The architecture of the english lexicon
Mora augmentation and the stem
Yüklə 2,22 Mb.
|
- Bu səhifədəki naviqasiya:
- Lapse- s
- Moraic Prominence
- 5.2.1 Constraint domains
- Stem-Coh
- SuffixMoras
- Onset
5.2 Mora augmentation and the stem The constraint hierarchy just developed needs to be further extended to account for a number of other structural types found among the /-ent/ and / al/ forms. For example, while forms with heavy prefixes of the type (com)(pr´)sal are correctly predicted by the above constraint hierarchy, attested outputs like c—nfident would incorrectly be rejected in favor of *(con)(f´)(dent), and the unconstrained ability to add moras to the representation (thus far constrained only by the low-ranking *m) could yield many unattested successful forms. However, for cases like so³nant, where the stem is monomoraic with no other means of expansion, the possibility of moraic augmentation in the optimal candidate is necessary. In fact, it is only in such cases, when the footable domain is a discrete monomoraic morpheme, that any unusual moraic adjustment (of either type, compensatory or augmented) occurs. To eliminate incorrectly predicted forms like *con(f´d)ent, and to rule out mora addition in most cases, the constraint *m (¤ 4.1.3) will be promoted within the existing hierarchy:77 (5.25)
The crucial constraint here is *m, which eliminates the lengthened *conf´dent due to its extra mora. The optimal form, c—nfident, is an example of a word in which the foot type (smm sm) is acceptable (¤ 5.1.1), indeed preferable within the context of this approach to (smm)sm, which would both incur Lapse-s violations and leave the stem syllable outside the main-stressed foot, violating Stem-Ft-R.78 At this point, the constraint Stem-Ft-L, which was useful in the simplified constraint hierarchy offered in the preceding chapter, is no longer significant and will be dropped from the discussion of relevant constraints.79 As was suggested above, the unconstrained addition of moras to feet, whether by augmentation or compensatory "borrowing", would produce many incorrect forms. For example, bisyllabic monomorphemic forms such as d’gital and —rbital show no lengthenings, although forms like *dög´tal and *orb´tal would avoid a violation of NoOrphans, unlike the attested forms; a similar lengthening might be expected for the same reasons in heavy stems like that seen infŽrnal, i.e., *infe³rnal. Furthermore, no lengthening is ever seen in any stems of two syllables or greater, even when they present unfooted syllables. Thus ma(l’g)nant surfaces with an initial unfooted syllable rather than the less Lapsed *ml’gnant, and we see co(l—s)sal (with geminate penult) rather than *c¯l—ssal. Turning to other types, the same situation is seen in cal‡mity and in unsuffixed words like ar‡chnid, appŽndix. In fact, even when relatively high-ranking constraints such as NonFin are violated, bare stems still fail to show lengthening as an alternative way of satisfying those constraints, e.g., we see (c—met), (br’gand) instead of alternatives like *(co³)met or *(brö)gand. The only cases where lengthening does appear (apart from those discussed above, i.e., monomoraic stems like so³nant) is in forms like (reø)(bœtt)al, (deø)(fŽnd)ant, (eø)(v‡po)rate, where the lengthened item is a monomoraic affix, and the lengthening is optional.80 This suggests a high-ranking constraint which maintains the moraic coherence of stems with two or more moras supplied by Moraic Prominence. Such a constraint could be formalized as follows: (5.26) Stem-Coherence: NI(m, Stem, m) This states that every mora is aligned to either edge of the stem. For each intervening mora, a violation will be incurred. Under this definition (which allows for alignment to either edge, rather than both), monomoraic and bimoraic stems will both satisfy this constraint (since there are two edges), while any further moras will incur violations: (5.27) Stem Stem-Coh /m/ Ã
/mm/ Ã /mmm/ *m
/mmmm/ *mm (etc.)
If surface long vowels were all regarded as underlyingly long, we’d expect to find long vowels anywhere in the word, and see alternations in various positions. Instead, what the data shows is long vowels predominantly appearing in monosyllabic morphemes of the shape /CVC/, and being especially prevalent at the right edge of the word, preceding affixes. Vowel alternation involving long vowels also occurs in stems that appear in these contexts. This, combined with the rarity of long vowels in multimoraic roots and the similar distribution of long vowels at the right edge of finally-stressed unsuffixed verbs and adjectives, suggests that the majority of these long vowels are not present in the lexical entry, but are rather the result of the morphologically conditioned lengthening process discussed above, enforced by the position of the root within the stem constituent, and the presence of the suffix / æ/, which appears on the right edge of many stems of this type.
To derive the correct results from the Stem-Coh constraint proposed above, it needs to be slightly reformulated, and a new concept and notation introduced for the Alignment/No-Intervening constraint family, that of the constraint domain: (5.28) Stem-Coh: NI(Stem: m, Stem, m) This states that for every mora in the domain of a stem, no mora intervenes between that mora and the stem. This limitation is important, because otherwise moras outside the stem (such as suffixal moras) would incorrectly incur violations of this constraint. With the domain now specified in the constraint, such moras will not be evaluated with regard to this constraint. Constraint domains offer a further refinement of the notation for expressing formal constraints, but adding them to the notation does not add further complexity to OT, because domains were in fact implicit (but ambiguously so, as was the case with alignment violation) in the theory already. For example, when an alignment constraint was presented, e.g., Align(Ft, R; PrWd, R) or Align(s, R; Ft, R), it was only evaluated in terms of the feet or syllables in a particular word. However, in an utterance, there could be other syllables or feet, for example in the following word, or in following clitics, inflectional endings, or "level II" suffixes, which could incur unwanted violations of such constraints.81 By limiting the OT tableau to the word, the domain for these constraints was being implicitly limited to the Morphological or Prosodic Word.82 Likewise, when a constraint is relevant for cross-word effects, such as the cross-word rhythmic effects seen in English (e.g., the well-known case of rhythmic stress adjustment, for example th’rteen mŽn), the domain of the constraint would be a supra-word phrasal constituent. It should be clear that it is not appropriate for constraints to apply outside their domains in some cases. Expressing the domain explicitly in the formalization specifies precisely how the constraint will operate. Certain correspondence constraints (McCarthy & Prince 1995) also implicitly refer to domains, as the correspondence governed by such constraints is evaluated only over a certain constituent type (e.g., the reduplicant), and not others (e.g., the base or stem). For the previous constraints offered herein, the constraint domain happens to be the next higher morphological category, the Morphological Word, rather than the Morphological Stem as in this case. Promoting Stem-Coh into the hierarchy of significant constraints as a undominated (except by Moraic Prominence) constraint restricts the possibilities for intrusive lengthening in the stem to footable monomoraic morphemes.83 There is one further restriction on lengthening in words, which is a stronger restriction than on the lengthening within stems. In English, suffixes are never augmented outside of stem constituents, even when they are monomoraic. For example, a candidate *(in)(fŽr)(nl) would be more optimal than the correct (in)(fŽr)nal on the basis of the current constraint hierarchy (since Lapse-s is then not violated at all), but such forms never surface. Suffix morphemes show lengthening rarely, and only when they become "stem-ized" due to subcategorization constraints, e.g., ({vic}){(to³ri)}ous (lengthened by CiV lengthening, ¤ 5.2.2), where the suffix becomes part of stem constituent as a result of the addition of /-ous/, which requires a stem on the immediate left.84 Thus, it is necessary to propose a faithfulness constraint SuffixMoras, which prevents material to the right of the stem (i.e., suffixes) from acquiring any more moras than those projected by Moraic Prominence: (5.29) SuffixMoras: NI-R(PrWd: PrWd, Stem, m) This states that no moras intervene between the right edge of each prosodic word and the right edge of some stem. By ranking this constraint below Moraic Prominence, the number of moras appearing to the right of the rightmost stem constituent will be restricted to the number projected by Moraic Prominence. This constraint will also encourage all "borrowable" moras to shift into the stem (reducing the number to the right of the stem), hence this constraint must also be ranked below Stem-Coh, or moras will shift into the stem even when the stem is already bimoraic: (5.30)
In the rest of the work, candidates which violate SuffixMoras will ordinarily not be presented in the tableaux. 5.2.2 Compensatory lengthening within the stem One well-known lengthening process in English is so-called CiV lengthening (Chomsky & Halle 1968: 47). In such words, a light open syllable is lengthened when followed by a sequence of /i/ and a non-high vowel. (5.31) cana³dian manage³rial adviso³rial colo³nial sta³tion pala³tial ara³bia come³dian The vowel /i/ itself in such a position, however, is not lengthened, e.g., off’cial, mil’tia. Furthermore, if the syllable in question is closed, lengthening does not occur, e.g., prov’ncial, advŽrbial. The appearance of a long vowel under the stated conditions, however, is nearly universal throughout the data. These facts suggest an adjustment inside the stem, that is, a mora internal to the stem is shifted to the lengthened syllable. The most likely source of the shifted mora is the vowel /i/ itself, which appears before a vowel of higher sonority and so can be conceived of as linking to the otherwise empty onset position of the following syllable. This is clear enough when the consonant preceding the /i/ is palatalized, as in pala³tial. The palatal place features of the /i/ can be understood as linking to the onset, the result surfacing as [°]. The syllable belonging to the /i/ has no surface instantiation; its mora has shifted to the preceding syllable. In derivational terms, this could be understood as a process: (5.32) s s s s s s s s | | | | | | \ / | Mm Mm Mm Mm þ Mm Mm m MMm / | /| /| | / | | |/ | / | p a l a t i a l p a l a t i a l
Of course, in an OT account, a serial explanation is not possible. Instead, the same insight can be captured by proposing a contrast between the way moras are projected by Moraic Prominence, and the way they are associated to segments by other constraints. This contrast is implicit in Zec’s (1994) approach to compensatory lengthening in various languages. Here, although the /i/ projects a mora, it fails to link to that mora.85 The constraint governing this process is likely to be Onset (Smolensky & Prince 1993), which demands that all syllables have an onset. It is the demand for an onset segment which takes the /i/ away from its syllable and mora: (5.33)
When /i/ joins the onset of the final syllable, the Onset constraint is satisfied. Additionally, the fact that the stem segments /t/ and /i/ are now outside the head foot is not relevant, as constraints like Stem-Ft-R refer to whether a syllable intervenes between the stem edge and the foot edge. So long as the misaligned constituent is only an onset, no violations of such constraints will occur. For candidates where /i/ remains in the stem foot, Onset will always be violated.86 For words like advŽrbial, we might expect a similar structure, namely ad(vŽmrmm)bial, where the extra mora links to the consonant /r/ and thus no detectable surface effect occurs. However, it appears that something besides the need for an onset is interacting with the constraints given thus far; otherwise, words like mil’tia cannot be explained. The vowel /i/ itself is never lengthened in the CiV environment, although the Onset constraint in the hierarchy given above would demand lengthening. Thus, the Onset constraint is not sufficient to trigger the mora shift, and it appears that there is a further requirement that the shifted mora move to a more sonorous segment than /i/ itself, that is, only to the vowels /a,e,o/. Also, the mora will always shift to the left, never to the right. These further conditions suggest that more constraints are relevant to the behavior of moras in this context. The most obvious reason for the mora to apparently shift to the left is the need for the foot to remain bimoraic, enforced by the high-ranking Min 2. A form *(pˆla)(tÆa³l) would not only violate Non-Finality, but would place the head foot outside the stem. In most words that display CiV lengthening, the /i/ in question is on the right edge of the stem, often as part of an / ity/ class suffix like / ion/ or the / i / extension extending many stems further suffixed by /-al/ suffixes (i.e., forms in / ial/, / ious/, / ian/, / ior/, etc.). Thus, the mora projected by the /i/ optimally aligns within the stem, and within the foot (which will align to the stem by Stem-Ft-R), even in candidates whose /i/ segment itself leaves the foot to form an onset. Likewise, in CiV lengthening cases when the /i/ is not stem-final, it is in foot-final position in what is often the only plausible foot in the word, e.g., (ma³ni)ac, which can be understood as {(ma³)nÆac}. However, when /i/ occurs in a CiV position, but is not foot-final, a very different effect is seen, similar to that seen when other vowels become adjacent, e.g., man´acal. The structure of such forms can be understood as follows: (5.34) s ( s s ) s | | | |
Mm Mm Mm Mm / | /| / | / | m a n i a c a l
Forms like man´acal offer one example where it is more optimal to multiply link /i/ to two subsyllabics than to delink its mora. Multiply linking the segment still satisfies the high-ranking Onset, and indeed all words where vowels are contiguous in this way show this apparent lengthening which indicates onset-filling, e.g., the³ater, Bo³as, cr«a³te, pa³ean. Thus, the constraint against multiple linking of a segment to two subsyllabic constituents must be fairly low-ranked. Such a constraint might be formulated as follows: (5.35) NoMultiLink NI-R(Seg, m, M) This states that the right edge of a segment should align to the right edge of a mora, with no margin constituent intervening. A segment multiply linked to a mora and a margin constituent (but not one linked to two mora constituents, as in a long vowel) would incur a violation. Another instance where this constraint appears to be violated is in words like mil’tia, where the structure is best understood as mi(l’ti)Æa, the /i/ being linked to both the final mora in the foot and the onset of the last syllable. A likely explanation for the failure of the preceding /i/ to lengthen is that the mora, to escape the segment that projected it, requires any segment it associates to to have higher sonority than the projecting segment. This might be understood as a constraint aligning moras to the feature [+low] when possible:87 (5.36) m->[+low] NI(Foot: m, [+low], Seg) Preventing a mass migration of moras to segments like /a/ at the expense of all other moraic segments would be faithfulness constraints ensuring that each segment has a subsyllabic constituent to correspond to, for example the constraint Moraic Coherence (5.17). The vowel /i/, however, cannot draw off the mora, since it is of the same sonority as the foot-final /i/ offering the additional mora. The same would be true of any moraic consonant found in a closed syllable preceding the /i/. Thus, only vowels containing the feature [+low], in open syllables within the same foot can undergo this type of compensatory lengthening:88 (5.37a)
(5.37b)
For cases where the /i/ is not foot-final, the superordinate constraints governing foot placement override these constraints: (5.38)
Thus the only instance in which CiV lengthening can take place is when a stressed open syllable with a low or mid vowel is followed by the weak member of the foot, a syllable with the nucleus /i/. If the head syllable of the foot is closed, no candidate expanded with the /i/’s mora will succeed. The phenomenon of CiV lengthening also gives further evidence for the gemination proposed in ¤ 3.3.2, as words with such geminates, e.g., rem’ssion, pot‡ssium, fail to show CiV lengthening, which is best explained by regarding the head syllable as closed and already bimoraic due to the geminate, i.e., (m’s),(t‡s). It might be asked why /u/ does not cause a similar lengthening, for example in words like gr‡dual, c‡sual, cont’nual. However, the segment /u/ in such cases is not simply the vocalic correspondent of the glide /·/, but rather shows a complex features set which includes palatal and velar elements. The palatal element often surfaces in the onset of /u/’s syllable, as in the words above. Both the palatal and velar features are evidently not able to align to a given subsyllabic at the same time, hence this /u/ cannot simply join to an onset and give up its mora in the way the simpler /i/ can. For such cases, the round and velar features of /u/, on the right edge of the segment, can act as the onset of the following syllable, while the palatal features on the left edge can act as the onset of the /u/ syllable itself, the mora still remaining within the foot. In this sense, /u/ is always multiply linked to mora and margin constituents. Yüklə 2,22 Mb. Dostları ilə paylaş:
|