniel Dinnsen, Heather Goad, Bruce Hayes, John McCarthy, Kathryn Tippetts, and Barbara Vance and from help with the data from Jean-Pierre Calais, Jehan Vasseur, and Jean-Luc Vigneux. Finally, we are indebted to the CatJL editors, Jesús Jiménez and MariaRosa Lloret, for their sizeable investment in proofreading our manuscript and correcting the mistakes, both big and small, in it. The normal disclaimers apply. Catalan Journal of Linguistics 4, 2005 127-154 Geminates and Picard Pronominal Clitic Allomorphy* Brian José Indiana University. Department of Linguistics Bloomington, Indiana 47405 (USA) [email protected] Julie Auger Indiana University. Department of French & Italian and Department of Linguistics Bloomington, Indiana 47405 (USA) [email protected] Table of Contents 1. Introduction 2. Theoretical and empirical issues 3. Four patterns analyzed 4. Conclusion References Cat.Jour.Ling. 4 001-252 7/2/06 11:45 Página 127 1. Introduction A growing body of literature has found that the numerous forms that certain pronominal clitics take on in various Romance languages can all be derived from a single underlying form per pronoun, with the alternations being due to such factors as surrounding phonetic context, syllable structure, the presence or absence of other clitics, and so forth (cf., e.g., Bonet and Lloret 2005 for Catalan; Popescu 2000 for Romanian). In this paper, we show that pronominal clitic allomorphy in Vimeu Picard (henceforth VP) is another such case.1Our focus here, however, is limited; we do not examine all of the VP pronominal clitics, only those which have a geminate consonant allomorph. We will see that while a unique underlying form can be posited for each of the clitics in question, a different type of structure must be posited in each of the four cases that we consider. Thus, we see that what are simply geminates in a language can, in fact, be quite diverse underlyingly, offering another view of Ham’s (2001: 12) claim that “all geminates within a given language may not have a unified phonological representation”. Each of these diverse structures is subjected to an OT constraint ranking (Prince and Smolensky 2004) which results in gemination, in geminate maintenance, or in degemination, thereby deriving the attested alternations. In this paper, we consider four cases in particular: a) the 1sg and 2sg accusative/dative/reflexive pronouns, b) the 3sg accusative pronoun, c) the 3pl accusative pronoun, and d) the partitive/genitive pronoun. Representative examples, introducing the relevant allomorphy, are presented in (1)-(5). While the alternations of primary interest in this paper are between a geminate and a singleton consonant, as in (1)-(3), any analysis of this alternation must also take into account two other clitics that have a geminate allomorph: the 3pl accusative pronoun and the partitive/genitive pronoun. The 3pl pronoun in (4), unlike the singular pronouns in (1)- (3), consistently surfaces as a geminate (e.g., *si j’z érouos déquértchès would not be possible for (4b)). Finally, the partitive/genitive pronoun in (5) (cf. Catalan, French en; Italian ne) is of interest because it exhibits a different type of alternation altogether; here, instead of a geminate ~ singleton opposition, we find [nn] ~ [nε˜]. (1) 1sg accusative/dative a. Tues mmé, [ty mme] si tu veux. kill me-ACC if you want ‘Kill me if you want to.’ b. Acoute mé bien, [akut me bjε˜] Dorine. listen me-DAT well, Doreen ‘Listen to me good, Doreen.’ 128 CatJL 4, 2005 Brian José; Julie Auger 1. Picard is a Romance language very closely related to French and spoken in parts of northern France and southern Belgium. The Vimeu variety is spoken in France’s Somme département and is bounded by the Somme River to the north, the Bresle River and Normandy to the south, departmental highway 901 to the east, and the English Channel to the west. Cat.Jour.Ling. 4 001-252 7/2/06 11:45 Página 128 (2) 2sg accusative/dative/reflexive a. Assis tté leu [asi tte lø] su ch’banc. sit you-REFL there on that bench ‘Sit here on this bench.’ b. Dépéque té [depek te] d’évnir. hurry you-REFL of to-come ‘Come quick.’ (3) 3sg accusative a. O ll’ avoéme [o ll avwe˜m] aveuc nous we 3sg-ACC had with us ‘It [the wind] was at our backs.’ b. J’ l’ avoais [ l avwε] intindue dire... I 3sg-ACC had heard to-say ‘I had heard her say...’ (4) 3pl accusative a. Pétète qu’a zz’ éroait [ka zz erwε] acatès. maybe that she 3pl-ACC would-have bought ‘Maybe she would have bought them.’ b. Si jé zz érouos [si e zz erwɔ] déquértchès. if I 3pl-ACC would-have unloaded ‘If I had unloaded them.’ (5) Partitive/genitive a. J’ énn’ ai foait [e˜ nn e fwε]. I PART AUX made ‘I made some.’ b. Cho’f fille blonde a nin rioait [a nε˜ riwε]. the girl blonde she PART was-laughing ‘The blonde girl was laughing about it.’ We describe the data more fully in the next section (§2.2); there, we also address the representation of geminates that we assume in this paper (§2.1). We present our analysis in §3. Finally, we conclude in §4. 2. Theoretical and empirical issues We begin this section by identifying the representation of geminates that we adopt in the analysis to follow (§2.1). Then we expand on our superficial presentation of the data above, including a description of pronoun-specific distributional idiosyncrasies (§2.2). Geminates and Picard Pronominal Clitic Allomorphy CatJL 4, 2005 129 Cat.Jour.Ling. 4 001-252 7/2/06 11:45 Página 129 2.1. The representation of geminates In this paper, we assume a doubly-linked representation of geminates instead of a possible, and arguably more popular, alternative: the moraic representation (e.g., Hayes 1989). The pros and cons of these two views and the history of the debate have been discussed extensively in the literature, and we will not repeat those arguments here (cf., e.g., Broselow 1995; Curtis 2003: 30-58; Ham 2001: 6-15; Kraehenmann 2003: 13-27). Suffice it to say that the debate persists even to this day.2Possible doubly-linked representations include a single root node linked to two skeletal slots (e.g., McCarthy 1979), or a single set of phonological features linked to two root nodes (Selkirk 1990). Because we believe it to have been amply demonstrated —particularly in the framework of Prosodic Morphology (McCarthy and Prince 1986; but also cf. Selkirk 1990)— that there is no need to posit a skeletal tier as part of the prosodic hierarchy, we opt for the latter of these two doublylinked representations here. It should be understood, though, that by adopting a Rt-Rt representation of geminates, we are not arguing against the moraic representation. Rather, our reasons for treating geminates as Rt-Rt —or, more precisely, for not treating them as moraic— are twofold. First, we can find no evidence of any weight-sensitive phenomena in VP which would allow us to argue for or against the moraic representation of geminates in this language.3Second, with respect to syllabification and vowel epenthesis, geminates pattern much like unsyllabifiable consonant clusters, a result which is difficult to obtain if geminates are single but moraic segments, but which follows straightforwardly if they are doublylinked segments. Next, we turn to a more thorough examination of the data. This will reveal the various positions where geminates can occur in VP —initially, medially, and finally— and it will demonstrate the similarities between geminates and clusters, both of which require two syllable positions, thereby underscoring the attraction of the Rt-Rt representation. 2.2. Data and overview of the problem The data described here and analyzed in §3 is taken from a corpus of written Picard texts supplemented with native-speaker intuitions. For arguments defending the use of written data for phonological analyses (in VP), see Auger (2001: 259-262). 130 CatJL 4, 2005 Brian José; Julie Auger 2. One representation that seems to have gone by unnoticed in these debates is Schmidt’s (1992) hybrid representation in which geminates are both doubly-linked and moraic. While we do not find the specific arguments that she offers in favor of her segmental-moraic representation over a simply moraic representation to be particularly compelling (also cf. Curtis’ 2003: 306-309 criticisms of the composite model as implemented by others), we nonetheless think that it is an intriguing hypothesis. It is also interesting that the two-root representation is not incompatible with mora projection (Selkirk 1990: 127), that Curtis (2003) does, in fact, combine Selkirk’s two-root representation with moras, and that Muller (2002) arrives at a very similar skeletal and moraic representation as Schmidt, all of which attest to the appeal of such a hybrid representation or composite model. 3. For languages where there is evidence that geminates contribute to syllable weight, we have adopted Moraic Theory (José forthcoming). Cat.Jour.Ling. 4 001-252 7/2/06 11:45 Página 130 Also note that Auger found written Picard to be largely representative of the spoken language with respect to both a phonological variable (Auger 2002) and a series of grammatical variables (Auger 2003a). The question at hand is what this written data tells us about VP pronominal clitic allomorphy and the distribution of geminate consonant forms and non-geminate forms. We begin, though, with a general description of VP syllable structure. 2.2.1. Syllable structure and consonant clusters The phonotactics of VP allow for fairly complex syllable structures. For instance, complex onsets are possible as long as sonority rises from the first segment to the second, the two segments do not occupy adjacent levels on the sonority hierarchy, and they have different places of articulation. Also, complex nuclei are possible as long as they are «light monomoraic diphthongs» (Steele and Auger 2002: 320), e.g., [wV] as in ploéyon [plwe.jɔ˜] ‘flexible branch’ or [ɥi] as in pluie [plɥi] ‘rain’. Finally, complex codas are possible word-finally as long as the first segment is a liquid and the second is less sonorous (e.g., pérle [perl] ‘speaks’). If consonants that cannot be properly syllabified according to the preceding phonotactic requirements come together, then some repair strategy must be invoked. The phonology of VP offers four possibilities. The first one, cross-syllabification, allows a word-initial consonant to be syllabified as the coda of a preceding vowelfinal syllable, as in passeu dvant [pa.sød.va˜] ‘to pass in front of’, or a word-final consonant to be syllabified as the onset of a following vowel-initial syllable, as in un triste étot [˜.tris.te.to] ‘a sad state’. The second possibility is vowel epenthesis. An example of this is when a word like dvant, with an illicit initial cluster, follows a consonant-final word such as s’assir ‘to sit down’: s’assir édvant [sa.si.red.va˜] ‘to sit down in front of’. The third possibility is deletion. This occurs almost exclusively with final (rising sonority) obstruent-liquid clusters; in such cases, the liquid is deleted (e.g., /øtr/ →[øt] eute ‘other’; cf. eutrémint [ø.tre.mε˜] ‘otherwise’). The fourth possibility, indirect licensing, is only available at the edges of prosodic boundaries. Examples of this are when a cluster like [dv] occurs at the beginning of an Utterance (Utt) or Intonational Phrase (IntPhr), as in Dvant qu’éch co i cante [dva˜ keʃ ko i ka˜t] ‘Before the cock crows’, or when a cluster like [st] occurs at the end of a Prosodic Word (PrWord), as in a fzoait juste [a fzwε yst] ‘it was close’. In these cases, the most marginal consonant in the cluster can be licensed by one of these higher levels of prosodic structure instead of the syllable. The actual licensing strategy employed in such contexts varies, though (see Auger 2000, 2001; Steele and Auger 2002). Of more relevance here is that there are important differences between initial and final appendices. For one thing, they are not licensed at the same levels; as indicated above, initial appendices are licensed by only the highest levels of prosodic structure (the Utt and the IntPhr), while final appendices are licensed by levels as low as the PrWord. Also, while any consonant can be licensed as an initial appendix, the only possible final appendices are the voiceless stops. Returning, now, to the specific question of geminates, we first provide general observations about where they occur in VP (§2.2.2) and then we address issues specific to the particular clitics analyzed here (§2.2.3). Geminates and Picard Pronominal Clitic Allomorphy CatJL 4, 2005 131 Cat.Jour.Ling. 4 001-252 7/2/06 11:45 Página 131 2.2.2. Geminates: general observations In VP, as in most other languages, geminates tend to occur in intervocalic position. The examples in (1)-(5), above, illustrate this issue. Here, the first half of the geminate is syllabified as a coda and the second half of it is syllabified as an onset. This preference for intervocalic ambisyllabicity is certainly understandable, considering the phonotactic restrictions on complex onsets and codas introduced at the beginning of §2.2.1: since geminates consist of two root nodes sharing a single feature geometric structure, they cannot respect sonority sequencing requirements for complex onsets and codas. This forces us to ask, then, how it is that VP, rather atypically, also allows geminates in phrase-initial and phrase-final positions. Examples are provided in (6) and (7), respectively. The answer resides in our fourth possibility for handling illicit clusters: indirect licensing. That is, in these positions one half of the geminate is licensed as an appendix by a higher level of prosodic structure, as discussed above with respect to clusters. So, the two halves of the geminates in (6) constitute appendix-onset sequences and in (7), coda-appendix sequences. We should point out, though, that forms such as those in (7) with final geminates are the exception rather than the rule. Not only is there a general prohibition against final geminates in VP, but even with post-verbal clitics (the only possible exception to this prohibition) final geminates are allowed only when they convey emphatic stress.4In these cases, the [mm] of (7a) violates constraints against final geminates and [m]-appendices (the voiceless stops, again, being the only possible word-final appendices in VP) while the [tt] of (7b) violates the constraint against final geminates. Both of these states of affairs are presumably achieved by ranking an emphasis constraint, call it *WEAK-IN-STRONG/EMPHATICSTRESS (see §3.2), above these constraints.5 132 CatJL 4, 2005 Brian José; Julie Auger 4. That phrase-final geminates are restricted to clitics in post-verbal position is evident in the following examples. Here, we see that the /tt/ of 3pl verbal inflection surfaces as [tt] when phraseinternally, as in (i) and (ii), but as [t] when phrase-finally, as in (iii). (i) Il avoait’t acatè [i.la.vwεt.ta.ka.tε] din la ville éch qu’o n’trouve point they had bought in the city that which one NEG finds not din chés piots poéyis. in the little countries ‘They were able to buy in the city what you can’t find in these little towns.’ (ii) Ch’étoait gramint des piots ménagers qu’i n’avoait’té [ki.na.vwεt.te] point d’auto. it was lot of-the little small farmers that they NEG had not of car ‘There were a lot of small farmers who didn’t have a car.’ (iii) Oz avoéme tout no temps pour accouter chés gins qu’i leu dvisoait’t [ki.lød.vi.zwεt] we had all our time for to-listen the people that they REFL were-chatting ‘We had all day to listen to the people who were chatting with each other.’ 5. That post-verbal clitics are able to exceed VP syllable structure, suggesting that at least some of the σ-STRUCTURE constraints are dominated, offers further evidence in favor of a *WEAK-INSTRONG/EMPHATICSTRESS constraint. Some speakers accept forms such as Dépèq’tt! [depεktt] ‘Hurry up!’, where the [k] could be the coda and the first [t] an appendix, but where there is no position for the second [t]. Similarly, in Donne-mmé [dɔ˜nmme] ‘Give me’, the [n] could be a Cat.Jour.Ling. 4 001-252 7/2/06 11:45 Página 132 (6) Phrase-initial geminates a. Ll’ as-tu rtrouvèe? 3sg AUX you found-fem ‘Did you find her?’ b. N’n’ êtes-vous bien certain? of-it are you well certain ‘Are you sure about that?’ (7) Phrase-final geminates a. Non, creuyeu mm’. No believe 1sg ‘No, believe me.’ b. Tais tt’! shut-up 2sg-REFL ‘Shut up!’ When geminates are not in intervocalic position or at the edge of some prosodic boundary —that is, when they occur in a consonantal context where both halves of the geminate cannot be licensed due to a shortage of available positions— one of two courses of action is available: either the geminate shortens, as in (8a), or an epenthetic vowel is inserted, thereby rescuing the intervocalic context so that the geminate can surface faithfully as a long segment, as in (8b). Thus, we see that with respect to syllabification, geminates pattern much like consonant clusters. How they differ is that clusters are not readily simplified (except under very specific circumstances (e.g., /øtr/ →[øt])) but geminates are fairly readily shortened when available positions are lacking. (8) Geminates in a consonantal context a. O l’ saveu. you 3sg know ‘You know it.’ b. O llé saveu. you 3sg know ‘You know it.’ Geminates and Picard Pronominal Clitic Allomorphy CatJL 4, 2005 133 coda and the second [m] an onset, but there is no position for the first [m]. In these cases where we get more segments than positions, an emphatic meaning again attaches; these same forms with a singleton —[depεkt] and [dɔ˜nme]— do not mark any emphasis. Cat.Jour.Ling. 4 001-252 7/2/06 11:45 Página 133 2.2.3. Geminates: pronoun-specific observations In addition to these general patterns, there are a few pronoun-specific distributional facts that need to be addressed. For instance, pre-verbally, the 1sg and 2sg pronouns occur only as singletons, never geminates, even intervocalically. Examples are provided in (9). The geminate ~ singleton alternations illustrated in (1)-(2), then, are limited to post-verbal position. Conversely, for the 3sg pronoun, there is no alternation post-verbally: here, we find only geminate [ll], never singleton [l]. Examples are provided in (10). For this pronoun, the geminate ~ singleton alternations illustrated in (3) and (8) are limited to pre-verbal position. Likewise, the [nn] ~ [nε˜] alternations of the partitive/genitive are restricted to pre-verbal position; as with the 3sg, we find only the geminate allomorph post-verbally. This is shown in (11). Finally, as mentioned in §1, the 3pl surfaces only as a geminate regardless of position; the examples in post-verbal position in (12), along with the examples in pre-verbal position in (4), show this. (9) 1sg, 2sg pre-verbally: singletons only a. A m’ est bien égal. (*A mm’est) that 1sg is well equal ‘It’s all the same to me.’ b. Tu t’ as bérlurè? (*Tu tt’as) you 2sg AUX mistaken ‘Did you make a mistake?’ (10) 3sg post-verbally: geminate [ll] only a. Dis llé. (*Dis l’) say 3sg ‘Say it.’ b. Donne é-llé à tin pére. (*Donne lé) give 3sg to your father ‘Give it to your father.’ (11) Partitive/genitive post-verbally: geminate [nn] only a. Pis rapporté n-né pour chint sous. (*rapporté nin) and bring-back PART for hundred monetary-unit ‘And bring back five franc’s worth.’ b. Prindez nné don ène cope éd bouteilles. (*Prindez nin) take PART thus a couple of bottles ‘So take a couple of bottles.’ 134 CatJL 4, 2005 Brian José; Julie Auger Cat.Jour.Ling. 4 001-252 7/2/06 11:45 Página 134 (12) 3pl post-verbally (also pre-verbally in (4)): geminate [zz] a. Acouteu zzés jacteu dérriére pour des pronnes. (*Acouteu z’jacteu) listen-to 3pl to-jabber behind for some prunes ‘Listen to them jabbering back there for some prunes.’ b. Mais ouve éz-zés, tes zius! (*ouve zés) but open them, your eyes ‘Open your eyes!’ To summarize, this problem presents us with two main challenges. One is explaining why geminates can occur phrase-initially and phrase-finally in addition to intervocalically, but only where there are two available syllable (or other prosodic) positions. The solution to this follows straightforwardly from geminates consisting of a single feature geometric structure simultaneously linked to two root nodes. The second is explaining why there are alternations where there are but not where there are not, as summarized in table (13). Cells in which there are no alternations are shaded. Given that a grammar, in terms of Optimality Theory, is a fixed, language-specific constraint ranking (abstracting away from various proposals for achieving the variable rankings required to analyze variable phenomena), it seems that this second challenge can be met only by attributing each of the four patterns in table (13) to a different underlying structure. We demonstrate what each of those underlying forms must be and how each is treated by the grammar in §3. (13) Positional considerations of (non-)alternations summarized 3. Four patterns analyzed In our preceding description of the data, we offered a preview of our analysis. In essence, where two syllable/prosodic positions are available, geminates surface as geminates; where only one syllable/prosodic position is available, either they shorten or a vowel is inserted so that the necessary second syllable position becomes available. The four different patterns depicted in table (13) result from four different underlying structures serving as inputs to the grammar, yielding gemination, geminate maintenance, or degemination. As we will see in the following sections, the 1sg and 2sg pronouns are underlying singletons, /m, t/, which are geminated post-verbally, syllable structure permitting; the 3sg pronoun is an underlying geminate, /ll/; the partitive/genitive is somewhere in between a singleton and Geminates and Picard Pronominal Clitic Allomorphy CatJL 4, 2005 135 Pronoun Pre-verbally Post-verbally 1sg [m][m] ~ [mm] 2sg [t][t] ~ [tt] 3sg [l] ~ [ll][ll] partitive/genitive [nε˜] ~ [nn][nn] 3pl [zz][zz] Cat.Jour.Ling. 4 001-252 7/2/06 11:45 Página 135 host, in violation of the applicable alignment constraint. Therefore, candidate (j) is optimal. (30) Post-verbal singletons: lengthen unless blocked by syllable structure Now that we have seen that something special happens to post-verbal clitics, we can return to the case of the 3sg and illustrate the final remaining aspect of that pronoun: its lack of alternations in this position. 142 CatJL 4, 2005 Brian José; Julie Auger Tais tt’! ‘Shut up!’ /tε t/ a. [tε´t]*! b. [tε te´]*!*** c. [tε e´t]*!***** d. [tε ete´] *! * * * ** ** ☞e. [tε´ tt]*!* ☞f. [tε tte´]*!*** g. [tε e´tt]*!****** h. [tε ette´] *! * * ** *** Acoute mé ‘Listen to me’ /akut m/ i. [aku´t m]*! * * ☞j. [akut me´]***** k. [akut eum]*!***** l. [akut e˜me´] *! * * * ** ** m. [aku´t mm]*!* * * * n. [akut mme´]*! ***** o. [akut eumm] *!* ****** p. [akut e˜mme´] *! * * ** *** σ-STRUCT ALIGN (VB, CL) *WEAK-INSTRONG *FINAL-APP DEP(V)/ STRESS ONSET DEP(V) DEP(RT) σ-STRUCT ALIGN (VB, CL) *WEAK-INSTRONG *FINAL-APP DEP(V)/ STRESS ONSET DEP(V) DEP(RT) Cat.Jour.Ling. 4 001-252 7/2/06 11:45 Página 142 3.3. The 3sg pronoun, continued Recall that the 3sg pronoun exhibits alternations pre-verbally but not post-verbally. Post-verbally, /ll/ never shortens. We might suspect this to be related to the process of post-verbal lengthening discussed in §3.2, but since /ll/ is already long, post-verbal lengthening has no effect. Nonetheless, there is an indirect connection between post-verbal lengthening and the lack of post-verbal degemination with the 3sg pronoun: they are both due to the same driving force, stress. Since /ll/ is underlyingly long, it, unlike /m/ and /t/, is subject to a faithfulness constraint that specifically prevents it from shortening in this privileged position even when there are insufficient syllable positions to accommodate it. This results in obligatory vowel epenthesis where degemination would otherwise be a possibility. In this case, /ll/ is not shortened post-verbally after a consonant (compare the lack of /m/ and /t/ lengthening after a consonant); instead, an epenthetic vowel is inserted between the verb and the clitic. This means that our new positional faithfulness constraint outranks ALIGN(VERB, CLITIC), which, itself, outranks the markedness constraint *WEAK-IN-STRONG, as established in tableau (30). Because this positional faithfulness constraint, MAX(ROOT)/STRESS in (31), and the various constraints responsible for syllable wellformedness, σ-STRUCT, outrank ALIGN(VERB, CLITIC), all candidates except for the optimal [war.del.lé] (candidate (d)) are immediately eliminated in the top half of tableau (32). Here, then, we see two unusual results: the clitic can be separated from the verb if necessary and an epenthetic vowel can appear in Utt-final position. Turning to the vocalic environment in the bottom half of tableau (32), we get the same result except that the clitic need not be separated from the verb by an epenthetic vowel. It is important to compare the optimal candidate here, candidate (k), with another very plausible candidate, candidate (i). We get an epenthetic vowel in Utt-final position, which is forbidden except with post-verbal clitics, because [l] is not a possible final appendix; in evaluating candidate (i) (also candidates (a), (b), (e), and (j)), we continue to count this as a σ-STRUCTURE violation, as we did for [m] in tableau (30). (31) MAX(ROOT)/STRESS: For a clitic in stressed (i.e. post-verbal) position, every root node present in the input should have a correspondent in the output (do not delete root nodes from clitics in stressed syllables). Geminates and Picard Pronominal Clitic Allomorphy CatJL 4, 2005 143 Cat.Jour.Ling. 4 001-252 7/2/06 11:45 Página 143 (32) Post-verbal geminates: obligatorily maintained This concludes our analysis of the basic pattern, as illustrated through the singular pronouns; we turn next to the other two clitic pronouns which have a geminate-consonant allomorph: the 3pl and the partitive/genitive. 3.4. The 3pl pronoun The 3pl pronoun consistently surfaces as a geminate. Even where only one syllable/prosodic position is available, no shortening is observed. Instead, vowel epenthesis is obligatory in such contexts. The reason for this follows from the underlying form of this pronoun. The 3pl pronoun is actually bimorphemic, consisting of a 3rd person morpheme and a plural morpheme, both of which must be realized, following Kurisu (2001). Specifically, the underlying form of this pronoun is /lz/, with the 3rd person morpheme /l/ assimilating to the plural morpheme /z/. There is 144 CatJL 4, 2005 Brian José; Julie Auger Warde é-llé! ‘Keep it!’ /ward ll/ a. [wárdl.l] *!* * b. [war.dél.l] *! * * * * c. [ward.llé] *! * * ☞d. [war.del.lé] * * ** e. [wárd.l] *! * * f. [war.dél] *! * * * * * g. [ward.lé] *! * * * * h. [war.de.lé]*! * * * ** * Disons l’lé ‘Let’s say it’ /dizɔnll/ i. [di.zɔ ul.l]*! * j. [di.zɔ˜.él.l] *! * * * * * ☞k. [di.zɔ˜l.lé] * * l. [di.zɔ˜.el.lé] *! * * ** m. [di.zɔ ul]*! * * n. [di.zɔ˜.él] *! * * * * * * o. [di.zɔ˜.lé] *! * * * * p. [di.zɔ˜.e.lé] *! * * * * ** * MAX(RT)/ STRESS σ-STRUCT ALIGN (VB, CL) *WEAK-INSTRONG *FINAL-APP DEP(V)/ STRESS ONSET DEP(V) MAX(RT) MAX(RT)/ STRESS σ-STRUCT ALIGN (VB, CL) *WEAK-INSTRONG *FINAL-APP DEP(V)/ STRESS ONSET DEP(V) MAX(RT) Cat.Jour.Ling. 4 001-252 7/2/06 11:45 Página 144 ample cross-Romance evidence supporting /lz/, including data from numerous dialects of Picard (cf. Dauby 1979: 33 and Mahieu 1984: 63 for Picard; Bonet and Lloret 2005: 47 for Catalan). Additionally, we know that many /l/-final clitics undergo assimilation and gemination in certain varieties of Picard, including VP (Cardoso 2001; Flutre 1977: 156-157; Morin 1995). If [zz] were to shorten in a consonantal environment, either the 3rd person morpheme or the plural morpheme would fail to be realized, so vowel epenthesis becomes obligatory in such contexts. Tableau (36) demonstrates. (33) ASSIM /lz/→[zz]: An abbreviation for the constraints responsible for categorical /l/ assimilation in the specific context of the 3pl pronoun. (34) REALIZEMORPHEME: Every underlying morpheme must receive some phonological exponence (Kurisu 2001). (35) /l/-FAITH: An abbreviation for the constraints that militate against /l/ assimilation. (36) 3pl /lz/: assimilates to [zz] and surfaces «faithfully» In tableau (36), we see that /lz/ must surface as [zz], otherwise one of the highranking constraints ASSIM /lz/→[zz] or REALIZEMORPHEME is violated. Between vowels (top half of tableau (36)), we get [zz] straightforwardly; next to a consonant (bottom half of tableau (36)), the elimination of (36f) to the preference of (36h) shows that the ranking σ-STRUCT » DEP(V) yields vowel epenthesis where needed. We turn now to our final pronoun: the partitive/genitive. Geminates and Picard Pronominal Clitic Allomorphy CatJL 4, 2005 145 a zz’éroait acatès ‘she would’ve bought them’ ASSIM REALIZE σ- /l/- DEP /a lz erwεakatε/ /lz/ →[zz] MORPHEME STRUCT FAITH (V) a. [al.zer.wε…] *! b. [a.zer.wε…] *! ☞c. [az.zer.wε…] * acouteu zzés jacteu ‘listen to them jabber’ ASSIM REALIZE σ- /l/- DEP /akutølz aktø/ /lz/ →[zz] MORPHEME STRUCT FAITH (V) d. [a.ku.tøl.zak.tø] *! * e. [a.ku.tøl.ze.ak.tø]*! * f. [a.ku.tøz.zak.tø]*!* g. [a.ku.tøz.ak.tø]*! ☞h. [a.ku.tøz.ze.ak.tø]** Cat.Jour.Ling. 4 001-252 7/2/06 11:45 Página 145 3.5. The partitive/genitive pronoun As with the 3pl, the idiosyncratic nature of the partitive/genitive pronoun is due to its underlying form. Here, the idiosyncrasy is more the particular alternation observed —[nn] ~ [nε˜] as opposed to [nn] ~ [n]— than the lack of alternations post-verbally, although we must also account for this latter issue. We propose that the underlying form of this pronoun is /nn/, where, again, the superscript «n» represents a floating nasal (Paradis and El Fenne 1995). Following Zoll (1998), we assume that the difference between a floating segment and an anchored segment is that the former is unassociated to any root node in the input while the latter is linked to a root node. Before a vowel-initial word (e.g., J’énn’ai foait in (5a)), the sequence /nnV/ is very normally syllabified as [n.nV] (although an epenthetic vowel may be required to introduce the coda position for the first, i.e. the anchored, [n]). Before a consonant-initial word, on the other hand (e.g., a nin rioait in (5b)), the sequence /nnC/ cannot be syllabified without the insertion of an epenthetic vowel; the floating nasal ultimately docks onto this inserted vowel, yielding [nε˜C].12 Before we delve any further into the discussion of this clitic, we should first note that there are two types of nasal vowels in VP (Vasseur 1998). In one of these, the nasal feature is primarily linked to the vowel itself (these are what Vasseur 1998 calls the «nasal» vowels); in the other, the nasal feature is primarily linked to a neighboring consonant and is associated to the vowel only by regressive spreading (these are what Vasseur 1998 calls the «half-nasal» —a.k.a. nasalized— vowels). (For discussion, see José and Auger 2004: §§2.2, 4.1.) Of particular relevance here is that in the case of an epenthetic vowel, it surfaces as [ε˜] when it is nasal and as [e˜] when it is halfnasal. (Indeed, the same is true of underlying /e/.) Both half-nasal and nasal vowels occur in the candidates in the following tableaux (e.g., (38d) vs (38e), (38i) vs. (38j)). 146 CatJL 4, 2005 Brian José; Julie Auger 12. An anonymous reviewer indicates that (s)he finds the UR we posit for this clitic surprising. Pointing out that sequences of adjacent identical segments within a single morpheme are normally analyzed as sharing a common feature structure (i.e., the representation of geminates that we assume here), the reviewer is understandably skeptical of a normal-nasal-plus-rootless-(but-otherwise-identical)- floating-nasal sequence. Frankly, we, too, were initially leery of this input; but we are now confident that this is the UR of this clitic. The principal reason for this is that /nn/ accounts for the distribution of allomorphs, whereas other URs that we might consider —/nn/, /n/, /nεn/, and /nεn/— are unable to. Also, the partitive/genitive pronoun is not the only one that patterns in this way. The masculine singular possessive adjectives also surface as [Cn] before a vowel-initial word and [Cε˜] before a consonant-initial word: e.g., (é)sn’ami [(e)sn ami] ‘his/her friend’ vs. sin copain [sε˜ kopε˜] ‘his/her friend’. Thus, we posit the same type of structure for these as we do for the partitive/genitive: /mn/, /tn/, and /sn/. And in the absence of morpheme structure constraints or some other such restriction on inputs in OT (cf. McCarthy 2003), if we allow /Cn/, Richness of the Base (Prince and Smolensky 2004) requires us to allow /nn/ as a possible instantiation of /Cn/, unless /nn/ is universally ill-formed rather than ruled out on a language-specific basis. This, though, would require the OCP to be a condition on inputs and not a violable output constraint, as Myers (1997) has convincingly argued that it is. An important caveat here is that Myers focuses on OCP violations across morpheme boundaries and not morpheme-internally. All that he says about the OCP within morphemes is that Lexicon Optimization (Prince and Smolensky 2004) will always select an OCP-obeying input over an OCPviolating input (Myers 1997: 889, fn. 31). McCarthy (2005), though, argues that other considerations can sometimes override the predictions of Lexicon Optimization. Cat.Jour.Ling. 4 001-252 7/2/06 11:45 Página 146 (37) MAX[nasal]: Every occurrence of the feature [nasal] in the input has a correspondent in the output. (38) Pre-verbal /nn/: [nn] before a vowel, [nε˜] before a consonant If we insert a vowel where it is not needed (i.e. before another vowel), we unnecessarily violate faithfulness, potentially creating a superfluous violation of ONSET, eliminating candidates (c), (d), and (e) in the top half of tableau (38). DEP(ROOT) does have to be violated in order for the floating nasal to surface as an onset, but this is better than the alternative: while «shortening» the input, as in candidate (b), does not violate MAX(ROOT), given that the floating nasal is underlyingly rootless, this does require us to delete an input [nasal], in violation of highranking MAX[nasal].13 Therefore, candidate (a) is optimal.14 Before a consonant, on Geminates and Picard Pronominal Clitic Allomorphy CatJL 4, 2005 147 13. For a more extensive discussion of the phonology of [nasal] in VP, including some other consequences of MAX[nasal] and evidence that it is a high-ranking constraint, see José and Auger (2004). Although tableau (38) may give the impression that MAX[nasal] is undominated, this cannot be the case. For instance, point [pwε˜] ‘not’ is regularly denasalized as poé [pwe] ‘not’ when in unstressed position: a n’insistouot point [pwε˜] ‘she didn’t insist’ vs. o n-n’oz ons poé [pwe] comprins ‘we didn’t understand each other’. Also, nasality can be deleted in what is probably a case of haplology: ‘nn’ and ‘nin’ represent both negative /n/ and the partitive/genitive complement in a nn’avoait point pérlè ‘she didn’t talk about it’ and j’nin sais rién d’plus ‘I don’t know anything else about it’. 14. Kathryn Tippetts (p.c.) asks why the floating nasal surfaces as an onset before a vowel and not as a nasal feature on that vowel (e.g., /…nna…/ →*[…nã…]), which would satisfy both MAX[nasal] and DEP(ROOT). Most often, this can be ruled out by ONSET (e.g., boin ‘good’ /bwen/ →[bwε˜], O no nn’aloéme ‘We were going’ MAX σ-DEP DEP /o no nnalwem/ [nasal] STRUCT ONSET (V) (RT) ☞a. [õ.nõn.nal.we˜m] [o] * b. [õ.nõ.nal.we˜m] *! [o] c. [õ.nõn.ne.al.we˜m] [o] [a]! * ** d. [õ.nõ.ne˜.nal.we˜m] [o] *! ** e. [õ.nõ.nε˜.al.we˜m] [o] [a]! * * A nin rioait ‘She was laughing about it’ MAX σ-DEP DEP /a nnriwε/ [nasal] STRUCT ONSET (V) (RT) f. [ãn.nri.wε] *! [a] * g. [ãn.ri.wε] *! [a] h. [ãn.ne.ri.wε] [a] * **! i. [ã.ne˜n.ri.wε] [a] * **! ☞j. [ã.nε˜.ri.wε] [a] * * Cat.Jour.Ling. 4 001-252 7/2/06 11:45 Página 147 the other hand, if we do not insert a vowel, either a σ-STRUCTURE violation results (candidate (f)) or we fail to realize the input floating nasal (candidate (g)). Candidate (j) is preferred to candidates (h) and (i) in the bottom half of tableau (38) because candidate (j) violates DEP(ROOT) only once, for the inserted vowel, while candidates (h) and (i) each violate DEP(ROOT) twice, once for the inserted vowel and once for the consonantal root node where the floating [nasal] surfaces. The lack of post-verbal alternations with this pronoun, where we get only [nn], never [nε˜], is most certainly related to the process of post-verbal lengthening that we introduced earlier. The different outcomes of vowel epenthesis after this pronoun when it is pre-verbal vs. post-verbal are particularly telling. Pre-verbally, as we have just seen, a vowel inserted after this pronoun yields [nε˜] (recall that the quality of [e] changes to [ε] when it bears the full force of a [nasal] feature); postverbally, a vowel inserted after this pronoun yields [nne]. This, of course, is expected if the addition of a root node has first lengthened /nn/ to [nn], just as it has lengthened /m/ and /t/ to [mm] and [tt].15 Even though we are not dealing here with opacity, of either the non-surface-apparent variety or the non-surface-true variety, we assume that faithfulness to intermediate [nn] is achieved by a high-ranking constraint requiring the output to be faithful to a sympathetic candidate (McCarthy 1999), MAX-O(ROOT)/STRESS, but it is possible that the same result could be achieved by means of some other mechanism such as constraint conjunction or strata-specific rankings (see McCarthy 1999: 382-391 for a summary).16 Tableau (41) illustrates the selection of the sympathy candidate and tableau (42) the selection of the optimal candidate based upon the sympathy candidate. The selector constraint is DEP(V), as indicated by the preceding this constraint. In sympathy candidate selection (tableau (41)), only those candidates that satisfy the selector constraint are considered; among these, the candidate that best satisfies 148 CatJL 4, 2005 Brian José; Julie Auger boin + appétit ‘enjoy your meal’ = [bwe˜.na.pe.ti]; both *[bwε˜.a.pe.ti] and *[bwe.ã.pe.ti] violate ONSET, which outranks DEP(ROOT)). But with /Cn/, the onset could and should be filled by the /C/ (e.g., /snami/ →*[sã.mi]). There must be some other reason, then, that a floating nasal does not dock onto a following vowel. At present, we do not have an explanation, but one possibility is that only leftward nasal linking is permitted. While seemingly convenient, there is some justification for this: pre-N but not post-N vowels are allophonically nasalized in VP. José and Auger (2004) attribute this directionality to markedness (*[oral]V[nasal]C» DEPPATH[nasal]), but if the lack of rightward nasalization is more general, then perhaps it should be attributed, instead, to faithfulness (e.g., NOSPREAD-R[nasal] » *[oral]VN, *NV[oral] » NOSPREAD-L[nasal]). 15. This is not to say that post-verbal [nn] is a geminate in the same sense that [ll], [mm], and [tt] are. Geminate [ll], [mm], and [tt] are truly geminates in the sense that they consist of a single feature geometric structure doubly-linked to two root nodes. On the other hand, [nn] is not a geminate, properly speaking, in that it has two feature geometric structures, each of which is singly-linked to its two root nodes. 16. This O-faithfulness constraint needs to be specified for stressed position as in (39); otherwise, the wrong candidate emerges as optimal in the bottom half of tableau (38). In (38), candidates (a, f) would be selected as the sympathy candidates, and if this sympathy constraint is simply MAX- O(ROOT), then the attested candidate (j) would be eliminated for violating it (see footnote 17). Alternatively, it could be that the selector constraint should be DEP(V)/STRESS instead of DEP(V), which would yield the correct results in both tableaux (41)/(42) and tableau (38); but the STRESS condition must be included somewhere. Cat.Jour.Ling. 4 001-252 7/2/06 11:45 Página 148 the (rest of the) constraint hierarchy, minus any sympathy constraints, is selected as the sympathy candidate. Here, this is candidate (a). For the selection of the optimal candidate (tableau (42)), all constraints, including sympathy constraints and the selector constraint, become relevant again. Candidates are evaluated normally, with sympathy constraints —and only sympathy constraints— evaluating correspondence between the sympathy candidate, here candidate (a), and each of the output candidates. In tableau (42) sympathy constraints rule out candidates (b), (d), and (e), leaving candidates (a) and (c).17 The former of these, candidate (a), is ruled out for violating syllable structure requirements: the sequence [nn] cannot be licensed in final position ([n] not being a possible final appendix); so, candidate (c), with its final epenthetic vowel, is optimal. (39) MAX-O(ROOT)/STRESS: For a clitic in stressed (i.e. post-verbal) position, every root node in the sympathy candidate has a correspondent in the output (between the sympathy candidate and an output candidate, do not delete root nodes from clitics in stressed syllables). (40) MORPHCONTIG-O: Segments that form a contiguous string within a morpheme in the sympathy candidate form a contiguous string in that morpheme in the output. (41) Post-verbal /nn/: sympathy candidate selection Geminates and Picard Pronominal Clitic Allomorphy CatJL 4, 2005 149 17. We are assuming that candidate (42e) deletes a consonantal root node present in the sympathy candidate and inserts a vowel, thus its violations of DEP(V) and DEP(RT) in addition to MAX- O(ROOT)/STRESS. A homophonous candidate (42e') would turn the sympathy candidate’s second consonantal root node into a vocalic root node; such a state of affairs could be ruled out by high-ranking constraints MAX-O[CONSONANTAL] and/or DEP-O[VOCALIC]. Prindez nné ‘Take some’ /prεnde nn/ a. [prε˜.de˜n.n]** b. [prε˜.de˜n]*!* c. [prε˜.de˜n.ne]**!** d. [prε˜.de˜n.ne˜n]**!** e. [prε˜.de˜.nε˜]**!* MAX-O (ROOT)/ STRESS MORPH CONTIG-O MAX[nas] σ-STRUCT *WK-IN-STR DEP(V)/ STRESS DEP(V) DEP(RT) Cat.Jour.Ling. 4 001-252 7/2/06 11:45 Página 149 (42) Post-verbal /nn/: optimal candidate selection 4. Conclusion In this paper, we have shown that the various allomorphs of select pronominal clitics in VP can all be derived by the grammar from a single underlying form per pronoun. We have also seen that four distinct underlying structures must be posited in order to account for the four different patterns observed. The fact that geminates occur only where two syllable or other prosodic positions are available follows from geminates consisting of two identical root nodes. Table (43) summarizes both the underlying forms of the pronouns examined here and our analysis accounting for how they are realized in pre-verbal as well as post-verbal positions. That analysis relies, in part, on applying Sympathy Theory where there is no opacity, on extending positional faithfulness constraints from IO-correspondence to O-correspondence, on and extending positional stipulations from faithfulness and markedness constraints to alignment constraints. (However, taking alignment constraints to be a type of markedness constraint, this final device does not constitute much of an extension of the theory.) Finally, the investigation that we have conducted here requires us to modify, ever so slightly, some of the underlying forms posited by Auger (2003b). Specifically, she suggests that “Many pronominal clitics consist of a single consonant: // ‘I’, /m/ ‘me’, /t/ ‘2sg.non-nom’, and /s/ ‘3sg.refl’. Others consist of geminate consonants: /ll/ ‘3sg.acc’, /nn/ ‘3sg.gen’, and /zz/ ‘3pl.acc’” (Auger 2003b: 7). It remains true that many pronominal clitics consist of singleton consonants and that others consist of surface geminate consonants, but we have seen here that the underlying forms of the 3sg.gen and the 3pl.acc must be /nn/ and /lz/, respectively, and not /nn/ and /zz/, as posited by Auger (2003b). 150 CatJL 4, 2005 Brian José; Julie Auger Prindez nné ‘Take some’ /prεnde nn/ a. [prε˜.de˜n.n]*!* b. [prε˜.de˜n]*! ** ☞c. [prε˜.de˜n.ne]**** d. [prε˜.de˜.ne˜n]*! **** e. [prε˜.de˜.nε˜]*! *** MAX-O (ROOT)/ STRESS MORPH CONTIG-O MAX[nas] σ-STRUCT *WK-IN-STR DEP(V)/ STRESS DEP(V) DEP(RT) Cat.Jour.Ling. 4 001-252 7/2/06 11:45 Página 150 (43) Underlying forms and how they are handled by the grammar References Auger, Julie (2000). «Phonology, Variation, and Prosodic Structure: Word-Final Epenthesis in Vimeu Picard». In: Fontana, Josep M.; McNally, Louise; Turell, M. Teresa; Vallduví, Enric (eds.). Proceedings of the First International Conference on Language Variation in Europe (ICLaVE 1). Barcelona: Universitat Pompeu Fabra, pp. 14-24. Auger, Julie (2001). «Phonological Variation and Optimality Theory: Evidence from Word-Initial Vowel Epenthesis in Vimeu Picard». Language Variation and Change 13: 253-303. Auger, Julie (2002). «Picard Parlé, Picard Écrit: Dans quelle mesure l’écrit représentet-il l’oral?». In: Pusch, Claus; Raible, Wolfgang (eds.). Romanistische Korpuslinguistik: Korpora und gesprochene Sprache. Tübingen: Gunter Narr, pp. 265-278. Auger, Julie (2003a). «Picard Parlé, Picard Écrit: Comment s’influencent-ils l’un l’autre?». In: Landrecies, Jacques; Petit, Aimé (eds.). Bien Dire et Bien Aprandre N° 21: Picard d’hier et d’aujourd’hui. Lille: Centre d’Études Médiévales et Dialectales de Lille 3, pp 17-32. Auger, Julie (2003b). «Picard Pronominal Clitics Revisited». In: Núñez-Cedeño, Rafael; López, Luis; Cameron, Richard (eds.). A Romance Perspective on Language Knowledge and Use: Selected Papers from the 31st Linguistic Symposium on Geminates and Picard Pronominal Clitic Allomorphy CatJL 4, 2005 151 Pronoun UR Pre-verbally Post-verbally 1sg, 2sg /m, t/ faithful lengthen to [mm] and [tt] except where insufficient prosodic positions 3sg /ll/ faithful except consistently [ll] due where insufficient to positional prosodic positions faithfulness 3pl /lz/ assimilates to [zz] assimilates to [zz] and surfaces and surfaces faithfully faithfully due to due to morpheme morpheme realization realization partitive/genitive /nn/ before a vowel, lengthens to [nn] floating [nasal] and surfaces faithfully becomes onset, due to O-FAITH surfacing as [nn]; before a consonant, floating [nasal] docks onto epenthetic vowel, surfacing as [nε˜] Cat.Jour.Ling. 4 001-252 7/2/06 11:45 Página 151