Lexical Morphology and Phonology Paul Kiparsky Mkt, 1. Lexical Morphoony The approach to word suture that I shall explore here represen 3 convergence of several enpinally independent strands of researeh. One isthe merging theory of morphology and the lexicon (. ArenolT 1976), snd tore paral the iden of eve-ordered morphology elaborated by D. Siege! (197, 1977, M. Allen (1978) and others. Another centers around the problem of constraining Ieseal representations and phonological rules, oginning with various verons af the Alteration Condition (Kiparsky 19st, T973) and continuing with the coueption of Cyeie Puonslogy ist proposed by Masear (1974) and subsequently pursed ina numberof ties {tthe phonologits of patel languages (ost extensively Rubach (1961). shall ao be drawing on aspects of the recent mera theory of sess (bern and Prine 197, Hayes 1981) and sable strueiure. When these ‘ews ave put together nd developed in a extan dzection they explain a ‘eves of properties of phonolgieal rues and thei ration to morphology nd the fexicon that have s for appeared as unexplained generalizations, ‘rin some cast even deed coherent formulation or escaped notice ak together “Th basic insight of Ievehorderod morphology is that the derivational and Inetonal processes of language canbe organized ina series of levels ach evel it asecated witha set of phonologial rule for which it defines the domain of application. The ordering of levels moreover defies the posible ordering of morphological procates ia word-formation. Following [propost!of Peeisky (1979) let wt assume that the output of each word. formation pnoces i submited within the lescon ist the phonological ‘ules of ts lve. This establishes a basi division ameng phonological rules ‘into thowe which ae assigned to one or more eel inte lexicon, and thooe Which operate after words have en combined into senfenes inthe syntax. “The former, the rules of leicol phono) are intrinsically eyelicbacause they reapply after each step of word-formaton at their morphoogia! eve 131 ‘The Linguistic Society of Korea (ed), 1982, Linguists in the Morning Calm, Seoul: Hanshin Publishing Co.4 ra ay ‘The ater, the eues of postexicalplonolog, are inwaszally noneyelc, The lexicon i accordingly sete nthe follwing way o speed ese ems [jes tne —=| [waning =—| apn] fio \ 4 etn morplony | evel a phonology —— rs ee} $f rococo] ‘The output ofthe eght-hand boxes eolectivelyconstuts the set of eed stems ofthe language: The point thatthe result of every layer of dersation {sie excl item wil have important consequenes late on Mods ofthis type have also been investigated by Strats (199), Bool (0981, Plleybink (1981, and especially by Harris (1982) for Spanish and ‘Mohanan (1981) for Malayalam, the later aso wih extensive theoretical justcation ofthe framework. For conreeness T add tentative sketch of how the English lexicon ‘might be organized Tt draws on previous explorations of Bagsh morphology ‘by Seg! (1978), Alla (1978), Selkirk (MS), and Willams (198), bt tere in some respects from each. eaculaates aaeas ; 7 ee | =n || [+ boundary” ntetion “| stress, shortening evel 1 cee | rcvoundary eration ——=| compound cess | | et 2 ese fe ‘poundary” intetion | tuning evel 3 poalleel plonoogy (OF the three levels in (2) the fis evel comprises the aes which have ‘usualy bron atest with the} Hoondary. They corespond to the “Seimaey aufins” of traiional descriptions such 26 that of Sanskrit by Whitney (189), This lve ncdes derational sues such a8 ou. us, “iat an ection vfs such a thse Inept, me, Mo, hen ‘adicnde, ines, oct aswel as “blast, “umlaut” and other steanging ‘morphology as in tate, bled, hare, vot, Uie. To the second Teel we ‘sign #-boundary secondary”) deviation and compounding Such deriv tional sofnes at hood nee, eri, belong hee. The third level takes fae of the remaining, “regula” inletion (leaped, pleated, books, com> ‘drs, indexes, crocuses). ‘Aithough the division between level {and level 2 fies coincides entirely ‘withthe famine distinction berween the “-+ boundary” and the "# bound {ny lie, we hall se ttt infact Bas deeper roots inthe morphoogial fate. la what follows Iwill frst motivate 2) on morpholopial grounds tnd then proseed to develop some of its consequences onthe phonological ide6 Pa easy Fist let us introduce « specie format for morphological rules ia the grammar Following Lieber (980) [shall dstingush two basi clases of ‘word-formation process, compounding and afiation, aid asume thet ‘wordformatin i endocentie By this lena tha the category ofa derived ‘word is alwaysnon-distinet from the category of ead in English wally the rightmost constituent (cf. Williams 1981)" The categories in question include no ony lexical categories but ao features like Transtve, Agent tte. In any approach to morphology its necessary to provide eetsin kinds of information Tor any given aft: to what sorts of things the ax nay be fdded, whether i is prefixed or sufied, and what are the properties ofthe ‘esting form. Lieber propos tha this information is encoded i the lexis ety of thea ise by means ofa subeategorization frame and an inherent ‘ateporial specication Which pereoltes spward fom an alto the eon strvtions whose head it i. For our purpotes it wll be convenient to fonsiue thes subeategoriations and inherent feature spciatons of axes asso many contextual restitions on the rules whi set then ‘Aes wil then not be lexial entries and they wil ave lexical features ithe inhereny ory peraation Exch af A inodued hy aut he {@).lnsert A in env. [X_ Zh where ¥,Z corespond 16 the “subsategorization frame” of A and X cor ‘sponds ots inherent categovalspecifeation in Lieber’ format, ‘Consider fst infection. Inflectional afies re added to stems inthe content of excl and paradigmatic features asocated with thve ses, To lutrate et the noun ox have ben inserted 8 [ones A MOFDROIR fal ruleat evel I then obligatory ser the satin“ ae i 4). Taset fn neo. OX homes Tee moun boys not subet oak > Pra undergoes ae 3) a ©) laser Jf env. Koons ‘The morphological processes at dierent ves are elated in certain charac tetitic ways which recur from language to language 1. One phenomznoa to be accounted for tthe "bioeking” fet, which hasbeen discussed for derivation merphetogy by Paul (1896), Esau (1973) ‘Aronof (197, Clark and Clark (1979), Toman (1980), and shows sp sl ‘more eles i infction. Words which are ifecte at level 1, for example, usually do aot receive the general suites a evel 3. Tere sno “fot, oer onside fect, oxen, and no *kesped, ected alongside kept, met, We sal however have co explain not only why such the normal case but alo how itis posible for occasional doublets to exist, eg, feclednet, dreamed ‘heat, crocuses|croi, indexes. In deivationl morphology doublets fre aetally quite common, tothe point that blocking thre can harly be eee ee ee eee ee eee eee eter Trt rr ey ase Marty a Poles 1 conde or than general nde. ‘Costing poromeaan flv om obliga of morbologa su Foretample xf irl, ware) an 0 te 00 ‘yw dre eves byte lr le (Fae of cing, sing Ie WEuze he vowel! and domed ih ne ak ote She maked ces cba by mang th ps ale onal fe hve “Ti c af them als To et the Mocking ee for deans iy, we tthe ao a8 imei he conte ofthe appopramerpeopea tego. We sal ‘pepe tht lca ns are hee) sand foetal Tame (he ie kibs bund of fetes We ve wih he evan of eet TREE cto A eed tf yrs low oun to be made at ve 18 TE ine sane ae denon (hich aisto bea phonology ‘Sip. Thee sts ar mond at eve! Ty sch le “One I nen Ve heroes wee pie Bre soi So fom fpeh eco we dev he nan Ue flan wich i fed teas it endocnt FV 9008 esta apy an Teatig soured ou bene tl tcc The ner ‘Sisyetslnon tha eon sujet one of hoe eel 1 rues tend Sere eri aternse potter ened y the eoroponing fue (7) at level 2: (0). nset ern en. [V_enn sae ®) soy. tae inbavitant: “inhabit ‘And wien both do exis, th grandhandy bandstand S grandstand level: grandstand, = pandstanded [Nouns ean only be formed from untenied (ia parla, [~ Past verbs, ‘boca tense most agrex witha nominative subjet ef. Kiparsky and Kip: sky 1971, 3962) and nouns do not have nominative subjects. Only when {he verbs derived fom the compound aounat level 2 can tnse be assigned; at that point level infection s no Tonge accesible and only the rep Infotonal sues canbe assigned. More generally, we are now in postion to explain why exocentric (bahay ‘ih compounds are charactersialy flected t evel 3 even If thie second ‘members are by themselves infected at level 1, whereas endocentic com> ‘pounds retain the inflection that thei second member has by iel. Consider gil eth endocenti) vs. saberiooths sabertooth tigers” (exocenti), ‘AS inthe verb compound just discussed, endoceatri noun compounds are formed at level 2 by combining words, including words derived at level 1 such as feth. Exocenticcompocnds however, must on ou assumptions be assigned zero deeivatonl suffer since they otherwise would share the prop fetes oftheir heads, Le be endocetrc. But as noted i Ta 3, desivational Sufies cant be added to derived poral, Therefore exoenti compounds Le Mngt a Pombo 4 come out of level 2 with exclusively singular morphology and can resive pluralending only at evel 3 where they areadjoned te the whol compound “The model of lexial phonology dicedy prods the eorrlaton between ‘poundaty strength” and afix order which was observed for English by Stes! (197), ands eppareniy a general propery of languages. The several {anion that fie of leven are not added to stems which already contain afes of evel n+ I. For example, consider the two negative reese and those asilates to following consonant (Sigh 1981) but an. does hot, In standard generative phonology this is dealt with by assigning them + and 2 boundaries espocively and restricting assimilation to pel cross + only: (US) in + legible > legible ton # legible» *rollegible “This phonological diference between non and ini lated to the fact that ranean be prefied to word with n-Btdn-cannot be prefixed 1 & werd sith om (15) nonilegible Sinnomlegible Inthe present theory both the phonology and the orering ofthe wo prefs follows fom plting iv and the assimilation rae on level | and nan on eve 2. No boundary i then needed to block the assimilation of non. In this Way, toundary symbols sich b+ and 3 ean be etl eliminated from phono- Topical representations. The requite information Is carted. by the ap- propriate ordering of levels and the morphological bracketing of the sting {G£ Straus (197), Mokanan (1981). “The ecic application of phonologel ules has generally been assumed to ‘beset tothe convention that infernal brackets are erased at the end ofa ‘yce (Chomsky a Halle 1968, 20) We shall assume here the weaker vet= Son (17). equivalent 10 the “Opacity Condition” of Mokanan (1981, (1) Bracketing Erasure Internal rackets are erased atthe end of aleve. Hence the we of even the limited boundary information encoded in the ‘morphological Bracketing i restricted to the level at which the morphology isl assigned. Moreover, morphological ules alo do not have aecess 10 internal morphological structure of ener levels. esisky (197) and Allen (978) have suggested more restrictive conditions but these appear to be tic to maintain in view ofthe English example in 2) below, the Mi lam cases cited in Mohsnan (1981), and the extensive material disused, fom diferent point of view, in Carstairs (198) Tn addition o giving a more resretive theory of morphological junctures, level phonology makes it posible to deal with phenomena where boundary SSmmbols fai. A wel-knowa problem fr eyeic assignment of word sess in Poa Ky “English arises in 2x0 derivation. When nouns ae formed ffom verbs they ‘may shift 0 the nominal sess pattern (18), but hen verbs are formed, fom novns they do not shift as expected othe verbal stress pattern (180) (U8) a. tory > torment b. ptlermy > *pattéeny (ear, wry) ‘Tis erence in sess behavior i dicey accounted for by forming nouns fom verbs at evel f and verbs from noun at level 2, where they ecape the level rules of word ses. This correlates inthe ist instance with te pro dctvty dileence that we expect between level [und level 2 dervational processes. N to V derivation enjoys gest productivity in English (Clark and (Clark 1979), while V to N desivation i comparatively rxrited in cope, AB 8 broad generalization it can be sid tht verbs ae freely seronderved from foun whenever ot blocked by a synonymous formation at level I sch ae _pstematize 0 sytem, while nouns ae 2r0- sons) 2. As for morphology sine compounds are form at level 2 we corey predic hat noun compounds can become verbs but verb compounds cannot become nouns. This entely coret (09) a to grandstand, to wallpaper, to snowball, to quarterback tan aironditon, a stage-manage 2. By assigning V+ N and N-» V sero derivation to levels ¥ and 2 re spectively we en ako soe why verb formed with suines never yeld zero derived none GO)" "a publicize, Pa demonstrate, claiy while nouns may yield zerodervedverbe if they ar formed with level 1 sufiacs (21a) though no if they are formed with level 2 suffixes (210) (@la) "to pressure, to pictur, to commision, to proposition, {0 requisition, to tril to engine, to reverent, 10 reference ib) “to singer,*o beating, "to freedom, Ho prompines, “ro sanpiomt, 0 ssa, Mo nao, Yo sisterhood SEE EEE EEE HE EEE eee eee eee oe eee eee eee Lek Morgen Phowos B Ie we stipulate that zero suits cannot beaded 1 sufxed forms am 1x) hese fat fallout from Bracketing Erasure (17) andthe level ordering we hhve assumed. The nouns in 20) and the vecbs in 2b) are excluded by (2) ‘cause they ate zero-drived fom stems which are sufixed at the same evel. The vets in (2a) do not volute 22) bueaee the internal brackets fom evel} ate erased by (17) at lvl 2 where N to V derivation apples. “f Nouns zoroderived from verb should be capable of receiving level sulle, while verbs zro-devived from nouns should not be so capable. The ‘aampls in (23) show that zero-drived nouns ean indeed get eel ules (G3) contractual, murderous, rebelius ‘And it dss also sem oe tre hat here are no cases which would involve {nding a level I deverbnl sls to 2r0-derived denominal ver (24) estration, "gue, *patternance, “rusadaory, *ementant ‘But, as predicted, level 2 deverba sults canbe added to zeroderived de minal verbs (25) placement, commissonsble, veer, masquerading 5. A special ete ofthis relation botweca level T and level 2 derivation i that V+ NV 2210 derivations are possible, while N> V > N zero der Frations are not, Examples of the forge type are (OH) prod —> prety > prt ‘stage protest (demonsrstors may pr but eid ean only prov) dsebuny > dicbunty > dsc sll ata discount gta» dds > dg make a digest ‘ond > cpu» conposmdy jon or becomejoined na compound” Convincing N-» V-» N ease, such as a ypotetcl triplet parton -> patirns > patterns do ot exit to my knowledge, Some eases that look Tie they might bo ofthat eype are probly more eoeecl taken 2s involving two distinct zero nominalzations from a verb, such ws “act and either “elcid object 27), “implement (27), o “agent? e) Ga. disshargeydichareey “set of dichargey “that which is ‘dssharging discharged spit, pity it shity hit it ralicey splices plese paddle, pad dale “implement ading for paddling? bran Brushy easy seruby scrub “aetof seus “one who scrubbing! serubs4 Put Kparty The reason for aking the verbs rather than one ofthe nouns a bases that various overt > N derivations expresting each of these categories mist be assumed anyway, eg shavings, spite (let obj), router, mixer Cpl ‘ment and agent), Also, oly on tat assumption wll leve-ordering corey predict thatthe verbs may undergo level | inection (spat, sweated), And that verbs which must be zero-derived ill not form zero nouns But ‘express these eatgories hy means of oer sie, ain 28)" (28) ively evety > riveting at, riveter (implement, agent) condition, > condition, > condoning, conditioner valet, water, watering, wale An esentil feature of our approach is that allation and compounding are interspersed. Here we diverge from Wiliams (1981), who orders all inflectional and derivationalaffation before compounding, We have seen thatthe interaction of compounding and indetion ip English morphology supports our proposal Additional support fori comes as we now tut ‘the relatonship of compounding and drivationalafiaton. We shall con- centate on deverba derivation in Englsh, with parcuar emphasis on So-alled synthetic compounds such a8 (29): (G9). pester, font-runner, watshaaker, churchgoes bulshting [all afxation precedes compounding, then theve words have to be treated iN ++ Neompounds comparable tothe type doghouse, The have, however, ‘numberof pecularites which have tadionally been taken to motivate & Sliferent analysis, namely that they ae “derivations from 2 verbal nexus" (Marchand 196, 15). Th is esentally the approach which we sll pareve here “The most obvious special feature of synthetic compounds is that their second member does aot have to ext asa word ini owa ight. The nouns ‘eter, goer do not occur on ther own atleast in the senses in which they fnter into the compounds typeeter,thatrgor ele. Ax Marchand (1989, 1 remarks “the lexial independence of the Second member a mater of secondary imporianse The points more obvious for synthetic buhwvrhis, Where the second member is routasly bound (reelgged, moon faced, Feirminded ee) As Blooms (193, 231) observe forms lke lng-taled of redBearded are ot aptly described as coniiaing the words fed, eordd (asin tated monkey, bearded lady) the ‘atualstarting-poatisrathera phase like lon aior re Beard, rom which {hey ile by the presence ofthe sified Thi ithe same thing gs ying ‘hat we we compounds ofthe type fon raed, red-earded regards of the xlslence of words lke zaled, bearded; wines forms like Bucyed, four {footed ub-nased. Anaer maser English yale type that of Hire ‘master, tonsa peer.” housekeeping, SEE Heer eee eee e eee ee ee rreerreree reer lc Maple at Pn 15 “The tational view, then, is thatthe special properties of ynttie com pounds are due to tice eing derived fom phrasal combinations. To my Enowledg, the only generative treatment so far which exploits thi ides i that of Reeper and Sigs! (978). They propose that synthetic atpureshas fr dried by “lsc tansformatons" whic operate on subeateorization fares, Trey idea i that in a compound of the form co x ving ed | 2 bas tobe the oa ster of V, hati, he nearestcomplement of Vin the verb phrase, I is actully not asesary to follow Roeper and Siegel tothe extent of etving synthetic tatprashas from syntactic representations of any sot. ‘Ther esental insight thatthe semantic relatlonship Between the members of such compound i sytematically determined by the argument truce ‘tthe verb Their principle eacompases te following generalizations, which 18 far a T can sz are core: (0), X in GO) can ever be the subject of V Gn" “population-prowing,*blood- table bere and ee indeod not encountered in synthetic compounds in ~r nd ing. Bt if the latter could ls be derive by (45), the wrong interpreta tions would be let in asi wee through the back door. ‘Secondly, Allen notes redundancy which s inherent in the Rosper-Szge solution, It requires two separate rules introducing the deverbl sues “er end ng: the above-mentioned leva transformation forthe compound, ‘lus a separate cule forthe simple eases Uke singer, singing Wt tre ta forme verbs undergo ony one or the eter rue" 68) a theatergoer be Mabchearer *eoer hear ‘Yetthey are related processes, an the grammar should obviously ditngush English foe a hypotbescal language in which entirely diferent endings fe involved in synthelic compounds and simple devived agent and action ‘Some would ike our account of English morphology to answer at Hast, the following suestions about synthetic compounds: (0) Why are hey Formed withthe same sufss that make the pro ductive (level 2) agent and action nouns from simple verbs? (@)_ Why are those the ony sufltes that ocuron bound second ‘nibs? That is, how come there are no nous i ant, avon, wl the ‘which ae lke "poor "gaming i hal they are restricted to compounds? {@)_ Why docs the FSP not hold for compounds i-th, son ee — et 6) and aT? (8) Why i the rule forming synthetic compounds disjunctive with rate (45)? (9). What isthe elation of “backformation” that links synthetic compound nouns and compound verbs lke (3, 42), e arcondtoner, ‘atscondtonng an acon? (6) Why can a speaker al once assign axed meaning in some cases ‘vo fted meaning) to syathetic compounds he has never beard. before hile oer types of compounds suchas N-+N allow an indefinite num- be of interpretations? ‘A modified version ofthe Roeper Sloe! analysis wll low us to answer ‘cae questions, Assume that all English compounds ae derived by insertion1s Poa Kay OF Z into categoria frame X: 4) Zhe ‘In principle, X, Y, Z are Feely chosen among the lexical categories. We are meresed her inthe ease where Z ~ Verb. Because of the FSP, ¥ mus then ‘be Noun, Adverb of Adjective, because verbs cannot be interpreted as Sst stpuments of verbs. Consider nov the possible Yalues of X. Suppose X ‘Noun (Agent or Action) or Adjective. Then an appropriately subeategorzed sfx Ger, ing etc) wl be neta ale it obligatory by rules sch a (50): 50) Taser er! i en. [Y Vena. ving such compounds a5 airconditoner,at-condtoning, our facing, ERrange-sunding, depending on the rule and the choice of ¥. Next, oppose X—V.Norule inverting an aflxin that frame exists ut the structure i permitted a1 auch eg ai-conditin), Finally, oppose X-— Adverb, Again {heres no insertable af, andthe structre ir ejeted because it isexccentce ‘onic between the head (V) and ue dominating category (Ad) ‘Weare now ready to anewer the questions raised above, (0) Rule (50) mast obvouly be combined by the standard notational convention of parentheses with the rale for simple agent noune (7 3D (St). nsert einen [V_Ivonanse ‘which ike (0) belongs an evel 2 or phonological reasons and enjoys the ‘sme productivity and semante predictability, The resin schema i (62). Insert erin e2¥-(Y) V—honanne ‘A similae generalization obtsins forthe thor sufxes that form synthetic ‘compounds ing and wd). The identity of safes and relationship of pro- ‘cscs js refed formally ina simplifcation ofthe grammar which simi- dates the redundancy noted by Allen (@). The special status of ~r and -ng that they are introduced ‘iret into compounds by rule (50. Therefore they can be rested 10 particular compounds. The ether agont and action sles am, ation ete) te added ony to verbs and they cannot therefore be esrited to particule ‘compounds into which the resulting n0Un5 might then ener. (@). The reson the FSP doesnot hold for expound ks (46), (47) ‘shat they are not verb compounds, i they ae nt formed by taking Z— ‘Vin(él. Rather, they are formed by 2 — N from two\nows. The second ‘noun happens to be derived from a verb a level 1 but tht information 5 Tost by Bracketing Erasure at level 2 where the compound is foemed, 10 the FSP has no chance to apy. @) The uniqueness of the corect derivation for synthetic com pounds is ensured by the principle (be jsied ina phonological connection below, seep. 57) which selects the shortest derations among equally simple alternatives. By tis principle, we do not derive watchmaer into ‘aps (Vly with blputory insertion ofr giving mater, followed by [NN esa Maly ad Phasaey » rng warchmaler) base we ca have ety in on tep (NV, with ‘bigtry rian ofr) {@). The proses which derive compound verb ke (9,42), 68 a= condom she sane which form al compounds nding sytete m= feos Tz compound vt ae obtained when Xia(¥ Ze es (6) Compounds dened rom verb bs where ¥ of =¥) ae ae sine hed ntepeation by th SP ante safe sem Whee SS has two subenteyonzation aes the FSP can corepondingy aign {no Inerpctannsf te compound, a Wit) optonal transl verb For camps nae nfs might be endertood a oc on he Enuloy of sree ots a objeto he analogy of Boker, because acon be te both tutely ta sss. Compound ered om oun re aged vaca itrpcaion ch can be fae ony rage trata ond by soveton Fr staple, the aipursha ofthe form Ny i owning 1977) bem to be inerrced by some open expreion ofthe Fein nde) N, whch Ce Nev Ti nse fe the css ke (7) ‘thre Nes uel dexvd fom Yr. id he prodete nates Agen and Aton compound ia and dng twat tome morte: We tae ey Gace te 10 wae ‘he sre a eve in (Vu whore Nis (+ Agent 6 ot Br) Gres Acton|(ee pe, Tt we gone thi sulle ab we id the sive 53). tse § ine. (8) V_Io ane visa cima i Sop hous pay a sameway Teepe then sh hte cond fe on which allo at ang sabre insering heen. [XV by be. er-dved pounded level 2 by the[N Nj is of scheme (8) ad escape the FSP ike thecompoands (46,47 a8 explained above. Siri, suse remain difeal tocanse they denote not ony the time bat abo the act ofthe suns ring oF sting. ‘By allowing the sui ¢ tobe inserted also inthe environment [V X_l ‘we derive the minor types shown in (5): (D4 Kiljoy, euthront, pickpocket (C67) showdown, Buckoat, shakeup, palltuck, puton (ston); Showof pickup, getway, runaway, standby, standout (een) “They should algo be sujet tothe FSP; this prediction seems comet (ee {B)andexpsins the limitation othe sesond members Nounsand Adverbs? "The remaining major type of synthetic tapurushas consists of passive cies wih the sues een (G8) sate-owned, mothesten, God-given, selF-taught, home- table =» anablity would violate level ordering and the derivation ale —> by» *unabity would violate the resrction that un i preBned 10 ‘djectves, not to nots There ae of cours various case of exceptions to this generalization, such as sngranmaicaiy, wninteligence, srt, Jodboletity, where level | afics ave seemingly Been added “outside” i a alae exe Moray ant Pokey » Jove 2 pretation and compounding. Liber and Willams adduce them 38 ‘idence that level semantics iv independent of morphology. However, Sic have two wayn of dealing with suc exceptions im ou Sramework. TRC [st would be to make ue of limited recursion from evel 2 back to level ngranmatical, drotette would be formed a evel 2 and thea fed back Jato level I whee the sixes ean then be added to them. The second hy pothesis which shall adopt her, sto allow exceptional defera of Brace {ng Erasure for particular words. On this hypothesis, grammariality and ‘etry ar formed Sto evel | but disynerateally retain ther internal ‘racking lgrammaticall, iy (eet, iy 80 that wordformation processes ca apply tothe inner coasttuent at level 2 “There are some factual arguments which favor the second of the two slteatives, The level recursion hypothesis predicts that other sufixes ould be added at level 1 to words like ungrammaria, drole, which fs not the cae: Hngranaticale,*hydroelectrilan, *hydvoleiiy. Toe bracket retention iypothesisprodits that eter sufxes and compounding prosestes could apply at feel 2 to the internal stems of grammatical, letrity, which 1s the ease: exrgrammatialiy, semgrammaical, thermockecmey. Moreover, basket retention is the more Testective pothesis, Level recursion could be extended to arbitrary violations of Tevetordering, which amounts to a substantial weakening of the theory. ‘racket retention could not be invoked in those eases where the inner afl tion provess feeds of Bleeds the outer one. For example, suppose that— ontary to fact—tevel 1 sles suchas - (det ~ at could be {nde outside of sf belonging to level 2, such a im (f,nonaion, paricalaviom) ot ite (6, Kemedye,pre-Rephelite, The resting level- ‘rdsting violations such at *natlonalimiy, “pre-Raphatify cbviowsly ould not be derived by defring Bracketing Erasure, for the required Sources, the level | words “national, *pre-Raphelify do not ex. How. then ie instead were to allow recursion from level 2 level 1, there would bb bo principled way to block these leve-ordering violations from the potential voesbalary of English. THis however resonable to separate those aspects of excl meaning which ae not steveturaly determined, ‘Suppose, as suggested above, tht ndocentrie Nj +N compound nouns ae interpreted within the grammar merely by a generalized open expression “(kindof N, which X's N,", Downing (197 points out thatthe denotatum has tobe “namenorthy”, whichis presumably special ese ofthe general fact that proper and common nouns denote recognized individuals andree- ‘gna inde of things. On the “kind of interpretation (.e. in « common, ‘oun, X isl i generically interpreted, og. particular shorebied i sll ‘horebird when it froporrly Mes inland or even when it spends its day inBEES ee eee eee Ree EE EE ee eee Pee Eee » lary the Las Vegas 20. In acta use they are subject to several kinds of specilztion. Fist, as Downing dociments, the choice of X is fied in new compounds by the speaker's and hearers imagination and knowledge of the worl. Secondly, ‘any compounds are istittionalied with particular sense of X: a shore= ‘ied i bird that Feguents the shor, a songbird is bed that has 2 song, 8 ‘samcbind isa bird that connttutes pune. Take this instittionalzation to ‘eprevent “encyclopedic” knowedgy of the form "there exists a recognize [ind of N, which XN," rather han a fact abs the Uingusticsystem tl. ‘Ag Muchand (1969, 56) remaths: “In itll there is nothing in the word voter rat to ease an anafyis such as *water-prodocing rat. With this ‘meaning the word would be quite posible as the name ofa 1, for ine ance. But asa serous word it docs not exist inthe norm of the language because ther so denottum for iin the extralingul word. The norm of cour language sees onty certain patterns from the system of possible re- Slaations according tothe Sentata ofthe extalingual weld, To a cetain, ‘extent, therefore, the exralingual danotatum also must te known if our Analysis to be core.” Thirdly, many N,N; compounds are mere des- Sgnatlone of kinds of N which ned not have anything to do with Ny, for example conbird,lrebird, lard, kingbird,seretary-bird. So assume ‘that the above rule of ftepretation is optional I its not applied to a Compound, the compound wil share the features of is bead but the left ‘Sonsitutes will make no contribution to its meaning. By convention, Ni Ne is then interpreted as meaning “a kind of Ni" ‘Certain ype of compounds, ssh as thos ofthe Form Adtive + Noun (whites, Blackbird, shortcake, greenhouse, longhouse, tightrope) ae inaly resected to sich vse a5 conseationl designation of kinds of Ns ‘Wecan now accouat fr this acti the gramma by #ot allowing those eyes cof compounds to be signed any compositional itepeetation. So the [grammar merely cays that ohh aa tatpernha compound is akind of fish (the minima thing which by endocentrcty it must sy) but does not ansign it the meaning “white fk” Adjective + Noun compounds will ‘ecoringly have nose excep as terminology. From ths point of view is landersandable that dervationl patterns which have fog lst thir com- poston interpretation inthe morphological system may persist as no- ‘enclture and even continue to sappy new Cems. Finally, ike aay other Kind of expression, compounds may become fully iiomati. These may be tweted by listing or in whatever way is appropriate for idioms in perl So what thie scheme of interpretation sas is that one may interpret ‘or coin a new compound eich a gerbind i two possible meanings, fis 10 ‘enote& particular bird or kind of bid that i coaneted to tigers in some ‘way to be pragmatically determined, and secondly for any kindof bid that Lee Mpeg and Phooey 3 rnceds designation, for example a newly discovered species. One may not Fnterpeet of coin it for example denote tiger or # ind of tae. The advantage is that we delimit the grammatically governed aspect of words ‘meaning from what is arbiteay and conventional. To be sre, our grammar ‘nds up witha lexicon that does uot say how catbizds ier from cowbids for that tigrbinds do not ex, ut in view of the arguments that sh i ormation belong in the encyclopedia anyway that may be more an ase than a laity ‘The second advantage thatthe pots of tis section taken together now Sugeest that lial tlationship reduces to derivational relationship, fendeing the ingenious ut roundabout definition of ‘Willams (1981) ‘tise There i of course no single test that determines fm any given case ‘whether two lexical ens are derivatonally related or not the dein wl Jn cach cme rest om the overall grammatical sytem of the language. We shall therefore want the structure of the morphology 10 be such that in ‘onjoncion withthe evasion measure i pedis he intuitive judgments fof lexical relatedness that are shared by Mueat speakers of English. For ‘example, we would want the theory tobe abl orale out a excl elation- ‘hip between such pi of words afer and patra, car and ear, 2a ‘emen and seminary a these cars presumably on grounds of phono, ‘morphol, and semantic complexity respectively. In order 10 ace ‘omplsh thse shall have to supply format forthe cules which express the relationships between words in each ofthese respects. For phonolon) ‘we may asm the theory of generative phoaalogy provides such «format. For mogplology and rsmantescurtent work is providing bepinning and ‘we have made tome conerete proposals of our own to this end above 2. Abstracts, Cyeisity and Lexical Phonology ‘This organization of the lexion, in conjunction with the proposls de ‘eloped eae, has further consequences onthe phonological sie. Item Dodie th claim that allan only lexical categories are cylie domains, and alland only lexical rules ae cylic. The prediction that thor should be no ‘li rule pplication above the word lve, Up to recently this conce- ‘Guence would by isl have sie to wreck the thor in Wow ofthe fe, that sentence stress was one of the castivon arguments for cyclic. How ‘er, Richa (1964, 1972) and Liberman and Prince (1977) have pointed cut {hat metrical theory eliminates the need for eyeicasgnment of sentence ‘tre In that framework the Nuclear Stress Rule, which assigns prom ence to the right breach of a phrasal constituent, can apply in any onde or Smltaneousy o all constituents in the seateae. AS far as T know there areno rues which hae co apply eylaly from the innermost phrasal con ‘Stents ovt and the theory of lxial phonology predicts that.2 Pea ary “There are howover cases where eles of word phonology sem to apply fonee at the lowest level of phrase syntax. T have argued (Kiprsky 1979) that the English Rhythm Rule applies both telow and above the word level, in eases lke expdet~ expecdton and abuirdct ~ dbtrdct. dt to- spectively. These applications take pos at dflerent stages inthe dev ‘on boas the dstesing of metricly weak ita lables mst crcial- Iy intervene between thet. A somevthat similar situation has been found in the tonology of Ewe by Claments (1977, 119). In nether case is there ev Aenoe of eyelie eration within the sytaxitsl: each al applies once at the ‘asl lore and Joes ot need 0 reapply ell on sascessivel higher Fyntactie constituents. "The mest saaightformard assumption is that these rues belong to both tna tect and poslexial system. weve, iis sorely significant that hey oncom what maybe called phrase phonology rather than full edad sen- {nce phonology, Both rus are subject to 8 constraint which blocks their pplication on Verb fllowed by objet Noun Phrases (nin ddr “Pllinae ender; for Ewe ef. Clements p. 122 ff). Both rules have lexical acptins (ee. prfdund trith-) "profound crit; f, again Clements». Taf, So one contusion that could be draws from tis i thatthe lowest level of phrase stracure can in some way be fed back into the lexicon, Quite Sart om phonologel considerations this would be suggssied by the ‘la that phrasal combination at this evel are subject to selections restc~ tions and Table to get aed a idioms and formulas. We aleady noted above that they ean be inputs {0 woré-fomation russ, e.g, American history teacher, 10 stonewall haxde-of pai. A speci example of leialization ip that dbrct has for many people acquited inherently the specialized Aechnial sense it bers in the hase abet dr, denoting a specie tps ot Sehool of art These speakers make a diference between the sentences (Q)e This arti Abtrit (aot representational) 1. This art i abtret (oot conest) Inthe ight of such fasta, the nature of phrase phonology and is relation {o word phonology and sentence phonology deserve dowpe investigation. As Tora tower bound on elt, hare is some evidence that non exica catagories are not eye domains. A gute competing ease has teen made by ST Haris (MS) for Spanish on the bass of a sudy of sess and syllabi ‘teocture. One of bis example i summarized in Q), ° desden + ey ‘isis’ (noun pra) esdet + et "youdidain’ | @.sg. ub) [iesde 8 hy (Kes ahve she SA 1 sybian [2] Cetiameall Romineods (osppiiey pee gee eae : y y The point concers the eppliation of a eyeic rule which depaatalizes ‘ho min the coda of a syllable, The cootast between the noun pura des (len + esand te subjunctive verb fom dead -}e arises because the unde {ping pelted fis sylabe-nl in the noun stem /desdeO{ bat saber Fnial inthe verb stent (desde +, These respective stems constitute the fst eyelic domains and the nasal depalatalization finds its correct syllable fmvitonment there, not atthe word lve. Hazris demonstrates thatthe {jolie domains required for both sess and eylabieation in Spanish ae ‘hal the lect eateries. Thee are indications that hii bo true for ‘Geiman (Kipasky 1966, 75-79, Freneb Bouchard MS), Tiv (Pulleyblsnk MMS) and Hebrew (Rapparport MS.)-A general condition to this ffs was proposed already by Brame (1974) on the bss of Arabic and Engin data Fle noted the Palestinian Arabic contrast fhimna ‘ve undersood” vs “fla she understood withthe same morphemes except that na is a0 rng inthe fist form and clit tho second As he pointed ont, the data Ean te explained by assuming thatthe rule syneopating unstressed yorl is ‘ordered eyeliclly alter the Tule assigning peult stress on the assumption that the bare verb r00t i not acyclic domain: ° [itm male him + 6 + maby 1 Sires (abi + al (chim | Syncope him na ae 1 Sires a Winim on Syncope = Brame proposed tha) ele subtrings must ocear as independent words, tnd Q)all and ony rules mentioning brackets are cyclic, The ist condition taken erally is etal to stong because stems, which must be lxally fateporzed as N, V or A and which do consiate eylie domains are nat necesariy capable of occurring as independent words in intial lane | guage, wherethey may requ an obligatory ate ending. Moreover, many Tanguages wordsormation mst 1 all appearance be taken ac to 004 ‘win themsches ate not aeesarly member of any lexical eatery. If this is tra, then ie is necessary to stipulate tht Jee! rues apply ony in domains dalimited as leniel eatgores [ya “Another previously problematic connection between morphology and phonology i that morphological rules may be sensitive tothe output of honolopeal roles (Siege! 1934, Aronoft 1976. This i naturally accom Imodated in this Geoty because phonological rules operate in tandem with ‘morphology in the lescon. For example, the nounforming sui -al is _Po Pal Kt only added to verbs which are strened on the lst spb, arrival, reviral,aequltal vs. *dcpésits,*evdveral. Therefore the derivation pro: ‘estat ads-alto verbs as to apply after the eeic ale hat signs sess to verbs. I the grammar is organized a in (2) of 1, nay this ordering is predicted. The claim i that only esi roles may be relevant to more plology, more specically that only eyelic les of evel may be flevant {to morphology of level m, where n =m Perhaps the most sgnicant consequences thatthe theory has for pho- ology have to do with the nature of lexical representations. Hert prom iss to aslve the longstanding isics that have boon ctcssed in terms of constants in abstracts, the Stet Cyele Condition, morpheme structure ul, and other notions. We shall summarize the problems in Btoical perspective and then show how they may be appoached from the perspee- five of lexical phonology “The theory Worked out in Chomsky and Halle (1968) cums that under: Iyingrepesetations are chosen so as to give the simplest totl grammar, here the grammar includes both lxico and rule. Theft so guarantes ‘tha excl representations will be at leat as abstract a the las pho- nomic level. But they will be tore abstract whenever, and to whatever ‘extent, the simplicity ofthe system requires it. The simplicity ofthe system ray require mere abstract representation for several kinds of reatons. The ‘most important kindof reaoa for setting up an abstract representation of & ‘mogpheme is sjstematialy governed variation in ite phonologisl shape Underlying representations are setup in such a way as to permiceuch ep ‘lar variation fo be characterized by the simplest and therefore most gene raf rules pssible. Let us istrate the pont with an example rom English ‘Phonology. The] in dsonan is oe derived by a depeninaton rule from Js-F5), on the steeagh of is component morphemes, which cewhere Appear as ise direpucable, dcourteous) and son + et (Cf sont, con sonant). Moreover, this rpresotation direty explains why the word docs ‘not undergo the Yozing rule to which single is normally subject in that envionmemt (f, resonant). Simlaly, the [3] a rite i derived from fk by a "Velar Softening” rule, which apie throughout the derivations Romance") vocabulary of Engi, and accounts here forthe repulse re= lationship between erie, encom and erlc, erica This does not imply tat che speaker or bearer ned in any way mentally “derive” thewords he sys or hears by means of such rales us Velar Softening. What it doer ‘ean is that the alterations they govern belong tothe regula phosologieal ptiern of English, while for example a hypothetical &~ «alternation inthe reverse conlet, such at ‘crise ~ vor, would be ireula, The ‘aim madeis that someone who knows English mpc knows tat patter, and will unde appropiate circumstances recognize the diference between ras, ae Lek Morgoagat Phneay 8 regular and ireglar alterations, though he may not beable, even after tefleton, to verbalize the rules that under ‘The sytem that is eet up inorder to account for phonological alteraa- tions may then in tur permit farther abstraction in the undedying rere- entations of nonalterating forms as well Oflen it may be postvely reduired by the evaluation measure tha slets the simplest gremmar “bstractoess” in tis sense cannot be ciezed on any legitimate a rie! grounds; ia the absence of further evidence it would indeed te demanded by sound scenic method. There i, however, evidence, which shows that at proiely this point the theory induces characteristic type of wong ‘analy. Identifying and cocecting the source of errr inthe theory was the Toca of the socalled abstracinss controversy resulting in various versions ‘of "conerte” phonology and more recently cjelic phonology. Perhaps the mos familiar English example of arule which cause culties ofthe sort which ace at su here i Trsllaic Shortening (or Teac [axing Ths ue shortens vows f followed by a eat two more vowel, ‘of which the fit is unstrested (G)_V->[—Hongl) CVC, where Vi not metal trong eapplies, for example, to shorten the fong vowels inte intl syllables of () opacity, decaratve, taboate (opaque, delare table) The rule ha tobe assigned to evel on the evi enor of words bike (©) mghily, bravery, weariness (Clearly all morpheopealy simple words such a5 (@) ivory, nisin, stvedore, Goolagong, Avell, Oedipus, Oberon snstsomiow be exempted from undergoing i Standard generaive pho- nology fores their phonological repretenations to be adjusted, If posible, fnsuoha Way that the sevctural description of the rule is not met. For ‘nory iis possible to postlte a final y/, whieh will Become # by a in- dependently needed rule of English phonology. For nirtingale, Chomsky ‘nd Halle (1968, 234 rather less persuasively suggest underlying /ixtVag), With >» ay By rules they claim are required on other grounds. The reason this i utimately unilluminating is that in the balk of the eases— Steredore and 0 forththe flute of Tveylabic Shortening cannot be cplained amay by changing the underlying form anyway. ‘Along with te problem that many ordinary words have no regular de. ‘vation in the grammar, there isthe complementary problem that words te (Galt, sycamore, camera, pian, enemy, Amazon, Pamela, calendar have vo posible derivations, while only one is eer nesded, They could be Aetved at fee value from an underlying representation with a short vowel Jn the frst syllable, But they could aso be asjned a Zong vowel inthe% Put Katy fist syllable and taken fora “free side on the Trisylabie Shortening rl Tet as now look at some possible solitons to these problems. [A priniivestempt a solving this problem was the socalled Alteration Condition, Init tong version, it went a5 fellows (Kiparsky 1968-1973, re. D: (©) Obligatory neutralization rules cannot apply to all occurrences of smorpheme. “The general flo isto mith “abtractnss” of underying representations {o eases motivated by phoaologieal alternations, With repard to our par tieatar example, the Alfemation Condition does to thing It reo}es the lndterminay inthe undslying rpresentation of (8) by fixing the vowels fof est sales a shor, Aad prohibits the possibilty of dealing with {by sock phionologal dvies a8 postulating (x in nightingale and non Syllabi p/n ory. Nevertbeles, the Alteration Condition 2s stated in (©) is unquestionably inadequate, Without attempting to do justice tothe complex dseision that broke out around i Te us summarize what are perhaps ite moet damaging Mw. Tr The Allration Constion is aot interpretable asa formal condition ‘on grammars In order to check whether it sated ina given grammar, it ‘would be osestary t0 inspect every dvivaton ofthat grammar. The only “ove thatean be made out oft as asttepyof language acquisition which, ‘Ss that a earner analyzes form “at face value” untess he has encountered ‘ariante of it which jst 2 more remote underlying representation. "The Aeration Condition leaves the theory with an inherent, redun> ane, For example, sys that words like Amazon, calendar enemy, Emily fannot be derived fom underlying representations wth fong vowels in the first lable. That fact oupht fo be of one pice wit theft that words Tike Oberon, nightngle, ivory, Averell, overture do not undergo Triylabic Shortening. Yet we shall il eter have to mark the iter as exceptions to the rule, or eee restate ts environment so that it applies only acorss mor. perme boundaries Tht is redundant became the theory already tells ws, by the Alteration Condition, that these words could not be subject 10 ‘Tiyllabie Shortning. The problem is thatthe contain: on undeiying ‘representations leads fo predictable restiton onthe application of rules ‘nich canna, in thi formation, be expressed systematically inthe gram- zmar. Apart ftom the redundancy with the Alteration Condition, neither “ternative is sisfactory in ts own ght. To mask the words 36 exceptions misses the genecalzation tht all non-derved words subject the Ti Tabi shortening rule fall to undergo and it furthermore wrongly claims that sich words age iepuarities, and should tend to get “regularized” by shortening thir fit syllable, whieh doesnot appeae to happen. On the other hand, testing the rue to apply ony across morpeme boundaries _ {ese Marly ad Phases a sequresin the standard theory of phonology complex dxjunstion of pluses fn the eavironment of the rule, vz 0) G4), CG VOD. Ca Condition: a oF 8 of ¢ or @ to allow for shortening in cases such as t+ meror (0), gran + war (0), ‘eval 19 (©), min + ous D. "The Alteration Condition is too weak in that it imposes no restric. tions whatever on posible pateas of alteration. It appears that a eeain {ype of ule of which Triylabie Shortening isan example, coald not apply rmorpleme-nteraally a in (7, 8 even if alternations di exist, To see What ist stake, consider the condition on Trillabie Shortening tht the ol- owing vowel must be unstressed. It correctly lock the rule fom shorten- ing fit vowel in such words as (I) (1) 8 qustiton axation ayedton ‘anie ‘What does not exist, and arguably could not exis, i 9 patter illustrated bythe hypothe cass i (12), where we would have Lo allow morpheme internal aplleation of Trsylabie Shortening, on the strength of the long ‘ows tha ows up in the dedvaive, where Trsgllabic Shortening Is Hed by stress, (12) Gxigia: —*oeginal Gein) shonym: *sfndnymous (SFn6nymo0s) “The Alenation Condition i incapable of predicting that such cases do not '4-On the other hand, the Akernation Condition is too song in that it cncludes analyses which ae woll modvated on internal grounds. That it ‘not be mantaned wihowt wnaceptable los of generalization, Is thi failing which attacted the most discussion ia the contoveesy over the ‘Alrnation Condition. For example, while no-one wll mis the putative fx) in mipagal, a Sn yf ie rther well motivated by the system of English word stress. Words ding inary and ary at wells words Uke golxy,sudsty, sieraticle iy behave as ithe final vowel was realy a consonant, with rspet to sever Of the sess rues (Chomsky and Halle 1968, 130, Liberman ang Prince SSS TTT TT a a al RRR MRa al Ki 1917, Hayes 1981, Ch. 5, The same finaly acount forthe fale of Teele Inbie Shorcning’in derived words ike mceny, platy gene). secrecy, where the rule would otherwise be expeted to apply (Ruch 198, 1958) ‘Also the only exception tothe generalization noted above that al is aed ater stressed syllables i Burial; whic can be resolved by taking the final ‘ya nonyllabic. But the required rule (yi) wordsinally isan obigatory neutralization rule sac 4s a phoneme of English, and so ‘cannot apply in the proposed estes consistent withthe Alteration Com. ion. ‘An alteratve approach is to deny the phonological character of rules sachs Trisylabe Shortening. I is commonly sui that cules ofthis pe se to be considered as “morphological” or “morphologized”, This clan may aetally mean a numberof things, ce there are several poaible ways ‘of teting morphology conditioned ele nthe phonology. But on any ofthe possible consti the properts of the rule seem to be obscured rather than explained by the propos. Lat us consider tee versions in ur, ‘The fist vesion would be to simply add to the phonological cavcon. ‘ment of (8) a morphological environment consisting ofthe it of formatves ‘fore which the shortening process can tae ace, 18) Sen ~ nous vate semen ~ atin ar ne ~ eer ition ~ ndvonal » penal ~ pinay “ion ‘complre ~ comparison Aitje + compete ~ comple “op We silfy con fer ~ ferent ‘This isis, however, redundant. Theses which have o be included in it are simpy all -boundary” suffixes which ean cause the phonologlal ‘conditions of Triplabie Shortening tobe me. The only safes of that lass ‘hat may be omited are those which happen never fo ceca he appropiate honologial cumstances, for example deverbal a, whichis only added {o endstresied words (aril, ons, betraya). The correct genet ‘on is hat all and only suis ofthe + -bounary class may trigger Try Inbic Shortening inthe environment of (4, ‘A second morphologized version of (4) would omit the phonological ida 33), all i the Best eel ‘A further ‘corollary is that rules which assign metrical structure wil be Blocked on the first eee to the extent thatthe inpot i metrically struc ‘red alteady ia che enon. In Spanish, ese ik ur (vo lables, ofemos (Cour can be simply marked $ inthe lexicon (as suggested by Haris) and the syllabieation rules will not apy. “These assumption aford a series of major simplifeationsin Hayes theory ‘ot English word stress, Hayes introdnces baie rules of sess assignment (oot formation: the Engh Stress Rae (ESR) and Strong Retraction Rule* Pea Key (SRR). The ESR applies at the right edge ofthe word, assigning maximally ‘inary feet labeled $ W, where W may not be a heay syllable (aot count ing “extametrica” materia), The SRR applies iteratively fom right oe, also assigning maximally Biaty fet labeled $ W, but without restitions ‘to spllble weight AS proposed by Selkirk (1980), stress ls not a feature butte property of teng the strong oe only sllable foot, Ths, we have such derivations as (39, where parentheses denote exrametialiy 64 (@ hamameltntber = _ r sww Sw sws Ww (ESR) hamareanteotn) —+(SRR) amano, sw Aw (©) Tonderoga—= Teondertgn)—~ Tondo ‘The resin fet ato then joined into a word ree, whose right branches are labeled strong unless they consist of single sabes ‘Two special sets of assumptions are required im order for the analysis to workout right, The fist shat the stress rales, though necessarily yc, do not observe the principle of sc eyeiity. Tiss forced by the need £0 spply stress rules onthe fist ele, Le. in non-deved environments. The ‘Stcond is thatthe grammar may specify in what manne ses ules ppl speifelly whether they may overdo existing foot structure inthe string, ‘or wheter they only assign meteal structure to syllables which are not yet organized into fet. The two English rules ifr on this sore, The FESR belongs tothe fist eye in that It applic repress of whether the string already provided with a metrical organization or ot, The SRR, on the other han, stretly respects existing seta tact The const can be seen in the following examples. Patentheses show exttameiricality, Pa x aw fe iow (9) mt sta a-e) y : ry use (ESR) (othe) sw SW fe Sungrd + SR) stnidard) tind fe (ESR, word ee) ena Map a Pony ® sw ws oy ea ube okie ow lg sw snd +e (ost) * san ~ edie oS ca ows ¥ Swi sw CESK, wordt sandngie 4 Alon CSR) * sundae +t) ‘Example (35 shows how the ESR wipes out the metrical structure assigned by itt on a previous eee. Example (8b) shows how the SRR respects the metrical sructire asianed by the ESR prevowly on the same yee the SRR were allowed to wipe out the metrical structure assaned by the ESR, we would decve the wrong form *ilistrite, and moze geerally all ‘viencs thatthe ESR oven existed inthe grammar would disappear. Ex- fmple (35) shows how n addition, the SRR respects the metrical structure signed by the ESR and by itself ona previos eye. This is why we do not Aerve the wrong final form as sndiated thee (standrdzdtin). inthe present theory, nothing at all need be stipulated fa the grammar of nglith about the mode of application of any ofthe stress rules. It fellows ‘outright from the Elewbere Condition, In the ist pcs, the ESR and the SRR thst be disjunctvely ordered, Secondly, beng clic (ie. lexcl) rules they ae blocked from appviag in a strectare-changng fonetion in ‘pon-derved environment. This oily defines the corre conditions fr their ‘operation and interaction. The detail are a follows "The SRR is applicable to any string of syllables and the ESR is applic ‘able to sight edges of constituents, Leto domains bounded by]. Thus the Stractural description of the ESR propery contains the sruetural deseripe tion of the SRR. By vie of the Elsewhere Condion, the rales must ‘ppl disjunctiney, with ESR, asthe special ae taking precedence over the SRR where they coflct. This explains ther interaction ia SD). On the rightmost domaia, ESR assigns a unary foot, taking precedence over the 'SRR which would assign a binary foot over the lst two splibles to give enul es,Proeeding lefts, we come tothe seng ir. The ESR is ‘ot appllenbe since the sting is not at the right edge of « constituent, find 0 the SRR puts a binary foot on i, stressing the inital sable ‘The wwwal word tree rule labels the Tet foot strong to complete the erosion. "At forthe applisation of the ESR and the SRR to strings with preeist-0 Pa Key Jing foot strvture, the theory requires that both rules should be subject to the peinciles thal govern the appliation of exci rules. Cyclic roles are sujet to “act eget", which by oar easier argument again reducer {o the Ekewhere Condition, Is effet i to block esi rues from applying Jn a structare-changing funtion in nondervedenvizonments. We shall om take up the varios possible caet for Both rules in turn and show that everthing works as predicted, ‘There aro thre hinds of eases to condor: 1. Nonstracurechanging applications AS predicted, both rules apply see eg. Gb). 2, Structure changing applications in non derived enveonments. Both als should be booked ‘AS far asthe SRR is concerned, i just this blocking which ss behind it flue to erase mtecal structure assigned (in whatever way) on caer ‘jes. Inthe derivation of standin c0 380), the pplication of the FESR io avon on the ted eyele yields the structure 30) FR OF owl Aw (66) Usitndartieh tne By virtue of our assumption that the output of every cyl i entered ino the BE A ow lexicon, the sng tote lt of stor, fandardey, consittes a eve) fexcal entry which fs fly organind Tato fot” The Ehewhere Con dition prohibits the SRR from overriding the foot structure ofthis lexical ey. 'Now we are commited to the pedistion thatthe ESR too behaves in cxactly the same way. Where docs the ESR encouoter preexiting foot Structure? The case occurs in words Uke Atta, Kenck,, Musso, Which fave 4 sree on a syllable whee it cant be astgned by the Standard stress rules. They are very reasonably treated as being repe- fented prior to the application of the srs rales with « foot on that slate DF r SD) oR x Aw Missal ‘What prevents the ESR from aplyng to such nouns minting the stress ‘on the second slab 10 give the iocoret forms GA)? Be FF Lee Mg mt 5 AY TAN aX la’ Sit eyty ie he Bhewhere Condition So we must assume tht sr turchngingappleaont fn aon-drved covroument are prohibited for the ESR ar wel efor the SRR. eis thas aot re, even in ess of Taye ‘own analy that the ESR is simply eto erase whatever metic sto {tris inis way Like the SRR tsa igorous}y tothe condos that oven the appiation of esi xa) rale. 3 Suuctuestangng aplication in derived eovironment. Both rls should apply Tis conse the ESR In explsite parental (a the ESR revit xstng meta sacar became tea app ina dened owen: Seat That th sing under consideration inte scond eye (marked Citi G9) i ota sbating ofa eal etry which beara contaictory foot sree & 6) apr “sr ‘The Eewhere Conon is inppisabe, snd the ESR fret form oot the ght ee of the word ting pete. Azsin we expet to nd corresponding cases with the SRR. The reason they were not bviows Before that i st ete, the intron of 4 word trl tave been analy orgrized in fall at any cyte butte fst So ‘Shsrvatonlly the gencralation that the SRR never rorgnins existing foot arvcture spproninates the tu iy clowly Butt only by seient that scm s teary te. Kaas dowa in the rail cue whee moe ‘teri than cn be somodstd bythe ESR is intros on a ven le "There the SRR completes the foot signment on the ex pla tothe it cod ear combo t with ny rater contained i exiting ft re ture, Consider fay, sl, derive from fle, wld. Coming oat a the fist eye we have (4) t & 40) shits ole itty ‘Tho ESR now assigns a fot to fp. If the SRR were now to form & foot solely on that part of the sting which does not azeady belong toa foot, namely the =i hich precedes -f, the final outcome would be the wrong forms "ju, “ald. Rather, the SRR mst form a Toot on fl, lid,2 Poa Kp vontuilly yielding ff, soi. This shows thatthe SRR too is free to ‘pply to material aleady contained witia fet, provided it docs so in & ‘erivedcavionment Tn sum, there ir no excntil difreace in functioning between the ESR. aad the SRR. Being eycic, both are rested by srt ey, and being in an inclusion relationship they apply dxunctively. Both propertion are _peranteedby the Elsenhere Condition and nothing nced be stipulated about their mode of appeation in the grammar of English nadaition, two sets of cases which eecape even the flaemcsied net of Hayes’ analysis are now ‘uted in along with th rest: the file of the ESR to apply in noa- derived eavironmens, st im tila (ce 37) and the applicabiliy of the SRR in derived environments, as in soldfy (ee 4. ‘As & rout, firther substantial simpliication of the stress rus bee comes possible, We are feet cllapie the ESR and SRR lato a single rule in which the ESR appeirs as 2 mere condition on the consitven-fna (4) Assign maximally binary $W fet from right ole, where W may not branch in ev. Finally under oor theory the decision to enter idasyncratic metrical, ssractre inthe lxoon assumes rather more signicane than before. At Teast the following empirical consequences are a sake. Fist it reds, surely correc, that arbitrary exceptions to ses rules add atively ess Complexity to the grammar in base sems than in derived stems, Farthet- rors, i follows that such lexical exceptions shouldbe typical of languages ‘with eli sress assignment, uch a English. Psteycic tres rales, sch as the inal tres ole of Finnish, cannot be contravened simply by inherent lexi specication of ses exceptions to them are costlier in thatthe lex- cal ems in addition mutt be marked at not undergoing the sires rules, Tr languages whero word sires is asoned posteylically it should tare fore be chaacterselly more regule. And this is obvious the cus. ‘his sue arose in an interesting way in the orginal diseussion of the SCC by Mascaré (1979, Matar was forced by his version of yeie pho nology to set up phonemic stress in Catalan stems. Yeti seen that there {an unmarked stress potter mich asin Spanish. This can now be done jstce to by aisigning t by a cyclic stress rue, with underlying stress 1o- fled to exceptional antepentt or fina tres (ee jt “Sei, paps “father Support fr this alteratve i provided by a sci pment rae that depends upon ste. In the environment V_I there is devising of 8 (and sporadically of ther top alo), eal ‘vi (cabs deviih), ‘The devocng ale must apply even if the vowel subsequently loss its tress, toa following suis, eg. plore. Hence its cyclic. But the envionment wa Mel sis always contained within the fiat eye. Now if sess is — Se eee es Marly a Ploy 8 tunded}ying, the envzonmeat of devicing wil be non-derved andthe rule ‘should not be able to apply (even on ur theory, sine ii eatare-changig) Bat ifsress applies onthe ist cyl, then ican fed devocing onthe fest ‘le and there is no dificulty wit these derivations. ‘A furber important point that secnwr application erates derived en ‘vonmeats in Mascar6's theory but ot ia ous. MascarS indeed found an Interesting casein bis Catalan material which seemed to work erally a bis version predicts, but it yields ently to the reanalysis just indicated, ‘Moving now to the problems involving lescal representations, we c2 resolve them in essentially the same way. We shall simply take the mre ‘Pheme strcture rues” of a language to be so many rules of its Jexieal ‘Phonology. Then lexical rue, rather than bein simply inapplicable onthe frst cyl, wil aply on the st eel as redundancy rules fil in lexically _unspeciod feature specications, in addition, of course, t applying as be- Tore in derived environments in a featurechanging function, For example, the distinction between mihtngale and sycamore ix thatthe fit vowel is ‘speed a long inthe former blocking Table Shortening ax shown ahove—and anipedfied for length rather than specified as [—loog, nthe fer. Then the Elewhere Condition will not block Triyllabic Shortening from applying to sycamore a its cause (i) snot met, 50 that the ue wl specily the vowel as [-longh, as desired "A prion the elimination of spacial category of “morpheme stractre rules" is weloome because the statu of these putative rules has almost ftom the beginning of generative phonology been tect with problems Kenstowicr and. Kisasherh (1977) idea Tour of them: 1. Condition or rle: ere MSRS fo be construed as rules tha ln p= Atal estore speciation left blank in the matrices entered inthe lexicon, for are they to be eonstrucd as sti well formedness conditions that check the acceptability of fly specie lexis entries? 2. The duplication problem: Why ee ropuartes expressed by morpheme structure rules often recapitulated by phonological rules proper, epying to derived forms? 3: The dormtn problem: on what sorts of ents are lexical constrains ‘etined—morphemes, stems, or Dnitbed words? “4 The level problem: to what stage of derivation are lexieal constrains applicableandelyng representations, the phonetic output, oF some i termediate evel? Tn the approach proposed bere, the answers to these questions must ran 1s fallow: 1, Predictable feature specication are It unspecified in lexical entries nd are ile in bythe system of universal and languagepariula rales of Texiea! phonology.4 Pal Key 2, There is no duplication problem becase the rus that apply t0 nom erved forms in a Blankflling function, governing the strocture of pri- mtv lesial entries, are the same lexical rules that apply after the Sst, {jl fentere-changing fonction, governing the structure of derived let- 3 forms, 3, The domains on which lexical constraints are defined ar lex ete sores 4s, the ejle consitonts NA, V. Lexialconsraats are thre fore only indrecly pertinent to morphemes (rots alles et) “Lexical conseants ae applicebie in eal (ec) phonology as dees sined by the ordering ofthe relevant rules. Thus they are not necessity ttucelter of underying representations or ofthe phonetic outpu. In pati~ ‘lar, the spplcation of posleial rules may totaly obscure the canonical steactare of leis ems rom our point of vow, “duplication” between morpheme structure rules and rule of lexical phonology fe from being «problem, i actly the preiciod normal case, We do not allow rules whose domain is defined fs the morphemes minimally they must belong to level 1 abd apply als to Such derived forms 1 met thir tractral description. This doesnot mean that all ules wil actualy exhibic duplication. 1 ean very well happen that the environment of level I rule occurs only in undoied lexical items or ty in ors dived at fevel 1; what predicted is that fit occurs in both then the rule will indeed apply to both if we are to allow unspecified feature values in the lexicon, them it be- comes incunibent upon us 10 answer the well-known objections of Stanley (G960 agaist the proceduse, We sal do thi by stipulating hat no featare fan appear marked both and ~ inthe same environment inthe lexicon. For ou tring pot we revert tothe natural assumption of erly genera tive phonology tht phonological features are uaspecie in undeying rep- ‘esentations i ther value ean be assigned bya rate, The theory of grammar ‘wil provide a set of univer redundancy cle functionally analogous to the markednes principles of Chomsky and Hall (1968), bt formally inte fal to ordinary phonological rules. In particular, assume that for every feature F thee 1s minimally a rule ( Joler whete « (4 oF —) isthe “unmarked” valve. In addition other rules may be epplicabl inspect syntagmati or paradigmatic contexts. For example, for voicing we may have the rues (3) uf J cbvoiced) ©. Eobate)-» [voted puting the unmarked value as [—voiced for obstrunts snd [+ vice) {ewhere. We now sy that voiceless bstrbents and voiced sonorants are Fepresented a6 (0 Yoke). that is, unspecified for voking, and tht their SEE eee eee ee ee exe Morbo a8 Peelers 55 respective speciation for voiing are filed in by the application of rule (43), This much is quite in the spirit of tutional markedness theory, ‘Suppose further thatthe lexical phonologieal rules of «language apply to lexical entries together with universal rules suchas (43), ae prt of the ‘stem of Fxical redundancy rule. For example, the English lexical rule ‘of regressve voicing assimilation applies on the fist cjele alexi = Sundancy rule that assigns fa voiced asthe normal vale to bstrucnts the context [a voiced This again means that obstructs in that context ae [0 voles) in Teil entries of Engh We thas oblain hierarchy of successively more specie ules, all but te fst rule 43a) applying in domains included n more general rules and supeseding them inthe shared domain. The portion of the hierarchy that wwe ave san s0 fr is shown tooguthr n (44) 4) © [dob void) (©) [+ obsty [voiced © Tog] te voice. The dijuctve ordering among such ses of rules comes bythe Elsewhere Condition, ‘Clearly, lescal redundancy rues most not be allowed to change lesiclly specied features. For example, role (4b) only “fl Banks" and does ot apply to segments inherently specie a (+ voiced) such a Jf. This does not have to be specially stipulated inthe theory, but falls oat det from the EC. if we construe each lexical item L asa ale as proposed above, Suppose we have the Eaglsh words sip and xp. We then have the roles (4S), where () and (b) ae disjunctive by the EC: but (and (6) ee not imine toa wrt ond fi 5) & [+ obst} > (— voiced) bf obste ip (ijuntive with (0) + cor iced [+ ebst ] # Conjuctve with ()) oor ee. 0 wie Tn exaety the same way the bof absent, lexically speci as [+ voiced, oes not undergo the voicing asimiation (le), while the unepecied of abdomen docs. Ths the lexical eves themselves are the end points of the abovementioned hierarchies of sucesvely more specie Soe eee eee ee eee eo o ee cocci aa oo ooo cacao oe6 Pe kpcs ries 4) ce) ao oo} Tahoe arBhent 1 follows that in every content C, only Wo lexical specications wil be possible for any feture Fi (OF] and fe} where oF ste unpredictable Yale. Therefore we eseape Stanley's (1967 objection that allowing unspexit {a features in the lexicon amounts to ntoduciag a thres-alued feature 5% "Not every context allows both possible feature speciatons. If only one speciation of F occurs in some context C, iif some branch of the rary for F does not txminate in any excl items Bat in the ral for its then we shall sy that Fi nom-dtntve in C. His eit speci tation in Cir then necessary (OF "We sil havea debt to pay of to the abstractness issue. Although we have teen able to derive the “yelic syndrome", including the properties SSiogiated with the Suit Cysle Condition, from independently motivated riciles governing the lescon, we have as yet no explanation for why rules Fold become lexical fo fhe int place. After all, sound changes enter Tangoage as posllescal rules and thete is no a peor! reason why they ‘honk in tnt tend to graduate to levica status, with concomitant reaaly- [Sr of theiesynchronically nonderived outputs. More particularly, as was fade explicit in several formulations of the Alteration Condition aad the Stet Cjle Condition, obligatory neutralization rules have a special lft for the eyelelexicon which sll needs fo be accounted for. "To begin with this last question, the answer is evidently that obligatory outalization rales ace preiey thove rules whose outputs are potentially ‘bec to lesclzation without complicating tbe grammar. Why neutaliza- tion rules? They are roles which merge one se of representations with, nother: an ao where possibly C= A or B. So inthis case C derived from A has by def inition another possible source a the lexicon, namely B or whatever the foutee of B. The lexicaization ofthe output of @ nom-eutalzation rale (Gay. aspiration In Enliph) rogues adding some redundant category to tindrsing representations, which other things boing equal wl be rejected 7 eve Meola ad Piney 7 ‘on grounds of simplicity. And why just obligatory neutralization rules? If ‘the output ofan apron rl slexalized thea it lationship tothe eter variant eannt be accounted for by the ul. This wil agin be reject on ‘rounds of siplcty Imagine for example hat the Tesylabie Sorting ‘ule in Faglsh was postlexical und optional. Then obviwsy the shortened ‘arian of nightingale would not be lxicaized becase it could ot then be Felted to the variant with & long vowel, So if posteal rue is non Dettralcng or optional, simplicity comsideratons will generally require {hati contiaus to apply o nos-derived forms, and the Ekewhere Condition then entails that remain in the postlexieal component. "Tht docs not mean that there could not be non-nevrazing rules oF optional rales in the eyeic phonology. Such eases have certainly been documented the fteratar. All tat is predicted i that postlexial rues ‘an sit into the lencon wihost ether overt changes in thee non-drved Inputs or loss of generality only if they are obligatory non-neutalzation ras ‘Assuming tis sccount of why i is obligatory son-neutrlztion cules sat may beome lial, Wenow come to the question why the les that may ‘become lexical 80 readily do 80. To postulate a principle that the “ane marked” statue of rule lexical is m0 more thin a restatement of the ‘biervaton that we wish ¢o explain, Asume instead that the language fearer is guided by © principle that selects the simples availabe dei ‘on he etteron of derivational simplicity being length. By “avaiable” 1 mein “allowed by the evauation measure” This means that dori ‘onal simplicity istry subordinated to grammatical simplicity, and only ome into pay when the evaloaton measure is indeterminate as between Sernative grammars. The priaciple that the shortest derivations prefered is elated to Zwichys (197) “No Free Ride” Pisiple and more Gistany to Post's Natoralnes Condition, which is formulated as con- ‘tion of adequacy on phonological theories, rather than asa principle for fetecing between allecative descriptions within a thory. Tho idea was ctually implicit in some generative weatmeats of anulogial leveling, =. the disusson of the loss of Hal devoicing in Swiss diatets in Kiparsky {0968 We shall ell the Derivaional Simpsty Criterion (DSC) and Formulate it 8 fellows (G8) _Derbvatonal Simply Criterion (DSC) ‘Among allerative maximally simple grammars select that which has the shortest derivations “The favored status of lesial phonological rules is esivable from the DSC ‘benise puting a al into the eel phonology aways enables non-drived Torn to be detved from the shallowest source in satisucion of (4). As fehematie ceample consider how Trisyllbic Shortening might have be:s Pt Kai ‘ome a Jesical rule. Suppose that ata cettn period there arose surface aceptiontoitsuchasniphrngal, rom degeneratecompounds vowelength- ening, and other soners. Faced with data such 25 nighngale two analyses, fare avalable to the learner. The fest possibilty is to ake the word “at face value" and set up flings, with an underlying long vowel. The second, assuming for the sake of the example that an jx) deleted with compensatory lengthening ean sil be motivated for this tage of English, isthe more abstract nintVngtTese two alternatives commit the leraet to diferent assumpdons about the phonological rales a6 well. Underlying [nstVngs) entails that Trisyllabio Shortening is Isical. Undrting tite ‘ng entalls tht there deleting [xi poslexia, Bat the DSC fixes the underying form as estVng, in tn forcing Tisylbic Shortening ito the lexical phonology. Sch retrcturing erodes the support for the j= deletion mle and eventually brags about I demise ‘To summarize, we have arrivd at the eonelsion that whatisrphtsbout the Strict Cjele Condition is derivable from the Ekewhere Condition oo the assumption that word phonology is integrated withthe morphology inthe lexicon. The resulting theory can be considered an advance from both the conceptual aad the empiri point of view. Coneeptally, achieves grdt- fr explanatory depth in that vatious piacples that had to be stipulated previously ae now derived from the itesetion of more elementary prac ples. Empirically, s¢ marks a step towards overcoming the tesion between fo goa, each legitimate in themscier but ro far curiously dial to recopele with each other in generative phonology: maximal generality and {legance of descriptions on the one hand, and maximal realism, naturalness tte. on the other, It no longer seems nese lo make compromise in one in onder to achieve the oer “The notable feature ofthis theory when compared to eatin approaches to the same problons in generative phonology, as well a o current trends in syntax is thatthe main explanatory borden eared by simplicity and the Sractre of the grammar isl, a8 opposed t0 onditions on rales or Fo presentations. The oly condition we required was the essntally Wii Elsewhere Condition, which may conceivably be reducible toa more general cogatve principle 3. Morpiume Strctre In this section we shall make an atfempt to relate our approach to m9¢- ‘heme stretute to previous dscusons of the problem and to show that the ions raed there can be satisfactory answered‘The presentation ‘organized around the four peoblems that we identified above, namely the o> Ina binary feature system, two segments Pand Q are dsinc if they have ‘opposite value + and ~ for some feature. This would make A and C dis tee in 2) and B non-stinetfom both Yel, irl ae allowed to apply to the matrices in @) they are inevitably capable of makingall three mt. Iy ditinet. Suppose we have the rules eb Rlot El b ROE) we assume the convention tat [OF] is not analyzable a [+-F]oras {Fl @) applies to @) to yield ® Ny BOIS a remem rhs 5 0 tee a Sarge aa ets oor wattaw a Eeraa Saleen o i I IC Bit po Rot 4 = 5 Eee el Meg ad Phas “ In both (4) and (all tree segments are technialy isin from cach other. But we eannot allow rules to apply in such way as to make not Alistint cepesentaions distinc. We then have to recognizs "" ay third featoe value, 50 that B is dsinet from A and C in 2). But this ‘unsatisfactory ince we ave no wse forthe extra feature value "0" inthe operation of the phonological component. The theory in that form has fexpresive power which isnot utilized in grammars, with reolant lose of explanatory adequsy. ‘Tho present theory i of course not open to this formal ecm of the banking interpretation of MSRs because it doesnot allow a three-way contrast between +, —, and O10 arse. In any envionment, ony to lel pecications are possible fora eatre F,naosly the mated specication [af and the uomarked spstision [OF], where the mot specie esc rule sppicable ia that envzoameat asians the value —ato[F} The eitence of cha rule is guaranteed by the stipulation that fr every F thre isin rmaly a universal mackedness rule [ ]—» JF] (P+ or —). Thus the system remains stil binary while incorporating the builtin asymmetry of ratkednen theory ‘Stanley further drew attention to certain types of constraints on morpheme steactare which apparently require state wel-formedness condone rather than feature-speciving rules. He clasifed MS conditions into thse type ‘then conditions, postive conditions, and negative conditions. Only the ‘fhe typeof condition i dee incerconverbe withthe rule format, the “i” ease ofthe condition corresponding to the structural description ‘ofa rule andthe “then clause coresponding to its stroctral change. The Pealiveand negative type ae etentially templates of adible and in Admisble feature mathe. In these types the transation into als is open to the charge of arbtrarinss at to which feature specications to take as redundant and which to take ax distinctive: “For example fin some environment E, the valu (+f) imple the valve [+g] and the value [+a] implies the value [+f], then it would be fanbtary which Yue we actually choose to insieate in the dictionary” (p 89. ‘Apart fom arbitrariness, positive and negative conditions alo underlay Stanley's more empirical cecisns ofthe Bank-filing interpretation. Name he ltimed that blanking MSRS are capable of describing certain Aypothatal types of moepheme structure configurations which are un- ie any that coud exist n any natural language, and conversely that they arin princi incapable of describing certain types of morpheme struct ‘configurations which do ext in natural Ianguages. Staleys own explana: tion ofthese ces rele erucilly on the availablity of poi and nega- tive conditions and on certain constraints on them tha his theory tps,im | ff a Poa ay ‘Asan impossible typeof sequential configuration Stanley cts pu linguage with the four types. of morphemes (@ LCVG “CVGL” VGLC. GLCV where L = liquid, C= consonant, V = vowel, and G = glide. Noting that it ould be easly characterized by the MS rules Dal Jsf-con/fF cond’ 1 BL [owe veoh he suggests tha the absence of uh cs in seal languages could be ex ined if we acept the conditions interpretation and spate that @) & Positive MS conditions aze admisble in the theory Restrictions involving sylable structure must be stated by postve ‘MS conditions. ‘A positive condition in Stanley's see is template consisting of an incom- petely speed mac which must bea submatrix of every Ixia en. ‘The four configurations in (6) do not have e unitary characterization in tezms of positive conditions; adjunction of four unrelated conditions would be required. Since the unnatural situation of (6 is easy to detrbe ‘by MSR ofthe tational blankfiling sort but hard to dese in Stanley's theory, we apparently have an argument in favor of the ler Its very much fo Stanley’s credit that he recognized the special Kinship of “postive conditions” to syllable structure at time when syllable rrue- ture ad sareely begun tobe explored in generative phonology, Asa ater of fact, “postive conditions” are to my knowledge oniy found in con rection with resctons involving syllable structure (or more precely prosodic stactre including fet aswell as sslubies). This would have to bbe added at third postulate, (@) 6. Only restrictions involving syllable structure [relly prosodic structure] may be stated by positive MS conditions. From the perspective of 15 yeas hindsight and more resent work onsyl- lable struture i snow obvious that no argument agian blanking MS rules can be had from these observation. Stil, because the conclusions ‘hemseves connie to be widely accepted iis worth pong over briely how reeant work on prosodic structure his uodermined thei foundations. ‘A basicinadequacy ofthe whole argument i tat syllable strut less {snot dfined upon morphemes but upon lexical ents (N, Ad, Ad, V). ‘We shall tua to ths point shortly in the broader context of clang that te sme i really true of al lexical redundancy. Let ue assume then tha the lexical entry isthe proper domain of sllabieaion a lest and that “post tive MS conditions” in (82) accordingly should be replaced by "postive ‘word structure conditions”. Let us consider whether (8 2), $0 modified, ould not themasves be explained, slong with the patent impoesiblity of (6, from more fundamental assumptions about the prosodic organization Lee Meg nl Pholoe e of language ‘Such an explanation canbe achieved if we assume (as proposed in Ki pesky 1979 that universal grammar provides a syllabic parsing algorithm, based on s metrical template which defines a prominence sete to which he Segments ae matched optimally according othe familar sonorty heathy. “This assigns an unmarked syllabiicaton to any string of seginents. The srammar of « particular language then specifies the uximam permissible Syllable stracture and such marked sjlbifestion types as may exit in the Tnnguage, eg that spt coatains one syllable and aot two in English. The later are usally quite ite in scope and in Kiparky (198) itis suggested thatthe ae always phonolopclly motivated (eg. the monosyllabic vale tion of spi is forced by stress and other rules of English phonology). The ‘ote’ of universal slgritim and language-parcuse sylabicstion rules characterizes syllable structure exacdy i exactly the same way that he interplay of universal markedness rules and language-patcular lexical ‘les characterize the segmental inventory of language, a8 discused in fee. 2 above. Assume further the universal condition (MeCeray 1979, Le 198): (©) Lexical items must be exhavstively syllbisable, ‘The explanation forthe impossibility of (6 is then tha, in accordance with, the socalled “logic of mariedaest™, no language wil sstematially prefer comples syle structure to simple syllable structure, this hypothetical Jnguage docs. In particular, ths hypothetical language gros votes smarkedness by roqtig all sylabs to contain casters (2 thas LCVG. ste. yt lacks CVC ate), and by consistently sequencing eriain consonant ‘Combinations against the sononty hierarchy (LCVG but not CLVG, GLCV but not CLG ete). he upshot i thatthe theory of syllable structure forces the equivalent of “positive condition” (azmely the syllable template play the role that Stanley's theory asigned to them by fat, but i places them (more correctly) tthe level of the lsc etry rater than atthe level of the morpheme ‘There is consequently no argument whatsoever tobe deived from sable structure against taking ll redundant feature values tobe specified by blank- filing rules. ‘Stanley aio claimed counterevidenoe of the converse tye, where Blak fing rules are incapable of characterizing actual constants on morphere steuture. This involves his negative type of MS condition. "A nepative MS condition i construed a an incompletely specified max [Nuch that ll matrices in the lexcon of whch Nisa submatix are 1e jected. Stanley cites an example from Proto-Indo-European, where obsta- ft tope atthe beginning and end of morphcines were subject to certain co- ‘ocueenceretrstion involving their manner of eticulation, Assuming theo Pra ks “classical” reconstruction with thee series (00). oiceless unaspirated: pt, Ky” voiced unaspirated: (yd 8 voiced aspirteds hyd gy g™ we have the following combinations wth stars inleting the impermissible ‘ypes tp. a pd pat bah sb. od beast bh. | ah bh. id set where pb, opisent the nial obstrents and dhe inal obtrents, Td "2 denotes intervening sonorans. For example, there were m soe sepa et bere wrens mores ed aaee at eps mse MS cnton (2) CoE tt cots oF) “ Sint | Bs (pel ..L-£ 20) asaya atin pe so MS so or Feeney rari hed nt EITHER sae org tt nt mara tv natin sett isa oon ih or oe sation aaa a Sates ens aan ae wh ere ‘sel inane ol eno atig at ee tnt mcary to tam ta ep” 9) rn a i pt fr hs rr ne poe 1 Sets tt eta te fi fs et ‘Re Ray sue he ever matte ne Ge seam Secs fovea, = 489 of 2) ad th ea ins aon ng Ga teael tel AB. espe er 3 by he EC Ta es te ad Sod eto a5) Troiced] asp] ——_soiced) sp] noe o> = aa re a a eC) Au 0 tH OD dexieal ep. output “The unwanted combinations of stops in (II) are excluded by adding the sale (16) (4 voiced) —> fe asp [25 | iors) which 7. 9/0. Ibs =» hp.» «dy and "Ph Lexa Morpseand Pony 65 >). th, as shown in (1: fed) fp. ny @) hh «td risiflang peso routs sy Sual(ote] [oumlL2ea) [29] (on (wy Can (eae) Fae a ott Eom Levy Cea tea ENG (Sa preqtiel poedieet ea on CGS) Ess) SSeS he Sa it ie it He etic Wa fn pve na rwsmtins Ta vat [Ov] gq [Ov [+10] oy [+ et] el [Staple] © feo] © [oe [= Sl and th BSC tht ie xp i) ore or Be bee GT frees ca and OB) ove (79 foe Fee ge a Se scp of meting ee sinh ini iy mc io sist my eae nt hf ee Mcafee teint ote DSC ne [Tipu oe contnaion, ach te resson of Au meson! by Keno an Ki sete (97 3), sae aol and dtl wap oe mata ale 12S ac woe ent gael ous ony Th com. sha hha crna mot bee Fl we Fie ore aap wapthcot at deat Ta sabes ty aigng eh ee ‘theta mre ar han ind vp ha te ote) [araanipng te uma aut wl spt adeeb Ste mores one mtd can) to vote Mu Tao'macy oe sah any cet Be aut seal tenet vant eng nd ht ey Sen asi vo date i rapt oot en paola erie toe What doi can sone fo 3 Sov eThe wcral etn anereatrtie 8 Veet cosequ oe bing itp of MSR ny Bowe ihn i bah solo inrpeton Nor doe epunent fom tetas aces poet ion ev ae ste Mot pe tay he dre fom inher we hol oie veo of SEC cyl cond tne th eo nd re ‘Stor eet soy ny chun fe fate tm 8 Seattle ta ile ection MSs oe an Uae Uo rr nee In oof KG photo,66 Pt Kips 22, The daptaton podem Inthe case ofthe problems just discussed the phonological literature for the most part gives the impression thatthe matter is more or lst under control. This pot so when we tar tothe duplication problem. Within Standard generative phonology there have buen virally no leds on hhow fo deal with the undeniable fat that much of morpeme ruta is recapitulate in word structure, so tat there i substantial over between (MRS and phonological rales on the common understanding of how the grammar is orpanized, The subease of the duplication problem which has recived the most, attention isthe problem af where sepmentstuctre rales (which may be ‘consrued as cotent-fee MSRS) are to appl. Stale (1967, 4 fist elle Attention to the base paradox: “For some parposes we want the sepnent sinictre rules to state ther redundancies EARLY inthe P rules [ori the MSR PX, so that these redundant features can he wed by oer P land MSR] rues, But for othe purposes we want the segment stroctre rules to state thee redundancles LATE in the Profs so tht they ca late the redundancies in segments introduced or changed by the P roles. Recase of this confit, nether the usual practice of ating the segment structure rules be scattered through the P rules, nor the practice we advoetted carer of Jeting the segment sireture roi appeal only in the MS rules, can be secepted as it sands”. Stanley's proposal was to adopt the convention “thatthe output of each P rae is automatically subjected to the segment structure res.” As Stanly noes, this accounts fr the stuctrepreserva- tion type of P rules, but i oo strong: "It sem certain that we do not ‘want all P rules, especialy the late ones, o have their output sbjeted to ‘he segment structure rules. ts sillan open suestion how we ae to enti ima non-ed-hoe manner, just which P rules are 0 subjected.” ‘A stab at solving that probiem was made by Chomsky and Halle (1968, ‘Ch, 8). Thee theory of markednes incorporates the “inking convention” ‘hat markedness conventions ae applied tothe output of phonological rls iand only if they are applicable to all the segments in that output, Ths ‘departs from Stanley’sconjctare in puting the burden on kind of pine ple of symmetry rather than on the ordering ofthe rules, and in allowing ‘only universal segment-structure rules (2 markedness conventions) 10 ‘reapply inthe course of phonofogical derivations. Nether positon addresses, ‘he more general question af duplication in context-snstve rls, Lexie! phonology entails that “dupeation” should be the normal ase for any lexical (Le. "yele") re, wheter contexte oF conten enstive ‘and whether universal or language-petular. (in sense this agrees with Stamey’s conjecture that it hat to do with “esly” rules). The predicted EEE ee ee eee exe Marpolyad Poliey relationship between the structure of simple and derived lexical items i therefore a follows: (@) Ifa rae hols for leicly derived tems it holds For nondeived items in the unmacked ease. This follows fom the fact that undeived forms ae fubjec to lesical sles ony with respect o unspecified feature speciation. (@) Ifa rule holds for lexically undevved stems, should old sso for lexically derived items. This fllows because we have no special eatgory ‘of rule thats applicable oly to nondeived forms and no general pringples ‘hat Block re fom applying to dvived forms. OF course, pos-encally is wid + 6) = with, We eonlude that it this rule which also fils n the voicing Speifettion in the non‘inal obsrasnts of ade et ‘A paralel argument holds for nsal assimilation: “Timp, [ok ee are cxclded by an obligatory lexical rue that also apliss to lexis! morpheme ‘Sneatenations, Another reason why the constat must be governed by & Teva ule is that it limited to the metecal span of foots acres afoot ‘boundary the ple of articulation may disagree, cf. efoauent vs. clag nt, The later is optionally asad to clnadent by a poster rule thal, beside it optional, difers from the lexical rue also in {eauirng simple striagaaoency (rather than foot- boundedness, /) uta, and in asimlating at a arteuavry level (rather than just the ajo place Teatures as tbe lexical rule does). Thus, in symphony we ‘exive fist +f)» [on by the lena eule and then optionally)» (mt) (lbiodentl nasa) by the postesca ule Note that this correctly predts that we have two ather than three pronunciations oft France: [a] Franc) {tm} France, depending on whether the poslesical asimilation appli, not *ifm) Franc, since excl assimilation inapplicable.” More complicated sth distribution off) and [on stm-fnal positon they contrast in nouns but verbs end in the voiced spiant SEE Ee eee6 Pat Kigaty (49) 2 wrath, wreath, Booth, moth, heath, sheath, death path, th (Op, seth, lathe, ithe, spate (oD. be. wrth, sethe, tithe, soothe, soothe (). ‘But again this is not a royulacty of morphemes but of stem, because it holds iso for vers derived from nouns (20) tothtecthe, bahfbathe, mout|menthe, breatbreathe For inital position the distribution ss: [] im leva categories (nouns, ses ae) ed in ow case onion, oma, {@i) throw, think, thick, thorough, thong, thigh 2. the, this, that, then, there, thus, though, thou, thy tee “The rule could be formulated uo sated, in which ease it could be alexa rule but equally well « MSR of the taditional type, Perhaps however, exploiting the uct that only lexical categories recive sess in the lexical (Gree) phonology (emphatic and conteastive tess being assigned pose Texiealyat the senence level) the rule i beter formulated as (2) {@2) G-»3in the onset of an unstressed syllable in which cate the voiced 8 of the, shire fall together with the voiced fo) ot father, mater, robe, heathen, frting, feather later, gather, bother, titer, neither, fathom, (ith excoptions in the “Greck” vocabulary, e. ‘uti, method, eter, Athen). Rule (2) 4 necessary leial Both be- aut refers tothe ouput of leases and because it governs lec 180 alternations like south-zouher, nrth:nrthe, crthearthen, sah thy, clothing. "A particularly inleresting case involves the constraints on consonant castors and vowel length. There are two relevant lexical cules to consider here: shortening and slabifcation. Long vowels are shortened Before de rived consonant cluster In level I eset phonology C3) fhap-ky > ept Iswep-ry > swept Imary meth tree Id y > fedp ted faeem-+ ey elm fa+y at jen meee fost) lest id +0) width Inside morphemes, long vowels do occur before clusters ending in coronas, ‘ea oil, beast, aint, count, coax, strange. The Elsewhere Condition wil cowecly block the Shortening rule from applying tothe lexaly specie Tong vowels in these words. The question then is why Tong vowels are batted altogether before noncoronal clusters, ie. why there are 0 sch aa a aaa alia, | | ileal aia deleted ata aka a ta —— een Manag an Poly ° nouns or verbs as ching, eang, *fupk. Th answers tht sich Isc items are barred Because cy cant beasigned a yale strctae. The syllable structure rules in English allow a peak followed by a coda ‘consisting of at most two segments in order of descending sonorty, and, in inal positon, by any suber ofexteametrcal coronal consonants: That Js, the thyme must Be of the form (24) (4 sal (4 cont ‘where the elements of the core ([+ seg) appear in order of decreasing onary, as defined by the weltknown scale: vows, ides r, J, mals, ‘obstrcnts. {take i that vowol and glides difer prosodcally rather than aie ‘segmentally and so the apex of the hierarchy should really wf [FRE] we can rep sont hy in ex Srey FF. with the inlrptation th pon i the ‘eet tidy hepment otto Fb Breed te tie heey cn 6 Supe (aS MUr dt coo (+ meta t ob] For nl ve ae gue Go aintmp ton Wh nh Hor nh iB nec de 1 frm te {Sua hme ohh sna hr veon a hal eer at ‘ese toe he ti eo poner ac rade ns Saami on hes) Thee fea thyme coe Chae Grd) tly aym (wim) ap (op) an (oar) mm (arm) ap. Ghar) cm (in pel) emp hemp) In addition, the syluble may coatuinextrameca coronas tacked on to the end, eg. cas, fies, seths, (how) believed “These ues of sylabification are inthe grummar anyway to agcount for such facts as that colori dsyllabi but car x monosyllabie, that Krerd ie diylabic but that reabm is monoslubi, ete Thus they have nothing {o do with the “morpeme structure” or Sequential constraints on segments in words dies. Yeu they ae indirectly the most important determinants” iy ‘of those constrains by virtue ofthe general principle (9) mentioned eater that lexical lems must be exhautvey sylaiable. A putative word which ‘snot be pared into sylables by the above procedure is rejed by ©) tnd is therefore inadmissible ‘Conarsats on sfies are likewise derivative of @) andthe syabiestion rules. Tho reason why consonantal sfzs are all corona in English is tht on-corontl consnatal sufines woul form words that could not be parsed, ty (8) For example, the noun-forming Sufix 1h could not be f insead rca # word “warm mould be inadmissible ia Engish syllable stustre. Treterste that at al rpulacitis of morpheme structure have to eappest ts alternation in derived forms in hi thoory. It could very wel happen that ‘Served fori seedentally donot provide a chance forthe lexis rule to fpply. What shouldbe tre on the peseat theory is however that no rules ‘mast be restricted to apply ony within morphemes. 1 do ct infact know of fny “morpheme structure rues” in English which ned tobe so resid T conclude tht at leat English confi the predton that systematic constraints on morphemes are a funcUon of lexical rules. An apparent Counterexarple of cited by Keartowicy nd Kissebeeth (1977, 148) rom ind: Nase agree ia pace wth following stop within morphemes and yet, heteocganc combination retain unaesimilated across morpteme bound fies, eB an fn unknown’ ci + er “having Kise’, c+ nis, Kam + dar "manager, and wen they aise by a vowel fycope rue, 6 fmol “shine, camk ‘a ‘shined. In fat, however, nase do undere0 "similation in rome kinds of morpheme combinations, f the prefix san: (@8) som-+ bandh “connected (6. Bid ~ na, /bindb “Ge fon def ‘command’ (f. &+ del “ule, iasteution’) fap + git “musi” (Cf. gi Song) ‘conjecture tha the sues kay ond le before which assimilation does not take place belong in the level 2 morphology (Le they are the counterparts of the Englah "s-boundary” suftes) If that can be maintained, then indi i uit compatible with the theor. This account requires that vowel Syncope be ordered after nasa asiatin, Ae far a6 T kno, his can be fo, and if vows syncope turns out to Be pote, it actually must beso in opposition (othe View that “"morpbere stuctre rules” can be as the phonological component (which they term the “ordering to the duplication problem) Keastowcr and Kisseberth oe the following argument "Suppose & language hs a role assimilating the voicing of an obstruct to. following obstruent and tht this rule both deseibes alterations and tthe same ime exreses constant on lexis representations. According {o the ordering sollion, the aosaital members of an obstruent cluster til $e 0 for voie andthe voicing assimilation rue wil simultaneously ft SURGGEGSISSSRISGRGSGRGGSEEEESECESEEESIUGERIESEESSESSIGGGAR: © (SisnEEtESEateunsrsacesssenis ele Merl ad Pooley 1 in 0 and alter distinctive pls and mings valves. Now suppose this lan- ‘gage has rule lengthening vowels before voiced obstuents, bat that thi rule ust be ordered before the voicing asimiation sla. Thus, we might find forms Uke the following: po, para, but paca; and map mp, mad= dl, but ma:b, mapta,masteda. Now if we were to find form like mae (and ma'skte, mas). tere would be no way ih which thelength re could coretly assign [long] to the a, since atthe pont where this le pois (before voicing assimilation), the 2 will be (0 voice). Although tis isonly& hypothetical example, ve would not be tal surprised to find 2 languige with similar proper” Iris quit teu that such a configuration of data could not be derived ‘ifthe grammar is organized as propose. If voicing assimilation and lngth- ning apply st lesa rales im that order, ax rouired to derive fab +t “yma;pia and /naSg ta) > ma-skta, then we would predict that map ¥ day bevomes ma:hda instead of mab o lap} da) (fae fee) i] a) 11, Voicing As. [J [=] [osza) maze} 2 Lengtbening (— | [macd) maze maze H.1. Voisng AS. mabds ] mapta = maskia 2 Lengthening mabe} ‘ade map mace mnie But it would te unreasonably defeats to reject thoory om the basis of hypothetical counterexamples. If real ones cannot be found, we rather havea prim facie argument forthe theory which reds tha hey do not 'K & K further note that rues which bot lin zeroes and change pluses and minases “wil often fil to corey state the constraints on undelye {ng morpheme shapes" They ete ete in point from Klamath. The cone trast between voiceless aspirated stops (writen p, 1,2, kg) and voiced ‘unaspirated tops J 4) i nutalized preconsonantaly into voieess unaspirated stops (2, 7. €, K, 0): (G0). body") “wrinkle” mbodi: + ek ie) mboty’ +a Moxsw) ‘covet’ oss bt oxtw a While aspiration is contasive in underlying reretentations in preconso= nantal postion, as these examples lostate there ere morphemes whose ‘underlying preconsonantal stops ae indeterminate a to aspiration bees they never come to stand before @ vowsl in any alemsnt .poP as “hoard”. Those stops shouldbe left unspecified for aspiration inthe lexicon and rectve their proper speciation by the neutalination rule, Av the {example shows, “sven that a phonological rule both supplies unspecified Teature values and lb alters feature vlues, one cam discover whether that rule expresses constraint on undedying morpheme shapes only By exee 2 Pat Kary mining sich lexi repvesentation to see whoter thre are any tht violate the rule, Thus,» the proper MS conszunts Would never be dell ex. presed” (K & K, p. 1. "The point ofthe examples thatthe neotalization re applies in a Banke filing function in nonderived eavicmments 2nd in featurechanging function in derived environments Asime tat slabifeaton applies at the ‘word lest and thatthe nestaiztion eule apples to sllabe-nal sons, nd that aspcation is itnesie, We then have the rules I): OD ut Jochen [obser] [ved] 6. [robe sat (asp 4. sylabieaion © Chott ep salen vet where (4 b) ate universal and (6 «) ae Klamath-partculae. By the EC, Gisjunctive onering hols between (8) and (2). S50 C2): (2) pfens mbodisttk mboTy'ta lo:8o-}t lo:tw'+a rer) Try eye 322] [owe] [oa io) [see ca) lowe! lowes) Lowel Lose oiy ee fae) a Gia Ceved Eeve Gere) (ved ed) ‘ a peabeh, fram oo Oy (Oa CS ES] i in| es cal - Esl From our point of view, to say that a feature Fis datintve ina given content Cis tosay thatthe specication + For F ocersiaCin telecon, ‘Otherwise (.e. if only OF occurs), Fis nondistneie. ta Klamath, aspira- thon i istneive in prevoealic and in preconsonantal posion Because [+ ‘sp oceurs in the lexicon in both those contexts, while voicing isnot dis- Hiptve beeuse everything i (0 ved} in the Ison. "A thicd general argument against hiving morpheme structure goverod by rules inthe phonoloieal component which K & K take tobe perhaps thet Strongest—is that the “duplication problem!” “i realy jst a speci fubease ofthe more general problem of “conspiracies”: "Tt is reasonable ‘to suppose that any adequate theory of conspiracies will naturally extend ‘elf fo the duplistion problem as «spor subease. On the otber hand, ‘here i simply no way in which the ondering solution can be generalized Sa aaa a all Lest Maroy ae Petey a to handle conspiracies (p. 14) Recent developments in phonology have done ite te suppor the idea that sconspitacies” ave a untay sategry of phenomena regirng some base theory of thee own. On the contrary, the various puzzles that have been ‘consigned to that entepory have proved 10 be ofa rather heterogenous sort ‘Some “onspiraciee” have founda piacipled expinaton in atosepmentat theory (ce eg, Goldsmith 1976 on "tone stability” In reference toa “eon spiraey” in Lomonge), Others soch a6 the Yawelmanl cae analyzed in issberth (19%) and cited-—togetier wih a similar Tonkawa case—in the present context by Kenstowies and Kiseherth (1977, 142-14) yield tothe theory of sllabie phonology in now obvious ways. The real unity in these ‘phenomena sees fo be that they involve prsedle organization ofthe pe {hot carice gonerative phonology was ill equipped 10 handle and which Uharsfore appeared particulary problematic, Tsun, eical phorology entails a pineipled and general form of the “oadetng slut’ tothe duplication problem. Ics well supported em piel the apparent wsakssesatributed oi fn te earle phonologieal Fterature are partly misapprehensions ane the remainder is defused by recent prosodic theory. 3. The domain pobie Kenstowiee and Kissberth observe that there are cases in which the domain of constant isthe word” rater than the morphsme (1977, 145. ‘Tunica (Hos 1940) requires all words to begin with CV, and root ‘norplcmes, which ovdinary caa oor in Word-intal position, naturally Conform tothe same constraint. That the word constraint i primary and Jmorpleme siryeluze i derivative fom its shown by the fact that posbely {hove root morphemes which happen never to occur wordiilly are no fubjeet to itt inaenably powsesed nouns, whch mast oseur with 3 postive prefix, may begin witha consonant cluster or with a vowel. An rset saalogous station is found In Odawa (Kaye 1974, ‘Like tound roots, afacs may also be exert from the constrains on crdinay sts, Thos, they seed not be ssabiable in themseves; what IS required is rather Hat when they are added to tems the result should be Syllable, For example, Enis has consonant sues but they ase coronal. Tis resection i derivative ofthe sable canon described above: Stch hypothetical words a8 wert +; “help + would be unslabifable Wile the coronal sullxe in worm i, help} cam be parsed as ext Inetieal. Therefore the retiton ga the ples of ateulation of Engh ‘ononanlal sufixes ned aot be sated a8 a “iorpheme structure con- “The literature is of course fll of proposed morpheme stusture onstraints [have not made a spxemati investgntion of them to see whether thy can, tothe extent that they are coret, all be formulated as lexial rules, at would be implied by the present theory. But Ido think that hose ‘Examples which have appeared in theoreti discusions as speci evidence for morpheme srcture conditions can te taken t0 be lexical rules, and indeed in many interesting cases mast be so taken, A number of ‘uch examples were already cited in the preceding section in conection ‘wth the "dopeation problem”. Some additional ones willbe taken up ‘Kaye and Nykiel (MS) note hat losnwords in Polish have adapted tothe underving phonotactics of the naive vocabulary, which in their view [selina interme of morpheme structure. The base consraint which they propos: excludes noen stems ending in -CR, whete Risa sonorant conso- ant (excepting macelies ending in nas, and -Aw clusters). Foreign ‘words are brought ino ine with ths constraint by breaking up the -CR ‘ster with "ex which they write 2s E), an underying vowel whichis Towered to.e before ancter E and otherwise deleted. Far example, eff "eb" Gn cards), bortomed from French traf, is analyzed as jet, as Cidenced by the diminutive tefetokfrefEI-} Ek + B. sagenko (1970) hha earlier made very similar proposil for Russian, involving a rule breaking up stem-fial CR clusters, oon 7 foie ea — [0 whore the qutity of the eri alo predictable (one jer next to velars and wh footer elewber). He too pointed out that cans have adapted to (his patter cing the examples G3) vila water ately) Mba. “saber Sela, bel ka ‘iis “Slpper tafe ka 1 am notin @ poston to comment on ether the Polish or the Rusian material with authori, but afew observations ie clove at hand which em, to support the positon that rule such as (32) can and must be a rule of lexi ponalgy, The reason t cas be a rle of lexical phonology is that the Jers mast be prevent thoughout the “cyclic” derivation and are ony de- Teed postyeialy, as shown by Pesesky (MS) and Rubach (198). The feason it mus be arle of lexical phonology is hat apis also to dese C1 R chsters that arise by sufBration, Rusian hus for example an ad iective forming sufi fl, seen in out eft behind, retarded” (tot, oes eft behind’, 2 "ied in Alexia “rupation” rae deletes (as wll sf) before consonants fv + 4" ie’, iy} ti, to ive “This ae als applies before i iv 4 1 4+ 6) > Hg, Because taneation recedes jerltion, which isin ny ease poston, this derivation shows ei Maly and Pheiey 8 ‘hat the sufi bare without proseding jr Now consider the adjective ‘hdl 'shining, Bight fom the root vel ween in sre, 10 shine rom the argument jus given we know thatthe undetlyiag form of the sem sist be fet + But the genitive plural is tt rom jvet-+ El + El, ‘ith jor inserted Before the , TheSoute of this er can only be rle (32), ‘pplying in this cae to a derived frm. This, then gives an independent basis for assigning rule (32) tothe lexical phonology, atleast Resin, Kaye (1970) ako adduces the ease of harmonic proceses, where the hasnionie feature ix characte Duta neces (because of ‘of harmony) af the entire word. These cases are however accounted for by the approach tothe decripion of harmony that was ris mentioned above Inher lel marking forthe harmo eatre is done atthe level of the ‘morpheme “The contrat between s and reteoex gin Sanskrit offers a particularly clear case ofthe domain and duplication problems, The rule hat neutrals itcthe sovalled “ruki” rule—shows some complications of exceptional feterest, and figured prominently In the anpiments forthe original RAC tnd later forthe SCC (Kiparshy 1973, Mascaré 1976; therefore it wll bean Sppropriste test cae for lexis! phoaotowy. "Pherak rae etofexs sto safle velar, nd aonlow vowels. Though the seul formulation ofthe rein tems of features isnot important for fur purposes, {shall assume that retolex consonants are defined by the Feature combination [--eoronal, high, back]. Since the mid vowels ¢, fare represented as ju the stage at which the oki rule apples, and r felongs ponolgialy to the clase of retolex consonants (whatever its ‘honest may have been exact) the envgonmeat can be tated as high]: [+ oont ) phigh (3 St) [mea] “t+ sah) “The ra ele apis repay to the many inflectional and derivations! Sufies beginning with sit lso applies in compounds and to elie, albeit with lexical exceptions and some variation (Wackernage! 1896, 233 f, Hock 1980), G5) 2s. sufi: dobdhsi you sie bitbhar+ gi “you any ors sutix: atyBts-tam ‘Iwent atbhrts+ ‘Tearried! future soi: beametsya4 the wil vakctpyatt ‘he wll Po sy" Assidertve: di-+di-+sa-+ti he want ni-aT4 s+ he wants woane toad” at psx: sents armies’ agn-tgu “res SCE ee eee eee eee eC% Pau Ray ‘compound: sad ‘siting’ parirgad ‘convention’ sie: sma “indeed” pacchant sma “they do (i) go “The muti rule exhibis the classe syndrome of lexical properties. Tn aniny morphomes,untetrofleted s appears in aot-dedved “Tuk” cavionments, e.Ksvlaya “sprout, bors ‘tip, barra socket of toot’, Tusona ‘ower pustaka "book, some already in the Rigveda, eg, bisa “Tous, fuse ‘mis rsa ‘a demon’. 2. The aki rule applies in derived environments. Wordintrally, iti victally exception ee, both across morpheme boundaries asin 33), and :morpheminterally when the environment is crated in various ways by the application of zero pre ablaut. For example, inthe root fas “instruc a is wotkened ton the weak grade environment, This pats the sito the ‘envionment ad we get + fa taugh rom (Jas +c Simi I, the root sha eu loses eg vowel in ceviain forms, for example inthe redupiested 3. pl. j-} har ant > Jatat. Asa rest ofthe los ofthe ‘owe, the rootinital velar (wsimilated to K) causes the root-fna sto tur nto. The root shine” and the homophonows root meaning “vel? duce to gin the weak grade «pl perl a+ vas + ws) ~>eqs shone, ‘duce, Similarly, the sux sas weakened 0 ue. ER. Sid v8 “Fas) “> vidas enowing. 3 The outpt of the maki rle 8 phonemic. Nominal stems and verb roots ‘ough not lines) san have gia eovironments wer keannot be derived from the rut rl, eg gost gph “iece af wood’, Base ea’, bhag ‘Spe’ api "pi fey“ ap "arate’. In these Teel items it ecesety fo set up undedying “The ruki rule must be ordered in the cycle because of its interaction with ether alsin the phonology and morphology. One caso fo his as to do withthe fact hat the rut rl scomigly applies “across” an augment or redapliatng syllable. The ease that interest us here isthe combination ‘ofa preverb ening in? with root ginning nx Tn this case, retoexion i [ener in Vedic and gradually acquires excl exceptions asthe language levels towards its classical state In pst ens forms where the augment a ie inserted before the root, retrofeon stil takes place, even though i conditions are rot overly met. For example, the rot sie ‘sprinkle’ forms ‘the compound verb root abheite“anoint is perfect is aby -+a-+elteat {G.sp). In order to derive gn Ghat ease, the rut rae eannot apply at the ‘word level or postyclicly, singe #8 environment would not then bs met ‘The corect output ean be derived, however, ifthe rukt rule applies eye clicaly, piven the further (independently jutied) ssumption that the reverb root combination is’ constvet. The denvation (suming Seain that eyelie nodes are those labled withthe lexis eatepores NY, ‘Adj, Aa) will proceed as follows Lee Merny a Pngy 7 (25) Cycle 2: Morphology: compounding Boia Re shy Phonology rit rule ‘ Cycle 3: Morphology: inflection [abhi ata A smite stoton athe inthe replated forms of Ure compound ‘eri, Unuly, bth the ot andthe replatin have fe erst ‘thorn ‘cre (rom abi se). Redupeation and vugneat can abo ‘eur lope 8, aberyatia ‘further incon tat the nee ji is ted on the fst te appiation across lie boundary in Vode Mocked by anater #0 the prseding or following lle Hj and Shara 197) This es way Itelf dened by woroteral eppliation ofthe aki ral Therefor, wich inating |. se, = ered is predcabe ke rats pid Selly and oly then (20) (2) sO} suena strigu sya Cy) Re. 7192. (©) tamgurgeta tse — tam sug Cau) oe 2208 Having provssonalyexablhed Be eycty of he rule, we can takes closerlooka the representations to whieh mst apps: Gv that thee ae bal underyng and the Guest aa Yo how src te should be salyzod inthe morpemes i which aay sand the Tuk environmen, and 30 never arash Un cla generate Phonology ght output would be produced fom ether uederyng form, ‘The RAC, a ll ae the SCC which replaced oth prohibit thera ele from apie in sordrved enioaments, and Thereby ental at the ndesg form in such eses mart bave ac For esimpl, such lexical items as ge pison san “ea, doa “vei "enjoy ar “a, Plow, “jy di hat’ yma’ pl “cra shoul, according, Ibe entered inthe lson with "The atracive capac of thi analy that word ke iso, sum do sot have tobe marked ax exceptions to the ro le, hut retin stra Simply by vito of bing entered wi fy te lexcon. By the same token, tbe dawtack tate exten of he marl ten cant exp why 3 isthe normale inthe rt environment even in non-deedenonnent, {otha we ae Teh without any scoount ofthe margin, "marked” sat for words Uke Bo, kann the Stok leon, "With the theory of ka plonloey we Bate the rst of oth wor “Tabing “tv aan atbrvition for ie etre eng reaexion el wt fay back, High ar sgt above), Ba enterd wid [=] an 0 docs not undergo the rl, wil i entered. with OR] (ospeiod for R) and gee seid 8 (R]by the ale gener we hve the fellewing ae 1 mae cass: etre as (OR) speciod, as (+R) by rake it aplcble (a), there a8 [-R] (es)® Poa Kies 2 marked ease: entered os (-R]or [+R] les cannot apply (ia, 929) 5 exceptions: marke as not undergoing rok rule (certain compount). ‘We mst now show that our new analysis seeounts equally well forthe evidence previously ced in favor of taking non-akermating 3} in the rukt venient as underlying [gs Thre arguments were given for doing 8 shih we shall now re-examine in Carn. : ‘One argument was based upen the fact tat the uk ule doesnot apply ithe folowing segments snr. For example, when usar “awn (rom as ‘shine) foros san the weak cases, we get 20 retrofleson, ef. fen. st tras, Fors ks 3p aor. jug - rae would then seem to indicate that the oting a undedsing, since underying fs! would give “gjsrn. This argu ‘ent compelling from the viewpoint of the SCC. To get gin avs rem the ui rule would have to apply i the fist eye, but the SCC prohibits it from applying there. However, lexical phonology entails precisely thatthe ‘ut rate docs apply t wnapoced jus in the first cycle, which gives the ‘eared rl. The fest cycle the verb stom, formed thematically diet fom tbe soot Hencforh we shall we capital S10 denotes unspecified for retroexoa, Le. (OR) (G8) Cele 1: Morphology lms uk ale : yete2: Morphology (inven Abiaat jug in Ober rues ‘juga ‘The only question is: why do we sot snilacy get retrofeion in the cli deriaton of way? Because the rut environment here deve by Zar prde abla, caused bythe sect infectonal sux inthe last eyo “Therefore the rigger w and blocker rof the ru rule both ome up sim taneously (has > ards) and 20 the raki rule never as chaos 10 ppl. Derivations of the gen. 8. and ac. 6g, forms and the denominal adjective are gen below: (G9) Cycle 1: Morphology {slaty (oaSlaly UvaSlere {Chto 2: Morphotogy [vaSarly 4S), UvaSSr}amls laSaehaly ‘Cull phonology: ‘bloat [loSeéS} — (uSinham WS rk Tosithame = rie dram vasara “The explanation thas tuenseruslly on the fat that de (inerenly unc ented) stom ofthe noun awn” 38+ ar) although de fst oflable “lvays seduced tow by vero grade inthe declension. Ths in Cur as (0 be the ene if sre grado triggered in certain morphologel conditions) by a fallowing accent, as inthe theory sketched out eg. in Halle and Ki- pansy (1981, Moreover the ul grade underlying form pred e conest evel Murphy an Peels » form inthe adjective derived withthe sux -, with strengthening of the first syllable: /vaS 4 ar > odsard relating to dawn, atti’. From putative stem (ug + asf we would wrongly derve aura, L eonlode that the contrast between was and ajisan is explained eniely on independent grounds even i both are given underlying forms unspecified for retrteaion, Ts, the cates whore blocks retrfeson can ety well be dstingised from the cass wheres doesnot een i the aki rule applies onthe fist. cle as required by lexical phonology. This answers the fst argument for the RACSCC theory. "The other two apparent poses of evidence for underlying in the ruké cnvigoaments concern gin sie, The facts ee quite interesting snd deserve ‘nother look "The eects ofthe rut! rule can be observed in the nominative singer, ‘nding = in the approprise environment. Before following voiced ses ‘ment, lal in ak environments surfues as reg nl gece ism {cchar The rs derived via yb a postylie voicing assimilation ule appl {able to all obstruents, followed by an ablgtory rule which convert the ‘output ¢(aoneristent in Sasi) (which we aleady know telongs pho- ological among the relrfiex consonants). When a voiced segment does fot follow, me get # in both rak and nonruk environments, e@. ai, riomh, devah, which assimilates in place of articulation to a folowing Yolcelss consonant, oligntoily i this is coronal, optionally otherwise, agri tara, apmifagns kart In certain close combinations, however, ‘a blocked and there + and ¢ again appear 26 per the rk rule, ep dar Dutra “the son of Heaven’, djaus pa ater Heaven". We const that final subj othe rk ul, although this is ony detectable in erin ‘environments because both and ar lable to turn into -o is assim fed variants Before voiceless segments and pause "Now we note thatthe deverba suflxs ia have exactly the same san- hi Behavior asthe sequences 5, w-ts, eg. [hav + sf “bltion’ a) ‘Fuso formolys ef, (nom. ap) owing} +> nuh and (lose combination) yojuy Karo “be males eI follows that these suas ‘mst be represented asi, up at Feast at tbe point at which the posteyehc saachi rules spply. “There is compelling evidence that these sufies are neverthces unde ingly unspeite iS), Jo In the fist place, the outcome of underlying tefre sis ifernt from that seen in thes us, eg noms. 0% +) + 4a wi external sndhi forms based on while nom spay + 18+ 5) and feaks + uS +S) have external sani forme based on -f, 2 show ‘shove. The oly basi for understanding this eiferene isto have, at the ‘elevant point, x distinction batween/S/sad jy and to havea rle with the “a