A CORPUS STUDY OF GREEK BARE SINGULARS: IMPLICATIONS FOR AN ANALYSIS

Neste artigo apresentamos os resultados de um estudo de corpus sobre o singular nu no grego. Este estudo disponibiliza pela primeira vez dados de corpus que inclui todo o conjunto de verbos que têm sido apontados como permitindo o nominal singular contável nu (BSCNs) em grego. Primeiro, discutimos duas das classes de verbo que permitem o BSCNs, verbos de criação e construções existenciais. Em seguida, argumentamos que BSCNs não são neutros para número e que eles parecem ter um estatuto argumental, nos levando então a argumentar que uma análise de pseudoincorporação “estrita” à La Espinal & McNally (2011), por exemplo, não encontra apoio no grego. Nossos dados e análise parecem indicar que um tipo de pseudoincorporação “liberal” no sentido de Dayal (2011) poderia fornecer uma análise para o grego.


Introduction
Languages that allow for bare singular count nouns (BSCNs) in the object position have given rise to different proposals in the semantic literature (Borthen 2003;Farkas & de Swart 2003;Espinal & McNally 2011;Dayal 2011), the majority of which embrace semantic incorporation. 2 A fi rst step towards an analysis has often been identifying the class(es) of verbs that license bare nominals.These verbs seem to form a spectrum, from which each language slices its own portion.In this paper we offer data from corpora that inform the theoretical decisions to be made.
In particular, in Greek, the language under study here, the scarcity of "real" data has led us to conduct a corpus study in order to bring fresh data into the ongoing discussion.This is in line with a recent change in perspective (see e.g. the discussion in Gilquin & Gries 2009), that recognizes that informally collected linguistic acceptability judgments are not the only data that would qualify as linguistic data, because a multitude of factors might infl uence them (Schütze 1996).Thus, corpus data should be made available as well and be treated as equally important.
Greek is a language that has both defi nite and indefi nite determiners.The defi nite article is present in a great variety of contexts, having one of the most expanded uses from a cross-linguistic point of view, appearing with demonstratives, possessives, generics and proper names among other contexts.As Alexiadou et alii (2007:64) claim, the ability of nominals to occur 'bare' as arguments (of verbs) is quite restricted. 3 According to Chierchia's (1998) Nominal Mapping Parameter, languages differ with respect to whether nominals are mapped directly as semantic arguments (type e), or semantic predicates (type <e,t>).His Nominal Mapping Parameter is implemented in terms of the binary features [+/-arg] and [+/-pred] that differentiate languages: Greek has been claimed to be similar to Romance languages in exhibiting the features [-arg] [+pred] (Sioupi 2002;Lazaridou-Chatzigoga 2011).According to Chierchia, in this type of languages nouns are mapped onto predicates 3 It is important to note that the bare singular in Greek is not compatible with a kind-level predicate like be extinct: #Dhinosavros ehi afanisti.
dinosaur is extinct Generic kind-level statements are expressed with defi nite plurals, as below: I dhinosavri ehun afanisti.
the dinosaurs are extinct 'Dinosaurs are extinct.' and since by defi nition predicates cannot appear in argument positions, this group of languages should disallow bare arguments.The data in Romance and Greek suggests that this prediction is not borne out and that in these languages we do fi nd bare nominals in object position, as has been observed in the literature on Greek (Sioupi 2002, Alexopoulou & Folli 2010, Lazaridou-Chatzigoga 2011), and similarly, on Catalan and Spanish (Espinal & McNally 2011, henceforth E & M), Brazilian Portuguese (Schmitt & Munn 1999) and Romanian (Dobrovie-Sorin et alii 2006).This paper presents the results of a corpus study we conducted on the Hellenic National Corpus (HNC) in order to obtain data.The study has had the following outcomes: a.It confi rmed that Greek allows for bare nominals in the object position 4 .
b.It made corpus data for the fi rst time available for the whole set of verbs that allow for BSCNs in Greek.
c.The research based on the results suggested a reconsideration of how a possible analysis of the Greek data should look like.
The structure of the paper is as follows: in section 1 we present the methodology and the results of the corpus study.Section 2 is devoted to the discussion of some of the results, focusing on a subset of the verbs found to license BSCNs.It also raises the issue of number neutrality. 4 Bare singulars in canonical subject position are not licit.Thus, the following is ungrammatical: i. *Pedhi irthe.
child came When items are stressed in contrastive focus and/or have undergone movement, as below, you can though fi nd a bare noun in a subject position: ii.ton exetase YATROS.him-CL examined-3SG doctor-NOM 'A doctor examined him.'(ex.from Kolliakou 2003) We will show later on in this paper that in some non-canonical subject positions one can fi nd bare nouns.
Furthermore, section 3 poses theoretical questions that arise from the discussion of the data focusing on the issue of whether Greek BSCNs are arguments and on their referential status.Finally, we present the conclusion.

Data collection from the Hellenic National Corpus and results
We will fi rst discuss the methodology used in the study and then present the results, with examples of data for each verb type found to license bare singulars.

Methodology
The Hellenic National Corpus (1999, henceforth HNC) is a corpus that is tagged for parts of speech (POS-tagged).It contains data of written Modern Greek language (book, Internet, newspaper, magazine, miscellaneous) dating from 1990 onwards (47.013.924words).It provides a query system, which helps one to make queries by selecting one to three successive words, lemmas or POS, adjusting the distance between every two of them.Additionally, it allows for selecting specifi c parts from the corpus and making queries in the resulting sub-corpora.
As this corpus enables making two-word/lemma/POS queries, the fi rst parameter was set to the POS Verb and the second one to Noun/common as well as the distance between them to 0, so as to get all V+Ncommon combinations.This is the closest to the desirable V+BSCN combination that could be asked for.Due to HNC's limitation of only giving up to 2.000 hits for each query, its facility of making sub-corpora was exploited, creating as many sub-corpora as required in order to pick out every single V+ Ncommon combination.Importantly, the corpus lacks tags for Number and for the mass/count distinction.Consequently, the results contained too much pollution, such as V+BPlural and V+Nmass combinations, which made 'cleaning' them of these constructions manually a necessary task.After following the above procedure, we were left with all V+BSCN combinations.

the MP had car but was-3SG broken
The MP had a car, but it was broken.' (4) Mu ipe oti arketo kero epsahne spiti stin periohi.

Discussion of results
On the basis of the above results, we are going to focus on the following issues: a) Greek BSCNs are licit with creation verbs, as seen in example (8).
b) Greek BSCNs are licit in existential constructions, as seen in example (9).
The fi rst observation leads to a comparison of Greek with Norwegian (Borthen 2003), while the second one to a comparison with Catalan/ Spanish (E & M 2011).

Creation verbs
Greek, like Norwegian (Borthen 2003), allows BSCNs with creation verbs.Borthen's examples (2003:154-155) are all in the progressive aspect, as can be seen in the English translations she provides.Here are some of them: (10) Hun baker kake.
she bakes cake 'She is baking a cake.' (11) Kanarifuglen min bygger rede.canary.bird-DEFmine builds nest 'My canary bird is building a nest.' Similarly, in the Greek sentence in (12) the verb can be translated in the progressive aspect too, although there is no aspectual marking in the verbal morphology.
(12) Tus ftiahni dhromo.them-CL is.constructing/constructs road 'S/he is constructing a road for them.'However, in Greek there is no restriction with respect to aspect when BSCNs appear with creation verbs.This is demonstrated by the following examples, where the aspect is perfective: (13) Ehtise erghostasio pu ihe sigekrimeni hrisi.
built-3SG-PRF factory that had-3SG specifi c use 'S/he built a factory that had a specifi c use.' Judging from the aspect present in the Greek examples above, we argue that Greek BSCNs are licit with creation verbs regardless of the perfective/imperfective aspect of the verb.Initially we could argue that creation verbs take an incremental theme object, which means that the object of creation verbs does not exist when the activity of creation takes place (see also Borthen 2003; Le Bruyn 2012) and, thus the noncanonical nominal argument might be expected.Nevertheless, we see that in Greek creation verbs can also appear with BSCNs in an aspect that indicates the completion of the event the verb describes, that is, in perfective aspect.We hypothesize that creation verbs originally combined with BSCNs only in imperfective aspect, but over the time they got generalized to the perfective aspect too, starting to pattern like other V+BSCN combinations, in which the verb can freely appear in both aspects, e.g.foresa/forusa fusta 'I put (past) on/was wearing a skirt'.
Based on the small number of data that Borthen (2003) provides and the fact that her observations rely on the English translations, we can only preliminarily conclude that Greek seems to impose no restriction with respect to aspect in creation verbs and, thus, be different in this respect from Norwegian.

Existential constructions
In the corpus we fi nd data that show that Greek existential constructions allow for BSCNs, an environment that receives special attention in the analysis of Catalan/Spanish data in E & M (2011).Specifi cally, in Greek we fi nd BSCNs in two existential constructions.The fi rst one is literally composed by the verb ehi 'has', as we saw in (9) above, repeated below as ( 16) for convenience: (16) Ehi ghamo sto dhiplano horio.
has wedding in.the next village 'There is a wedding in the next village.'Interestingly, as we see in the above example, no clitic like the Catalan hi (see a similar example in ( 17)) or expletive item like the English there (see translation of ( 17)) is present: (17) Avui hi ha casament.
today there has wedding 'Today there is a wedding.' (ex.from E & M 2011:121) Moreover, these contexts provide a ground to test whether BSCNs in Greek are number neutral.Following E & M (2011) for a diagnostic of number neutrality in a language that is not like Hungarian (Farkas & de Swart 2003) -where we fi nd predicates like the English collect to take BSCNs -we see that the continuation of the following discourse is unsuccessful: (18) Ehi ghamo sto dhiplano horio.#Enan stin eklisia ke enan sto dhimarhio.has wedding in.the next village one in.thechurch and one in.thetown.hall'There is a wedding in the next village.One is in the church and one in the town hall.'It is worth noting that this is not a special feature of BSCNs in existential constructions, but a characteristic of BSCNs in Greek in general: they are not number neutral, but rather receive an atomic interpretation (cf.Alexopoulou & Folli 2010), as can also be seen in the infelicitous continuation of the discourse below: (19) psahno aftokinito; #ena mikro ya tin poli ki ena fortighaki ya ekdhromes am.looking.for/look.for.1SGcar one small for the city and one van for trips 'I'm looking for a car.A small one to ride in the city and a big one for trips.' (ex.from Alexopoulou & Folli 2010) Respectively, Dayal (2011) argues that non-canonical bare singular complements in Hindi are not number neutral, which seems to be the fi rst indication for considering a "liberal" pseudoincorporation treatment à la Dayal for BSCNs in Greek.
The second existential construction that was found to allow for BSCNs is the one formed with iparhi 'exists', as seen below (for a discussion of existentials in Greek see Koufaki 2012): (20) Kato apo tin skini iparh limni 400 tetraghonikon metron.
below from the stage exists lake 400 square meters 'Below the stage there is a lake of 400 square meters.' The two existential constructions in question differ as to the case the bare noun is assigned.When a bare noun appears with ehi, it appears in the object position and, thus, it receives accusative case, while, when a bare noun appears with iparhi, it appears in the subject position and, thus, in the nominative case. 7o sum up, BSCNs in Greek appear in two existential constructions with different syntactic structures, that is, as objects of ehi 'has' and subjects of iparhi 'exists'.Also, compared to the corresponding Catalan construction, the existential construction with the predicate ehi involves no clitic.Greek existential constructions seem to suggest a different syntactic analysis then, which would have implications for their semantics.Finally, BSCNs in Greek are not number neutral.All the above constitute the fi rst piece of evidence that a "strict" pseudoincorporation like E & M's semantic analysis of Catalan/Spanish BSCNs might not be suitable for Greek BSCNs.

Interim summary and conclusions
As the corpus study revealed, Greek displays a large set of verb classes that license BSCNs.Crucially, a considerable overlap between Greek and Catalan/Spanish verb classes is observed, including transfer, ownership, intensional, usage verbs as well as existential constructions, all subsumed by the so-called 'HAVE'-predicate class (E & M 2011).Considering this overlap, BSCNs in Greek could be accommodated along the lines of E & M's (2011) analysis of 'HAVE'-predicates combining with BSCNs, and thus, be treated as an instance of pseudoincorporation (for a fi rst discussion see Lazaridou-Chatzigoga 2011).We have already seen two barriers to this idea, i.e. that Greek existential constructions are syntactically different from Catalan/Spanish, and that BSCNs are not number neutral.In what follows we will be dealing with the argument and referential status of BSCNs in Greek in order to determine whether we could adopt a pseudoincorporation for Greek BSCNs.

Argumental and referential status of Greek bare singulars
If bare singulars occupy the object position, then one would like to know whether they have argumental status or not.Pseudoincorporation analyses claim that bare nouns do not have an argumental status, but that they are rather incorporated nominals.Treating BSCNs in Greek as arguments seems to be suggested on the basis of differences between Catalan/Spanish and Greek that are observed on a wide range of contexts, that license Greek BSCNs, but not Catalan/Spanish.The contexts include the following: a) the subject position of passivized 'HAVE '-predicates, as in (21), the controller of an implicit subject, as in ( 22), locative modifi ers, as in ( 23), and the subject position of secondary predicates, as in (24).What is special about the contexts is that they allow for a bare singular, even though it seems to be in a subject position, a position, where BSCNs are in general not licit in Greek.We already saw that BSCN can appear in the subject position of the existential iparhi.The above contexts show that bare singulars in Greek can function as subjects.Given that subjects are arguments, this leads us to conclude that BSCNs are to be treated as arguments.
Furthermore, we see that BSCNs can be modifi ed by adjectives similar to the ones that can modify canonical arguments in object position, such as anihtohromi 'light-colored' in the following example (see also Alexopoulou & Folli 2010) Incorporated nominals are claimed to be referentially weak.If we argue though for an argumental status of BSCNs in Greek, that would have the following consequence: the contents of subject positions are normally associated with a referent, so bare singulars in Greek would seem to be able to instantiate discourse referents -at least in some contexts.This is confi rmed by the continuation of the discourse, as illustrated below: (26) Htes to vradhi dhothike dheksiosi i stin presvia tis Vulgharias.Dhen tin i efharistithike omos kanis, yati dhen ihan krasi.yesterday the evening was.given reception at.the embassy the-GEN Bulgaria-GEN notit-FEM-CL enjoyed-3SG though nobody because not had-3PL wine 'Yesterday evening, there was a reception at the Bulgarian embassy.Nodoby enjoyed it though, because they didn't have wine.'If bare singulars in Greek are arguments, then we cannot adopt a pseudoincorporation analysis in the style of E & M (2011) or Farkas & de Swart (2003).It remains open though whether Dayal's "liberal" kind of pseudoincorporation could provide a potential analysis for BSCNs in Greek.

Conclusion
We conclude that a) BSCNs with creation verbs are not limited to the progressive aspect as in Norwegian, b) existential constructions are different from Catalan/Spanish, so a different syntactic analysis might have implications for the semantics.Furthermore, c) BSCNs are not number neutral, and d) BSCNs in Greek can function as subjects and seem to have a referential status in this position that makes it easier to associate a discourse referent to them.They furthermore allow modifi cation in a similar way to canonical arguments.All the above facts indicate that a "strict" pseudoincorporation analysis is not suitable for Greek.