92 415p. MFO1 /PC17 Plus Postage.

ESCOL '911-22ON THE METRICAL UNITY OF LATINATE AFFIXESLuigi BurzioThe Johns Hopkins University1.IntroductionThe Latinate subset of English suffixes breaks down into twosubclasses, which I will refer to as Class I and Class II, followingestablished terminology. The suffixes of Class I cause changes inwhere underscore(1),as illustrated inthe stress pattern,identifies the stress of the unsuffixed stem.(1)CLASS I AFFIXES: Restressinga.-al: accidental, medicinal, original, parental, prefixal,pyramidal, triumphal, universalb.-ic: 'inguistic, realistic, allergic, oceanic, astronomic,gymnastic, homeric, idiotic, prApheticc.-ion/ -ation: congregation, consecration, demarcation,insulation, integration, intimidation, affirmation,allegation, centralization, combination, compilationd.-ous: advantAgeous, courageous, outrageous, incestuous,momentous, voldminous, ridiculous, tempestuousIn contrast, the suffixes of Class II generally preserve the stemstress, as in all the cases in (2).(2)CLASS II AFFIXES: Stress-neutrala.-able: accgptable, adaptable, affordable, expandable,oppAsable, refAndable, respgctable, surpassable, sustainableabolishable, alterable, answerable. cherishable, colorable,deliverable, inhabitable, inhgritable, interpretable,perishableb.ist: pharmacAlogist, perfectionist, empiricist, geneticist,romanticist, extremist, h6morist, propagandist, terroristamericanist, capitalist, generalist, individualist,modernist, monarchistc.absentgism, alarmism, defgatism, escapism, extremismmodernism, monarchism, capitalism, fgderalism, liberalism,literalism, naturalism, radicalism, amgricanism, fAvoritismd.ant/ ent/ ance/ ence: consultant, contgstant, defgndant,11)V3

2dependant, determinant, expectant, inhabitant, registrant,relaxant, resistant, resultant, visitant, compliance,deliverance, inheritance, resemblance, resistance,severance, sufferance, absorbent, antecedent, coexistent,consistent, convergent, depAndent, reminiscent, subsistent,transcendente.ment: abandonment, accomplishment, achievement, advancement,amendment, berAavement, bombArdment, devAlopment,embodiment, envelopment,accompaniment, gAvernment, impoverishmentf.ize: propagandize, anthopomorphize, europeanize, italicize,romanticize, compAterize, gAlatinize, monopolize,revolutionizecapitalize, municipalize, familiarize, popularize,americanizeg.may: felony, matriarchy, monarchy, butchery, orthodoxy,Accuracy, confAderacyIn the past, from Chomsky and Halle's "SPE" to Halle anc Kenstowicz(1991), stress neutrality has consistently been analyzeo as a formof evasion of stress or immunity to the stress principles, thoughthis idea has been implemented in different ways. In SPE, thedistinction between the two classes was a difference in the type ofboundary that the suffix occurs with. Class II suffixes wereassociated %ith a word boundary "#", while Class I had a morphemeboundary " ". The restressing character of Class I suffixes thenfollowed from the fact that the stress rules reapplied at eachsuccessive "cycle", and had the power to alter previously assignedstresses. Neutrality, on the other hand,followed from theassumption that rules of word stress, while applying across morphemeboundaries, did not apply across word boundaries, hence leaving thestem unaffected by later cycles, as stated in (3).(3)Chomsky and Halle (1968) "SPE":i)parent alstress rules reapplyii)american # iststress rules fail to reapplyIn Kiparsky's (1982) influential "Lexical Phonology" model, thedistinction was not one of boundary type, but rather one of levelof affixation in a sequentially organized derivation. Class Isuffixes are in this system attached at "Level 1", followed bycyclic application of tie stress rules, while Class II ones areattached at "Level 2", which is beyond the reach of the stressrules, as in (4).

3(4)Kiparsky (1982) "Lexical Phonology":Level 1i)I Morphology: Class I affixationPhonology:Level 2ii)Stress rules, .Morphology: Class II affixationPhonology:In the alternative model developed by Halle and his associates,which we refer to as "Cyclic Phonology", the Class I/ Class IIdistinction is not due to the different level of affixation, butrather to the different sets of rules that each class triggers.Class I suffixes have a lexical diacritic specifying that they are"cyclic", and as such trigger cyclic rules. In contrast, Class IIsuffixes are "non-cyclic", triggering only non-cyclic rules, whichapply after all cyclic ones, only once per word. This schema isillustrated in (5).(5)Halle and Vergnaud (1987a,b), Halle and Kenstowicz (1991)"Cyclic Phonology":i) Cyclic Phonology, triggered by cyclic affixes: Class I(main stress rules)ii) Non-cyclic Phonology, triggered by NC affixes: Class II(other stress rules)Note that the organization in (5) does not immediately account forstress-neutrality, unless the set of non-cyclic stress rules istaken to be null. In Halle and Vergnaud (1987a,b), this is not so,however, and the exact account of stress-neutrality is not madeexplicit. An explicit account of neutrality is given in Halle andKenstowicz (1991) extending the general framework of Halle andVergnaud (1987b). Their analyses are illustrated in the derivationsin (6) below.(6)Iparenta.b.american cycliccyclicpareI nta'alb'american ist cyclicnon-cyclicCyclic rules: r-1 parse; "Stress erasure convention"1-r parse; "Crossover constraint"ii) NC rules:i)1.1

4In (6a), a first application of the cyclic stress rules, which applyright-to-left, parsing syllables ir. ) feet, results in the metricalgrid given. In (6a'), a second apolication of the cyclic rulescorresptndingly results in the new yid. There is no preservationof earlier metrical structure here, because cyclic rules areassociated with the "Stress Erasure Convention", which eliminatesthat structure at each new cycle. In (6b), the cycle is just as in(6a), but in (6b') the suffix is non-cyclic, hence triggering thenon-cyclic rules only. The latter are in fact presumed to be rathersimilar to their cyclic counterparts in overall composition, butwith three crucial differences: i) non-cyclic rules parse thestructure left-to-right (Halle and Kenstowicz' innovation to theHalle and Vergnaud framework); ii) they are not associated with theStress Erasure Convention; iii) they are subject to the "CrossoverConstraint", which states that a parsing procedure cannot jump overpre-existing metrical structure. As a result of the conjunction ofi) and iii), Class II suffixes, like 1st in (6b'), remain unparsed,hence evading stress much as in earlier models. As a relt of ii),the stem stress will surface uncnanged.IIn this paper, I present an alternative account of stressneutrality based on the hypothesis that neutral suffixes are subjectto normal parsing into feet, like non-neutral ones, though twodifferent modes of metrification distinguish the two classes. Thestructure of my proposal is schematically illustrated in (7) below.One important aspect of it is the assumption that metrical structureis part of the underlying representation of words, subject to outputwell-formedness conditions, that is the assumption that there is nostress-assignment, but rather stress-checking. I assume furthermorethat thereisarequirement that morphemesbemetrically"consistent", that is that they surface with the same metricalstructure to the extent possible, just as they tend towardconsistencyof segmentalandsemanticstructures.Metricalconsistency of both stem and suffix will occasionally be achievable.In the more general case, however, the two will be incompatible,hence requiring that either the stem or the suffix metrify in a newway, inconsistent with other occurrences of the same morpheme. Theproposed distinction between Class I and Class II is now that withClass I suffixes the suffix wins, impo:ing its own metrification onthe stem, as in (7a), whereas with Class II suffixes the stem wins,forcing the suffix to metrify accordingly.(7)a.parentparentalalVISIaNSINNONC b. american istamerican istWe will see later on that, from this perspective, it is in facttlf

possible to predict which suffixes will behave which way. The mainpart of this paper is a collection of arguments for this approachIf correct, thoseand against any form of "stress evasion".arguments will suggest that models of the lexicon or of thephonological component that have stress evasion as an availableoption are overly permissive in rather significant ways.2. Analytical FrameworkTo make the proposal in (7) concrete, it will be necessary tointroduce a certain amount of analytical machinery. This will bedrawn from Burzio (1987), (1990), (1991), (in progress). We beginby considering the two well-known generalizations of English stress,as in (8) and (9).(8)(9)a. Heavy penultimate:agenda, appendix, horizonb. Antepenultimate:america, asteriska. Superheavy final:pervert, decideb. Penultimate:inhabit, imagineThe pattern in (8) is the one characteristic of nouns and suffixedadjectives, while that in (9) is characteristic of verbs andunsuffixed adjectives. The "super-heavy" syllables of (9a) aresyllables whose structure exceeds that of normal heavy syllables byone consonant, while a heavy syllable is standardly defined as onethat either ends in a consonant or has a long vowel.In Hayes (1982), (1985) and much subsequent work, the twopatterns in (8) and (9) are analyzed in terms of a single parsingalgorithm, by supposing that English word ends are metricallyambiguous, in the sense that the parsing, proceeding right to left,can start at two different points in the structure. Specifically,Hayes has proposed that at the right edge of a word either a fullsyllable or a single consonant can be "extrametrical", namelyignored in the parsing, so that the parsing will then consistentlyconstruct either a monosyllabic foot (H) or a bisyllabic one isLI,where "H, L" stand for heavy and light syllables respectively. Thiswhich has been very influential, is schematicallyapproach,summarized in (10).(10)After Hayes (1982), (1985):Word-ends metricallyambiguous: .1,4/Feet al/ iglia.a(gen)afire ri)dacab.per(ver)in(ha bi)ttc.(per)(as te)vertrisk

6In Burzio (1987)-(in progress), I present a different approach. Thelatter maintains Hayes' intuition that there is a single parsingmechanism at work and that the difference between (8) and (9) is dueto the ambiguity of word ends, but it iLplements it in a ratherdifferent way. A central assumption of that approach is thatsyllabification is in general free to parse phonetically emptystructure beyond the visible edges of words. Given the well-knownprinciple that requires consonants to syllabify as onsets wheneverpossible, that assumption will entail that final consonants willalways be onsets, since they can be followed by a null vowel, in themanner of (11).(11)per.ver.to,this hypothesis, there willthus never be truly "final"consonants, all English words ending in a vowel, overt or null. Itis easy to see that this analysis now reduces the stress pattern ofthe items in (11) to the penultimate/ antepenultimate pattern ofa.oen.da/ a.me.ri.ca. More generally, it will reduce the pattern of(9) above to that of (8), as we see below.In addition, thisOnhypothesis, which extends to other languages, accounts for thephenomenon of "super-heavy" syllables, and the fact noted by Hayes(1982, p.229) and otherwise unaccounted for, that in English andother languages ".superheavy syllables may occur only in -"superheavies" being sequences of two syllables, one with a nullvowel. Their occurrence only peripherally reflects the distributionof null vowels. General motivation for Hayes' extrametricalitydisappears as well, since feet can now be taken to be uniformlybinary (Ha) or ternary (oLo), as in (12). Further advantages forthis approach are discussed in Burzio (in progress).(12)Burzio (in progress): Feet (Ho)/ (ola)a.a(gen da)a(me ri ca)b.per(ver to)in(ha bi to)However, some notion of extrametricality is necessary in thisapproa-h as well, in order to handle nouns ending in a consonant(that is, a null vowel), in the manner illustrated in (13).(13)(per ver)(as to ris)t.kSpecifically, we must suppose that syllables with null vowels aremetrically ambiguous, that is optionally extrametrical, much assyllables in general were optionally extrametricalinHayes'1EST COPY AVAILABLE

7framework. But consider now that the type of extrametricalityinvoked by (13) is in fact attested independently of the null-vowelhypothesis, with a certain class of special syllables that we referto as "weak" (W). The class of weak syllables, in which we thusinclude those with null vowels, is illustrated in (14).(14)-Metrical ambiguity of "weak syllables": . WIA/ ,10-W class: o, y, ive, ure, [son]a.b.c.d.e.f.per(ver to)an(ti pa thy)ob(jec tive)ad(ven ture)a(pos tle)de(cem ber)/(per ver)t/(ef fi ca)cy(ad jec)tive(a per)ture,(ve ge ta)blerac)ter////(tem pe ra) tureThe extrametricality of the right-hand examplesin(14b-f)isessentially "theory neutral". That is, it must be postulated in anytheory that takes the basic generalizations of English to be as in(8) and (9). The reason is that the stress pattern here nsthoseextrametricality of (14b-f) must thus be postulated in Hayes' extrametricality, as for example in (effi) ca fcvl, where " " isand "[J" thesystem,extrametricality of Hayes'the normalextrametricality of this special class of syllables. Our system in(12) thus accounts for the two patterns in (8) and (9) by minimallyextending the extrametricality needed for (14b-f) to the case in(14a), whereas the system in (10) does so by introducing unrelatedmachinery. As shown by (14), weak syllables are then syllables withvowels, syllables with the high vowel corresponding tonullorthographic y, syllables ive and ure, and syllables with sonorantnuclei. As discussed in Burzio (1991), (in progress), I presume, albeit tentatively- that the distinguishing characteristic of weaksyllables is in fact acoustic weakness, their metrical behavior thenreflecting the general alignment of metrical structure and acousticprominence.Weak syllables have another property beside the ability to beextrametrical. When they are metrified as part of a binary foot(He), that foot, which we will refer to also as "weak", fails toattract primary stress, as illustrated in (15a-f), which describesthe same spectrum of weak syllables as (14a-f).(15) Weak Foot: (o W)a.bernar (di no)d. archi (tec ture)b. contro (ver sy)e.taber (na cle)c.inno (vA tive)f. alli (gA tor)We interpret the phenomenon of weak feet as also related to theactual, namely acoustic, weakness of weak syllables. Weak feet are

8presumed to be acoustically weak because they are structurallyminimal, namely binary, and they also incorporate a weak syllable.Primary stress in English is then taken to fall on the rightmostfoot which is not weak --the latter restriction resulting again fromthe general alignment of stress and acoustic prominence.3. Stress Neutrality and Weak SyllablesWe are now ready to make the proposal in (7) above moreconcrete. Specifically, we propose, as in (16) below that stressneutrality is a by-product of the rather general ambiguity of wordends in English, alias the double option of metrificatioh or nonmetrification of final weak syllables.(16)Stress-neutrality results from the ambiguity of word ends,namely.)W/ versus .W).Assuming then as in (7b) above that with neutral suffixes the stem"wins", we consider the cases in (17) to see how suffixes can beintegrated into the metrical structure of the stem under the thesisin (16).(17) a. ac(cep to)b. propa(gan da)ac(cdp ta)blepropa(gan dis)tc. a(me ri ca)na(me ri ca)(nis to)The case in (16a) is rather straightforward. Here, the firstsyllable of the suffix able supplants the final syllable of the stemaccept (i.e. the null vowel), while its second syllable ble --a weaksyllable (compare (14e)) is allowed to remain extrametrical. As aresult, the rightmost foot of acceptable is identical to the one ofaccept, whence the identical stress. The case in (17b) is rathersimilar, as the first syllable of the suffix also supplants thefinal one of the stem. while the second syllable of the suffix,being again weak, remains extrametrical. The case in (16c) is a bitdifferent, since here none of the suffix can be incorporated intothe final foot of the stem. But the other option available to weaksyllables can be resorted to. By metrifying that syllable, thesuffix will form a separate foot of its own, still leaving thepreceding one undisturbed. This of course predicts a stress on ist,but only a secondary one, due to the weak foot. The primary stressis then correctly predicted on its stem position. Note that whileperceptual evidence does not inchdendently support the view that istin (17c) has a secondary stress, it is nonetheless compatible withit. For perceptual prominence of syllables with secondary stressesappears to be rather systematically non-distinct from that ofunstressed heavy syllables with unreduced vowels.2We will see later on that the type of account of stressneutrality illustrated in (17) and based on the ambiguity of weaksyllables can be upheld in all cases.3 One might ask, however, whythe same ambiguity should not suffice more generally, lable,

9incorrectly. We will see that there is a principled answer to thisquestion, but in the meantime we may consider the following as apreliminary account. Let us suppose that Class I suffixes arelexicallyspecifiedashaving metrically mflambiguous ends,specifically by means of a foot boundary which either includes orexcludes the final weak syllable depending on the suffix, as in(18).(18)Non-neutral suffixes are specified with unambiguous ends:0115A,a. (pd ren)tpa(ren tall,a' (py ra mi)dpy(rA mi dal',b. (lin guis)tc.(mó men)tmo(men toutsc'(ri di cu)lelin(guis ti colb' (home)roho(mericolri(di cu loulsNote that comparable markings seem required in any theory, so as todistinguish for instance al or ous, which place stress according tothe pattern in (8) above, from ic, which follows the pattern in (9),placing stress always on the immediately preceding syllable, as (18)shows. In our analysis, the same, independently needed, diacriticwill also serve to distinguish non-neutral suffixes from neutralones. The latter will simply lack the diacritic.4. Arguments against Stress-evasionandWenow turn to the arguments for our appro

Focus in English"; "Phrasal Input to Derivational Morphology in . Department of Cognitive Science at Johns Hopkins University, the Department of Linguistics at the University of Maryland, College Park, and the Department of . Anna & Maria Tere