A Parameter of Syllabification

PAPERS AND STUDIES IN CONTRASTIVE LINGUISTICS, Vol. XXIV P L ADAM MICKIEWICZ UNIVERSITY PRESS • POZNAN ISSN 0137­2459 A PARAMETER OF SYLLABIFICATION* ROLAND NOSKE Univertity of Amsterdam 0. Introduction In the litcrature concerning the syJIable (e.g., Kalm 1976), essentially one principle of assignment of syllable structure has been advanced: an algorithm which links elements to syllabic nodcs in the following order: a. every vowel; b. a maximal number of consonants preceding every vowel; these eonso­ nants must form a permissible word­initial cluster; c. a maximal number of consonants following the vowel; these consonants must form a permissible word­final cluster. Although there exists, according to this principle, a well­deJined order of appli­ oation within the syllable, nothing is said about the way the algorithm applies to a larger order of segments (e.g. a prosodie unit). Apparently, all the vowels are linked first, then the consonants which are going to form the onsets, and finally the codas. Thus, within a prosodie group which constitutes a domain for syllabification, all syllables are assigned simultaneously. This is why we wish to call this type of syllabification non-directional. However, Kaye and Lowenstamm (1981:306­11) propose another principle: according to them, syllabification is directional and can take place from lelt to right in some languages, and from right to left in others. The principle of * We wish to thank Norval Smith for much valuable discussion, as well as for a num­ ber of corrections to our English. This work was supported by tho Stichting Taalwetenschap, which is funded by the Netherlands Organisation for the Advancement of Pure­ Soientific Research (ZWO). 48 R. Noske ayllabifieation which they advance is given (1) and (2) (1981:307­9): (1) The Rightward Strategy Scanning a word frora left to right, make t h e first syllable as unmarked as possible. If the resulting syllable on the right conforms t o the formal and substantive constraints of the Ianguage, then there is a syllable boundary a t t h a t point. If the resulting right syllable violates a eonstraint, move the syllable boundary over one segment to the right and try again. Repeat until the resulting right syllable is Heit. Alter the first syllable boundary has been found, repeat t h e process for each successive syllable until the end of the string is reached. (2) The Leftward Strategy Scanning a word from right to left, make t h e last syllable as unmarked as possible. If the resulting syllable on the left conforms to the formal and substantive constraints of the Ianguage, then there is a syllable boundary a t theflpoint. If the resulting left syllable violates a eonstraint, move the syllable boundary over one segment to t h e left and try again. Repeat until t h e resulting left syllable is licit. After the first syllable boundary has been found, repeat t h e process for each successive syllable until the beginning of the string is reached. We refer the reader to K a y e and Lowenstamm (1981) for an explanation of their concept of markedness. I t will suffice here to mention the concrete exam­ ples mentioned by the authors as motivation for their theory. As a Ianguage using the Rightward Strategy they mention English of which they give the following examples: (3) a. command b . canteen c. astute ka­maend ksen­tiyn a­stuwt Although in English the segmeuts or clusters of segments m, nt, st are per­ missible word­finally, the syllable boundaries are not located after these seg­ ments or clusters of segments b u t at the first possible place (going from left t o right) (provided it is possible to syllabify the whole word, which is the reason t h a t in (3b), the syllable boundary is placed after the n). I n (3b, j ) the syllable­ initial position of the dental piosives can be proved by the fact t h a t they are aspirated. Aa a Ianguage which uses the Leftward Strategy Kay and Lowenstamm men­ tion Polish. Speakers of Polish generally agree t h a t the syllabification of words Jike wyspa 'island', ospa 'smallpox', tykwa 'pumpkin', stacja 'station' is as in (4): (4) wys­pa os­pa tyk­wa stac­ja A parameier of eyUabifloation 49 This syllabification can be explained if the Leftward Strategy (2) is adopted. The Rightward Strategy (1) would give rise to syllabifications like *wy-spa, *ty-kwa, etc. Independently of the proposals b y K a y e a n d L o w e n s t a m m , T e r Mors (1982) proposes a directional syllabification for Klamath, a Penutian language of Oregon. W h a t is interesting liere is t h a t the direction of syllabification can be demonstrated in terms of epenthesis site. I n Klamath, an epenthesis process applies in order to resolve unsyllabifiable consonant clusters. Consider the underlying form in (5): (5) /snogwk/ In this form, the sequence gwk is not a permissible word­final cluster. The syllable template of K l a m a t h is given in (6): (6) syllable template for Klamath (O c v (C(O) [—son] [—sou] I n order t o make the sequence syllabifiable ,there are two conceivable epenthe­ sis sites: between g and w, and between w et k. I n fact the process of epenthesis insert the schwa between g and w, creating the form snogdivk. After having undergone an independently motivated vocalisation process, which changes dw t o o: the form surfaces as (7): (7) [snogo:k] The form in (8), which would result if epenthesis were to take place between k, is ungrammatical. (8) *[snogwak] As in m a n y analyses concerning epenthesis, Ter Mors conceives of this process as the creation of a V on the level of the skeleton. This empty V is then filled u p with t h e neutral vowel by a rule (or. according to the proposals of Archan­ geli (1984), by a complement rule): (9) V The V on the skeletal level is assumed to have been inserted by the process of syüabification proper. I t is thus t h a t the reiationship between epenthesis 50 R. Noske a n d syllabic structure is estabUshed (recall t h a t epenthesis functions in order to resolve otherwise unsyllabifiable sequences). The following syllabification mechanism is proposed b y Ter Mors (1985:316): (10) Ruk of ASSOCIATION: Map in a R h [ = r i g h t ­ t o ­ l e f t , R N fashin t h e syllable t e m p l a t e onto t h e segmental trier; maximize clusters. Because of the directionality, the following structure is created for /snogwk/; (11) a a C C V C V C C s n o g w k C C First, | and | are syllabified. Then the syllabification mechanism expects k w a V. Because of the fact t h a t the element which it then encounters is not a V, a V is inserted: (12) o A\ V C C w k If one were to apply the well­known Maximal Onset Principle (incorporated in the proposal by K a h n ) , one would derive erroneously: 1 (13) a a C C V C C V c s n o g w k which, after application of (9) would produce the ungrammatical form in (8). We would get the same result if we were to adopt a left­to­right syllabification. H a ving outlined the essence of the proposal by Ter Mors, we will now show t h a t the principle of directional syllabification accounts for certain processes in Yawelmani. We will see t h a t the principle of directional syllabification re­ 1 I n fact, this is not completely true. Strictly speaking, the maximal onset principle would fail to syllabify t h e second syllable altogether: in order to make an onset maximal it should be able to refer to the nucleus. Since there is no V to perform this function, the proeesa of syllabification would be blocked. A pa?'ameter of syllabification 51 ceives additional motivation because it explains certain syllabic processes other thaii epenthesis. Then, applying the principle to a third language, Tigrinya,. we will see t h a t the direction of syllabification is parametrised. 1. The syllabic processes in Yawelmani Yawelmani, also a Penutian language (but not particularly closely related to Klamath), spoken in South Central California, has attracted a lot of atten­ tion from linguists. We mention here only the theses of Kuroda (1967), Kisse­ berth (1969) and Archangeli (1984). The main source is Newman (1944). Se­ veral processes affecting syllable structure take place in this language. The ana­ lysis we present here can be found in a more extended form in ÏSToske (1985). Yawelmani has three types of syllable: CV, CVC and CV: (0V| V t ). Like Kla­ math, Yawelmani has an epenthesis process which serves to "repair" the syl­ labic structure. This is why we propose t h a t in a certain phase of the deriva­ tion the nucleus 2 can be empty, into which vowels will be epenthesised at a later stage. These are the well­formedness conditions for the onset and t h e rhyme in Yawelmani: (14) possible Yawelmani onset / O syllabic structure C skeleton c consonantal segmental tier (15) possible Yawelmani rhymes b. R /N^ N V V v 2 . > syllabic structure Cd J skeleton vocalic segmental tier consonantal segmental tier We assume here the well­known onset-rhyme bipartition, with the rhyme being subdivided into nucleus-coda, for reasons that are outside the scope of this article, but which are given in Noske (1985). 52 R. Noske I n (15b) the coda is present, b u t obligatorily empty. The reasons for this do not concern us here, b u t are given in Noske (1985). We purpose the following syllabification principles for Yawelmani (cf. Noske 1985:347): 16) Syllabification principles for Yawelmani syllable structures are mapped onto the skeleton a. from right to left b . in such a way t h a t the number of empty syllabic nodes is mini­ mised. There are three reasons for the directionality. The first two concern the rules of Shortening and Elision postulated for Yawelmani b y Kuroda (1967) a n d Kisserbeth (1969): (17) Shortening: V ­>■ [—long] / (18) Elision: V ­ 0 / V i # The functioning of Shortening can be seen in (19): (19) / t a x a a + t / [taxat] taxaa-, 'bring'; -t, passive aorist The functioning of Elision can be seen in (20): (20) / l a g a a + i n + h i n / [laginhin] lagaa- 'spend the night'; -in-, mediopassive, -hn9 aorist Applying the syllabic well­formedness conditions given in (10, 11) as well as the syllabification principles in (16), one obtains the syllabic structure in (21): (21) 0 / O \ R / \ N Cd C V V c \/ I v c I c We now see t h a t the process of shortening follows auxomatically from the re­ trograde character of syllabification: the mechanism will try to syllabify ac­ cording to the templates in (15a, b). Because of the fact t h a t it encounters a C, it will select (15a). Then, (continuing to go from right to left) it encounters the rightmost V, and links this to the nucleus node. Now a second V is en­ A parameter of syllabiflcation 53 eountered. Because of the faet t h a t (15a) allows only one V t o be linked to the syllabic structure, this V is then skipped and is therefore not realised, since we assume t h a t in order to be pronounced an element on the CV­tier has to be linked to the syllabic structure. We obtain a similar result to t h a t in (21) with vowel elision in the case of a closed syllable. Cf. (22): (22) a f ! /\ N Cd 1! G V V C j v v i I n (22), the retrograde syllabiflcation links only the rightmost V to the sylla­ bic structure, thus the V to its left is not realised, which is the correct result. For an open syllable, it is the requirement t h a t the two contiguous V s be lin­ ked to a single element on the vocalic segmental tier t h a t rules out the form in (23): (23) R O /\ N Cd / \ c V V We thus see t h a t two processes in Yawelmani which up till now have had to be explictly stated in terms of rules, follow from the assumption of a directio­ nal syllabification mechanism. We now come to a third reason for the di­ rectionality. Consider the foUowing two rules, proposed by Kuroda (1967) and Kisseberth (1969): (24) Epenthesis: 0 ­» i / o{*o B . Koske 54 (25) Two Sided Open Syllable Deletion (Kuroda 1967:32): V [­long] 0/VC CV The derivations in (26) ülustrate t h e functioninf og these two rules: (26) paH-, 'fight'; ­Tm, aorist; -t, passive aorist a. p a ? t + h n pa2iti+hin pa?it+hin pa^ithin b. p a ^ t + t pa?it+it pa^it+t pa?tit Underlying Representation Epentiesis (24) Two Sided Open Syllable Deletion (25) Surface Representation I n s t e a d of Epenthesis (24) we propose the following rule, which fills an e m p t y node w i t h t h e neutral vowel (which is i in Yawelmani): (27) Epenthesis: N V 0 ­+|/_ This rule differs from rule (9) only in t h a t i t refers to a n e m p t y syllabic node (nucleus), instead of a n e m p t y V. L e t us now look a t t h e structures t h a t will be generated b y onr syllahification principles in (16) for t h e structure in (26): (28) (29) O O R R / \ N V a P G O N C O R / \ / \ Cd I ! C CT N Cd I C C V O R R / \ / \ Cd N c N Cd C Cd I c C n n a I We see t h a t the e m p t y nodes are precisely in those places where we find an epenthetic vowel on t h e surface. We now see t h a t we are able to posit a more general rule for t h e epenthesis process t h a n in (24). Because our rule refers to syllable structure, it is endowed with explanatory power: it explains t h e fact t h a t epenthesis takes place only in those cases where it is necessary to "repair" the syllabic structure. N o t e also t h a t we do n o t need the Two Sided Open Sylla­ A parameter of syllabification 55 ble Deletion (25) here. I t would indeed be strange t h a t a vowel should be in­ serted first, only to be deleted later, as in the derivations in (26).3 If one assumes a syllabification from left to right, one would derive the structure in (30), which would give rise to the ungrammatical phonetic form. in (31): (30) G O R O / \ N Cd C V c c R O N R / \ / \ N Cd Cd (31) *[pa?tihni] V c a n We thus see t h a t the right­to­left directionality can account for three processes in Yawelmani. I t is useful to mention a fourth reason here, of a more theoreti­ cal nature, which argues in favour of a retrograde syllabification. Consider what would happen if syllabification were to take place from left to right. If C the mechanism encountered a postvocalic |, it would not be able to decide to c which syllabic node this element were to be linked. For this, it would have to C V C know whether the following element were a |, or a |. I n the same case, the I c v c would have to be assigned to the coda (of the former syllable), in the latter case to the onset (of the next syllable). This amounts to saying t h a t we would have to be able to look ahead, which would entail a partial bidirectionality. This would make the mechanism unnecessarily powerfull: a right­to­left appli­ cation does not produce this kind of problem, at least not for CVC languages in which the onset is obligatorily filled. I n order to see this consider what would happen if (going from right to left) the mechanism were to encounter a C after having linked the node of nucleus to a V. Because of the fact t h a t the onset is obligatorily filled, it would always assign onset status to this C. There is thus 3 A p a r t from cases Itke (26), t h e only motivation adduced for t h e rule of " T w o Sided Open Syllable Deletion" (27) is constituted b y t h e baheviour of t h e mediopassive morphe­ m e — i n — , However, as pointed o u t in Noske (1985 : 359), t h e deletion of i which can occur in this morphene is t h e result of t h e analogical reanalysis of t h e verb stem foliowed b y ~~in— as a monomorphemic verb stem (the so­called "fake b a s e " (Newman 1944:75)). T h e i t h e n alternates w i t h zero, because it is reanalysed as a n epenthetic vowel. E.. Noske 56 no need to "know" the nature of the element on the CV­tier (i.e., a C or a V) that is to the left of the C in question. By assuming right­to­leftdirectionality, the mechanism can be kept as simple as possible.4 We can now raise the question whether right­to­lefb directionality is universal, or, to put the same question less strongly, whether this directionahty is uni­ versal for CVC languages (i.e., a language in which the maximal syllable is CVC). This could be tested by investigating which stie is is selected for the in­ sertion of an epenthetic vowel in order to "repair" syllabic structure, like the one in (24). This boils down to asking the question in (32): f# (32) Is the environment G<n universal in syllable repair in CVC lang­ uages? We will see in section 2 that the answer to this question is negative. To see this, we will examine the process of Epenthesis in Tigrinya. 2. The syllabic processes in Tigrinya Tigrinya, a South Semitic language of northern Ethiopia, possesses an epen­ thesis process which can operate word­internally as well as word­finally. Pam (1973) mentions two different rules, which he later combines. The first rule is givenin (33) (1973 : 116): (33) 0 - > i / C C _ ( C ) # The functioning of this rule can be seen in (34) (1973:114): (34) a. /kalb+n/ [kA.bin] "dog"+sunixed conjunction b. /kalb/ [kAlbi:] "dog" In (34b) the i has been lengthened by a lengthening rule given in (35) (1973 : 115): (35) i - > i : / _ # * We are indebted to Professor Werner Winter for indicating to us that an earlier version of our text was unelear on this point. If a CVC language allows empty onsets, but does not allow syllabifications of the type CVC$VC (where $ = syllable boundary), a syllabification excluded by its high degree of markedness in terms of Kaye & Lowen­ starnm's proposals (whatever the direction of syllabification), we can make our prediction more general and posit that left­to­right syllabification is the marked setting for CVC languages in general. As the reader will have noted, the suggestions made by Kaye and Lowenstamm also imply a partial bidirectionality. However, we think t h a t a languago tends to choose the direction of syllabification which implies the fewest complications, other things being equal, and that therefore a CVC language would tend to select the right­to­left syllabification. A parameter of syllabification 57" The second epenthesis rule is given in (36) (1973:111): (36) 0 ­ > i / # C C An example of the functioning of this rule can be seen in (37): (37) /sbar/ [sibAx] 'break' P a m combines these two rules as in (38) (1973:117): (38) 0 ­> i / [ ­ s y l l ] [ — s y l l ] _ [ ­ s y l l ] P a m thus has to resort to the expression [syll], i.e. he has to treat word b o u n d a r i e s o n a p a r w i t h c o n s o n a n t s . H e t h u s h a s to specify these elements ne­ gatively with regard to their syllabicity. This is not very satisfying for iri faet word botmdaries and consonants have very little in common. This is the reason why phonologists have stopped referring to word boundaries as [syll], One can now ask the question whether in Tigrinya, as in Yawelmani,, the epen­ thesis process can be analysed as the result of the process of syllabification. Por this, it is necessary to know what the maximal syllable in Tigrinya is, and whether epenethesis takes place only if the process of syllabification is confron­ ted with an otherwise unsyllablfiable sequence. The answers to both questions are straightforward: the syllable structure of Tigrinya is CV (V) (C), the maximal syllable being thus CVVC, and Epenthesis operates only in those structures where otherwise a more complex consonantal syllabic structure t h a n O W C would ensue. We can now see t h a t the rules in (33) and (36) can bedispensed with if we as­ sume a left-to-right syllabification, in the same way as the right­to­left syllabi­ fication in Yawelmani we proposed above: if a C is encountered by the sylla­ bification mechanism a t a place where it only expects a V, a V is projected, a n d later filled b y the neutral vowel (in the case of Tigrinya an iï).ïï we adopt such an analysis, we can express the process in a unitary way, without having to resort to references to [syll]. The only difference from the Yawelmani case is t h a t the directionality of syllabification is revcrscd, and t h a t the quality of the epenthetic vowel is süghtly different (/ instead of i). I t can thus be stated t h a t the parameter of the directionality of syllabification is set differently for Tigrinya and Yawelmani. Not only does the epenthesis process provide motivation for the directionality of the syllabification process, but this is also the case with a process of vowel deletion, operating in an opposite fashion to Yawelmani. Thus is formulated by P a m (1973:76) as in (39). (39) Vowel Elision (Pam 1973:76): V V ­> 0 [+long] 58 K. Noske The stractural description of this rule is the mirror image of t h a t of rule (18), formulated by Kisseberth for vowel elision in Yawelmani, except in its requi­ rement t h a t the vowel be long, which follows from the fact t h a t Tigrinya al~ lows two V s in a elosed syllable (the two V s represent a long vowel or a diph­ thong). An example of the application of rule (39) is given in (40) (Pam 1973:77). (40) Base a­|—prefixation Infixation Vowel Elision Output r i*s a+ a: 0 a+ra:?s 'heads' "This form surfaces as [ara : Hs], (The place where the epenthesis takes place seems t o contradict rule (33), as well as our reanalysis of it. However, P a m (1973:117­8) points out t h a t the epentheiss site is exceptional and is re­ stricted t o the class of words to which the form in (50) belongs, thus there seems to be a morphological conditioning here). We will now make a final point. As was mentioned in section 1, given a di­ rectionality in the syllabification mechanism, one would expect it to be right­to­ left rather than left­to­right because of the fact t h a t in the latter case the mechanism would have to look ahead. The right­to­leffc directionality would thus constitute the unmarked case. This seems indeed to be confirmed b y the facts. There are many CVC languages where G<^ is the environment for syllable repair. A situation such as the one in Tigrinya, however, where the environment for syllable repair could be formulated as ^yC a seems to be very rare, and must be assumed to be the marked case. This conjecture is confirmed if we look at the situation in languages closely related to Tigrinya, e,g., Tigre. I n Tigre, the equivalent of (34b) is as in (4): (41) [k.Alib] This form is also found in other related languages. Hence it must be conclud­ ed t h a t in Tigre, syllabification takes place from right to left, the unmarked direction, and t h a t Tigrinya is exceptional in its left­to­right syllabification. «3. Gonclusion We have shown t h a t the principle of directional syllabification proposed by Ter Mors (1982) in order to explain the process of epenthesis in Klamath, as well as b y K a y and Lowenstamm (1982) in order to account for the dif­ ference in the distribution of syllable boundaries between English a n d Polish, A parameter of syllabification 59 receives additional motivation from the syllahic processes in Yawélmani such as epenthesis and vowel elision. I t was shown t h a t in Tigrinya too, syllabification must be assumed to take place directionally, but t h a t in this language it applies from left to right, in contrast to K l a m a t h and Yawélmani. The implicit conclusion in Kaye and Lowenstamm, i.e. t h a t the direction of syllabification is a parameter, is thus shown to be correct. I t is plausible t h a t the left-to-right syllabification is the marked setting of the directionahty parameter lor 0 V 0 languages with a obligatorily filled onset. REFERENCES Archangeli, ü . 1984. Underspecificalion in Yawélmani phonology and morphology. U n p u ­ blished P h . D . dissertation, t h e M.I.T. Belletti, A. et al. (eds). 1981. Theory of markedness in generative grammar. P i s a : Scuola Normale Superiore di Pisa. K a l m , D . 1976. Syllable-based generalizations in Mnglish phonology, Unpublished P h . D . dissertation, the M.I.T. K a y e , J . a n d L o w e n s t a m m , J. 1981. "Syllable s t r u c t u r e a n d m a r k e d n e s s t h e o r y " , I n Belletti, A. et al. (eds). 1981. 287­315. Kisseberth, G. W . 1969. Theoretical impUcations of Yawélmani phonology. Unpublished P h . D . dissertation, University of Illinois, U r b a n a . K u r o d a , S.­Y. 1967. Yawélmani phonology. Cambridge, M a s s . : The M.I.T. Press. N e w m a n , S. 1944. Yohuts language of California. N e w York: Viking F u n d Publications in Anthropology. Noske, R. 1985. "Syllabification a n d syllable changing processes in Y a w é l m a n i " . I n V a n der H u l s t , H . a n d Smith, N . S . ' H . (eds). 1985. 335­61. Pain, M. D. 1973. Tigrinya phonology. U n p u b l i s h e d P h . D . dissertation, City University of New York. Ter Mors, 0. 1985. Empty-Vnodes and their role in Klamath vowel altemations. I n V a n der H u l s t H . a n d S m i t h N . S. H . (eds). 1985. 313 ­ 33. V a n der Hulst, H . a n d Smith, N . S. H . (eds). 1985. Advances in nonlinear phonology. Dordrecht: F o r i s Publications.