The of Hungarian auxiliaries: a account

András Imrényi Jagiellonian University, Cracow Chair of Hungarian Philology Poland [email protected]

(2009). In particular, it will be proposed that Abstract while utazni acts as the governor of el (licens ing its appearance), the latter element takes the This paper addresses a hot topic of Hunga auxiliary as its head (a case of rising). Formal rian syntactic research, viz. the treatment of evidence in favour of the account will come “discontinuous” constructions involving from , coordination, prosodic structure, auxiliaries. The case is made for a projective and the placement of adverbs. dependency grammar (DG) account built on Secondly, with the above syntactic analysis the notions of rising and catenae (Groß and Osborne, 2009). Additionally, the semantic in mind, I will turn to the issue whether the basis of the dependency created by rising is dependency created by rising has any asso described with a view to analogy and con ciated meaning or function. It will be argued structional meaning. that it does, but in a way which crucially in volves aspects of (clausal) constructional se 1 Introduction mantics. The paper is concerned with a syntactic con The topic of this paper is the word order pat struction rather than the word class of auxilia tern illustrated below. ries. It has to be mentioned, though, that both (1) János el fog utazni Párizsba. traditional (Lengyel 2000) and generative ap John away will.3SG travel Paris.to proaches (Kenesei 2008) to Hungarian tend to ‘John will travel to Paris.’ narrow down the group to a few elements (in cluding fog ‘will’ but excluding akar ‘want’, (2) Részt akar venni a kiállításon. for example). I side with Kálmán C. et al. part. ACC wants take the exhibition.on (1989), however, who identify Hungarian aux ‘He/she wants to take part in the exhibition’ iliaries on the basis of syntactic and prosodic Both examples include a discontinuity, with behaviour; roughly, appearance in the kind of the auxiliaries fog ‘will.3 SG ’ and akar ‘wants’ construction illustrated in (1) and (2) above. I intervening between two parts of the complex regard verbs which participate in this construc verbs elutazni ‘to travel away’ and részt venni tion (in other words, which are collexemes of it ‘to take part’, respectively. Under the standard in terms of Stefanowitsch and Gries, 2003) as assumption that the finite auxiliaries are the auxiliaries, when and to the extent that they do roots here, taking lexical verbs as their infini so. However, this does not prevent them from tival complements, the simplest DG analysis being verbs, i.e. “auxiliary” is not viewed here incurs a projectivity violation: as a distinct (let alone closed) word class of Hungarian. (3) fog In section 2, I will present the relevant data, János utazni and make three observations against which the el Párizsba analyses will be matched. Section 3 compares four syntactic accounts, two each from the tra János el fog utazni Párizsba ditions of phrase structure grammar and de The goals of the paper are twofold. pendency grammar. Section 4 addresses the Firstly, I will compare possible analyses of relationship between rising and constructional the construction, and argue for a projective DG meaning. Finally, summary and conclusions account along the lines of Groß and Osborne follow in section 5.

118 Proceedings of the Second International Conference on Dependency Linguistics (DepLing 2013), pages 118–127, Prague, August 27–30, 2013. c 2013 Charles University in Prague, Matfyzpress, Prague, Czech Republic 2 Data and observations verb with the base verb” (Kiefer 2003: 17). Thirdly, it can be added that the VM + verb In this section, I make three observations about sequence tends to behave as a single phonolog the construction, which will serve as a basis for ical word, with the wordinitial stress of Hun evaluating analyses in section 3. These obser garian falling on the first syllable of the unit. vations are highlighted below for convenience. VM s come in two subgroups, illustrated by 1. There is a syntactic relationship between the expressions in (4) and (5). the verb modifier ( VM , e.g. el, részt ) ap (4) a. moziba megy pearing to the left of the root auxiliary cinema.to goes and the infinitive (e.g. utazni, venni , ‘[he/she] goes to cinema’ with the ni infinitive suffix) on its right. b. újságot olvas 2. There is also a syntactic relationship be newspaper. ACC reads tween the VM (e.g. el, részt ) and the root ‘[he/she] reads newspaper’ / ‘[he/she] auxiliary (e.g. fog, akar ). is engaged in newspaperreading’ 3. The three elements (i.e. the VM , the root (5) a. kimegy auxiliary and the infinitive) form a outgoes grammatical unit, which, however, is ‘[he/she] goes out’ subject to word order variation. b. elolvas awayreads 2.1 The link between VM and infinitive ‘[he/she] reads [to the end]’ The first, rather trivial observation is that in patterns like el fog utazni ‘he/she will travel Whereas the VM s of the complex verbs listed away’ and részt akar venni ‘he/she wants to in (4) satisfy an of the base verb, so take part’, there is a syntactic relationship be called verbal particles such as el ‘away’, be tween the first and the third element. This rela ‘in’ and ki ‘out’ fail to do so (cf. Kiefer ibid.). tionship is one of licensing: the socalled verb Nevertheless, there is considerable agreement VM modifiers ( el ‘away’, részt ‘part. ACC ’) could in the literature that the two types of s are not occur in these structures were it not for the amenable to essentially the same syntactic lexical verbs appearing in an infinitive form. analysis, cf. the analogous examples in (6). The two elements form a semantic unit with (6) a. moziba fog menni a higher or lower level of compositionality (cf. ‘[he/she] will go to cinema’ the oftcited example berúg ‘get drunk’, where b. újságot akar olvasni the VM be literally means ‘in’, and rúg literally means ‘kick’). In addition, it is noteworthy that ‘[he/she] wants to read newspaper’ there is often a morphological dependency be c. ki fog menni tween the two elements: for example, the t ‘[he/she] will go out’ accusative suffix of részt ‘part. ACC ’ is as d. el akarja olvasni signed by venni ‘to take.’ While morphological away wants. DEF .OBJ read dependencies are considered separable in prin ‘[he/she] wants to read it’ ciple from syntactic ones (cf. Mel’čuk 1988), there is a clear tendency for such dependencies In conclusion, it would be hard to deny that to hold between elements which are also syn there is a relationship between VM s and infini tactically related. tives in the construction under study. The link In Hungarian linguistics, the term “verb is evident at several levels of analysis includ modifier” 1 (also known as “preverb”) denotes ing the lexicon, morphology, syntax and se a category of elements with the following mantics. From a syntactic perspective, the rela properties: “(i) they occupy the position im tionship can be defined as licensing, a point mediately preceding the verb,2 and (ii) in the that will be taken up later in section 3. typical case they form semantically a complex 2.2 The link between VM and auxiliary 1 As a reviewer points out, the term may be misleading as Less immediately apparent is the fact that there VM s are not in fact modifiers (in the sense of being ad VM juncts). However, I still adopt it, following standard prac is also a syntactic relationship between the tice in Hungarian grammar (cf. É. Kiss, 2002: 67). and the root auxiliary. Although the two ele 2 At least in socalled neutral clauses, cf. section 2.3 .

119 ments are adjacent, adjacency alone is clearly the root auxiliary form a tightly integrated unit. insufficient to establish the link as syntactical For example, the epistemic adverb talán ‘per ly significant. For instance, in this obviously haps’ cannot occur between the VM and the contrived example , this and obviously have auxiliary (8a), only between the auxiliary and little to do with one another. the infinitive (8b) or externally to the VM + However, the following data strongly sug auxiliary + infinitive pattern (8c, 8d). gest that the VM and the root auxiliary are (8) a. *János el talán fog utazni Párizsba. more intimately related. b. János el fog talán utazni Párizsba. (7) A: János el fog utazni Párizsba? John away will.3SG travel Paris.to c. János talán el fog utazni Párizsba. ‘Will John travel to Paris?’ d. János el fog utazni talán Párizsba. B: Igen, el fog. ‘John will perhaps travel to Paris.’ yes away will.3SG Finally, the following coordination pattern also ‘Yes, he will.’ suggests the existence of a direct link between In speaker B’s utterance, the VM and the root the VM and the auxiliary. Coordinating el akar auxiliary together form a wellformed clause. and el is fog (where is means ‘also’) would This would hardly be possible in the absence hardly be possible if VM + auxiliary sequences of a direct syntactic relationship (more specifi were not grammatical units. 3 cally, a dependency) between them. In partic (9) J. el akar és el is fog utazni Párizsba. ular, the analysis in (3) is rendered unlikely, J. away wants and away also will travel Paris.to since it implies the possibility of eliding an ‘John wants to, and also will, travel to Paris.’ intermediate element (utazni ‘travel’) while preserving the phonological content of ele All in all, ellipsis and coordination facts, pros ments both above and below it in the tree. We ody, and the distribution of adverbs such as will see in section 3 that this goes against what talán ‘perhaps’ provide converging evidence seems to be a valid generalization about the that the adjacency between the VM and the root relevant cases of ellipsis. auxiliary is syntactically significant. Precisely A second argument for a direct syntactic how this can be incorporated in a DG analysis link between the VM and the root auxiliary is an issue to be addressed in section 3. comes from prosodic structure. As noted above, VM s immediately preceding their base 2.3 Evidence that the three elements form verbs form a single phonological word with a grammatical unit them; for example, 'elutazik ‘[he/she] travels Finally, a third observation about the construc away’ has a single stress assigned to the first tion is that the VM , the auxiliary and the infini syllable. Importantly, a similar situation holds tive form some kind of grammatical unit. In when the VM is followed by an auxiliary. For this regard, note first that strings such as el example, in 'el fog 'utazni ‘[he/she] will travel akar utazni and el fog utazni can be substituted away’, el and the first syllable of utazni are by oneword predicates with a similar dis stressed, while fog is unstressed, presumably course function (10, 11). because el and fog belong to the same phono logical word. Under the reasonable assumption (10) a. János el akar utazni Párizsba. that elements forming phonological words tend ‘John wants to travel to Paris.’ to be syntactically closely related, this suggests b. János elutazna Párizsba. that there is a direct link between el and fog in John away.travel. COND .3 G Paris.to the syntactic hierarchy. ‘John would [gladly] travel to Paris.’ Thirdly, the distribution of certain adverbs (11) a. János el fog utazni Párizsba. also supports the conclusion that the VM and ‘John will travel to Paris.’ b. János elutazik Párizsba. 3 As a reviewer observes, disjointed elements may appear John away.travel.3SG Paris.to in answer fragments, cf. German [Wem gefällt das? ‘Who likes that?] Mir gefällt das nicht ‘Not me.’ However, ‘John is [soon] travelling to Paris.’ speaker B’s utterance in (7) crucially includes the root Secondly, the strings mentioned can be coordi auxiliary, whereas in the German example, the root verb is elided. It seems plausible to suppose that remnants nated (12) or elided by (13). In the which do include the root must be continuous. latter example, pedig is a marker of topic shift.

120 (12) János el akar utazni és el is fog utazni P.ba. tary block that always appears in exactly the ‘J. wants to, and also will, travel to Paris.’ same form. Rather, it is subject to significant variation regarding the word order of its ele (13) J. el fog utazni Párizsba, Mari pedig Rómába. ments. In section 3, I will argue that this unit ‘J. will travel to Paris, and Mary to Rome.’ status combined with a high degree of flexibili Such facts are easiest to explain if the VM + ty can be best captured with the notion of cate auxiliary + infinitive pattern is treated as a nae as proposed by Osborne et al. (2012). grammatical unit. However, it is important to observe that the unit in question is highly flex 3 Competing analyses ible. In particular, the word order of its ele We are now in a position to assess competing ments is subject to variation, as demonstrated by the examples below. syntactic analyses of the construction. The main criterion for evaluation will be the extent (14) János el fog Párizsba utazni. to which they comply with the observations ‘John will travel to Paris.’ made in the previous section. Of the four ac (15) JÁNOS fog elutazni Párizsba. counts to be considered, the first two come ‘It is John who will travel to Paris.’ from the tradition of phrase structure grammar. These will be presented in 3.1 , followed by a As (14) shows (compared with (1)), the rela comparison of two DGbased solutions in 3.2 . tive position of the infinitive and its dependent is not fixed by the construction: Párizsba ‘to 3.1 Phrase structure grammar Paris’ may precede as well as follow its head In the last decades of the 20 th century, phrase utazni ‘travel.’ And as (15) illustrates, certain structure grammar enjoyed a virtual monopoly sentence types may also rearrange the order of in analyses of Hungarian word order, so much the VM and the auxiliary. When an identifica so that even those not committed to Choms tional focus (cf. É. Kiss 1998a) such as JÁNOS kyan generative grammar chose to adopt it for is present in the structure, it attracts the finite descriptive purposes. Thus in their classic pa auxiliary to its right, and the VM is attached to per on the system of Hungarian auxiliaries, the infinitive. More precisely, it is attached to Kálmán C. et al. (1989: 52) assigned the tree the infinitive which licenses it, a qualification diagram in (19) to the sentence below. made necessary by examples such as (17). (18) A MIGÉRT részt akart venni a kiállításon. (16) János el fog tudni utazni Párizsba. ‘MIGÉRT [name of Hungarian company] John away will be.able travel Paris.to wanted to take part in the exhibition.’ ‘John will be able to travel to Paris.’ (19) Sentence (17) JÁNOS fog tudni elutazni Párizsba. ‘It is John who will be able to travel to P.’ Subject Predicate Phrase

In Hungarian linguistics, examples such as Complement (14) and (16) are often called neutral clauses, whereas patterns like (15) and (17) are known Carrier Verb as nonneutral ones. Roughly, whereas a neu tral declarative clause answers the question Carrier Nistem What happened? or What is the situation? , a nonneutral one is felicitous under more spe A MIGÉRT részt akart venni a k.on cial communicative circumstances. The gene Dated as it undoubtedly is, the account is not ralization that VM s immediately precede the without merits. Firstly, it captures the intuition finite verb or auxiliary is constructionspecific. that the three elements form a grammatical unit Clauses with identificational foci, a negative (2.3 ): specifically, részt akart venni ‘wanted to particle, an interrogative pronoun, etc. display take part’ is analysed as a VP within the predi a different word order (see also section 4). cate phrase. Secondly, the relationship between To conclude this section, the facts are fairly the VM részt ‘part. ACC ’ and the infinitive venni complex but substitution, coordination and ‘take’ is signalled (cf. 2.1 ), with the two form ellipsis tests do suggest that the VM + auxiliary ing a constituent called “carrier” in the VP. + infinitive pattern forms some kind of gram On the other hand, the link betweeen the VM matical unit. However, this unit is hardly a uni részt ‘part. ACC ’ and the auxiliary verb akart

121 ‘wanted.3 SG ’ is not directly indicated, despite (21) S evidence from prosodic structure ('részt akart 'venni ), coordination ( részt akar és részt is fog TopicP PredP venni ‘wants to, and will, take part’) and the placement of adverbs. 4 Pred VP1

(20) a. *részt mindenképpen akart venni V1 VP2

b. részt akart mindenképpen venni V2 CaseP c. mindenképpen részt akart venni VM V2 d. részt akart venni mindenképpen ‘wanted to take part by all means’ János el i fog t i utazni Párizsba As shown in (20), the adverb mindenképpen This analysis has the advantage of being more ‘by all means’ has the same distribution visà restrictive, and therefore theoretically more vis részt akart venni ‘wanted to take part’ as appealing, than the proposal of Kálmán C. et talán ‘perhaps’ did with respect to el fog utazni al. (1989). 5 The price paid for this is the intro ‘will travel away’ in (8). This suggests that duction of underlying representations and részt and akart form a tightly integrated unit. transformations, which rival theories such as The biggest problem with (19) though is that LFG and HPSG reject on account of their per it violates the No Crossing Branches principle ceived lack of psycholinguistic plausibility and widely adopted in the tradition of phrase struc practical (computational linguistic) utility. ture grammar. Kálmán C. et al.’s flexible ap More importantly for the present discussion, proach to what passes as a wellformed tree is while (21) is consistent with the observation problematic because it grossly overgenerates that there is a syntactic link between the VM the set of possible sentences. In the absence of and the infinitive, and also goes some way to clearly defined restrictions on the emergence ward recognizing the relationship between the of discontinuities, any word order is predicted VM and the auxiliary, 6 it fails to reflect the unit to be possible, and the analysis is lacking ex status of the VM + auxiliary + infinitive pat planatory power. tern. To the extent that the argumentation in The second phrase structural analysis consi section 2.3 was sound, this puts the account at dered here is couched in transformational ge a disadvantage. nerative grammar. Rather than presenting a specific account found in the literature, I will 3.2 Dependency grammar attempt to come as close as possible to com As noted in the introduction, the simplest DG plying with the observations made in section 2 representation of the construction involves a as well as the basic assumptions of the theory. projectivity violation. 7 The analysis is repeated Also, the analysis will only make use of ideas in (22) below. that are present in one or another version of the (22) fog standard generative model of Hungarian (see in particular É. Kiss, 1998b, 2002). János utazni Transformational generative grammar al el Párizsba lows one to recognize the link between the VM János el fog utazni Párizsba and the infinitive at an underlying level of re presentation, and to let movement rules pro duce the surface word order. Thus, under the account in (21), the VM and the (nonfinite) 5 The tree in (21) is simplified in ways that do not verb form a constituent at “deep structure” be crucially affect the argumentation. Technically, the VM is fore the VM is moved out of the VP into a in Spec,PredP, and Pred 0 may be the landing site of the finite verb (cf. É. Kiss 2008: 131). Thus, the VM and the phrase called PredP, cf. É. Kiss (2008). finite verb may enter a SpecHead configuration. 6 This is so if the VM and the auxiliary are in a SpecHead relationship, cf. footnote 5. 7 In Nivre’s formulation, “A dependency graph satisfies the constraint of projectivity with respect to a particular 4 The kind of ellipsis shown in (7) works perfectly with linear order of the nodes if, for every arc h [head]→ d verbal particles (such as el ’away’, ki ’out’, etc.) but it is [dependent] and node w, w occurs between h and d in the rather marginal with VM s like részt . linear order only if w is dominated by h” (2005: 10).

122 The diagram signals the syntactic relationship core ought to be a “network within the net between the VM and the infinitive, with utazni work”, with its internal structure describable ‘travel’ identified as the head of el ‘away.’ Se by a continuous set of dependencies. condly, the root auxiliary, the infinitive and the For this reason, and the further points made VM form a dependency chain (cf. Hudson in 2.2 , I propose the following representation 1990: 99), hence a unit of DG. However, the of the syntactic structure of (1), following tree in (22) implies that the adjacency of el and Groß and Osborne (2009). 9 fog is merely a fact of word order; there is no (24) fog direct dependency between them. As is well known, versions of DG can be János el utazni g built either with or without the assumption of Párizsba projectivity (cf. Nivre 2005: 10). Here, what János el fog utazni Párizsba needs to be established is whether there are any empirical reasons for rejecting (22). As Groß and Osborne (2009: 53) crucially sepa suggested in 2.2 , the main counterargument rate the notions of governor and head. A comes from the following type of ellipsis: word’s governor is the word licensing its ap pearance. By contrast, its head is the word that (23) A: János el fog utazni Párizsba? immediately dominates it. Although by default, John away will travel Paris.to the governor and the head are the same word, ‘Will John travel to Paris?’ the two functions may also be associated with B: Igen, el fog. different nodes of the structure. In such cases, yes away will.3SG however, the head must be higher up in the ‘Yes, he will.’ tree than the governor; in other words, only “rising” can occur, not “lowering.”10 Ellipsis is a hugely complex phenomenon, and The analysis in (24) expresses that the gov a fully predictive account of when it is or is not ernor of el is utazni ; this is marked by the g possible may be an elusive research objective.8 subscript of the latter. The dependency pro However, it seems fairly clear that given the duced by rising is distinguished by a dashed structure in (22), one does not expect utazni to dependency edge. Importantly, rising is un be elided while both its head fog ‘will.3SG ’ and derstood only metaphorically here, since Groß its dependent el ‘away’ are unaffected. and Osborne’s approach is strictly non According to Rosta (2006: 176), “[e]llipsis derivational (cf. Groß and Osborne, 2009: 54). involves the deletion of the phonological con Hence, there is no such claim that the head of tent of some syntactic structure, and it seems to el should have been utazni at an underlying operate rather as if (the phonology of) a branch level of representation. For arguments support of the syntactic tree were snipped off. Thus if ing risingbased analyses of linguistic pheno the phonological content of one node is de mena, see Groß and Osborne (2009: 56–64). leted, then so must be the phonological content By separating governor and head, the analy of all nodes subordinate to it.” Although this sis conforms to the observation that the VM is formulation is almost certainly too restrictive, syntactically related to both the infinitive and as Rosta himself concedes (note especially the root auxiliary. Especially significant is the gapping phenomena, cf. (13) and Osborne, fact that the kind of ellipsis seen in (23) fol 2005: 275–280), it does seem to be a valid ge lows naturally from the proposal, which was neralization for the case at hand. When a sen not the case with (22). What is yet to be seen, tence is reduced to a combination of elements though, is whether the grammatical unit status including the root (let us call it its “core”), the of the VM + auxiliary + infinitive pattern is accounted for under these assumptions. 8 This is especially true for crosslinguistic predictions. As a reviewer remarks, similar word order configurations 9 For a parsingoriented approach along similar lines, see to the ones discussed in this paper exist in French, cf. Barta et al., 2004. Jean l’a vu ‘John has seen it’/’John saw it’, where the 10 Groß and Osborne’s concept of rising has many prece object l’ ‘it’ is licensed by the past participle vu dents in the literature including Duchier and Debusmann, ‘seen’ but it precedes and arguably depends on the aux 2001, Gerdes and Kahane, 2001, and Hudson, 2000 (cf. iliary a ‘has.’ Still, the past participle cannot be elided Groß and Osborne, 2009: 51). I adopt their approach (*Jean l’a vu ). I assume that this is motivated by inde because of its descriptive appeal; other frameworks may pendent properties of French; parallel structures in differ be seen as better developed from a model theoretic or ent languages need not permit the same kinds of ellipsis. computational linguistic perspective.

123 Broadly speaking, the issue is what kinds of what is important is that el fog utazni is a cate units larger than the word a syntactic DG anal na (marked by italics in (24)). Hence, the anal ysis can recognize. One traditional unit type is ysis conforms not only to the observations the DG equivalent of a phrase or constituent. made in 2.1 and 2.2 but also to the point that In contrast with phrase structure grammar, DG the three elements form a grammatical unit treats constituents as units implied by a net (2.3 ). Moreover, since the concept is defined in work of wordtoword relations (Hudson, terms of dominance relations only, it is suffi 2007: 121) rather than as unique nodes of the ciently flexible to accommodate word order tree. A theoryneutral definition of constitu variation. Thus, the examples in (14) and (15) ents, also applicable to DG, is as follows: can receive the following analyses, in which the three elements still form catenae. 11 (25) Any node plus all the nodes that that node dominates. (Osborne, 2005: 254) (27) fog

In (24), there are only two multiword consti János el utazni g tuents: utazni Párizsba , and János el fog utazni Párizsba Párizsba . By contrast, el fog utazni does not János el fog Párizsba utazni count as a constituent, since it does not include all the nodes that its root ( fog ) dominates. (28) fog Another established unit type recognized by DG is the dependency chain, i.e. a continuous JÁNOS utazni non line of h → d relations. Accord el Párizsba ing to Hudson (1990), “a word’s phrase con JÁNOS fog el utazni Párizsba sists of the union of all its downchains.” In (24), the following complete downchains of The proposal results from a happy marriage of fog ‘will.3 SG ’ can be identified: fog → János ; empirical and theoretical considerations. On fog → el ; and fog → utazni → Párizsba . the one hand, there is strong empirical evi Again, el fog utazni as analysed in (24) is not dence for a direct link between the VM and the captured by the concept. root auxiliary (cf. 2.2 ), as signalled in (24) and Recent years, however, have seen the recog (27). On the other, the independently moti nition of a new, more inclusive unit type im vated theory of rising and catenae provides a plied by the dependency network. Building on simple way of accounting for this as well as previous work (notably O’Grady, 1998, and other relevant observations. Osborne, 2005), Osborne et al. (2012: 359) Also noteworthy is the fact that DG fares introduce a unit type called catena (Latin for much better than phrase structure grammar in ‘chain’) , defined over a Dtree as follows: expressing the unit status of the VM + auxiliary + infinitive pattern. Constituencybased ap (26) A word, or a combination of words which proaches either struggle to reflect this intuition is continuous with respect to dominance. but fail to produce a satisfactory account, cf. The catena concept is more inclusive than that Kálmán C. et al. (1989), or ignore the issue of constituents/phrases because it does not re altogether, cf. the analysis couched in trans quire the unit to include all the nodes dominat formational generative grammar. By contrast, ed by a given element. Also, it is more inclu the proposed DG account is flexible and re sive than traditional dependency chains since it strictive enough to be faithful to the facts while also captures combinations of words consisting also having strong theoretical appeal. of a head and multiple dependents (schemati At the same time, a possible objection to the cally: d1 ← h → d2). Finally, single words also rising analysis still remains. In particular, un count as catenae, which is again an extension der the assumption that dependencies ought to on the previous concept of chains. have an associated meaning or function, it is In this paper, it is not my goal to defend the catena concept (for this, see e.g. Osborne and 11 Groß, 2012, and Osborne et al., 2012). Suffice Two reviewers make the point that VM s may be analys able as . If this is indeed the case, then the vertical it to say that there is considerable evidence projection lines of VM s have to be removed under the (especially from ellipsis, analytic predicates, conventions of Groß (2011: 60). Since the dependency and idioms) suggesting that the concept is edges would still be the same, the basic validity of the highly operational. For the present discussion, analyses is not at stake. Whether a clitic analysis is ne cessary is an issue left for future research to resolve.

124 yet to be seen if the dependency created by tivation (if not justification ). Since they contri rising also conforms to this requirement. In bute a key aspect of constructional meaning what follows, I argue that rising has a key role (making a whquestion what it is), their promi in coding aspects of constructional meaning. nent and distinctive linear position is natural. And while in nonprojective versions of DG, 4 Rising and constructional meaning the attested word order would not entail a de pendency between the whword and the root Since the inception of modern DG, the idea auxiliary, there are independent reasons for that dependencies have an associated meaning subscribing to that account (e.g. the ellipsis in or function seems to be shared by most depen What can you see and what can’t you? ). In the dency grammarians. Tesnière already claimed final analysis, the dependency created by ex that “there is never a structural connection traction or rising can be seen as “bringing to without a semantic one” (1959: 44, my transla gether” both a semantic property (the special tion). And while Hudson rejects the view that function of whquestions as endowed to the dependencies are primarily a matter of mean construction by the whelement) and formal ing, he does contend that meaning is one of the ones (distinctive word order and prosody). properties that they “bring together”, “along Generalizing from this, one may hypothes with word order, agreement, case choice, and ize that (certain) dependencies created by ris so on” (2007: 130). Witness also the conver ing play a part in coding “global” aspects of gence between DG and construction gram constructional meaning, independently of any mar/CxG (Osborne and Groß, 2012, and refer “local” (lexically motivated) semantic relation ences therein), which hinges on the notion that ship between the two elements. More specifi dependencies have a semantic side to them. cally, there may be a significant correlation After all, the basic tenet of CxG is that lexicon between rising and sentence types (grounded and syntax form a continuum, with syntactic in illocutionary force distinctions).12 constructions as well as morphemes, lexemes, As we return to Hungarian, it seems plausi etc. described as pairings of meaning and form. ble to develop a similar account of the seman Exceptions, however, have also been al tic background to the word order of VM s. To lowed by some theorists. Thus, Hudson argues begin, note that whereas VM s immediately pre that the “subject or object of a verb need not cede their base verbs in neutral positive declar have any semantic relation to that verb at all” ative clauses lacking auxiliaries (30a), they (2007: 130), It seems to be raining being an follow them in sentence types which depart example. Even more importantly for the from this function in terms of illocutionary present discussion, he posits an “extractee” force or polarity: dependency between what and can in the sen tence below (2007: 131). (30) a. János elutazott Párizsba. John awaytravelled.3SG Paris.to (29) ‘John travelled to Paris.’ b. Hova utazott el János? where travelled.3SG away John?

‘Where did John travel?’ c. PÁRIZSBA utazott el. English nonsubject whquestions involve ris ‘It is to Paris that he/she travelled.’ ing according to Osborne and Groß (2009: 52). d. Nem utazott el Párizsba. In What can you see? , can is the head of what not travelled.3SG away Paris.to just as Hudson’s surface analysis (above the ‘He/she did not travel to Paris.’ string of words) has it. Therefore, it is signifi cant that for Hudson, “such dependencies [as extractee] are concerned with very little but 12 A similar reasoning may apply to other “meaningless” word order, and have little claim to semantic dependencies such as the subjects of English and German justification” (2007: 131). This suggests that weather verbs (it rains , es regnet ). As Jespersen remarks, “the need for this pronoun [English it , German es, etc.] the dependency created by rising perhaps does was especially felt when it became the custom to express not, and need not, have an associated meaning. the difference between affirmation and question by Clearly, though, the word order of English means of word order (er kommt, kommt er? ), for now it whelements can at least receive semantic mo- would be possible in the same way to mark the difference between es regnet and regnet es? ” (Jespersen, 1924: 25).

125 The function of a neutral positive declarative positive negative clause such as (30a) is to profile the occur past elutazott nem utazott el rence of an event relative to a mental space (in ‘he/she travelled ‘he/she did not the sense of Fauconnier 1985). Here, the lis away’ travel away’ tener learns about the occurrence of a travel present elutazik nem utazik el ling event in the past (a mental space distinct ‘he/she is travel ‘he/she is not from the present), involving John as the mover ling away’ travelling away’ and Paris as the goal. At the core of the con future el fog utazni nem fog elutazni struct is the predicate elutazott , which has the ‘he/she will tra ‘he/she will not function of a schematic clause. It may also vel away’ travel away’ stand by itself meaning ‘He/she travelled away.’ With respect to this clausal core, János Table 1. Polarity and word order in Hungarian ‘John’ and Párizsba ‘to Paris’ simply elaborate In all three tenses, VM + finite verb/auxiliary the mover and the goal, respectively. order is associated with positive polarity, and a By contrast, (30b,c,d) depart from the func different linearization with its opposite. If the tion of (30a) in one or another way. (30b) is VM did not precede the root auxiliary in the used to inquire about John’s destination; (30c) future tense, fog elutazni and nem fog elutazni identifies Paris as the goal to the exclusion of would stand in opposition, and the semantic other possibilities; and finally, (30d)’s speaker contrast would be coded less saliently as well denies the occurrence of the travelling event. as less regularly across the paradigm. Although not all deviations from the neutral To conclude, I have argued in this section positive declarative clause type are signalled in that the word order (and assuming projectivity, this way, it is reasonable to suggest that the the rising) of VM s codes important global as inversion of VM and finite verb plays a promi pects of constructional meaning. Firstly, it es 13 nent role in coding clause type distinctions. tablishes a formal parallel between catenae From this perspective, the word order (and with analogous functions (cf. the lefthand col by implication, the rising) of the VM in the umn in Table 1). Secondly, the VM + auxiliary Hungarian auxiliary construction can be moti pattern of neutral declarative clauses allows for vated by two interrelated facts. Firstly, the aux a salient and regular way of coding sentence iliaries in question set up mental spaces in type distinctions (cf. the three rows in the ta which an event unfolds. For example, fog ble). It seems likely that other “meaningless” ‘will’ sets up a space for talking about future dependencies such as Hudson’s “extractee” events, akar ‘want’ a space for discussing and the subject of English weather verbs (cf. somebody’s intentions, etc. Since mental spac footnote 12) may receive a similar motivation. es are also implicit in the semantic structures of oneword predicates, it is natural to roll 5 Conclusions spacebuilding verbs and verbs denoting events in those spaces into complex predicates. As In this paper, I made the case for a projective noted in 2.3 , VM + auxiliary + infinitive pat DG analysis of the Hungarian auxiliary con terns have a function analogous to that of VM + struction. In 2, evidence was presented that in V sequences. The word order of the VM in the VM + auxiliary + infinitive patterns, there were former can be seen as a reflex of complex pre syntactic links both between the VM and the dicate formation motivated by such analogies. infinitive and between the VM and the aux Secondly, the resulting word order has the iliary. In addition, it was argued that the three advantage of allowing for a salient and regular elements formed a grammatical unit. In 3, four way of expressing clause type distinctions. analyses were compared, with the result that Consider the following parallels: only the DG account based on rising and cate nae conformed to all of the above observa tions. Finally, section 4 highlighted aspects of 13 Compare also Goldberg’s (2006: 166–182) account of constructional meaning and analogy as moti English subjectauxiliary inversion (SAI). According to vating factors for the form of the construction. Goldberg, SAI as a “sytematic difference in form” signals a “systematic difference in function” (178) visàvis pro totypical sentences (which are positive and declarative). Acknowledgments However, “it is certainly not the only possible device” The research reported here was supported by (181) in this capacity. For a more detailed account of English and Hungarian inversion, see Imrényi (2012). the Hungarian Scientific Research Fund

126 (project no. K100717). I also thank Timothy Imrényi, András. 2012. Inversion in English and Osborne and three anonymous reviewers for Hungarian: comparison from a cognitive pers many insightful comments and suggestions. pective. In: Hart, Christopher (ed.), Selected All remaining errors are my own. Papers from UK-CLA Meetings , 1. 209–228. Jespersen, Otto. 1924. The philosophy of gram- References mar. George Allen & Unwin Ltd, London. Barta, Csongor, Ricarda Dormeyer and Ingrid Kálmán C., György, Kálmán László, Nádasdy Fischer .2004. Word order and discontinuities in nd Ádám and Prószéky Gábor. 1989. A magyar a DG for Hungarian. Proceedings of the 2 segédigék rendszere. Általános Nyelvészeti Conference on Hungarian Computational Tanulmányok XVII: 49–103. Linguistics . Juhász Nyomda, Szeged. 19–27. Kenesei, István. 2008. A segédigék. In: Kiefer Duchier, Denys and Ralph Debusmann. 2001. To Ferenc ed. Strukturális Magyar Nyelvtan 4. A pology dependency trees: A constraint based ac count of linear precedence. Proceedings from the szótár szerkezete. Akadémiai Kiadó, Budapest. th 615–620. 39 annual meeting of the ACL. 180–187. Kiefer, Ferenc. 2003. A kétféle igemódosítóról. É. Kiss, Katalin. 1998a. Identificational focus ver Nyelvtudományi Közlemények 100: 177–186. sus information focus. Language 74: 245–273. Lengyel, Klára. 2000. A segédigék és származéka É. Kiss, Katalin. 1998b. Mondattan. In: É. Kiss ik. In: Keszler, Borbála (ed.), Magyar gram- Katalin, Kiefer Ferenc and Siptár Péter, Új ma- matika. Nemzeti Tankönyvkiadó, Budapest. gyar nyelvtan. Budapest: Osiris. 1–184. 252–258. É. Kiss, Katalin. 2002. The syntax of Hungarian. Mel’čuk, Igor. 1988. Dependency Syntax: CUP, Cambridge. Theory and Practice. The SUNY Press, Alba É. Kiss, Katalin. 2008. Tagadás vagy egyeztetés? ny, N.Y. Magyar Nyelv 104: 129–143. Nivre, Joakim. 2005. Dependency grammar and Fauconnier, Gilles. 1985. Mental spaces: Aspects dependency parsing . Vaxjo University. of meaning construction in natural lan- O’Grady, William. 1998. The syntax of idioms. guage. MIT Press, Cambridge MA. Natural Language and Linguistic Theory 16: Gerdes, Kim and Sylvain Kahane. 2001. Word or 279–312. der in German: A formal dependency grammar Osborne, Timothy. 2005. Beyond the constituent: A using a topology model . Proceedings from the dependency grammar analysis of chains. Folia 39 th annual meeting of the ACL. 220–227. Linguistica 39: 251–297. Goldberg, Adele. 2006. Constructions at work: Osborne, Timothy and Thomas Groß. 2012. Con the nature of generalization in language . structions are catenae. Construction Grammar OUP, Oxford. meets Dependency Grammar. Cognitive Lin- Groß, Thomas. 2001. Clitics in Dependency Mor guistics 23 (1): 163–214. phology. Depling 2011 Proceedings : 58–68. Osborne, Timothy, Michael Putnam and Thomas Groß, Thomas and Timothy Osborne. 2009. To Groß. 2012. Catenae: Introducing a novel unit of ward a practical dependency grammar theory of syntactic analysis. Syntax 15 (4): 354–396. discontinuities. SKY J. of Ling. 22: 43–90. Rosta, Andrew. 2006. Structural and distributional Heine, Bernd. 1993. Auxiliaries: cognitive forces heads. In: Kensei Sugayama and Richard Hud and grammaticalization . OUP, Oxford. son (eds.), Word Grammar. New perspectives on a theory of language structure . Conti Hudson, Richard. 1990. English Word Grammar . nuum, London. 171–203. Blackwell, Oxford. Stefanowitsch, Anatol and Stefan Th. Gries. 2003. Hudson, Richard. 2000. Discontinuities. In: Ka Collostructions: Investigating the interaction be hane, Sylvaine (ed.), Les grammaires de tween words and constructions. International dépendance. Traitement automatique des Journal of Corpus Linguistics 8 (2): 209–43. langues 41 . Hermes, Paris. 7–56. Tesnière, Lucien. 1959. Éléments de syntaxe Hudson, Richard. 2007. Language networks. The structurale. Klincksieck, Paris. new Word Grammar . OUP, Oxford.

127