Singular they and the syntactic representation of gender in English

Singular they enjoys a curious notoriety in popular discussions of English grammar. Despite this, and though its use with quantificational, non-specific, and genuinely epicene antecedents dates back at least to the 1400s (Balhorn 2004), it has been little discussed in formal linguistics. This squib suggests an analysis of this longstanding use of they, while also describing a more recent change in they’s distribution, whereby many speakers now accept it with singular, definite, and specific antecedents of known binary gender. I argue that the distribution of they, in both conservative and innovative varieties, has implications for our understanding of the syntactic representation of gender in English, the structure of bound variable pronouns, and the regulation of coreference.


Introduction
This paper addresses general issues of pronominal binding and coreference, though its empirical focus is comparatively narrow, being mainly concerned with the distribution of so-called singular they in Modern English, and the evidence that distribution provides for the syntactic representation of gender features in the language.
For a functional element, singular they enjoys a curious notoriety, particularly among non-linguists. Though sometimes discussed as though it were a twentieth century innovation, introduced due to feminist backlash against epicene he, in fact it has been used in contexts like those in (1), with quantificational, non-specific, and genuinely epicene antecedents, going back at least to the 1400s (Balhorn 2004). Bodine (1975) notes that prescriptive edicts against singular they, by contrast, date back only to the 1700s, and never successfully eradicated singular they from standard English usage. Theoretical work on the English pronominal system has largely overlooked the use of they in (1), though it has been described in sociolinguistic and some psycholinguistic studies. The focus of this paper, however, is not on this longstanding use of they, but instead on a more recent change in they's distribution, which appears to have been overlooked even in descriptive work. This change is visible in contexts like those in (2), where some speakers Glossa general linguistics a journal of Bjorkman, Bronwyn M. 2017. Singular they and the syntactic representation of gender in English. Glossa: a journal of general linguistics 2(1): 80. 1-13, DOI: https://doi.org/10.5334/gjgl.374 accept they even with an antecedent that is singular, definite, and specific, referring to an individual whose binary gender is known to both speaker and hearer. 1 (2) a. %The professor i said they i cancelled the exam. b. %Our eldest child i broke their i leg. c. %I'll let my cousin introduce themselves. 1 Not only has this extension of they gone unreported, it also presents an additional puzzle, in that even speakers who accept singular definite specific they as in (2) generally do not accept sentences like those in (3), where the antecedent remains singular, definite, and specific, but is now a proper given name or gender-specific noun. 2 (3) a. *Janet i said they i cancelled the exam. b. *Thomas i broke their i leg. c. *I'll let my sister i /father i /aunt i introduce themselves i .
The contrast in acceptability between (2) and (3) has been made more striking by increased cultural visibility of nonbinary individuals-individuals who identify with neither masculine nor feminine gender, and so who cannot be referred to with either the singular masculine pronoun he or the singular feminine pronoun she. Many nonbinary individuals prefer singular they as a pronoun of reference, and they is sometimes said to have the advantage of being already part of English grammar, in contrast to fully innovative alternatives. 3 It is thus surprising that even for innovative they users, the current status of they in English is apparently insufficient to render sentences like those in (3a-b) automatically acceptable, even for speakers who accept sentences like those in (2), and even if Janet and Thomas are known to name nonbinary individuals.
Internalized cultural assumptions that all humans can be sorted into binary gender categories likely form part of the obstacle to adopting they as a singular pronoun of reference-but the contrast between (2) and (3) reveals that there may also be a grammatical obstacle as well. This paper is not intended to argue against adopting they as a singular pronoun of reference for nonbinary individuals, but instead to clarify the grammatical status of they among contemporary English speakers, and so to clarify what changes are involved in extending they further to examples like (3a-b).
From a theoretical perspective, this paper argues that the distribution of they has implications for our understanding of both bound variable pronouns, and the role of gender in regulating pronominal coreference. In the first case, I argue that we need a syntactic account in which bound variables (optionally) instantiate fewer feature distinctions than their full referential counterparts. The distinction among pronoun types proposed by Déchaine & Wiltschko (2002) offers the right kind of distinction in this respect, while the widely adopted mechanism of Feature Transmission (Heim 2008) used in much semantic work to account for features on bound variables cannot account for the attested distribution of singular they.
1 From this point forward in the paper, I consistently use the reflexive form themselves even as the non-plural reflexive of they, though it should be noted that some speakers prefer themself. There is also independent variation with theirself/selves. 2 For at least some speakers, they is possible with given names if the referent is of unknown gender, or of known nonbinary gender. But because the empirical focus of this paper is on the contrast between (2) and (3), and to avoid a proliferation of marks to indicate different distributions of grammaticality, in the remainder of this paper examples like those in (3) are marked as ungrammatical. 3 A number of gender non-specific pronouns have been formally proposed in various sources, including e/em/ eir (Spivak 1990), hu/hum/hus, ze/zir/zir, ze/hir/hir, thon/thon/thons, and many others.
Accounting for the innovative use of they in (2), by contrast, requires an account of the status of gender features in English, and the role of these features in regulating pronominal coreference. I suggest that for innovative they users, gender has ceased to be a contrastive feature on pronouns, instead becoming a fully optional semantic feature. But even for these speakers, a subset of English nouns-proper names and gender-specific nouns-remain syntactically specified for gender features, and mechanisms of coreference resolution in discourse prevent they from referring back to any antecedent that was previously specified for gender.
The paper begins in sections 2-3 by briefly reviewing previous work on the distribution of singular they in English, and by describing in more detail the innovative use of they with singular definite specific antecedents. Sections 4-5 then turn to details of the proposed analysis.

A brief history of they
As noted above, they has been possible with quantificational, indefinite, and epicene antecedents throughout the Modern English period. 4 The examples in (4) Though prescriptive objections to singular they may have reduced its use in much formal writing, especially in more linguistically conservative American sources, singular they has remained well attested in speech and less formal writing through to the present day (Bodine 1975;Matossian 1997;Baranowski 2002). Studies as far back as Bate (1978) have shown that singular they is used even by speakers who self-report that they do not use it at all, and a long record of psychological studies has shown that English speakers do not in fact interpret generic he as gender-neutral (Martyna 1978;MacKay & Fulkerson 1979;Gastil 1990;Foertsch & Gernsbacher 1997;Miller & James 2009). Bodine (1975) provides the following examples of naturally-occurring singular they "collected by the author from the ordinary conversation of native speakers of American English holding bachelors, masters, and doctoral degrees." (139): (5) a. Did everyone say they missed you like mad yesterday? b. Somebody left their sweater. c. Not one single child raised their hand. d. When you call on a student, it's better if you can remember their name.
Haegeman (1981)-the sole generative paper to comment on this use-gives a number of examples similar to those in (5), and directly states: "It is important to add that this use of they/their is restricted to generic nominals which are unmarked for sex: no instances of such a neutral singular they are to be found with (i) referentially specific nominals (my teacher); (ii) sex-specific nominals (sister, mother, etc.)." (Haegeman 1981: 237) The examples from Shakespeare and Austen above already show that the second of these restrictions is too strong, as we see bound variable they with nominals like man or sister. We will return to the first restriction below, in discussing the innovative use of singular they. Recent studies of pronoun use have continued to find an association between they and quantificational, indefinite, and epicene antecedents, even when not using this semantic terminology. Newman (1992), analyzing a corpus of spontaneous speech from nine television shows, suggests that singular they prefers "nonsolid" antecedents, where "solidity" is intended to combine referentiality, specificity, and definiteness. Newman states this as a preference rather than an absolute requirement, but none of the examples of potentially solid antecedents in the paper-all of which appear in (6)-involve truly definite and specific singular antecedents for they. In (6a) the antecedent one is nonspecific (though contextually referring only to women married to men); in (6b) it is a free relative whoever is chosen; in (6c) it is an nonspecific partitive one of them (in context, one of a set of male doctors).
(6) a. If one is signing and thinking that their husband is representing that there isn't any other relationships, there isn't any problems there, and they do that as part of the consideration for the agreement, then, of course, you can test it and determine and develop the information that shows that that wasn't so. And that they were being misrepresented. b. I'm sure that whoever is chosen, probably their viewpoints will be at least reasonably known on those issues. c. Just before the program, we looked up one of them, and they're not in there. Matossian (1997), perhaps the most comprehensive study of the colloquial use of singular they to date, adopts Newman's distinction of the "solidity" of antecedents in her study of epicene pronouns in sociolinguistic interviews conducted in Philadelphia and Minneapolis. Again, though Matossian describes the association of they with nonsolid antecedents as a preference, and does not directly comment on whether they ever appears in her data with a specific definite antecedent of known binary gender, all of the examples given in the dissertation of potentially solid antecedents appear to be either indefinite or nonspecific. A selection of Matossian's examples of singular they appear in (7), including the single example given of a singular definite antecedent for they (7a) (where the antecedent is nonspecific the other).
(7) Matossian (1997: 164-166) a. Usually, if one [spouse] is Catholic, as long as the other's Catholic, whichever nationality they are doesn't matter, as long as they're Catholic. I mean if they're a good Catholic, you know. b. [Q: What would happen if a girl asked a boy to dance?] They-some people would dance with them, but, like, others wouldn't. [Q: How come?] It depends on the girl-who the girl is. c. I consider [childbirth] the most important event of my life. An' it's somethin' I'll never forget, an' I'm glad I had the opportunity to have a child an' to see everything. I can't imagine anyone just being knocked out an' wake up an' have the baby in their arms. I think they miss so much.
Throughout the modern English period, then, it appears that they has been possible as a bound variable (with quantificational or indefinite antecedents), and with genuinely epicene antecedents, those of indeterminate, mixed, or unknown gender (as in the case of non-specific definites like the ideal student). 5 Finally, note that we cannot account for this use of they-what we might call the conservative distribution of singular they in English-simply by proposing that gender or number features are optionally absent in these contexts, because they is restricted to singular animate antecedents. When the referent of a pronoun is inanimate, the pronoun it is obligatory. 67 (8) a. Every book lost its/*their cover. b. I like my local grocery store. I recommend it/*them to you. 7 c. (Seeing a distant object, which cannot be identified:) I think it's/*they're a tree.
Interestingly, this restriction to animates remains true even for speakers for whom they is possible with specific definite singular antecedents of known binary gender, what we might call the innovative distribution of singular they, to which we turn in the next section.

Innovative they with specific definite antecedents
Though previous descriptions have reported that they is not possible-or is strongly dispreferred-with singular definite antecedents, or as a referential pronoun when the gender of the antecedent is known to both the speaker and hearer, we already saw in the introduction that for some contemporary speakers they is fully grammatical in these contexts.
(9) a. %I really love their costume. (e.g. while pointing someone out) b. %My friend left their sweater here. c. %Your research assistant said they'll be joining the call later.
The judgements reported in this paper on the innovative use of they are those of the author, confirmed in consultation with other native English speakers from the US and Canada, and via observation of the use of they in speech and writing over several years. Approximately 20 speakers have been consulted in detail regarding the judgements overall, and though not all of them accept the use of they illustrated in (9), all sentences marked with '%' have been systematically accepted by a subset of speakers consulted. While firm generalizations are not possible based on this small sample, it appears that younger speakers more often accept the innovative use of they, while older speakers are more likely to find it ungrammatical or pragmatically odd. Confirmation of this age-grading, or further conclusions regarding possible regional or social factors, must await a larger scale study.
If we were to inspect only the examples in (9), we might hypothesize that for innovative speakers gender specification has become optional even on pronouns, the sole context where it remains morphologically reflected. As a consequence, the distinction between singular she/he and plural they would also be lost, bringing the third person in line with second person number-neutral you.
But this simple explanation could not explain the fact that there are speakers who produce examples like those in (9), and who judge them to be fully grammatical, but who neither produce nor judge grammatical examples like those in (10) (assuming that Janet and Thomas name individuals of known binary gender). 8 (10) a. *I just saw the lead actress i , and I really love their i costume. b. *Janet i left their i sweater here. c. *Thomas i said they i 'll be joining the call later.
The examples in (9) and in (10) (9) were grammatical due to the loss of gender features in English, then we would expect they to be possible with all singular definite antecedents, including those in (10). The contrast between (9) and (10) also cannot be attributed to the increased "solidity" of proper names, i.e. their referential specificity and definiteness when compared with other nouns, because innovative speakers do generally accept they with last names introduced by non-gender-specific titles. Some innovative they users also report that they improves with names that have both masculine and feminine associations, like Chris or Alex, with internet handles, even those that are gender-specific (but are nonetheless not traditional names). 9 (11) a. %I just saw Chris i , and I really love their i costume. b. %Professor Smith i left their i sweater here. c. %moongirl17 i said they i 'll be joining the chat later.
These names are as referentially "solid" as the names in (10); they differ in lacking the traditional gender associations of English given names. I suggest below in section 5 that though they is a pronoun not specified for gender (or for number), its extension into the space of definite specific pronouns reveals the vestiges of syntactically-represented gender in English, as well as the role of gender features in regulating pronominal coreference.

Accounting for bound variable and epicene they
I argue in this section that the conservative distribution of singular they reflects a contrastive three-way distinction among gender features ([masc] v. [fem] v. ∅), with its quantificational and indefinite uses arising because bound variable pronouns correspond to a smaller structure than do their referential counterparts (following e.g. Déchaine & Wiltschko 2002, 2015. I assume a realizational view of the syntax-morphology interface, in which potentially underspecified morphological items compete to realize a fully specified syntactic representation. In this type of system, they emerges as the elsewhere realization for English 8 The judgements in (10b-c) become more complex if the names refer to individuals of known nonbinary gender-though it is striking that even many innovative they speakers have difficulty extending they to nonbinary individuals, and go through a period of making errors in pronoun use despite efforts to avoid them. This suggests that using they in contexts like those in (10b-c) requires genuine grammatical adjustment even for innovative they users. 9 Though speakers do report a contrast between (10) and (11), these judgements are somewhat more variable than others reported in this paper. It is also possible that when presented as abstract examples, speakers treat the examples in (11) as involving referents of unknown gender. These potential confounds make this a potentially fruitful area for future large-scale studies of singular they in naturally-occurring discourse.
pronouns, occurring in the absence of any Ф-features that would trigger insertion of another pronoun. They cannot be specified for number, as it occurs in both singular and plural contexts. 10 It cannot be specified for gender, because it is not gender-specific. And it cannot be specified for animacy, as it occurs as both the animate and inanimate plural. 11 The realization rules for third-person pronouns in English are as in (12), restricting ourselves to nominative forms for the sake of simplicity: 12 ). That feature is either present or absent in the syntax, but its absence results in a positive semantic interpretation through the mechanism of contrast. A language like French, with a classic two-way contrast in gender, is naturally accounted for in these terms: feminine gender corresponds syntactically to a positively specified feature ([fem]), while masculine is not [-fem] but instead corresponds to the absence of a gender feature, and thus masculine is also the default gender used in cases of gender mismatch and in epicene contexts. We have already seen, however, that neither he nor she can be used in English in a gender-neutral or epicene context. This can be explained if neither masculine nor feminine is unspecified (featurally "unmarked") in English, and if the features [masc] and [fem] are both contrastive: because conservative varieties require either he or she for singular definite specific antecedents of known gender, the absence of [masc] on a pronoun must be interpreted as ¬masc, and the absence of [fem] as ¬fem.
What if a singular pronoun bore neither of these gender features? Contrast would result in such a pronoun being interpreted as ¬masc&¬fem (i.e. neither masculine nor feminine). This interpretation conflicts with the strong pragmatic assumption that animate individuals (especially humans) can be exhaustively categorized as either masculine or feminine. A singular pronoun with no gender features thus would not be interpreted as referring to an individual who is neither masculine nor feminine, but instead with an inference that the gender of the referent is unknown, nonspecific, or otherwise indeterminate. 13 Here contrastive ternary gender ([masc] v. [fem] v. ∅) accounts for epicene uses 10 Note that even singular uses of they fail to trigger third person singular agreement forms on the finite verb (is, was, has, or -s), instead triggering the default "plural" forms (are, were, -∅). 11 Whether they is also underspecified for person depends on the representation of third person. While there is a long tradition arguing that third person corresponds to the absence of any more specific person features ( of they in the same way that contrastive binary gender (most often [fem] vs. ∅) accounts for epicene uses of masculine singular pronouns in languages such as French. The above account does not yet address the use of they with quantificational antecedents: if gender is contrastive, we might expect bound variable pronouns to be obligatorily marked for gender, at least when their domain is restricted to either masculine or feminine entities. And yet, contrary to this expectation, we have already seen that even in its conservative distribution, bound variable they occurs with gender-specific antecedents, in the examples from Shakespeare and Austen in (4).
The use of they with antecedents like no man or both sisters suggests that gender features are (perhaps optionally) absent on bound variables. Rather than stipulating this optionality, however, we can explain it in terms of a structural difference between bound variables and referential pronouns. Déchaine & Wiltschko (2002) propose that pronouns fall into at least three classes, distinguished by their semantics and morphosyntax, and further that these three classes correspond to different sizes of syntactic projection: DP pronouns are interpreted as definite referring expressions, ФP pronouns as variables, and NP pronouns as predicates. Déchaine & Wiltschko (2015) argue directly that English pronouns can realize either DP structures (in their referential use) or ФP structures (as bound variables). As the name suggests, ФP pronouns include a projection associated with Ф-features, but lack a higher DP layer. The distribution of they as a variable can be accounted for with a minimally richer structure than Déchaine & Wiltschko propose, with one additional head in the ФP domain. Following work such as Steriopolo & Wiltschko (2010), I assume here that gender on English pronouns is located high in the nominal spine-that it corresponds to what they call discourse gender, i.e. gender that is determined by properties of the discourse referent. 14 While for Steriopolo & Wiltschko discourse gender is located on D, I suggest that in English it occurs lower, within the ФP domain though nonetheless higher than the heads associated with number (Num) and animacy (n 0 ), neither of which is ever optional on bound variables. 15 (13) 11 between bound variables and referential pronouns. Déchaine & Wiltschko (2002) propose that pronouns fall into at least three classes, distinguished by their semantics and morphosyntax, and further that these three classes correspond to different sizes of syntactic projection: DP pronouns are interpreted as definite referring expressions, ΦP pronouns as variables, and NP pronouns as predicates. Déchaine & Wiltschko (2015) argue directly that English pronouns can realize either DP structures (in their referential use) or ΦP structures (as bound variables). As the name suggests, ΦP pronouns include a projection associated with Φ-features, but lack a higher DP layer. The distribution of they as a variable can be accounted for with a minimally richer structure than Déchaine & Wiltschko propose, with one additional head in the ΦP domain. Following work such as Steriopolo & Wiltschko (2010), I assume here that gender on English pronouns is located high in the nominal spine-that it corresponds to what they call discourse gender, i.e. gender that is determined by properties of the discourse referent. 14 While for Steriopolo & Wiltschko discourse gender is located on D, I suggest that in English it occurs lower, within the ΦP domain though nonetheless higher than the heads associated with number (Num) and animacy (n 0 ), neither of which is ever optional on bound variables. 15 ) are located on a higher head than either number or animacy. Bound variable This is higher than the position typically assumed for grammatical or semantic gender, most commonly associated with the root itself or with n 0 . See Kramer (2016) for an overview of arguments in favour of a low position for grammatical gender, across a number of languages. To the extent that some speakers disprefer he or she as bound variables, it might be possible to locate gender on English pronouns on D. This would eliminate the need for a projection ΦP in (13), which treads quite close to a dedicated Gender Phrase. Such a projection has been proposed by authors such as Picallo (1991), but is argued against by Kramer (2015;2016), among others. arbitrary grammatical gender features, a cultural or pragmatic assumptions that all persons are either male or female would eliminate the previously available category of animate neuters, leaving genuinely epicene antecedents to instead be realized by an elsewhere form. 14 This is higher than the position typically assumed for grammatical or semantic gender, most commonly associated with the root itself or with n 0 . See Kramer (2016) for an overview of arguments in favour of a low position for grammatical gender, across a number of languages. 15 To the extent that some speakers disprefer he or she as bound variables, it might be possible to locate gender on English pronouns on D. This would eliminate the need for a projection ФP in (13), which treads quite close to a dedicated Gender Phrase. Such a projection has been proposed by authors such as Picallo (1991), but is argued against by Kramer (2015Kramer ( , 2016, among others. The crucial assumption here is that gender features in English ([masc] and [fem]) are located on a higher head than either number or animacy. Bound variable pronouns can correspond either to a full ФP, in which case they will be specified for gender, number, and animacy, or to NumP alone, in which case they will lack any gender specification. I assume that referential pronouns always correspond to the full DP structure, and so are obligatorily specified for gender (except in the above-noted cases, where contrastive non-specification of gender gives rise to an epicene interpretation). This departs from Déchaine & Wiltschko (2002), who assume that any given pronoun in a language will reliably correspond to DP or ФP or NP. The structural variability of individual pronouns is predicted on a realizational approach to morphology, however, and so represents a modest extension of their proposal.
This proposal departs significantly from the standard treatment of features on bound variables in the literature, via the mechanism of Feature Transmission proposed by Irene Heim (detailed, for example, in Heim 2008). Heim notes that person and number features appear to be formally uninterpreted on bound pronouns, on the sloppy identity reading in examples such as (14): (14) Only I love my mother. (Nobody else loves their mother.) Feature Transmission addresses this issue by proposing that the Ф-features of bound variables (e.g. [1sg] features on my) are morphosyntactically copied from the binder to the variable, but totally absent on the LF branch, and thus without semantic effect. Feature Transmission has a number of drawbacks as a morphosyntactic mechanism. In particular, instances of variable binding across finite clause boundaries would require feature copying in violation of syntactic locality conditions-but it also faces a serious empirical challenge in the distribution of bound variable they. To account for the fact that they alternates with both he and she as a bound variable in English, we might be tempted to suggest that Feature Transmission is simply optional, and that in its absence we find the elsewhere pronoun they. What this would not account for is the fact that it does not alternate with they-and neither does my in examples like (14). For this reason, Feature Transmission cannot account for the distribution of they as a bound variable-and also cannot account for its occurrence as a referential pronoun, either with epicene antecedents or in the innovative distribution.

Accounting for singular specific definite they
While the use of they as a bound variable and epicene pronoun is long standing, its extension to specific definite contexts appears to be a recent innovation. Any account of the grammar of innovative they users must explain how they can refer to specific individuals of known binary gender, while nonetheless also explaining why it remains impossible when its antecedent has been introduced by a gender-specific noun or proper name.
Consider again the examples in (15), repeated from (9) but now showing the variation between they and he or she. In these examples the referent of they is animate, singular, both definite and specific, and of known binary gender. For speakers with the conservative distribution of they, for whom they is uniformly unacceptable in (15), gender is grammatically obligatory on referential pronouns, in the sense that it must be expressed whenever the gender of the referent is known.
For speakers with the innovative distribution of they, by contrast, specifying gender on referential pronouns is possible (and often pragmatically preferred) but evidently grammatically optional. Why then do innovative they speakers pattern with conservative speakers in requiring he or she in examples like those in (10), repeated in (16)?
(16) a. *I just saw the lead actress i , and I really love their i costume. b. *Janet i left their i sweater here. c. *Thomas i said they i 'll be joining the call later.
The simple fact that innovative speakers can use they for referents of known binary gender, interchangeably with he or she, is sufficient to demonstrate that for these speakers gender is no longer contrastive on pronouns: for such speakers the absence of gender does not convey that gender is unknown or indeterminate (though it is compatible with those interpretations). Following Wiltschko (2008), I refer to this type of semantically interpretable yet optionally specified feature as an adjunct feature, notated as <f>. For speakers with the innovative distribution of they, then, the contrastive features [masc] and [fem] have been replaced (at least on pronouns) by the optional and non-contrastive features <masc> and <fem>. The realization rules for pronouns remain the same, however, so that he and she surface whenever these optional features are present. The non-contrastive status of gender features does not explain why they cannot refer back to a singular gender-specific noun or proper name in (16), however. To account for this, I suggest that though English lacks arbitrary grammatical gender, nouns like actress and given names like Janet and Thomas are nonetheless syntactically associated with gender features. The restriction of nouns like actress to feminine reference is thus not a purely semantic fact, but instead reflects the presence of a syntactic gender feature. 16 Nouns associated with gender features include gender-specific nouns, including kinship terms (boy, girl, man, woman, nephew, neice, father, mother, etc.), some names of professions (actress, waitress, stewardess, etc.), and proper given names. 17 Associating these nouns with gender features is not in itself enough to account for their incompatibility with they, however: we might expect that because they is underspecified, it could freely occur as a default pronoun regardless of the features of its antecedent. To rule out examples like those in (16), we need to ensure that pronouns are never specified for fewer features than their antecedents: if a pronoun refers back to an individual introduced by a noun like actress, which bears a feminine gender feature, the pronoun 16 For conservative speakers, these would remain the contrastive features [fem] and [masc], but for innovative speakers we can ask whether gender on nouns and names, like gender on pronouns, has become an adjunct feature. In favour of this possibility, we have already seen that nouns like friend can be referred back to by singular they without any implication that the gender of the friend is unknown. This suggests that for innovative speakers gender cannot be contrastive on gender-non-specific nouns like friend, and so given the logic of contrast it is not expected to be contrastive on any other nouns. Simply because a feature is not contrastive, however, does not mean that it is not obligatory on some nouns. Kramer (2015) implements arbitrary lexical gender in terms of selectional requirements: in the case of a noun like actress the suffix -ess itself would realize a <fem> n 0 head; other gender-specific nominals, like mother or nephew, may be selected by a similar (but unpronounced) n 0 head, or may realize a root that is directly specified for a gender feature (though this requires that formal features can occur on roots). This is a different structural position for gender than was assumed above for pronouns; similarly, Steriopolo & Wiltschko (2010) locate the the gender of nouns such as actress or mother, which they refer to as as semantic gender (determined by properties of the nominal itself), on the root itself, while the discourse gender features of pronouns (determined by properties of the discourse referent) occur higher. 17 I list feminine profession terms because it is not clear whether the non-feminine variants of these nounse.g. actor-imply a masculine referent. In some cases the previously-masculine term has been extended to cover all members of a profession, but in other cases terms like steward do seem to be restricted to men. It is interesting that in some domains a novel gender-neutral term has been adopted (e.g. server in place of waiter/waitress) perhaps suggesting that the previous "neutral" (i.e. non-feminine) term had a masculine association for many speakers. must similarly bear a feminine feature, and be realized as she. This feature matching requirement cannot be attributed to a syntactic mechanism such as Agree, moreover, because the inability of they to refer back to gender-specific antecedents also holds across clauses, as in (16a). Hence the second part of the explanation, which is that coreference (as opposed to binding) is subject to a dynamic condition on feature matching: 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.
Consider the case where a referent is introduced with a noun phrase like my friend, as in (15b). For innovative they users, gender features are syntactically optional, and so the new discourse referent established by my friend has no gender information associated with it. It can therefore be referred back to by a pronoun that similarly lacks any gender feature (they)-or by he or she, either of which would then introduce gender information (and so prevent subsequent reversion to gender-non-specific they).
By contrast, if a referent is introduced by a gender specific noun like actress, or by a proper name such as Janet, then it is immediately associated with the gender feature borne by those nominals, here <fem>. Here subsequent reference to the same entity would require the pronoun she, even for innovative they users: they is excluded because it does not maximally match the formal features previously associated with the referent in the discourse.
For conservative they users, by contrast, all referential nouns must be associated with gender features (when the gender of the referent is known), because for these speakers gender features remain contrastive, even when they are not morphologically expressed. For these speakers they will always fail to maximally match the features of any specific referent of known gender, and so will never be felicitously used in examples like (15).

Coda: they as a nonbinary pronoun of reference
The goal of this paper has been to draw attention to the complexities of singular they in modern English, both in its long-standing use as a bound variable and epicene pronoun, and in its apparent recent extension to use with definite and specific antecedents.
Returning to the difficulty that many speakers have in accommodating they as a pronoun of reference for nonbinary individuals, I am proposing that this arises not only from pragmatic or cultural assumptions about the binarity of gender, but also from a grammatical property of given names in English. What is necessary in order to adopt they as a singular pronoun of reference, on this account, is to unlearn the generalization that given names are uniformly syntactically associated with gender features.