We assume, with Bjorkman (2017), that in the English pronoun system at Stage 1, there are three privative, contrastive features: the gender features masculine [MASC] and feminine, [FEM], as well as the feature inanimate [INANIM]. It is not clear from Bjorkman’s discussion whether she considers [INANIM] to be a gender feature per se. As we shall see, both Bjorkman’s analysis and our own place [INANIM] on a different syntactic head from [MASC] and [FEM], suggesting that it may not belong to the same class of features.15 For the purposes of this paper, we will assume that [INANIM] is not properly a gender feature, and that in English, the semantics of both [MASC] and [FEM] include something like animacy as well as their particular gender specification. Each of these three features contrasts with the other two and with the absence of any gender/class feature. We also assume, with Bjorkman (2017), that singular [SG] is the marked number feature.

However, we follow Kramer (2009; 2015) in placing the gender features on the nominal categorizing head n. We assume that the φ head above Number, where Bjorkman places [MASC] and [FEM], is actually the locus of person features like [PARTICIPANT] and [AUTHOR].16 Third-person pronouns either have no marked features on φ, as we assume for English, or depending on the language, may lack person specification altogether (Déchaine & Wiltschko 2002; Béjar 2003).

We also assume that at this stage, English nouns like sister, policeman, and turnip, as well as names like Susan, Richard, and Thailand, carry the contrastive features [FEM], [MASC], and [INANIM], respectively. Further, we assume that when nouns like teacher, doctor, and person, and names like Kelly and Pat, are used to refer to individuals of known binary gender at this stage, the contrastive feature matching the individual’s binary gender is obligatorily inserted on the nominal head n.17

The third-person pronominal vocabulary items are given in (11). These are essentially the same as those proposed by Bjorkman (2017: 7), the only difference being the syntactic position of [FEM] and [MASC].

As Bjorkman notes, the features [MASC] and [FEM] behave differently from [INANIM] at Stage 1. Specifically, for pronouns bound by quantified antecedents, as in (12), [MASC] and [FEM] are realized only optionally.18 In contrast, as shown in (13), [INANIM] must be realized.

We propose the structure in (14) for Stage 1 third-person pronouns.20

We propose that for quantified antecedents like those in (12) and (13), there is a close relationship between the quantifier, which occupies a D_Q head above Number, and the Number head itself. We provisionally stipulate that the features of the number head are copied to D_Q, and thus become features of the quantified nominal as a whole, while features of the n head lower in the structure are only optionally copied to D_Q.21 The referential properties of a definite, non-quantified DP presumably require that the gender features of n, when present, are obligatorily copied to the D level.22 Under this view, if a quantified antecedent bears the feature [INANIM], that feature will always be accessible to the binding relation between the antecedent and the quantifier-bound pronoun, and will have to be spelled out, while the lower gender features on n may or may not be accessible.

At Stage 1, then, all three features [MASC], [FEM], AND [INANIM] are contrastive in the pronoun system. Many nouns, like actor, actress, mother, and brother, and proper names like Susan and David, also carry contrastive gender features, and other nouns, when used to refer to individuals of known gender, bear an obligatorily inserted contrastive gender feature. Except for pronouns bound by quantified antecedents, as described above, coreference requires that the features of the pronoun match those of its antecedent. Singular they in other contexts, whether A-bound by a non-quantified antecedent or not bound at all, is thus possible only when the gender of the antecedent is epicene or unknown, and the antecedent thus bears none of the three gender features; in all other cases, either he or she is required.

Let us turn now to Bjorkman’s Innovative variety, which we refer to as Stage 2. We claim that in fact, there is no change whatsoever in the status or position of gender features in the pronoun system between Stage 1 and Stage 2: all three gender features remain contrastive, and for nominal antecedents bearing gender features, Stage 1 and Stage 2 are identical. The difference between Stage 1 and Stage 2, we claim, lies in the way nouns are specified in each speaker’s lexicon, and in whether gender-nonspecific nouns referring to human beings are obligatorily assigned a binary gender feature. Recall that at Stage 1, nouns referring to people are, in any given sentence where the gender of the referent is known or assumed, systematically assigned either [MASC] or [FEM] corresponding to that referent. At Stage 2, however, the rule inserting a contrastive gender feature is optional: nouns like teacher and friend need not be assigned gender features even if the gender of the referent is known or assumed. Nonetheless, there remains at Stage 2 for any individual, a subset of nouns, like mother, father, sister, brother, actress, and policeman, as well as certain proper names, that lexically bear contrastive gender features; i.e. the gender feature is present on the noun in the speaker’s lexicon, not assigned based on a given referent.23 Since gender features are contrastive, a singular pronoun whose antecedent carries either [MASC] or [FEM] – either lexically or based on the known or assumed gender of the referent – must be spelled out with he or she. Individual speakers will differ, based on their own experience, as to which nouns, and which proper names, lexically bear contrastive gender features, and in how readily unfamiliar proper names are assigned a lexical gender specification. In addition, the possibility exists that a Stage 2 speaker may encounter an individual of non-binary gender whose pronoun of reference is they but whose name (e.g., Mary), carries a contrastive gender feature in that Stage 2 speaker’s lexicon.

A Stage 2 speaker has several options in this situation. First, they could adjust their lexicon by deleting the contrastive gender feature from that particular proper name. Mary would then join the class of given names like Kelly, Alex, and Morgan, which are not marked for gender, and (15) would be grammatical. This predicts, somewhat implausibly, that once a Stage 2 speaker has accommodated their lexicon as just described, any instance of Mary, regardless of the gender of the referent, should be able to serve as the antecedent to they. A second, more plausible, option would be to add a new lexical entry for Mary to their lexicon, homophonous to the existing entry, but with no marked gender feature. This new lexical item would be specified as denoting an individual whose pronoun of reference is known to be they, or might even be restricted to refer only to the particular non-binary individual in question. A sentence like (15) would thus remain ungrammatical, unless it referred specifically to the individual in question, or to some other individual with the relevant property. Essentially, this amounts to marking well-formed uses of (15) as exceptions.

A third option – which we would characterize as unhelpful (and even bigoted) but nonetheless a logical possibility – would be for the Stage 2 speaker to simply refuse to use the individual’s correct pronoun, on the grounds that to do so would be grammatically ill-formed. As can be confirmed by a cursory online search, this option is frequently chosen (even on occasion by some linguists), quite explicitly and publicly, by people with an agenda that has very little to do with grammar, and pertains more to the attempted exclusion of transgender and non-binary individuals.24

The final option would be to change the status of gender features altogether, making them optional modifiers25 <masc> and <fem> rather than contrastive features whose absence carries meaning, moving the speaker fully to Stage 3. At this stage, which is to our knowledge the last stop on this trajectory, they can be used to refer to any non-inanimate singular individual, whether or not the antecedent is linguistically present, and whether or not the antecedent is a nominal bearing semantic gender features. He and she are also available, spelling out <masc> and <fem> respectively, and may indeed be required in certain situations (as when the speaker knows that the referent’s pronouns are he, or she), but their use is always grammatically optional, since the gender features they bear are grammatically non-contrastive. This stage aligns most closely with our own judgements, and with the judgements of our consultants.

An important point to be made here is that this fourth option is also available to a Stage 1 speaker faced with data like (16), repeated from (6).

We suspect that many Stage 3 speakers, and in particular those who have greater familiarity with non-binary individuals (e.g., those who have non-binary acquaintances, friends, or family members; see Ackerman et al. 2018) may actually never have passed through Stage 2 at all, having instead simply changed the status of the features [MASC] and [FEM] to optional adjuncts <masc> and <fem> in the first instance. This move is essentially the grammatical manifestation of an understanding that so-called “natural gender” is not a binary property.

For those speakers who have instead made case-by-case changes to the gender specifications of nouns in their lexicon and are thus at Stage 2, it would seem that shifting into Stage 3 involves making a separate conceptual leap away from the cultural assumption that all persons can be categorized according to a gender binary, a leap that requires engaging with wider socio-cultural changes regarding gender and gender identity. But no matter which path is taken, at Stage 3, the morphosyntactic and sociocultural changes dovetail.

However, for a Stage 2 speaker, faced with a single situation that their grammar does not accommodate, the second option described above – creating homophonous entries for a single form like Mary, one with a gender feature and the other without – has the smallest effect on the grammar as a whole, and this is plausibly the option that a Stage 2 speaker would choose in such a situation. We speculate that this is where many speakers of English are currently situated: effectively marking exceptional cases to accommodate non-binary individuals of their acquaintance, but without making overall structural changes to the status of gender in their grammars. In the long run, however, given the increased visibility of individuals of non-binary gender, this approach leads in principle to a less efficient lexicon, rife with homophonous pairs of lexical entries, one member of each pair carrying specific idiosyncratic pragmatic constraints. At the same time, these same exceptional cases, having expanded the distribution of singular they, reduce the empirical pressure for a systemic change in the status of gender features in the pronoun system.

Using a realizational model of morphology, Bjorkman argues that in the Conservative variety (our Stage 1), there is a contrastive three-way gender distinction, with marked masculine (MASC) and feminine (FEM) contrasting both with each other and with an unmarked neuter. Other features in the system are singular (SG) and INANIMATE. Third-person pronouns are realized as in (17) (Bjorkman 2017: 7).

All of the features shown in (17) are contrastive, and marked on the syntactic heads shown in (18), adapted by Bjorkman from work by Steriopolo & Wiltschko (2010) and Déchaine & Wiltschko (2015) (Bjorkman 2017: 8).

For Bjorkman, quantifier-bound pronouns (also called bound variables) like those in (19) optionally realize either φP or NumP. If the pronoun realizes φP, then if the quantified antecedent is animate and either masculine or feminine, either he or she will be spelled out, as in (19a) and (19b). Bjorkman assumes that if neither [MASC] nor [FEM] is present, an unspecified φP can receive an epicene interpretation, as required in (19c). However, since a bound variable can also be a bare NumP, it can lack gender specification entirely, even if its antecedent is singular and of known gender, as in (19e) and (19f).26 The pronouns in (19c) and (19d) are thus in principle ambiguous between a φP with no gender specification on φ and a bare NumP, while the pronouns in (19e) and (19f), whose antecedent is lexically marked with [MASC], can only be a bare NumP.

At Stage 1, then, English has a fully contrastive, ternary gender system, both for us and for Bjorkman. A singular pronoun whose antecedent is marked as MASC or FEM must be realized as he/him/his or she/her, respectively.27 Epicene they arises with singular animate referents of unknown or non-unique gender, realizing a φP bearing neither MASC nor FEM. The only exception is with quantifier-bound pronouns, which can, for Bjorkman, consist of a bare NumP with no gender specification.

While Bjorkman’s account, by our estimation, correctly describes the distribution of pronouns at Stage 1, the stipulation that quantifier-bound pronouns optionally spell out either φP or NumP is theoretically worrisome.28 Ideally, it should be possible to identify some other properties (we know of none) that fall out from the syntactic category of the pronoun in these instances; if there are indeed no such properties, then the account is ad hoc. Empirically, this stipulation corresponds to our proposal that in quantified DPs, the features of the Number head are obligatorily copied to D_Q, while the features of n are copied only optionally. Both approaches require further theoretical and empirical support, a question we leave for future work.

In Bjorkman’s Innovative variety (our Stage 2), the contexts in which singular they appears include a wider range of animate antecedents of known gender, as in (20) repeated from (3) above.

Crucially, in each of these sentences, the gender of the antecedent is known to the speaker, and may be binary. In the Conservative variety, the epicene interpretation of singular they would not be available here, and the pronoun would obligatorily realize contrastive MASC or FEM, as applicable.29

At the same time, the gender features seem not to have not lost their contrastive force entirely, since Bjorkman points out that sentences like (21) remain ill-formed for Stage 2 speakers.

Faced with this apparently inconsistent behaviour of the features [MASC] and [FEM], Bjorkman proposes that for speakers who accept the sentences in (20), but reject those in (21), the two gender features in the pronoun system are no longer contrastive, but are optional, interpretable, adjunct features in Wiltschko’s (2008) sense. The absence of both <masc> and <fem> on φ thus does not imply that the referent of φP is neither masculine nor feminine, and thus does not restrict such a φP to epicene contexts.

This account straightforwardly predicts that sentences like those in (20) are well-formed, but does not explain why those in (21) are unacceptable. If the gender features are truly optional adjuncts, they should pattern as other optional semantic content does: omitting them should never cause ill-formedness.30 On the assumption that the nouns man and mother carry the features <masc> and <fem> respectively, while the nouns person and parent carry no gender features, the choice to use they instead of he or she in (21) might be expected to have the same status as using person or parent instead of man or mother in (22).

Given Bjorkman’s analysis, then, the account of the ill-formedness of the sentences in (21) will not be found in the narrow syntax. Bjorkman proposes that it follows instead from a dynamic condition on feature matching in discourse, which requires that “referential pronouns can only be interpreted as referring to a previously-introduced referent if they bear a superset of the features that have already been associated with that referent in a discourse. In other words, a pronoun can add to the linguistic features associated with a referent, but it cannot underspecify them.” (Bjorkman 2017: 11). The effect of this proposal is essentially to smuggle contrastive gender features back into the system. By definition, adjunct features cannot be syntactically obligatory. If Stage 2 speakers do not have grammatical, contrastive gender features in their pronoun systems, then the choice of gender-marked pronouns like he and she cannot be enforced by the grammar.

It must be noted that this requirement holds only of the featural structure being spelled out, and not of the vocabulary items themselves, since Stage 2 speakers have no trouble using the underspecified vocabulary item they to refer to a plural, gendered antecedent like women, as in (23).

In (23), the effect of Bjorkman’s proposed dynamic requirement here is that the φP spelled out by they must bear the feature <fem>. But since there is no vocabulary item in the system that spells out <fem> without also spelling out the contrastive feature [SG], they is inserted.

There is a problem with the requirement, however. As Bjorkman (2017: 11) describes it, its effect is that “a pronoun can add to the linguistic features associated with a referent, but it cannot underspecify them.” In other words, the featural representation that a pronoun spells out cannot be less specific than that of its antecedent, but it can be more specific. Consider the sentence in (24).

Assume, with Bjorkman, that singular is marked relative to plural in the pronoun system, and that women bears a gender feature <fem>. Featurally speaking, the women is specified as <fem>, but bears no number feature. Suppose that the node spelled out by the pronoun bears both <fem> and [SINGULAR]. The pronoun will therefore be spelled out as she. In accordance with Bjorkman’s requirement, the pronoun bears a superset of the features of its antecedent, wrongly predicting that (24) should be acceptable.

Another question about the proposed dynamic condition on feature matching is its status in the theory. Is it a (presumably universal) property of how coreference is resolved in discourse, or is it instead a (necessarily learnable) property that one language may have and another may lack? Bjorkman does not discuss the question, but on the assumption that the condition lies outside the grammar proper, we will assume that it is intended to be universal.31 One could imagine a revised version of Bjorkman’s dynamic requirement, such that its effect is limited to only adjunct features. Under such a view, it would be instructive to look at number marking on nominals and pronouns in a language like Halkomelem, where number has been shown to be an optional adjunct feature (Wiltschko 2008), but where both singular and plural pronouns are available (Wiltschko 2002). If Bjorkman is correct, the presence of plural marking on a noun should force any pronoun coreferential with it to take plural form, while a plural-referring but unmarked noun should be able to serve as the antecedent of a plural-referring but unmarked pronoun. Without a condition such as this one, one might expect there to be no particular connection between the presence of plural marking on a pronoun and its presence or absence on a coreferential pronoun.

However, one need not go so far afield as Halkomelem number marking to test whether a universal condition of this sort can be maintained. It should, in fact, be applicable in the English of Stage 3 speakers, as long as (a) these speakers still have the pronouns he and she in their system, spelling out optional adjunct gender features, and (b) these speakers also have, in their lexicon, some nouns (like stewardess, policewoman, actress, chairman, etc.) that can only felicitously refer to individuals of known binary gender, with some notable exceptions (see Ackerman 2019 for felicitous examples based on situational context, as in At the Halloween party, I saw that the stewardess_i left their_i hat in the kitchen, but I don’t know who_i it was). As we will see in the next section, this prediction is not borne out. Furthermore, though Bjorkman describes the judgements we have reported as our Stage 2 as the Innovative pronoun system, we claim that this is not in fact the most innovative variety in current use.

It is at Stage 3 that the differences between our account and Bjorkman’s come fully into focus. Recall that at Stage 3, they is completely unrestricted in referring to third-person, singular, animate antecedents. In particular, it can be used in sentences like (25), where the antecedent is a proper name with strong gender association (25a) or a semantically gendered noun (25b).32

For us, the difference between Stage 2 and Stage 3 is clear: while at Stage 2 the masculine and feminine gender features were still fully contrastive, at Stage 3 they have lost their contrastive status and have become completely optional modifier features.

This straightforward account of the difference between Stage 2 and Stage 3 is not available to Bjorkman’s analysis, since there, even at Stage 2 the features are claimed to be optional and non-contrastive. Nor, we assume, can it plausibly be that Stage 3 speakers have lost the dynamic condition on feature marking in discourse, since that condition is not parametric. Nor can it be that gender features are entirely absent from the Stage 3 pronoun system, since even at Stage 3 he and she are optionally available. The difference can therefore only be that words like grandmother and Sir, and proper names like Paul, have lost their gender feature specification altogether. However, even for Stage 3 speakers, nouns like grandmother and chairman are still strongly gendered, though they can serve as antecedents to singular they when the occasion calls for it (such as in the case of an individual who uses either of the labels grandmother or chairman, and uses the pronoun of reference they). We conclude from this that Bjorkman’s account of what we call Stage 1 and Stage 2 cannot easily be extended to account for the Stage 3 system, and that the account proposed here is to be preferred.