Pushed out of arm’s reach: Pronouns and spatial anaphora in Hungarian

This paper investigates hitherto unnoticed variation in the linguistic coding of spatial anaphora in locative PPs in Hungarian. While Hungarian primarily employs reflexives in these configurations, it is well-known that pronouns are the default strategy in English, and the reflexive anaphor is allowed in locative PPs only in the presence of certain licensing factors. One such factor is the availability of body-oriented readings (Rooryck & Vanden Wyngaerd 2007, 2011), and we argue here that this plays an important role in Hungarian, too. The paper reports the findings of a corpus study and an online questionnaire study and shows that pronouns are not only acceptable in Hungarian spatial anaphora, but either outperform or form a viable alternative to reflexives when the location denoted by the PP is not close to the referent of the antecedent. A secondary effect of structure building is also observable in two configurations of the extended PP. We argue that the employment of a possessive structure in certain PPs, and moving a P-element to a CPPP cap may also contribute to saving pronouns in contexts of spatial anaphora.


INTRODUCTION
Locative PPs represent a well-known construction type wherein the complementarity between pronouns and anaphors is not immediately apparent. In English, both pronouns and reflexives are grammatical in configurations of spatial anaphora (a.k.a. snake sentences), though the former have been claimed to be the unmarked or the preferred choice (see, among others, Faltz 1985, 100, suffixes, case-like postpositions, and case-assigning postpositions. 3 Case suffixes and case-like postpositions bear agreement morphology if their non-case-marked complement is a (prodropped or overt) pronoun, and case-assigning postpositions take a case-marked complement. The pronoun strategy has been described as ungrammatical in each of the three types of spatial PPs in snake sentences in the syntactic literature. The following example with a case-like P is from É. Kiss (1987, 185-186). 4 (3) János i letette a könyvet p {(ő)mellé i } / maga i mellé. John down.put.3SG the book.ACC (he.)to.beside.3SG himself to.beside ʻJohn put down the book beside him/himself.ʼ While we essentially agree with É. Kissʼs evaluation of this particular example, we will argue here that the pronominal strategy is a genuine option even in standard Hungarian. In fact, if the right configuration is available, pronouns may even fare better than reflexives in spatial anaphora.
We take Rákosi (2010Rákosi ( , 2012 as our starting point, where it is shown that pronouns are an alternative for the majority of speakers in non-third persons. Here we report the results of our corpus and questionnaire studies to provide solid evidence that pronouns are in fact acceptable across the board and may even be preferred if the right conditions are met. The major factor is the nature of the location denoted: while reflexives are the only option in body-oriented readings 3 We provide an example for each type of P below in (i-iii). Lexical noun phrase complements precede P-elements in Hungarian (the a-examples), and these complements bear case-morphology in the case of case-assigning Ps. The postposition együtt ʻtogetherʼ, for example, requires its complement to be marked with comitative case (iiia). Pronominal complements are by default realized as agreement-morphology on the P head if it is a case suffix (ib) or a case-like P (iib). Case-assigning Ps take the pronominal form of the appropriate case suffix. The pronoun complement itself is usually pro-dropped.
in practically any syntactic context, the pronoun becomes a true alternative if the Figure is to be located at some distance away from the individual denoted by the complement of the P-element. With some extra structure added to the usual PP configuration, the pronoun actually is the better alternative for everyone. The structure of the paper is as follows. Section 2 spells out the empirical background behind our questionnaire studies, including a discussion of previous work on pertinent variation in constructions of spatial anaphora in Hungarian, as well as the results of our corpus studies and pilot studies. In Section 3, we report the results of our main questionnaire study. We use these data in Section 4 to argue for a differential syntactic treatment of reflexive and pronominal spatial anaphora in Hungarian, and we present our respective analyses. We follow Rooryck & Vanden Wyngaerd (2007 in our analysis of the reflexive variety of snake sentences, but we will argue that the pronoun construction is only available to the extent that some extra syntactic structure is added to the default PP core. Section 5 concludes the paper.

The previous literature on variation within Hungarian
It has been noted in the pertinent literature that some Hungarian dialects prefer the pronoun to the otherwise standard reflexive in the following construction (see Hegedűs 2012, 180-181, den Dikken et al. 2001).
(4) Én nem viszek % velem / magammal esernyőt. I not take with.1SG myself.with umbrella.ACC ʻI do not take an umbrella with me.ʼ The pronoun version is found mainly in the western part of Hungary (mostly along the Austrian border), and in some archaic Hungarian dialects on the eastern slopes of the Carpathians in Romania. Note that (4) contains a case marker (comitative case), and Hungarian dialect studies do not offer a comprehensive discussion of spatial anaphora. This particular example is also a somewhat atypical specimen of spatial anaphora, since the pronoun is obligatorily co-construed with the subject. 5 Rákosi (2010Rákosi ( , 2012) is a first attempt aimed at probing into the scale of variation that is attested in standard colloquial Hungarian in the use of the pronominal strategy in snake sentences. Rákosi (2012) in particular reports the results of a questionnaire study which included 5 Such predicates require the pronoun in English, and the reflexive is ungrammatical, irrespective of the availability of body-oriented readings or of potentially favourable discourse conditions for logophoric uses (see Smith 2004).
(i) Mary i brought her lunch with her i / p herself i .
(ii) He i looked about him i / p himself i .
(iii) The tramp i had a lot of money on him i / p himself i .

We did not include such examples in our current survey.
target examples with inflected spatial case markers (see (4) above) and case-like Ps (illustrated by (5) below) (N 5 45). The sentence that received the highest rating (1.29 on the average on a 5point scale from 2 to À2) is the following: Mi láttunk mellettünk valamit az út szélén. we saw.1PL beside.1PL something.ACC the road side.POSS.on ʻWe saw something next to us at the side of the road.ʼ The dominant factor influencing the judgements was the agreement feature content of the P-element, and the level of acceptability correlated with the following scale: 1PL > 1 SG > 3PL > 3SG. 6 These results confirm the judgements on the 3SG example in (3), since an instance of this construction scored only À0.87 on the average in this survey. But these results also inform us that the pronominal strategy is a viable (though perhaps marked) option for many Hungarian speakers. To be precise, it is an option in PPs headed by case-like Psthe pronoun version of (4) was barely acceptable for the participants of this survey (with mean judgements at À1.47 on the 5-point scale from 2 to À2). Thus while the pronoun in (4) is indeed a dialectal phenomenon with restricted scope, the relative acceptability of (5) is characteristic of a much wider population of speakers of standard colloquial Hungarian.
In our current inquiry, our aim is to dig deeper into this domain by the inclusion of caseassigning Ps and by paying special attention to the location factor mentioned in Section 1. Before concluding this subsection, we note that the pronominal coding of spatial anaphora was acceptable across the board in earlier stages of Hungarian, including even 3SG forms in Old Hungarian. The following example is from Hegedűs (2014, 137), and it contains both an inflected case marker (nala) and a case-like postposition (mellette) in a syntactic context where only the reflexive anaphor is grammatical in current Hungarian.  One possible reason behind the variation that we see in modern Hungarian in pronoun acceptability might actually stem from differing historical paths of development: more archaic dialects might have preserved this older stage of the language. Thus the easternmost dialects that prefer the pronoun in (4) may simply do so because they are archaic. 7 In Section 4, we return to this issue to argue that variation in current standard Hungarian may also be related to a potentially uneven spread of PP grammaticalization across the Hungarian speaking community in a way that affects judgements concerning examples like (5).

Background data collection
To get a firmer grip on the data, we conducted a corpus study and two pilots, and we used the results of these inquiries in setting up the design for our main questionnaire. The corpus study was performed on version v2.0.5 of the Hungarian National Corpus (Oravecz et al. 2014), and it produced the following major results.
First and foremost, the corpus data provide clear evidence that the pronoun strategy is available in Hungarian snake sentences even if it is infrequent in comparison to the reflexive strategy. Table 1 contains some illustrative data involving four case-like and two caseassigning postpositions. The case-assigning postposition szemben ʻin front of meʼ requires comitative case on its complement, whereas the complement of keresztül ʻthroughʼ is in superessive case.
For this set of data, we used search expressions to get hits which contain the 1SG form of either the pronominal or the reflexive PP in the vicinity of a 1SG verb form. The results were then manually checked and selected to include only the target examples, that is, only cases where the PP and the verb are indeed in the same clause. It is evident that the pronoun version is an existing, albeit less frequent alternative to reflexives in Hungarian, and thus the construction that we are investigating in this paper is a learnable trait of Hungarian, and not an experimental artefact.
The pronoun hits frequently include some extra material on top of the target PP: another VP-internal PP, a particle on the verb, or the locative adverbials itt ʻhereʼ or ott ʻthereʼ. Example (7) below illustrates this latter construction, and it is one of our test items from the main questionnaire: Köszöntöm itt velem szemben Kovács Péter közgazdászt. greet.1SG here with.1SG opposite Kovács Péter economist.ACC ʻI welcome the economist Kovács Péter here opposite me.' The corpus search also provided evidence for the relevance of the location factor that Rooryck & Vanden Wyngaerd (2007 show to be a key determining factor in the grammar of spatial anaphora. The pronoun hits typically describe situations in which the Figure is located at some distance away from the referent of the antecedent, as in the case of fölém ʻ(to) above meʼ 3 magam fölé ʻ(to) above myselfʼ 45 15:1 case-assigning velem szemben ʻin front of meʼ 6 magammal szemben ʻin front of myselfʼ 271 45:1 rajtam keresztül ʻthrough meʼ 1 magamon keresztül ʻthrough myselfʼ 10 10:1 watching something in the sky above oneʼs head, or in the case of looking at things at the bottom of the sea while one is swimming. Following the literature on the cognitive aspects of the linguistic coding of the near/far distinction, we define being far as being out of armʼs reach (see Kemmerer 1999 for an overview). Thus (7) instantiates our far condition since the public event it describes will typically involve the two participants located at a distance that reaches beyond the personal space around either individual. Our body condition involves situations in which this distance is either within arm's reach (as in laying a towel under oneʼs body to sleep on it, or pulling somebody elseʼs hand close to oneself), or when the event described directly targets the body of the referent of the antecedent (for example, when one runs low tension current through oneʼs body in an experiment, or when looks down on oneʼs own body). Some of the corpus hits describe situations in which the referent of the pronoun and the referent of the antecedent are referentially distinct or, in the terminology of Heim (1998), they spell out different guises of the same individual. (8) is one such example: Rajtam kívül éltem. on.1SG outside lived.1SG ʻI was living outside of me.ʼ The predicate here is anti-reflexive and it only supports coreference-based interpretations. We avoided such examples in our questionnaire study, and included only items where both coreferential and bound variable readings are possible (see Subsection 4.3 below for more on this).
The literature on spatial anaphora reports restrictions on the thematic type of the PP. In Hebrew, for example, the reflexive is obligatory if the PP describes a path, but not when it describes a place (Bassel 2018). We found no particular evidence for such thematic restrictions in our corpus study, since the pronoun hits instantiate all types of PPs, including places, paths, sources and goals. Our pilot questionnaires did not reveal an obvious sensitivity to this issue in Hungarian, and we therefore did not include this parameter in our study.
In sum, the results of our preliminary corpus and questionnaire studies were incorporated in the design of our main questionnaire as follows. First, we tested only case-like and case-assigning postpositions in clauses whose predicate is neither obligatorily reflexive (I brought the umbrella with me), nor anti-reflexive (I lived outside of me). Second, we paid special attention to the location parameter by carefully separating what we call here the body condition from the far condition. Third, we only used 1SG test items, since non-third person snake sentences fare much better for native speakers in the pronoun condition (see the previous subsection). Thus if we want to investigate the relevance of the location factor in Hungarian, we may expect more solid judgements with 1SG test items. Four, we only used pro-dropped pronominal examples. This is the usual case in Hungarian, the pronoun itself is only overt in PPs if it has an obvious discourse function. Overt pronouns are also a better fit for coreference-based interpretation, but since our focus was on testing whether pronouns may function as a true alternative to reflexives in snake sentences, we selected only pro-dropped examples. If judged acceptable, such examples provide a stronger case for the genuine availability of the pronominal strategy in snake sentences in Hungarian.

The design
The aim of the online questionnaire was to test the overall assumption that besides reflexives, pronominals are also acceptable in snake sentences. Participants saw contrastive pairs of sentences on the screen, and the items in each pair were identical except for the choice of the PPcomplement (pro-dropped pronoun vs. reflexive). Their task was to rate each sentence separately on a 5-point Likert scale from 1 (totally unacceptable) to 5 (totally acceptable). The underlying intention here was to force the subjects to be aware of the competition between pronouns and reflexives. If pronouns receive high ratings even in the obvious presence of the reflexive alternative, then that provides a strong case for the availability of the pronoun construction in snake sentences. The hypotheses tested are given below: H1 Pronouns are more acceptable than reflexives when the location denoted by the PP is not close to the referent of the antecedent. 8 H2 Pronouns are more acceptable with case-assigning Ps than with case-like Ps.
The rationale behind Hypothesis 2 is that, other things being equal, we expect a visible increase in the complexity of PP structure (the case-assigning P head plus the case suffix on its complement) to correlate with extra structure building in the syntax. This extra structure may help avoid a Principle B violation in case the pronoun is used. We used a mixed design in the experiment, we tested two within-subjects factors with two levels each: pronoun type: pronoun vs. reflexive, and location: body vs. far. To minimize the chance of participants' being accustomed to structures which have been described as marginal before and to avoid a decline in performance due to fatigue, we also had a between-subjects variable: one group of participants rated examples with case-like postpositions, while the other group saw test items only with case-assigning postpositions (P-type: case-like Ps vs. case-assigning Ps).
In both conditions of P-type, the questionnaire contained 12 test items (6 pairs of sentences), and there were 4 pairs of fillers. Two of the latter contained sentences that were totally unacceptable, while the other two pairs consisted of fully acceptable sentences. These were inserted to check whether participants were paying attention or were just clicking randomly while filling in the test. As a result of this screening procedure, we had to exclude eight subjects out of 103 in the case-like P condition, and three out of 108 in the case-assigning P condition, since they provided controversial judgements in the fillers. Tables 2 and 3 summarize the gender, age and typical childhood location data of the remaining participants. 9

The results
The overall results are shown in Table 4. Preliminary findings indicate that the highest average rating occurred with case-like Ps in the reflexive, body condition, while the lowest average rating 8 Being far is taken to be anywhere in the zone which is beyond armʼs reach, see Subsection 2.2. 9 In both groups of subjects there were participants who failed to provide information on their regional background, i.e., they did not specify where they spent their childhood.
can be found with case-assigning Ps in the pronoun, body condition. 10 It is also revealed that reflexives received higher ratings with case-like Ps, and pronouns fared better in the far condition in general.
We carried out a mixed design ANOVA to analyse the results. There was a significant twoway interaction between pronoun type and location: F(1, 198) 5 531.006, P < 0.001, h 2 5 0.73. This means that pronouns got higher ratings when the location is far (i.e., beyond arm's reach) from the referent of the antecedent, while reflexives received higher ratings when the location denoted by the PP is close (within arm's reach, our body condition). Hence, our first hypothesis has been confirmed.
We also found a significant three-way interaction: F(1, 198) 5 37.975, P < 0.001, h 2 5 0.16. The interaction is represented in Fig. 1. Fig. 1 shows that with case-assigning Ps, pronouns are more acceptable in the far condition than in the body condition ("near"), while reflexives show just the opposite pattern. In fact, pronouns outcompeted reflexives in the far condition in this case. With case-like Ps,  In principle, one would expect pronouns in the body condition to be slightly better with case-assigning Ps than with caselike Ps (see section 4 for a discussion). The reason why our actual data show the inverse of this expected pattern is probably due to the fact that two of our test items strongly evoked the body schema in the case-assigning P condition (running low tension current through oneʼs body in an experiment and looking down on oneʼs own body). We did not have such examples in the case-like P condition, and this lack of balancing across these two conditions is most probably the factor that is responsible for this result. Note that reflexives in the body condition were somewhat better in the case-like P construction (4.86) than in the case-assigning P construction (4.47) as expected, even with the biasing factor just described.
pronouns do not outperform reflexives in the far condition, nevertheless they represent an alternative, since the average ratings are almost the same for pronouns and reflexives in the far condition. This means that the second hypothesis has been partially confirmed. We illustrate the empirical landscape behind these data with two quartets of test items, one from the far condition, and one from the body condition. Consider this set from the far condition first, with the mean judgements given after the respective items: Köszöntöm itt ……………… Nagy József közgazdászt. greet.1SG here Nagy József economist.ACC ʻI welcome the economist Nagy József next to/opposite me.ʼ The pronoun was the more acceptable choice in both the case-like P condition and in the caseassigning P condition, but whereas the gap is minimal in the former, it becomes much more pronounced in the latter case. The next two pairs illustrate our body condition. As is clear, the pronoun is a weak competitor in this case, and it fared much worse in both construction types than the reflexive.

Some notes on individual variation
We have also tried to interview individual native speakers on several occasions on how they react to Hungarian snake sentences, and the following two trends have emerged. First, speakers tend to be divided in their judgements pertaining to pronouns in case-like P constructions: many find such examples marked or "weird", whereas other speakers may accept them without reservation. Second, even the rejecters tend to find pronouns in case-assigning P constructions more or totally acceptable. The weight of these observations is also detectable in our questionnaire data. In Fig. 2 below, we illustrate individual variation in terms of the individual ratings for each participant, i.e. the variable represented in the figure shows the respective value of the calculated difference between the mean of reflexive and pronoun ratings. 11 In the case of those participants who provided higher average ratings in the pronoun condition than in the reflexive condition, the value in question is below zero (collapsing the results of the body and far conditions). As Fig. 2 shows, only 6 subjects gave higher ratings for pronouns than for reflexives in the case-like P condition (out of 95), whereas in the case-assigning P condition this number raised to 27 subjects (out of 105). In visual terms, this is represented by a slight shift to the left in the case of case-assigning Ps in the figure. This gives at least partial confirmation to our second experimental hypothesis, namely, more subjects gave higher ratings to pronouns in the case-assigning P condition.
We add finally that we tried to find statistical correlations between the individual ratings and the social variables that we collected, but none of these were significant. In absolute terms, the lowest ratings for pronouns (and the highest ratings for reflexives) come from the youngest generation (19-29) and from subjects from the eastern part of Hungary (the Tiszántúl region). Further data collection of a much wider scale is needed to test whether these tendencies are real.

Interim summary
The most important result of our empirical data collection is the evidence that pronouns are acceptable in snake sentences in Hungarian. Nevertheless, they represent a true alternative to reflexives only in our far condition, i.e., when the Figure is beyond the reach of the referent of the pronoun. There is no real competition when it is not, since reflexives tend to be the only acceptable option in our body condition. We also see a secondary effect of syntactic structure: reflexives are somewhat worse in case-assigning P constructions, and pronouns are somewhat The histograms in Fig. 2 show the number of those participants (indicated by the height of each bar) whose calculated difference of average ratings fall into the ranges given. better in the self-same context at least in the far condition, with substantial variation in individual judgements.
In this section, we discuss what these results mean for the syntactic analysis of snake sentences in Hungarian. We adopt Rooryck & Vanden Wyngaerdʼs (2007 analysis for the reflexive cases, but we present an alternative analysis for the pronoun construction. Our account rests on the assumption that PPs in the pronoun condition can have a more complex syntactic structure than in the reflexive condition, and this structural variation is also supported by factors independent of our current concerns.

Reflexives and the body
As discussed in the introduction, Rooryck & Vanden Wyngaerdʼs (2007 analysis of snake sentences aims at, among other goals, capturing the interpretive difference between pairs of the following kind: (12) a. Mary kept her childhood dolls close to her. (5 proximity/vicinity) b. Mary kept her childhood dolls close to herself. (5 against her body) They note that the choice of the pronoun allows for a relatively abstract understanding of the space denoted (Rooryck & Vanden Wyngaerd 2011, 235). The doll could be kept, for example, at Maryʼs home. The reflexive (12b), however, requires a situation in which the doll is kept close to Maryʼs body.
They draw on work that incorporates spatial axes or axial parts in the grammatical representation of nominal expressions and spatial prepositions. 12 They argue in particular that the -self morpheme of English reflexives "is semantically rich enough to provide the semantics of axial dimensions in the appropriate syntactic context", and it comes with a valued dimensional feature (op. cited 247). It can thus value the unvalued dimensional feature in the AxPartP layer of the decomposition of the PP. (12b) is analysed in this system as follows, using their notation (op. cited 254): They assume that the axial dimension that the reflexive provides accounts for the body-oriented interpretation which we see attested in (12b). The PP is not a binding domain in their analysis (as they explicitly claim, it is not a phase) in (13), and the reflexive can straightforwardly enter a local referential dependency with its subject antecedent. We adopt this account for Hungarian under the assumption that (i) the PP is not phasal in Hungarian in the default case (but see Subsection 4.3 for more on this), and that (ii) the Hungarian reflexive magam ʻmyselfʼ is also lexically equipped with axial features. This second assumption is especially motivated in Hungarian, since the Hungarian reflexive is a grammaticalized possessive 12 See Svenonius (2006) and subsequent work. For arguments supporting the postulation of AxPartP in Hungarian PPs, see Asbury (2008) and Dékány (2011). We use the term PP informally as a shorthand for any syntactic variant of fullfledged adpositional phrases. structure with a nominal head that probably meant ʻbodyʼ in the period preceding Old Hungarian (see Rákosi 2011 for an overview discussion). We use one of our test items from the case-like P condition for illustration, focusing on the essentials: Since the reflexive is the default case in Hungarian snake sentences, we do not claim that it always contributes axial dimensional features, but when it does so, the body-oriented interpretation is positively there. Irrespective of this question, the Hungarian PP is not phasal by default, and the binding domain for the reflexive is thus the embedding clause. The pronoun construction, however, is different. We present our arguments for the differential syntactic treatment of pronominal snake sentences in the next subsection. Rooryck & Vanden Wyngaerd (2007 argue that pronouns, unlike reflexives, lack axial dimensions in English, and they are thus unable to value the features of AxPart. Here therefore AxPart enters the derivation with valued features. They assume furthermore that AxPart also hosts a Speaker-variable in this case, which is bound by the structural representation of the speaker in Evid(ential)P in the left periphery of the clause. The emerging structure is as follows (focusing again, on the relevant details, op. cited 254): On the one hand, this analysis accounts for the semantic difference between (12a 5 15) and (12b). An observer-centered interpretation arises from the binding of the speaker variable in AxPart, which is concomitant with a broader and more abstract construal of space, lacking the body-oriented focus that the reflexive structure in (12b) manifests. On the other hand, they argue that (15) involves two occurrences of the speaker (in EvidP and in AxPart), and it creates a structure which is "identical to […] two consecutive sentences in the discourse" (op. cited 248). They explicitly claim that the PP here is a phase, and therefore the dependency between the pronoun and the subject antecedent is not local. 13 13 Their major empirical argument for this is that the pronoun in snake sentences behaves like an E-type pronoun in cross-sentential anaphora. Compare the following two examples.

Pronouns and syntactic complexity in snake sentences
(i) Every boy saw a snake near them/ p him.
(ii) Everyone came in. They/ p He sat down.
Hungarian is different in this respect, see (18) below.
The syntactic structure in (15) comes, so to say, for free. This is welcome inasmuch as it is indeed the pronoun strategy that is the default in English snake sentences both in terms of frequency and in terms of its less specific semantics in comparison to reflexives. The Hungarian facts are, however, different. Pronouns are only a marked alternative in snake sentences in Hungarian, and their use shows substantial inter-speaker variation. We therefore do not adopt Rooryck & Vanden Wyngaerdʼs (2007 analysis for Hungarian. We argue instead that there are two, independently motivated, structure-building strategies for Hungarian PPs that help save the pronoun, at least in the far condition. The logic of the argument is that the Hungarian PP is not phasal by default, but it may become so equipped if there is a need to express the beyond-arm's-reach readings that are incompatible with the reflexive strategy. The first of these structure-building processes is the employment of a possessive structure in certain PPs. Many of the Hungarian P-elements originate from possessive structures historically, and as Hegedűs (2014) shows, they have lost their possessive character gradually during the grammaticalization process. Some of their possessive traits are in fact still available synchronically, and Rákosi (2010Rákosi ( , 2012 argues that this is what makes pronouns in snake sentences acceptable. In particular, he postulates a silent PLACE predicate in an LFG-based syntactic analysis in the structure of these PPs, so that mellettem ʻbeside meʼ essentially is equivalent both syntactically and semantically to ʻbeside my placeʼ. This possessive structure is a marked alternative to a more run-of-the-mill, non-possessive PP configuration, and it can be seen as a late fossil of the possessive diachrony available for those speakers who are more permissive towards pronouns in snake sentences. Dékány (2011Dékány ( , 2018 provides a list of arguments for the synchronic possessive analysis of all Hungarian PPs. 14 She assumes in particular that they contain a silent light noun PLACE, which is "less lexical and referential than ordinary nouns and depend[s] on a higher functional head for licensing and interpretation" (op. cited 384). This silent noun is the possessum inside the complement of the PP, and the pronominal "possessor" is also spelled out in this domain: The pronoun then moves to an FP cap at the top, where the agreement morphology of the postposition is also spelled out. The pronoun itself can also be pro-dropped, as happens with the examples cited in this paper.
Since Dékány (2011Dékány ( , 2018 accounts successfully for the morphosyntactic linearization facts, among other things, we have no reasons to reject it. At the same time, we intend to maintain the spirit of Rákosiʼs (2010Rákosiʼs ( , 2012 account in assuming that some PPs can be less grammaticalized possessive constructions for some speakers at least. We suggest therefore that the silent PLACE predicate has a (more) referential version that is in fact detectable. Consider the following minimal pair for illustration: 14 We refer the reader to Dékány (2011Dékány ( , 2018 for a comprehensive discussion of the literature that postulates a possessive structure for PPs. that the best place ʻI always keep scissors with me. #That is the best place.ʼ (17a) contains a PP that has this referential PLACE predicate, which can be referenced in discourse. (17b) has a PP with the run-of-the-mill, non-referential variety of this silent noun. Note that the PP in (17) is headed by a case suffix, but the referential possessive structure is still available. We assume that it is even more available for case-like and case-assigning Ps in the right context, which is our far condition. Some of these PPs may be stored lexically as idiomatic units (as is probably the case for (17a)), but this more referential possessive PP structure can also be created on the fly. The speakers who avail themselves of this option more readily are the speakers who are more permissive towards the pronoun strategy in snake sentences. 15 If these (pro-dropped) pronouns find themselves buried in possessive phrases inside the PP structure, then we expect them to license both bound-variable and coreferential readings. 16 This is indeed the case, the speakers who accept (18) Admittedly, the referential contrast manifest between (17a) and (17b) is not always as pronounced as here with each potential pronoun-reflexive minimal pair in the domain of snake sentences. We believe nevertheless that it is often there, and when it is, then it is the pronoun version wherein the silent place noun is more referential. 16 Pro-dropped possessors generally license both bound-variable and coreference-based readings, see (i). The reflexive anaphor in the complement position of PPs can only be interpreted as a bound variable for the majority of speakers, see (ii).
(i) Csak én etetem a tyúk-ja-i-m-at napraforgóval. only I feed.1SG the hen-POSS-PL-1SG-ACC sunflower_seed.with (a) ʻNobody else feeds my chickens with sunflower seeds.ʼ coreference (b) ʻNobody else feeds their chickens with sunflower seeds.ʼ binding (ii) Csak én láttam magam előtt szellemet az úton. only I saw.1SG myself in_front_of ghost.ACC the road.on (a) p ʻNobody else saw a ghost in front of me on the road.ʼ coreference (b) ʻNobody else saw a ghost in front of them on the road.ʼ binding We therefore conclude that a referential possessive structure inside the PP is an available option in Hungarian, and it serves to protect the pronoun, avoiding a Principle B violation.
Our findings also indicate that there is possibly another structure building option which may contribute to the same effect, but this is on the left edge of the PP, rather than in the inside. Reflexives are worse, and pronouns are somewhat more acceptable in case-assigning P constructions or when there is some other PP-material present, like the locative adverbial itt ʻhereʼ. These effects also include the so-called prepositional use of certain case-assigning postpositions, first described by Dékány & Hegedűs (2015). Consider the following set of examples from one of our pilot studies (we used the same 5-point Likert scale as in the main study): prepositional use near myself.to ʻI drew the pram close to me/myself.ʼ The acceptability of the reflexive decreases in the prepositional construction (4.16→3.06), and the pronoun is rated somewhat better (3.06→3.12) to the extent that it actually fares better again than the reflexive. This is important because (19) represents our body condition, where reflexives should consistently be better than pronouns.
We follow Dékány & Hegedűs (2015) and assume that the P-head may move to a functional projection CP PP on the left edge as in (20a), and that results in the prepositional order. 17 We postulate furthermore that the locative adverbial itt ʻhereʼ can also occupy the head of CP PP and form a constituent with the rest, as in (20b) (2010) and den Dikken (2010) for the claim that a CP layer analogous to clausal CPs can be projected in the extended PP. We thank an anonymous reviewer for calling our attention to this literature. What our experimental data show is that this kind of structure building is secondary to the main effect obtained in our far condition. The extra structure possibly helps in licensing the referential PLACE predicate that triggers the possessive core needed to protect the pronoun, but it may also have a direct effect of its own. This latter assumption implies that the movement of a P-element to the head of CP PP (20a), or the insertion of a P-element in that position (20b) makes the extended projection of the PP a phase. Whether this is indeed the case is an issue that merits further data collection and research.

And what about 3SG?
We started this paper with a 3SG example (3), which was judged to be ungrammatical, and our survey focused on non-third person data therefore. But now, equipped with what we have learnt on the way, we are at a vantage point to view this issue from another perspective. If we create a context that includes the factors that favour pronouns in snake sentences in general, then pronouns become an option for speakers that we have consulted specifically on this issue: János i is látta ott vele i szemben a szellemeket. John too saw.3SG there with.3SG opposite the ghosts.ACC ʻJohn too saw the ghosts opposite him.ʼ The restriction against 3SG instances of pronominal anaphora in snake sentences may thus only be epiphenomenal, possibly driven by the functional pressure on the system to avoid a fundamental referential ambiguity (which does not arise in first and second persons, since speech participants are kept constant in a given speech situation). If this is the right interpretation of the data, then there is actually no ban against 3SG, or at least any such restriction is not internal to the computational system itself.

CONCLUSION AND OUTLOOK
The default strategy in contexts of spatial anaphora is the employment of reflexives in Hungarian. Though dialect studies note some specific instances of pronouns used in locative PPs in some varieties of Hungarian, standard Hungarian has been mostly described in the pertinent syntactic literature to lack the pronoun strategy. Our aim in this paper has been to show that there is in fact more extensive variation in this domain, and the pronoun becomes an available option for all speakers in some syntactic configurations, and in some cases pronouns even outcompete reflexives. The paper describes the finer details of this variation, building on the results of our corpus and questionnaire studies.
We discuss two important licensing factors. First, building on Rooryck & Vanden Wyngaerd (2007 on spatial anaphora in English, we argue that in Hungarian, too, the nature of the location matters. The pronoun strategy is available in Hungarian when the Figure is located at some distance away from the Ground, and the reflexive is the only choice in the case of bodyoriented readings. Second, we argue that this variation in meaning is concomitant with two different kinds of structure building in the PP in Hungarian, which are available to save pronouns in these configurations, and which possibly render the extended PP a phase and thus a local domain for the purposes of binding theory. These are the employment of a referential possessive structure in certain locative PPs, and the projection of a CP PP cap in others (see Dékány 2018, andHegedűs 2015 for independent arguments that such structure building is available in Hungarian PPs). Our experimental results also indicate that there is substantial variation across speakers in the extent to which they tolerate the pronoun strategy. We propose that this variation correlates with variation in the syntax of the PP in terms of the availability of the two structure building operations that we have described here.
We tested first person examples, because the effects we describe are strongest in first and second person. We concluded nevertheless that the pronoun strategy may become an option in spatial anaphora if all the possible licensing factors are present, but generally it is still a marked option in contrast to reflexives. Further investigations are required to understand the finer details of variation in each person and number, but the factors we have identified here play a crucial role across the board.
In addition to the factors we have studied, there are also other possible licensing factors. These include focusing of the PP or a part of it, or the potential effect of an extra verbal particle, if one is present. Our preliminary investigations did not reveal an obvious influence of these factors, but here too, further data collection is required to arrive at a better understanding of the facts.