Inanimate antecedents of the Japanese reflexive zibun: experimental and corpus evidence

1 Introduction

The study of the Japanese reflexive pronoun, zibun ‘self’, has generated copious analyses in the syntactic (Aikawa 1999; Iida 1996; Katada 1991; Nishigauchi 2014), semantic (Oshima 2006; Tenny 2006), pragmatic (Sakakibara 1995), cognitive (Hirose 2014, 2018; Uehara 2003), and functional (Kuno and Kaburaki 1977) realms. Its theoretical importance for anaphoric binding has been underscored by considerable experimental work in first-language acquisition (Fujiwara 2013; Ito 2012), second-language acquisition (White et al. 1997; Yoshimura et al. 2012), psycholinguistics (Aoshima et al. 2009; Nagata 1995), and neurolinguistics (Okabe et al. 2011). This wealth of research has captured many different aspects of zibun, however, there is an important feature that has escaped detection. This new development relates to the well-known animacy constraint of zibun, whereby zibun is unable to be anteceded by an inanimate antecedent as discussed by Kuno (1973), (1) being a typical example:^[1]

The animacy requirement was formalized by Kuno (1973: 291) stating, “The antecedent of zibun ‘self’ in Japanese must be something that is animate and has willpower”, which naturally excludes objects such as trees or stones. In the literature on Japanese reflexive pronouns, this has been typically repeated or noted that zibun requires an animate antecedent, as usually the focus has been understanding zibun’s binding behavior using typical human antecedents. However, our data (in Section 2) provides evidence contrary to the animacy constraint. In what follows, there are examples of inanimate zibun that have a verb designating an action that must or can be performed by non-human entities.

We quickly note that the data excludes cases of objects that have been personified, acquiring human-like qualities (cf. Dorst 2011), as exemplified by (2) below.

Personification has an effect of describing non-human entities as taking on human nature. For example, the ability to talk is an essential attribute that pertains only to human beings, not to the moon, hence personification at work in (2). As this example illustrates, a personification effect is typically discernible if the predicate taking the inanimate subject describes a human action (cf. Bloomfield 1963: 165–166). Consequently, no example of a personified inanimate zibun forms part of our dataset.

The type of zibun investigated here is the reflexive use of zibun, and not the viewpoint (empathic) or logophoric use of zibun.^[2] The data that we deal with sits firmly with the reflexive use of zibun, due to the fact that an inanimate object is unable to adopt a perspective (logophoricity), and it is difficult to adopt an empathic perspective for the same object (empathy). This naturally leaves reflexive zibun as our target of discussion, given that it simply reflexive-marks its predicate. Note that when the term zibun is used henceforth, we refer to it in its reflexive usage, unless otherwise stated.

In the next section, we first demonstrate the three uses of zibun as mentioned above to orientate the reader, and then present new data that demonstrate that reflexive zibun can indeed take a genuine inanimate antecedent (without the presence of personification),^[3] contrary to the predictions of Kuno’s animacy constraint. This, however, should not be interpreted as a complete falsification of the animacy constraint as there are innumerable cases where the animacy distinction is clearly corrected. Rather, instead of taking the constraint as inviolable, we present the case further on that a revised animacy constraint plays a key role in restricting zibun’s potential antecedents, as it is possible to overcome it given the right context.

2 The data

3 Judgments and processing of zibun

As discussed in the introduction, while there is a breadth of research on zibun, the idea that zibun cannot be anteceded by an inanimate antecedent has gone thus far unchallenged. As the evidence presented above represents the real use of zibun taking an inanimate antecedent, it is now important to investigate how native Japanese speakers interpret and process such sentences. That is, do native speakers find these types of examples acceptable? Will processing inanimate zibun be an effortful undertaking?

Given the discussion of the data above, there appear to be two factors at play. The first evident one is that of animacy (given the animacy condition), as it is unmistakable that the status of the antecedent’s animacy will affect the binding probability. The second issue is that of episodic versus generic sentences, for as discussed, the type of sentence zibun occurs in with an inanimate antecedent may affect its acceptance. As there are no experimental studies to draw upon,^[8] and working with the information that we have on hand, we initiate a self-paced reading (SPR) experiment combined with judgment ratings (along with reaction timing) to better understand inanimate zibun, which is detailed in the next section. The three hypotheses assessed by the experiment undertaken are presented thus:

4 Results

The first analysis presented is on the acceptability judgments of the sentences tested to certain how acceptable inanimate zibun is, as structured here. This is then followed by observing zibun’s processing profile. Analyses are completed using SPSS 25.

5 Discussion

This article began with the introduction of the unexpected allowance of an inanimate antecedent for zibun, theorized not to be possible yet found in corpus materials. We then explored the acceptability of inanimate zibun and investigated its processing profile. We now revisit the experimental hypotheses, discuss the experimental results together with the corpus data in more general terms, and finally shift to the theoretical side of the problem.

Beginning with the hypothesis assessment, H1, inanimate antecedents are judged slightly less acceptable than their animate counterparts but are still found to be acceptable, in line with the hypothesis. Regarding H2, it is clear that the Episodic and General sentence types do not contribute to a different processing profile of inanimate zibun, nor in the judgment reaction timing, falsifying the hypothesis. Finally, assessing H3, while the hypothesis holds at the by-subject and by-item levels for the CRT and PST regions (inanimate zibun takes longer to process), this was unable to be finally generalized across both levels.^[15]

A reviewer notes that it is impossible to rule out that a metaphorical interpretation was not assessed, i.e., the antecedent was personified by the participant, and not assessed as an inanimate antecedent. While it is not possible to fully understand what the participant is thinking at the time, on the balance of probabilities we believe that indeed the participants were consistently assessing the antecedent as inanimate without personification for the following reasons: (i) the lead-in contexts for inanimate interpretations were of factual nature, as found in our corpus data, where we have specifically avoided contexts that support a metaphoric reading (such as fables);^[16] (ii) some of our examples involve verbs that cannot take an animate object, e.g., sakuzyosuru ‘erase, delete’ in (5b) or kesu ‘delete’ due to the nature of the designated actions (e.g., *The man deletes himself); therefore, such sentences would not make any sense if the antecedent and its self were interpreted as animate; (iii) the antecedents given in our examples can be followed by the existential verb aru, which is generally used with inanimate subjects, whereas it cannot be by the verb iru, which only take animate subjects in (15).

The distinction between aru and iru in Japanese is related to a more general cross-linguistic pattern of classification by verb on animacy (Croft 1994), and can be utilized as a diagnostic for evidencing the speaker’s conceptualization of objects; (iv) as metaphorical interpretations are commonly found in everyday discourse, that then the inanimate antecedent has degraded acceptability is odd – unless there is something causing the hesitation – the consideration of the inanimate antecedent non-metaphorically; (v) the SPR effect found started at the CRT region (before sentence end), suggesting that the participants are engaged with antecedence assignment, extending to the following region PST (sentence end); if the participants are dealing with a metaphorical interpretation, we would expect the effect to start at PST and extend across the spillover regions, as the literal interpretation of the sentence would be processed and compared against a metaphorical interpretation – this process should start once the sentence is finished, but it did not; (vi) regarding judgment reaction timing, although it is possible that the participants are now dealing with a metaphorical interpretation (the sentence is now fully parsed), considering the context provided that has steered them towards an inanimate antecedent judgment, an inanimate antecedent assessment fits better with our understanding of the antecedent judgment and prior parsing.

Put together, these results suggest that inanimate zibun appears to be processed slightly slower than animate zibun,^[17] with timing differences in the judgments showing (inanimacy judged slightly slower), but these judgments are nonetheless found to be acceptable (but at a slightly lower level than the animate counterparts).^[18] Thus, the experimental results reinforce what the corpus data shows, that an inanimate antecedent of zibun is acceptable within the right context (without personification), contrary to what has been presented in the previous literature. We hasten to add that this does not disprove Kuno’s animacy constraint, as sentences such as (1), among others, do not suddenly become acceptable. Rather, this constraint is clearly active in Japanese, but its strength of application is found gradable rather than binary, as it is possible to override it. In fact, animacy has often been perceived as a gradient notion in the linguistic literature and often represented as a category along a hierarchy or continuum (Aissen 2003; Bossong 1984; Yamamoto 1999; see also de Swart and de Hoop 2018). A reviewer also points to the following examples (16a) and (16b) as problematic if one does not assume the animacy constraint (which we do not in its current form).

The reviewer points out that, without the animacy constraint, one might think it would be hard to account for the clear difference in grammaticality. While we propose a revision to the animacy constraint further below in (20), these examples suggest the particular importance of the interaction between animacy and the speaker. As we have discussed in Section 1, according to Hirose (2000, 2002, 2014, 2018 and Oshima (2004, 2006, in addition to the reflexive use, there are two other uses of zibun, which serve as non-clause-bounded reflexive pronouns: viewpoint (empathic) and logophoric. Viewpoint zibun is used to signal the speaker’s perspective being placed or projected onto that of its referent and typically appears in an adverbial subordinate clause or a relative clause. Logophoric zibun, on the other hand, “occurs in the indirect-discourse complement of a saying or thinking verb and refers to the original speaker of indirect discourse” (Hirose 2018: 379).

In both uses, the speaker plays a pivotal role. In the viewpoint use, the speaker projects his/her viewpoint onto the referent of zibun. His/her viewpoint is therefore projected onto an animate participant only, as illustrated in (17), where we find it difficult to empathize with John:

Likewise, in a logophoric context, the antecedent must be conscious as the subject of thinking or communicating, and inevitably animate, just as the narrator of a story must be more or less regarded as animate.^[19] For instance, (18) does not make any sense because the corpse, as an inanimate entity, cannot be understood as conscious:

In (16a) and (16b), we see that zibun appears in relative clauses, i.e., contexts for the viewpoint zibun, where the speaker’s viewpoint is cast onto its inanimate antecedent. It is unlikely to place one’s perspective on and empathize with inanimate entities, hence the unacceptability with zibun.

A reviewer, however, does present an interesting counter example showing the speaker being able to empathize with an inanimate object, as (19) below:

The animacy constraints on the viewpoint and logophoric uses of zibun is thus naturally drawn from the fact that the speaker in general is supposed to be animate; however, as the above example shows, even this can be overruled under special circumstances.^[20] By contrast, in our earlier examples, e.g., (6a)–(6c), zibun is the co-argument of the antecedent and expresses reflexivity of the action, without any viewpoint or logophoric aspect being present. We can then generalize that the reflexive use productively allows for inanimate antecedents. We now discuss the significance of this observation in relation to the corpus data and corresponding theoretical issues.

Beginning with the empirical data, it is curious that generic sentences are found with inanimate zibun but not episodic sentences in the corpus data assessed. The absence of episodic sentences, however, does not predict difficulty in the processing or interpretation of these types of sentences: the distinction between the generic and episodic sentences, exemplified by (9) and (10), is not an important factor for inanimate zibun. The question then becomes, why haven’t episodic inanimate zibun sentences been found in the wild?

We simply might consider that a generic sentence is more likely to be found because it is simply reporting on the state of the inanimate antecedent – which is not something uncommon. However, it is uncommon for an inanimate to perform an action on itself – it must have some type of agency. The examples used in the experiment here focused on a new species of tree and a computer program that could act upon themselves, supported by a context that suggested some agency.^[21] This shows that part of the puzzle is that if one offers a context for agency for an inanimate object, an inanimate zibun can be acceptable (whereas example (1) does not meet these criteria).^[22] In this relation, it is a well-known fact that Japanese has a tendency to disallow inanimate referents to occur as transitive agents as in (20) (e.g., Kuno 1973):

As we have seen in Section 2, however, the corpus data suggest that generic sentences and other generalizing constructions more easily allow for inanimate subjects in a transitive sentence (Kageyama and Shen 2012: 40). If agency is the key to allowing inanimate antecedents of the reflexive zibun, it then follows that attested examples of inanimate zibun are more commonly found in generic contexts, as inanimate referents with agency in general tend to be allowed only in those linguistic environments. In this sense, the perceived inanimate zibun problem is less syntactic and more semantic/pragmatic.^[23] We now turn to the theoretical issues surrounding the allowance of inanimate zibun.

A positive step taken concerns theoretical clarity: inanimate zibun should not be accounted for as an inviolable phenomenon in the syntactic or semantic interfaces. Specific syntactic/semantic proposals accounting for inanimate zibun from early (e.g., Aikawa 1993) to more recent ones (e.g., Tenny 2006), in this light, are too strong as they cannot capture the new data presented here. Consequently, by adopting the hypothesis that inanimate antecedents with a high degree of agency can bind zibun, and incorporating Oshima’s (2007) empathy considerations, Kuno’s (1973) animacy constraint (while covering the general data well) is reformulated as thus (21):

Moreover, we are in agreement with a reviewer that rather than a lexically specified restriction, the binding of zibun in relation to animacy is due to an amalgamation of factors, which the above condition tries to capture. This thus shifts the research direction to specifying the general type of context and circumstances, and what the threshold of agency is for inanimates. Based on the sentences we have found and tested, it appears that this insentient agency is attributable to, for example, evolutionary pressures (involving plants) or programmed behavior (involving computers); in a sense, agency related to preprogrammed behavior which occurs automatically when a particular set of conditions are met. In fact, entities with insentient agency are construed as ‘independent instigators’ (Fauconnier 2011: 539) who autonomously act upon themselves without being manipulated by an implied controller. Inanimate entities endowed or equipped with any preprogrammed behaviors, for example, trees or computer applications, are regarded as agents in this sense. By contrast, it would be highly unlikely for concrete objects such as ‘rock’ or ‘book’ to ever serve as an inanimate antecedent for zibun whatever the context, or abstract nouns such as ‘honesty’ or ‘history’, as they cannot be adapted in the same manner above. Under this approach, the fluid nature of animacy surrounding zibun stems from how we conceptualize inanimate entities in terms of their agency depending on the context that it is used within.

Incidentally, the reader will notice that in Kuno’s example in (1), repeated here as (22), the reflexive is used as a maker of a middle situation, i.e., occurs in a context of derived intransitivity (unlike (3a)–(3b)).

Itself in the English translation marks the intransitivity of the verb; more accurately, the verb repeat has been detransitivized. In fact, it is omissible with no change to the sentential interpretation. As discussed by Siemund (2010: 814–815), such an optional occurrence of itself has important commonalities with intensifying self-forms of the adverbial type, as with The budgie opened the cage door itself. However, zibun does not have such intensifying functions; it is another reflexive pronoun, mizukara, that takes on an intensifying function in Japanese and therefore more suitably functions as a maker of a middle situation, as evidenced by the acceptability of Rekisi-wa mizukara kurikaesu ‘History repeats itself [without any help].’ The main point is that zibun has a reflexive use where it acts as an argument of the verb and refers to an event participant involved in a transitive eventuality; that zibun lacks a middle marker usage does not however, implicate that it disallows inanimate antecedence.

It is then predicted that if zibun takes on such an intensifying function, it can more readily be used with an inanimate antecedent. In fact, as seen in Section 2, there are intensifying adverbial instances where zibun can occur with an inanimate subject, zibun-kara ‘from self’ (23) and zibun-de ‘by self’ (24) (cf. Kogusuri 2018, 2019):

Notice here that the sentence in (23) has only an episodic reading, as the verb is used in the simple past form. This adverbial use of zibun is much more commonly found than accusative zibun with inanimates, as the intensification function focuses on actors and their corresponding actions (the agency of itself, without others, see Hirose (2014: 107)), rather than the action’s effect on an antecedent (even though the reflexive and subject are one of the same).^[24] Here, we see another path of allowing inanimate zibun, not by extending the categorical boundary of ‘animate’ with a high degree of agency, but by shifting its function from anaphoric reference to intensifying an actor role in the event.

6 Conclusions

This study has presented data on reflexive zibun being able to take an inanimate antecedent, something that was said to be impossible since Kuno (1973). This was further verified through an experimental paradigm which explored the processing and interpretation of inanimate zibun, showing that while it is processed slower and has lower acceptability than animate zibun, it appears that this is allowed as part of the grammar given the proper context. Based on Hirose’s (2014) trichotomy of uses of zibun, we have argued that, only the reflexive use, but no viewpoint or logophoric ones, can allow for inanimate antecedents if they are conceptualized as entities with a high degree of agency, or if zibun is used as an adverbial intensifier emphasizing the antecedent’s actorhood. We then offered a reformulation of the animacy condition to reflect the inanimate binding possibilities (considering Oshima 2007), and explored reasons why episodic sentences with inanimate zibun are difficult to find in corpora. Finally, we explored theoretically the base reasons why the animacy condition exists and how generalizing contexts, as attested in the corpus data, are related to inanimate zibun.

The important implication of our conclusion is that animacy, governing the use of zibun, is a conceptually flexible notion, dependent on the speaker’s conceptualization as well as the context where it is used. In this sense, zibun’s binding is semantic and pragmatic in nature.