Janez Orešnik Ljubljana ON THE LA.CK OF PALA.TALISATION BEFORE -END- IN THE PLURA.L OF ICELA.NDIC NOMINALISED PRESENT PARTICIP'Ll;š SuCH AS LEIKAl\Dil Summ~. This paper discusses the velar pronunciation of the root inal segments in the plural of the Icelandic nominalised .present participles, e.g. in leikend-, the plural of leikandi "actor". Since k(k) and ~are pronounced palatal be:fore e in other old word:š'"(e.g. in akkeri "anchor"), the velar pro= nunciation of leikend- and sJ.!Ililar words requires an expla- nation. I see in the non-application of the palatalisation rule (while that rule operated before e; it is by no means certain that it still does) to leikend- and similar words the effect of Kiparsky's phonological universal ("Neutralization processes apply only to derived inputs") as modified in Orešnik 1979 (substrings crucially involving word-derivational morpheme boundaries are not derived inputs). Thus the /:k8'.0+e/ of leikend- is not a derived input, and the palatalisation rule, a neutralisation process, does not apply to it. In Icelandic, there are quite a few nouns in -andi, almost all derived from verbs, e.g. leikandi m. "actor11 , derived from leika "to act", VerQ.andi f. the name of one of the norns, from verQ.a "become 11 (J6hmi.nesson 1927:18-19). Nouns in this group that are masculine (i.e. the majority) and admit of both the singular and the plural, are inflected as leikandi in (1), q.v. (Gu4- mundsson 1922:77). 1 My thanks are due to Stefful Karlsson. M.A. and to Dr. Magnus P~tursson for their help with some of the examples, and to Miss Margaret G. Davis for the correction of my English. The printing process used for the present number of the journal has made it necessary to disregard the usual rules for dividing words at the end of a line. 245 (1) N. Sg. leikandi Pl. leikendur G. leikanda leikenda D. leikanda leikendum A. leikanda leikendur It can be seen from (1) that the derivational suffix of these nouns is -and- in the singular, -end- in the plural. The -end- of the plural displays a peculiarity: if it is immediately preceded by k(k) or ~), the sounds that these graphemes stand for are pronounced velar (GuQfinnsson 1946:41). Examples: leikendur (Southern [-k-], Northern [-kh-J , huggendur [-k:-], eigendur [-_.l'~]. The expected pronunciation would be palatal ([eh, c, jJ ): other words of the same age as the leikendur type2 display the palatal pronunciation of k(k), ~) before ~ (Ofeigsson in Blondal 1920-24:XXI-III), e.g. kefja [eh-] "suffocate", akkeri [-hc-J "anchor", l.p.sg. ~ [c-J of ganga "go". (I have not any OLD examples of V~ or of ~ out- side the leikendur type.) Thus palatalisation of velar .obstru- ents HAS taken place generally before ~' just as before Old Icelandic 2:,, ,!, z, f, ~' ~, ~, ei, ::z, ..ii.' and this is the generally accepted opinion among the scholars (Noreen 1923: 190). The fronting of the velar obstruents was due to the application of a palatalisation rule, except in the cases in which the palatality of the obstruent had been lexicalised.3 The purpose of the present paper is to suggest a historical ex- planation f or the lack of palatalisation of the root final obstruents in the leikendur type. 2 The expression THE LEIKENDUR TYPE refers to the plural forms of such nominalised mascuiine present participles in -andi as contain a velar obstruent immediately in front of t~ derivational suffix -~-· 3 There is evidence for progressive palatalisation in old Ice- landic as well (Noreen 1923:191). This matter will be dis- regarded in what follows as irrelevant to the issue at hand. 246 I suggest that the lack of palatalisation of the root final velar obstruents in the leikendur type is due to the phonologi- cal universal which limits the application of the so-called (non-automatic) neutralisation processes to the so-called de- rived inputs. This universal was proposed by Kiparsky 1973, and will be called Kiparsky's universal in what follows. Here I repeat my description of the universal given in Orešnik 1979: 225-26 almost verbatim: Kiparsky 1973 formulated the phonological universal (2a) and an alternative stronger statement (2b), qq.v. (2) (a) Neutralization processes apply only to derived forma. (b) Non-automatic neutralization processes apply only to derived forms. To illustrate the universal, I shall briefly repeat one of Kiparsky' s own examples, namely the Finnish rule t-+ s / ~i. This rule/process is neutralising in the following sense: it creates an output, namely ~ before _!, which is also present in the input of the rule, i.e. which exists at the stage of deri- vation immediately preceding the application of the !-to-~ rule. (Example of underlying ~ before _!: nousi impf. of nousta "stand up".) The rule applies (a) across a morpheme boundary (e.g. /halut+i/-halusi "wants, i.nipf.", (b) morpheme-internally, with derived _! (e.g. /vete/-/veti/-vesi 11water"). The rule does NOT apply (c) morpheme-internally, with underlying ,! (e.g. /koti/-koti "home"). To account for this state of affairs, Kiparsky introduced the notion DERIVED FORM or DERIVED INPUT, which he defines as follows: (3) I will refer to an input which is created either by combin- ing morphemes through derivation or inflection, / ••• / or by applying a phonological rule, / ••• /asa DERIVED input. He then combined the notions NEUTRALIZATION RULE/PROCESS and DERIVED FORM/INPUT into the universal (2a). The universal 247 predicts that the ,!-to-~ rule applies in /halut+i/, where the crucial substring has been created through inflection (the form crucially involves an inflexional morpheme boundary), and in /veti/, where the crucial substring has been created by apply- ing a phonological rule; it does not apply in koti, whose ti is nota derived input in the sense of (3). The discussion of the notion NON-AUTOMATIC added in (2b) will be postponed until the end of the paper. In Orešnik 1979:230 I proposed a modification of Kiparsky's universal: it is only to strings crucially involving an IN- FLEXIONAL morpheme boundary and to strings created by applying a phonological rule that neutralisation processes apply. This excludes strings crucially involving WORD-DERIVATIONAL morpheme boundaries from the domain of the neutralisation processes. 4 4 Apparently contradicting my claim that no (non-automatic) neutralisation rules operate across a derivational morpheme boundary are those many instances of derived words in Ice- landic and other languages that display what at least on the face of it are the effects of the application of (non-auto- matic) neutralisation rules at crucial places. Such counter- examples are to be ex:plained away in alternative ways, some of which are listed here (Orešnik 1979): (a) In addition to phonological rules which apply to derived inputs (in my sense of the term), there are word-derivational rtlles (which are partly repetitions of certain phonological rules) whose domain of application is limited to single word- derivational processes. To put it slightly differently, some counterexamples are due to analogy with existing examples of the same type. (b) Some counterexamples may be due to the application of phonological rules at a tirne when those rules were not yet (non-automatic) neutralisation rules. After that, the changed forms were passed from generation to generation of speakers in toto, because the speakers treated those words as un- črerived. (Notice, however, that failure to recover the underlying shape is extremely unlikely in the nominalised present participles, because of the -and-/-end- alternation in the infle::f(ion.) - -- ( c) Some counterexamples can be deliberate creations of lingu- istically minded persons who consciously applied a phono- logical rule to that subpart of a new derived lexical item crucially involving a word-derivational morpheme boundary. (d) Some counterexamples may be spelling pronunciations of wide acceptance. 248 My modification of Kiparsky's universal was necessitated by the Icelandic word slafneskur "Slavic". This lexical item contains the consonantal cluster ~ [vnJ. This cluster is otherwise not found outside Icelandic compound words, because there is a phonological rule in the language changing /vn/ to /pn/, e.g. strong nom. pl. m. hrifnir [-pn-J , of hrifinn [-v-] "impressed", has the following derivation: ( 4-) vowel syncope vn--+pn (h)riv+In+Ir o (h)rivnir . (h)rip:nir " The vn--+pn rule is neutralising: it creates consonant clusters that exist in the language also independently of the vn--+pn rule. E.g. there is ~ [-pn-J "avenge", whose [-pn-J there is no reason to derive from /vn/, therefore it is here consjdered lexicalised as /pn/. (Historically speaking, the [pn] of hefna is ultimately due to the application of an older version of the vn--.pn rule, viz. vn--+bn (P6r6lfsson 1925:XXVIf.), at a tirne when that rule was not yet neutralising, i.e. when there were not yet any [bn] in the language.) The vn--+pn rule does not apply to slafneskur, although this word has a morpheme boundary between ! and !!; (seeing that it is a derivative of Slaf-i "Slav": /slav+nesk+Yr/), and its /v+n/ is a derived input in the sense of Kiparsky, see (3) above. To account f or this apparent exception I suggested in Orešnik 1979 that strings crucially involving WORD-DERIVATIONAL morpheme boundaries do not eo ipso constitute derived inputs, and there- fore neutralisation processes do not apply to them. Cf. foot- note 4-. This modification of the notion DERIVED INPUT accounts for the lack of application of the vn--+pn rule in slafneskur and, it turns out, for the lack of application of the palatalisation rule in the leikendur type. The situation in the leikendur type is comparable to that of slafneskur. Each deep phono- logical representation of the words of the leikendur type 249 contains a root final velar obstruent f ollowed by a word-deri- vational rnorpherne boundary, which is in turn followed by /e/, e.g. Southern /leik+end/. At the tirne that the palatalisation rule still applied before all front vowels (Noreen 1923:190), although it was already a neutralisation process,5 the pala- talisation rule should have applied to the root final velar obstruent and turn it into a palatal, according to the original formulation of Kiparsky's universal. This rnust be avoided. My rnodification of Kiparsky's universal rernedies the situation: the string, root final velar obstruent + word-derivational rnorpherne boundary + /e/, is nota derived input, consequently the neutralising palatalisation rule should not apply to it. It does not, and this is the correct result. Since when has the palatalisation rule been a neutralisation process? Answer: Since the tirne when palatal obstruents came into being that were no longer analysed as due to the appli- cation of the palatalisation rule. The change of the palatalis- ation rule into a neutralisation process rnay thus have happened when the rule was still produeing subphonernic palatals only (i.e. when the palatals occurred autornatically before the front vowels and .ši_, and only there; J6hannesson 1923:14-9,· 151). Ergo the palatalisation rule rnay have been a neutralisation process since pre-historical Icelandic. If this is true, we expect not to find root final palatal reflexes in the leikendur type even in the oldest texts. Unfortunately this expectation carmot be verified, because the orthography of the oldest texts did not mark palatalisation of velar obstruents (Noreen 1923:190). However, roughly since the fourteenth century the palatal obstruents have been optionally indicated in at least the following ways bef ore front vowels: 5 The modem Icelandic palatalisation rule has a more limited domain of application, or - to put it differently - numerous exceptions. For a formulation of the modem Icelandic pala- talisation rule see Orešnik 1977, where it is assumed that now the palatalisation rule fronts velar obstruents only if they are immediately f ollowed by an inflexional morpherne boundary + /I/. 250 (1) An 2: was inserted between the grapheme for the correspond- ing velar and the immediately following grapheme for the front vowel, except ~, i:_, e.g. kiefia for kefja (J6hannesson 1923: 149' 151). (2) An expected ~ [jJ was omitted, e.g. seia for segja (Bandle 1956:128f.). (3) An expected ~ [j] was spelled 2:, e.g. seiia for segja (Bandle 1956:128f.). (4) The diphthongisation of monophthongs was indicated before ~ [j], e.g. seigia for segja. Historically, the diphthong- isation followed the previous palatalisation of [t] to [j] in the chronology (Bandle 1956:46) • . (5) The inverted spelling E.:h for etymological ~ was used, e.g. nygia for nyja (Noreen 1923:217). The phenomena listed sub (1-5) above have not been observed in the leikendur type: I am not aware of any spellings such as leikiendur for leikendur, eiendur for eigendur, veiendur or veiiendur or veigendur for vegendur, heygendur for heyjendur. On the basis of this negative fact-finding and on the basis of the modern pronunciation of the leikendur type it must be assumed that the root final velar graphemes in the leikendur type have stood for velars, not for palatals, since the begin- ning of the (optional) explicit notation of the palatals. 6 What the state of affairs was even earlier can be only guessed at.7 6 The velar pronunciation of the root final segments in the leikendur type is here assumed to be essentially older than the velar pronunciation of ~2{), _g_(g) before ~in some other words (e.g. froken, orgel, kelST}. 7 I know of two lines of Scaldic poetry whose inner rhyme in- vol ves the ~ of ei5endur and the reliably velar ~ of fa~r: Katrinardr~pa 15 eigendr tokv idran fagra, and 50 eigen rna firi idran fagra (J6nsson 1912-1°5:519, '.)2'6). However,tnese exampleS-are late (J6nsson ibidem dates the poem to the fourteen~h century; Kalund 1894:128 dates its manuscript, AM 713 4 , to the first half of the sixteenth century) and therefore not of any use for our discussion. 251 That the morpheme boundary bef ore -end- in the words of the leikendur type, and generally in the nominalised present parti- ciples, is word-derivational (J6hannesson 1927:18f.), not inflexional, follows from (1) the fact that only some verbs have nominalised present participles made from them: the formation of the nominalised present participles is not at all so predictable as the forma- tion of those verbal forms whose membership in the inflexional paradigms of verbs is beyond doubt; 8 , .. (2) the fact that the meaning of the nominalised present parti- ciples is in general not predictable from the meaning of the verbs from which they are derived. Rather, the semantic re- lationship between the verbs and the corresponding nominalised present participles is that usually observed between verbs and the corresponding nomina agentis. Example: leikandi does not cover the whole semantic range of the verb leika (it does not mean "one who does the action of leika in whatever of its meanings"), but only one meaning of ~' namely "act (in the theatre, etc.) 11 • The correct translation of leikandi is "actor" (Zo~ga 1951 s.v.). The circumstance that -and-/-end-, now established·as a de- rivational suffix, changes its form in the inflexional paradigms of the nominalised present participles, is slightly atypical (the typical case being that the derivational suffix remains basically unchanged throughout the inflexion of the word whose part it is), but far from unique. A comparable case is the Classical Greek synchronically unmotivated alternation between the derivational suffix variants -nu- og -~- in the present and imperfect tenses of the ve:rbs in -nu-mi, e.g. deiknUili "I point". Quite common is the extreme alternation, in inflexional 8 In dictionaries of modem Icelandic, the nominalised present participles have their own lexical entries or subentries. 252 paradigms, between a derivational suffix and its absence, e.g. Serbo-Croat sg. grad:an-in- "citizen", pl. gradan-. M:y explanation of the pronunciation of the root f inal segments in the leikendur type is based on the assumption that the ~ of -~- has not been produced by a rule (say, by the i-umlaut rule) from ~ at least since the palatalisation rule assumed the character of a neutralisation procesa. For if -end- were syn- chronically from -~-, -end- would be a derived input in the sense of (3) and in my sense, and the palatalisation rule should apply before its ~· Since I cannot prove that -~- is not from -~- synchronically speaking, the above assumption has to be listed among the presuppositions of my explanation. It has been suggested in passing by Arnason (1978:188) that the lack of the palatalisation of the root final velar obstru- ents in the leikendur type can be explained as due to the ana- logical pressure of the root final velar obstruents of the singular case forms (and, I add, in the Icelandic of the past centuries also of the genitive plural in -anda and of the dative plural in -ondum) upon the one-time root f inal palatal of the leikendur type. While this explanation cannot be discarded, two argumenta can be adduced against it: (1) The assumed pressure of the velar on the palatal is un- parallelled in Icelandic, although inflexional paradigma con- taining the alternation palatal II) velar abound (for examples see Orešnik 1977); the normal case is that the alternation palatal~ velar remains intact regardless of the numerical relationship between the two types of segmenta. (For instance, in the in- flexion of the weak masculine nouns such as skagi "promontory", the palatal reflex occurs in one case form, the nominative singular, only, whereas the remaining seven case forma contain the corresponding velar, skaga(r), skogum, yet there is no tendency to replace the palatal of the nominative singular by 253 o the velar of the remaining case forms.)/ (2) The explanation of the velar pronunciation of the root final segment in the leikendur type by analogy is ad hoc. On the whole, ad hoc explanations not supported by arguments should be fallen back on only if no principled explanation for the same range of data is available. A principled explanation of this kind HAS been suggested in the present paper. Point (2) above is also valid for the alternative suggestion made in passing by Arnason (1978:188) that the lack of pala- talisation in the leikendur type is ascribable to a special kind of boundary before -end-. Finally, I have to answer the question as to why the blocking e.f.fect of my modification of Kiparsky's universal is not seen in a greater number of derived words than it is. The answer is given in footnote 4, and the following can be added. If Kiparsky's formulation (2b) is tpken as a starting-point for 9 If anything it is the palatal segment that gets analogically extended at the expense of the velar one. Cf. the weak mas- culine noun far]?egi "passenger", whose oblique cases should contain [.~r] , but are now at least optionally pronounced wi th [eij] (Blondal 1920-24 s.v.). Another example of the same kind is the nom.pl. Vij (of hig "battle") adduced in Helgason 1977:405 from a seventeent century manuscript. The form must be due to the analogical extension of the root in tRgi4, vigin, vigi. The fact that it is only e palatal segment that gets ana- logically extended, never the velar one, can be understood in the light of the fact that all inflexional paradigms display- ing such analogy contain palatal segments followed by /+I/ only (where /+/ is an inflexional morpheme boundary). Pala- talisation before /+I/ is an active process even nowadays, and any attempt at introducing the velar segment before /+I/ ends with the palatalisation rule applying to it. Since in old Icelandic palatalisation before /e/ was also an active process, the same result would have obtained in the leikendur type if its velar segment had been introduced from the singular at that tirne. 254 the modification of the universal, the notion NON-AUTOMATIC can be invoked to explain why there are no examples of lacking pala- talisation rule application across derivational morpheme bound- aries in the not so few _!-initial derivational suffixes. Ki- parsky defined as AUTOMATIC those rules having no exceptions. This definition has been questioned, e.g. by Stampe in Kiparsky 1976:184-85; Stampe advocates the equation, automatic rule = natural rule, in the sense of his own Natural Phonology and of Dressler 1977; that is, rules are more or less natural (rather than natural or unnatural), thus more or less automatic. Also, some parts of a rule can be more natural than others. That Kiparsky's definition leads to spurious results can be shown with the Icelandic palatalisation rule. That rule has only one systematic exception, as far as I know, pointed out by Mag- n-6.sson 1959:15: saggi, laggi [-k:-] , which are sometimes used in Vestfir4ir instead of sagg4i, lagg4i = Standard Icelandic sag.Qi, lag4i, preterite of segja 11say 11 , ~a "lay". Is this single exception really enough to change the palatalisation rule from automatic to non-automatic, with the potent conse- quences that this entails? It seems to be more promising to adopt Stampe's line mentioned above, and assert that palatalis- ation before /I/ is more natural/automatic than, sa:y, palatal- isation before /e/. The advantage of this assertion is that it makes the effects of my modification of Kiparsky's universal predictably absent before /I/ immediately preceded by a deri- vational morpheme boundary (because palatalisation is automatic before such /I/) , but predictably possible before / e/ immedi- ately preceded by a derivational morpheme boundary (because palatalisation before /e/ is non-automatic, more precisely, less automatic, than palatalisation before /I/). The fact that the palatalisation rule has lost some of its force on the wa:y from old to modem Icelandic (Arnason 1978: 187) does not su:f:fice to explain the leikendur type, at least not under the assumption made in the present paper that the lack of palatalisation in the type is old, that is, from the time that the palatalisation rule was at its peak. 255 There are quite a few modem words containing a velar obstruent before jej. For examples see Or~šnik 1977. Further I can mention that Stef§n Ka:rlsson (per litteras, 1977) pronounces Eggert, ekkert, Rakel with a velar obstruent\ in his youth he heard velar s. in the geitin of ki4a-ki4a-ki4a-ki4, koma geitin call to goats, in Fnj6skadalur (Su4ur-Pingeyjars~sla, North Iceland). These examples are different from the leikendur type in that they do not involve derivational morpheme boundaries. For a suggestion about how such cases are to be treated in modem synchronic grammar, see Orešnik 1977 (and footnote 5 above). The historical explanation, however, is not clear. Most probably the velar pronunciation of the crucial segments in these examples is much younger than the similar phenomenon in the leikendur type (cf. footnote 6). It cannot be understood in the .light of the palatalisation rule's gradually having had its domain.of application limited in historical Icelandic (Ama- son 1978:178), because the improbable implication would then be that words like Eggert, ekkert, Rakel have always contained an underlying velar opstruent although such an obstruent never appears in the surface representations of those words. My tentative guess is that Icelandic now possesses a phonological rule which depalatalises palatal obstruents before jej, and that that rule is especially potent in substandard Icelandic. To sum up, the historical development of the palatalisation rule has been as follows. For a time the rule was at its peak (and the leikendur type existed already then), then the rule began losing its force (Arnason 1978:187), limiting its domain of application to velar obstruents before j+Ij (Orešnik 1977), so that naturalised words do not any longer, except by fiat, palatalise velar obstruents before jej. (Thus arose orgel, fr6ken, etc.) Fip.ally, an automatic depalatalisation rule came into being that depalatalises palatal obstruents before jej, and maybe bef ore some other vowels (but only extremely rarely before jij, cf. sa.ggi, la.ggi above). The depalatalisation rule has not yet reached all the palatal obstruents before jej\ 256 for instance, it has only exceptionally operated upon palatal obstruents before accented /e/. The standard language has so far ignored the results of the operation of the depalatalisat~on rule. R E F E R E N C E S: Arnason, Kristjful. Palatalisation in modem Icelandic. Lingua 46, 1978, 185-2?3. Bandle, Oskar. Die Sprache der Gu4brandsbiblia. Bibliotheca arna- magnooana 17. Copehhagen, 1956. Bl5ndal, Sigf'fi.s. !slensk-dBnsk or4.ab6k. Reykjavik, 1920-24. Dressler, Wolfgang u. Grundfragen der Morphonologie. Wien, 1977• GuQfinnsson, BjBrn. M~ltzkur I. Reykjavik, 1946. GuG.mundsson, Val~r. Islandsk Grammatik. Islandsk Nutidssprog. Copenhagen, 1922. Helgason, J6n. T6lf ann~lagreinar fr! myrkum !rum. SjBt1u rit- ger4ir helgaaar Jakobi Benediktss;ypi 20. jtl.11 1977, 400-18. Reykjavik, 1977• J6hannesson, Alexander. !slenzk tunga i fornBld, vol. 1, Reykja- vik, 1923. -----· Die Suffixe im Isl!:tn.dischen, Arb6k H~sk6la !slands. Halle (Saale), 1927. J6nsson, Finnur, ed. Den norsk-islandske skjaldedigtning, A I-II. Copenhagen, 1912-15. Kiparsky, Paul. Phonological representations. Three dimensions of linguistic theo:ry (ed. o. Fujimura), 1-136. Tokyo, 1973. -----· Abstractness, opacity, and global rules. The ordering and application of phonological rules (ed. A. Kout- soudas), 160-86. The Hague, 1976• KSlund, Kristjan. Katalog over Den arnamagnE.eanske handskrift- samling, vol. 2.Copenhagen, 1894. Magntiss:on, Asgeir-BlBndal. Um frambuN.inn rd, gd, fd. !slenzk tunga 1, 1959, 9-25. 257 Noreen, Adolf. Altisli:lndische und altnorwegische Gramrnatik (Laut- und Flexionslehre) unter BerUcksichtigung des Urnordischen. 4th edition. Halle (Saale), 1923. Orešnik, Janez. On the modem Icelandic palatalization rule. Dialectology and sociolinguistics. Essays in honor of Karl-Hampus Dahlstedt, 137-45. Umea, 1977· -----· On the pronunciation of modem Icelandic rovl(a) and slafneskur. Islenskt m~l 1, 1979, 225-32. ~6r6lfsson, Bjorn Karel. Um islenskar orQmyndir ~ 14. og 15. old og breytingar Peirra ur fornmAlinu. Reykjavik, 1925. Zo~ga, Geir T6masson, Islenzk-ensk or4ab6k. 3rd edition. Reykja- vik, 1951. Povzetek ODSOTNOST PALATALIZACIJE PRED -END- V MNOŽINI ISLANDSK]H SUBSTANTIVIZIRANIH SEDANJIŠKIH DELEŽNIKOV KOT LEIKANDI V stari islandščini sta se ~ in ~ pred svetlimi samoglas- niki izgovarjala na trdem nebu. (Trdonebnost je razen v leksi- kaliziranih primerih podeljevalo t.i. palatalizacijsko pra- vilo.) Izjema so bile (in so) množinske oblike odglagolskih samostalnikov moškega spola na -~: leikandi "igralec", množ. leikend-, z mehkonebnim.!• Palatalizacijsko pravilo torej tu ni delovalo, in sicer po mnenju pisca zaradi fonološke univerza- lije P. Kiparskega, ki pravi: "(Neavtomatični) nevtralizacijski procesi delujejo samo na izpeljane oblike. 11 To univerzalijo je delno spremenil Orešnik 1979, tako da je iz pojma izpeljana oblika izločil nize segmentov, ki vsebujejo besedotvorno mor- .femsko mejo. in bi imeli -0b njej doživeti fonološko spremembo. V leikend- in drugih takih primerih narekujejo morfemsko mejo med~ in -~- besedotvorne razmere, in čez to mejo torej v stari islandščini palatalizacijsko pravilo ni delovalo, ker je bilo (neavtomatični) nevtralizacijski proces. 258 Fonološkim spremembam, ki jih opažamo ob besedotvornih morfem- skih mejah, so vzrok 1) besedotvorna pravila (tj. analogija po starejših primerkih istega besedotvornega vzorca), 2) fonološka pravila, ki niso (neavtomatični) nevtralizacijski procesi, 3) fiat besedotvorcev, 4) izgovarjava po črki. 259