Comparing reflexive and personal pronouns in Chinese locative prepositional phrases

. In many languages, the complementarity captured by syntactic binding principles breaks down within locative prepositional phrases (LPPs), where both reflexive pronouns and personal pronouns can be coconstrued with a local subject. LPPs are therefore an ideal testing ground for non-syntactic factors influencing pronoun use. Focusing on Chinese, we experimentally tested the extent to which acceptability of reflexives ziji and ta-ziji and personal pronoun ta depends on two semantic factors: event type and spatial relation type. Findings reveal effects which follow the trend of those previously reported for English, but which affect different forms to different degrees. Along with advancing understanding of binding in LPPs, this work contributes to comparisons between ziji and ta-ziji and the typology of pronouns more broadly.

1. Introduction.In many languages, both reflexive pronouns and personal pronouns can be used to refer back to someone mentioned earlier in the sentence.Often times the choice between pronoun forms is entirely determined by syntactic binding conditions.We see this for instance in the English sentences in (1a), where Condition B rules out use of the personal pronoun her to refer to Michele,and (1b), where Condition A rather renders the reflexive herself ungrammatical.
b. Michele i 's brother congratulated her i /*herself i .
However, in many languages, the kind of strict complementarity shown in (1) breaks down in locative prepositional phrases, or LPPs.We see this for instance in the English example in (2), where the reflexive and personal pronoun are both acceptable under the same interpretation.
(2) Michele i set a glass next to her i /herself i .
Following Reinhart & Reuland (1993), Büring (2005), and Bryant (2022b) and among many others, we assume that Conditions A and B are both satisfied in sentences such as (2), meaning both reflexives and personal pronouns are syntactically licensed.LPPs are therefore the ideal testing ground for factors outside of narrow syntax that help shape pronoun choice.
This paper focuses on two semantic factors: the type of event denoted by the clause containing the LPP-either motion or perception-and the type of relation denoted by the LPP itselfeither contact or non-contact.Both factors have previously been shown to have a robust effect on pronoun choice in English LPPs.However, cross-linguistic research on LPP-internal pronouns remains limited, and it is an open question whether and to what extent similar effects are found in other languages.
In order to address this question, we investigated the effects of event type and relation type on LPP-internal pronouns in Chinese.Like English, Chinese allows reflexives and personal pro-nouns within LPPs to be co-construed with the sentence subject, as shown in (3).But unlike English, Chinese has two reflexive forms, simplex ziji and complex ta-ziji, in addition to a personal pronoun, ta.We tested the relative acceptability of ta, ziji, and ta-ziji in sentences such as (3) in a joint presentation acceptability judgment experiment.Findings reveal effects of event type and relation type which follow the general trend of those observed in English, but which affect different pronoun forms to different degrees.Along with advancing understanding of binding in LPPs, this work contributes to a growing literature on differences between ziji and ta-ziji and the typology of pronouns more broadly.
Section 2 introduces the effects of event type and relation type in English, which serves as the empirical starting point our investigation of Chinese.Section 3 introduces ta, ziji, and taziji in more detail, setting up the central empirical questions driving this work.The acceptability judgment experiment is presented in Section 4, followed by discussion of the results in Section 5. Section 6 concludes.

Event type and relation type in English.
While it is the case that subject-oriented reflexives and personal pronouns are both generally allowed in English LPPs, it is not the case that both pronoun forms are equally favored.Judgments can vary widely from one sentence to the next, with a host of factors coming together to push preferences one way or another.
In their seminal work on English pronominalization, Lees & Klima (1963) observe that pronoun choice within LPPs may depend on the nature of the meaning relation between the LPP and its host clause, which in turn depends on the type of event the clause denotes, illustrating with the minimal pair in (4) (see also Wilkins 1988;Hestvik 1991).
(4) a.The men i threw a smokescreen around themselves i .
b.The men i found a smokescreen around them i .[Lees & Klima 1963:(12)-( 13)] As stated in Bassel (2018Bassel ( , 2023)), the descriptive split appears to be between sentences that express motion and those that do not.Based on examples like (4), we should expect reflexives to fare better in motion sentences than in non-motion sentences (all else being equal), while personal pronouns should fare better in non-motion sentences than in motion sentences.
2 Event type alone does not determine which pronoun form will come out the winner in English LPPs.Kuno (1987) observes that preferences also depend on the spatial relation denoted by the LPP, in particular whether or not the relation involves direct contact between figure and ground.Th effect of relation type is exemplified in (5), originally from Wechsler (1997) (see also van Hoek 1997;Strahan 2006;Lederer 2013).
(5) a. Corporal Crump i pinned the medal onto *him i /himself i .
b. Corporal Crump i pinned the medal beside him i /*himself i .[Wechsler 1997:(39)] Such examples suggest that reflexives should be most natural in LPPs that express direct contact between figure and ground (e.g., between Corporal Crump and the medal) while personal pronouns should be most natural in the expression of more distal relations.The impact of event type and relation type on English LPP pronoun acceptability were experimentally confirmed in Bryant (2022a,b) in a joint-presentation sentence rating study.Results revealed robust effects that were largely symmetrical for reflexives and personal pronouns: the former were most natural in the expression of motion and contact, while the latter were most natural in the expression of non-motion and non-contact.Importantly, despite the suggestion of complementarity in some prior work (cf.( 5)), neither factor was found to fully determine relative preferences.Rather, relative preferences were gradient across sentence conditions, and even when one form was reliably favored over the other, contrasts in acceptability were significantly less stark than those observed for sentences like (1a-b), where syntactic binding conditions render one form ungrammatical.The overall preference pattern turned up in Bryant (2022a,b) is captured in (6). 3 (6) a. Chloe i poured some glitter on ??her i /✓herself i .
b. Chloe i poured some glitter on ?her i /?herself i .c. Chloe i noticed some glitter on ?her i /?herself i .d. Chloe i poured some glitter next to ✓her i /??herself i .
3. Chinese pronoun forms.With the English picture in place, we turn our attention to Chinese.
As introduced in Section 1, Chinese has two reflexive pronoun forms, simplex ziji and complex ta-ziji, as well as a personal pronoun form, ta.Each of these forms is third-person singular and restricted to animate antecedents.Note that while ta and ta-ziji do not show a gender distinction in the spoken or transliterated forms, there is a feminine and masculine distinction in Chinese orthography.
Just as in English, Chinese reflexives and personal pronouns are in complementarity distribution in simple transitive sentences like (7a-b).In (7a), use of ta to refer to Lijuan is ruled out by Condition B, such that either ziji or ta-ziji must be used instead.In (7b), both reflexive forms are ruled out by Condition A, leaving ta as the only grammatical option.( 7 To the best of our knowledge, no prior work has investigated Chinese pronouns within LPPs.Our primary question therefore concerns the relative acceptability of reflexives and pronouns in Chinese LPPs.In particular, we ask whether the acceptability of ta, ziji, and ta-ziji in sentences like (8) displays a sensitivity to event type and relation type and, if so, how this sensitivity compares to the one reported for English.
Pursuit of this question also bears on a growing body of literature comparing simplex ziji with complex ta-ziji more broadly.As shown in (7a) above, both ziji and ta-ziji can be anteceded by a local subject.Ziji is also compatible with non-local antecedents, as shown in (9).
(11) Zhangsan i Zhangsan shuo said nabenshu that-cl-book hai-le hurt-PFV ta-ziji i himself 'Zhangsan i said that book hurt himself i .' [Pan 1998:(4)] In fact, experimental work by Lyu & Kaiser (2023) shows that a non-local antecedent is also possible, and even preferred, across an animate intervener when coconstrual with the latter is ruled out by the semantics of the embedded verb.This is for instance the case in ( 12), where the verb chuipeng 'flatter' biases towards an other-oriented reading, and the preferred referent of ta-ziji is the chairman rather than Director Li.Nevertheless, contrasts like those between ( 9) and ( 10) suggest that ta-ziji may have a stronger locality bias than ziji.In line with this, Lu (2011) and Lyu & Kaiser (2023) report that ta-ziji is less sensitive to discourse factors that can push ziji to favor non-local antecedents, including topic prominence and contextual plausibility.Interestingly, however, ta-ziji has been shown to be preferred over ziji in some instances of non-local antecedence.Dillon et al. (2016) found that ziji was associated with lower naturalness ratings and longer reading times than ta-ziji in so-called sub-commanding configurations where the target antecedent is embedded within an inanimate local subject, as in ( 13).( 13 Furthermore, Liu (2022) reports that ta-ziji is rated as 'acceptable' more often than ziji in sentences such as ( 14), where the non-local antecedent is introduced as the topic of the sentence in an initial paranthetical.
(14) Shuodao speak-of XiaoLi i , XiaoLi zhejianshi event shanghai-le hurt-PFV ziji i /ta-ziji i self/himself 'Speaking of XiaoLi i , this event hurt himself i .' [Liu 2022:(17b),(26b)] In short, the relative acceptability of ziji and ta-ziji depends on the particular constructions in which they occur in ways that are not yet fully transparent.Ta-ziji displays a stronger locality bias in some cases, while ziji does in others.While we do not assume that LPP sentences such as (8) involve non-local antecedence, we may still find differences in the relative acceptability of LPP-internal ziji and ta-ziji.On the one hand, the reflexives are c-commanded by their antecedent, as is the case in non-local examples where ziji has the advantage.On the other, there is no animate intervener, like in examples ( 13)-( 14) where ta-ziji is preferred.Investigating reflexives in LPPs can therefore help clarify the conditions favoring one reflexive form over the other.
4. Acceptability Judgment Experiment.The goal of this experiment was to test the relative acceptability of subject-oriented ta, ziji, and ta-ziji within LPPs.We were specifically interested in whether and to what extent acceptability depends on event type (motion / perception) and relation type (contact / non-contact).
4.1.STIMULI.16 sets of target sentences were created by crossing event type, relation type, and pronoun form (ziji / ta-ziji / ta), resulting in 192 target sentences in total.One sentence set used in this experiment is shown in (15) below.Motion sentences all had the structure NP 1 -ba-NP 2 -V-PP while perception sentences all had the structure NP 1 -V-NP 2 -PP.Direct objects (NP 2 , for instance liang-pian 'glitter' in (15)) were always inanimate, such that the sentence subject was the only possible sentence-internal antecedent.
( Sentences were designed to be as similar to their English counterparts as possible in order to allow comparison between the two languages.However, there are three ways in which the surface structure of the Chinese sentences included in this experiment diverged from the English sentences tested in Bryant (2022a,b), exemplified in (6) above.
First, the particle ba was included in motion sentences as its use was deemed most natural by the first author and additional native speaker consultants.
Second, all Chinese LPPs include a preposition (e.g., dao 'to' and zai 'at' in (15a-d)) as well as a postposition (e.g., shang 'on' and pang 'next').Following Wu (2015), we assume that the surface differences between Chinese and English arise from differences in head movement and that the underlying structures of the two languages are roughly equivalent.
Third, in most sentences pronoun forms were followed by a body term, either shen 'body' (see ( 15)) or tou 'head'.The Chinese possessive particle de was not used in order to minimize structural and interpretive differences betweeen the Chinese stimuli and previously tested English sentences.Nevertheless, the difference in the complexity of the LPP complement could lead to differences in the relative acceptability of reflexives and pronouns in the two languages.We return to this point in Section 5.
As shown in the sample question in Figure 1, minimal triplets varying only in pronoun form were presented side-by-side.As in Bryant (2022a,b), joint presentation was used to help draw out relative preferences between pronoun forms and to help constrain interpretation of the personal pronoun ta.Each sentence appeared next to two-point acceptability rating scale, where 'yes' indicated that the sentence was acceptable and 'no' indicated that the sentence was unacceptable.Sentence triplets were preceded by a short supporting context intended to support consistent readings across participants.Contexts named only one person, the subject of the target sentences, to further ensure that personal pronouns were correctly interpreted as referring to the sentence subject rather than someone else in the discourse.
In addition to target sentences, we also tested four simple transitive control sentences, exemplified in (16a-b).In reflexive controls such as (16a), the sentence subject is the only possible antecedent for the direct object pronouns, and either ziji or ta-ziji must be used.In anti-reflexive controls such as (16b), the context introduces a second discourse referent that contrasts in gender with the sentence subject.Use of an other-oriented verb (here, zhui 'chase') and gender marking on ta and ta-ziji6 rule out reference to the sentence subject and, hence, rule out use of ziji and ta-ziji.Control sentences thus provided a baseline for (un)acceptability against which ratings for target sentences could be compared.( 16 74 adult Chinese speakers were recruited online via Prolific Academic.71 were self-reported native speakers of Mandarin, 2 were native speakers of Cantonese, and 1 was a native speaker of the Wu dialect.Participants were instructed that they would be comparing sentences with the same meaning by choosing whether each is something they would say in the context provided.Training trials demonstrated that more than one sentence could be endorsed with a 'yes' response per question.Following training, participants were randomly assigned to one of four question lists, each containing 16 target questions (1 per sentence set, 4 per condition) along with 4 control questions and 20 filler questions, randomly presented.4.3.PREDICTIONS.We expect to see complementarity between reflexives and personal pronouns in control sentences, such that ziji and ta-ziji are judged as acceptable only in reflexive sentences while ta is judged as acceptable only in anti-reflexive sentences.If Chinese pronouns show a similar sensitivity to event type and relation type within LPPs as has been observed in English, we would expect reflexive forms ziji and ta-ziji to be judged as acceptable more often in motion sentences than in non-motion sentences and in contact sentences than in non-contact sentences, and we would expect the personal pronoun ta to show the opposite pattern.4.4.RESULTS.Results for target and control sentences are shown in Figure 2. Data were analyzed in R using logistic mixed-effects models (lme4 package, Bates et al. 2015) with acceptability judgment (1 = 'yes') as the dependent variable and subject, item, and sentence set as random intercepts.Starting with control sentences, we found very clear evidence of complementarity between reflexives and personal pronouns.There was no significant difference in endorsement rates (rates of 'yes' responses) between ziji and ta-ziji in reflexive controls (β=0.74,p=0.18), giving us confidence that participants were comfortable endorsing more than one pronoun form per question.Collapsing reflexive forms, we found that endorsement rates were significantly lower for ta than for ziji and ta-ziji in reflexive controls (β=-5.45,p<0.001), while they were significantly higher in anti-reflexive controls (β=5.97,p<0.001).
Turning to target sentences, we did not find evidence of complementarity between reflexives and personal pronouns in Chinese LPPs: endorsement rates were above 60% for all three pronoun forms across all four target conditions.This is in stark contrast to what was found in control sentences, where endorsement rates were less than 10% for disfavored forms.
We first ask to what extent endorsement rates for each pronoun form depended on event type and relation type.Endorsement rates for ziji did not significantly depend on event type (β=-0.25,p=0.30) or relation type (β=-0.17,p=0.45).Conversely, endorsement rates for ta depended significantly on both event type (β=1.29,p<0.001) and relation type (β=0.44,p<0.01).Specifically, endorsement rates for ta were higher in perception sentences than in motion sentences, and they were higher in non-contact sentences than in contact sentences.Finally, endorsement rates for ta-ziji depended significantly on relation type (β=-0.51,p<0.01), with higher endorsement rates in contact sentence than in non-contact sentences, but did not depend on event type (β=0.23, p=0.20).
We next ask how endorsement rates compared across pronoun forms.Endorsement rates for ziji were high across target conditions, around 91%, on par with the endorsement rates observed for ziji in reflexive controls.Unlike in reflexive controls, endorsement rates were significantly higher for ziji than for ta-ziji in all four target conditions (β=1.56, p<0.001).Endorsement rates were also higher for ziji than for ta in all target conditions except perception/non-contact, where the difference between forms was not significant (β=0.23,p=0.45).Finally, while endorsement rates were significantly lower for ta than for ta-ziji in motion/contact sentences (β=-0.52,p=0.01), they were significantly higher in perception/non-contact sentences (β=1.59,p<0.001) and marginally higher in perception/contact sentences (β=0.58,p=0.05), and there was no difference between ta and ta-ziji in motion/non-contact sentences (β=-0.08,p=0.70).
5. Discussion.The overall high endorsement rates found for subject-oriented ta, ziji, and ta-ziji confirm that all three pronoun forms are syntactically licensed in Chinese LPPs.In fact, the relative preferences between forms captured in this experiment were never as strong as those turned up in prior work for English sentences such as (6a) and (6d), where speakers reliably favor the reflexive and personal pronoun, respectively.The relative weakness of the contrasts found in the present study could relate to the difference in LPP complement structure flagged in Section 4.1-that is, the use of an additional body term in Chinese LPPs.While the inclusion of the body term does not appear to affect the accessibility of the sentence subject as a suitable antecedent for LPP-internal reflexives, it may provide an additional locality buffer for LPP-internal personal pronouns.
7 It could also be a product of the response method, as binary choice ratings are a less sensitive measure than the continuous sliding scales used in Bryant (2022a,b) on English LPPs (see Marty et al. 2020 for methods comparison).Of course, it is also possible that the differences that emerge between English and Chinese are rooted in the grammars of these two languages.More work is needed to tease these explanations apart.Most importantly for our purposes, we found that relative preferences between LPP-internal pronouns in Chinese are shaped in part by event type and relation type.Unlike in English, the effects of event type and relation type in Chinese are form-specific.They are mainly apparent for the personal pronoun ta, endorsement of which was found to depend on both factors.In contrast, only relation type affects the acceptability of the complex reflexive ta-ziji while neither factor affects acceptability of the simplex reflexive ziji.Nevertheless, where effects of event type and relation type were found in Chinese, they ran in the same direction as they do in English.In particular, ta was accepted more often in perception and non-contact sentences than in motion and contact sentences, like English her, while ta-ziji was accepted more in contact sentences than in non-contact sentences, like English herself.This suggests that there is something universal about the way event type and relation type can can shape LPP-internal pronoun acceptability, if they have an effect at all.
Our results also shed new light on the comparison between simplex ziji and complex ta-ziji.While both reflexive forms are generally acceptable in LPPs, ta-ziji appears to be more marked 10 than ziji, even in the absence of an animate intervener.Furthermore, only ta-ziji showed any sensitivity to the semantic factors tested here.Under the assumption that the sentences investigated in this study involve local binding of LPP-internal reflexives, this outcome suggests that ziji is the default local anaphor in Chinese, subject only to syntactic constraints, while use of ta-ziji may require additional semantic or pragmatic motivation.

Conclusion.
The experiment presented in this paper expands the cross-linguistic view on pronoun use within LPPs, demonstrating that two semantics factors at play in English-event type and relation type-are also active in shaping pronoun acceptability in Chinese.On the one hand, our findings reveal variation in the extent to which these factors matter across and even within languages.On the other, they point to a possible universal in the role these factors can play.This study thus sets the stage for future work investigating the impact of semantic constraints on pronoun use cross-linguistically.

Figure 1 .
Figure 1.Sample question from the acceptability judgment experiment.Survey materials were presented entirely in Chinese characters; English translations are included here for clarity.

Figure 2 .
Figure 2. Rates of sentence endorsement ('yes' responses) by pronoun form across control conditions (left) and target conditions (right).