Cross-linguistic considerations on preverb stacking (with special reference to Bulgarian)*

Alessio Muro Cross-linguistic considerations on preverb stacking (with special reference to Bulgarian) The term preverb stacking (PS) designates the co-occurrence on one verbal base of two or more prefixes bearing spatial, aspectual, or quantificational meanings. The phenomenon is best known from its high productivity in the Slavic languages. However, PS is also attested in several other Indo-European branches, and it is found even in genetically unrelated and geographically remote languages. This paper will provide a first attempt at a cross-linguistic typology of PS, but it will also pay special attention to problems typical of Slavic languages (such as the interaction of PS and the aspectual value of the verb in terms of the typically Slavic perfective vs. imperfective dichotomy). Special attention will be paid to Bulgarian, where the phenomenon is especially productive.

found in different languages, paying special attention to processes of lexicalization at work between the inner preverbs and their host roots or between members of compound preverbs. We will devote special attention to Bulgarian, a language where PS is exceptionally productive. However, the comparison between the Bulgarian data and superficially similar data from other Slavic languages such as Russian will also show that PS is not just a marginal phenomenon; rather, it is important in order to delineate the aspectual profile of a particular Slavic language, since superficially similar preverb strings trigger different syntactic configurations in terms of the perfective vs. imperfective dichotomy.

Non-Slavic Indo-European languages
Apart from the innermost preverbs (which can function as telicity operators), preverbation contributes a semantic content which in most cases is spatial in nature. The only non-spatial adverbial meaning found in several Indo-European languages outside of the Slavic branch is, to my knowledge, 'together' (cf. Gothic, Lithuanian, Greek, Sanskrit). Let us consider the following sentences from the Gothic and the koiné Greek versions of the New Testament (John 18:15) 1 : (1) sa=h þan siponeis … miþ-inn-ga-laiþ (Gothic) that:m:nom:sg=and then disciple:nom:sg … together-in-off-went:3sgs miþ iesua in rohsn þis gudjins with J:dat into palace:acc:sg the:m:gen:sg high.priest:gen:sg '(And then) that disciple … went in with Jesus into the palace of the High Priest.' In the above example, the form miþ-inn-ga-laiþ shows the stacking of no less than three preverbs: the innermost one, ga-, appears to be lexicalized (we have forms like galeiþan 'start, depart' or with other preverbs, but we never find *leiþan alone in the Gothic corpus) 2 . The other two preverbs, miþ-and inn-, on the other hand, show totally different syntactic properties: they are both doubled by independent PPs, headed by their corresponding prepositions (miþ and in, respectively). 1 The following abbreviations are used in morpheme glosses: 1, 2, 3 -persons of verbal agreement; A -agent; acc -accusative; aor -aorist; cis -cislocative; compl -completive; dat -dative; decl -declarative; distr -distributive; f -feminine; gen -genitive; imp -imperative; inf -infinitive; ingr -ingressive; iter -iterative; m -masculine; narr -narrative; nom -nominative; pl -plural; refl -reflexive; S -subject; sg -singular; stat -stative. Superscript P,I in Slavic forms indicate perfective and imperfective aspect. 2 Cf. German mit-be-gleiten, where -gleiten in turn comes from ge-leiten 'to escort '. Things are partly different in the Greek version: (2) sun-eis-ēlt h en tō̃i iēsoũ eis tḕn aulḕn (Koiné Greek) together-to-go:aor:3sgs the:m:dat J:dat to the:f:acc palace:acc toũ ark h ieréōs the:m:gen high.priest:gen Here the stack is made up of only two preverbs: only the inner one, eis-, is doubled by a PP headed by the corresponding preposition eis. The comitative preverb sun-has no double; nevertheless, it seems to govern the dative case shown by the following NP. PS seems to have been productive in Ancient Greek, already from the Homeric stage. Imbert (2008)  Note that the ordering of the preverbs in (3c) differs from ex. (2) above in that eis-comes outside of ana-, whereas in (2) it comes inside of sun-. The available data do not allow to establish a hierarchy as yet, but variable ordering is one of the facts to be taken into account when describing PS.

b. sam-ab h i-vy-ā-hṛ
together-upon-apart-to-take 'mention together; associate together' b'. hṛ 'take, bear, fetch, carry, bring' ā-hṛ 'fetch, bring, offer, deliver, bring, put on, use, utter, speak' vy-ā-hṛ 'pronounce, mention, converse, call by name, confess' ab h i-vy-ā-hṛ 'utter, pronounce, converse about sth' sam-ab h i-vy-ā-hṛ 'mention together; associate together' As can be noted, the addition of each preverb causes a semantic drift in the meaning of the whole verbal base: in (4a), ā-functions mainly as a telicity operator, but a few idiosynchratic semantic extensions can also be observed, as can be seen from the translations. The further addition of anu-and sam-are however only compatible with the basic interpretation ('grasp'). The addition of sam-without anu-3 , instead, has the effect of making new idiosynchratic meanings appear once again. This shows that conventionalization is at play at each and every stage of preverbation, which seems to point in the direction suggested by Papke, i.e. that preverbation proceeds incrementally, one layer at a time. In (4b) we can see a partly similar situation: ā-may function as a mere telicity operator or else generate a whole array of secondary meanings, only some of which pertain to the field of thought and speech. The addition of vi-, ab h i-, and sam-, instead, restricts the semantics of the complex base exclusively to the linguistic field (i.e. one of the secondary meanings, unlike 4a): the resulting base samab h ivyāhṛ is used exclusively in metalinguistic discourse. Again, the consistent semantic shift associated with each layer of preverbation seems to indicate some degree of lexicalization at each stage; the preverbs seem not to be added simultaneously, but incrementally, and most likely at different chronological stages.
To conclude our (partial) survey of PS in Indo-European languages, it will be interesting to have a quick look at the situation exhibited by a language belonging to the branch that is most closely related to Slavic: Baltic. Let us consider the following Lithuanian data (Nevins, Joseph 1993:95-96): (5) a. žìn-ti b. pa-žìn-ti c. pri-pa-žìn-ti (Lithuanian) know-inf /pa/-know-inf in.front-/pa/-know-inf 'to know (sth)' 'to know (sb)' 'to acknowledge, admit, recognize' Surprisingly, we find that PS is not productive at all in Baltic: (5c) is one of an extremely limited set of examples 4 . Moreover, stacks of more than two prefixes are not found in Baltic. In (5b), the prefix pa-restricts the set of possible objects to humans only (simultaneously triggering an inchoative reading of the 3 In Sanskrit preverbs ending in -i and -u (such as ab h i-, vi-, anu-) are subject to a sandhi rule according to which the final vowel of the preverb is replaced by the homorganic semiconsonant (thereby yielding ab h y-, vy-, anv-). 4 I am grateful to Peter Arkadiev for bringing this fact to my attention. verb). In (5c), pri-adds the meaning 'publicly' 5 . The derivation of (5c) clearly presupposes the lexicalization of (5b), which functions as its starting point.

PS in Native America
The ability of preverbs to stack is not confined to the Indo-European language family. The phenomenon is also attested in different areas such as the Americas. In Cora, a Southern Uto-Aztecan language spoken in Mexico, we can observe a quite elaborate system of preverbs 6 . These quite often come in stacks of two or three units, and quite often the semantics of the stacks is not strictly compositional. Let us consider the following example, taken from a narrative about a female toad (Casad 1984:457): The translation mentions an implied concept ('in their hands'); the nominal expression for 'hands' is m w áhka'a, in Cora, but this form does not appear in (6). Similar observations can be made about (7): (7) ú pú=é'-e-h-n y eeri-'i there decl:3sgs=far.away-outside-on.slope-be.in.sight-stat '[The sky] over there above the town is (all) lit up.' Again, the translation shows that a town is intended as a part of the background, but no explicit mention of the 'town' (čah) is made. We can thus see that conventionalization and lexicalization play an important role in the preverb system of this language.
the other examples we have considered so far, and it means that we can also expect the strings of preverbs in the two examples to perform different functions: in (6) the preverbs describe a complex path, whereas in (7) we have a case of PS used to represent the extensional domain of a static location configuration. This case is by no means rare in the language, as (8) illustrates: In these examples, the addition of stacks of preverbs does not alter the meanings of the verbal bases: all the forms describe configurations of static location. This suggests that the lexicalization process, in cases such as these, might have affected the strings of preverbs, rather than the base and each preverb incrementally.
PS of the Cora type, which allows for the expression of static location as well as complex paths, though unusual from an Indo-European point of view, is found elsewhere in America. I can report an example from Pawnee, a Caddoan language genetically unrelated to Cora (Mithun 1999:372): (9) ri-kata-iri-itik narr:3a-against-horizontally-hold 'She holds (her/him) in her bosom. ' We can thus hypothesize that the use of PS for defining the extensional domains of static situations could be an areal trait typical of the Americas. However, given the paucity of our data, this remains to be proven by further research.
Another more general difference between PS in the two native American languages we have seen and Indo-European languages is the fact that the preverb systems of Cora and Pawnee bear no etymological relation to their adpo-sition systems: Cora preverbs are rather related to adverbs (a situation which reminds of Hungarian), whereas for Pawnee it is even questionable whether the language has adpositions at all (see the discussion on the closely related Wichita in Baker 1996).

PS in Bulgarian
As already hinted at in the introduction, in some languages preverbs grammaticalize, evolving into telicity operators or aspectual markers. Slavic languages are the most typical example of this process.
PS too is especially productive in Slavic, and most especially in some Southern languages such as Bulgarian. Focusing on the colloquial language, Istratkova (2004) claims that up to seven preverbs can stack on some verbal bases. Atanasova's (2011) study of the Bulgarian National Corpus yielded forms with up to four preverbs, while Rojzenzon's (1974) monograph on PS in a cross-Slavic perspective reports forms with five preverbs. This number is what my research yields, too: despite a few differences in the opinions expressed by the sources and some disagreement among speakers, my consultants seem to agree on the fact that the Bulgarian language can tolerate quite well stacks of up to five preverbs on a given root, as in (10)  In this form, the perfective base kaz-('to show') is modified by the lexical prefix raz-('around') to yield the meaning 'tell, narrate'. The form thus obtained is further modified by the repetitive pre-, after which the addition of the fixed quantifying preverb sequence iz-po-na-takes place. In this sequence, the cumulative na-teams up with the external completive iz-to yield the idea of an action affecting a massive amount of material (na-) and carried out until the complete exhaustion of the object (iz-). The distributive po-further adds the concept of an action performed incrementally.
The study of PS in Bulgarian involves different levels of analysis, as shown by Atanasova (2011): there are formal issues, such as which preverbs take part in stacking, how many elements a stack can include, and in which order they can appear. But there are also semantic issues, such as the problem of the lexicalization of compound preverbs like o-po-in o-po-vestjavam I 'to announce, proclaim'. Other semantic issues are iteration (where a preverb appears twice in a stack, as in the sequence po-po-), and variable ordering (as in iz-po-draskvam I 'to scratch all over' vs. po-iz-draskvam I 'to scratch a little'). In what follows, a discussion of these issues will be provided.

Formal issues
Atanasova (2011) is a descriptive study of literary Bulgarian. It analyses 2,680 verbal forms with multiple prefixes, mainly taken from Bălgarski Tălkoven Rečnik, plus additional material from the Bulgarian National Corpus (http://search. dcl.bas.bg/), as well as other sources.
Preverbation in contemporary  . All of these except v-/vă-and ot-can be found as a second-layer preverb (or further to the left) in PS constructions. Other prefixal morphemes (such as zad, prez-, bez-) are not involved in PS. As with the other Slavic languages, most Bulgarian preverbs are etymologically related to prepositions; exceptions are pre-, pro-and raz-. 'enchant', denominal, from čar 'spell') D. Bound verbal roots: văz-pri-émam I /văz-pri-éma P 'perceive, agree' (cf. priemam I /priema P 'receive, accept, give shelter', but emvam I /emna P 'take, catch, attack') With simple, underived verbal bases (Type A) the preverbs may be added either incrementally or as a lexicalized preverb compounds. The semantic drift, if there is any, may be slight, and it is not always clear whether a given combination of preverbs forms a compound or not. In Type B the preverbs are clearly added as a compound, as a singly-prefixed form *po-vestjavam I does not exist. Types C and D are instead by definition cases of incremental preverbation (a simple stem *čarovam I is not available for Type C, nor is an underived form *emam I available for Type D).
Atanasova's approach provides an insightful, fine-grained classification methodology for dealing with PS as the phenomenon manifests itself in Bulgarian. However, a few observations are in order. The author correctly excludes from her study verbs formed by prefixed nominal roots, such as o-bezcvetjavam I /obezcvetja P 'bleach, discolor' (cvjat 'color' > cvet-en 'colored' > bez-cvet-en 'bleached' > o-bez-cvet-ja 'to bleach' 8 ). Loan verbs with etymological preverbs are also not considered (quite understandably). 7 The common Slavic prefixes *vy-and *nizъ-do not have productive reflexes in modern Bulgarian. 8 It must be noted that in forms such as o-bez-cvetja only o-is a preverb in our sense, bez-being a negative prefix that modifies an adjectival stem at an intermediate stage of the derivation of the form at issue. Thus, we cannot speak of PS in this form.
On the other hand, the author does not deal with the negative prefix ne-in the combination ne-do-, since this prefix is not a preverb strictu sensu (verbs such as ne-do-viždam I /ne-do-vidja P 'be short-sighted' are not considered instances of PS). Problems may emerge if this assumption is carried on to the level of cross-Slavic comparison, however, since we have formations like OCS vъz-ne-naviděti 'despise, hate' (Rojzenzon 1974:152). Even though we can agree with the author that ne-is not a preverb, this element does in fact take part in PS in OCS (and other Slavic languages). Most likely, it forms compound preverbs with the elements occurring to its immediate right: OCS vъz- [ne-na-]viděti could then be analyzed as an instance of double preverbation, and Bulgarian [ne-do-]viždam as an instance of single preverbation.

Semantic issues
3.2.1. lexical/superlexical distinction. Following Smith's (1991/1996) distinction between lexical (perfectivizing or resultative) and superlexical (phasal or adverbial) preverbs, Istratkova (2004) proposes that only the innermost preverb in a stack can be lexical, all the others being superlexical. The meanings of superlexical preverbs are given as follows by Istratkova (2004:312) 9 : (11) bulgarian superlexical preverbs (istratkova 2004): pre-'to do again' raz-'to do in excess, to the very end, in many directions' na-cumulative; requires plural or mass nominal arguments po-distributive over subjects and objects iz-'to do completely' po-attenuative: 'do to a certain extent, with low intensity' za-'to begin' do-'to finish' po-delimitative: 'do for a while' As can be noticed, po-is listed three times (with three different semantic values). The semantic range covered by Bulgarian superlexical preverbs varies somewhat with respect to other Slavic languages: e.g. the perdurative pro-and the saturative na-found in Russian have no Bulgarian counterpart.

preverb compounds. The distinction between PS on derived verbal
vs. nominal bases is of crucial importance, since it allows to distinguish two 9 The list of superlexical preverbs reported is not to be conceived as a hierarchy in Cinque's (1999) sense. Istratkova (2004:318), however, does in fact propose such a hierarchy for Bulgarian. See further (3.2.3, 3.2.4 and the conclusions) for specific issues relevant to phenomena suggestive of hierarchical properties. key procedures giving rise to PS: recursive/incremental preverbation vs. adjunction of compound preverbs. Derived verbal bases can build stacks of prefixes incrementally, i.e. for the doubly prefixed verb po-văz-măžeja P 'become a little more virile' we also have a corresponding verb văz-măžeja P 'become a man', with one single preverb. Things are different with o-po-vestja P 'announce, proclaim', for which no equivalent *po-vestja exists in Bulgarian, as we have seen. The class of compound preverbs can be determined precisely. Atanasova (2011) reports 18 combinations: iz-pre-, o-na-, o-po-, o-pre-, ob-za-, po-do-, pod-s-/să-, pre-o-/ob-, pred-u-, raz-po-, raz-pre-, raz-pro-, s-po-, săpri-, să-v-, u-s-, za-o-/ob-, za-v-. The preverbs văz-, ot-and nad-do not form compound preverbs.
The semantic value of some of these prefixal compounds can be compositional. An example is iz-pre-(numerous.agents-prolonged/tiresome.action): iz-pre-vărvjam I se/iz-pre-vărvja P se 'for many or all to go, passing one after another'. As a further example, we can quote za-o-/ob-(ingr-extensive.action): za-ob-lačavam se I /za-ob-lača se P 'become cloudy, overcast'. Other compounds instead seem to be processed as a unit, not compositionally. This seems to be the case of the string să-v-('for the agents to perform the action simultaneously' or 'to turn out to be the same as sb or sth'): să-v-padam I /să-v-padna P 'concur, clash, coincide'. Another such string is pred-u-('action is performed before a given limit') : pred-u-preždavam I /pred-u-predja The innermost preverbs of a stack (generated by primary prefixation) can express meanings related to space as well as aspectuality and some adverbial functions (such as i.a. the attenuative).
The preverbs occupying the second slot (generated by secondary prefixation) can also express such types of meanings, but locative concepts are drastically reduced.
The third (and fourth) preverb layers show a further semantic reduction, with the aspectual meanings being reduced to phase-related concepts; the adverbial meanings such as the attenuative are predominant in this domain.
In some cases, the polysemy of a base verb can be inherited by the PS construction (e.g. po-za-silvam I se/po-za-silja P se I. 'to become a little stronger'; II. 'to walk a little faster'). Quite often, though, PS contributes to restricting the semantics of a verb. The prefixed verb ot-minavam I /ot-mina P has four meanings: A. 'to go away from sth/beyond sth'; B. 'to pass by without stopping or greeting sb'; C. 'to ignore, pay little attention'; D. 'for a pain/illness to disappear'). But, as Atanasova (2011) notes, po-ot-minavam I /po-ot-mina P is only compatible with meanings A and D (i.e., it can only mean 'to go a little further away from sth' or 'for an illness/pain to decrease a little'). This situation reminds of the Sanskrit facts illustrated by Papke (2010) and seen above.

iteration.
There are only four preverbs that can undergo iteration in Bulgarian (po-, iz-, pre-, o-/ob-). We can talk about two types, differing in terms of adjacency: A. Adjacent: po-po-gleždam I 'have a look every now and then' B. Non-adjacent: pre-raz-pre-deljam I /pre-raz-pre-delja P 'redistribute, replan' Iteration is merely formal, never semantic: the sequence [po-po-] in po-pogleždam I conveys a distributive sense that can be ascribed either to the outer preverb (which could be placed in a hierarchically higher position in the derivation, as Istratkova claims) or the preverb sequence as a whole (since distributivity features are commonly expressed by the iteration of a morpheme, cross-linguistically). The inner preverb seems to be delimitative, rather than attenuative 10 . In pre-raz-pre-deljam I , on the other hand, the inner pre-is lexical and the outer one iterative. There are also combinations of different prefixes with very similar meanings, which at first sight might be taken as suggestive of semantic iteration 11 : The form prevăzproizvežda se 'it is reproduced' shows four preverbs stacked on the root √ved 'lead' (which continues OCS vesti and only in some dialects surfaces as a pair veda P /veždam I ): the preverbs of the first two layers are however lexicalized, as is evidenced by the semantic drift (iz-veda P /iz-veždam I 'take out, lead' > pro-iz-veda P /pro-iz-veždam I 'produce, carry out, promote'). The preverb văz-in văz-pro-iz-veda P /văz-pro-iz-veždam I 'reproduce, renew' indicates the production of an object similar to an original, whereas pre-can be taken to mean 'again, anew', or else it could emphasize the idea of similarity to the original already expressed by văz-12 . Anyhow, we cannot speak about semantic iteration, even in such cases.
Turning our attention to other Slavic varieties, there is even one attested case of what seems to be the non-adjacent iteration of a preverb stack similar to the one seen in (12). It is described as the Perm dialect of Russian, as recorded in the 1930s (SRNG 6:27, cited in Ludwig 1995): 10 This could be a counterexample to Istratkova's (2004) claim that the delimitative po-does not allow for stacking.

. [vy-iz-]po-[vy-iz-]obi-chodit'
[compl-compl-]distr-[compl-compl-]around-walk I analyze the form vyspovyizobichodit' as a case of a compound preverb made up of two different completive preverbs (vy-+iz-) and added to a base containing a lexicalized preverb (obi-chodit' 'tidy up'); the compound preverb seems to be iterated after the merger of a distributive po-. The overall impression one gains from this form is that we have to do with nothing less than 6 preverbs, even though the synchronic operations at play seem to be no more than 3. The iteration of the completive preverbs, once again, could be explained as the expression of a distributivity feature, with no need to hypothesize semantic iteration.
The antonimy effect seems to fall out neatly from Istratkova's (2004) hierarchy of superlexical preverbs: (14) HierarcHy of superlexical preverbs (istratkova 2004): The hierarchy predicts that po-in iz-po-draskvam and po-iz-draskvam should spell out two different syntactic projections (distributive in the former case and attenuative in the latter): given the meanings reported for the two forms as reported above, the prediction seems to be borne out 13 .
However, we must distinguish grammaticalization from lexicalization: whereas the former is likely to generate hierarchy effects, the latter is apparently at work in the remaining cases described above. The problem is too complex to be dealt with in this paper, but the first step toward a solution should be a thorough semantic analysis of each form involved. 13 I am grateful to an anonymous reviewer for drawing my attention to the possibility of a hierarchy effect in the point at issue.

PS and viewpoint aspect
The issue of viewpoint aspect, as applied to verbs with PS, has been a matter of controversy. The most widely accepted guiding principle is that the last added aspectually relevant affix (including suffixes) determines viewpoint: thus, from piša I , we obtain na-piša P . Is then the prefix na-a marker of perfective viewpoint ( Because a negative imperative context excludes the perfective viewpoint in Bulgarian, the form must be considered imperfective. As noted by Istratkova (2004), then, stacked preverbs do not uniformly contribute to perfectivity in Bulgarian. The so-called perfectivizing preverbs actually quantize the predicate; preverbs from the second layer on are all superlexical. In this respect, Bulgarian differs from other Slavic languages such as Russian, Ukrainian and Polish 14 . Ludwig (1995) shows how these languages do in fact allow some outer prefixes to perfectivize a singly-prefixed verb, even when the base verb is imperfective. To see this, we can consider the Russian aspectual pair na-birat' I / na-brat' P 'gather a lot, assemble': a further preverbation layer results not in an aspectual pair, but rather in two aspectually equivalent perfectives: po-na-birat' P = po-na-brat' P 'to gather a lot, little by litte'. This situation is a contrast to Bulgarian, where po-na-biram I /po-na-bera P can be argued to form a pair with opposite aspectual values. In Russian, Polish, and possibly also in Ukrainian, this phenomenon of reperfectivization of imperfective prefixed verbs mainly happens with poand na-. As Tatevosov (2008) notes, reperfectivization even involves verbal bases with PS and the secondary imperfectivization suffix -(y)vat', if the second preverb has been merged after the suffix, as in [na-[[za-pis] P -yvat'] I ] P diskov 'record a lot of CDs'. In other words, for Russian one has to know the history of a particular form to know its aspectual value; such a problem does not exist in Bulgarian.
As far as the interactions between viewpoint aspect and PS are concerned, then, we have thus unveiled an important difference between languages like Rus-14 I am grateful to Francesca Fici for helpful discussion on this point. sian and Polish on the one hand and Bulgarian on the other hand: while in Russian and Polish PS some preverbs of the second layer and higher can still maintain their perfectivizing/quantizing force, preverbs of these layers are aspectually inert in Bulgarian (as Istratkova 2004 andAtanasova 2011 show). This means that PS can actually take part in defining the aspectual profile of any Slavic language.

Conclusions
The cross-linguistic considerations on the phenomenology of PS exposed in this paper show that the Slavic preverb system, although unique in the constellation of its properties, works according to principles which are not unique to Slavic, especially as far as the grammaticalization of aspectual meanings is concerned. Given the high degree of similarity between the phonological shapes of preverbs and adpositions in Indo-European languages generally, it is tempting to equate the two systems: in the case of Slavic, this means equating preverbs with prepositions. On a typological level, however, we must bear in mind that the Indo-European phenomenology is a rare case, and that the most privileged interaction of the category of preverbs is with adverbs. Even so, a grammaticalization path leading from preverbs to markers of aspectual meanings can be observed in typologically very different languages: in Cora (Casad 1984) the spatial preverbs wa-'throughout' and ta-'across' often function as perfectivizers (or quantization/telicity operators?). They can even stack in the sequence wa-ta-to emphasize the completion of an event.
In a cross-linguistic perspective, Indo-European PS obeys a semantic constraint: it is limited to dynamic predicates. The mechanism by which it is generated is mainly the (incremental) conventionalization and lexicalization of compounds of preverbs and verbal bases. The phenomenon may well go back to the protolanguage, but we cannot know whether other mechanisms (such as the formation of compound preverbs) were already active at that stage. In any case, it is only in the Slavic languages that PS developed into the phenomenon we can observe now. In these languages, preverbs extended their semantics, very likely entering more than one grammaticalization path; these paths determined their evolution from spatial particles to superlexical preverbs, and from telicity markers to markers of quantization and, later on, perfectivity.
These paths, however, seem to have taken different directions in different languages such as Bulgarian and Russian: superlexical preverbs, in particular, have not gained any quantizing or perfectivizing power in Bulgarian, whereas they have in Russian. This difference is responsible for the tendency of Russian PS to be associated with perfectivity, whereas Bulgarian PS tends to be aspectually inert.
Some phenomena, such as variable preverb ordering and iteration, may be explained by means of a hierarchy, as proposed by Istratkova (2004). This approach is promising; however, working out the details of the cross-Slavic (as well as cross-Indo-European and more generally cross-linguistic) variation in the