A great deal about the human sentence processor has been learned from investigating how pronouns are integrated into sentence comprehension in real-time. Pronouns require referents, and these can be sourced from the extra-linguistic context or linguistic antecedents. The form of a pronoun—encoded by number, person, and gender features—constrains its reference, something that can be detected in the earliest moments of processing (Arnold et al., 2000). Some English pronouns, however, do not bear features that easily and unambiguously determine their referents, tolerating a range of antecedents with differing features. One particularly interesting case is the English pronoun series they/them/their and its reflexive counterpart (themselves/themself). These anaphors admit several different types of grammatical antecedents, subject to differences in semantic interpretation, discourse context, and speaker variation. They/them/their accepts a plural antecedent (1a), and readily takes a singular antecedent given certain discourse and semantic properties (1b).

Known as “singular they”, this second usage is undergoing changes in progress in the types of linguistic antecedents it takes and the individuals it references (Camilliere et al., 2019; Conrod, 2019; Konnelly and Cowper, 2020). It is also the pronoun of reference for many individuals.

There is some evidence that they is underspecified along several dimensions and that this confers processing advantages upon it. Moxey et al. (2004) found, in comparison to instances where they lacked a salient plural antecedent, earlier disruptions in reading instances of she/he lacking a salient singular antecedent. Similarly, using ERP methods, Filik et al. (2008) found evidence of a cost for unheralded she/he but not for they. Sanford et al. (2008) looked at so-called “institutional they”, where the reference to an individual or group is highly underspecified (as in On the train, they served coffee), and found no processing costs compared to unheralded singular pronouns. These authors suggest that they is an underspecified pronoun and so will tolerate a wide range of antecedent types, posing no immediate processing difficulty [perhaps under shallow or good-enough processing (Ferreira et al., 2002)], yet possibly requiring greater resources in later processing (Moxey et al., 2004). Other such underspecified pronouns have been attested by Poesio et al. (2006) and the time course of their integration into sentence comprehension suggests similar conclusions.

At the same time, there is also evidence that in other contexts they exhibits a preference for plural antecedents. In an eye-tracking while reading study, Sanford and Filik (2007) found evidence of processing difficulty for they when only a singular antecedent was available. They argue that they expects a plural antecedent but can later accommodate a singular antecedent when no plural is found. In an ERP study, Chen et al. (2021) elicited a P600 effect for they with a singular antecedent in comparison to a plural antecedent, suggesting some processing difficulty for singular they. Similarly, in a Maze task study, Van Handel et al. (2021) report results that suggest that reflexive themselves preferentially links to a plural antecedent over a competing singular antecedent.

The gender of the antecedent also plays a role in the processing of they and themselves pronouns¹. Prasad et al. (2018) found that the reflexive themselves elicited a P600, often taken to arise from syntactic anomaly, with singular antecedents that are associated with a gender stereotype (John) but not with non-gendered antecedents (the participant). They argue that they/themselves is underspecified for gender, and the processing cost is gender rather than number mismatch.

The studies cited above are limited to instances of “referential” they. As documented by Conrod (2019); Camilliere et al. (2019), and Konnelly and Cowper (2020), referential singular they is both undergoing many changes in present day English and is subject to several interacting discourse, pragmatic and sociolinguistic factors (Bjorkman, 2017; Camilliere et al., 2019; Conrod, 2019; Konnelly and Cowper, 2020). We focus on bound variable singular they, which has been attested for centuries (Balhorn, 2004; Bjorkman, 2017) and is stably acceptable among English users (Conrod, 2019; Konnelly and Cowper, 2020). Bound variable pronouns are unlike referential pronouns in that their denotation is not fixed. While it (its in possessive form) in (2a) picks out the one dog introduced by the antecedent noun phrase, the meaning of it in (2b) varies, picking out each of the individual dogs described by the antecedent noun phrase.

In the linguistics literature, the use of it in (2b) is described as a “bound variable” pronoun. This is appreciated by formulae such as For every dog x, x likes x's owner, where the third variable x stands for it in (2b). The pronoun in this case is said to be “bound” by the quantified phrase every dog.

Singular they can likewise be understood either as referential (3a) or as a bound variable (3b).

That they is interpreted as singular in both sentences in (3) is illustrated by the plausibility of the thought attributed to each runner: that they themselves, not a plurality of runners, are the unique fastest runner. Previous studies (Camilliere et al., 2019; Conrod, 2019; Han and Moulton, in press) have shown that English users generally prefer they over alternatives such as he or she as bound variables, particularly when the antecedent phrase does not express gender, as with runner in (3b). The processing studies on bound singular they suggest that it also has an advantage over referential they, at least for the English speakers who participated in those studies. For instance, Foertsch and Gernsbacher (1997) found that they can resolve to singular quantified antecedents without apparent difficulty regardless of the gender of the antecedent. Ackerman (2018) compared sentences employing themself with gendered and non-gendered antecedents, finding a processing advantage using eye-tracking while reading for gender stereotyped quantified antecedents (indefinites like a mechanic) as compared to referential antecedents (stereotyped proper names).

What has not been asked yet is whether there is any evidence that the number properties of the antecedent in quantificational sentences have an effect on the acceptability and processing of bound variable they, as has been suggested in the case of referential singular they. There is some suggestive evidence that they do. Using reading times and acceptability ratings, Han and Moulton (in press) compared singular they and singular gendered pronouns in sentences containing both gendered and non-gendered quantified (4a) and referential antecedents (4b).

In reading time measures, Han and Moulton (in press) did not find any evidence of processing difficulty for bound singular they regardless of the gender of the quantified antecedent in the target region containing the critical pronoun, while they did find evidence of processing difficulty for referential singular they with gendered antecedents. However, in the spillover region, there was a main effect in which the they-sentences were read more slowly across the board, in comparison to the he-sentences. This was in contrast to the acceptability ratings that showed that bound singular they was rated as high as or even higher than he, while referential singular they was rated lower than he. Han and Moulton (in press) make the tentative suggestion that there may be a weak cost in initial processing when bound they/them finds only a singular antecedent.

No study reported to date, however, compares bound variable singular they with a plural vs. singular antecedent, and so cannot directly answer the question of whether bound variable singular they incurs a cost when taking a morphologically singular quantified antecedent like every. In the studies that follow, we make this comparison by capitalizing on configurations in English where a morphologically plural quantified noun phrase, headed by all, can bind they with a singular interpretation. This is shown in (5a) (Rullmann, 2003; Sudo, 2014). As with the bound variable sentence in (3b), repeated in (5b), this sentence attributes a pragmatically plausible thought to all the runners—that only one singular individual, they themselves, is the fastest runner.

Sentences such as (5a) provide a near minimal contrast with (5b), allowing us to hold constant bound they but vary the number on the quantified antecedent. All else being equal, if they more readily accepts a plural antecedent at least initially—as has been hypothesized for referential singular they by Sanford and Filik (2007) and Van Handel et al. (2021)—we should expect an advantage for singular they bound by all as compared to every.

While contrasting these two quantified expressions offers a novel way to examine the role of morphological number in processing bound variable they, the two quantifiers differ along a semantic dimension, namely the difference between collective and distributive interpretations. Every often resists a collective reading, while all generally allows one. With a predicate such as swarm, which requires a plural argument, a noun phrase quantified by every is not acceptable, whereas one quantified by all is (6) (Morgan, 1984; Winter, 2002).

The more distributive nature of every could make singular interpretations of they more available in comparison to all, on the assumption that it promotes an interpretation in which the predicate that corresponds to the nuclear scope of the quantifier holds of singular (atomic) individuals. This could have the effect of elevating the acceptability of a binding dependency between every and singularly-interpreted they while depressing that between all and they. As a consequence, this might obfuscate an advantage in number congruency between all and they². Support for this concern can be found in the existing experimental literature. Patson and Warren (2010) test the interpretation of indefinite expressions such as a box when presented as the direct object of a sentence with either a distributive (with each) or a collective (with together) subject. Their study shows that participants are more likely to interpret the indefinite singular as referring to multiple different objects mapped to covarying antecedents under the distributive subject vs. the collective one. This is direct evidence of a distributive subject invoking multiple individuals, which could each in turn serve as an ideal referent for singular they in a binding context, as noted above. In short, the salience of the multiple atomic individuals invoked by a quantifier like every or each may be promoting a preference for a bound singular reading to a degree that outmatches a preference for a plural reading of they when anteceded by a subject with all.

In light of these considerations, one goal of our studies was to investigate the role of distributivity in the availability of bound singular they. To do so, we additionally tested quantified phrases headed by each, which shows very strong tendencies toward distributivity (Vendler, 1967; Ioup, 1975; Tunstall, 1998).³ We reason that if distributivity makes singular bound variable they more available, we should find a cline in acceptability that mirrors the cline in distributivity from all to every to each. That is, in terms of acceptability judgments, we predict that singularly-interpreted they bound by each will be rated as more acceptable than that bound by every, which will be in turn rated as more acceptable than singular they bound by all. On the other hand, if congruence in superficial morphological number alone regulates the acceptability of bound variable they, then we predict sentences where all binds singularly-interpreted they will be rated higher than those involving each and every.

As for predictions concerning the incremental processing of sentences involving bound they under these different quantifiers, we contrast the predictions of the view that, for at least some speakers, they initially seeks plural antecedents and resists singular ones (Sanford and Filik, 2007; Chen et al., 2021; Van Handel et al., 2021) with those of an underspecification approach. Under the former view, we expect to find longer reading times for the pronoun when only a singular antecedent is available. (We use morphologically singular pronouns like he to test for a baseline reading time penalty for pronouns that can find only a number mismatched plural antecedent.) On the other hand, under the underspecification account, if at the earliest moments of processing, they is underspecified for number as suggested by Moxey et al. (2004) and Sanford et al. (2008), we do not expect any disruption in reading times at the pronoun in sentences where it can only find an antecedent headed by singular every/each in comparison to plural all. Under either account, however, if distributivity is a factor in incremental processing, then we expect some processing penalty for all in comparison to every/each.

We first report two acceptability rating studies in which we determine the acceptability of bound variable singular they under three types of quantifiers: plural collective (all), singular collective/distributive (every), and singular distributive (each). Crucially, we presented participants with sentences that forced a singular interpretation of they. One study tested sentences with non-gendered quantified antecedents and the other tested sentences with gendered quantified antecedents. In both studies, we observed a cline in acceptability of bound variable singular they such that each is better than every, while every is variably better than all. While distributivity does affect the acceptability of they, bound they under all is relatively acceptable, particularly in comparison to a non-quantified baseline. As for morphological number, we found no evidence of reduced acceptability for they taking a singular quantified antecedent.

We then turn to two self-paced reading studies to investigate whether there are any processing costs associated with they taking a singular as compared to a plural quantified antecedent. The two studies tested sentences with non-gendered antecedents and sentences with gendered antecedents, respectively. While the acceptability judgments suggest no such difference, incremental processing at the pronoun may be delayed if the pronoun's morphological number is not congruent with the only available antecedent, a cost that is potentially overcome by the time readers make considered acceptability judgments. Furthermore, given our acceptability judgments, distributivity could also be reflected in processing costs. As we report below, in the two self-paced reading studies, we found no evidence of a difference between plural and singular quantified antecedents for they. Similarly, we found no processing costs reflecting the distributive/collective distinction. Rather, for all types of quantified antecedents, encountering a singular pronoun (he) rather than they registered a processing delay, in contrast to non-quantified antecedents.

We suggest that the absence of number effects in both the online and offline measures is most compatible with the view that they is underspecified for number, and the presence of distributivity effects only in the acceptability ratings is related to a perhaps late construal mechanism that is required to fully semantically interpret singular-denoting pronouns under non-distributive quantifiers like all (Rullmann, 2003; Sudo, 2014). Furthermore, we take the processing effect associated with bound singular pronouns (he) to reflect a general fact about the English pronoun system, revealed here at the early moments of processing: morphologically singular pronouns resist bound interpretations, possibly because they is the language's preferred alternative.

Experiment 1 tested the acceptability of bound singular they in the context of non-gendered antecedent quantifier phrases headed by all, every, and each. As a baseline, we also tested the acceptability of singular they in sentences with definite plural antecedents. We predict that if bound singular they prefers a plural antecedent, all should be more acceptable than every and each. By contrast, if bound singular they prefers a distributive antecedent, each should be the most acceptable and all the least acceptable of the quantifiers under investigation. As we detail below, our stimuli were created in a very specific way to help ensure that readers interpreted they as a singular bound variable.

Experiment 2 examined the naturalness of singular they with gendered antecedents. Previous studies show that the gender of the antecedent can affect the acceptability of they (Doherty and Conklin, 2017; Ackerman, 2018; Ackerman et al., 2018; Prasad et al., 2018). At the same time, Han and Moulton (in press) found that the gender of the antecedent had no effect on the acceptability and processing of singular they bound by the quantifier every. We therefore set out to determine whether gender modulates the acceptability of singular they with the plural (non-distributive) quantifier all. This will help delimit the role of gender in regulating the acceptability of bound singular they. We expect that when they is bound by all, then sentences with gendered common nouns should exhibit the same pattern as the ones with non-gendered common nouns as found for every in Han and Moulton (in press).

Experiment 3 investigated the incremental processing of they in comparison to a superficially singular pronoun (he) in sentences with non-gendered quantified antecedents. Using a self-paced reading task, we examined the processing profile of they in the context of three types of quantified antecedents (all, every, and each), with the goal of addressing whether bound singular they incurs a processing cost with singular as compared to plural quantified antecedents, and additionally, whether the semantic distributivity of the antecedent quantifier has any effect. We intended the singular pronoun (he) to serve as a control because we expected it to exhibit processing difficulty in the context of all but not in the context of every or each.

Here, we follow studies like those in Foertsch and Gernsbacher (1997) and Sanford and Filik (2007) where we measure the reading time of pronouns that are given only one antecedent. Similarly, we will interpret longer reading times at the pronoun as showing that the processor has more difficulty integrating—or linking—the pronoun to the one available antecedent. Such a difficulty is expected if they initially accepts only a plural antecedent, and induces a cost when only a singular antecedent is available (Sanford and Filik, 2007). In that case, we expect slower reading times at they (or soon after) following each/every as compared to all. If they is underspecified for number in initial processing, we do not expect an impact of number in the form of longer reading times at the point of the pronoun.

We test these predictions using the stimuli from the acceptability rating study Experiment 1. As in the acceptability rating studies, we included material in our self-paced reading stimuli to enforce a bound singular interpretation of they. Recall that in those stimuli a bound singular interpretation is essentially forced by the predicate be the only one. While this does not force a bound interpretation in advance of encountering the pronoun (since it follows the pronoun), we sought to use this material as a probe to determine if readers made an early commitment to a plural interpretation of they. If readers simply pursued a plural interpretation of they—referentially anaphoric to the plurality introduced by, e.g., the all phrase—they should demonstrate difficulty at only one.

Experiment 4 used the same self-paced reading procedure as Experiment 3 with male gendered antecedents. This experiment compared definite plural antecedents for they against quantified antecedents (all, each). The change to gendered antecedents was made to remove the possible confound discussed in Experiment 3.

Our goal in this article was to investigate whether the acceptability and processing of bound singular they are sensitive to the morphological number and the semantic distributivity of the antecedent quantifier, by comparing quantifier phrases headed by all, every, and each. If they initially prefers to form a dependency with a morphologically plural antecedent (Sanford and Filik, 2007; Van Handel et al., 2021)—and if this extends to bound variable uses of they—then we can expect to find a penalty when the antecedent phrase is headed by morphologically singular every or each in comparison to plural all. On the other hand, if the distributivity of the antecedent quantifier plays a role, we should see a cline in acceptability and/or processing ease among the tested antecedent quantifiers that mirrors the cline in distributivity from all to every to each.

In the self-paced reading studies (Experiments 3, 4), regardless of the gender of the antecedent, we found no evidence that bound singular they prefers a plural antecedent. Bound singular they with the morphologically singular quantifiers every and each was read no slower than with the morphologically plural all. In fact, they exhibited faster reading times than he with all the tested quantifiers. These results also bear on suggestive evidence in Han and Moulton (in press) that bound variable singular they may give rise to a weak number mismatch effect with singular quantified antecedents. However, the number of the antecedent was not directly manipulated in those studies. In the present studies, where the antecedent number is manipulated, we do not find such an effect.

As for the role of distributivity, while the acceptability studies (Experiments 1, 2) showed that bound singular they was preferred under more distributive quantifiers, the reading time measures from our self-paced reading studies did not detect any effects of antecedent quantifier distributivity. We did not find any differences in the processing profile of they among the tested quantifiers. In Experiment 3, the singular region exhibited no processing difficulty for they in comparison to singular he for every and each, as well as for all. In Experiment 4, while the they-sentences showed no hint of difficulty for any of the tested antecedent types (all, each, and definite plural), the he-sentences showed some difficulty for all and each in comparison to the definite plural.

These results taken together suggest that they is underspecified for number, which can be predicated by only one in sentences with either morphologically singular or plural quantifier antecedents, and that they has processing advantages over he as a bound variable.

Turning to the acceptability rating studies (Experiments 1, 2), we found that bound singular they is highly acceptable, regardless of the gender of the antecedent noun. The each-sentences were the most acceptable, followed by the every-sentences and then the all-sentences. These results show that bound singular they is not more acceptable with a morphologically plural antecedent quantifier; on the contrary, it is more acceptable with a morphologically singular antecedent quantifier. This is consistent with the findings from the reading times that showed no evidence of bound singular they being specified to prefer plural antecedents. But unlike the reading time measures, the acceptability ratings showed an effect of distributivity: the most distributive quantifier, each, had the highest acceptability ratings.

The finding that singular they bound by a singular quantifier poses no more difficulty nor a reduction in acceptability compared to singularly-interpreted they bound by a plural quantifier comports well with the proposals in the theoretical (Kratzer, 2009; Bjorkman, 2017; Konnelly and Cowper, 2020) and processing literature (Sanford et al., 2008) that they can be underspecified for number and gender features. As an underspecified pronoun, therefore, they should not clash with either singular or plural quantified antecedents, whether gendered or not. In fact, according to our reading time measures, not only did they easily retrieve singular quantified antecedents, they exhibited more processing ease than he for all the tested quantifiers. Thus, our online data further affirm the finding in previous offline studies that English users prefer they over singularly gendered pronouns (such as he or she) as bound variables, even with gendered antecedents (Camilliere et al., 2019; Conrod, 2019; Han and Moulton, in press).

Our findings stand to some extent in contrast with those in the literature regarding referential singular they, which has been reported to exhibit processing difficulties with singular antecedents (Sanford and Filik, 2007; Van Handel et al., 2021). Han and Moulton (in press) suggest that bound variable dependencies favor underspecified pronouns, whereas referential dependencies are less likely to, at least for more conservative speakers. We must remain cautious, however, in making comparisons between bound and referential singular they, as the latter is undergoing changes in the language that may impact both acceptability and processing. For instance, researchers have documented acceptability and processing differences relating to participants' exposure to referential singular they in non-binary accepting environments (Conrod, 2019; Chen et al., 2021). Future research should directly compare bound and referential singular they with attention to speaker variation in the case of referential they.

The underspecification approach is further bolstered by the differences between the acceptability rating studies and the reading time results. The online reading time measures did not detect any distributivity effects in the processing of bound singular they, though the offline acceptability ratings did. In particular, we found a cline of acceptability such that the each condition was rated highest, followed in turn by the every, all, and def.plural conditions. As we noted above, the reduction in acceptability can be traced to the need to interpret the all-sentences and the definite plural sentences distributively. This outcome is expected since distributive readings are generally dispreferred to collective ones (Frazier et al., 1999; Dotlačil and Brasoveanu, 2021).

The role of anaphora with they adds some further nuance to this picture. Repeating (5a) from above, we begin the discussion here with an example paralleling cases discussed in Rullmann (2003):

To obtain the plausible singularly-interpreted bound variable reading of (17), most researchers posit a distributive operator DIST (Rullmann, 2003; Sudo, 2014). The question is whether a collective interpretation is at all possible for (17), and when (and if) readers accommodate a distributive interpretation.

A collective interpretation would force a referential interpretation of they—where (17) perhaps describes a scenario in which the runners compare themselves as a group to a set of walkers—and does not require this operator. If this interpretation is possible, then the need to invoke a DIST operator is not clear on a first pass parse of (17). Indeed, it may only be upon later consideration of the sentence, and perhaps with other contextual information highlighting the bound variable reading, that a processor would even posit the DIST operator.

In our target items, the use of the only one who explicitly marks the embedded clause as a predicate over individuals. As discussed above, this forces participants toward a singular interpretation of they, but differences in antecedent quantifier distributivity, most crucially a distinction between all and each, only emerged as significant in the offline rating of sentences. We propose that this indicates the positing of the DIST operator is part of a later stage of computing the truth conditions of a sentence as a whole, which impacted sentence acceptability ratings¹⁴.

With an antecedent containing the quantifier each, a salient marker of distributivity, participants are primed for a predicate over individuals, and need not posit a covert operator. For the def.plural condition, the covert operator is the most likely way for the sentence to be felicitously resolved to a coherent meaning. That this is the condition rated significantly lowest in both Experiment 1 and Experiment 2 is not surprising: it is the one condition where additional structure (the DIST operator) must be assumed in the higher clause after the embedded clause is fully interpreted.

In the online study, they does appear to be processed with more difficulty in the def.plural condition relative to he. However, we believe that this appears more likely to be a property of some advantage for he, possibly arising from participants having already accommodated an unheralded antecedent for the more specified pronoun, as discussed above. If there were some impact of distributivity on they in this condition, we would expect this to manifest in a penalty on they in the def.plural condition relative to the other antecedents, and this is not what we observed. That we have not observed this in online processing speaks again to the underspecified nature of they: participants are not fixing on a plural interpretation of they, even when one is readily available. While it is true that other studies (Dotlačil and Brasoveanu, 2021) have found online evidence of a processing penalty when an antecedent is forced from a collective reading to a distributive one, these studies were targeting only the interpretation of an ambiguous sentential subject and its relationship to a verbal predicate, not the interpretation of a pronoun bound by that subject.

Keeping in mind that offline studies collect sentence acceptability judgments, measures of the sentence considered as a whole, the cline in ratings we have observed may be reflective of the different degrees to which a silent operator is necessary to arrive at a coherent interpretation of the sentence once it is considered in full. In the each condition, at most, a “floating” operation is necessary in order to have the quantifier in the right position to limit the domain of the verb to singleton sets. Nothing additional needs to be posited, and not surprisingly, this is the condition rated highest. The middle conditions, every and all, show an initially surprising behavior, not even being significantly differentiated in Experiment 2. This may be a reflection of participants' intuitions that the two quantifiers have some qualities in common with each, yet neither is its perfect equivalent. The lower ratings may not reflect so much difficulty (as in the def.plural condition) but more of a Gricean Maxim of Manner reaction. Participants are aware that a less ambiguous option is available (i.e., each), and maybe rate these middle cases slightly lower for this reason.

In the studies reported in this article, we investigated whether the processing and acceptability of bound singular they are sensitive to the morphological number and the semantic distributivity of the antecedent quantifier. Based on the findings from our reading time data and our acceptability rating data, we argue that (i) they is underspecified for number; and (ii) the construal mechanism of distributivity is part of a later stage of comprehension involving the computation of the truth conditions of a sentence.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by Research Ethics Board, Simon Fraser University; Research Ethics Board, University of Toronto. The patients/participants provided their written informed consent to participate in this study.

C-hH and KM designed the experiments. C-hH, KM, TB, HG, DS, JW, and SW constructed the stimuli and wrote the manuscript. TB, HG, JW, and SW programmed the experiments and collected data. C-hH performed statistical analysis. All authors contributed to the article and approved the submitted version.

This study was supported in part by Social Sciences and Humanities Research Council (SSHRC) Insight Grant 435-2014-0161 and Academy of Korean studies Grant AKS-2016-LAB-2250004 to C-hH, as well as SSHRC Insight Grant 435-2018-1012 to KM.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.