103 Marko Simonović * UDK 811.163.6'366'373.611 University of Graz/ DOI: 10.4312/linguistica.60.1.103-117 University of Nova Gorica, Austria/Slovenia CATEGORIES, ROOT COMPLEXES AND DEFAULT STRESS: SLOVENIAN NOMINALIZATIONS REVISITED 1 1 INTRODUCTION Standard Slovenian is a language with lexical prosody in nouns, verbs and adjectives (the default prosodic pattern assigned when no lexical prosody is present or available is discussed in section 3.1). Further, derivational morphemes which behave as nominal- izers, verbalizers and adjectivizers can have stress-shifting effects. The interaction between lexical stress and the structure of words in Slovenian is discussed in Marvin (2003), where a detailed analysis of stress assignment in Slovenian nominalizations is presented. The nominalizations analysed by Marvin fall into two broad categories in terms of prosody: the nominalizing affix either leaves the prosodic pattern of the base unaffected or it causes the word stress to shift to a new position. In some cases, the same nominalizing affix can appear in both prosodic types, giving rise to minimal pairs that differ only in stress. It is exactly such cases that make a struc- tural analysis of stress assignment all the more necessary, since the problem cannot be solved through lexical specification on either the base or the affix entailing multiple suspiciously homonymous affixes belonging to the same category. The prosodic dichotomy between affixes is illustrated in (1). The nominalizations in (1a) inherit the prosodic pattern of the base, but those in (1b) have a different prosodic pattern from the base. (1) Deadjectival and deverbal nominalizations in Slovenian a. Stress-preserving nominalizations Nominalization Base Nominalization Base stár-ost stár anketíran-ec anketíran old-ost interview.pass .ptcp -c interview.pass .ptcp ‘oldness’ ‘old’ ‘interviewee’ ‘interviewed’ b. Stress-shifting nominalizations Nominalization Base Nominalization Base star-óst stár napadál-ec napádal star-ost attack.lptcp -c attack.lptcp ‘old age’ ‘old’ ‘attacker’ ‘attacked’ * rkicma@gmail.com 1 I acknowledge the financial support by the Austrian Science Fund FWF (grant I 4215) and the Slovenian Research Agency ARRS (grant N6-0113). Linguistica_2020_1_FINAL.indd 103 Linguistica_2020_1_FINAL.indd 103 22. 12. 2020 15:07:37 22. 12. 2020 15:07:37 104 Note that only the deadjectival nominalizations (stár-ost and star-óst) form a true minimal pair, as the deverbal nominalizations involve different participles. This is why Marvin’s analysis will first be summarised based on the pair stár-ost vs. star-óst. The crux of the analysis is that the stress pattern of complex words is negotiated as a function of Phasal Spell-out. An affix can only influence the prosody of the root if it is in the same phase with that root. This means that stress-preserving nominaliza - tions contain several phases and the potentially stress-affecting affix is merged in a different phase from the root, i.e., it arrives ‘too late’ to change the spell-out of the root. For the pair stár-ost vs. star-óst, this means that the former is a truly deadjec- tival nominalization containing an adjectival and a nominal head, whereas the latter is a root nominalization (with no adjectival head). As illustrated in the trees in (2), in both stár-ost and star-óst the only two pieces of structure that have phonological content are the root √star and the nominalizer -ost. However, while in the dead- jectival nominalization stár-ost they are separated by a phase boundary (aP being a phase), in the root nominalization star-óst they are adjacent. Assuming that -ost has underlying stress and is able to shift the stress from the adjacent head yields the right result. As Marvin points out, the difference between the two structures plays a role not only at the PF, but also at the LF interface. Indeed, in such pairs, the true deadjectival nominalizations tend to have a more transparent interpretation: stár-ost means the property of being star ‘old’, whereas the root nominalizations tend to have rather idiomatic meanings. (2) stár-ost vs. star-óst: The analysis in terms of Marvin (2003) a. nP 3 n aP ost 3 a √ STAR b. nP 3 n √ ost STAR As far as the deverbal nominalizations in (1) are concerned, it has already been noted that they do not form a perfect minimal pair, as they incorporate different partici- ples. The logic of their analysis is therefore somewhat different, as the structure of the nominalization depends on the structure made available by the participle. As Marvin convincingly shows, only the passive participle has a truly adjectival structure, whereas the l-participle does not introduce a categorial head. This adjectival vs. non-adjectival contrast leads to a dichotomy similar to that between stár-ost and star-óst discussed above: the passive-participle nominalizations contain an adjectival phrase and there- fore the nominalizer cannot be stress-shifting, whereas in the case of the l-participle nominalizations all pieces which have phonological content are in the same phase and the nominalizer can impose its own prosodic pattern. Linguistica_2020_1_FINAL.indd 104 Linguistica_2020_1_FINAL.indd 104 22. 12. 2020 15:07:37 22. 12. 2020 15:07:37 105 One important assumption necessary for this account to work is that when catego- rial affixes and roots have conflicting prosodic requirements, the affixes win. This is not explicitly spelled out in Marvin (2003). However, since categorizers are heads, any model that assumes the phonological privilege of syntactic heads predicts this result. Such models have indeed been proposed, see Revithiadou (1999) for a cross-linguisti- cally grounded proposal. 2 PREDICTIONS OF THE MODEL AND UNEXPECTED GAPS Marvin’s analysis makes several predictions concerning the prosody of nominaliza- tions. In this section I will consider Slovenian data against the background of these pre- dictions and establish a list of empirical issues which need to addressed when revisiting Marvin’s model. Reasoning from the type of base, the predictions are as follows: A1) If the base is a root, we expect differences between stress-shifting and stress- preserving nominalizers (reflecting the lexical specification of the affix). A2) If the base is a root accompanied by a non-categorial affix (as with l-participle bases), the situation should be as with roots, i.e. we expect differences between stress-shifting and stress-preserving nominalizers (reflecting the lexical speci- fication of the affix). A3) If the base has a category, we should always observe the stress-preserving be- haviour of the nominalizers (as the lexical specification of the affix cannot be enforced across the phase boundary). Some of the predictions above are easier to test than others. A1 is extremely hard to test because a root cannot easily be distinguished from a combination of a root and a phonologically mute categorial head. On the other hand, A2 can be tested using deriva- tions which incorporate the l-participle. Here, the results are quite unexpected: all af- fixes, nominal and adjectival, which combine with the l-participle cause a stress shift. This is illustrated in (3) using the family of words which incorporate the l-participle méril ‘measured’. (3) Derivations incorporating méril ‘measured’ Nouns: meríl-ec meríl-ka meríl-o measure.lptcp -c measure.lptcp -ka measure.lptcp -o ‘timekeeper’ ‘timekeeper’ (fem.) ‘measure, scale’ Adjectives: meríl-n-a meríl-sk-a measure.lptcp -n-fem measure.lptcp -sk-fem ‘measuring’ (fem.) ‘related to measuring’ The lack of stress preserving derivations with l-participles is the first empirical issue (EI1) which I seek to account for in this article by building on Marvin’s model. Linguistica_2020_1_FINAL.indd 105 Linguistica_2020_1_FINAL.indd 105 22. 12. 2020 15:07:37 22. 12. 2020 15:07:37 106 EI1: L-participle derivations are always stress-shifting. A3 can be tested by isolating those bases which have an overt categorial head. If we remain focused on deadjectival ost-nominalizations and participle nominalizations, we expect the following: a) Derived adjectives (i.e. those adjectives which contain an overt adjectivizer) should never derive ost-nominalization with the stressed ost. b) Passive participles should never derive nominalizations with a stress-shift. In order to test the first prediction, I extracted all ost-nominalizations which had more than 300 attestations in the Slovenian national corpus Gigafida. I annotated the 1,231 items obtained for their stress pattern (whether -ost is stressed) and the visibility of adjec- tivizing morphology (whether the base contains a common adjectivizing suffix). Table 1: Stress and the visibility of the adjectivizer in ost-nominalization Stress Stress not on -ost Stress on -ost Visibility of an adjectivizer Adjectivizer visible Adjectivizer not visible Adjectivizer visible Adjectivizer not visible Example oséb-n-ost person-adj-nom ‘personality’ pogóst-ost frequent-nom ‘frequency’ last-n-óst property-adj-nom ‘feature’ nor-óst insane-nom ‘insanity’ Number 1079 126 4 22 Percentage (within the stress type) 90% 10% 15% 85% The first observation to be made based on the above data is that the relative frequen- cies observed fit the model quite well, which means that what we need to handle are essentially exceptions. The only ost-nominalizations which contradict the model are the four nominalizations with a stressed affix -ost which cannot be analysed as root nominalizations since they contain clearly derived adjectives. Beside last-n-óst ‘fea- ture’, these are skriv-n-óst ‘secret’, dolž-n-óst ‘obligation’ and uč-en-óst ‘erudition’. While both their form and their meaning clearly point in the direction of idiomatized items, it is still an issue how we represent such items, as introducing new roots (√lastn , √skrivn etc.) just to fit the data seems a counter-intuitive move. EI2: There are (clearly exceptional) items in which the nominalizer -ost is stressed, although there is a visible adjectival head under it. The nominalization uč-en-óst ‘erudition’ actually contains a passive participle uč[έ]n ‘learned’ and is therefore an exception relevant for the next prediction we are Linguistica_2020_1_FINAL.indd 106 Linguistica_2020_1_FINAL.indd 106 22. 12. 2020 15:07:37 22. 12. 2020 15:07:37 107 considering: passive participles should never form nominalizations with a stress-shift. Just like with the previous set, we are again dealing with exceptions, which require a representational solution. The examples of stress-shifting nominalizations from passive participle are very few, but they are attested in every type of nominalization. In the ex- amples below, I use the nominalizations of the passive participle mérjen ‘measured’ in order to illustrate the regular pattern and the full list of exceptions I have identified (and which were confirmed as familiar to more than one third of the speakers I consulted). (4) c-nominalizations from passive participles a. Regular pattern (faithful prosody) b. Exceptional pattern (shifted prosody) Base Nominalization Base Nominalization mérjen mérjen-ec múčen muč[έ]n-ec ‘measured’ ‘measured one ’ ‘tortured’ ‘marthyr’ (5) ka-nominalizations from passive participles a. Regular pattern (faithful prosody) b. Exceptional pattern (shifted prosody) Base Nominalization Base Nominalization mérjen mérjen-ka múčen muč[έ]n-ka ‘measured’ ‘measured one’ (fem.) ‘tortured’ ‘marthyr’ (fem.) (6) je-nominalizations from passive participles a. Regular pattern (faithful prosody) b. Exceptional pattern (shifted prosody) Base Nominalization Base Nominalization mérjen mérjen-je míšljen mišlj[έ]n-je ‘measured’ ‘measuring’ ‘thought’ ‘opinion, thinking’ vprášan vprašán-je ‘asked’ ‘question’ This list of frequent exceptions leads us to the next empirical issue. EI3: There are (clearly exceptional) items in which the nominalization of a passive participle is stress shifting. Another issue that arises from these data is the exceptional stress-shifting behaviour of many different affixes. In the examples above, beside -ost, these are, -c, -ka and -je. Especially the last one seems to behave as a stress shifter only in the two examples quoted above (plus potentially several words with a variable stress pattern). This makes it considerably less plausible to specify the stress-shifting behaviour in their lexical entry. Therefore, we can identify the last empirical issue. EI4: Several affixes (e.g. -ost, -c, -ka and -je) act as stress-shifters in very few items. Linguistica_2020_1_FINAL.indd 107 Linguistica_2020_1_FINAL.indd 107 22. 12. 2020 15:07:38 22. 12. 2020 15:07:38 108 3 DESIDERATA FOR AN IMPROVED MODEL AND A NEW PROPOSAL If the model of stress assignment in Slovenian discussed so far is to be improved, it is crucial not to lose its many virtues. The correlation between root-selecting be- haviour and stress-shifting effects on the one hand, and adjective-selecting behaviour and stress-neutrality on the other is a robust fact confirmed by corpus data. I used the same methodology as for ost-nominalizations above on two more nominalizers the stress-shifting -ota (e.g. in lep-ota ‘beauty’) and stress-neutral -ež (e.g. in perverz-n-ež ‘pervert’). Ota-nominalizations show no clear examples of derived bases (based on 23 items), whereas ež-nominalizations have derived bases in 94% of the cases (based on 48 items). There is, however, an overgeneration problem. The original model predicts the ex- istence of root-selecting stress-neutral affixes. Recall that in this model the prosody of affixes was lexically specified and, while category-selecting affixes are necessarily stress-neutral, root-selecting affixes can be either stress-shifting or stress-neutral. Since I have not identified any stress-neutral nominalizers which generally select roots, I will present a proposal which excludes such a structure. My proposal dispenses with lexical prosody on derivational affixes altogether. In- stead, all the prosodic effects follow from the structural position in which the affix appears. Before I define the relevant positions, I first take a closer look at the prosodic effect which I have referred to as stress-shifting in the presentation of the data so far. 3.1 Stem­Final Stress as the Default in Slovenian All the data in which we found stress-shifting behaviour so far share the same stress pattern: stress is always stem-final, i.e. falling on the syllable preceding the inflectional ending. In some of the examples above, the stem-final position of stress was somewhat obscured in the citation form due to an epenthetic process in forms without overt in- flectional morphology (e.g. in muč[ɛ́]n[ə]c, where the schwa is epenthetic and absent from all other forms, which have overt inflection). In (7) I provide the dual forms of examples of all the classes considered as stress-shifting in the text so far, since the dual ending always has an overt exponent. Note that in (7) the hyphen only separates the stem from the inflectional ending. (7) a. ost-nominalizations staróst-i učenóst-i ‘old ages’ (dual) ‘eruditions’ (dual) b. derivations incorporating the l-participle méril ‘measured’ Nouns: merílc-a merílk-i meríl-i timekeepers’ (dual) ‘timekeepers’ (fem.) (dual) ‘measures, scales’ (du.) Adjectives: meríln-i merílsk-i ‘measuring’ (fem.) (dual) ‘related to measuring’ (fem.) (dual) Linguistica_2020_1_FINAL.indd 108 Linguistica_2020_1_FINAL.indd 108 22. 12. 2020 15:07:38 22. 12. 2020 15:07:38 109 c. nominalizations incorporating passive participles muč[ɛ́]nc-a muč[ɛ́]nk-i ‘martyrs’ (dual) ‘marthyrs’ (fem.) (dual) mišlj[ɛ́]nj-i vprašánj-i ‘opinions’ (dual) ‘questions’ (dual) d. ota-nominalizations lep[ɔ́]t-i groz[ɔ́]t-i ‘beauties’ (dual) ‘horrors’ (dual) As the reader can testify, in all the examples above the stress is stem-final. This is by far the most common stress pattern in Slovenian and I argue that this is the Slovenian default prosodic pattern. Essentially, this means that stem-final stress is assigned to all words in which lexical prosody is not present or not available. In order to substantiate the claim that stem-final stress is the most common one in Slovenian, Simonović and Mišmaš (this volume) isolated the 3,000 most frequent nouns, verbs and adjectives and marked the stress pattern of each category. Items that can either have the stem-final stress or another stress pattern were excluded from the count. In each of the categories the stem-final stress pattern is by far the most common one and a majority of words has this pattern. The stem-final stress pattern was encoun- tered in 63% of the verbs, 70% of the nouns and 73% of the adjectives. 3.2 Derivational Affixes as Roots Now that I have identified the stress pattern common to all the stress-shifting environ- ments, I turn to the task of identifying what these environments have in common and considering why this should lead to the assignment of the default stress pattern. As it stands now, this class of environments looks quite heterogeneous, encompassing vari- ous nominalising affixes (-ota, -ost, -c, -ka, -je) with an environment which can only be described as a combination of the l-participle and any suffix which is added to it. In a sense, following the intuition from Marvin’s proposal that stress-shifting is a phenomenon related to the root domain, I propose that what triggers the default proso- dy in Slovenian is indeed a root phenomenon, but of a kind that was not available in the theory used so far. Following Lowenstamm (2014), I assume that derivational affixes are transitive roots, some of which select categories, whereas others select other roots (see Simonović 2020) and Simonović and Mišmaš (this volume) for applications of this model to Slovenian). Lowenstamm’s model, originally proposed to resolve several is- sues in English nominalizations, predicts the existence of complex root phrases with no intervening functional structure. These structures are termed radical cores by Lowen- stamm. My central assumption concerning the prosody of Slovenian radical cores is that they always receive default prosody. The idea is that the regular algorithm that computes compositional prosody requires the presence of functional projections that can decide which morphemes win in the event of conflict (much as in Revithiadou 1999, where heads win). Due to the disruption of the usual asymmetry caused by the Linguistica_2020_1_FINAL.indd 109 Linguistica_2020_1_FINAL.indd 109 22. 12. 2020 15:07:38 22. 12. 2020 15:07:38 110 lack of functional structure, the prosody in radical cores cannot be computed composi- tionally and therefore the assignment of default stress ensues. We can now revisit our first nominalization examples stárost and staróst and apply the new model to them. My analysis of stárost is basically still that of Marvin (2003) in that stárost is a deadjectival nominalization. The technical innovation illustrated in (8a) is that -ost is now a root that selects an adjective as its complement, which naturally leads to preserving the prosodic pattern of the base. The root nominalization staróst is still a root nominalization, but it now contains two roots, √star and √ost , which form a radical core, triggering default stress. (8) a. nP 3 n √P 3 √ aP OST 3 a √ STAR b. nP 3 n √P 3 √ √ OST STAR Note that, unlike Marvin’s model, the proposed model can represent the exceptional ost-nominalizations with an overt derivational affix preceeding -ost, such as last-n-óst ‘feature’ (thereby resolving the Empirical Issue 2 identified in Section 2). In the four items like last-n-óst, n will be part of the radical core, as shown in (9). (9) nP 3 n √P 3 √ √P OST 3 √ √ N LAST While such items are representable, they are also expected to be rare, as they require a specific Encyclopedia entry. This prediction is borne out, as we only found four such items. 4 CONSEQUENCES With the new proposed model in place and having applied it to the first portion of the data, I now turn to the more general picture. Linguistica_2020_1_FINAL.indd 110 Linguistica_2020_1_FINAL.indd 110 22. 12. 2020 15:07:38 22. 12. 2020 15:07:38 111 The stress-shifting affixes we have seen so far fall into three different categories when it comes to their selectional properties: a) Affixal roots like -ota always select other roots so it seems safe to assume that they have the selectional feature [√P]. b) Affixal roots like -ost, -c, -ka, and -je select roots only in a subset of their uses, which can be seen as exceptions. These roots are then generally category select- ing, having the feature [xP], but they can be stored as part of radical cores as well. c) Combinations of the ending -l and other affixes generally shift stress. Roots like -ota are entirely unproblematic and their stress pattern follows from the structure imposed by their selectional properties, as illustrated in (10). (10) nP 3 n √P 3 √ √ OT(A) LEP The account for the other two types needs to be rather more elaborate, as there is more variation and interaction between several affixes. I address the exceptional forma- tion of radical cores in 4.1. and the interactions of √l with other affixes in 4.2. 4.1 Exceptional Radical Cores The exceptional radical cores are readily analysed as cases of storage of a simplified structure, which contains just the bare minimum of functional projections, i.e. the top- most categorizing head, without which no well-formed word can exist. Note that I am now considering the ost-nominalizations with stressed ost (e.g. staróst) as exceptions. This is justified by the quantitative data presented in Table 1, which show that less than 3% of all ost-nominalization have the shifted stress pattern. While in examples of the type staróst only one functional projection was removed from the structure, other examples involve several removed projections. In order to illustrate this, I quote a minimal pair that some of the speakers have involving two c- nominalizations of the passive participle múčen ‘tortured’. (11) Two possible c-nominalizations of múčen ‘tortured’ Base Nominalization 1 Nominalization 2 múčen múčenc-a muč[ɛ́]nc-a ‘tortured’ ‘tortured persons’ (dual) ‘martyrs’ (dual) Linguistica_2020_1_FINAL.indd 111 Linguistica_2020_1_FINAL.indd 111 22. 12. 2020 15:07:38 22. 12. 2020 15:07:38 112 Based both on the stress pattern and on meaning, it is clear that Nominalization 1 contains more functional structure than Nominalization 2. Nominalization 1 therefore also has more structure in common with the passive participle than Nominalization 2. The relevant trees for múčenec and muč[ɛ́]nec, respectively are provided in (12). (12) múčenec ‘tortured persons’ vs muč[ɛ́]nec ‘marthyr’ a. nP 3 n √P 3 √ aP C 3 a √P 3 √ vP EN 3 v √ MU Č(I) b. nP 3 n √P 3 √ √P C 3 √ √ EN MU Č(I) Note that the proposed analysis is clearly distinct from the more traditional analysis which would simply state that the muč[ɛ́]nec is stored as a single item. Crucially, there is no new root √ mučenc and the two items still contain the same roots, which are con- nected in a different structure. This subsection addressed the Empirical Issues 3 and 4 from Section 2 by providing a model which can handle the exceptions, that is items exceptionally lacking functional layers present in more compositionally built related words. In the following subsection I address Empirical Issue 1, which concerns the stress-shifting behavior of the deriva- tions that contain the l-participle. 4.2 L-participle, √l , or Both? Turning now to the nominalizations which appear to be derived from l-participles, I can already formalize what was formulated as Empirical Issue 1 in Section 2: the fact that “L-participle nominalizations are always stress-shifting.” All that needs to be speci- fied is that √l is a root-selector and the result of √l always triggering default stress is achieved. Now we can go through the contexts in which √l appears and see how plau- sible this assumption is. While a combination of √l with other affixes is very common, there actually is a class of nominalizations which show no other (overt) derivational affix on top of √l . Linguistica_2020_1_FINAL.indd 112 Linguistica_2020_1_FINAL.indd 112 22. 12. 2020 15:07:38 22. 12. 2020 15:07:38 113 One example quoted in (3) above was merilo ‘measure, scale’. Nominalizations of this type usually have the structure verbal root + theme vowel + l + o (o being a case end- ing). Marvin (2003: 108) notes that the theme vowel in some cases is not the same as the one attested in the verb. Furthermore, there are cases in which the root is not verbal at all. All three types are illustrated in (13). (13) Three types of lo-nominalizations a. verbal root + theme vowel attested with the same root + lo Nominalization Related l-participle merílo méril ‘measure scale’ ‘measured’ pisálo písal ‘writing device’ ‘written’ b. verbal root + theme vowel not attested with the same root + lo Nominalization Related l-participle barvílo bárval ‘pigment’ ‘coloured’ tolkálo tólkel ‘percussion’ ‘banged’ c. non-verbal root + theme vowel + lo Nominal Related l-participle Related noun glasbíl-o / glásb-a ‘musical instrument’ ‘music’ zrcál-o / / ‘mirror’ This variety of patterns attested in lo-nominalizations suggests a structure which is not necessarily deverbal, although it does seem to host roots which also show up in verbs more often than others. The analysis I propose is then that in (14). The inflec- tional ending -o is shown between brackets, which is a clear oversimplification: the -o is an exponent of a higher functional head (presumably K, the case head) which is not visible in the tree built so far. (14) nP 3 n √P 3 √ √ L(O) MER(I) Linguistica_2020_1_FINAL.indd 113 Linguistica_2020_1_FINAL.indd 113 22. 12. 2020 15:07:38 22. 12. 2020 15:07:38 114 I remain agnostic as to whether the root √l is the same as the morpheme at the end of l-participles, which Marvin (2003: 90) convincingly shows to be the Elsewhere allo- morph in the verbal inflection. If this is the case, the picture is very well comparable to that in the nominal domain found by Simonović and Mišmaš (this volume) and it would necessitate a more general reconsideration of the distinction between the Encyclopedia and the Vocabulary. Note, however, that the lexical prosody of l-participles (e.g. méri-l ‘measured’ versus morí-l ‘murdered’) indicates that l-participles contain a vP under the position where the Elsewhere allomorph is inserted, so that there is no radical core in these forms and no default prosody emerges. The most prominent feature of the root √l beyond lo-nominalizations is that it is very often selected by other roots, which then leads to whole families of related words as the one shown in (3) and repeated here in (15). (15) Derivations containing méri+l related to mériti ‘measure’ Nouns: meríl-ec meríl-ka meríl-o ‘timekeeper’ ‘timekeeper’ (fem) ‘measure, scale’ Adjectives: meríl-n-a meríl-sk-i ‘measuring’ (fem) ‘related to measuring’ Especially productive are the agentive nominalizations in which √l is selected by √c and √k (a ): out of 3000 most frequent verbs in the corpus Gigafida, more than 500 have nominalizations in √l +√c (-lec) or √l +√k (a ) (-lka), making this the most pro- ductive pattern for agent nominalizations in Slovenian. This being so, it may appear more plausible to assume a verbal structure under √l . While this may be an option worth exploring, there are several good reasons for sticking with Lowenstamm’s initial model in which no categories are allowed under radical cores. First, nominalizations in √l +√c (-lec) or √l +√k (a ) (-lka) resist productively combining with the frequent ver- balizer -irati, typically used in borrowed verbs (e.g. analizirati ‘analyse’, promovirati ‘promote’, etc.). Out of 125 verbs derived with -irati which figure among the 3000 most frequent verbs in the corpus Gigafida, only two have attested nominalizations of this type, and both of them are attested less than three times: kopiralec ‘copier’ and parkiralec ‘parker’. Second, there are nominalization in which √l +√c (or √l +√k (a )) is preceded by verbal-looking material which is not attested in any actual verb. Such √l +√c -nominalizations are illustrated in (16). (16) √l +√c -nominalizations lacking the base verb Nominalization Implied verb Actual verb Nominalization of the actual verb posnemovalec *posnemovati posnemati ? posnemalec ‘imitator’ ‘imitate’ ‘imitator’ obračunovalec *obračunovati obračunavati *obračunavalec ‘calculator’ ‘calculate’ obveščevalec ? obevščevati obveščati *obveščalec ‘informant’ ‘inform’ Linguistica_2020_1_FINAL.indd 114 Linguistica_2020_1_FINAL.indd 114 22. 12. 2020 15:07:38 22. 12. 2020 15:07:38 115 A remarkable feature of the nominalizations in (16) is that they all contain the se- quence ova/eva, which is readily analysed as the root √ov in combination with the theme vowel a, a pattern widely attested in actual verbs (see Simonović/Mišmaš, this volume). So, similarly to the examples in (13b) and (13c), √l selects a structure which contains a theme vowel, but this structure is not necessarily an existing verb. Before concluding this contribution, it should be pointed out that the model present- ed predicts the existence of words in which root-selecting and category-selecting af- fixes co-occur in the same structure. I have not presented any such words so far because the presentation has focused on the dichotomy between radical cores and their default stress, on the one hand, and structures which intertwined roots and functional structure with their faithful stress, on the other. However, words which have both radical cores and category-selecting affixes are amply attested in Slovenian. In (17) two such nouns are presented which are related to the verb mériti ‘measure’. (17) Derivations containing méri+l related to mériti ‘measure’ (part 2) merílnic-a merílnik ‘measuring room’ ‘measuring device’ Note that these nominalizations do not have stem-final stress, but they crucially contain a radical core (meriln) which has received the default stress. The analysis is shown in (18). (18) nP 3 n √P 3 √ aP IC(A) 3 a √P 3 √ vP N 3 √ √ L MER(I) 5 CONCLUSION In this contribution, a principled account has been presented of stress assignment in deadjectival and deverbal nominalizations in Slovenian, explicitly addressing both Linguistica_2020_1_FINAL.indd 115 Linguistica_2020_1_FINAL.indd 115 22. 12. 2020 15:07:38 22. 12. 2020 15:07:38 116 regular cases and exceptions. The analysis has been couched in the model proposed by Lowenstamm (2014), which views derivational affixes as transitive roots. The main ad- dition to the model was the proposal that parts of the structure which only contain roots with no intermediate functional structure (the ‘radical cores’) always receive default prosody. The analysis presented dispenses with lexical stress in Slovenian affixes and derives the prosodic properties of affixed words solely from their structure. The proposed model has been applied only to a limited number of derivational pat- terns, but it makes rather clear predictions as to what is to be expected in the rest of the language(s). It is therefore a worthwhile endeavor for future research to test the proposed model using additional data both from Slovenian and other languages. Finally, since the main focus has been on prosody, at least two aspects of the data analysed have received less attention than they deserve. The first is the semantics of the nominalizations discussed, which has only been addressed in passing. The second is the nature of the theme vowels in Slovenian (and beyond). For the sake convenience, I have followed Marvin (2003) in assuming that theme vowels belong to the root. It, however, remains a fact that theme vowels only play a role in analyses which relate mainly to the verbal domain. It is therefore necessary to explore further their nature and their lexical representation. It is my hope that further research will tackle these issues and bring new insights to bear on the matter. Primary Sources Gigafida. 24 February 2020. http://www.gigafida.net/. References HALLE, Morris/Alec MARANTZ (1993) “Distributed morphology and the pieces of in - flection.” In: K. Hale/S. J. Keyser (eds), The View from Building 20: Essays in Linguis- tics in Honor of Sylvain Bromberger. Cambridge, MA: MIT Press, 111–176. HALLE, Morris/Alec MARANTZ (1994) “Some key features of distributed morphol- ogy.” In: A. Carnie/H. Harley/T. Bures (eds), MIT Working Papers in Linguistics 21. Cambridge, MA: MITWPL, 275–288. LOWENSTAMM, Jean (2014) “Affixes as roots.” In: A. Alexiadou/H. Borer/F. Schäfer (eds), The syntax of roots and the roots of syntax. Oxford: Oxford Univer- sity Press, 230–259. MARANTZ, Alec (1996) “‘Cat’ as a phrasal idiom.” Manuscript. Cambridge, MA: Massachusetts Institute of Technology. MARVIN, Tatjana (2003) Topics in the stress and syntax of words. Doctoral disserta- tion. Cambridge, MA: Massachusetts Institute of Technology. REVITHIADOU, Anthi (1999) Headmost accent wins: Head dominance and ide- alprosodic form in lexical accent systems. Doctoral dissertation. Leiden: Leiden University. SIMONOVIĆ, Marko (2020) “Derivational affixes as roots across categories.” [In press.] Linguistica_2020_1_FINAL.indd 116 Linguistica_2020_1_FINAL.indd 116 22. 12. 2020 15:07:38 22. 12. 2020 15:07:38 117 Abstract CATEGORIES, ROOT COMPLEXES AND DEFAULT STRESS: SLOVENIAN NOMINALIZATIONS REVISITED A new account is presented of stress assignment in deadjectival and deverbal nomi- nalizations in Slovenian, explicitly addressing both regular cases and exceptions. The analysis is an extension of the account by Marvin (2003) and is couched in the model developed by Lowenstamm (2014), which views derivational affixes as transitive roots. The main addition to Lowenstamm’s model is the proposal that parts of the structure which only contain roots with no intermediate functional structure (the ‘radical cores’) always receive default prosody. The presented analysis dispenses with lexical stress in Slovenian affixes and derives the prosodic properties of affixed words solely from their structure. Keywords: stress, nominalizations, Distributed Morphology, default stress pattern, derivational affixes as roots, radical cores Povzetek KATEGORIJE, KORENSKI SKUPKI IN PRIVZETI NAGLAS: PONOVNO O SLOVENSKIH NOMINALIZACIJah V prispevku je predstavljen nov pristop k določevanju naglasnega mesta izpridev- niških in izglagolskih nominalizacij, ki neposredno naslavlja tako običajne kot izjemne primere. Pristop nadgrajuje analizo Tatjane Marvin (2003) z modelom Jeana Lowen- stamma (2014), v katerem so besedotvorni morfemi obravnavani kot prehodni mor- femi. Glavni prispevek k Lowenstammovemu modelu je predlog, v skladu s katerim deli strukture, ki vsebujejo le korene brez vmesne funkcijske strukture (t. i. korenski skupki), vedno pridobijo privzeto prozodijo. Predstavljena analiza tako odpravlja leksi- kalen naglas v slovenskih morfemih in razlaga prozodične lastnosti tvorjenk izključno z vidika njihove strukture. Ključne besede: naglas, nominalizacije, razpršena morfologija, privzeti naglasni vzo- rec, besedotvorni morfemi kot koreni, korenski skupki Linguistica_2020_1_FINAL.indd 117 Linguistica_2020_1_FINAL.indd 117 22. 12. 2020 15:07:38 22. 12. 2020 15:07:38