Nonargumental Mixed Projections

Syntax 13:2, June 2010, 165–182 DOI: 10.1111/j.1467-9612.2009.00134.x Nonargumental Mixed Projections E. Phoevos Panagiotidis Abstract. This paper takes a comparative look at idiosyncratic instances of mixed categories in Korean, Japanese, Hebrew, and Greek, arguing them to be genuine mixed projections, despite their inability to function as arguments—which in turn is a well-known characteristic of mixed projections, such as English gerunds. After their syntactic behavior is examined, it is argued that these nonargumental mixed projections are embedded within prepositional phrases headed by (null) temporal prepositions. This derives their peculiar properties while successfully capturing the differences between them and purely verbal/clausal projections such as infinitivals. 1. Introduction This paper inquires into two issues. First, it argues that the so-called gerunds in Greek, Hebrew gerunds, and verbal nouns in Korean and Japanese are indeed mixed projections, consisting of a nominal and a verbal/clausal part. This will lead us to a second question: if the above configurations are truly instances of mixed projections, then why can they not appear in argument positions, as expected from more familiar cases of mixed projections, such as English gerunds, Turkish nominalized clauses, and Dutch nominalized infinitives? 2. Gerunds as Mixed Projections Let us start with the first question: are Greek and Hebrew gerunds, as well as Korean and Japanese verbal nouns, truly mixed projections? To answer this question, we need to clearly state what we mean by the term ‘‘mixed projection,’’ of which gerunds constitute a well-studied case. By way of definition, we can say that mixed projections are XPs displaying properties from more than one category—say, both nominal and verbal/clausal properties.1 Research on mixed projections, and gerunds in particular, is vast; they were the focus of intensive research throughout the 1970s and 1980s, although less so more recently. For reasons of space, I will rely on the meticulous overviews and critical reviews of the relevant literature by Bresnan (1997), Borsley & Kornfilt (2000), I wish to thank Winnie Lechner, Rita Manzini, Adriana Belletti, Sabine Iatridou, Anna Roussou, Ianthi Tsimpli, Georges Tsoulas, Io Sitaridou, Gillian Ramchand, Elena Anagnostopoulou, and Vayos Liapis, as well as the audiences in Protaras, Salonica, York, and Harvard for their feedback, criticism and suggestions. Three anonymous reviewers provided valuable discussion and ideas. Of course, all errors, omissions, and misconceptions are mine. 1 An anonymous reviewer reports that Latin grammarians already refer to the category of gerundivium (and supine and participle) as of ‘‘mixed nature.’’ Here, the term ‘‘mixed projection’’ is preferred over ‘‘mixed category’’ exactly because mixed projections are not primitives such as ‘‘noun’’ or ‘‘verb’’ but syntactic structures containing material belonging to different categories. A disclaimer is also called for regarding the term ‘‘gerund,’’ which is simply used for convenience (and following the literature on Hebrew and Greek) and carries no theoretical implications, as is going to become clear. I wish to thank an anonymous reviewer for discussion of the terminology.  2010 The Author Journal compilation  2010 Blackwell Publishing Ltd 166 E. Phoevos Panagiotidis Malouf (2000), and Hudson (2003) in order to zoom in on two important properties of mixed projections. The first one is that the markers of the two categories (e.g., nominal and verbal) that give a mixed projection its categorially dual character never intersperse. Put differently, mixed projections consist of (typically) two categorially uniform subtrees. In Bresnan 1997 (also in Malouf 2000 and Ackema & Neeleman 2004:174), we find this observation expressed as the following generalization: (1) Phrasal Coherence The mixed projection ‘‘can be partitioned into two categorially uniform subtrees such that one is embedded as a constituent of the other.’’ (Bresnan 1997:4, after Malouf 2000) The state of affairs this generalization describes is that chunks of different categories, say, nominal and verbal, in a mixed projection are distinct and occupy different ‘‘sides’’ thereof. For instance, we never have alternating nominal and verbal projections making up a mixed projection. Consequently, there is a cut-off point where verbal/clausal characteristics end and nominal ones begin, and there is no attested case of a mixed projection where verbal and nominal heads actually alternate or intersperse (Bresnan 1997). This purported state of affairs is schematically illustrated in the following diagram: (2) a. A mixed projection abiding by Phrasal Coherence nominal nominal nominal verbal verbal verbal b. An impossible state of affairs * nominal nominal verbal nominal nominal verbal Furthermore, the generalization in (1) holds for mixed projections in a range of typologically unrelated languages: Hebrew, Arabic, Turkish, Kikuyu, Italian, Dutch, German, Dagaare (a Gur language of the Niger-Congo family), and others (Bresnan 1997).  2010 The Author Journal compilation  2010 Blackwell Publishing Ltd Nonargumental Mixed Projections 167 A typical example of Phrasal Coherence can be seen in the extensively studied English POSS-ing gerunds, as in (3), where the higher part of the projection is headed by a possessive D (a nominal element), whereas the lower part assigns accusative (the signature of the category v). There is no interspersion of verbal elements within the nominal ‘‘chunk,’’ or vice versa. (3) [BobÕs obsessively scolding us] traumatized everyone. A second generalization can be surmised by surveying the literature on mixed projections (Borsley & Kornfilt 2000, Malouf 2000, Hudson 2003), from which it emerges that mixed projections externally behave as nominals: (4) Nominal external behavior Externally, mixed projections behave as nominals. Externally, mixed projections display straightforward nominal behavior; for instance, English gerunds and Turkish nominalized clauses can be arguments of verbs, Spanish and Dutch nominalized infinitives are headed by articles, and so on. However, we have no clear evidence of bona fide mixed projections behaving externally as verbs or clauses, while containing a ‘‘real’’ nominal element.2 These two generalizations bring us to the first problem this paper will look into: how are we to analyze elements that look like mixed projections, such as Greek and Hebrew gerunds, or Korean and Japanese verbal nouns, but can only appear as adjuncts? Three options present themselves: perhaps these are not mixed projections and their behavior is to be explained otherwise; alternatively, they are indeed mixed projections and their peculiarity—that is, their not appearing as arguments—automatically falsifies (4); or, finally, they are indeed mixed projections and their not appearing as arguments is to be explained away. I argue here that the third solution is the correct one, taking the constructions in question to be genuine mixed projections and claiming that their nonargumental status results from their involving more structure than meets the eye. But to finally understand why such constituents do not appear in argument positions, we need to look at their internal structure in detail. 3. Korean /Japanese Verbal Nouns Verbal nouns (VNs) in Japanese and in Korean share some very remarkable properties. They are morphologically simplex nominal elements (see Yoon & Park 2004 for a detailed argument on why VNs cannot possibly be derived from verbal sources), and no special nominalizing morphology is attached to them, something that 2 This is a matter that needs further research. Although I have so far not managed to find unambiguous instances of mixed projections that externally behave as verbs, an anonymous reviewer raises the issue of nominalized verbs in Australian languages. Given the relevance of such a possibility for the proper evaluation of Phrasal Coherence (1), this is a question that will feature prominently in follow-up research.  2010 The Author Journal compilation  2010 Blackwell Publishing Ltd 168 E. Phoevos Panagiotidis contrasts them, in the case of Korean at least, with ‘‘syntactic’’ nominalizations suffixed by -um (Yoon & Park 2004, Ackema & Neeleman 2004:179). However, VNs display two prototypically verbal/clausal characteristics: the ability to assign verbal Case (Iida 1987) and the projecting of full argument structures (see Tsujimura 1992, Manning 1993 for overview and analyses). An example from Korean (adapted from Yoon & Park 2004) illustrates this state of affairs: (5) [Kim-paksa-ka woncahayk-ul yenkwu]-cwung-ey cencayng-i Kim-Dr.-nom atom.nucleus-acc research-midst-loc war-nom ilena-ss-ta. broke.out-pst-decl ÔThe war broke out while Dr. Kim was researching the atom nucleus.Õ In (5), the VN yenkwu ÔresearchÕ, a noun, is associated with functional material that assigns accusative to woncahayk-ul Ôatom nucleusÕ and nominative to Kim-paksa-ka ÔDr. KimÕ. Interestingly, and as already glimpsed in the example above, VNs cannot be arguments but are typically embedded within modifying expressions with a temporal interpretation. Observe the following Japanese example. As in the Korean example in (5), the VN ryokoo ÔtravelÕ is inside a temporal expression (Ôon the occasion ofÕ) from which it receives genitive Case, while itself assigning both nominative and accusative to its two arguments, sensei ÔteacherÕ and kagai ÔabroadÕ, respectively. (6) [sensei-ga kaigai-o ryokoo]-no sai... teacher-nom abroad-acc travel.vn-gen occasion Ôon the occasion of the teacherÕs traveling abroad…Õ (Shibatani 1990:247) Alternatively, VNs can combine with a copula/light verb to yield the light-verb construction (see Shibatani 1990:247 for Japanese and Yoon & Park 2004 for Korean): (7) John-i Yenghi-lul simha-key kongyek(-ul) hay-ss-ta. John-nom Yenghi-acc severe-adv attack(-acc) do-pst-decl ÔJohn severely attacked/criticized Yenghi.Õ (Yoon & Park 2004) We will not look further into this particular syntactic context because too much depends on the proper analysis of light-verb constructions in Japanese and Korean. Let us then return to the case where VNs appear embedded within a temporal expression. A number of such expressions in Japanese are listed here (adapted from Shibatani 1990:247 and Tsujimura 1996:139): (8) VN with a temporal nominal VN-no ori Ô(on) the occasion of VNÕ; VN-no setu Ô(at) the time when VNÕ; VN-no akatuki Ô(at) the happy occasion of VNÕ  2010 The Author Journal compilation  2010 Blackwell Publishing Ltd Nonargumental Mixed Projections 169 (9) VN with a temporal postposition VN-tyuu Ôin the middle of VNÕ; VN-go Ôafter VNÕ; VN-sidai Ôas soon as VNÕ; VN-gatera Ôthe same time as VNÕ; VN-izen Ôbefore VNÕ Generalizing, we can say that VNs can appear as complements of temporal expressions and postpositions like those in (8) and (9), as illustrated in (5) and (6). In brief, nouns functioning as VNs externally still behave like nominals, as they can be complements of postpositions and/or receive Case. At the same time, they have full argument structure and must possess the relevant functional structure necessary to assign accusative and even nominative to their arguments. Thus, VNs qualify as mixed projections, with a nominal external character, despite their not occupying argument positions. 4. Hebrew Gerunds Gerunds in Hebrew display a very telling behavior, similar to that of VNs in Japanese and Korean. Closely following Siloni (1997:chap. 5) throughout this section, I will sketch some of their properties and characteristics, through a comparison with both infinitives and nominalizations in the language. Let us begin by comparing gerunds and infinitives. First, Hebrew gerunds form a morphological minimal pair with infinitives: gerunds look exactly like infinitives but lack the preverbal infinitival marker le-. They consequently completely lack nominal morphology and/or /-feature marking.3 (10) le-Õašen Ôto smokeÕ Õašen ÔsmokingÕ infinitive gerund Second, gerunds take obligatory overt subjects, unlike infinitives, where overt subjects are impossible (Siloni 1997:164–165): (11) Bi-rÕot *(-o/Dan) Õet Õim-o, Õalac lib-o. in-seeing subj.cl/Dan acc mother-his exulted heart-his ÔWhen he/Dan saw his mother, his heart exulted.Õ Third, the negation lo, which is a verbal/clausal negation par excellence (see, e.g., Shlonsky 1997:chap. 2), is disallowed with gerunds, as illustrated in (12). (12) a. Dan bikeš Õotanu lo le-Õašen po. Dan asked us not to-smoke here ÔDan asked us not to smoke here.Õ 3 Yoon (1996) and Ackema & Neeleman (2004:175–181) have shown that neither -ing nor similar endings in mixed projections of Dutch, Spanish, and Italian constitute nominalizing morphology— that is, category-changing affixes.  2010 The Author Journal compilation  2010 Blackwell Publishing Ltd 170 E. Phoevos Panagiotidis b. *Õim lo havin-o Õet ha-macav, hitragaznu. with not understanding-subj.cl acc the-situation we.got.angry ÔWe were angered with him not understanding the situation.Õ (Siloni 1997:165, (12)) Finally, and crucially for our inquiry here, gerunds in Hebrew cannot appear in argument positions and can only be found in the complement of a temporal preposition, like in (11); ‘‘they constitute temporal adjuncts’’ (Siloni 1997:164): (13) *Lo keday [Õašen Dan Gitane]. not worth smoking Dan Gitanes ÔIt is not worth Dan smoking Gitanes.Õ Actually, gerunds in Hebrew appear to stand in complementary distribution with infinitives: it is simply not possible to have an infinitive in the complement of a temporal preposition, as (14) shows. (14) Lifney (*la-)Õalot ha-nosÕim la-matos, hexel la-redet gešem. before (to) ascending the-passengers to.the-plane started to-fall rain ÔBefore the passengers boarded the plane, it started raining.Õ (Siloni 1997:163, (8)) A summary (adapted from Siloni 1997:165) is given in Table 1. Let us now briefly turn to the differences between gerunds and nominalizations in Hebrew. Nominalizations are bona fide nouns and can (usually) pluralize, be modified by adjectives, and appear in the complement of articles such as ha-; they can also license šel (‘‘genitive’’) arguments. On the other hand, gerunds cannot pluralize, they take adverbs instead of adjectives, they cannot appear in the complement of ha- or other determiners, and they cannot license šel genitives. The interested reader is referred to Siloni 1997:172–175 for details and examples and to Hazout 1994 for a different take on the issue. This state of affairs is reminiscent of Korean, where -um nominalizations contrast with VNs; crucially, Hebrew nominalizations, like Korean -um nominalizations, can of course appear in argument positions, unlike gerunds and VNs, respectively, as shown in (15), adapted from SiloniÕs (1997:174, (29), (30)): Table 1. Comparison between Hebrew infinitivals and gerunds Distribution Overt subject Lo negation Infinitives Gerundival clauses Argument position Impossible Possible With temporal Ps Obligatory Impossible  2010 The Author Journal compilation  2010 Blackwell Publishing Ltd Nonargumental Mixed Projections 171 (15) a. *Hikans-o la-misrad hiftiÕa Õoti. entering-subj.cl to.the-office surprised me ÔHis entering the office surprised me.Õ b. Knisat-o la-misrad hiftiÕa Õoti. entering-his to.the-office surprised me ÔHis entrance to the office surprised me.Õ gerund nominalization To capture the characteristics of Hebrew gerunds as well as their differences from both infinitives and nominalizations, I will claim along with Siloni (1997:177–184) that Hebrew gerunds are indeed clausal chunks, albeit headed by a null D. In other words, Hebrew gerunds are DPs consisting of a null D with a clausal chunk as its complement: they are syntactic nominalizations. Their being DPs externally—at this point recall the nominal external behavior generalization in (4)—explains how they can be complements of temporal Ps and why they do not tolerate the article ha-: they are already headed by a null D. Turning to the nature of the ‘‘lower’’ clausal chunk that Hebrew gerunds contain, we observe the following. Gerunds can take overt subjects—in fact, they must take overt subjects, as in (11). Whatever the reason for the obligatoriness of subjects, their presence would force us to think that the verbal/ clausal part of the gerund contains a TP projection. However, the impossibility of lo negation with gerunds, illustrated in (12), suggests that this must be a Tense head that is somehow defective, either because it lacks some feature or because it cannot be associated with a complementizer (Siloni 1997:165–175). In whichever direction the answer lies, observe that the morphological similarity between Hebrew infinitives and gerunds is compatible with an analysis of gerunds as containing the projection of a somehow defective T head. From the sketch above, the status of Hebrew gerunds as mixed projections combining a nominal (determiner) part and a verbal/clausal one (a defective TP) should have become more than plausible. 5. Greek Gerunds I will turn now to Greek gerunds, describing them first and then proceeding to offer an analysis. In this, I will mainly follow Tsimpli (2000), Haidou & Sitaridou (2002), and Tantalou (2004). Greek gerunds cannot be arguments and can only function as adverbial adjuncts; unlike English gerunds, they cannot appear in the complement of complementizers, determiners, or prepositions. As far as their morphology is concerned, they are formed from the imperfective stem of active paradigm verbs plus the indeclinable suffix -ondas.4 They display no nominal morphology, either. Regarding the rest of their properties, Greek gerunds can be divided, according to their function and position in the clause, into manner gerunds and absolute gerunds. Manner gerunds are impossible with any overt subject. Actually, they can only have a bound-variable null subject, say, PRO, which must be bound by the subject of 4 See Manolessou 2005 on the history of the ending -ondas.  2010 The Author Journal compilation  2010 Blackwell Publishing Ltd 172 E. Phoevos Panagiotidis the main clause (Tsimpli 2000:156, Spyropoulos & Philippaki-Warburton 2002, Haidou & Sitaridou 2002:190–191, Tantalou 2004:4), as the following example illustrates: (16) I ghonisi prosehan ta pedhiaj [PROi/*j pez-ondas skaki]. the parents watch.impf the children play-ger chess ÔThe parents were watching the children while (the parents were) playing chess.Õ Manner gerunds do not tolerate temporal adverbs or the aspectual auxiliary eho ÔhaveÕ, with which periphrastic perfect tenses are formed in Greek. A manner gerund cannot be negated. Finally, manner gerunds typically follow the main verb. Absolute gerunds permit a range of interpretations: temporal, causal, and other. Their exact interpretation in each context depends largely on pragmatics (Tsimpli 2000:137–139). As for the subjects of absolute gerunds, these can be null and they are usually coreferential with the main clause subject, but not obligatorily so. This characteristic makes them more like pro subjects; observe the following quite telling contrast:5 (17) a. [pro Trogh-ondas] erhete i oreksi. eat-ger comes the appetite Ô(OneÕs) appetite grows by (oneÕs) eating.Õ b. #Erhete i oreksii [PROi/*j trogh-ondas]. comes the appetite eat-ger ÔAppetite grows while it (= the appetite) is eating.Õ absolute gerund manner gerund Whereas the pro null subject of an absolute gerund in (17) takes a generic reference, such as ÔoneÕ, the PRO subject of a manner gerund must be bound by the main subject, yielding the rather odd interpretation of (17b). Interestingly, absolute gerunds also license overt subjects in nominative Case, as well as true temporal adverbs (like ÔnowÕ, ÔyesterdayÕ, etc.). They also co-occur with the aspectual auxiliary eho (ÔhaveÕ), meaning that absolute gerunds can appear in the present perfect. They can be negated using mi(n) negation—the negation reserved for nonveridical modalities. Finally, as also seen in (17), absolute gerundival projections usually precede the main clause. The following pair of examples illustrates the different behavior of the two classes regarding negation: (18) O Manosi irthe [PROi (*mi) mil-ondas sti Nina]. the Manos came not talk-ger to.the Nina ÔManos came (not) talking to Nina.Õ manner gerund 5 See Tsimpli 2000:154–156 for discussion on controlled gerundival subjects. I wish to thank Anna Roussou and an anonymous reviewer for discussing this matter with me.  2010 The Author Journal compilation  2010 Blackwell Publishing Ltd Nonargumental Mixed Projections 173 (19) [ proi idhi (mi) mil-ondas sti Nina] o Manosi efiye. absolute gerund already not talk-ger to.the Nina the Manos left ÔAlready (not) talking to Nina, Manos left.Õ Only absolute gerunds, like the one in (19), can be negated. On top of that, example (20) illustrates that absolute gerunds can take nominative subjects and be modified by temporal adverbs, thus establishing quasi-independent temporal reference; neither of these options is available for manner gerunds. (20) [Vlep-ondas htes o Manosi ti Nina eki] proi efiye simera. see-ger yesterday the Manos the Nina there left today ÔManos seeing Nina there yesterday, he left today.Õ A way to account for these differences between manner and absolute gerunds is to say that only the latter have a T projection, roughly following Haidou & Sitaridou (2002:194–196). The Tense head must be responsible for: • • • • • • licensing temporal (not just aspectual) adverbs, sanctioning quasi-independent temporal reference, enabling the appearance of mi-negation, licensing periphrastic perfect tenses with an auxiliary,6 permitting a pro subject, and assigning nominative to an overt subject.7 Such an approach, in the spirit of Tsimpli (2000), views Greek gerunds as clausal chunks and not as anything like mixed projections. Before examining some problems with a ‘‘clausal chunk’’ approach, it is necessary to say a few more things about the Tense head in the next section. 6. A Crucial Excursus on Tense Supposing Haidou & Sitaridou (2002) are on the right track regarding their distinction between gerunds not projecting a TP (manner) and those that do (absolute), observe the following parallel in the behavior of absolute gerunds and periphrastic tenses under conjunction, with respect to pronominal clitics: (21) I Nina ta idhe ke *(ta) akuse poles fores. the Nina cl saw and cl heard many times ÔNina saw them and heard them many times.Õ 6 7 I am grateful to an anonymous reviewer for this observation. For Greek, nominative could nevertheless be a ‘‘red herring’’: see Tsimpli 2000:153.  2010 The Author Journal compilation  2010 Blackwell Publishing Ltd 174 E. Phoevos Panagiotidis (22) I Nina ta ihe [dhi ki akusi] poles fores. the Nina cl had seen and heard many times ÔNina had seen and heard them many times.Õ (23) [[Vlep-ondas (ta) ki akugh-ondas] ta] i Nina… see-ger cl and hear-ger cl the Nina ÔSeeing (them) and hearing them, Nina…Õ Regarding the clitic-coordination possibilities above, we notice that absolute gerunds (23) resemble the perfective (sometimes called ‘‘participial’’) verbal forms such as dhi ÔseenÕ and akusi ÔheardÕ combining in (22) with an auxiliary to give perfect tenses in Greek. Unlike the situation with finite verbs in (21), it is obligatory for the verbal forms, and possible for the gerunds, to be conjoined under the scope of a single clitic. Taking clitics to attach to T (Kayne 1989, 1991; Terzi 1999), the conclusion is that the verbal forms must, and the gerunds can, stand as units smaller than TP, either because these units do not consist of TPs (possibly true of ‘‘participles’’), or because their T heads are somehow defective.8 Generally speaking, there is much more to tense marking than a simplex T head carrying, for instance, monadic [past] or [present] features. Tsimpli (2000) analyzes gerunds in this spirit. An alternative, and much more convincing, range of treatments of the syntactic representation for Tense has been developed by von Stechow (1995), Stowell (1996), Heim (1997), and Demirdache & Uribe-Etxebarria (2000), among others. Let us follow Stowell (1996:278–283) here in arguing that temporal interpretation involves the interaction of two categories: T(ense) and Z(eit). We can think of them the following way, grossly oversimplifying: Tense is a predicative category that relates two time-denoting expressions. In that respect, Tense is a bit like a temporal preposition (e.g., before or after). Zeit, on the other hand, heads timedenoting expressions, which may be covert. Thus, Z is a temporal referential category, akin to D (determiner), and when the ZP is covert, it is parallel to phonologically covert pronominals (e.g., PRO). Generally speaking, ZPs denote time intervals that T orders. The account is sketched in this phrase marker: (24) The Tense schema TP ZP1 T′ T ZP2 With the above in mind, we can now divide Greek gerunds into two categories: a. manner gerunds, lacking T; these are ZP2s, expressions denoting a time interval; 8 As Adriana Belletti (p.c.) points out, this situation is reminiscent of Italian absolute small clauses (Belletti 1990). See also footnote 10.  2010 The Author Journal compilation  2010 Blackwell Publishing Ltd Nonargumental Mixed Projections 175 b. absolute gerunds, which contain a TP headed by a (defective) T (Haidou & Sitaridou 2002:194–196). By (24), these would consist of a T with a ZP2 in its complement and a null (PRO-like) ZP1 in a specifier of T. 7. What about ‘‘Mixed’’ Projections? Suppose that the above arguments are on the right track and that the two types of Greek gerunds essentially differ in size—manner gerunds being ZPs and absolute gerunds TPs. We still do not need to consider Greek gerunds as anything more than verbal/clausal chunks, like infinitives are: at this point, it would not seem necessary to go all the way to saying they are mixed projections. Still, by arguing that Greek gerunds (or Hebrew gerunds, for that matter) are just clausal chunks (call these TPs, ZPs, or whatever), we miss a number of points. As suggested earlier, gerunds within the same language can differ in terms of the number and the type of projections they contain. That much is clear for Greek, where manner gerunds behave like they contain just a ZP2, whereas absolute gerunds appear to also contain a TP projection. Despite this difference, however, both types of gerunds always behave adverbially, as modifiers, never as clausal arguments. Elaborating, consider that Greek gerunds are unable to function as anything but adverbials even if they contain Tense—a Tense head that licenses nominative subjects and temporal adverbs. Compare the situation with purely clausal chunks headed by a defective T, such as infinitives in other languages, and the range of positions they can appear in: these TPs can certainly function as arguments. This takes us to another piece of evidence that Greek gerunds are nothing like simple clausal chunks. Such evidence comes from the fact that they are unable to function even as ‘‘nexus constructions’’ (Svenonius 1994), which are sometimes taken to be vP chunks—hence certainly smaller than either type of Greek gerund. Compare: (25) Me drink alcohol? Never. (26) *(Egho) pin-ondas alkool? Pote.9 I drink-ger alcohol never If we leave things here, we need extra assumptions about why Greek gerunds do not function at least like infinitives or nexus constructions, let alone as arguments. Otherwise, the suggestion that a purely clausal projection, especially one headed by T, must be inert is very odd. In other words, more structure than just the clausal chunk—regardless of this clausal chunkÕs size and makeup—must be involved. Thus, if we go for the minimal hypothesis—namely, that there is nothing special with these 9 Greek uses na clauses (na being an irrealis particle, informally speaking) instead of nexus constructions, as two anonymous reviewers point out. See section 9 for some more discussion.  2010 The Author Journal compilation  2010 Blackwell Publishing Ltd 176 E. Phoevos Panagiotidis ZPs and TPs that would prevent them from functioning as arguments, infinitives, or nexus constructions—we are led to presume that they cannot do so because of the structure they are embedded in. This structure should consequently make the clausal chunks it embeds unavailable to function as a (quasi-) clause or as an argument.10 The suggestion I am going to put forward here is that Greek gerunds, like Hebrew gerunds as well as Japanese and Korean VNs, are mixed projections, and their properties, including their nonargumenthood, can be derived from exactly this fact. Toward this end, I will compare and analyze Greek and Hebrew gerunds on a par with each other. 8. A (Null) Temporal P Let us now look at some common properties of Greek and Hebrew gerunds: (a) they contain clausal chunks but they are not infinitives; (b) they cannot be arguments; and (c) they cannot be in the complement of overt Ds or Cs. The last point can be easily explained: in both Hebrew and Greek gerunds there is no C layer, as the clausal chunk finishes with T (or Z) and then is dominated by a nonclausal projection. As far as Hebrew is concerned, this nonclausal projection is none other than DP (Siloni 1997:177–190). So Hebrew gerunds, far from being plain clausal chunks, are mixed projections (D is a nominal element) and look like this: (27) Hebrew gerunds, Take One DP D TP Ø As noted, if (27) is correct, then the absence of C is justified. The absence of articles is derived, as well: the gerundival projection is already headed by a null D and therefore closed off regarding any more nominal material. Given that the properties reviewed earlier are shared by Greek gerunds as well, we now only need to extend the 10 Interestingly, Romance gerunds (the -ndo forms in languages like Spanish, Italian, and Portuguese) seem to behave like purely clausal chunks: they can act like complements to perception verbs, they can be introduced by prepositions, they function in the manner of a nexus construction, they may modify an object (adverbially but also attributively as secondary predicates), they can function as absolutes, and, crucially, they participate in the formation of periphrastic tenses. They must, however, minimally differ from infinitives, as they cannot function as arguments (except in the context of a perception verb). This is illustrated by Portuguese: (i) O João viu a Maria cantando o fado. the João saw the Maria sing-ger the fado ÔJoão saw Maria singing the fado.Õ (ii) *O João apreciou falando sobre o cinema. the João liked talk-ger over the cinema My suspicion is that such Romance gerunds form a minimal pair with Greek and Hebrew gerunds, in that the former are not mixed projections but bare clausal chunks (a topic for future research). I wish to thank Rita Manzini, Ioanna Sitaridou, Vayos Liapis, and Adriana Belletti for discussing the issue with me and an anonymous reviewer for extensive commentary (including the above examples).  2010 The Author Journal compilation  2010 Blackwell Publishing Ltd Nonargumental Mixed Projections 177 beginnings of the Hebrew analysis in (27) to Greek, claiming that Greek gerunds are clausal chunks (TPs or ZPs) embedded within a DP. In the case of Greek, the presence of a D (and its specifier) provides us with a way to explain away the following: as Tantalou (2004) observes, some speakers seem to (marginally) accept topicalization, focus movement, and wh-movement within gerundival projections. If we posit a D layer in Greek gerunds, like in (27), then limited wh-movement, focus movement, and the presence of topic positions come for free, as Greek D is indeed associated with Focus and Topic projections (Grohmann & Panagiotidis 2005). At the same time, the lack of a C (complementizer) field, a fullfledged discourse-oriented set of projections, explains the limited acceptability of such operations within gerundival expressions, precisely as is the case within (other) DPs in Greek (for examples and an analysis, see also Horrocks & Stavrou 1987 and much subsequent work). Nevertheless, this cannot be the whole story. Turning back to Hebrew, the picture is clearer because Hebrew gerunds ‘‘must be introduced by [overt] temporal prepositions’’ (Siloni 1997:164). Recall that this is also the common way of introducing Japanese and Korean VNs (see section 3). In Greek there is no overt temporal preposition heading the gerundival projection. But maybe there still is a preposition there after all, albeit a null one. This is not as ad hoc as it may sound, given that in Greek there are adverbial expressions that look like bare noun phrases: (28) Irthe [B #(to) proi]. came the morning ÔShe/he came in the morning.Õ Expressions like the bracketed one in (28) cannot act as arguments, either: a plausible analysis would be that they are actually PPs, as Emonds (1987) claims for English (but see Larson 1985 for discussion of such expressions in a different spirit). So, we could extend the Hebrew analysis to cover Greek gerunds as well: they are clausal complements of a D, which in turn is the complement of a (null in Greek) temporal P.11 Finally, in Greek we would have a manner gerund when the clausal complement is a ZP and an absolute gerund when it is a TP. Otherwise, Hebrew and Greek gerunds are structurally identical: mixed projections embedded in a temporal PP.12 Mutatis mutandis, this is the general schema for Korean and Japanese VNs as well. 11 A question at this point is how we know that it is a null D and not a null C that is selected by a P. The general lack of (overt) temporal Ps selecting overt Cs—in Greek at least—provides support against the idea. Moreover, in an example like (28), any overt P is impossible—a fact compatible with the hypothesized null P being a lexical item with specific interpretation. The D, as illustrated, may be dropped in certain registers/ idiolects. 12 Given that their clausal chunk is a (defective) TP, Greek absolute gerunds are virtually identical to Hebrew gerunds. This in turn brings up the issue of the differences between the two languages, as two anonymous reviewers point out. It must be admitted that the obligatoriness of subjects in Hebrew gerunds, as in (11), although a useful telltale sign of TÕs presence, is hard to explain. However, the impossibility of lo negation can probably be linked to a comparable state of affairs in Greek absolute gerunds, where only nonveridical negation mi may be used (see section 5).  2010 The Author Journal compilation  2010 Blackwell Publishing Ltd 178 E. Phoevos Panagiotidis (29) Greek and Hebrew gerunds, Take Two PP P (Ø) DP D TP/ZP Ø The question that arises now is what the interpretation of this null temporal P is. Given that it is null, we would probably be correct in guessing that it must be an ‘‘elsewhere’’ preposition and take the default interpretation. At the same time, we can plausibly argue that the default interpretation of a temporal P would be that of containment. Looking at the null P in (28), we see that the null P relates two time intervals: that of arriving (expressed by irthe ÔcameÕ), and that of the morning (expressed by to proi Ôthe morningÕ). The null P arranges the two times in a containment relation: the arriving interval, the main event s(ME), is contained within the morning interval, the ‘‘prepositional’’ event T(PE): s(ME) ˝ T(PE).13 Let us claim that the same null P is present in Greek gerundival constructions. Clarifying, consider the following example with a manner gerund: (30) Irthe [B traghudh-ondas]. came sing-ger ÔShe/he came singing.Õ Once more, the two intervals, that of arriving and that of singing, are in a containment relation. The postulated null P in (30) relates two time intervals, that of arriving (expressed by irthe ÔcameÕ) and that of singing (expressed by the gerund), with the null P arranging the two intervals in a containment relation: the arriving interval, the main event s(ME), is contained within the singing interval, the ‘‘gerundival’’ event T(GE): s(ME) ˝ T(GE). So, the null temporal P in (30), taking a mixed D-ZP projection as its complement, is identical to that in (28), which takes an ‘‘ordinary DP’’ as its complement. The presence of null P in all Greek gerunds explains how the gerund in (16), although qualifying as a manner gerund, can receive a temporal interpretation—also witnessed by the while in the gloss of (16), for instance. However, there is still a clear difference in interpretation of manner and absolute gerundival projections. Where does this difference stem from? 13 I wish to thank Winnie Lechner for discussing the semantics of the null P with me, with errors and misconceptions remaining, as ever, mine.  2010 The Author Journal compilation  2010 Blackwell Publishing Ltd Nonargumental Mixed Projections 179 The null temporal P orders the main time interval and that of the gerund ZP in a containment relation s(ME) ˝ T(GE): this is the manner reading. Turning to absolute gerunds, the temporal P c-commanding the gerundÕs T will also interact with it, albeit differently: recall from (20) that absolute gerunds have (quasi-) independent temporal reference anyway. In the spirit of Tsimpli (2000:142–148), the available readings of absolute gerunds arise from pragmatics. However, the different adverbial readings an absolute gerund may receive must supervene on the interaction between the temporal P and the gerundÕs T (and its imperfective aspect), and also on the interaction between the whole gerundival PP and the main event T. Finally, viewing gerunds as, ultimately, temporal PPs naturally answers the question of why they cannot function as arguments. It is for the same reason that expressions like to proi Ô(in) the morningÕ in (28) cannot: exactly because they are temporal PPs. 9. Two Loose Ends Before concluding, it is necessary to address two of the broader consequences raised by the analysis, as also noted by the reviewers. One concerns the status of the null D and why it must be null in gerundival projections. As already noted, ‘‘nominal-style’’ limited wh-movement, focus movement, and the presence of topic positions within gerundival projections suggest that they contain a D layer instead of a C layer. Now, in a language like Greek, overt articles can be found with a wide range of complements, from full-fledged clauses to proper names; they are, however, impossible with gerunds. Departing from Siloni (1997:183–184)—who, addressing the equivalent question for Hebrew, calls these null Ds ‘‘inert’’—it could be speculated that there is a variety of null determiner in Greek (and Hebrew) exclusively for ‘‘adverbial’’ uses; that is, for appearing inside the complement of the semantically default and phonologically null temporal P. This null D could be either the lexical entry for the indefinite D, which is null in Greek, or a morphologically zero variant of the underspecified article to, which takes clausal complements; consider example (28). In other words, gerunds would take the ‘‘temporal adverbial’’ article. If this is independently on the right track, then the following minimal pair could be analyzed as the contrast between ‘‘direct’’ adverbial modification (31a) and modification by an adverbial inside a temporal PP (31b). (31) a. Irthe [argha]. came slowly ÔShe/he came slowly.Õ b. Irthe [PP B [DP B [argha]]]. came slowly ÔShe/he came late.Õ  2010 The Author Journal compilation  2010 Blackwell Publishing Ltd 180 E. Phoevos Panagiotidis A second matter is why the particular types of TPs and ZPs show up only in the complement of temporal Ps (embedded in the projection of a null D). I think that, again, we have to consider the bigger picture, as follows. We saw in section 4 that the Hebrew gerund is, essentially, the version of the infinitive for use with temporal prepositions, a situation reminiscent of the alternation between infinitive (a clausal chunk) and gerund (a nominalized clausal chunk) in Latin. We also saw in (26) that Greek gerunds cannot function as nexus constructions, a role taken up by na clauses (see fn. 9). Now, Modern Greek possesses no infinitives and every role that infinitives fulfill in other languages (and Greek gerunds cannot fulfill) is taken up by na clauses in this language: na clauses can even appear in the complement of an article (like other clauses). We could once more draw a rough sketch of why we only find ‘‘gerundival’’ TPs and ZPs inside a PP: whatever the status of na (a Mood head or C; see Philippaki-Warburton 1994 and Roussou 2000), a na phrase is apparently the smallest possible clausal chunk that can stand alone or as a verbal complement in Greek. Smaller clausal chunks, like ‘‘gerundival’’ TPs and ZPs, can only be licensed inside the null temporal preposition expressing containment: na clauses and the clausal chunks within gerunds are in complementary distribution in Greek. In brief, I suggest that the existence of mixed projections appearing exclusively inside temporal Ps must be correlated with the role (or lack) of infinitives and the size of the smallest possible clausal chunk that can stand alone or act as a complement.14 10. Conclusion We have seen that Hebrew and Greek gerunds are PPs headed by a temporal preposition. This preposition is null in Greek, ordering its two arguments in a containment relation. Crucially, the complement of this temporal preposition is a mixed projection comprising a D layer and a clausal subtree (a T-less ZP or a defective TP). We hence unify them with Japanese and Korean VNs, which are subtrees under nominal structure, in turn selected by temporal expressions such as adpositions. None of the above elements can ever be arguments or (quasi-) clauses, as they externally behave as temporal expressions, which captures their exclusive function as adverbial modifiers. 14 A comparable state of affairs seems to hold with Korean and Japanese VNs: apparently, outside their (overt) adpositional shell they can only act as complements of light verbs; see (7) and the discussion about it.  2010 The Author Journal compilation  2010 Blackwell Publishing Ltd Nonargumental Mixed Projections 181 References Ackema, P. & A. Neeleman. 2004. Beyond morphology: Interface conditions on word formation. Oxford: Oxford University Press. Belletti, A. 1990. Generalized verb movement: Aspects of verb syntax. Turin: Rosenberg and Sellier. Borsley, R. & J. Kornfilt. 2000. Mixed extended projections. In The nature and function of syntactic categories (Syntax and Semantics 32), ed. R. Borsley, 133–166. San Diego, CA: Academic Press. Bresnan, J. 1997. Mixed categories as head sharing constructions. In Proceedings of the LFG97 Conference, ed. M. Butt & T. Holloway King. Stanford, CA: CSLI Publications. Demirdache, H. & M. Uribe-Etxebarria. 2000. The primitives of temporal relations. In Step by step, ed. R. Martin, D. Michaels & J. Uriagereka, 157–186. Cambridge, MA: MIT Press. Emonds, J. 1987. The invisible category principle. Linguistic Inquiry 18:613–631. Grohmann, K. & P. Panagiotidis. 2005. An anti-locality approach to Greek demonstratives. In Contributions to the 30th Incontro di Grammatica Generativa, ed. L. Brugè, G. Giusti, N. Munaro, W. Schweikert & G. Turano, 243–263. Venice: Cafoscarina. Haidou, K. & I. Sitaridou. 2002. The licensing of subjects in Greek gerunds. In SOAS Working Papers in Linguistics 12, ed. C. Colella & S. Hellmuth, 189–197. London: SOAS. Hazout, I. 1994. Nominalizers in theta theory. The Linguistic Review 11:5–48. Heim, I. 1997. Tense in compositional semantics: Introduction. Lecture notes, MIT. Horrocks, G. & M. Stavrou. 1987. Bounding theory and Greek syntax: Evidence for wh-movement in NP. Journal of Linguistics 23:79–108. Hudson, R. 2003. Gerunds without phrase structure. Natural Language & Linguistic Theory 21:579–615. Iida, M. 1987. Case-assignment by nominals in Japanese. In Working papers in grammatical theory and discourse structure, ed. M. Iida, S. Wechsler & D. Zec, 93–138. Stanford, CA: CSLI Publications. Kayne, R. 1989. Null subjects and clitic climbing. In The null subject parameter, ed. O. Jaeggli & K. Safir, 239–261. Dordrecht: Kluwer. Kayne, R. 1991. Romance clitics, verb movement, and PRO. Linguistic Inquiry 22:647–686. Larson, R. 1985. Bare-NP adverbs. Linguistic Inquiry 16:595–621. Malouf, R. 2000. Verbal gerunds as mixed categories in Head-Driven Phrase Structure Grammar. In The nature and function of syntactic categories (Syntax and Semantics 32), ed. R. Borsley, 133–166. San Diego, CA: Academic Press. Manning, C. 1993. Analyzing the verbal noun: Internal and external constraints. In Japanese/ Korean linguistics 3, ed. S. Choi, 236–253. Stanford, CA: Stanford Linguistics Association. Manolessou, I. 2005. From participles to gerunds. In Advances in Greek generative syntax, ed. M. Stavrou & A. Terzi, 241–283. Amsterdam: John Benjamins. Philippaki-Warburton, I. 1994. The subjunctive mood and the syntactic status of the particle na in Modern Greek. Folia Linguistica 28:297–328. Roussou, A. 2000. On the left periphery: Modal particles and complementisers. Journal of Greek Linguistics 1:65–94. Shibatani, M. 1990. The languages of Japan. Cambridge: Cambridge University Press. Shlonsky, U. 1997. Clause structure and word order in Hebrew and Arabic: An essay in comparative Semitic syntax. Oxford: Oxford University Press. Siloni, T. 1997. Noun phrases and nominalizations. Dordrecht: Kluwer. Spyropoulos, V. & I. Philippaki-Warburton. 2002. Subject and EPP in Greek: The discontinuous subject hypothesis. Journal of Greek Linguistics 2:149–186. von Stechow, A. 1995. On the proper treatment of tense. In SALT V: Proceedings of the 5th conference on semantics and linguistic theory, ed. M. Simons & T. Galloway, 362–386. Ithaca, NY: CLC Publications. Stowell, T. 1996. The phrase structure of Tense. In Phrase structure and the lexicon, ed. J. Rooryck & L. Zaring, 277–291. Dordrecht: Kluwer.  2010 The Author Journal compilation  2010 Blackwell Publishing Ltd 182 E. Phoevos Panagiotidis Svenonius, P. 1994. Dependent nexus: Small clauses in English and the Scandinavian languages. Ph.D. dissertation, University of California, Santa Cruz. Tantalou, N. 2004. Gerunds in Modern Greek. Ms., Johns Hopkins University. Terzi, A. 1999. Clitic combinations, their hosts, and their ordering. Natural Language & Linguistic Theory 17:85–121. Tsimpli, I.-M. 2000. Gerunds in Modern Greek. Journal of Greek Linguistics 1:133–169. Tsujimura, N. 1992. Licensing nominal clauses: The case of deverbal nouns in Japanese. Natural Language & Linguistic Theory 10:477–522. Tsujimura, N. 1996. An introduction to Japanese linguistics. Oxford: Blackwell. Yoon, J. H.-S. 1996. A syntactic approach to category-changing phrasal morphology: Nominalizations in Korean and English. In Morphosyntax in generative grammar, ed. H.-D. Ahn, M.-Y. Kang, Y.-S. Kim & S. Lee, 63–86. Seoul: Hankwuk Publishing. Yoon, J. H.-S. & C. Park. 2004. Process nominals and morphological complexity. Ms., University of Illinois, Urbana-Champaign. https://netfiles.uiuc.edu/jyoon/www/Papers.htm. E. Phoevos Panagiotidis University of Cyprus Department of English Studies Kallipoleos 75 P.O. Box 20537 1516 Nicosia Cyprus [email protected]  2010 The Author Journal compilation  2010 Blackwell Publishing Ltd