Abolishing Gender on D

Abstract

This article presents a Minimalist syntactic analysis of sociopragmatically conditioned gender features on pronouns. To account for inter- and intra-speaker variation, I locate the parameter for social gender in the presence or absence of an unvalued gender feature on the phase head D. Supporting this analysis, I show that variation in English speakers’ acceptability and use of definite, specific singular they, as in (i), is sensitive to reference; this sensitivity is robustly explained by the location of gender features on D.

(i) Taylor_i is writing their_i own autobiography.

For speakers who report (i) as ungrammatical, a crash results from the uGender feature on D remaining unvalued. For innovative speakers, uGender is not present on D and no crash results from a lack of gender features. This analysis explains why a pragmatic feature like social gender can cause true syntactic ungrammaticality, since the narrow syntax encodes certain pragmatic features as obligatory.

Résumé

Cet article présente une analyse syntaxique minimaliste des traits de genre pronominaux conditionnés sociopragmatiquement. Pour tenir compte de la variation inter- et intra-locuteurs, je localise le paramètre de genre social dans la présence versus l'absence d'un trait de genre non valué sur la tête de phase D. Je soutiens cette analyse en montrant que la variation dans l'acceptabilité et l'utilisation par les anglophones du they singulier défini et spécifique, p. ex. (i), est sensible à la référence; cette sensibilité s'explique de manière robuste par la localisation des traits de genre sur D.

(i) Taylor_i is writing their_i own autobiography. ‘Taylor_i rédige sa_i propre autobiographie.’

Pour les locuteurs qui signalent (i) comme agrammatical, l’échec résulte du fait que le trait uGenre sur D reste sans valeur. Pour les locuteurs innovants, uGenre n'est pas présent sur D et aucun échec ne résulte de l'absence de traits de genre. Cette analyse explique pourquoi un trait pragmatique comme le genre social peut provoquer une agrammaticalité véritablement syntaxique, puisque la syntaxe étroite encode certains traits pragmatiques comme étant obligatoires.

1. Introduction

This article provides a syntactic account for an ongoing change in the third person pronominal system, in which singular they – which has been previously restricted to indefinite or generic antecedents – has expanded for many speakers to allow use with specific, definite antecedents, including proper names.¹ The first section reviews the evidence for this change and elaborates the different types of singular they. Section 2 outlines the syntactic proposal, which consists of two components: first, a head-raising analysis of pronouns; and second, a proposed microparameter that differentiates speakers who can and cannot use singular they for specific antecedents. Section 3 shows the consequences of this proposal, including confirmation of a previously-proposed typology of pronouns. Section 4 briefly compares this proposal with related alternative proposals, and section 5 concludes.

1.1 Singular they

Variation in the grammaticality of singular they with different antecedents has been informally reported for some time, but linguistic analyses have typically been in psycholinguistic studies (Moulton et al. 1978, Hughes and Casey 1986, Hyde 1984, among others) which deal with potential processing cost, ambiguity, or the alternation between generic singular they and the generic use of he. More recent work on they has directly addressed the grammaticality of they with a singular, definite, specific antecedent. Bjorkman (2017) observed that speakers report different grammaticality/acceptability judgments for definite, specific uses of they in the singular, as in (1). Bjorkman differentiated between these specific, definite uses and the (otherwise well-established) generic or indefinite uses, as in (2).² See Curzan (2003) for a thorough review.

1. (1) ? That syntax professor_i loves their_i job.

2. (2) A syntax professor_i must always love their_i job.

Acceptance of singular they depends on context and specificity. Speakers may reject (1) but accept uses like (3), even though both antecedents are definite. Uses like (3)–(4) are well-attested and have been in use for much longer than those like (1). Curzan (2003) reviews the history of generic and indefinite uses like these, as well as the history of prescriptive grammarian pushback against those uses, and their alternation with generic uses of he.

1. (3) The ideal student$_i$ never forgets their$_i$ homework.

2. (4) Every student$_i$ should do their$_i$ homework.

What Bjorkman (2017) crucially observed is that the judgments for singular they become even more variable (and less generally accepted) when anteceded not just by a specific antecedent, but in particular by a proper name. Proper names seem to be the newest possible antecedent for singular they. I group proper names like (5) and definite specific antecedents like (1) together as definite specific singular they, or dsT.

1. (5) % Jayden${}_{1}$ loves their${}_{1}$ job.

In a large-scale grammaticality judgment survey targeting they with indefinites, generics, and proper names, Conrod (2018) found that speaker age correlated with ratings of singular they only when used with proper names; their data support an apparent-time analysis in which this variable is currently undergoing a change in real time (Conrod 2019). The goal of this article is therefore to propose a syntactic account of a single parameter that differentiates speakers who have dsT as part of their grammar of English from those who do not, while accounting for particular restrictions on the use of dsT and other English pronoun phenomena.

2. Proposal: uGender parameter on referential D

I propose that the synchronic variation found in English speakers who rate dsT as grammatical or ungrammatical is due to the presence or absence of a uGender feature on the D head in the pronominal DP complex. Innovative speakers who allow dsT lack the uGender feature on D, while conservative speakers maintain the uGender feature on D. The presence of the feature on D requires valuation for referential pronouns in particular (as distinct from other types, such as bound anaphors).

The core of this proposal is that for non-dsT speakers, what makes dsT (but not other types of singular they) ungrammatical is gender, not number. This is corroborated by metalinguistic comments from various speakers who lack dsT (e.g., Pullum 2003), and explains why the grammaticality variation is around specific referents, not all singular antecedents.

As a direct result of their participation in this grammatical change, this also implies that for dsT speakers, gender is always optional. This goes further to explain sociopragmatic variation where gender features can be omitted or included to adhere to different Gricean maxims (Grice 1968).

In this section I show how this proposal can be operationalized; I am working within the Minimalist framework (Chomsky 1995, 2000) and assume the Borer Conjecture (Chomsky 1995, Borer 2014) for the purposes of the microparameter which I am proposing here. Head movement will be modeled after Matushansky (2006). I assume a Y-model grammar in which the narrow syntax first operates (Merge and Agree being the only available operations), after which syntactic structures are sent to the Conceptual-Intensional (CI) and Sensorimotor (SM) interfaces. I assume that Vocabulary Insertion happens either at or after the split into the interfaces.³

This proposal relies on an updated formulation of the three-part pronoun typology proposed by Déchaine and Wiltschko (2002). While the pronominal typology I maintain here is very similar to Déchaine and Wiltschko's, I assume that all pronoun types are full DPs, whereas Déchaine and Wiltschko assume that different pronouns represent different levels of projection, constituting nPs, $\phi$Ps, and DPs. In order to reflect the same syntactic differences between more and less noun-like pronouns observed by Déchaine and Wiltschko, I instead analyze pronouns as starting lower in the nominal domain (at little n) and optionally raising to intermediate or higher positions in the DP structure.

I review the three-way split of pronoun types in section 3 below. Here I primarily focus on identifying the microparameter that differentiates dsT-users from non-dsT users (who reject definite specific singular they). In (6) below, non-dsT users are speakers whose grammars maintain an uninterpretable uGender feature on referential D (circled). This feature, when unvalued by the point of spell-out, causes a crash that results in non-dsT speakers reporting ungrammaticality only when singular they is specific and referential. This is an important distinction, because very few speakers overall find singular they truly ungrammatical in its non-referential (generic or indefinite) uses. The ongoing grammatical innovation, therefore, is by dsT speakers who have no issue hearing or producing singular they with a specific referent – these speakers have the grammar represented in (7) below, where the referential D head crucially lacks a uGender feature. Thus, even when they raises to D without any gender features, no crash is caused.

1. (6) Referential dsT for non-dsT speakers

   

2. (7) Referential dsT for dsT-speakers

   

The numeral notation D$_5$ in (6)–(7) denotes specific referential determiner; this is a direct reference to an entity in discourse context, rather than an operator picking a set of entities out of a context. (The numerals are arbitrary; I use them for convenience of differentiable indexation.)

An important feature of this proposal is that the “problem” causing ungrammaticality for non-dsT users is not number, but gender; it is also important to note that loss of the uGender feature is facilitated by ambiguous uses of they, which I discuss further in section 3.3. This follows the proposal from Bjorkman (2017), also adopted by Konnelly and Cowper (2020), that singular they lacks gender (and number) features altogether. It also correctly predicts that metalinguistic comments about dsT will target gender, not number, which seems generally borne out in comments presented by Conrod (2019).

The structural difference between dsT in (7) above and an epicene singular they as in (3) is based on different syntactic/semantic types of determiner, compared in (8) below.

1. (8)

In (8), $\iota$ signifies a definite determiner that quantifies over a set, which I take as distinct from a specific referential determiner like D$_5$. Sentences like (3) can be produced by non-dsT speakers, even if the lexical antecedent contains some gendered meanings; I discuss this further in section 3.1.⁴

I take the $\iota$ operator to be equivalent to a definite determiner like English the; as such, it is an $\iota$ operator that appears in full DPs such as ‘the teacher’. The direct referential determiner, notated as D$_5$ in (8b), is distinct from the definite determiner; the only full DPs D$_5$ would appear in are proper names. Chapter 2 of Conrod (2019) discusses the relationship between proper names and pronouns more extensively.

In the next section I will briefly summarize how gender features of referential pronouns are evaluated for appropriateness. This is relevant to the proposal primarily because it diverges from most other work on how gender-matching is determined between pronouns and their antecedents (e.g., Kučerová 2018, Sigurðsson 2019, among many others).

2.1 How gender is evaluated

An important aspect of this proposal is that the gender of pronouns is evaluated in different ways for different types of pronouns (the types are detailed further in section 3). For this section, I will differentiate between Condition B (free) pronouns, which I refer to as referential pronouns, and locally-bound variables, including Condition A anaphors, which I refer to as variable pronouns.

Referential pronouns in this proposal are constructed from a specific, referential determiner that is directly indexed to a referent, denoted by a numeral subscript: D$_5$ in (6) and (7) is a determiner of this type. These determiners enter the derivation already indexed; their index is based on the speaker's intent to refer to a particular entity. If a pronoun successfully raises to D (see section 3 for more on head-raising), whatever features that pronoun carries are evaluated for sociopragmatic appropriateness with respect to the referent in the discourse context. One possible formalization of this by Conrod (2019) relies on the use-conditional semantics detailed by Gutzmann and McCready (2014); that denotation is repeated in (9).

1. (9)

   1. a. pronouns $ = x + \phi$ (Gutzmann and McCready 2014: 63)

      she5 ≈

   2. b. Paraphrase: an entity $x$ such that $x$ is the entity indexed as 5 AND such that it is appropriate to refer to 5 as she

Crucially, computation of pronoun–antecedent matching is not a matter of syntax, but rather a matter of sociopragmatics in this model. As far as the syntax proper is concerned, if the uGender feature is present, it simply needs some value, and if the uGender feature is absent, pronouns do not need gender. I will discuss further the implications of this proposal in section 4.3, where I also show why this is an advantage over previous proposals that attempt to determine gender-matching in the syntax.

Variable pronouns are different: they require gender to be checked for matching or appropriateness based on an operator that locally binds a D$_x$ determiner. This difference explains why non-dsT speakers easily tolerate what would otherwise be a gender mismatch when the antecedent is indefinite or quantificational, rather than referential.

In section 3, I show some of the important direct consequences of this proposal, and further delineate the types of pronouns and different empirical predictions made by each structure.

3. Details and consequences

The proposal targeting uGender on D predicts that different types of pronouns (e.g., referential vs. variable) will have different restrictions for necessity of feature valuation. This builds on previous accounts of pronominal typology which separate pronouns into more N-like or more D-like, in order to robustly explain why uGender must be valued on referential pronouns, but not other types.

Under this analysis, all types of pronouns consist of a full nominal structure, including n (a low functional head, following, e.g., Borer 2005), Num (a number head, along the lines of Ritter 1992), and D (a determiner head, following Abney 1987). The only element that pronouns lack is a lexical root below n; they are purely functional in this analysis.⁵ In what follows, when I refer to lexical nouns I am referring to non-pronominal nouns that have a root node under n – and by contrast, pronouns refer to any nominal structure lacking that root node.

I specify that the uGender feature on D is uninterpretable (and the iGender feature on n is interpretable) for several reasons. First, I take uninterpretable features to be features that are not (yet) valued; strong uninterpretable features may act as probes for Agree, and must be valued before spell-out. The manifestation of gender-related morphology on determiners is frequently controlled completely by gender features on nouns (though see King 2016 for some interesting exceptions). This suggests that determiners should not have inherent gender that is pre-determined, but that it should be valued over the course of the derivation. Conversely, in languages that have robust systems of grammatical gender, it is clear that lexical nouns have idiosyncratic gender features already present in the lexicon, and these are the features that may provide value to other nominal elements in their local context. In addition, I take gender features on n as interpretable (and their counterparts on D as uninterpretable) because predicative pronouns (which remain in a ‘low’ noun-like position) maintain a semantically meaningful sense of gender, for instance, ‘the other she’. In those cases, it is clear that the low nominal use of the pronoun still contributes meaningful gendered information, which would not be explicable if the low nominal gender features were uninterpretable. This system is a departure from Kramer (2016) and others, but is the most advantageous for explaining the facts here, as well as those described in Conrod (2019) and elsewhere.⁶ For my purposes, uGender should in general be able to be valued by either Agree or movement – but I argue that in English pronouns, only movement values it (and the uGender feature remains only on referential pronouns, not variable or predicative or even definite ones).

The three-way typology I use here is similar to that of Déchaine and Wiltschko (2002) : the ‘smallest’ pronouns are the most noun-like, ‘intermediate’ pronouns can act either as predicates or as entities, and ‘full’ pronouns replace entire DPs. In my translation of this three-way separation, I will refer to predicative pronouns (which can co-occur with nominal modifiers, including external determiners); variable pronouns (which include anaphors and bound variables); and referential pronouns, which refer directly to an entity in the discourse context (with or without a linguistic antecedent). The main difference between this analysis and Déchaine and Wiltschko's is that I assume that all pronouns are full DPs, rather than various nominal projections. This more robustly explains the facts of English that I discuss here, in two ways: first, Déchaine and Wiltschko (2002) cannot explain predicative pronouns like the one in (10c) and described in chapter 2 of Conrod (2019). Second, unlike some of the differences shown by Déchaine and Wiltschko (2002), all three types of pronouns in English shown in (10) can occupy all positions in a clausal structure: they can all act as arguments, for example, and do not show obvious morphological differences. Thus, it is an advantage for the present analysis to assume that they are all full DPs, and that their differences are primarily internal.

The structure proposed in Conrod (2019) constructs this three-way division using head-raising, rather than constituents of different sizes. In (10), I give examples of each type, showing the various head-raising options.

1. (10)

(Conrod 2019: 232) I follow Matushansky's (2006) formulation of head-raising, which, when applied to the fully raised n in (10a) above, proceeds as follows:

Chapter 5 of Conrod (2019) discusses the details of the head-raising mechanism in significantly more detail; for my current purposes, any analysis of cyclic head-to-head raising should suffice.

In addition to the difference in head-raising, each type of pronoun is selected by a different (null or overt) D head; the referential D head in (10a) is directly indexed to a discourse entity. The variable D head in (10b) represents a variable that must be bound by an operator, and lacks the uGender feature present on the referential D head. Predicative pronouns appear with overt, garden-variety determiners like articles (10c).⁷ For the purposes of this investigation, I will focus primarily on the differences between variable pronouns and referential pronouns in the structures in (10) above, since these are the structures that underlie generic singular they and referential (i.e., definite, specific) singular they (dsT), respectively.

3.1 Variable vs. Referential structures of they

Because D$_x$ lacks uGender in this model, it is predicted that variable pronouns will not cause a crash/ungrammaticality when there are no gender features on the pronoun to value uGender, as is the case with they. Variable pronouns in the three-way system proposed in Conrod (2019) include pronouns anteceded by indefinites or quantificational DPs, such as (12) below:

1. (12) Every syntactician$_i$ loves their$_i$ own analysis the most.

It is important to note that this analysis also predicts that (singular) they should be possible and grammatical with quantificational antecedents even when the antecedents appear to have clear gendered denotation. This is also borne out:

1. (13) By some miracle, a woman$_i$ can feed a baby with their$_i$ body.⁸

The example in (13) is not only attested – it is quite common. Zimman (p.c.) has found instances of singular they used with gendered antecedents as part of ongoing real-time corpus work, which suggests that the variable pronoun itself does not need to value an unvalued gender feature in order for the derivation to converge. Data like these (and many presented by Konnelly and Cowper 2020) also counter proposals (e.g., Bjorkman 2017, Kučerová 2018, and Sigurðsson 2019) that gender features are valued by grammatical features on antecedents, particularly in the case of contemporary English, which otherwise shows essentially no sign of grammatical gender features on lexical nouns. I apply this further to English lexical nouns with apparent morphological gender, such as ‘stewardess’ – the same laxity of coreference is observed by Ackerman (2019).

As Ackerman (2019) discusses, the anomaly observed in the use of the pronoun his in (14a) is ameliorated by context in (14b); the same effect can be observed for words like ‘actress’, ‘bachelorette’, and other words containing the -ess and -ette suffixes, among others. I take this as evidence that the remaining so-called gender effects of these nouns are purely social and pragmatic, rather than instantiations of a forced matching effect of grammatical features between antecedents and pronouns.

As further evidence that lexical nouns lack gender features – in my proposal regarding singular they, dsT speakers are predicted to accept singular they for any lexical antecedent, no matter what gendered semantics exist – including proper names. However, even for non-dsT speakers, this insensitivity should also persist for any antecedents that are coindexed with variable (not referential) antecedents, even for very gendered antecedents. The example above in (13) is one such example, but variable singular they can even coexist with (gendered) proper names – provided that the pronoun is still variable, not referential.

Since variable pronouns must co-occur with an operator to bind them, quantifiers and disjunction should license variable they, even when the members of the set being quantified over are all of the same gender. In (15), several speakers I consulted have shown a preference for their over her, even when they generally agree that woman denotes a particular gender of person. In (16), consultants also preferred their, even though the antecedent contains two proper feminine names. In fact, when given sufficient context – informants were asked to construct this sentence using names of people they knew, so it could be established that the gender identities of both referents were agreed upon by consensus. Even so, with two universally-agreed-upon names of mutually-known female referents, the disjunction licensed their as perfectly acceptable in this construction.⁹

1. (15) Any woman$_i$ who wants to combat sexism should examine ?her$_i$/their$_i$ own internalized prejudices.

2. (16) Either Barbara or Alicia always assigns their/?her own papers (though I don't remember which)

In the analysis I am putting forward, gender feature matching is allowed but not required by the binding antecedents, in contrast to the examples in (17) and those that follow. If it were the case that lexical (non-pronominal) antecedents had grammatical gender features that controlled pronominal features, the data in (13)–(16) would be quite impossible to explain. For this reason, I generally discard the notion of grammatical gender features on (any) lexical nouns in contemporary English, again following Konnelly and Cowper (2020). Again, I am proposing that iGender persists (optionally) on pronominal n heads, but is not lexically specified by root nouns. However, in languages other than English with robust grammatical gender systems, gender features on n in lexical nouns would only be able to combine with noun roots with lexically specified gender.¹⁰

However, this does call into question why any speaker of English would retain a uGender feature at all, even if only on a particular referential-pronominal D head. In fact, retention of the unvalued Gender feature (uGender) on referential D, and its interpretable counterpart (iGender on $n$, which I have thus far largely glossed over) predicts that pronouns, when coindexed with other pronouns, should have significantly stricter matching requirements than pronouns anteceded by lexical nouns in English. And this does indeed turn out to be the case.

While lexical antecedents generally don't clash with pronominal gender (“My son $_i$ ... she $_i$” is very uncommon), they allow a non-match with ungendered pronouns (“Any woman $_i$ ... they $_i$”). Pronoun to pronoun matching has much stricter conditions, particularly on locally-bound pronouns such as reflexives. Even for dsT speakers, singular they cannot be in a local binding relation with a different pronoun:

1. (17) * He$_i$ likes themself$_i$

2. (18) * They$_i$ like himself$_i$

Locality is important, because locally bound anaphors in this proposal constitute variable pronouns (not referential pronouns). This suggests that bound and predicative pronouns may retain gender features, but based on data like (15) it is apparent that those features are not obligatory. This also implies that not only is there a matching requirement on bound reflexives, as in (17), but predicative pronouns may also be able to locally bind and force gender matching effects.¹¹

1. (19)

   1. a. Every she loves her mother.

   2. b. * Every she loves his mother.

   3. c. ? Every she loves their mother.

Because the forced matching between pronouns with gender features appears to apply primarily to locally bound pronouns or anaphors, it is therefore also predicted that pronoun-to-pronoun matching requirements will be laxer between two or more free (Condition B) pronouns. In fact, switching between pronouns over the course of a conversation is well-attested in sociolinguistic interviews:

1. (20) RRA: His partner at the time was also dating this other person that was in our group. Um, and they have a very, um, he's a very strong and kind of controlling personality, and so he had kind of taken over like the whole thing, […] Ha. Yes. it kind of, that was kind of one of those things where it just- and that same person, I would see them more often than I would see [RRB] and they were trying to like convince me of these like negative things [...]

What this suggests is that, first, variable pronouns and referential pronouns are indeed constructed differently in the syntax; and second, variable pronouns are dependent on their antecedents for feature valuation in some way, but this only manifests as ungrammaticality when the antecedent is also a pronoun. This is further evidence that pronouns retain iGender in contemporary English but that lexical nouns do not, even when their denotation is directly gendered in some way.

The important takeaway from these data is that the rules for matching featural gender are sensitive to binding, and that gender features do exist on pronouns (but are not obligatory on pronouns for all speakers).

3.2 Aside on predicative pronouns and they

I have thus far not discussed predicative pronouns in this typology, in large part because under the model from Conrod (2019), predicative pronouns don't act like pronouns at all – they behave more like nouns. Their denotation, however, is closely linked with the social meaning of pronouns, which is itself interrelated with the presence or absence of dsT in a grammar. Conrod identifies predicative pronouns in English as including pronominal relative clauses (which I leave aside for the purposes of this article), and depronominalizations, as in (21).

1. (21) I've never met a they before.

The denotation given in Conrod (2019) for depronominalizations (in which a pronoun is modified by an external determiner) can be paraphrased as “the kind of person who is called they.” In order for this type of depronominalization to appear as a they (rather than as the much more common a she or a he), the speaker must have a legible category of kinds of people who get called they.¹² Thus, the microparameter that I have proposed in this article does not predict that dsT speakers and non-dsT speakers will have differing judgments about the grammaticality of (21), but rather that non-dsT speakers will lack a legible referent for the denotation given. Anecdotally, this seems to be essentially correct: when presented with (21), non-dsT speakers will volunteer comments instead about non-predicative uses of they (as reported by Pullum 2003, among other language commentators).

I will largely leave aside predicative pronouns for the remainder of this article, except to comment that my proposed microparameter predicts infelicity, not ungrammaticality, of a they for non-dsT speakers – and that this is an advantage of the proposal generally.

3.3 Ambiguity enables extension

As I alluded to briefly in section 2, part of the proposal's strength for explaining variation in English is that many instances of the pronoun they are ambiguous, and thus provide a possible route for acquisition of the grammatical change that has been underway. Furthermore, desiderata of explanation for singular they generally includes why an apparent grammatical difference doesn't cause constant interspeaker conflict.

While singular they is a hot topic among non-linguists, lay metalinguistic commentary around dsT doesn't suggest that this is causing significant grammatical conflicts except in very particular circumstances (such as non-discrimination protections of nonbinary students).¹³ What this implies is that there must be sufficient overlap in possible structures of utterances of singular they in ambiguous contexts to enable subtle extension and eventual microparameter resetting.

One important factor that occasionally masks grammatical conflict is that conservative non-dsT speakers will accept singular they in most contexts. Sociolinguistic data from Conrod (2019) show that singular they is not judged significantly worse when anteceded by a quantificational or indefinite DP, and in fact they was judged higher than other singular pronouns (he, she) in these contexts.

1. (22) Any person who wants to succeed ought to try their best.

Crucially, non-dsT speakers also accept definite, epicene singular they when it is not referential. It is for this reason that I have proposed that the microparameter is based on referential D, not on definiteness. Epicene definite NPs can contextually be quantifier-like, even though definite determiners aren't generally considered to be quantificational. The example in (23) is an instance of a definite antecedent (my math teacher) which, due to world knowledge, may potentially refer to a set of possible entities rather than a specific entity; in this instance, such a reading is supported by a context where the speaker has different math teachers at different times (as, for example, when they advance from algebra to calculus). In this context, there is a readily-available construal that allows an epicene reading of they that doesn't refer to a specific individual.

A second reading of (23) is also supported, however, and this ambiguity requires only a context change, not a change in the utterance. If it is the case that discourse context allows a hearer to identify a particular math teacher bounded within, for example, the present semester, then a definite and specific referent whose gender is unknown may be the referent.

1. (23) Utterance: My math teacher always gives me a lower grade for doodling, they are so unfair!

   No-dsT meaning: No matter what math teacher I get, that math teacher grades me down for doodling – they (epicene) are unfair!

   Possible-dsT meaning: This quarter my math teacher (whom you've never met, and don't know the gender of) graded me down every time I doodled – they (specific) are unfair!

My math teacher/they in (23) is ambiguous between a non-dsT meaning (definite epicene) and a possibly-dsT meaning, which is referential but pragmatically conditioned. This ambiguity can be captured in the syntactic/semantic structure with underlyingly different determiners – a true definite article in the non-dsT reading (as in Elbourne 2013), but a referential index with somewhat underspecified context in the dsT reading.

This distinction does contradict the proposal by Elbourne 2013 that all pronouns are themselves built upon definite determiners. However, based on the varying judgments by dsT and non-dsT speakers regarding definite vs truly-referential antecedents, this distinction is well-supported for the purposes of this proposal. Thus, the inventory of D heads which can contribute to the composition of pronouns must include, at least, a D$_{def}$ that is distinct from a directly referential D$_5$ (where the numeral is an index to a discourse entity).

Furthermore, these overlapping and potentially ambiguous readings of (23) provide a possible explanation for why we do not see more conflict around singular they between speakers whose grammar includes dsT and those whose grammar does not. The overlapping ambiguous uses may also have given rise to the reanalysis that enabled resetting the microparameter to begin with, which would be necessary for both the synchronic variation and evidence for ongoing change observed by others (Bjorkman 2017, Conrod 2018, Ackerman 2019, Konnelly and Cowper 2020, among others).

In this section I have focused primarily on the direct consequences of the proposed microparameter, presence or absence of uGender on D, for the (un)grammaticality of singular they within the framework of a pronominal typology that differentiates referential, variable, and predicative pronouns. In the next section I compare this proposal with other recent proposals regarding innovative variants of singular they, and discuss potential advantages of the present proposal.

4. Comparison with alternative accounts

This section reviews two similar previous accounts of innovative singular they: both Bjorkman (2017) and Konnelly and Cowper (2020) use the Distributed Morphology (DM) framework, with feature-based insertion rules and matching conditions between pronouns and lexical items.

Bjorkman (2017) proposes that the innovative form of singular they which allows it to co-occur with definite singular antecedents relies upon a reconfiguration of the morphosyntactic makeup of pronouns to allow gender features to either be present or absent. In this way, Bjorkman's proposal is similar to the one I make here: the conflict with speakers who reject (definite, specific) singular they is gender, not number. Bjorkman suggests that for innovative speakers, Gender has become an optional adjunct feature rather than an obligatory feature required for spell-out, citing Wiltschko (2008) for precedent of the existence of optional adjunct features.

Bjorkman assumes the structure for English pronouns roughly based on the typology by Déchaine and Wiltschko (2002), though with the addition of a $\phi$P projection that hosts the gender features relatively high in the nominal spine, following Steriopolo and Wiltschko (2010). This structure is shown in (24), where each feature is hosted on a head along the nominal spine. In Bjorkman's account, features are privative and are not differentiated as interpretable or uninterpretable – they are simply present or absent.

1. (24) (Bjorkman 2017: 8)

The insertion rules Bjorkman proposes for third person pronouns are given in (25).

1. (25) (Bjorkman 2017:7)

As in Déchaine and Wiltschko (2002), different subconstituents of the structure shown in (24) are spelled out for different pronouns; quantifier-bound pronouns spell out as pro-$\phi$P or pro-NumP, and full (referential) pronouns are spelled out as pro-DPs. Bjorkman (2017) proposes that the difference between innovative (singular they) speakers and conservative speakers is a difference in the featural makeup of the pronoun: for conservative speakers, [MASC]/[FEM] are obligatory features, but for innovative speakers, <Fem>/<Masc> are optional adjunct features (Wiltschko 2008). In the conservative system, absence of gender features is impossible; in the innovative system, absence of gender features results in a gender-neutral (but animate!) form. The conservative system therefore enforces a gender binary, where $\neg$[MASC] implies [FEM] (and vice versa).

Bjorkman's proposal has a few crucial differences from the proposal I have presented in this article. First, the use of privative features is based on the common insight that singular they is ungendered and lacks gender features in English; in my proposal, however, I locate interpretable gender low in the nominal spine (on n) and uninterpretable gender much higher (on D); thus, in order for a derivation to converge, the uninterpretable feature must be valued locally (via head movement) when it is present. Second, Bjorkman does not differentiate between definite epicenes and definite referential pronouns – both are full DPs in her system, and should therefore be subject to the same constraints; additionally, Bjorkman reports some judgments based on ‘mismatches’ between lexically-gendered antecedents and singular they which do not reflect the empirical findings of later work (Conrod 2019, Konnelly and Cowper 2020).¹⁴ Ultimately, the insights from Bjorkman (2017) are retained in my current proposal, but the syntactic structure and microparameters I give in section 2 more robustly predict some of the finer-grained distinctions between different uses of singular they than does Bjorkman's original squib.

Konnelly and Cowper (2020) provide an important follow-up to Bjorkman's (2017) squib; in their article, Konnelly and Cowper not only adjust for some of the irregularities in Bjorkman's reported grammaticality judgments, but also propose three distinct stages of development in the grammar of singular they. These stages are paraphrased in (26).

1. (26)

   1. a. Stage 1: speakers use and accept singular they with quantified, generic, or indefinite antecedents.

   2. b. Stage 2: speakers use and accept singular they with definite or specific antecedents so long as the antecedents lack lexical gender specification.

   3. c. Stage 3: speakers use and accept singular they with any antecedent, regardless of lexical gender specification.

   (Konnelly and Cowper 2020: 4)

Konnelly and Cowper note that not all speakers participating in the grammatical change progress through all three stages; while Stage 1 describes the ‘conservative’ starting point, they argue that speakers can and do pass directly from Stage 1 to Stage 3 without the intermediate step of Stage 2. Their description of Stage 2 is largely based on the judgments reported by Bjorkman (2017), which do seem to reflect a general state for some speakers and not an idiosyncrasy of Bjorkman herself.

Like Bjorkman, Konnelly and Cowper's proposal also uses the DM assumption that morphophonological form will be determined by (mostly post-syntactic/post-spellout) lexical insertion rules based on grammatical features; also like Bjorkman, Konnelly and Cowper take singular they to be lacking gender features (and take gender features to be optional adjunct features). However, the syntactic structure they propose differs from Bjorkman's in the location of these features on syntactic nodes in the nominal spine, as shown in (27) – this structure is intended to represent the grammar of their Stage 1 speakers.

1. (27) (Konnelly and Cowper 2020: 8)

In Stage 1, features of the Num head are copied to D$_Q$ for quantifier-bound pronouns; as in Bjorkman's (and my) proposal, MASC/FEM are not binary, obligatory features in innovative varieties (Stages 2 and 3).

The three analyses are similar in that they all propose some kind of route wherein singular they loses its gender features in innovative varieties. For Bjorkman (2017), MASC/FEM are contrastive and obligatory in the conservative variety, but non-contrastive in the innovative variety. However, Bjorkman retains an insertion rule wherein pronouns must spell out a superset of the features of its antecedent: thus, if the antecedent is explicitly gendered, then pronominal gender features become effectively obligatory.

For Konnelly and Cowper (2020), in Stages 1 and 2 the MASC/FEM gender features are contrastive, and must be present on a pronoun that has a gendered antecedent. The difference between Stages 1 and 2 is that, for Stage 2 features, fewer antecedents carry gender features, and thus force agreement less frequently. Because Konnelly and Cowper attempt to explain ungrammaticality purely through features of antecedents, their analysis cannot explain why conservative speakers will still reject dsT when used without a linguistic antecedent. In Stage 3, they propose that the MASC/FEM features lose their contrastive status throughout the system, both for pronouns and lexical nouns.¹⁵

For Konnelly and Cowper, then, the difference between Stage 2 and Stage 3 is that gender features, for Stage 3 speakers, are not only optional on pronouns, but also optional on all lexical items: for innovative speakers, gender represents “completely optional modifier features” (Konnelly and Cowper 2020:15).

Translating the three stages to the microparameter I propose here, Stage 1 speakers are non-dsT users, meaning they retain uGender on referential D, while Stage 2 and 3 speakers are dsT speakers, meaning they have lost the obligatory uGender on referential D.

The differences between my proposal and Bjorkman's (2017) also largely apply to Konnelly and Cowper (2020); while the latter make a stronger distinction between quantificational determiners and non-quantificational determiners, they still do not differentiate between definite and specific uses, and still explain grammatical crashes experienced by more conservative speakers by means of gender features on lexical nouns. In the next section I will investigate this particular difference in more detail, and show evidence that lexical features are not necessary to explain the variation found. However, as with Bjorkman, many of the valuable insights made by Konnelly and Cowper (2020) are compatible with the present proposal.

4.1 Lexicalist approach

An important commonality in how Bjorkman (2017) and Konnelly and Cowper (2020) explain conservative speakers is that both accounts rely on the presence of gender features on lexical nouns in contemporary English, including (or especially) on proper names. In both accounts, singular they can only appear (for some speakers) with a proper name if that proper name lacks gender features. This correctly predicts the judgments reported in (28)–(29), which are representative of the Stage 1 or 2 speakers for Konnelly and Cowper, and of the innovative variety for Bjorkman.

1. (28)

   1. a. Mary$_i$ likes herself$_i$

   2. b. * Mary$_i$ likes themself$_i$

2. (29)

   1. a. Taylor$_i$ likes herself$_i$

   2. b. Taylor$_i$ likes themself$_i$

In this approach, which I refer to as the lexicalist approach, the contrast between (28b) and (29b) is due to lack of gender features on certain lexical nouns and names; Taylor is a gender-neutral name, so it doesn't have any gender features and doesn't conflict with they. (Notably, Konnelly and Cowper's Stage 3 speakers lack (obligatory) gender features altogether, which is one of the major differences between their approach and Bjorkman's.)

There are a few problems with the lexicalist approach, which I will first outline, and then show how my microparametric proposal solves them by relieving singular they of lexical gender features altogether. First, it is not clear how proper names get assigned gender features in the first place – this is a bit of an issue because, as Ackerman (2019) has pointed out, not only do proper names often show considerable synchronic variability in gender assignment, but the gender frequency of some proper names change over time. If it is the case that proper names carry formal features denoting gender (even if they are optional features) the synchronic and diachronic variation in proper names presents significant difficulties for explaining the role of those features in grammaticality judgments like those shown in (28)–(29).

A second issue is that the nature of lexical storage for gender-neutral or gender-ambiguous names is problematic if it is assumed that lexical entries (for proper names or indeed any common nouns) include formal MASC/FEM features – again, even if these features are optional. There are two possible explanations for how a name like Taylor would have to be stored in the mental lexicon if lexical nouns and names did indeed carry gender features. The first option is that Taylor (and other ambiguous names) would need to be underspecified in the lexicon, even for the most conservative speakers – because even Stage 1 or non-dsT speakers will generally accept such names with either he or she – meaning there must then be some additional mechanism that allows conservative speakers to apply gender features to lexical items. The second option is that gender-ambiguous names like Taylor would need to constitute multiple homophonous lexical entries (up to three!), and speakers would need to determine whether a male, female, or other type of Taylor is under discussion when determining whether the pronoun is matched appropriately or not. Both of these possibilities are not only computationally somewhat cumbersome, but fail to robustly explain why conservative speakers would in fact object to the use of they with gender-neutral names.

A third issue with the lexicalist approach is that featural mismatch does not robustly explain why even non-dsT speakers will occasionally produce (and accept without comment) singular they with explicitly gendered generic or quantificational antecedents like (13), repeated here as (30).

1. (30) By some miracle, a woman$_i$ can feed a baby with their$_i$ body.

If, indeed, any lexical nouns can carry gender features at all, it is not clear why those gender features should influence pronominal matching for proper names but not quantified antecedents – and, again, it is worth noting that neither Bjorkman (2017) nor Konnelly and Cowper (2020) make a syntactic differentiation between definite antecedents and specific definite antecedents. Data from Lal Zimman (p.c.) also suggest that examples like (30) are not at all rare, nor are they restricted to speakers who would accept dsT otherwise. I include here one more such attested example, to show that this restriction does not explain a great many actual utterances from otherwise conservative speakers.

1. (31) I'm the kind of man that likes to know who's buying their drinks.¹⁶

No version of the lexicalist approach, or indeed any proposal that gender features persist on lexical nouns in contemporary English, can account for utterances like (30) or (31) without apparently stipulating that sometimes, such features are optional or can be ignored. The lexical items in question, woman and man, would not be reasonably described as gender ambiguous either generally or in the contexts in which they were spoken.

Under Konnelly and Cowper's account, utterances like (30) and (31) are only possible under the assumption that the speakers are in Stage 3, because the antecedents contain explicitly gendered material; indeed, their Stage 2 was modeled after the judgments given by Bjorkman (2017). My own account leaves open several possibilities to explain Stage 2-type judgments. First, it is possible that Stage 2 speakers do retain a uGender feature on D, and are potentially valuing that feature with a value like GENDER-NEUTRAL – which can only fulfill that feature for sufficiently gender-neutral referents. The second possibility is that the ungrammaticality reported by Stage 2 speakers is in fact an error in the use-conditions: that is, Bjorkman's judgments (in 2017) were actually a matter of inappropriateness, rather than morphosyntactic ill-formedness. The third possibility is similar: it is possible that Stage 2 speakers find they inappropriate or semantically infelicitous when a more specific pronoun (like she or he) would be permitted. This third possibility is essentially an instantiation of Maximize Presupposition!, as formulated by Heim (1991). For an extended discussion of variability of how semantic or pragmatic constraints such as this are ranked (within and between speakers), see chapter 4 of Conrod (2019).

In the next section, I will discuss briefly the alternative explanation that Konnelly and Cowper put forward to explain dsT/Stage 3 speakers, and discuss some of the (mainly theory-internal) disadvantages of that approach within the context of the lexicalist approach for their Stage 1 and 2 speakers.

4.2 Radical feature-free Lexicalist approach

For Konnelly and Cowper (2020), the difference between Stage 1 and Stage 2 speakers, on the one hand, and Stage 3 speakers, on the other, is that the most innovative speaker group always allows they as an option, regardless of antecedent; in their system, this is an effect of the non-contrastive status of gender features on both lexical nouns and pronouns.

[T]he change in the English pronominal system at Stage 3 […] is subtle. In fact, the pronominal system remains exactly as it was; all that changes is the status of the gender features themselves, specifically, whether they are required contrastive features or optional modifier features.

(Konnelly and Cowper 2020: 16)

While this approach, which I refer to as the radical feature-free lexicalist approach, solves the problems of the lexicalist approach which I discussed above, there are some issues in how Stage 3 might be operationalized in the same general language system as Stage 1 and 2. These issues are primarily theory-internal, as the feature-free lexicalist approach makes otherwise very generous empirical predictions that do not conflict with the data I have presented so far.

The first issue with the feature-free lexicalist approach is that Konnelly and Cowper do not give a model for how adult learners might move between stages; if speakers do often start out in Stage 1 and progress eventually to Stage 3 (with or without an intermediate period in Stage 2), this would require speakers moving from 1 to 3 to reanalyze gender features on both pronouns and lexical nouns as non-contrastive. Speakers who pass through Stage 2 (either from 1 to 2, or from 2 to 3) would also need to significantly reorganize their lexicon, so that lexical items that once carried (contrastive) gender features lost those features. This loss in the course of an overall grammatical change in the syntax of a language is plausible, but predicts a greater amount of intervening idiosyncrasy. At present there is very little data on the particulars of individual speaker change over the lifespan with respect to singular they; future work should probe individual lifespan change in order to test whether that change could be robustly explained by eventual radical loss of gender features that were previously present in a speaker's grammar. Nevertheless, it is to the advantage of my own proposal that I have not relied on the presence of gender features on lexical nouns in the first place, and therefore do not need to explain how or why they might be erased by an individual (adult) speaker.

One additional consequence of featural explanations, which is shared by the present proposal as well as those of Konnelly and Cowper (2020) and Bjorkman (2017), is that non-contrastive gender features in innovative varieties are not necessarily privative. Thus, all three analyses allow for the possibility not only of items that carry neither MASC nor FEM, but potentially both MASC and FEM appearing on a single item. In a lexicalist approach, where these features could potentially coincide on lexical nouns, this may arguably be the case for certain lexical items (which I will not print due to their transphobic connotations). In all three approaches, a doubly-gendered pronoun might consitute ‘mixed’ forms (if one wishes to analyze he/she or s/he as a single pronoun).

Alternatively, it is possible that doubly-gendered pronouns are obligatorily spelled out as they in English – and that this is a matter of syncretism in plural they. Languages with more robust gender morphosyntax, such as Spanish, do show apparent gender marking on plurals, and languages tend to have idiosyncratic patterns of marking the gender of a mixed group of plural referents.

In the system I have proposed here, MASC and FEM are values of an interpretable iGender feature, rather than separate privative features. Thus, it is not necessarily predicted that there should exist items with both MASC and FEM present. Indeed, the valuation-of-iGender system that I use implies that iGender could ultimately attain possible values other than MASC and FEM; for example, iGender:NONBINARY is a logical possibility in such a situation where NONBINARY becomes a legible and widely-recognized ‘answer’ to the gender ‘question.’ This may also extend to neopronouns like those discussed by Miltersen (2016): pronouns like fae/faer/faeself could potentially develop an iGender value of FAE. Neopronouns in English are underexamined, however, and I hope that future research in both approaches incorporates more data relating to them.

Finally, the radical feature-free lexicalist approach presents an issue for parsimony in a Minimalist framework (which I recognize neither Konnelly and Cowper nor Bjorkman were particularly striving for): namely, in my proposal, the microparameter proposed is consistent with the Borer Conjecture (Borer 2014). This does not necessarily make different empirical predictions from Konnelly and Cowper's, but provides a slightly easier explanation for lifetime change, since resetting a featural parameter on a single functional head (referential D) is less burdensome than rewriting one's entire lexicon.

In the next section, I will detail further how the approach I take in this article does not rely on features of lexical nouns, and can additionally much more robustly explain sociopragmatic variation in gendered pronouns as well as the presence or absence of dsT in any given speaker's grammar.

4.3 Reference approach

In section 2, I showed that my proposal is specific to referential pronouns, and depends on the presence or absence of a uGender feature on D. Crucially, what causes ungrammaticality for non-dsT speakers is not an insufficient match in features between a pronoun and its antecedent, but rather that there simply must be some kind of gender feature on referential pronouns, whether or not it is “correct”. What this predicts is that non-dsT speakers will err on the side of misgendering referents rather than using dsT; this does in fact happen with great regularity (e.g., Pullum 2017).

My proposal also reflects the metalinguistic commentary made by Pullum (2017): the difference between dsT grammars and non-dsT grammars is that gender features are either optional or they are obligatory. For referential pronouns, the pronominal gender features enter the derivation independent of an antecedent; indeed, neither Konnelly and Cowper nor Bjorkman much discussed the possibility of a pragmatic (not a linguistic) antecedent. The features are then evaluated with regard to the referent, based not on “match” but rather on appropriateness.

In other words, dsT speakers will occasionally avoid or reject singular they based not on grammatical constraints but on social relational knowledge about what pronoun is appropriate to the context. Likewise, non-dsT speakers will make their own best attempts to use a pronoun that is appropriate to the pragmatic context – but non-dsT speakers, as Pullum self-reports, are at a disadvantage, since their grammar constrains them to a forced binary choice. As such, non-dsT speakers will also occasionally hyper-correct by switching pronouns mid-conversation for the same referent, especially when faced with potential uncertainty or difficulty in expressing an appropriate gender for the context (see Conrod 2019: 169 for examples of this from sociolinguistic interviews).

One important aspect of this proposal is that, in discourse contexts where gender is not already clear, a pronoun can do the work of introducing gendered meaning into that discourse context. Pullum (2017) introduces he into a discourse context where no precedent otherwise existed – except, as Pullum notes, the gendered bias introduced by a first name. This proposal also robustly accounts for instances of pronoun-switching, by dsT and non-dsT speakers alike, throughout a conversation; since gender-mismatches are not here considered grammatical anomalies but rather sociopragmatic ones, there is, under my proposal, no grammatical constraint that forces matching for even the most conservative speakers.

This proposal captures the same insights as those made by Konnelly and Cowper and by Bjorkman, but uses a different mechanism: that an unvalued uninterpretable feature (gender) causes a crash is, in some ways, a different way of saying that certain (contrastive/privative) features are obligatory. Rather than having grammar-external metafeatural information of obligatoriness, this is represented directly in the syntax (specifically on the specification of D).

This approach has several advantages besides those detailed above. First, the matter of proper names is much more flexible under the reference approach – there is no need to worry about names changing in gender frequency over time, and learning new proper names depends on social world knowledge about the referent rather than formal features. Speakers needn't have multiple lexical entries for gender-ambiguous names – instead, the appropriateness of any given pronoun depends on the sociopragmatic context (e.g., are we talking about Taylor Swift or Taylor Lautner?). No formal features for lexical nouns or proper names need to be learned or unlearned by speakers as their grammars undergo change throughout their lifetime – what pronoun to use depends entirely on who we're talking about, not the particular words or morphemes we're using to talk about them.

A second advantage is that the single microparameter proposed here explains very robustly why conservative (non-dsT) speakers will reject or avoid singular they, but only with proper names, not with other definite antecedents. Since the difference between D heads is, in my account, a difference between referential D and all other Ds, definite antecedents needn't cause a crash unless they pick out a sufficiently specific referent. This microparameter also very cleanly explains how individual speakers can move between stages in the grammatical change without significantly reorganizing their lexicon or rewiring their featural inventory – only a single D head needs to be reanalyzed for a single feature.

Finally, I believe a significant advantage of the proposal I have made here is that it maintains the validity of the important insights made in previous work on singular they. While I have framed it in the terms most familiar to those working in Minimalist spaces, this microparameter may also be transliterated into other frameworks, so long as the core proposal I have made can be maintained: what differentiates dsT speakers from non-dsT speakers is whether gender features are obligatory on referential pronouns – that is all.

5. Conclusion

In this article, I have proposed a new account for existing synchronic variation among English speakers whose grammaticality judgments vary with regard to singular they when anteceded by a proper name or specific referent. This account proposes that the deciding factor is not the presence or absence of gender features (either on pronouns or on lexical nouns that may antecede them) but rather on the presence or absence of an uninterpretable Gender feature on referential D heads. The proposal is made within the context of a three-way pronominal paradigm that differentiates referential pronouns from variable and definite pronouns (and from predicative pronouns).

The core of this proposal is that reanalysis of ambiguous singular they has allowed speakers to acquire the pronominal system of English without uGender on D, and that the absence of uGender on referential D is what has allowed singular they to expand to more specific uses than were previously observed. This formulation of the current proposal shares the insights from Bjorkman (2017) and Konnelly and Cowper (2020) – namely, that grammatical anomaly around definite singular they results from a crash due to lack of gender features – but has some advantages over these previous accounts. One advantage is that the (micro)parameterization may more robustly explain the sociolinguistic variation observed by these and other authors (Conrod et al. 2016, Ackerman 2017, among others). An additional advantage of the particular division of the pronominal typology is that the head-raising account given here correctly explains the different restrictions on different types of pronouns, without over-restricting or erasing existing sociopragmatic variation.

Finally, this analysis provides support for some more general proposals about pronouns and syntax given elsewhere. The three-way distinction used here maps very closely onto the pronominal typology proposed by Déchaine and Wiltschko (2002), and is largely compatible with their observations about cross-linguistic variation. The analysis also supports the framework of use-conditional semantics as in Gutzmann and McCready (2014), wherein syntactic structures can be evaluated for sociopragmatic anomaly in a way that does not over-attribute all anomaly to the narrow syntax, but still appropriately reflects the compositional nature of use-semantic meaning.

Future work focusing on dsT and pronouns more generally can build off of two major aspects of the present proposal. First, the predictions made by the microparametric proposal suggest that lifespan change of individual speakers’ grammars from non-dsT to dsT may occur relatively quickly, since only a single parameter needs to be reset – if the microparametric approach is correct, speakers should not show slow development of lexical remapping, but rather should be able to wholly master dsT within a year or two. Second, the predictions made by the reference approach suggest that speakers who do acquire dsT will not be significantly different in terms of how and when they deem pronouns to be appropriate or inappropriate in a given context, except that dsT speakers will have an unmarked option at their disposal and non-dsT speakers grammatically will not. Thus, non-dsT speakers may employ alternate strategies such as pronoun avoidance in contexts where a dsT speaker might use they; and dsT speakers are unlikely to start using they for any and all referents unless it is deemed socially appropriate. Future sociosyntactic work may verify these predictions.