2.1 Allomorphy and allosemy

The framework of Distributed Morphology (DM) holds a realizational view of exponence and interpretation in which structures generated in narrow syntax map to interfaces that realize these structures (see Embick & Noyer 2007 and references therein). This is often represented with the standard Y-Model, as depicted in (4). In this model of the grammar, contextual allomorphy is modelled via a mapping relationship at PF between (some set of) abstract syntactic terminals to exponents called Vocabulary Items (see e.g. Halle & Marantz 1993; Halle 1997; Embick & Noyer 2007; Embick 2015; among many others). In a parallel view of allosemy, abstract syntactic terminals map onto LF entries comparable to Vocabulary Items; these are sometimes referred to in the literature as entries from the Encyclopedia (e.g. Marantz 1997; Harley & Noyer 1999; Harley 2014).

In the standard formulation, each Vocabulary Item at PF is specified to be inserted when the input contains a certain set of features and a particular local context. When multiple Vocabulary Items are compatible with the input, the one that is selected is specified for the largest subset of features (the Subset Principle) and/or has the most specific contextual restriction; this formulation follows the logic of the well-known Elsewhere Principle.

The Vocabulary Items in (5) illustrate how allomorphy is conditioned in this type of system using the example of the English past tense, whose suffixal form corresponds to several distinct realizations.

In (5), the total number of features does not vary across Vocabulary Items that realize the past tense node, but the contextual restriction does vary, specifying the list of roots with which each exponent can appear. The “default” plural exponent -d (often written -ed) lacks a contextual restriction: it is selected when the root in the input does not appear in the lists of the other Vocabulary Items’ contextual restrictions (including for e.g. nonce verbs and novel coinages) or when the root in the input is not local, as in the case of denominal verbs which are argued to cyclically separate the root from T (e.g. stringed/*strung the snap peas; see Adamson 2018 and references therein).

The literature on allosemy has suggested that contextual restrictions for meaning alternations can be modelled in a parallel fashion. Most of the existing literature concerns contextual allosemy of roots and a set of functional heads (e.g. Voice), but not with nominal features per se. An important exception is Dali (2020), who pursues an analysis of the representation of feminine gender in Tunisian Arabic in terms of contextual allosemy: according to her analysis, the interpretive contribution of feminine gender in Tunisian Arabic varies depending on its morphosyntactic environment, giving rise sometimes to a feminine inference but at other times a singulative inference.

For an allosemy-based approach to meaning alternations, we submit that there are two core principles of the theory of allomorphy that should meaningfully extend to allosemic alternations: elsewhere ordering and locality, which in conjunction with each other produce a domain effect (on which, see Embick & Marantz 2008; Embick et al. 2023; among others). We address these principles and their interaction in turn.

In DM, elsewhere exponents are Vocabulary Items that have a “default” distribution, appearing when the input does not match the feature content or the contextual restriction for any more specific Vocabulary Item, as in the case of the aforementioned English -d in (5). Extending this concept to gender allosemy, suppose we have a syntactic terminal with a gender feature [m] (masculine) that corresponds to two allosemic entries, one of which is selected in specific environments, while the other is selected elsewhere. This is roughly schematized as in (6), with a male alloseme specified for some set of roots and an animate alloseme appearing elsewhere. We will make the interpretations somewhat more precise below;³ (6) is intended currently only for illustration of the concept.

According to (6), a node with an interpreted masculine feature is interpreted as male in the context of specific roots, but is interpreted as animate elsewhere. The elsewhere item will thus be selected if i) a root that it appears with is not specified for the male entry, or ii) there is no root in the local context of [m] at all. While this is currently intended as a hypothetical, we will show how entries along the lines of (6), in conjunction with explicit assumptions about semanticopragmatic competition, give rise to correct predictions for the behavior of the masculine gender in Greek below.

Regarding the issue of locality, the DM literature has advanced the idea that contextual allomorphy can only be sensitive to certain elements within a domain, though researchers have offered various perspectives on exactly how much context is available at a given node (e.g. Bobaljik 2000; Embick 2010; Moskal 2015b; among many others). While details differ, the shared hypothesis is that the choice of exponent cannot be conditioned by elements that are too far away.

Research on allosemy has similarly advanced the idea that contextual allosemy is limited to a local domain. For example, Marantz (2013) suggests that contextual allosemy is constrained both cyclically and by “semantic adjacency”, and we adopt this view presently. For concreteness, we assume a strict view in which lexical heads are cyclic (e.g. Embick 2010), and the interpretation of such heads cannot be conditioned by context outside of the cycle they delineate, though we would like to stress that many less restrictive theories of allosemy would also be compatible with the findings we explore below. The cyclic constraint is illustrated for a cyclic nominalizing head n in (7), which can “see” the root with which it combines for purposes of allosemy, but cannot see β or anything higher.⁴

In combination with each other, elsewhere ordering and locality restrictions produce a domain effect. In essence, it is possible for some elements to appear either inside a domain or outside of it; for a node (or set of nodes) that is within the domain and whose allomorphy is sensitive to this element, the element is visible for the purposes of allomorphy for the former, but not for the latter. Consequently (in the absence of other conditioning factors), the element within the domain may condition a specific allomorph whereas the latter will result in the selection of an elsewhere item.⁵

In order for entries like (6) to be able to derive gender meaning differences across different root contexts, gender features must be local to roots. We follow much of the recent literature within DM according to which gender features reside on a nominalizing head n, and combine with a category-less root (among others, Lowenstamm 2008; Kramer 2015; 2016a; Adamson & Šereikaitė 2019; Adamson 2024a; and references therein).⁶ The n-based view of gender features reflects the notion that gender is a property of nouns, and has been employed to account for, among other things, patterns of gender agreement with nominalizations. The basic decomposition is as in (8), where n bears a feminine feature [f].

While local to roots, gender features are not structurally local to everything. Consequently, some elements are expected not to be able to condition gender allosemy. For example, gender features on n will not be local to features from DP-external structures, and this is reflected in the generalization that nominal gender interpretation is not sensitive to the aspect or the tense of the clause in which the nominal appears; no such sensitivity seems to appear in Greek, and this is moreover a cross-linguistically valid generalization, as far as we are aware. Interestingly, gender features under the current view are not even structurally local to other elements within the nominal they belong to, under the received view that nominal structure is complex. In particular, we assume a DP structure (following Abney 1987; Ritter 1991; and much subsequent work), whereby D and Num are separate projections within the nominal spine (9); within this nominal structure, a gender feature (like [f] in (9)) is local to a root, but is not local to D: the former is semantically adjacent and within the cycle delineated by n while the latter is not.

Consider both the elsewhere item and the locality property together for the entries in (6). There is a contextual restriction for the male interpretation, dictating that this entry be selected only when the nominalizing head occurs with particular roots. The expectation is then that the elsewhere item animate will be selected for [m] if i) it appears in the context of roots that do not select male; ii) the head appears with no root; or iii) if a root is present but not local. We will demonstrate below that this type of prediction for the interpretation of masculine gender in Greek is borne out. However, in order to do so first requires an elucidation of the relationship between gender interpretation and semanticopragmatic competition, a task which we undertake in the next subsection.

To summarize so far, a contextual allosemy account of meaning alternations can generate predictions when it is constrained in a comparable way to contextual allomorphy. In particular, a default (elsewhere) alloseme is expected to surface when the conditions for a more specific alloseme are not met. This can happen either when there is no conditioning element or when a conditioning element is present but nonlocal. Within the domain of gender allosemy, gender features on n are expected to be able to take the specific interpretation (male in this case) in the context of particular roots, but otherwise require the default alloseme (animate), when roots are either absent or nonlocal. We will demonstrate below how this is instantiated for gender allosemy in Greek.

2.2 Semanticopragmatic competition

A substantial literature has identified a form of semanticopragmatic competition whereby more semantically narrow interpretations preclude the use of more semantically general forms. Implementations differ, but this idea underpins various principles proposed in the literature, including Maximize Presupposition (Heim 1991; Sauerland 2002; 2008; Singh 2011; among many others), Lexical Complementarity (Harbour 2016; Toosarvandani 2023; Adamson & Anagnostopoulou 2025), principles of contrast in feature geometries (Harley & Ritter 2002; McGinnis 2005; among many others), and specifically for gender, the Principle of Gender Competition (Sudo & Spathas 2020; Spathas & Sudo 2020).

Such competition-based principles have been invoked in the domain of nominal features to account for the relation between markedness patterns and interpretation. For example, the idea that third person is interpretively unmarked relative to first and second persons has been captured by analyzing third person as having an underspecified interpretation,⁷ which can nevertheless not be applied when referring to discourse participants (i.e. the speaker and/or the addressee) because participant pronouns are semantically “stronger” in their interpretation, thereby blocking use of the “weak” third person forms (see relevant discussion in Toosarvandani 2023). For concreteness, we adopt the formulation of this type of competition from Harbour (2016), called lexical complementarity (LC), which has been invoked in accounting for how person features come to be contrastively interpreted (see especially Toosarvandani 2023 for an analysis of person with privative features rather than with Harbour’s binary features).⁸ While we adopt LC, other models of semanticopragmatic competition could also be employed to derive similar results, and we make no claims regarding which type of theory of contrastive inference is to be preferred (see e.g. Bobaljik & Sauerland 2023 for critique of LC in the domain of clusivity).

The effect of Lexical Complementarity can be seen for how person features are interpreted on pronouns in (11)–(12). The denotations in (11) characterize the interpretations of each person feature: [speaker] is restricted to first person reference, whereas [participant] refers to discourse participants, conceivably either to the speaker or the addressee(s); the extension of [participant] is thus a subset of the extension of [speaker]. For singular reference to the speaker, Lexical Complementarity dictates that [speaker] be used; it preempts the use of [participant] (by itself, without [speaker]) even though the speaker belongs to the extension of [participant]. The same pattern holds for [π], which denotes a general set of individuals, but cannot be used in reference to discourse participants due to Lexical Complementarity. This is represented schematically for singular reference in (12), where the letter i stands for the speaker, the letter u stands for an addressee, and letter o stands for other individuals (following Harbour’s 2016 convention), with struck-through text indicating that reference to these individuals is blocked by Lexical Complementarity.

Semanticopragmatic competition along these lines has been suggested for the use of feminine vs. masculine genders, when the former has a sociocultural gender inference (female) that the latter lacks, in cases where masculine does not have a male-specific interpretation (Sauerland 2008; Sudo & Spathas 2020; Spathas & Sudo 2020; and relatedly Percus 2006). Denotations are sketched as in (13) (in terms of assertion for simplicity). This competition is represented in LC-terms schematically in (14) for singular individuals: the extension of the feminine feature includes only female individuals, while the extension of the masculine feature includes animate individuals more broadly, including both male and female individuals. However, LC effectively removes female individuals from consideration for the use of masculine, such that reference to an individual who is a woman cannot employ a masculine form, as in (15).⁹

Consider now how allosemy is expected to interact with Lexical Complementarity. Suppose there is a gender value with a more general extension than the masculine: [m] is under its default interpretation animate but sometimes male, while the other feature value – which we can provisionally call [γ] – corresponds to entities (broadly construed), which encompass both animates and inanimates – this meaning is thus effectively vacuous.

In (16), there is an allosemic alternation of the masculine, but not for [γ]. However, the effect of Lexical Complementarity for [γ] will be different depending on which [m] alloseme is selected: the result will be that [γ] is used either for non-male reference or for non-animate reference, respectively.¹⁰ We will suggest that this is the correct characterization of the neuter gender in Greek, though we will first motivate our approach to gender in Greek more explicitly.

We now have both components in place: i) allosemy is like allomorphy in that its alternations are subject to locality and elsewhere ordering, and ii) semanticopragmatic competition dictates that stronger meanings preempt the use of weaker ones. We now proceed to apply this type of account within the context of the Greek gender system – with special reference to neuter gender – before turning to some of its more interesting predictions in Sections 3 and 4.

2.3 The Greek gender system

Greek is a language characterized as having three grammatical gender categories – feminine (fem), masculine (masc), and neuter (neut) – which are reflected in agreement with adjectives, determiners, and other categories (on definitions of grammatical gender, see Hockett 1958; Corbett 1991; Kramer 2015). This three-way distinction in gender categorization can be observed with determiner agreement as in (17), as Greek determiners agree in both gender and number (as well as case) with the corresponding noun. (On the relation between gender and declension in Greek, see Alexiadou 2004.)

For the vast majority of human-denoting nouns, fem and masc genders are correlated with inferences about conceptual gender: feminine human-denoting nouns typically carry a feminine inference (see Merchant 2014; Alexiadou 2017; Sudo & Spathas 2020; Spathas & Sudo 2020; Adamson & Anagnostopoulou 2025), which we label as female. Though there are exceptions, other human-denoting nouns are overwhelmingly masc, and some of these nouns carry male-specific inferences while others carry no male (or female) inference.

Whether a human-denoting noun carries gender inferences like female or male can be diagnosed in various settings, such as focus-sensitive environments like only-constructions and superlatives (Sudo & Spathas 2020), by ellipsis (Bobaljik & Zocca 2011; Merchant 2014; Alexiadou 2017; Sudo & Spathas 2020; Sprouse et al. 2022; among others), by unknown-gender environments (e.g. Jakobson 1984; Bobaljik & Zocca 2011; Adamson & Šereikaitė 2019; Sudo & Spathas 2020; Adamson & Anagnostopoulou 2025), and by pluralization (Jakobson 1984; Wechsler 2008; Bobaljik & Zocca 2011; Sudo & Spathas 2020; Adamson & Anagnostopoulou 2025).

As an illustration, consider the plural nouns in (18)–(19), whose masc plurals behave differently from each other, as mentioned above for (1). In (18), the noun meaning ‘teacher’ has both fem and masc variants: the fem variant gives rise to the inference that every member of the group is a woman, whereas the masc is compatible not only with an exclusively male group, but also with a gender-heterogeneous group, e.g. one consisting of men and women. This is a fem/masc markedness asymmetry: the fem gender brings in a (sociocultural) gender inference, whereas the masc gender does not.¹¹

Merchant (2014) and Sudo & Spathas (2020) identify the nouns in Table 1 as examples of (what we are calling) asymmetric nouns, on the basis of their behavior in ellipsis and other focus-sensitive expressions, respectively (see also Alexiadou 2017).¹² (We have excluded the examples where the masculine and feminine do not share a common phonological profile.)

In contrast, the noun meaning ‘lady’ or ‘gentleman’ in (19) exhibits a symmetric relation between feminine and masculine plurals: the feminine gives rise to the inference that every member of the group is a woman while the masculine gives rise to the inference that every member of the group is a man. Thus neither alternant can be used to refer to a gender-heterogeneous group consisting of both women and men.

There is thus a different interpretive association for the use of the masculine gender with (18) versus (19). The list in Table 2 contains what Merchant (2014) and Sudo & Spathas (2020) identify as symmetric nouns, again on the basis of their behavior in ellipsis and other focus-sensitive expressions, respectively. (According to Sudo & Spathas 2020, for both asymmetric and symmetric nouns, membership in these classes for some items can be variable among speakers, e.g. thios/thia ‘uncle/aunt’.)

In the current system, interpretable gender features are interpreted on n, which combines with a root (Kramer 2009; 2015). We assume that a root combining with an n with interpretable gender denotes a predicate of individuals of type <e,t>.¹³ Under this view, roots have a consistent interpretation regardless of which gender they occur with (see Sudo & Spathas 2020 for a different perspective, and Harris 1991 for a related view in a different framework); we set to the side the possible issue of root allosemy. (20) offers examples for the roots in the nouns meaning ‘teacher’ and ‘gentleperson’.

According to the current view, roots themselves do not carry gender inferences, and the differences in how masculine gender comes to be interpreted for asymmetric vs. symmetric nouns is a result of contextual allosemy of gender on n. (On the requirement for a root to combine with n, see e.g. Embick & Marantz 2008 on the Categorization Assumption.) (21) sketches interpretations for nominalizing heads with either feminine or masculine features, assuming they are also semantically of type <e,t> and combine with <e,t> roots via Predicate Modification.¹⁴ The contribution of the female inference is consistent across feminine nouns, but the meaning of masculine depends on which root it occurs with; it is male in the context of the root meaning ‘gentleperson’, though in other contexts it lacks this inference. One way to model this would be to say that masculine in such cases lacks gender inferences altogether (see Sudo & Spathas 2020); we instead propose following Adamson & Anagnostopoulou (2025) that masculine bears the interpretation animate. (On the choice of the latter as an elsewhere alloseme rather than the reverse ordering, see Section 4.)

The animate interpretation of the masculine reflects the generalization in the language that human-denoting nouns are overwhelmingly either masculine or, if denoting female individuals, feminine. The status of masculine as animate will also have consequences for the interpretation of the neuter gender, as we will see below.

Adamson & Anagnostopoulou (2025) (henceforth A&A) propose that the neuter gender in Greek is unmarked relative to the other genders, and comes to be interpreted as inanimate via LC. A&A suggest that neuter’s unmarked status is reflected, for example, in neuter predicative agreement appearing with agreement with clausal subjects, a default diagnostic (e.g. Kramer 2015; Adamson & Šereikaitė 2019; among others).

While other authors have suggested that neuter is a morphosyntactic default or “residue” gender in three-gender languages like Greek (Kramer 2015; Corbett 1991; see Tsimpli & Hulk 2013; Alexiadou et al. 2020; among others on Greek), A&A argue that neuter gender in Greek can also bear an inanimate interpretation, accounting for data from coordination resolution facts; they also suggest that it is consistent with other facts about neuter gender in the language. For example, the majority of (non-derived) neuter nouns denote inanimates and there appears to have been a diachronic pressure for inanimate nouns in the masculine and feminine categories to become neuter.

In the current system, where gender features occur on nominalizing heads, we can model noun gender as in (23), where feminine nouns are formed as in (23a), masculine nouns as in (23b), and the less marked neuter nouns as in (23c).¹⁵

A&A suggest an inanimate interpretation is derived for neuter gender if neuter is very general in its meaning, much like third person, but reference to animate individuals with the neuter is precluded due to LC. The denotations they offer are in (24), with the effect of LC schematized for singular individuals in (25). As shown in (25), the neuter association with inanimacy comes out via LC with the animate meaning of masculine; the animate meaning also interacts with the feminine to derive i) reference to men using masc and ii) the use of masc in gender-unknown contexts with humans.¹⁶ (We assume that the femaleness inference of the feminine feature also entails an animacy inference.)

While (24) derives the association of neuter gender with inanimacy, it fails to reflect the male inference seen with some masculine nouns, as discussed above. Adding our allosemy of the masculine including the interpretation male, we arrive at the denotations in (26). Notably, a choice between more than one alloseme is restricted to the masculine.

We will see shortly how the male alloseme nevertheless impacts the interpretation of the neuter, though it will be helpful in this process to consider how LC applies in the case of plural individuals.

We have heretofore only considered how the meanings of our gender features, in conjunction with Lexical Complementarity, characterize singular individuals. For plural individuals, consider how gender interpretation is relevant to groups. Several authors have suggested that (sociocultural) gender is a distributive property for a group (Corbett 1991; Wechsler & Zlatić 2003; Wechsler 2008; Sudo & Spathas 2020; Toosarvandani 2023; Adamson & Anagnostopoulou 2025), such that any gender inference must hold for every member of that group.¹⁷ This is consistent with the plural test described above. Starting first with asymmetric nouns, in the example in (18), repeated in (27), the feminine is associated with a female inference, and therefore requires that every member of the group be a woman. In contrast, the masculine with the asymmetric noun in (27b) is associated with an animate interpretation, so this plural noun is applicable for either a homogeneous group consisting exclusively of men, or a heterogeneous group consisting of (for example) men and women.¹⁸

This plural behavior is in line with the combination of i) the female and animate gender inferences for feminine and masculine nouns, respectively; ii) the distributivity of gender inferences; and iii) the effect of LC. We can again schematize extensions and the effect of LC as in (25) and consider possible two-individual plural groups formed from singular individuals (28) Maria, Sofia, Petros, and Christos (two women and two men, respectively) and two inanimate objects (setting aside the combinations of inanimates with animates).

Because feminine bears a female inference and gender is distributive, only the group consisting exclusively of women (Maria and Sofia) belongs in the feminine grouping. Because masculine bears an animate inference, every group belongs in the masculine grouping, including the group consisting of Maria and Sofia, though this group will be impacted by LC. That this is so is supported by (29), which indicates that the masculine plural noun is infelicitous with a group consisting exclusively of women. (See relatedly Sudo and Spathas’s discussion of their Principle of Gender Competition.)

With respect to the use of the neuter in (28), Lexical Complementarity rules out the use of the neuter in reference to human groups. Inanimate reference would be permissible, though (by hypothesis) inanimates are not in the extension of the predicate teacher;¹⁹ thus the noun meaning teacher can never appear in the neuter (unless there is some further derivational option such as diminutivization, on which, see Section 5.2).

The preceding discussion concerns plurals of asymmetric nouns, where the animate interpretation of the masculine is compatible with heterogeneous groups whereas the female interpretation of the feminine is compatible only with exclusively female groups. For symmetric nouns where the interpretation of masculine is male, the masculine extension will only include exclusively male groups; consequently, the effects of Lexical Complementarity will diverge from the masculine-as-animate environments. To see this, recall from (19), repeated in (30), that symmetric nouns do not allow the masculine with heterogeneous groups; these are presently analyzed with the male alloseme of masculine.

For such nouns, the gender inferences of feminine and neuter are the same, but for masculine, instead of animate, the meaning is instead male:

The behavior of the male inference in plurals is consistent with the distributivity requirement: every member of the group must be male in order for it to be a masculine group. Strikingly, however, Lexical Complementarity interacts with the male inference for plural groups in such a way that the result for neuter nouns is different from masculine-as-animate, as observed by A&A. In particular, gender-heterogeneous groups are not removed by LC in the neuter class.

The expectation is that, if a neuter plural form existed for the noun meaning ‘gentleman/lady’, it could be employed for reference to heterogeneous groups. However, like the English nouns for ‘lady’ and ‘gentleman’, there is no related corresponding gender-neutral term (except the awkward term ‘gentleperson’ we have been employing here). We assume that the absence of a neuter form of the noun with the root can be attributed to PF gender-licensing conditions à la Kramer 2015: in essence, there are formal demands on the noun that rule out the use of the neuter gender, independently of what its interpretation would be.

That being said, an account with both a male interpretation of the masculine, alongside a principle of Lexical Complementarity or a comparable competition-based principle, makes predictions for human-denoting nouns for which there is a neuter form, both in terms of their interpretation, and as we show, in an asymmetry between plurals and singulars. It is shown in the next section that this is borne out for a small set of kinship nouns in the language.

4.1 Pronominals

We assume nominals introduce gender consistently at the same locus n for both lexicals DPs and pronominals (e.g. Adamson & Šereikaitė 2019), as is consistent with the view that pronominals build nominal structure in parallel ways to lexical nominals, albeit with potentially less structure under some perspectives (see Cardinaletti & Starke 1999; Déchaine & Wiltschko 2002; Panagiotidis 2002; and much subsequent work). We assume further that, unlike lexical DPs, pronominals can lack roots altogether (e.g. Moskal 2015a; b).

In this case, within pronominals, there is no conditioning element for the specific male alloseme: there is no root and everything else is cyclically nonlocal. The expectation is therefore that only the default animate alloseme is available. As a consequence, reference to an adult human group is predicted to have to be either feminine (if the group consists exclusively of women) or masculine (for all men or a heterogeneous group). This is borne out, as evidenced by the pronominal clitic example in (44). If the male alloseme were also available, we would expect that neuter could also be used for mixed groups; however, this is not the case.

One crucial caveat here concerns pronominals with (potentially non-overt) linguistic antecedents. If there is a DP with a root that does select the male alloseme and has a neuter plural form, then a pronominal can refer to the same group with the neuter form, as in (45) with the pronominal clitic.

In such cases, we suggest that the root (and perhaps additional structure) is present but elided (and therefore unpronounced) for the pronominal.²⁶ This distinction between structural absence vs. elided material is reminiscent of the so-called deep vs. surface anaphora dichotomy (following Hankamer & Sag 1976; see Wurmbrand 2016 for a recent take on this distinction in the context of grammatical gender). For present purposes, it is important to emphasize that even with this optionality, the male alloseme is not readily available across pronominals; rather, it only appears when such a linguistic antecedent licenses ellipsis of a root.

Two further pieces of evidence support this contention. First, if the linguistic antecedent is not neuter plural, then the pronominal cannot be neuter plural, either (46). That is, if a plural noun employs the animate alloseme and refers to a mixed group with the masculine, pronominal reference to the same group in a subsequent clause cannot be with the neuter plural. This is expected under the elliptical view; there is no structure in the antecedent for (46) which permits the male alloseme, so the neuter plural is not available for pronominal reference.

Second, if the antecedent uses the neuter plural but pronominal reference makes use of a non-elided pronoun instead, then the masculine plural should also be available for a gender-heterogeneous group, given that the latter structure only has reference to the default animate interpretation of [m]. This is borne out for some speakers; (47) is a variant of (45), with the only difference being that the masculine plural pronominal appears in the second clause rather than the neuter plural.²⁷

In sum, pronominals without linguistic antecedents lack roots; they are therefore expected to select only the default animate alloseme of the masculine. This is consistent with the choice of a masculine pronoun – and not a neuter pronoun – to refer to a gender-mixed human group in an example like (44). In contrast, pronominals with linguistic antecedents may have elided roots, particularly the roots that appear in their antecedents, but they do not have to (a related idea is pursued by Sigurðsson 2019: 738). In the former case, the root can locally condition the allosemy of masculine gender, and as a consequence, masculine plural is used when the root combines with the animate alloseme (46), but neuter plural when it includes the male alloseme (45); alternatively, a root need not be elided, in which case the default animate is used, as in (47). This account thus captures the licensing of neuter gender for pronominals referring to mixed groups whose antecedents are exceptionally neuter plural, while also capturing the generalization that masculine plural pronouns are otherwise employed for gender-mixed groups.

4.2 Coordination resolution

We now turn to coordination resolution, which we argue provides evidence that contextual allosemy is locally restricted. In Greek, resolved agreement with mismatched nominals depends on the animacy of the conjuncts (Kazana 2011; Anagnostopoulou 2017; Adamson & Anagnostopoulou 2024; 2025) – the basic generalization is that for mismatched human nominals, resolved agreement on predicative adjectives is masculine (48) and for mismatched inanimates, resolution is neuter (49).

Various authors have suggested that the resolved values for person, gender and/or number are interpreted with respect to the entire coordinated phrase (at least for a subset of environments); see Corbett 1991; Sauerland 2003; Wechsler & Zlatić 2003; Wechsler 2008; Harbour 2020; Adamson & Anagnostopoulou 2025. For example, plural can be a property of the entire group denoted by the coordinated phrase but not necessarily of any of the individual conjuncts if both are singular. In structural terms, this can be implemented in terms of having interpretable features at or around the level of the conjunction phrase (&P). This is schematized by the tree in (50) for the example in (48): each conjunct has its own gender feature, and at the level of the &P, there is an interpreted gender feature, which is applicable to the entire group denoted by the coordinate phrase; see Adamson & Anagnostopoulou (2025) for one detailed analysis.

As Adamson & Anagnostopoulou (2025) argue (cf. Corbett 1991; Wechsler & Zlatić 2003; Wechsler 2008), the gender of &P is interpretively distributive, as it is for plurals. Thus a coordinated phrase in which each nominal denotes a woman will be feminine, a homogeneously male group will be masculine, and a heterogeneous group will also be masculine, as the animate ascription to each member of the group is appropriate. As Adamson & Anagnostopoulou (2025) show for Greek, this characterization also holds even when, for example, the conjunct is grammatically neuter but refers to a woman (see relatedly Wechsler & Zlatić 2003; Kazana 2011).

More controversially, Adamson & Anagnostopoulou (2025) argue that inanimate mismatch as in (49) yields neuter resolved agreement because neuter is interpreted essentially as inanimate; this stands in contrast to alternative theories that might attribute neuter agreement to a type of default insertion (e.g. Willer-Gold et al. 2016). As discussed by A&A, having a very general meaning of the neuter gender, in combination with Lexical Complementarity, correctly predicts that inanimate groups yield neuter resolved agreement. (On resolution with same-gender inanimates and various other patterns, see Adamson & Anagnostopoulou 2025 and references therein.)

How does our contextual allosemy account bear on the issue of coordination? The gender features at &P are not local to the roots in each conjunct. Therefore, only the elsewhere animate alloseme of the masculine should be available for interpretation on &P.

This is consistent with i) the choice of masculine instead of neuter for mixed-human groups as in (48) (all members are animate); and ii) the choice of neuter for inanimate groups as in (49). The prediction is that it should never be possible for coordination resolution to yield neuter plural for a heterogeneous human group.

A striking example indicating that this prediction is borne out comes from coordinating ‘cousin’-type nouns. As discussed by A&A, coordinating nominals with ‘cousin’ nouns in Greek yields masculine resolved agreement (51). This is expected under the current account: the male alloseme is available for the nominalizing heads occurring with each root – but with each root being deeply embedded within each conjunct – the roots are not visible for gender allosemy at &P, being too far away. Therefore, despite the existence of the neuter plural forms of ‘cousin’, coordination resolution with neuter agreement is unavailable in this context; instead, the masculine alloseme animate is selected at &P and correctly applies to the heterogeneous human group.²⁸ This contrast in locality between the features at n and the features at &P is depicted in (52).

To summarize, the male alloseme is only selected when it is local to particular roots. As we saw in 4.1, pronominals lacking roots do not have access to this alloseme, and resolved gender features at the level of the coordinate phrase can never be close enough to such roots for the male alloseme to be selected. For both pronominal elements and coordination resolution, only the default animate interpretation of the masculine is available, and the account correctly captures that it is used for heterogeneous groups.

4.3 Allosemy and cross-linguistic variation: the case of icelandic

A further benefit of the allosemy account is that it offers a way to account for points of cross-linguistic variation. Whereas for Greek, our analysis says that male is the more specific allosemic entry of masculine while animate is the default, we could imagine that another language could still have both allosemes but flips the relation between the two, such that animate is the specific alloseme while male is the default. Our account permits both types of languages, and we in fact suggest that Icelandic instantiates the second type. We emphasize, however, that our discussion of Icelandic is more tentative; we leave a more complete analysis of the Icelandic gender system to future research.

Icelandic is another Indo-European three-gendered language with a distinction between genders also labelled fem vs. masc vs. neut. Unlike in comparable expressions in Greek, it has been observed for Icelandic that neut is used for mixed groups in a variety of contexts (Corbett 1991; Wechsler & Zlatić 2003; Adamson & Anagnostopoulou 2025). As discussed by Adamson & Anagnostopoulou (2025), this happens for plural pronouns (53) referring to mixed groups, in greetings (54), and for coordination resolution (55). (For extensive discussion of Icelandic coordination resolution, see Thorvaldsdóttir 2019.)

Under the present system, this would fall out naturally if [m] is interpreted as male in these environments (cf. Wechsler 2008): mixed groups appear as members of the neut set and are not excluded by presence in the masculine or feminine sets, being neither exclusively male nor female, respectively.²⁹

However, there is also evidence that Icelandic has a fem/masc asymmetry like Greek with particular nouns (e.g. Grönberg 2002; examples from Adamson & Anagnostopoulou 2025). This can be seen with professional nouns like ‘actor’; in a gender-irrelevant context like the question in (56a), the masculine can be used neutrally in a question such that a felicitous response could include either a female or male individual. In contrast, the use of the feminine in the question cannot be felicitously responded to with reference to a male individual (57a) (see Jakobson 1984; Bobaljik & Zocca 2011; among others, on this diagnostic).

We suggest that the Icelandic pattern is best understood as the reverse relation between allosemes in Greek: whereas in Greek the male interpretation of the masculine only appears with particular roots (58), in Icelandic the animate interpretation only appears with particular roots (59).

This allosemy account captures the neuter plural facts for heterogeneous groups with pronouns and plurals. fem/masc asymmetries then arise in a “small corner” of the language with specific nouns that require the animate alloseme of [m]; this is parallel to the “small corner” of Greek for which neuter is used for mixed groups.

While our account correctly captures the set of differences described here, it is necessarily more tentative for the Icelandic system. As discussed by Grönberg (2002), there are in fact gender-neutral uses of the masculine with some quantificational elements, a property which is not readily captured under the current account (assuming such elements are rootless like pronominals). Sigurðsson (2019: fn. 24) in fact notes an instructive difference between masculine allir ‘all’ and neuter öll ‘all’: the first one can be used to refer to people in general (irrespective of gender), whereas the second one is used for a group of mixed gender. The interaction with quantification and/or genericity suggests that a more complex set of factors are at play; we leave the resolution of this issue to future work.³⁰

5.1 Children-denoting nouns

Consider the children-denoting nouns in Greek in (61), which are all neuter, as evident from e.g. their determiner agreement.

Such nouns are often described as having a hybrid status. For example, while neuter agreement is obligatory with determiners, adjectives, and other categories, there is some optionality with other gender-matching elements such as bound pronouns (62), as observed by Spathas (2010).

In a dual-feature system (like that of Smith 2015; 2017; 2021; Wurmbrand 2016; Adamson & Anagnostopoulou 2025; among others), children-denoting nouns could have uninterpreted neuter gender feature at the same time as having an interpreted feminine or masculine feature. We speculate instead that the neut gender of children-denoting nouns is not accidental, but rather, is also interpreted. In particular, we suggest that the neuter gender seen with children-denoting nouns reflects the intuition that children lack the animacy status of adults, being non-fully fledged humans in some relevant sense.³¹

There is cross-linguistic empirical basis for this idea. In particular, we observe that other comparable languages use neuter gender for nouns denoting children as well, suggesting that the association is not arbitrary. To name just a couple examples, the nouns meaning ‘child’ and ‘infant’ are neuter in Tamil (Dravidian), a language with feminine and masculine categories used for adult-denoting nouns (Corbett 2000: 60–61); and the nouns meaning ‘child’ and ‘girl’ are neuter in German (Kind and Mädchen, respectively), which also has a three-way gender distinction. The animacy distinction manifests itself in languages with two genders that generally distinguish humans vs. inanimates; for example, in Swedish, a language with two genders for nouns (common vs. neuter), the noun ‘child’ is neuter whereas the vast majority of human (adult) terms are common (see e.g. Holmes & Hinchliffe 2013).

There is also evidence from within a language suggesting that a referent’s (young) age can modulate preference for an animate vs. non-animate gender choice in pronominal reference. Braun & Haig (2010) report that with the German neuter noun Mädchen, survey results (302 participants) show that age modulates the preference of pronominal reference with the neuter or the feminine pronoun as in (63). This evidence could be taken to suggest that children at a younger age are easier to treat (linguistically) as non-animate in the relevant sense.

According to our allosemy account for Greek, neuter comes to be interpreted essentially either as non-animate or as neither male nor female, depending on which masculine alloseme is selected. Since the cross-linguistic pattern suggests even in languages lacking fem/masc gender categories for nouns that children are treated as non-animate, we will opt for the former characterization in Greek. This is stated more formally in (64).³²

Consider how (64) applies in the case of a noun like koritsi ‘girl’, which is neuter. This root bears a childhood inference (in a sense not spelled out here), so it categorizes individuals as non-animate. Assume that all the nouns in (61) are not listed for the male alloseme of the masculine; instead, the default animate alloseme is selected.³³ The neuter gender with nouns like koritsi then correctly reflects that the referents are treated as if they do not belong to the class of animates.

It may seem counterintuitive that a noun like koritsi ‘girl’ has neuter gender as a reflection of being linguistically non-animate while simultaneously having lexical semantics that reflect female status. But while being female entails being animate in the real-world sense of being alive (or perhaps having been alive), it does not entail being animate for the purposes of the linguistic opposition of animate vs. non-animate (with perhaps a better term being “rational”). For example, animals can be female but non-animate, and this comports with the fact that their grammatical gender is not (obligatorily) interpreted in Greek: hence kuneli ‘rabbit’ is neuter but can refer to a rabbit that happens to be female. (See Spathas & Sudo 2020 for extensive discussion on such nouns in Greek.) The female lexical semantics of koritsi thus need not clash with linguistic non-animacy. This is unlike the female inference that comes from a feminine feature; as stated above, we assume that the interpreted property of female from [f] implies animate. Thus the constraint in (64) derives the incompatibility of koritsi with feminine agreement (*i koritsi ‘the.f.sg girl’): the animate interpretation of the feminine feature is incompatible with the non-animate requirement of the constraint.

One might also expect that our Binary Assumption stated in (35) requires a noun like koritsi to be feminine. However, this assumption is simply not applicable because the constraint in (64) means that child-denoting nouns are not animate, and are therefore not subject to this requirement.

Because (64) is restricted to the formation of nouns with childhood roots, it has no bearing on gendered reference to children with pronominals (lacking a child-root) or other nominals, which can indeed refer to individuals who happen to be children, treating them as animate individuals.³⁴ Hence the bound pronoun in (62) may be feminine or neuter, being a distinct nominal from the quantified nominal kathe koritsi ‘every girl’.³⁵

What if a root can have a childhood reference but does not have to? The nouns in (61) all seem intrinsically children-oriented, but some kinship nouns like egonos ‘grandchild’ have a childhood component to them but can readily refer to an adult that bears the relevant kinship relation. We speculate that this situation characterizes some set of kinship nouns that select for the male alloseme of the masculine gender. For such nouns, the childhood component of the root is optional: when it is present, the only choice is for neuter gender, but when it is absent, the Binary Assumption applies and individuals are categorized as male or female, and hence the noun can be masculine or feminine, respectively.

Recall from Section 3 that many kinship nouns select the male alloseme. As alluded to above, some of these nouns are acceptable with the neuter singular. Our contention is that it is precisely the case when these nouns can have childhood inferences, as for ‘grandchild’ (65) and for some speakers, ‘niece/nephew/nibling’ (66).

In summary of this more speculative subsection, children-denoting nouns can conform to the Neuter Hypothesis from (60), given the additional constraint (64) placed on children-denoting nouns. In addition to capturing the neuter status of children denoting nouns like koritsi ‘girl’, the account can correctly capture which neuter kinship nouns can be singular, in contrast to the nouns discussed in Section 3, which resist (unmarked uses of) the singular.

5.2 Diminutives

We suggest that the allosemy account can further be extended to the interpretation of neuter diminutives. Adamson & Anagnostopoulou (2025) observe that human-denoting nouns occurring with a diminutive suffix can often either retain the gender of their base or can be neuter, with a systematic difference in interpretation. For example, with human nouns like andras ‘man’, the gender-matching diminutive has an endearing interpretation while the neuter diminutive is pejorative, as in (67).

Adamson & Anagnostopoulou (2025) suggest that the pejorative interpretation arises as a result of a dehumanization effect that comes from neuter being interpreted as “inanimate” in combination with a human-denoting noun. Following Kramer (2015), we take diminutive suffixes that change the gender to be instances of nominalizing heads, which bear their own gender features (see relatedly Wiltschko 2006; Wiltschko & Steriopolo 2007; Steriopolo 2008; Gouskova & Bobaljik 2022; among many others). We assume that the diminutive n adjoins to the existing nominal structure; hence neuter diminutivization is a form of denominal nominalization. Following Kramer and others, we assume that diminutivization of a noun corresponds to this type of denominal structure for gender-changing diminutivization but not for the gender-matching option; both are illustrated in (68). We follow Kramer (2015); Adamson & Šereikaitė (2019); among others in assuming that the gender features of the outermost nominalizing head are the ones applicable for agreement.

We assume that the gender features on the diminutive n head are also interpreted. Because they are separated from the root by the intervening n head that combines with the root, the diminutive heads are nonlocal for the purposes of root-based allosemy of gender. Therefore, the only interpretation is with the default alloseme animate. For the gender-matching case, there is only one interpreted gender feature since the diminutive head does not combine with its own set of gender features. For the neuter case, there will be a clash between the animacy inference from the inner n head and the inanimate type of meaning from the outer n. While this may appear to a present a contradiction, we suggest that the addition of the diminutive semantics allows for a type of coercive interpretation, whereby the animate referent is treated as if it is inanimate, giving rise to the pejorative dehumanization effect (though we do not spell this out more explicitly).

This account leads to predictions for other diminutivized nouns. Recall from (64) that children-denoting nouns are considered non-animate; because of this constraint, there should be no pejorative effect for neuter diminutivization when it applies to children-denoting nouns. Adamson & Anagnostopoulou (2025) in fact note that there is no pejorative effect for child-denoting nouns such as ped-aki ‘child-dim’, which remain neuter when they appear as diminutives.

The other prediction concerns inanimate nouns, which should not bear a pejorative interpretation when the neuter diminutive n combines with it. Grammatically neuter examples such as kapel-aki ‘little hat’ and xer-aki ‘little hand’ support this claim. This can also be extended to animal nouns treated as non-rationals (see Spathas & Sudo 2020 for extensive discussion of such nouns); as also noted by Adamson & Anagnostopoulou (2025), gat-aki ‘little cat’ and skil-aki ‘little dog’ (respectively formed from feminine and masculine nouns) are not pejorative.

5.3 Other neuter human-denoting nouns

There are also other nouns in Greek that are necessarily neuter, which can be – and are often – used to refer to humans. A list of well-known and fairly frequent nouns with this property can be found in (69).

We do not have a clear answer for how neuter comes to be interpreted on these nouns. One intuition to pursue is that what is most relevant about these roots for their neuter status is that they do not require reference to humans. Atomo is used for individual people, but it is also used for inanimates to mean ‘atom’; the same is true for prosopo ‘character/person’ and melos ‘member’.

Arguably the most problematic example for this intuition is thima ‘victim’; perhaps its neuter status may suggest something was previously a living human or animate being, but might not be anymore (for example, if they are a murder victim). In any case, we can state the intuition somewhat more formally as in (70).

The effect of (70) is that such nouns are ineligible to be masculine or feminine because the gender semantics would entail being animate.

In relation this constraint, we would like to highlight that nouns with these meanings in other languages are also known for not having gender interpretations associated with animates, which seems not to be coincidental. For example, in Italian, vittima ‘victim’ is feminine (as is true of its relatives in other Romance languages).³⁶