Little pro’s, but how many of them? — On 3SG null pronominals in Hungarian

While Hungarian 3SG individual reference null pronominals are in free variation with their lexical counterparts, 3SG generic reference null pronominals do not show such variation. This follows from the fact that Hungarian 3SG generic null pronominals behave like bound variables, i.e. they always require a 3SG generic lexical antecedent in an adjacent clause. Both the 3SG generic lexical antecedent and the 3SG generic null pronominal must be in the scope of the GN operator, which is seated in SpeechActParticipantPhrase (SAPP), the leftmost functional projection of the left periphery in the sentence (see Alexiadou & D’Alessandro, 2003; Bianchi, 2006). GN binds all occurrences of the generic variable in accessible worlds (see Moltmann 2006 for English one/oneself). These properties distinguish Hungarian from the four major types of Null Subject Languages identified by Roberts & Holmberg (2010).


Hungarian as a Null Subject Language
Hungarian has all the properties of so-called Null Subject Languages (Jaeggli & Safir, 1989;Egerland & Sigurðsson, 2009;Roberts & Holmberg, 2010). It allows expletive null subjects in meteorological sentences, individual reference null subjects as well as generic reference null subjects, and any other null arguments in active finite sentences: Expletive null subject Már hajnal-od-ott proEXPL, amikor el-alud-takk a gyerekek. already dawn-RFL-PAST3SG EXPL when PFX-SLEEP-PAST3PL the children 'It was already beginning to dawn when the children fell asleep.' Referential null subject Verai fél-t, [hogy proi/j le-kés-i a film-et]. Vera fear-PAST3SG that (s/he) PFX-MISS-PRES3SG the movie-ACC 'Verai feared that shei/j (herself or someone else) would miss the movie.' Generic inclusive null subject Despite these properties, Hungarian differs from the four major types of Null Subject Languages (NSLs) established by Roberts & Holmberg (2010) in crucial ways. First, while 3SG individual reference null pronominals are in free variation with their lexical counterparts, no such variation is found with 3SG generic reference null pronominals. Second, the 3SG generic reference null pronominal functions as a variable, which must be coreferential with a 3SG generic lexical antecedent in an adjacent clause, while 3SG individual reference null pronominals function quite well without any lexical antecedent. Third, 3SG generic inclusive null pronominals show no scope interaction with quantifiers, as they must be in the scope of GN, which always take widest scope. These properties will be discussed in the rest of the paper. This paper is organized as follows. Section 1.1 describes the syntactic and semantic properties of 3SG null pronominals with the generic inclusive interpretation and the way they differ from 3SG null pronominals with the individual reference interpretation in Hungarian. Section 1.2 explains how 3SG generic inclusive are distinguished from 3PL generic exclusive 2 Existentially quantified NPs can take scope over 3PL generic exclusive NPs. This indicates the quantificational properties of 3PL generic exclusive NPs. 3SG generic inclusive NPs, on the other hand, show no similar scope interaction, as GN always takes widest scope (see Moltmann, 2006): (i) Az emberek néha furcsa e-mail üzeneteket kapnak. (= Some x's are such that….) ∃>GN 'Sometimes people receive strange e-mail messages.' (ii) Az ember néha furcsa e-mail üzeneteket kap. (≠ Some x is such that…) *∃>GN 'Sometimes one receives strange e-mail messages.' On 3PL generic exclusive null pronominals in Hungarian see (Tóth, 2011). On the semantic differences between Hungarian 3SG generic inclusive vs. 3PL generic exclusive null arguments see Dalmi (2013;2014). null pronominals with respect to their scope interaction options and binding requirements. Part 2 discusses a recent typology of Null Subject Languages (NSLs) given by Roberts & Holmberg (2010) and explains that Hungarian does not fit in any of the four major types of NSLs as far as 3SG generic inclusive null pronominals are concerned. Part 3 introduces Harley & Ritter's (2002) feature geometric account of pronominals and Alexiadou & D'Alessandro's (2003) proposal to include impersonal SI in Italian in that system. Finally, it explains why Hungarian 3SG generic inclusive null pronominals do not require a feature geometric approach and proposes to give merely a feature composition of such pronominals.

3SG generic inclusive pro vs. 3SG individual reference pro in Hungarian
The general consensus concerning Hungarian 3SG individual reference null pronominals is that they are in free variation with their lexical counterparts (É. Kiss, 1987;2002;Kenesei, 1989): 3 Péter nem 0 biztos abban, hogy ő/pro átmegy a vizsgán. Peter not COP sure in_it that he/(he) through_go.3SG the exam-on 'Peter is not sure that he/pro will pass the exam.' Ő/pro nem 0 biztos abban, hogy ő/pro átmegy a vizsgá-n. he/(he) not COP sure in_it that he/(he) through_go.3SG the exam-on 'He is not sure that he/pro will pass the exam.' The 3SG generic reference null pronominal, proGN, differs from 3SG individual reference pro inasmuch as it shows the properties of bound variables (see Moltmann, 2006 on English one/oneself and Kratzer, 2000 on German man). Unlike the 3SG individual reference pro, this type of 3SG null pronominal must be bound by the 3SG lexical antecedent az ember 'the manGN', (8); it does not alternate with the 3SG individual reference pronominal (lexical or null), (9); it cannot be bound by the 3SG individual reference pronominal, whether this pronominal bears structural or inherent case, (10)-(11). Notice that unlike the 3SG individual reference null pronominal, the 3SG generic reference null pronominal always involves the speaker, hence it is 1 st person-oriented, (8)-(11): Az emberGN nem tudja, mennyit kell proGN / *ő-neki / *pro the man not knows how much must (the man-DAT) / he-DAT / (he-DAT) a nyugdíj-ig dolgoz-ni-a. the retirement-TERM work-INF-3SG 'One does not know how much one/*he/(*he) must work till retirement.' Az emberGN nem 0 biztos abban, hogy *ő / *pro / proGN át-megy a vizsgá-n. the man not COP sure in_it that he/ (he) / (the man) through-go.3SG the exam-on 'One is not sure that one / *he will pass the exam.' *Ő/*pro nem 0 biztos abban, hogy az emberGN / proGN átmegy a vizsgá-n. He/(he) not COP sure in_it that the man / (the man) through_go.3SG the exam-on 'He is not sure that one will pass the exam.' Ha az emberGN sejtené, mennyi megpróbáltatás vár if the man guess-COND3SG how many adversities wait proGN rá / *ő-rá a házépítés során, proGN /*ő /*pro el sem kezdené. (one) on / he-on the housebuilding during (one) / he /(he) PFX not even start-COND3SG 'If one could guess how many adversities are awaiting one in the course of building a house, one would not even start.' As the examples in (8)-(11) show, 3SG proGN clearly differs from 3SG pro in its distribution, binding requirements, and semantic interpretation: The properties of 3SG proGN (i) It requires the 3SG lexical antecedent az ember 'the man'. (ii) It does not alternate with 3SG lexical, individual reference pronominals. (iii) The 3SG lexical antecedent az ember 'the man' cannot have individual reference. (iv) It is 1 st person-oriented, i.e. it includes the speaker.
These properties justify isolating 3SG generic reference proGN from 3SG individual reference pro. The semantic properties of 3SG generic inclusive proGN also distinguish it from 3PL generic exclusive pro. This will be shown in the next subsection.

The semantic properties of 3SG generic inclusive vs. 3PL generic exclusive null pronominals
Cinque (1988)  It is important to segregate the 3PL generic exclusive reading of impersonal SI from the 3SG generic inclusive reading (see Roberts & Holmberg, 2010 for a recent typology of Null Subject Languages, NSLs). As Moltmann (2006) points out, 3PL generic exclusive people in English shows scope interaction with quantifiers but 3SG generic inclusive one never does: People often wear fashionable clothes at parties.
OFTEN > MOST or MOST > OFTEN One often wears fashionable clothes at parties.

GN > OFTEN but *OFTEN > GN
While (14) can be interpreted as "there are many x's, such that…." Such interpretation is not available for (15). The two lexical items also differ pragmatically: You can't imagine how much one suffers during cosmetic surgery! You can't imagine how much people suffer during cosmetic surgery.
Generic inclusive one in (16) conveys the speaker's intention to arouse sympathy in the addressee, while (17) with generic exclusive people is merely a statement of the fact. The existentially quantified DP in (18) does not c-command the 3SG pronominal, and therefore it cannot serve as an antecedent for it. This leads to a Weak Cross-Over (WCO) effect.
In (19), on the other hand, GN takes scope over all occurrences of one and no Weak Cross-Over effect shows up: WEAK CROSS-OVER EFFECT *Someonei's mother always gives himi a birthday present. One's mother always gives one a birthday present.
A similar scenario is found in Hungarian: mindig ad-0 ajándék-ot az proi any-ja [DP ti]]. someone-DAT always give-PRES3SG present-ACC the s/he.NOM mother-POSS3SG 'Hisi mother always gives someonei a present.' NO WCO-EFFECT [SAPP GN [TOPP Az ember-nekgen mindig ad ajándék-ot the man-DAT always give.PRES3SG present-ACC az [DP progn any-ja]. the (the man.NOM) mother-POSS3SG 'One's mother always gives a present to one.' GN in Moltmann's (2006) sense is a group-inducing, first person-oriented modal operator that binds generic inclusive one in accessible worlds. GN does not enter into scope interaction with quantifiers. Furthermore, GN always has widest scope. This verifies its operator status. 4 Tóth (2011) also discusses the quantificational properties of the Hungarian 3PL generic exclusive az emberek 'the people' and its interaction with quantificational adverbials like általában 'generally', rendszerint 'usually'. This behaviour follows from the fact that 3PL generic exclusive az emberek 'the people' (and its null counterpart) is bound by a quantifier like MOST: most x's are such that…: A középkor-ban általában az emberek féltek a villámlás-tól. the Middle Ages-INESS generally the people feared the lightning-from 'In the Middles Ages, people generally feared lightning.' A középkor-ban általában féltek pro a villámlás-tól. the Middle Ages-INESS generally feared (the people) the lightning-from 'In the Middle Ages, people generally feared lightning.' (Most x's in the Middle Ages were such that they feared lightning.) 3SG generic inclusive az ember 'the man' and its null pronominal counterpart, on the other hand, do not show any scope interaction, as they must both be in the scope of the GN operator, which takes scope over the whole sentence: Ha az emberGN munkanélkülivé válik, gyakran éhezik proGN. when the man unemployed becomes often starves (the man) 'When one becomes unemployed, one often starves.' (≠Most x's are such that when they become unemployed, they starve.) Cross-linguistic studies on pronominals mostly discuss individual reference null pronominals, expletive null pronominals but not generic reference ones. By studying the properties of 3SG generic reference pronominals across languages, a more fine-grained typology of Null Subject Languages can be provided. In Consistent NSLs, 3SG generic inclusive pronouns must always be lexical, whereas 3SG individual reference pronouns can be freely dropped. Partial NSLs, by contrast, require 3SG individual reference pronouns to be overt but 3SG generic reference pronouns to be null. Holmberg (2005; explains these facts in terms of parametric variation across languages. Wherever the T0 head has a [+D] feature, the 3SG pronominal does not spell out. In Italian sentences with the 3SG generic inclusive interpretation, the T0 head has a [-D] feature, which forces the 3SG generic pronominal to be overt.

A cross-linguistic outlook on null arguments
Roberts & Holmberg (2010) establish a typology of Null Subject Languages (NSLs), based on the kinds of null subject such languages allow:  Holmberg (2005; observes the following correlation between 3SG referential vs. 3SG generic null subjects in Type 2 Partial NSLs and Type 3 Consistent NSLs: Partial NSLs: 3SG referential subjects must not be null; 3SG generic subjects must always be null; Johni-wa kono beddo-de-wa yoku nemu-reru-to proi /proGN iu. John-TOP this bed-in-TOP well sleep-can-COMP (he / one) say 'Johni says that hei/one can sleep well in this bed.' (Japanese, Holmberg & Sheehan, 2010) Although Hungarian is unquestionably a Null Subject Language (NSL), it does not perfectly match any of these major types of NLSs, as far as the properties of 3SG generic inclusive arguments are concerned. In several respects, it patterns with Icelandic, which is renowned as a non-Null Subject Language. In Icelandic, 3SG generic inclusive subjects are expressed lexically, by maður 'the man'. Yet, in certain contexts, a 3SG generic inclusive null pronominal can also show up: Ef maður tapar, thá er maður úr leik. if one/I loses then is one/I out 'If one loses, one is out.' 'If I lose, I am out.' (Icelandic, Jónsson, 1992) [ (Icelandic, Egerland & Sigurðsson, 2009) Variation between the 3SG generic vs. 1SG individual interpretations of maður 'the man' is contextually determined. The 3SG generic inclusive interpretation of the null pronominal in (37) is licensed by the non-veridical operator.
It is worth noting that in (37), generic inclusive maður must be repeated in the matrix clause, just as English one or German man (see Moltmann, 2006 andKratzer, 1996, respectively). This is not the case in Hungarian, where it is the first occurrence of 3SG generic inclusive arguments that must always be lexical but the other occurrences must be null, irrespective of whether they appear in matrix or subordinate clauses, i.e. precedence is crucial but dominance is not. This fundamentally distinguishes Hungarian from all the four major types of NSLs.

Harley & Ritter (2002)
Harley & Ritter (2002) adopt Noyer's (1992;1997) Universal Feature Hierarchy (UFH), and propose a system of pronominals in which pronouns are built up of geometrically arranged feature bundles, including pragmatic features such as speaker and addressee, as well as morphological features like person and gender: While this system provides sufficient room for cross-linguistic variation among individual reference pronominals, it does not have much to offer for generic ones.

Alexiadou & D'Alessandro (2003)
Alexiadou & D'Alessandro (2003) provide a feature-geometric account of impersonal generic SI in Italian. Their system draws heavily on Harley & Ritter's (2002)  We have seen that Icelandic, a non-NSL, has variation in the interpretation of maður 'the man', however this variation does not hinge on aspect. Hungarian patterns with Icelandic in this respect, and calls for a non-feature geometric analysis.

The feature composition of the 3SG generic null pronominal in Hungarian 5
Hungarian is a NSL, in which variation between the generic vs. individual reference interpretations of 3SG null pronominals is not conditional on the aspectual properties of the verb. These interpretive differences can be reduced to the presence or absence of the [GN] feature, licensed by the GN operator in SAPP, the leftmost functional projection of the left periphery of the whoel ssentence. 3SG vs. 3PL agreement on the verb clearly indicates that generic inclusive and generic exclusive null pronominals are specified for person/number features in this language. The locus of licensing these features in the clause is the TP projection. 6 (As opposed to Finnish, Hungarian imposes no obligation on 3SG individual reference subjects to be lexical, i.e. they can be freely dropped. The feature composition of 3SG generic null proGN vs. 3SG individual reference pro is given in (48a) and (48b), respectively. These features are licensed in the clausal architecture given in (47) via Cyclic Agree (Bèjar & Rezac, 2009 The differentiated feature specification of individual reference vs. generic reference pronominals enables us to place Hungarian in its proper place among Null Subject Languages (NSLs). Given that 3SG vs. 3PL generic null pronominals have turned out to be crucial in the cross-linguistic study of null arguments (see Holmberg & Phimsawat, 2013), the present study is an important contribution to a more fine-grained picture of null arguments, in general, and in this way, to the typology of Null Subject Languages.

Summary
The