Phonological and Morphological Domains In Kyirong Tibetan

Phonological and morphological domains in Kyirong Tibetan* 2 Abstract A number of phonological generalizations in the Kyirong dialect of Tibetan are argued to require reference to rules or constraints referring to the syllable or the phonological word (pword). The latter domain is significant because it requires three distinct representations for suffix-stem combinations, depending on the particular suffix involved: (a) the stem and suffix form one pword, (b) the suffix lies outside of the pword of the stem and is attached to a higher prosodic constituent, and (c) the stem and suffix are separate pwords. In addition, one phonological process will be argued to operate at the left edge of a morphological domain, i.e. the ʻstemʼ. While the parsings in (a)-(c) above are attested in many other the languages of the world, Kyirong Tibetan is unusual typologically because all but one of the stem plus suffix combinations is either (b) or (c). By contrast, in many other languages stem plus suffix sequences are typically represented as in (a). Kyirong Tibetan will also be argued to be unusual typologically because it treats both parts of compounds as single pwords and not as two separate pwords, which is probably the default option cross-linguistically. 1 Introduction In this paper we discuss a number of phonological generalizations in the Kyirong dialect of Tibetan (henceforth KT) on the basis of the excellent description of that language by Huber (2002). These generalizations subsume both segmental and suprasegmental phenomena, e.g. 3 the distribution of aspirated consonants, long vowels, contour tones, and various consonant clusters. We argue that all of these generalizations require rules or constraints referring to the two prosodic domains ʻsyllableʼ and ʻphonological wordʼ, as well as to the morphological domain ʻstemʼ. We posit a number of explicit prosodic structures for mono- and heteromorphemic words of KT. For stem-suffix combinations we argue for three structures: (a) cohering, i.e. the stem and suffix form one pword, (b) noncohering, i.e. the suffix lies outside of the pword of the stem and is attached to a higher prosodic constituent, and (c) double-pword, i.e. the stem and suffix are separate pwords. In the few words of KT which contain prefixes we argue that they are represented with the mirror image of the noncohering representation in (b). Finally, we show that compound words of KT are represented as a single pword. The present study is important for several reasons. First, although Huber (2002) refers to both the syllable and the pword in her account of KT, the relevance of these two domains is not always treated explicitly in the larger analysis. Thus, our analysis is intended to clarify what the domains are for the various phonological generalization one can observe in KT. Second, in her description of KT Huber (2002) does not include the morphological category ʻstemʼ; hence, we show the necessity of that constituent. Third, the three prosodic structures described above in (a)-(c) and for compounds illustrate the importance of KT for research in typology. Although 4 representations like the ones in (a)-(c) are not uncommon in the languages of the world, what is typically the case is that languages parse a suffixed word in a consistent fashion , in particular with the cohering structure in (a); see the literature on Prosodic Phonology, e.g. Nespor & Vogel (1986), Hall & Kleinhenz (1999). KT seems to be unique in treating almost all of its stem-suffix sequences as either noncohering (i.e. b) or double-pword (i.e. c). KT is also unusual typologically because it treats compound words as single pwords and not as two separate pwords, which is probably the default option cross-linguistically. Our account will add to existing literature on the phonology of Tibetan varieties. While there are other accounts of Tibetan phonology, most are diachronic in approach (Delancey 2003; Hari 1979; Mazaudon 1977; Sprigg 1966; 1990; Vollmann 1999). Other treatments of Tibetan phonology include grammatical descriptions of other dialects, including Häslerʼs (1999) grammar of Dege, Sunʼs descriptions of Amdo (1986), Zhongu (2003) and Sgerpa (2006), Denwoodʼs (1999) grammar of Lhasa, van Driemʼs (1992) grammar of Dzhongkha (the national Language of Bhutan), a phonetic (acoustic) analyses of tone in Dolpo and Mugom (Watters 2002), and a comparison of tone in Central and Kham dialects (Haller 1999). Virtually all of these studies cited focus either on multiple dialects from the historical-comparative perspective, or approach synchronic phenomena from a particular theoretical model (e.g. feature geometry or Optimality Theory), or leave aside questions of the intersection of phonological and 5 grammatical domains. Here we focus instead on the question of rule-domains in a single dialect. The organization of this paper is as follows. The pword domain referred to above and the relationship between this prosodic unit and morphological constituents (i.e. stems, prefixes and suffixes) form the topic of section 2. In section 3 we provide background information on the phonology and morphology of KT (on the basis of Huber (2002)). In section 4 the syllable domain for KT is discussed. Here we posit the syllable template and demonstrate that the distribution of long vowels requires reference to the syllable. In section 5 we provide evidence from the segmental and tonal phonology (i.e. the distribution of aspirated consonants, onset clusters, contour tones and coda consonants) for the pword as defined in section 2. In section 6 we posit that the domain for ʻintrusive nasalsʼ is the morphological constituent ʻstemʼ. In section 7 we provide a summary of the prosodic structures posited in our treatment of KT, compare these structures with similar ones in other languages and discuss the relevance of our analysis for the research program known as Prosodic Phonology (Nespor & Vogel 1986 and many subsequent studies). Section 8 concludes. 2 Prosodic and morphological domains In our treatment of KT we discuss a number of phonological generalizations that hold within either phonological and morphological domains. The morphological categories referred to here are the ʻstemʼ, ʻprefixʼ, ʻsuffixʼ, and the ʻgrammatical wordʼ (henceforth ʻgwordʼ); the relevant 6 phonological domains include only the ʻsyllableʼand the pword. In this section we illustrate the intricate relationship between the pword and the morphological constituents referred to above because this is one aspect of our analysis that is not made explicit in Huber (2002). Our goal here is not to posit explicit algorithms which parse pwords on the basis of morphological categories (see, for example, Nespor & Vogel (1986), as well as many more recent studies within the Optimality Theoretic framework). Instead, we simply show what the pword for KT looks like with respect to the morphology of that language. In section 7 we return to these structures and compare them with the corresponding ones for other languages. The relationship between the morphological domains ʻgwordʼ, ʻstemʼ and ʻprefixʼ and the phonological domain ʻpwordʼ is illustrated in (1). Here we can observe the structure of monomorphemic words (i.e. bare stems) in (1a), compound words in (1b) and prefixed words in (1c).1 In these structures ʻCʼ and ʻVʼ represent segmental material (i.e. a consoant and a vowel respectively) and ʻσʼ the syllable. 7 (1) Relationship between pwords and bare stems, compounds and prefixed forms: a. Bare stem: gword b. Compound: gword gword stem stem CVC C VC CVC σ σ c. Prefixed word: stem σ prefix stem C VC CVC σ σ pword pword pword pphrase pphrase pphrase In (1a) we see the structure of a (monomorphemic) word, which consists of a stem and a gword. The prosodic constituents include the syllable, the pword and the next highest level in the prosodic hierarchy, e.g. the phonological phrase (=pphrase in (1)). Compound words of KT will be argued to have the structure in (1b), where the two parts (i.e. stems) make up both a gword and a single pword. Prefixed words have the structure in (1c): The stem is a separate pword and the prefix is noncohering in the sense that it is situated outside of it. We assume that a full noncohering representation is one in which the syllable of the prefix is linked to the prosodic constituent higher than the pword, which we assume here to be the pphrase.2 8 Suffixed words of KT are more complex because there are three possibilities depending on which suffix is involved. These three possible structures are presented in (2a-c): (2) Relationship between pwords and suffixed forms: a. Cohering gword b. Noncohering c. Double-pword gword stem suffix gword stem suffix CVC CVC CVC CVC σ σ σ pword pword pphrase pphrase σ stem suffix CVC C VC σ σ pword pword pphrase In (2a) we see that the stem and the suffix together form a pword, in (2b) that the suffix is situated outside of the pword and that it is linked to the next highest category in the prosodic hierarchy and in (2c) that both the stem and the suffix form their own pwords. We refer henceforth to (2a-c) as ʻcoheringʼ, ʻnoncoheringʼ and ʻdouble-pwordʼstructures respectively. 9 It will be shown below in section 7 that one cannot predict which suffix of KT belongs to the three structures in (2); hence, the language requires suffixes to be lexically marked for the type of prosodic structure is required. 3 Background information on KT The Kyirong dialect (skyid-gron) of Tibetan is spoken by under one thousand people in Kyirong County, Western Central Tibet, just across the border from Nepal. KT is a Bodish, Tibetic language of the Tibeto-Burman language family. Geographically it is classified as one of the Tsang dialects of Central Tibetan. It has similar phonological properties to other Tibetan dialects spoken in Nepal, including Langtang, Helambu and Kagate Tibetan (Huber 2002: 3-5). Unlike other U-Tsang dialects like Lhasa Tibetan, however, (Denwood 1999), Kyirong Tibetan posseses onset clusters. The nature of these clusters is discussed in more detail in section 4. KT is in contact with and has borrowed lexical items from Tamang (Tibeto-Burman) Nepali (Indo-Aryan) languages, and presumably also shares some vocabulary with other Central Tibetan dialects (e.g. honorific terms) and Standard Spoken Tibetan (Huber 2002: 11-12). In the present study we include only those words which correspond with written Tibetan forms, which means that the word is traceable within the Tibetan group at last 500 years back. In (3) we have listed the consonant and glide phonemes of KT (Huber 2002: 15). 10 (3) KT Consonant and glide phonemes Bilabial Alveo-Dental Retroflex Alveo-Palatal Palatal Velar Glottal Stops plain voiceless p t  c k aspirated p t  c k voiced b d   g Affricates plain voiceless ts t aspirated ts t voiced dz d voiceless s  h voiced z   Fricatives Nasals m n Lateral l Lateral Fricative  Vibrant r Glide w  ŋ j Three of the consonants (i.e. /m n l/) may be found as geminates, but only in word-medial position, e.g. [ɲēnnɛ:] ʻfastingʼ (Huber 2002:19). In section 4.2 we argue that KT also possesses the four prenasalized stops [ndz nd ndʑmb], which Huber does not include in her chart of phonemes. 11 The KT vowel phonemes are listed in (4): (4) KT Vowel phonemes FRONT I FRONT II CLOSE i, i, y, y, y u, u, ũ HALF-CLOSE e, e, ẽ ø, ø, ø o, o, õ HALF-OPEN OPEN CENTRAL BACK ɛ, ɛ, ɛ: a, a, ã: Two important groupings reflected in (4) are what Huber (2002) calls “primary” and “secondary” vowels. The latter segments can be equated with the FRONT II column, while all remaining vowels belong to the primary category (p. 21).3 The distinction between primary and secondary vowels is important because the latter segments show various distributional regularities, e.g. short secondary vowels are followed by a plain voiceless consonant (e.g. [p]) but long secondary vowels by a voiced consonant (e.g. [b]) (Huber 2002: 21). Note that the categories ʻprimary vowelsʼ and ʻsecondary vowelsʼ are not natural classes in traditional phonology and that these groupings therefore cannot be captured with a single set of distinctive features. This is a theoretical problem we do not deal with below. Note that (4) nasalized vowels are phonemic in KT, although they are always long. A reviewer points out to us that the inventory of KT is unusual typologically in that it possesses only long nasal vowels but no short ones. One might therefore want to analyze the long nasal 12 vowels as underlyingly short and write a context-free default rule specifying that nasal vowels are long. This is an option we leave open. KT also has a tone system, based partially on the melodic properties of vowels and partially on vowel phonation type. There are three relative pitch heights (i.e. register tones): /v/ (high pitch with a modal vowel phonation), /v/ (mid, slightly rising pitch with a modal vowel phonation) and /v/ (low, slightly rising pitch with a breathy or lax vowel phonation) (Huber 2002: 23). There are also three falling contour tones in KT, all with lengthened vowel durations in comparison with the register tones: /v/ (high falling pitch with a creaky vowel phonation), /v/ (mid-falling pitch with a creaky vowel phonation) and /v/ (low falling pitch with a creaky vowel phonation). Due to the nature of diachronic tonogenesis in KT, there are some restrictions as to which onset consonants appear before particular tones. Concerning morphological structure, KT possesses many inflectional and derivational suffixes but only a small number of prefixes. The majority of disyllabic gwords in KT (as well as other Sino-Tibetan languages) are compounds, which are analyzable into (mostly) monosyllabic constituent morphemes (Matisoff 1991: 492). 4 The syllable domain In this section we discuss the role of the syllable in KT, namely the syllable template (section 4.1) and the distribution of long vowels (section 4.2). 13 4.1 The syllable template The KT syllable consists of an obligatory vocalic nucleus (either short V or long V) and an optional onset consisting of one nonsyllabic segment (i.e. consonant or glide) and an optional coda consisting of a single nonsyllabic segment. The syllable template for KT is presented in (5). Since our analysis does not require reference to subsyllabic structure (i.e. the constituents ʻonsetʼ, ʻnucleusʼ, ʻcodaʼ), we simply employ the flat structure in (5). ʻCʼ is an abbreviation for a nonsyllabic segment and ‒ unless otherwise noted ‒ ʻVʼ designates either a short or a long vowel. (5) KT syllable template: σ (C) V (C) The final C in (5) is restricted to /p m ŋ j r/ (Huber 2002: 27); we ultimately argue that the domain for this restriction is the pword (see section 5.4) and not the syllable. All consonants in the inventory in (3) are found in word-initial position in single onset CV(C) syllables (Huber 2002: 26). Two representative examples for [n] and [l] are presented in (6a). In (6b) we can see examples of KT words of the form ʻ V.CV ʼ, in which the ʻCʼ is occupied by [n] and [k] respectively. We follow the uncontroversial view that an intervocalic consonant is (universally) in onset position, hence ʻ VCV ” is parsed ʻ V.CV ʼ and not ʻ VC.V ʼ.4 14 (6) Examples of CV(C) syllables: a. word-intitial [lū] ʻsongʼ Glo., p. 12 [nāŋ] ʻthe day after tomorrowʼ Glo., p. 14 b. word-internal in ʻ VCV ʼ: [ā.nī] ʻpaternal auntʼ Glo., p. 1 [jā.kā] ʻsummerʼ Glo., p. 6 There are various restrictions concerning the kind of C which can and cannot occur in V.CV, the most important being that aspirated consonants are not permitted in this position. These systematic gaps are discussed in section 5.1 below. The syllable template in (5) can include a second onset consonant in word-initial position. The possible word-initial CC sequences are listed in the table in (7). In (8) we list an example of all five clusters. 15 (7) Word-initial CC clusters: p pʰ b (8) r j + + + + + Word-initial CC clusters of KT: [brɛ̖ː ] ʻriceʼ Glo., p. 1 [pru̖ː ] ʻto shake out, shake somethingʼ Glo., p. 18 [pʰrā.mī] ʻnarrowʼ Glo., p. 19 [pjāŋ] ʻto suspendʼ Glo., p. 64 [pʰjàː] ʻto sweepʼ Glo., p. 88 We show in section 4.2 that all two member consonant clusters situated between vowels (see (9) below) are consistently heterosyllabified in KT and therefore conclude that initial CC clusters like the ones in (8) are only possible at the left edge of a pword (section 5.3). In (9) we provide a table illustrating all word-medial CC clusters: 16 (9) Word-medial CC clusters: p k b t d ts + dz ʈ g t dʑ  + p + s m + + + + m + ŋ + j + + + + + + + + + + n r + b j l + + + + + + + + We take no position concerning which of the gaps in (9) are systematic and which are accidental. Examples of words containing the sequences in (9) are presented in (10). In (10a) the first C is an obstruent and in (10b) it is a sonorant. (10) Word-medial CC clusters of KT: a. [tsōkpā] ʻdirty, badʼ Glo., p. 28 [caʱktsɛ:] ʻmuch, manyʼ Glo., p. 2 [t īksona] ʻprobablyʼ Glo., p. 30 [t ūpty:] ʻseventeenʼ Glo., p. 31 [roptsø̅:] ʻroughlyʼ Glo., p 20 [kuʱpt ū] ʻninetyʼ Glo., p. 9 [ʈāp ɛ:] ʻcomb, hairbrushʼ Glo., p. 26 [gablɛ̅:] ʻafterʼ Glo., p. 5 [t abrù:] ʻyoung birdʼ Huber, p.c. 17 b. [t ōbjɛ:] (+[ o]) ʻto go to a girlʼ Huber, p.c. [bøjbō] ʻsoftʼ Glo., p. 1 [pʰɛjt e:] ʻprobably, most likelyʼ Glo., p. 19 [pɛj ɛ:] ʻwool combʼ Glo., p. 17 [sījʈīp] ʻshade, shady sideʼ Glo., p. 21 [tām ā] ʻbeansʼ Glo., p. 23 [cʰīmdzè:] ʻneighborʼ Glo., p. 3 [sūmgā:] ʻthe threeʼ Glo., p. 22 [āmdʑī] ʻdoctorʼ Glo., p. 1 [lamsā:] ʻimmediate, instantlyʼ Glo., p. 11 [sāmlō] ʻwith taŋ- in 'to thinkʼ Glo., p. 21 [doŋbō] ʻanimal with a blazeʼ Glo., p. 4 [āŋdũ:] ʻwomanʼs dressʼ Glo., p. 1 [braŋɟɛ:] ʻbutter offeringʼ Glo., p. 1 [tiŋsā:] ʻnowadaysʼ Glo., p. 24 [riŋmō] ʻlongʼ Glo., p. 20 [cāŋlāŋ] ʻalways?ʼ Glo., p. 2 [kaŋrī] ʻsnow mountainʼ Glo., p. 8 18 [den. ɛ:] ʻlike thatʼ Glo., p. 4 It should also be noted that KT also possesses word-internal [nd mb] in words like mendā ʻgunʼ (Glo., p. 13) and jāmbā ʻfrying panʼ (Glol, p. 46). As we show in section 6, there are good reasons for assuming these nasal-stop sequences to be single segments. 4.2 The distribution of long vowels The examples in (11) are intended to illustrate that long and short primary vowels contrast in KT in word-final position, both in disyllables (in 11a) and in monosyllables (in 11b).5 (11) Long and short primary vowels contrast in word-final position: a. In disyllables: [ātsī] ʻan exclamationʼ Glo, p. 1 [am.di] ʻdoctorʼ Glo., p. 1 [cip.e] ʻgingerʼ Glo, p. 3 [jaŋ.de] ʻlight weightʼ Glo, p. 6 [tsī.ū] ʻrespectʼ p. 21 [ū.ū] ʻpaperʼ p. 21 [za.ō] ʻweekʼ p. 20 [c.ō] ʻmain doorʼ p. 20 19 [kʰā.bā] ʻloudspeakerʼ p. 20 [kʰā.bā] ʻtranslatorʼ p. 20 b. In monosyllables: [mi] ʻeyeʼ Glo, p. 13 [mi] ʻmanʼ Glo., p. 13 [pe] ʻto openʼ Glo., p. 17 [pē] ʻexample, sayingʼ Glo., p. 17 [ku̖:] ʻto waitʼ Glo, p. 9 [ku] ʻnineʼ Glo, p. 9 [lo] ʻto readʼ Glo, p. 12 [lo] ʻyearʼ Glo, p. 12 [pʰā:] ʻthither, awayʼ Glo., p. 65 [ʈʰā] ʻbird of preyʼ Glo., p. 89 The contrast illustrated in (11) only holds for oral primary vowels because nasal primary (and secondary) vowels are always long. In the remainder of this section our comments regarding the distribution of long and short vowels of KT therefore hold only for oral vowels. Huber does not say so explicitly, but she writes (p. 21-22) that only the long secondary vowels [y ø ] seem to occur in final position. This can be illustrated with the words in (12), in 20 which [y ø ] occur in disyllabic words (including compounds), as in (12a), and in monosyllabic words, as in (12b): (12) Only long secondary vowels in word-final position: a. [ta.ky] ʻwireʼ Glo., p. 30 [ty.tsø] ʻtimeʼ Glo., p. 25 [pj.] ʻwool combʼ Glo., p. 17 ʻto give, offer (honorific) Glo, p. 20 [tø] ʻto show, take outʼ Glo, p. 25 [k] ʻlanguage, voice, soundʼ Glo, p. 8 b. [py] By contrast, the corresponding short secondary vowels [y ø ɛ] are nonoccurring in word-final position ‒ a gap we discuss briefly in section 6.6 The additional data in (13) illustrate that the contrast between long and short primary vowels also holds word-medially before a single consonant: (13) Long and short primary contrast before a word-medial C: [ɲi:ba] ʻsecondʼ Glo., p. 15 [ɲī.bo] ʻthe twoʼ Glo., p. 15 21 [dze:.bō] ʻbeautifulʼ p. 20 [dze.bō] ʻleperʼ p. 20 [tʰu:tē] ʻThupten (person)ʼ Glo., p. 26 [t ʰu.bā] ʻtraditional dressʼ Glo., p. 32 [kò:.la] ʻunderʼ Glo., p. 9 [ko.lē] ʻslowlyʼ Glo., p. 9 [ca:.cà:] ʻwith ma- to jokeʼ Glo., p. 2 [ca.sā] ʻtownʼ Glo., p. 2 The reason the examples in (13) are important is that the first vowel is syllable-final (due to the parsing of VCV as V.CV); hence, the data in (11) and (13) together show that short and long primary vowels contrast syllable-finally.7 The examples in (14) illustrate that the short vs. long contrast in word-medial position before a single C can also be observed for secondary vowels: (14) Long and short secondary vowels contrast before a word-medial C: [kȳ.tø:] ʻthreadʼ Glo., p. 9 [ky:bō] ʻexpensiveʼ Glo., p. 9 [pø:.bā] ʻstrong, mightyʼ Glo, p. 18 [pø:.pā] ʻTibetan Glo, p. 18 22 [kɛ̅:.cū:] ʻtranslatorʼ Glo., p. 8 [kɛ̅.t ā] ʻspeechʼ Glo., p. 8 The data in (11 and 13-14) together show that long and short vowels ‒ both primary and secondary ‒ contrast syllable-finally with the provision that short secondary vowels are nonoccurring word-finally. The data in (15) (some of which are repeated from (10)) illustrate that the contrast between long and short vowels is neutralized to the corresponding short segments before ʻCCVʼ. Put differently, the Va in ʻ...VaCCVb...ʼ is consistently short.8 In (15a) the first C in ʻ...VaCCVb...ʼ is an obstruent and in (15b) it is a sonorant. Given the parsing ʻ...VaC.CVb...ʼ as indicated below, the generalization is that only short vowels ‒ both primary and secondary ‒ can occur in a closed syllable. (15) Va is always short in ...VaCCVb... a. b. [tsōk.pā] ʻdirty, badʼ Glo., p. 28 [rop.tsø̅:] ʻroughlyʼ Glo., p 20 [gab.lɛ̅:] ʻafterʼ Glo., p. 5 [t ab.rù:] ʻyoung birdʼ Huber, p.c. [t ōb.jɛ:] (+[ o]) ʻto go to a girlʼ Huber, p.c. [cʰīm.dzè:] ʻneighborʼ Glo., p. 3 23 [ām.dʑī] [ām.dʑò:] ʻdoctorʼ ʻearʼ Glo., p. 1 Glo., p. 1 [sām.lō] ʻwith taŋ- in 'to thinkʼ Glo., p. 21 [doŋ.bō] ʻanimal with a blazeʼ Glo., p. 4 [braŋ.ɟɛ:] ʻbutter offeringʼ Glo., p. 1 [riŋ.mō] ʻlongʼ Glo., p. 20 [duŋ.mō] ʻchurnʼ Glo., p. 4 [cāŋ.lāŋ] ʻalways?ʼ Glo., p. 2 [kaŋ.rī] ʻsnow mountainʼ Glo., p. 8 [den. ɛ:] ʻlike thatʼ Glo., p. 4 The important point is that the absence of long vowels before ʻ...CCV....ʼ makes sense if the consonants are heterosyllabified. This absence would not derive a straightforward explanation if they were tautosyllabic, i.e. ʻ...Va .CCVb...ʼ The data in (15) demonstrate that KT avoids word-internal closed syllables which contain a long vowel. It is important to stress that this generalization extends to word-final closed syllables as well. Thus, in (16) we can observe that the contrast between short and long vowels is neutralized to the former in word-final closed syllables: 24 (16) Va is always short in ...VaC# [tʰīp] ʻto lightʼ [cʰīm] ʻhouseʼ [nāŋ] Glo., p. 26 Glo., p. 3 ʻthe day after tomorrowʼ Glo., p. 14 [pʰȳj] ʻto give, to offer (honorific)ʼGlo., p. 20 [tēr] ʻto giveʼ Glo., p. 24 The data in (15) and (16) together tell us that KT avoids closed syllables with a long vowel and that it does not matter whether or not the closed syllable is word-internal or word-final. Following Huber (2002: 20) we hold that long vowels can only occur in an open syllable ‒ a generalization stated in (17): (17) Long Vowel Constraint: Long vowels occur (and are contrastive only) in an open syllable. It should be noted that Huber does not posit a phonotactic constraint like the one in (17), nor does she transcribe KT words like the ones in (15) with syllable boundaries. Additional evidence for the VC.CV parsing in (15) and for the necessity of the Long Vowel Constraint is that an underlying long vowel (primary or secondary) will shorten if it is situated in a closed syllable. Examples illustrating the process we refer to below as Closed Syllable Shortening are provided in (18). This process is not discussed in Huber (2002), but all of the 25 examples have been gleaned from her glossary. In the examples below we have listed words in which the closed syllable comes into being by morpheme concatenation. See the third column under (18), in which the respective underlying forms with the corresponding morpheme boundaries (represented here and below as ʻ‒ʻ) have been provided. Since these examples are not discussed in Huberʼs (2002) text we do not know whether or not the morpheme parsings in the third column of (18) are productive synchronically or if they are simply a historical remnant.9 The underlying forms will be justified below. (18) Closed syllable shortening examples: [ʈu̖:] ʻsixʼ /ʈu:/ Glo., p. 27 [ʈup.t ū] ʻsixtyʼ /ʈu:-pt ū/ Glo., p. 27 [ʈuk.pā] ʻsixthʼ /ʈu:-kpā] Glo., p. 27 [pɛ:] ʻwoolʼ /pɛ:/ Glo., p. 17 [pɛj. ɛ:] ʻwool combʼ /pɛ:-j ɛ:/ Glo., p. 17 [pȳ̃:] ʻrelativeʼ /pȳ̃:/ Glo., p. 19 [pȳŋ.ɟà:] ʻrelativeʼ /pȳ̃:-ɟà:/ Glo., p. 19 The alternant with the long vowel in (18) cannot come about by a rule lengthening underlying short vowels in word-final position because KT contrasts short and long vowels in this position 26 (recall the examples in 11). The alternations in (18) therefore require underlying forms with long vowels and the rule of Closed Syllable Shortening in (19): (19) Closed Syllable Shortening: V ⇒ V / ̲̲ C] Apparent counterexamples to the analysis proposed up to this point are presented in (20). These words illustrate that long vowels can stand before CCV under the condition that CC are a homorganic nasal plus stop (NS) cluster: (20) Long vowels before homorganic NS clusters: [cī:ndzù:] ʻmiddle fingerʼ / cī:-ndzù:/ Glo., p. 3 [ɦo:ndō:] ʻpitʼ / ɦo:-ndō:/ Glo., p. 6 [ba:ndʑām] ʻmask danceʼ / ba:-ndʑām / p. 48 [ŋɛ̅:mbø̅:] ʻpillowʼ / ŋɛ̅:-mbø̅:/ p. 27 We consider the NS in (20) to be a single segment, i.e. a prenasalized stop. The reason for our assumption is twofold: First, if the NS in (20) consisted of two separate heterosyllabic segments (e.g. [cī:ndzù:] for the first example) then the otherwise exceptionless generalization concerning the distribution of long vowels would be lost. Were the NS in (20) two separate tautosyllabic segments (e.g. the nasal [n] plus affricate [dz] in [cī:ndzù:]) then our treatment could not explain why these are the only word-medial tautosyllabic consonant-clusters in the 27 language at all and that they blatantly violate sonority sequencing. It is also worth noting that in other Tibetan dialects prenasalized stops are attested, e.g. Dege (see Häsler 1999) and Lhasa (Denwood 1999: 76). 5 The pword domain In this section we consider four areas of KT phonological structure in which the pword plays a decisive role, namely the distribution of aspirated consonants (section 5.1), contour tones (section 5.2), onset clusters (section 5.3) and coda consonants (section 5.4). 5.1 The distribution of aspirated consonants The following words illustrate a contrast between voiceless aspirated stops (in 21a) and plain voiceless (i.e. unaspirated) stops (in 21b) in pword-initial position. (21) Contrast between aspirated and plain voiceless stops in pword-initial position: a. [pʰā:] ʻthither, awayʼ Glo., p. 65 [tʰā] ʻto surround (with [kō(r)-]ʼ Glo., p. 25 [ʈʰā] ʻbird of preyʼ Glo., p. 89 [cʰàː] ʻyou (pl.)ʼ Glo., p. 3 [kʰā] ʻmouthʼ Glo., p. 9 [tsʰā] ʻsaltʼ Glo., p. 29 28 b. [t ʰā] ʻto prepareʼ Glo., p. 31 [pāː] ʻpictureʼ Glo., p. 17 [tā] ʻto look, watchʼ Glo., p. 23 [ʈā] ʻhairʼ Glo., p. 26 [cā] ʻto change place, bringʼ Glo., p. 2 [kā] ʻorderʼ Glo., p. 7 [tsàː] ʻto filterʼ Glo., p. 28 [t āː] ʻa littleʼ Glo., p. 30 The vowel following the pword-initial segment in (21) is [a] or [a]. The additional examples in (22) illustrate that there is no correlation between aspiration and the quality of the following vowel, since any vowel can follow an initial stop or affricate. (22) Contrast between aspirated and plain voiceless stops in pword-initial position: a. [pʰȳj] ʻto give, to offer (honorific)ʼ Glo., p. 20 [tʰīp] ʻto lightʼ Glo., p. 26 [tʰøː] ʻto hearʼ Glo., p. 26 [ʈʰī] ʻseat, chair, throneʼ Glo., p. 27 [ʈʰū] ʻto washʼ Glo., p. 28 [cʰīm] ʻhouseʼ Glo., p. 3 29 b. [kʰùː] ʻable to doʼ Glo., p. 10 [tsʰūː] 'hither' Glo., p. 29 [t ʰū] ʻwaterʼ Glo., p. 32 [pi̖ː] ʻto doʼ Glo., p. 18 [pū] ʻbody hairʼ Glo., p. 18 [tĩ̖ː] ʻbottom of vesselʼ Glo., p. 24 [tu̖ː] ʻpoisonʼ Glo., p. 25 [ʈi] ʻknifeʼ (n.coll) Glo., p. 27 [ʈu̖ː] ʻsixʼ Glo., 27 [cīː] ʻmiddleʼ Glo., p. 3 [kuʱː] ʻnineʼ Glo., p. 9 [tsī] ʻvarnishʼ Glo., p. 28 [tsùː] ʻto pierce a holeʼ Glo., p. 29 [t ī] ʻwhatʼ Glo., p. 30 [t ū] ʻtenʼ Glo., p. 30 An important phonotactic generalization regarding KT is that aspirated consonants occur in no other context, i.e. neither pword-internally (in syllable-initial or in syllable-final position), nor 30 pword-finally. We posit the phonotactic constraint in (23), according to which aspirated consonants only surface in pword-initial position:10 (23) Aspiration Constraint: Aspirated consonants only occur in pword-initial position. The examples in (24) illustrate that the Aspiration Constraint is responsible for causing an underlying aspirated consonant to lose its aspiration if it occurs in the second half of a compound word.11 The aspirated consonant in the second part of the compounds is present in the underlying form because it surfaces as such if it is a free morpheme, i.e. /t ʰū/ ʻwaterʼ, /kʰāŋbā/ ʻhouseʼ, /t ʰūŋī/ ʻsmallʼ. The voicing of the deaspirated consonant in (24) is caused by an independent process of Voicing Assimilation we do not discuss here. In (24) and below ʻ>ʼ indicates a historical change. Not that the final syllables of the two stems in the final line and the final syllable for the morpheme ʻhouseʼ are deleted in the phonetic form; we do not discuss these clippings below because we cannot determine their productivity. (24) Deaspiration and Voicing Assimilation: /nākòː/ ʻnoseʼ + /t ʰū/ ʻwaterʼ > [nā.dzū] ʻnose mucousʼ p. 28 /mẽ̅ː/ ʻmedicineʼ + /kʰāŋbā/ ʻhouseʼ > [mē.ŋgã̅] ʻhospitalʼ p. 28 /sømbā/ ʻshowʼ + /t ʰūŋī/ ʻsmallʼ > [sø.ndʑũ̅ː] ʻsmall shoeʼ p. 28 31 What the examples in (24) suggest is that both parts of the compound form a single pword, as in (1b), and that this prosodic structure triggers a (historical) rule of Deaspiration, which comes about because the aspirated consonant is no longer in pword-initial position.12 The examples in (25) illustrate that the negation prefix ma- can stand before an aspirated consonant. There are few prefixes in KY; the prefix ma- is the only one we have found that occurs in this context: (25) Aspirated consonants after the prefix ma- NEG: [kʰùː] /kʰùː/ ʻto be able to doʼ Glo., p. 10 [ma.kʰùː.so] /ma-kʰùː-so/ ʻNEG-be.able- AOR. SENSʼ p. 219 [pʰòː] /pʰòː/ ʻto touchʼ Glo., p. 19 [ma.pʰòː] /ma-pʰòː/ ʻNEG -touchʼ p. 241 The prefixed examples in (25) require that ma- be analyzed as phonologically noncohering; hence, the words in (25) have the prosodic structure in (26), which is the equivalent of (1c):13 (26) Prosodic structure of the examples in (25): ma-(kʰùː)PWD-so ma-(pʰòː)PWD The examples in (27) illustrate that KT possesses two suffixes beginning with an aspirated stop: the Possibility auxiliary -tsʰø:nu: (27a), which must follow a genitive variant of the 32 nominalizer -ba, and also the a suffix glossed ʻuntilʼ, -tʰø:, which must follow an imperfective, reduplicated verb and must precede the locative suffix, in (27b): (27) Examples of suffixes with aspirated consonants: a. [ta: kʰō lɛ:gā pi:-jo-ba-tsʰø:nu:] now 3.SG work do- IPFV- NOM- SEEMS p. 191 ʻHe seems to be workingʼ b. [tō sa-sā-tʰø:-la lakpā ʈʰỳ:] food eat-until- LOC hand wash p. 224 ʻWash your hands before eating!ʼ The examples in (27) are significant because they pose a potential problem with the Aspiration Constraint. Were the suffix to belong to the same pword as the stem, the generalization concerning the distribution of aspirated consonant would be lost (i.e. that aspiration only occurs pword-initially). What the examples in (27) therefore suggest is that the suffixes -tsʰøːnuː and - tʰøː form their own pwords, as in (28). Recall from section 2 that this is the ʻdouble-pwordʼ representation in (2c). (28) Prosodic structure of suffixed words: a. taː kʰō lɛːgā (piː-jo-ba)PWD -(tsʰøːnuː)PWD b. tō (sa-sā)PWD -(tʰø ː)PWD -la lakpā tʰỳː 33 Independent evidence that -tʰøː is a separate pword will be discussed in section 5.2.14 The words in (27) are important because they show that a reanalysis of the Aspiration Constraint in terms of a morphological domain will not account for all of the KT data. Two alternatives along these lines is that aspirated consonants only surface (a) at the left edge of a stem or (b) at the left edge of a gword. Both (a) and (b) are problematic because they cannot capture the examples in (27), i.e. aspirated consonants occur here at the left edge of a suffix, which is, by definition, neither a stem, nor a pword. 5.2 The distribution of contour tones According to Huber (2002: 25) the falling tones of KT are restricted to monosyllabic words and to the word-final syllable of polysyllabic words. All vowels bearing falling tones are long. Examples of gwords in which falling tones are situated on the final syllable are presented in (29). The examples in (29a) consist of two syllables and the ones in (29b) of one syllable. In all of these examples the vowel bearing the contour tone is long. In (29) and below we only include data with the high falling tone, but Huberʼs observation also holds for the mid falling and low falling tones as well. (29) Contour tones in gword-final position: a. [pɛj. ɛː] ʻwool combʼ p. 25 [ʈiː.màː] ʻshadowʼ p. 25 34 b. [tʰī.cùː] ʻrulerʼ p. 25 [ u.rìː] ʻkinds of wheatʼ p. 25 [jik.t èː] ʻbeliefʼ p. 25 [sāŋ.bøː] ʻnext yearʼ p. 25 [kʰøː] ʻhe.ERGʼ p. 76 [cʰøː] ʻyou (middle grade, honorific)ʼ p. 81 [ òː] ʻcome!ʼ p. 87 [ɲìː] ʻtwoʼ p. 90 We argue that the distribution of the falling tones is captured by reference to pword-final position, as expressed in the following constraint:15 (30) Falling Tone Constraint: The falling tones may occur only on the final syllable of a pword. Note that all of the words in (29) are uncontroversially single pwords, since a single gword is a pword by default (recall 1a). We consider and reject possible morphological domains for the Falling Tone Constraint below. The following examples consist of a stem plus suffix. For transparency we include the morpheme boundary in the phonetic forms. The significance of these words lies in the fact that the contour tone is situated on the final syllable of the stem. 35 (31) Contour tones gword-internally: [mɛː-de] ʻsaid-NFʼ p. 70 [tsùː-de] ʻplanted-NFʼ p. 194 [pèː-de] ʻopened-NFʼ p. 77 [tsàː-gonuː] ʻfilter-IPFV.SENSʼ p. 76 [lòː-gonuː] ʻread-IPFV.SENSʼ p. 153 [tʰòː-la] ʻtop-LOCʼ p. 78 [kòː-la] ʻunder-LOCʼ p. 112 [nākòː-la] ʻnose-LOCʼ p. 170 [ o-n ō-tʰùː-la-ta] ʻgo-REDUPL-until-LOC-EMPHʼ p. 109 [cʰīmtsèː-ɟa] ʻneighbor-PLʼ p. 83 [cʰøː-ɟa] ʻyou (sg.)-PLʼ p. 83 [mɛː-o] ʻsaid-FUT.GUARʼ p. 85 [jɛː-o] ʻitched-FUT.GUARʼ p. 122 [mɛː-bamano] ʻsaid-ASSERʼ p. 87 [cèː-bamano] ʻbe.born-ASSERʼ p. 108 [mɛ:-so] ʻsaid-AOR.SENSʼ p. 118 [tʰøː-so] ʻheard-AOR.SENSʼ p. 119 36 [kỳː-jøː] ʻstolen-PRESPERFʼ p. 150 [kàː-jøː] ʻstopped-PRESPERFʼ p. 150 [mi- èː-jøː] ʻNEG-know-IPFV.EXPERʼ p. 209 [mi-kʰùː-bɛː] ʻNEG-be.able-IPFVʼ p. 179 [mi-tòː-bɛː] ʻNEG-feel.hungry-IPFV.SENSʼ p. 219 [tʰùː-gẽ] ʻmeet-NOMʼ p. 79 [tʰùː-si-nu] ʻbecome.deaf-ACCOMPL-PERFʼ p. 87 [pùː-nu] ʻpierce-PERF.SENSʼ p. 109 [ èː-bo] ʻtell-AOR.SENSʼ p. 120 [paʱkpàː-de] ʻmotorbike-DEFʼ p. 138 [cʰàː-gojøː] ʻfeel.cold-IPFV.EXPERʼ p. 153 [ʈʰỳː-bajimbɛː] ʻwashed-AOR.INFERʼ p. 165 [tʰøː-ba] ʻhear-NOMʼ p. 180 [mɛː-ba] ʻsaid-AOR.VOLʼ p. 209 [søː-bi] ʻremember-NOM.GENʼ p. 187 [t ùː-gē-nā] ʻlet-IPFV.GENER-Qʼ p. 207 [tsèː-nā] ʻplay-ifʼ p. 211 We argue that the suffix in all of the examples in (31) is noncohering, as in (32) (=(2b)). 37 (32) Prosodic structure of a representative example of the words in (31): (mɛː)PWD de (i.e. stem-suffix) The pword of the stem in (32) derives evidence from the Falling Tone Constraint. Although we have no positive evidence from other aspects of KT phonology for this parsing, it is worth noting that the structure in (32) is consistent with the analysis in section 4.2. For example, we see in (32) that the long vowel stands in an open syllable, in conformance with the Long Vowel Constraint. An alternative to (32) is that these suffixes have the double-pword representation in (2c). This is an alternative which is possible in the sense that there is no evidence against it. We therefore adopt the noncohering structure in (32) until positive evidence can be adduced that the suffixes in (31) are their own pwords. The following examples are two member compound words. A comparison of the compounds in the first column with the input forms in the third column reveal that the falling tone on the first morpheme reduce to a high level tone in the corresponding morpheme in the first column: (33) Contour tone lost in the first member of a compound: [māːmī] ʻsoldierʼ [t ākỳː] ʻwireʼ /t àː/ ʻironʼ + /kytòː/ [ŋɛ̅ːmbø̅ː] ʻpillowʼ /màː/ ʻfightingʼ + /mi/ /ŋɛː/ ʻpillowʼ+ /bøjbō/ ʻpersonʼ p. 29 ʻthreadʼ p. 66 ʻsoftʼ p. 27 38 [t āːɟā] ʻiron ladleʼ /t àː/ ʻironʼ + /ɟā/ ʻwooden pot ladle Glo., p. 30, 2 We posit that compounds like the ones in (33) are single pwords, as in (34). Note the contrastive prosodic structure in (34) vs. the one in (32). (34) Prosodic structure of a representative example of the words in (33): (t ā-kỳː) PWD If this is the correct prosodic structure then the deletion of the falling tone in (33) follows from the rule in (35), where F = falling tone, H = high, and T = any tone: (35) Falling Tone Deletion: (F T) PWD ⇒ (H T) PWD This rule says that a falling tone is converted into a high level tone before another syllable bearing tone if the two tone bearing units belong to the same pword. Note that the elimination of the falling tone via rule in (35) is a way KT repairs potential violations to the Falling Tone Constraint, which prohibits a falling tone from occurring within a pword. It should be noted that Falling Tone Deletion seems to apply in productive compounds (as in the first two in (33), Huber personal communication). The following examples show that Falling Tone Deletion applies before the nominalizer suffix /-pa/. These data are significant because the falling tone is retained on the stem before 39 the [-ba] suffixes presented in (31), i.e. [tʰøː-ba] ʻhear-NOMʼ and [mɛː-ba] ʻsaid-AOR.VOLʼ. Thus, there is a contrast between the preceding words in (31) with [‒ba] and the ones in (36). 16 (36) Contour tone lost before suffix ‒ba: [t ø̅ːbā] ʻtemple-keeperʼ /t øː/ ʻreligionʼ + /-pa/ p. 29 [døpā] ʻwishʼ /dø̖/ + /-pa/ Glo., p. 4 The examples in (36) require that the suffix ‒pa cohere with the stem, as in (37a-b). That the homophonous suffix ‒pa does not cohere in the words in (31) can be accounted for if we consider the two ‒pa morphemes to be two separate suffixes. Thus, only the ‒pa that derives agentive or abstract nouns from verbs or nouns coheres, as in (37a), but the other ones do not, as in the phonetic representations in (37b): (37) Cohering and noncohering structures for ‒pa a. (døpā)PWD (t ʰø̅ːbā)PWD b. (tʰøː)PWD -ba (mɛː)PWD -ba ʻwishʼ ʻtemple keeperʼ ʻhear-NOMʼ p. 180 ʻsaid-AOR.VOLʼ p. 209 In (38) we present examples illustrating that certain suffixes can end in the falling tone: (38) Suffixes ending in a contour tone: [tʰùː-sãː] ʻmet-AOR.Qʼ p. 83 40 [ɲɛ̅(d)-sãː] ʻfound-AOR.Qʼ p. 120 [ɲōp-t ɛː] ʻreach-NOMʼ p. 88 [kʰùː-t ɛː] ʻcan-NOMʼ p. 185 [sā- òː] ʻfirm-SUPERLʼ p. 98 [tsʰā-sèː] ʻwarm-ELʼ p. 98 [ja-løː] ʻgood-degreeʼ p. 99 [pri-dʑɛː] ʻwrite-FUT.VOLʼ p. 168 [oŋ-bàː] ʻcame-AOR.Qʼ p. 194 [lùː-bàː] ʻpour-CONSʼ p. 195 [ o-àː] ʻgo-CONSʼ p. 218 [t ùː-mɛː] ʻlet-AOR.NEGʼ p. 259 The examples in (38) require that these suffixed words have the double-pword representation in (1c). Thus, these words have the structure as in (39): (39) Prosodic structure of a representative example in (34): (tʰùː)ω (sàː)ω Recall from (28b) that the suffix -tʰøː was analyzed with a double-pword structure because it begins with an aspirated consonant. An additional reason for treating this suffix as a pword is that it ends in a falling tone, as the suffixes in (38). 41 A momentʼs reflection reveals that neither the Falling Tone Constraint nor Falling Tone Deletion can be reanalyzed in terms of morphological domains. One alternative along these lines is that both the constraint and the rule only holds at the right edge of a stem, thereby capturing the examples in (29) and (31). The reason this treatment will not work is that it incorrectly predicts that the falling tone at the right edge of the first member of a compound (see 33) should be retained. Equally problematic are the examples in (34), which show that the falling tone can occur at the right edge of certain suffixes, which by definition are not stems. 5.3 The distribution of onset clusters As noted in section 4.1, the KT syllable template in (5) can be augmented with a second onset consonant in word-initial position. Examples of two-member consonant clusters in word-initial position are presented below in (40). The five two-member clusters listed here (from (8)) are the only ones attested in KT. Clusters of three or more members are nonoccurring. (40) Complex word-initial onsets of KT: [brɛ̖ː ] ʻriceʼ Glo., p. 1 [pru̖ː ] ʻto shake out, shake somethingʼ Glo., p. 18 [pʰrā.mī] ʻnarrowʼ Glo., p. 19 [pjāŋ] ʻto suspendʼ Glo., p. 64 [pʰjàː] ʻto sweepʼ Glo., p. 88 42 Recall from (15) that two consonant clusters between vowels are heterosyllabified. In (41) we have repeated some of these examples for convenience. In (41a) we see that the VC.CV parsing holds if CC is a cluster that is not attested word-initially in KT. In (41b) we have provided an example of the VC.CV parse if CC is a permissible word-initial cluster: (41) Va is always short in ʻ...VaCCVb... ʼ a. b. c. [tsōk.pā] ʻdirty, badʼ Glo., p. 28 [rop.tsø̅] ʻroughlyʼ Glo., p 20 [sām.lō] ʻwith taŋ- in 'to thinkʼ Glo., p. 21 [pʰɛ̅j.t ēː] ʻprobablyʼ Glo., p. 19 [riŋ.mō] ʻlongʼ Glo., p. 20 [gab.lɛ̅ː ] ʻafterʼ Glo., p. 5 [t ōb.jɛː] (+[ o]) ʻto go to a girlʼ Huber, p.c. [t ab.rùː] ʻyoung birdʼ Huber, p.c. The generalization expressed with the data in (40) and (41) is captured in the following constraint: (42) Complex Onset Constraint: Two member consonant clusters only occur in pword-initial position 43 There is one prefix to our knowledge which occurs before consonant clusters like the ones in (40), namely negative mi-. The example in (43) is the only one we have found of this structure in Huber (2002). (Words with two consonants in the onset are not common in KT.) (43) mi- plus stem with a complex onset: [mi-pri-gẽ̅ː ] p. 230 NEG-write-NOM ʻnot writingʼ The example in (43) implies that the prefix mi- is non-cohering, as in (1c). In contrast to the constraints and rules posited in sections 5.1 and 5.2, the Complex Onset Constraint in (42) could in principle be reanalyzed in terms of a morphological constituent, i.e. ʻTwo member consonant clusters only occur at the left edge of a stemʼ. One could test whether or not the prosodic constraint in (42) or the alternative just suggested is correct by considering compound words. If KT allows for complex onsets to occur at the left edge of the second member of a compound then this would lend support to the stem-based alternative (since both parts of compounds are stems). If, however, KT does not allow complex onsets at the left edge of the second member of a compound then this would suggest that the pword domain in (42) is correct, since compounds in KT consist of a single pword. We leave this issue open for further research. 44 5.4 The distribution of coda consonants Recall from section 4.1 that the final C in the syllable template in (5) is restricted to /p m ŋ j r/. Examples of KT gwords ending in these five segments are presented in (44): (44) Examples of ..VC syllables in gword-final position: [tʰīp] ʻto lightʼ Glo., p. 26 [cʰīm] ʻhouseʼ Glo., p. 3 [nāŋ] ʻthe day after tomorrowʼ Glo., p. 14 [pʰȳj] ʻto give, to offer (honorific)ʼ Glo., p. 20 [tēr] ʻto giveʼ Glo., p. 24 The generalization to be captured is that if a gword is closed with a consonant then it can only end in one of the five consonants listed above and no other. A corollary of this generalization is that gword-internally a closed syllable can end in consonants other than /p m ŋ j r/, as illustrated in (45): (45) Examples of gword-internal syllables closed with a C other than /p m ŋ j r/. [tsōk.pā] ʻdirty, badʼ Glo., p. 28 [gab.lɛː] ʻafter, behindʼ Glo., p. 5 [t ab.rùː] ʻyoung birdʼ Huber, p.c. [t ōb.jɛː] (+[ o]) ʻto go to a girlʼ Huber, p.c. 45 [den. ɛː] ʻlike thatʼ Glo., p. 4 The constraint necessary to account for the restricted set of coda consonants in final position in (44) is stated in (46): (46) Closed Syllable Constraint: /p m ŋ j r/ are the only consonants allowed in pword-final position Huber (2002: 27) herself notes that /p m ŋ j r/ are restricted to ʻword-finalʼ position, although, as noted above, she does not draw a distinction between the gword and the pword.17 Note that the words in (45) attest to the fact that the domain of the Closed Syllable Constraint cannot be the syllable.18 6 The stem domain: Intrusive nasals In this section we demonstrate that the stem is the domain for intrusive nasals. Huber (p. 29) writes: “When syllables with initial voiced stops or affricates stand in the second position of a compound .a homorganic nasal appears before them”. She adds that this process also holds for reduplicated verbs. Examples illustrating what we refer to below as ʻintrusive nasalsʼ in compounds are presented in (47a-e) and reduplicated forms in (47f).19 (47) Examples Illustrating intrusive nasals in compounds and reduplicated forms: a. [ba] ʻmaskʼ [ba.ndām] ʻmask-danceʼ Glo., p. 1 /ba-dām/ Glo., p. 1 46 b. c. d. e. f. [sā] ʻground, floorʼ [sā.ndām] ʻborderʼ [me] ʻfireʼ [me.ndā] ʻgunʼ [ŋ] ʻpillowʼ [ŋ.mbø] ʻpillowʼ [pū] ʻupper valleyʼ [pū.ndo] ʻupper and lower valleyʼ [a] ʻto be similarʼ [a.nā] ʻsimilar, of same kindʼ Glo., p. 21 /sā-dām/ Glo., p. 21 Glo., p. 13 /me-dā/ Glo., p. 13 Glo., p. 16 /ŋ-bø/ Glo., p. 16 Glo., p. 20 /pū-do/ Glo., p. 20 Glo., p. 4 /a-ā/ Glo., p. 4 In the third column of (47) we have provided our interpretation of the respective underlying forms. The underlying form of the second stem in each two stem sequence does not have an initial nasal component because there is no contrast in this context between stems beginning with a voiced stop or affricate and ones beginning with a prenasal. According to Huber (2002: 18, 29) the process of nasal intrusion in examples like the ones in (47) also affects (word)-initial voiced stops and affricates. Huber (personal communication) informs us that all voiced stops in KT have a slightly prenasalized quality when they stand in gword-initial position. When pre-nasalization is realized gword-medially as in (47) then it is 47 realized phonetically like a true NC onset cluster and not like an nC pre-nasalized segment. By contrast, in g-word-initial position (underlying) voiced stops and affricates are realized as the corresponding prenasalized segment. For example, a narrow phonetic transcription of the word ʻarrowʼ [da] in isolation in KT would be [nda] and when it is part of a larger phonological domain, as in the second part of a compound, as in (47), that means ʻgunʼ, it is phonetically [nda]. This type of nasal intrusion also holds for the first member of the stems in (47); for example, the [b] in ʻmask-danceʼ in (47a) is [mb] in a narrow transcription in isolation. The examples in (47) contrast with the words in (48) which illustrate that a voiced stop beginning a suffix does not have an intrusive nasal. In (48a-c) we see examples of vowel-final stems plus suffixes beginning with /b d / respectively:20 (48) No intrusive nasal between stem plus suffix: a. b-initial suffix: [mɛː.ba.ma.no] ʻsaid-ASSERʼ /mɛː-bamano/ p. 87 [cèː.ba.ma.no] ʻbe.born-ASSERʼ /cèː-bamano/ p. 108 [søː.bi] ʻremember-NOM.GENʼ /søː-bi/ p. 187 [ ɛː.bo] ʻtell-AOR.SENSʼ / ɛː-bo/ p. 120 [t ʰø̅ː.bā] ʻtemple-keeperʼ /t ʰø̅ː-pā/ p. 29 48 b. d-initial suffix: [mɛː.de] ʻsaid-NFʼ /mɛː-de/ p. 70 [paʱk.pàː.de] ʻmotorbike-DEFʼ /paʱk.pàː-de/ p. 138 [kū.gẽ̅ː] ʻsteal-IPFV.GENERʼ /kū-gẽ̅ː/ p. 150 [tsàː.go.nuː] ʻfilter-IPFV.SENSʼ /tsàː-go.nuː/ p. 76 [tʰùː.gẽ̅ː] ʻmeet-NOMʼ /tʰùː-gẽ̅ː/ p. 79 [cʰàː.go.jøː] ʻfeel.cold-IPFV.EXPERʼ /cʰàː-go.jøː/ p. 153 c. -initial suffix: Thus, the intrusive nasal can only be observed before a voiced stop at the left edge of the second of two adjacent stems (in 47) but not at the left edge of a suffix (in 48). The examples in (49) illustrate that the intrusive nasal also occurs at the left edge of a stem if a prefix precedes (see Huber 2002: 29). The only prefixes to our knowledge which precede intrusive nasals are mi- and ma-. (49) Nasal Intrusion across prefix plus stem: a. b. [ o-] ʻto goʼ p. 29 [mi-n o] ʻI donʼt goʼ p. 29 [dʑi̖ː-] ʻto be afraidʼ p. 29 [ma-ndʑi̖ː] ʻdonʼt be afraidʼ p. 29 49 In contrast to the generalizations expressed in the constraints and rules posited in the previous sections, we argue that intrusive nasals surface in a morphological domain in the sense that they only surface in stem-initial position. This morphological environment is captured in the following rule of Nasal Intrusion: (50) Nasal Intrusion: Ø⇒ [+nasal, αplace] / STEM[ ̲̲ [+vc, -cont, -son, αplace] Nasal Intrusion says that a nasal is inserted at the left edge of a stem before a voiced stop and that the nasal inserted is homorganic with that stop. The phonological motivation behind the process of Nasal Intrusion in (50) warrants further discussion. Thus, why is a nasal and not some other segment inserted and why are voiced stops and affricates the trigger as opposed to other sets of sounds ? Although these questions are highly interesting, in a sense they detract from the main goal of this section, namely to account for the domain restrictions, i.e. why do we have intrusive nasals in (47) and (49) but not in (48). For this reason we leave open for further research the question of what phonological and/or phonetic factors might motivate the intrusion of nasals before voiced stops and affricates. This phenomenon is given a diachronic treatment in other studies of Standard and Central Tibetan (cf. Tournadre 2003; Denwood 1999: 75). 50 Nasal Intrusion is illustrated in the examples in (51), which are representative of the words discussed above. In the second column we can oberve the morphological structure, which is the input present in the underlying form and in the third column the phonetic form. (51) Sample morphological structures: Morphological structure phonetic form a. bare stem [da]STEM [nda] (= [nda]) b. stem + stem (=47) [sā] STEM [dzām]STEM [sāndzām] c. stem+suffix (=48) [søː]STEM [bi]SUFFIX [søːbi] d. prefix+stem (=49) [ma]PREFIX [dʑi̖ː]STEM [mandʑi̖ː] Nasal Intrusion applies to the voiced stops in the stems in (51b, d), while nothing happens in (51c) because the voiced stop is not at the left edge of a stem. As formulated in (51), Nasal Intrusion applies to bare stems as in (51a) as well and to the first member of a compound in (51b) if it begins with a voiced stop. As noted by Huber (see above) the intrusive nasal in absolute initial position is shorter durationally than the word-internal ones in (48b, d). The contrast between (51a) and (51b, d) is captured with a rule of phonetic implementation which we do not formalize here. At this point it is important to show why Nasal Intrusion does not have a prosodic domain. An obvious alternative is that the rule applies at the left edge of a pword. This analysis requires 51 that the compounds in (47) be analyzed as two separate pwords and that the suffixes in (48) and the prefixes in (49) be noncohering. There are two problems with this alternative. First, there are many suffixes of KT that have been argued to have the double-pword representation, e.g. in (38), and yet nasal intrusion never applies to voiced stops in these suffixes. Second, the prosodic analysis sketched above requires that each part of a compound be a separate pword, but we demonstrated in the preceding sections that KT compounds consist of a single pword. In this section we have argued that KT possesses a single rule (i.e. Nasal Intrusion) which refers to a morphological constituent. We assume that the constituent ʻstemʼ derives further evidence from the morphology of KT, although we do not provide that evidence here. In the literature on prosodic phonology one sometimes encounters proposals that there is a prosodic domain which corresponds to the morphological constituent ʻstemʼ. See, for example, Inkelas (1989), who propose the constituent ʻpstemʼ (in addition to pword) and Downing (1999), who shows that both constituents are active in Bantu languages. We leave open for further research whether or not it would be possible to extend this idea to KT. If this were the case then Nasal Intrusion would apply at the left edge of a pstem and there would be no evidence (to our knowledge) for phonological generalizations referring to constituents that are clearly morphological. 21 52 7 Typological relevance of KT In this section we summarize briefly the prosodic structures posited above for KT and then discuss the typological relevance these structures have for research in prosodic phonology. An important aspect of the analysis we presented above is that the various constraints and rules for KT require reference to pwords as well as to morphological structure. A simplified structure of stems, compounds and prefixed forms (in which no reference to the gword is made) is presented in (52). In the second column we present the constraint that requires the respective prosodic structure and the corresponding reference number. (52) Prosodic structure a. (stem)PWD Evidence Aspiration Constraint (23) Complex Onset Constraint (42) Falling Tone Constraint (30) b. (stem+stem) PWD Aspiration Constraint (23) Falling Tone Constraint (30) c. prefix + (stem) PWD Aspiration Constraint (23) Complex Onset Constraint (42) 53 Suffixed forms in KT can have one of the three representations in (53). In the second column we provide the constraint that requires the respective structure and in the third column the reference to the examples of the respective suffixes. (53) Prosodic structure Evidence Suffixes a. (stem) PWD + suffix Falling Tone Constraint see (31) b. (stem) PWD + (suffix) PWD Aspiration Constraint see (27 & 28) Falling Tone Constraint see (38) Falling Tone Constraint see (37a) c. (stem + suffix) PWD In (54) we provide a list of the suffixes of KT discussed in the examples above which require the three prosodic structures in (53): (54) a. Noncohering suffixes (=53a): -de -si -onuː -nu -la -bo -ɟa -de -o -ojøː -bamano -bajimbɛː -so -pa 54 -jøː -bi -bɛː -nā -ē b. Double-pword suffixes (= 53b): -tsʰøːnu -sèː -tʰòː -løː -sàː -dʑƐː -t ɛː -bàː (and also -pà: AOR.Q, which is diff. from CONS) -øː -àː -mɛː c. Cohering suffix (=53c): -ba Two points can be made with respect to the list in (54). First, there appears to be a single suffix of KT with the cohering representation, i.e. the one in (54c). Second, none of the remaining suffixes cohere with the pword of the stem to which they attach. Thus, the suffixes in (54a) have the noncohering representation in (53a) and the ones in (54b) have the double-pword representation in (53b). Recall that we adduced positive evidence that the latter suffixes are pwords (i.e. they have contour tones), but for the suffixes in (54a) there exists no evidence for 55 or against treating them as pwords. What this means is that one could in principle analyze the suffixes in (54a) as double-pword-suffixes (since there is no evidence to the contrary). This is a question we leave open for further study. The paucity of cohering structures like the one in (54c) is surprising because this is usually the ʻdefaultʼ structure assumed in other languages (e.g. Nespor & Vogel 1986, Hall & Kleinhenz 1999, Dixon & Aikhenvald 2002). That KT possesses suffixes of the (53b-c) type is not unusual cross-linguistically but what is odd is that it treats virtually all of its suffixes this way. A similar point worthy of note is the representation of compound words in (52b). The cross- linguistic studies on pwords cited above provide evidence from a wide variety of languages that each part of a compound must be analyzed as its own pword. By contrast, languages in which compounds are single pwords are rare. What we conclude is that the relationship between pwords and morphological structure in KT is unusual in the worldʼs languages. The reasons for why KT treats compounds and suffixed forms in this marked fashion is a question worth considering in the future. 8 Conclusion In the preceding sections we have highlighted both the syllable and the phonological word as important prosodic constituents in KT. We have shown that a number of phonological generalizations require a prosodic domain of pword, and that suffixes, far from behaving as a 56 uniform prosodic group, require three distinct prosodic representations: cohering, noncohering and double-pword. Prefixed forms were argued to require the mirror image of the noncohering representation, while compound words were showed to consist of a single pword. Finally, the domain of one phonological process (i.e. Nasal Intrusion) requires reference to the ʻstemʼ domain. The prosodic structures posited above for suffixed forms and for compounds are important because they illustrate the unusual position of KT with respect to typological research. In particular, KT is unusual in treating almost all of its suffixed forms as either noncohering or double-pword and its compounds as single pwords. In the future one might want to investigate whether or not KTʼs aversion to cohering representation for suffixed words might be derivable from some other aspect of its grammar. References Booij, G. (1996). Cliticization as prosodic integration: the case of Dutch. The Linguistic Review 13: 219-242. Breen, Gavan and Pensalfini, R. (1999). Arrernte: A language with no syllable onsets. Linguistic Inquiry 30: 1-25. Clements, G. N. (2001). Representational economy in constraint-based phonology. In: T. A. Hall (ed.) Distinctive Feature Theory. 71-146. Berlin: Mouton. 57 DeLancey, Scott (2003). Lhasa Tibetan. The Sino-Tibetan Languages, ed. by G. Thurgood & R.J. LaPolla, 270-286. London: Routledge. Denwood, Philip (1999). Tibetan. Amsterdam: Benjamins. Dixon, R. M. W. and Aikhenvald, Alexandra Y. (2002). Word. A Cross-Linguistic Typology. Cambridge: Cambridge University Press. Downing, Laura. (1999). Prosodic Stem ≠ Prosodic Word in Bantu. Studies on the Phonological Word, ed. by T. A. Hall & U. Kleinhenz, 73-98. Amsterdam: John Benjamins. Hall, T.A. and Kleinhenz, Ursula, eds. (1999). Studies on the Phonological Word. Amsterdam: John Benjamins. [Current Issues in Linguistic Theory 174]. Haller, Felix. (1999). A brief comparison of register tone in Central Tibetan and Kham Tibetan. Linguistics of the Tibeto-Burman Area 22.2: 77-97. Hari, Maria. (1979). An investigation of the tones of Lhasa Tibetan. Language data, Asian- Pacific series 13. Huntington Beach: Summer Institute of Linguistics. Häsler, Katrin (1999). A Grammar of the Tibetan Dege (Sde dge) Dialect. Bern: Bern University Dissertation. Huber, Brigitte (2002). The Tibetan Dialect of Lende (Kyirong): A grammatical description with historical annotations. Bern: Bern University Dissertation. (To appear in 2005 in Beiträge zur tibetischen Erzählforschung, vol. 15, ed. by D. Schuh). 58 Inkelas, S. (1989). Prosodic Constituency in the Lexicon. Ph.D. Dissertation: Stanford. Matisoff, James A. (2003). The handbook of Proto Tibeto-Burman. Berkeley: University of California Press. Matisoff, James A. (1991). Sino-Tibetan linguistics: Present state and future prospects. Annual Review of Anthropology 20: 469-504. Mazaudon, Martine. (1977). Tibeto-Burman tonogenetics. Linguistics of the Tibeto-Burman area 3:2.1-123. Nespor, Marina and Vogel, Irene (1986). Prosodic Phonology. Dordrecht: Foris. Peperkamp, Sharon (1997). Prosodic Words. The Hague: Holland Academic Press. Selkirk, E. (1995). The prosodic structure of function words. In: J. Bekman, L. Walsh Dickey & S. Urbanczyk (eds.) Papers in Optimality Theory. University of Massachusetts Occasional Papers in Linguistics 18. Amherst, Mass: GLSA. 439-469. Sprigg, R.K. (1966). Lepcha and Balti Tibetan: tonal or non-tonal languages? AM, New Series 12.2: 185-201. Sprigg, R.K. (1990). Tone in Tamang and Tibetan, and the advantages of keeping register-based tone systems separate from contour-based tone systems. Linguistics of the Tibeto-Burman area 13.1: 33-56. 59 Sun, J. T.-S. (1986). Aspects of the phonology of Amdo Tibetan: Ndzorge Shæme χəra Dialect. (Monumenta Serindica No. 16). Tokyo: Institute for the study of languages and cultures of Asia and Africa. Sun, J. T.-S. (2003). Phonological profile of Zhongu: A new Tibetan dialect of Northern Sichuan. Language and linguistics 4.4: 769-836. Sun, M. T.-S. (2006). Special linguistic features of gSerpa Tibetan. Linguistics of the Tibeto- Burman area 29.1: 107-125. Tournadre, Nicolas and Dorje, S. (2003). Manuel de Tibétain Standard. Paris: Langues & Mondes/LʼAsiathèque. van Driem, George. (1992). The Grammar of Dzongkha. Thimphu, Bhutan: Dzongkha Development Commission, Royal Government of Bhutan. Vollmann, Ralf. (1999). Zusammenhänge in der tibetischen Phonologie, Tonologie, Morphonologie und Orthographie. Grazer Linguistiche Studien 52: 121-139. Watters, Stephen A. (2002). The sounds and tones of five Tibetan languages of the Himalayan region. Linguistics of the Tibeto-Burman area 25.1: 1-66. * The present study was supported in part by funds from the German DFG Grant BI 799/2. We thank Brigitte Huber for access to her detailed grammatical description of Kyirong Tibetan, without which 60 such an analysis would not be possible. We thank Brigitte Huber, Balthasar Bickel and two anonymous referees for their input and feedback regarding this manuscript and the data contained within. Any errors are the responsibility of the authors of this manuscript. 1 Some general references on pwords include Nespor & Vogel (1986), Peperkamp (1997), Hall & Kleinhenz (1999) and Dixon & Aikhenvald (2001). In all of these studies there is agreement that the gword is not always coterminous with the pword. 2 Another conceivable option for (1c) is that the syllable of the prefix is linked not to the pphrase but instead to the Clitic Group (see Nespor & Vogel 1986, who motivate that constituent). For arguments against the Clitic Group see Booij (1996) and Peperkamp (1997). Yet another option for (1c) is that the syllable of the prefix is linked not to the pphrase (or to the Clitic Group) but instead to a recursive pword. See, for example, Selkirk (1995), who motivates recursive pwords on the basis of data from English. Our analysis does not crucially depend on any of these options; hence, this is a question we leave open for further research. Noncohering representations like the one in (1c) are well attested in languages outside of Sino- Tibetan. See, for example, Peperkamp (1997) for Italian and the studies in Hall & Kleinhenz (1999). 3 In her discussion of the KT vowels (p. 19) Huber considers only the vowels in the FRONT I column to be primary but later on (e.g. p. 21) she assumes that central and back vowels belong to this category as well. It is the latter interpretation that we adopt in the present study. 61 4 On occasion certain languages are cited as exceptions to this generalization, e.g. the Australian language Arrernte (Breen & Pensalfini 1999). Also, In (6) and in other examples we list in the final column the page number for the corresponding word in Huber (2002). ʻGlo.ʼ stands for ʻglossaryʼ. 5 Note that the examples in (11b) involving [e e] do not illustrate a true contrast because the tones differ. We consider the lack of contrasting words between [e e] in which the tones are identical to be purely accidental. 6 One exception to this generalization is the word [pɛj.cɛ] ʻwool combʼ, (Huber 2002: 25). 7 One complication is that long vowels seem to be nonoccurring before a (tautomorphemic) vowel. Examples of hiatus sequences in which the first vowel is short are in (i): (i) Short vowel plus tautomorphemic vowel sequences: [pi.ā] ʻmouseʼ p. 27 [lī.u] ʻblanketʼ Glo., p. 12 [pī.ō] ʻafternoon, eveningʼ Glo., p. 19 [ju.ā] ʻhandleʼ Glo., p. 7 [a.u] ʻpaternal uncleʼ Glo., p. 1 Our assumption is that the lack of VV is due to an independent constraint which we do not discuss here. 8 The only exception to this generalization we have found is [o:rt e] ʻthank youʼ (Huber 2002: Glo., p. 17). 62 9 At present we have no explanation for the tonal alternations involving the stems in (18). One example in Huber (2002) appears to display closed syllable shortening due to the addition of an optional syllable-final [j], i.e. [pʰȳ: ˜ pʰȳj] ʻto give, offer (honorific) Glo, p. 20. We omit this word from the data in (18) because it seems to be a unique example of this type rather than representing a regular pattern. 10 The Aspiration Constraint also accounts for the distribution of the voiceless glottal fricative [h] (Huber 2002: 27), which is also an aspirated segment. Huber also notes that the lateral fricative [ɬ] only occurs in ʻwordʼ initial position (p. 27). In their description of Standard Tibetan, Tournadre & Dorje (2003: 25, 52) note that medial syllables do not carry aspiration as a contrastive feature, although there is a plain vs. aspirated contrast in word-initial position. 11 The processes of Deaspiration in examples like the ones in (24) are attested as historical processes in other Tibetan dialects. For example, Häsler (1999: 90) gives the example /mʰēː/ ʻmedicineʼ + /kʰõ̅ː/ ʻhouseʼ > [mʰēːkʰõ̅ː] ʻhospitalʼ (the diacritics for tone are ignored here). 12 There seem to be few examples like the ones in (24). This may simply be an accidental fact of KT, or alternatively, these could be vestiges of a once productive historical process which is no longer productive synchronically. We leave these questions open for further research. 13 In section 5.2 we show that the suffix -so ʻAOR.SENSʼ must be noncohering. 63 A referee suggests that ma- in (25) is really a proclitic. We consider ma- to be a prefix because of its grammatical sensitivity (it only occurs immediately before verbs and copulas). It is also worth noting that negatives are clearly prefixes in other Tibeto-Burman languages. 14 A reviewer suggests that -tsʰø:nu: could be an auxiliary verb and -tʰø: an inflected conjunction, in which case they would not provide evidence for the double-pword representation we are endorsing. Interestingly, Huber (2002: 191) herself refers to these morphemes as ʻauxiliariesʼ, but Huber uses the word ʻauxiliaryʼin her grammar in a rather liberal fashion. More to the point, -tsʰø:nu: and -tʰø: are morphemes expressing tense/aspect/modality and evidentiality and are grammatical affixes. 15 A reviewer suggests that the Falling Tone Constraint is really conditioned by the fact that shortened vowels are not long enough to bear (perceptible) falling tones. This alternate explanation for the distribution of falling tones cannot be correct because there are pword-internal long vowels which cannot bear falling tones (e.g. the odd numbered examples in (13) above). 16 Recall from the discussion after (4) that voiceless consonants like [p] only occur after short secondary vowels, whereas voiced consonants like [b] occur after the long secondary vowels. This generalization accounts for the alternation between [p] and [b] in (36). Since [p] and [b] in examples like these are predictable we posit underlying forms with /p/. 17 According to Delancey (2003: 272) the only permissible coda consonants in Lhasa Tibetan are /p m k ŋ/. Denwood (1999: 71) apparently assumes that a similar phonotactic generalization in the variety of Tibetan he discusses is the syllable and not the pword. 64 18 An alternative to the pword domain for the Closed Syllable Constraint is either the gword or the stem. An argument against the former would include suffixes that end in consonants other than /p m ŋ j r/, but suffixes with closed syllables are absent in KT. We hold that the burden of proof for the gword domain (or the stem domain) now lies on the shoulders of those who do not believe the pword domain to be correct. 19 At present we have no explanation for the tone alternation on the stem (47d). We have found two isolated examples of noncompounds in which an intrusive nasal surfaces: a. [ka.nā] [ka.n ùː] b. [ta.lō] [ta.ndā] 20 ʻwhereʼ /ka-nā/ Glo., p. 8 ʻhowʼ /ka- ùː/ Glo., p. 8 ʻthis yearʼ /ta-lō/ Glo., p. 24 ʻnowʼ /ta-dā/ Glo., p. 24 One exception to this generalization is the derivational suffix -de, which optionally surfaces as -nde in [tsʰāndē] ʻhotʼ /tsʰā-dē/ (Huber 2002: 93). 21 A reviewer suggests an alternative analysis for Nasal Intrusion. According to their suggestion the nasal intrusion facts require a purely morphological solution: A nasal element is analyzed as a ʻcompounding affixʼ consisting of a placeless nasal consonant or glide (i.e. N). This element is licensed by a following voiced stop or affricate (perhaps after place assimilation) and is otherwise deleted. The deletion rule is very natural as ND (D = voiced stop) is the favored type of NC cluster crosslinguistically, and many languages have rules deleting nasals in other types of NC clusters. 65 The analysis is also morphologically plausible, as many languages have compounding affixes located between the two members of the compound (e.g. Japanese; see Clements 2001: 129). (On this analysis the retention of N after the prefixes mi- and ma- in (45) requires that N be analyzes as a part of the underlying representation of the prefixes, i.e. /miN/, /maN/.) The problem with this alternative is that it cannot explain why N is always present in gword-initial position, e.g. in (47a). Data like these tell us that N is not a morphological affix because it is also epenthesized in free morphemes.