This paper investigates a conflict in the realm of agreement that pertains to mismatches between formal and semantic φ-features on pronouns and their manifestation in agreement with different types of predicates. Formal (henceforth: grammatical) features determine the pronoun’s form (e.g. person and number), while semantic (henceforth: natural) features are not visible in the pronoun’s form, but nevertheless appear on predicates that agree with this pronoun (e.g. gender on 1^st and 2^nd person pronouns). Mismatches in agreement occur when the natural features do not correspond to the grammatical ones with respect to their values, for instance when a formally plural pronoun is used to denote a single entity.

When an agreement controller contains conflicting natural and grammatical features, the predicates that agree with it can differ with respect to the features they reflect. Systematic crosslinguistic differences have been observed in the previous literature, with some of the first observations coming from Comrie’s (1975) study on the curious behaviour of the pronoun for polite address, or the honorific pronoun. In languages such as Slavic, some from the Romance family (French, Italian, Romanian), as well as Modern Greek, the second person plural pronoun is used when politely addressing a single person. Some predicates in these languages, such as finite verbs, always show plural agreement, i.e. they agree with the grammatical features of the second person plural pronoun. However, other predicates, such as participles and adjectives, can show either plural agreement, consistent with the finite verbs and auxiliaries, or they can show the remarkable singular agreement and agreement in the natural gender of the referent, which has traditionally been termed semantic agreement.

Based on their agreement patterns, crosslinguistically different types of predicates have been shown to align according to the following hierarchy:

This paper will review the evidence for the existence of the hierarchy effects based on data from number agreement in the languages mentioned above. We will see that the honorific pronoun controls [π:2, #:pl]1 agreement on finite verbs cross-linguistically. However, the languages will differ in agreement on participles and predicate adjectives; in some languages, they may, or must, control singular and gender-dependent agreement. This paper will concentrate on the verbal predicate types on the Predicate Hierarchy by exploring the causes of variation on finite verbs and participles.2

The formal analysis will rest upon the assumption from the previous literature that the honorific pronoun formally encodes both the grammatical features (plural number and person) and the natural features (gender and singular number) (cf. Despić 2017). These features will be argued to have a hierarchically organised complex internal structure, formally modelled in the spirit of Harley and Ritter’s (2002) feature geometries. Moreover, the features will be argued to occupy different positions in the complex structure of the pronominal DP, grammatical features being higher than the natural ones. This will set the stage for the proposal whose central claim is that predicates differ with respect to the φ-features they probe for, and the order in which this probing applies. In particular, I will propose that finite verbs agree in person and number, in that order. Unlike them, participles and adjectives perform number and gender agreement. Variation among and within languages emerges from the order of these operations (strict or underspecified). If gender agreement is carried out first, the natural gender and number features will be copied from the lower parts of the DP. If number agreement is carried out first, the grammatical number features will be copied from the higher phrases in the DP, which will lead to an intervention effect for natural gender and number agreement.

The paper is structured as follows. Section 2 introduces the empirical puzzle with an overview of different types of languages and their respective predicate agreement patterns. The presentation of the theoretical background occupies Section 3. The main theoretical assumptions for the system to be developed in this paper are outlined in Section 4, after which the patterns of agreement with finite verbs and participles are presented in Sections 5.1 and 5.2. Section 6 discusses some implications of the analysis for regular agreement patterns and similar instances of mixed agreement in other contexts and Section 7 summarizes and concludes.3

Mixed agreement effects with honorific pronouns in Slavic, Romance (French, Romanian, Italian dialects), Greek and Icelandic have been documented and analysed in the works of Comrie (1975); Corbett (1983); Wechsler (2011); Wechsler & Hahm (2011); Arsenijević (2014); Despić (2017), among others. Slavic languages, the focus of this paper, present a convenient study ground due to the extensive range of variation that can be observed within a single language group. Most Slavic languages use the second person plural pronoun Vy/Wy/Vi/Vie for polite address. As this pronoun is grammatically specified for [π:2] and [#:pl] features, these features are expected to be reflected in agreement. However, different predicates in different languages may allow singular agreement (when addressing an atomic individual), accompanied by agreement in natural gender. Based on the patterns allowed on their predicates, the languages above lend themselves to a three-way classification.

The first group comprises languages in which finite verbs show [π:2, #:pl], while the participle, predicate adjective and predicate noun show agreement in singular number and the natural gender of the pronoun’s referent. Czech is one of the typical representatives of this pattern (2),4 together with French, some Italian dialects, Romanian and Modern Greek.5

The second group comprises languages such as Ukrainian, Belorussian (cf. Corbett 1983: 51) and Russian, whose finite verbs consistently show [π:2, #:pl] agreement as in Group 1 (3a), but unlike in Group 1, the participle now agrees in plural number (3).

Predicate adjectives pattern with the ones in Czech, showing semantic singular agreement (4).7

Before proceeding to the final group, a brief comment on Russian is in order. Russian owes its Group-2 status to the behaviour of its so-called long-form (LF) adjectives. While the LF adjectives always agree in natural gender and singular number, short-form (SF) adjectives admit only plural agreement, just like participles.8 There is a general consensus in the literature that LF predicate adjectives are more “adjectival” while SF predicate adjectives are more “verbal” in nature (Babby 1975; Bailyn 1994; 2012; Pereltsvaig 2007; Geist 2010; Borik 2014: 139). Following previous literature that treats the syntactic structure of SF adjectives as essentially verbal, I assume that they agree in the same way that verbs do, while LF adjectives are true adjectival predicates in the narrow sense.9

Languages that belong to the final group (Slovak, Upper and Lower Sorbian, Slovenian, Bulgarian, Macedonian) optionally permit formal or semantic agreement either on the participle, or on the predicate adjective, or on both of them. In these languages it is often the case that the prescriptive rules of the standard language require the usage of a particular form, which the usage of some speakers diverges from. For instance, in Bulgarian, even though the prescribed patterns mirror those from Group 2 languages, singular agreement can be found on participles as well (5).10 The same situation can be observed in Slovak.

Unlike Bulgarian and Slovak, Macedonian is strict about its participles, which resolutely agree in the plural, while the predicate adjectives keep their right to disagree from the norm by allowing plural agreement.

Bosnian/Croatian/Serbian (BCS) is tentatively placed it in this group due to variation present in some dialects, notably those spoken in the west of Croatia, around Zagreb, Rijeka or Gorski Kotar (see Comrie 1975; Corbett 1983; 2006; Stevanović 1989; Wechsler 2011; Wechsler & Hahm 2011; Despić 2017; Puškar 2017 for more detail on the possible variation).11 Yet, plural is the standard option, and the only one acceptable to most of the native speakers consulted.

To sum up, Table 1 presents the agreement options for different predicates in the languages mentioned above. What the languages have in common is constant grammatical plural agreement on the finite verb. They diverge in agreement on the participles – while the first group admits only agreement in natural gender and singular number, and the second group opts for the grammatical plural agreement, the third group optionally allows both possibilities.

The patterns from Table 1 raise several important questions for any theory of agreement, such as whether gender is encoded on local-person pronouns, and if the answer is yes, how exactly it is encoded if it is not visible in the pronouns’ morphology. Furthermore, if gender features do in fact exist on pronouns, does this mean that in Czech-type languages these are natural gender features, but in standard-BCS-type languages the honorific pronoun carries inherent grammatical masculine gender and plural number? Or is the masculine feature the result of default agreement? Finally, what principles of agreement force natural gender and number agreement on participles in Group 1, while restricting Group 2 participles only to formal agreement? Finding answers to these questions is going to be the task of the sections to follow.

Even in languages which do not accept natural gender agreement with the honorific pronoun, such as standard BCS, elements that stand in an agreement relation with local-person pronouns bear gender and number morphology that reflects the properties of the pronoun’s referent (8). This poses the crucial question whether gender belongs to the pronoun’s feature inventory.

A negative answer to this question is advocated by unification-based approaches, as proposed by Pollard & Sag (1994) or Wechsler (2011) and Wechsler & Hahm (2011). The latter two accounts argue that the agreement controller in (8) is underspecified for gender features, and its interpretation as male or female comes through interpreting the morphology of the agreement target.

The proposal put forward by Wechsler (2011); Wechsler & Hahm (2011) focuses particularly on honorific agreement. They make a distinction between two sets of features: concord features, which denote purely grammatical properties of an element (case, number and gender) and index features (person, number and gender), more closely tied to semantics and related to an element’s referential index, e.g. whether a noun denotes a male or female entity (cf. Wechsler & Zlatić 2003). A pronoun is argued to contain both sets of features. Concord features participate in agreement with targets which themselves also express gender and number (but not person) features, e.g. adjectives, while index features participate in agreement between the pronoun and the finite verb, which agrees in person and number. The main difference between Czech-type languages and BCS-type languages lies in the feature specification of the honorific pronoun, such that in the latter this pronoun contains formal masculine plural features. Since the honorific pronoun in standard BCS controls masculine plural agreement, it is assumed to have both masculine plural [ɣ:M, #:pl] concord features and 2^nd person masculine plural [π:2, ɣ:M, #:pl] index features, copied by adjectives and finite verbs, respectively. In contrast, the honorific pronoun in Group 1 languages has the same index features, but it lacks concord features altogether. Agreement is regulated by the Agreement Marking Principle (Wechsler 2011: 1009) which demands that if a pronoun lacks φ-features, the features of the agreement target must be interpreted such that it specifies the denotation of the controller. This principle ensures that the pronoun lacking concord features receives the correct interpretation.

This account faces certain challenges in accounting for variation across the Slavic family due to the lack of discussion of participial agreement. Given that participles agree in number and gender, they can be assumed to agree in concord features like adjectives. However in languages that allow the combination of a plural participle and a singular predicate adjective, there emerges a conflict in the representation of the features of the pronoun. The plural agreement on the participle in languages like Bulgarian (9) requires the presence of a [#:pl] concord feature on the honorific pronoun, whereas the singular agreement on the predicate adjective would require either a complete absence of concord features, or the contradictory [#:sg] concord feature. Patterns of this type are not discussed in these accounts.12

A further challenge is posed by patterns identified in BCS by Arsenijević (2014) and Despić (2017). Even though the honorific pronoun is assumed to contain two masculine features (concord and index), it is nevertheless capable of controlling natural gender agreement in certain environments. According to Despić (2017), the evidence comes from agreement in coordination. In BCS, singular conjoined nouns of different gender control default masculine plural agreement (10a). Two feminine nouns, however, control feminine plural agreement (10b).

Assuming the presence of two masculine features, the honorific pronoun should be expected to pattern with masculine nouns and force masculine plural agreement when coordinated with another nominal. Despić (2017) shows that this is not borne out. If a female-referring honorific pronoun is coordinated with another feminine noun, it can control feminine agreement (11), just like in coordination of two feminine nouns in (10b).13 This at the very least indicates that there should be a [ɣ:F] feature encoded on the pronoun.14

Moreover, the honorific pronoun can control natural gender and number agreement on non-nominative adjectives, such as the case with the secondary predicate in (12). In addition to that, and contrary to the claims of Wechsler (2011) and Wechsler & Hahm (2011), Arsenijević (2014) demonstrates that semantic agreement is available in the nominative as well, as long as the adjective is non-restrictive. According to Arsenijević (2014), the possibility of semantic agreement depends on whether the adjective is used restrictively and whether the honorific pronoun is used as a strong pronoun (13) or a clitic (12), and not on whether or not it contains concord gender features. As my account below will mostly be concerned with verbal agreement, I refer the reader to Arsenijević (2014) for further detail on adjectives.

If the honorific pronoun does contain natural gender and number features in addition to the grammatical 2^nd person plural, the final issue left to solve is how hybrid agreement comes about. I will follow Despić (2017) who argues for the “same pronoun hypothesis”, whereby the features encoded by the pronoun should be the same in different types of languages and the difference lies in the way in which they are copied by agreement targets. His proposal holds that each of the pronouns below encodes formal person and number features, as well as semantic gender and number, whereby the formal features [π:2, #:pl] do not match the semantic ones [ɣ:F/M, #:sg]. An agreement target that can optionally show either formal or semantic features must agree either in fully formal (grammatical) (14a) or fully semantic (natural) features of the hybrid controller (14b–c). There is never a situation (14d) where the participle agrees with the honorific pronoun such that it copies grammatical number (plural) and the natural gender (feminine/masculine).

According to Despić (2017), the predicate in standard-BCS-type languages will copy the formal plural number, which restricts agreement only to formal features. A default masculine exponent is inserted in the absence of a gender feature to satisfy the well-formedness requirements of a predicate. On the other hand, a predicate in a language like Czech will copy the semantic number, which will further restrict it only to semantic features, forcing it to copy the semantic gender of the pronoun as well.

What is left unsolved in this account is how semantic and formal number features are represented and in what way they are exactly encoded on the pronoun. Moreover, what forces agreement targets in Czech, as opposed to ones in BCS, to copy exclusively semantic features (i.e. the exact nature of the agreement restrictions) is another matter left without an explicit proposal in this account. I will propose a derivational account that will address exactly these issues.

The following sections introduce the theoretical tools for the subsequent analysis. Recall that the three main questions addressed by this paper are (i) how pronouns are structured, (ii) how φ-features are encoded on them and (iii) what kind of an agreement mechanism is responsible for the patterns above. I will propose that the pronominal DP consists of several sub-phrases, each of them hosting different φ-features. The lower phrases will encode natural gender and number bundled together, while the higher phrases will encode person and grammatical number. Finally, I will show how separate Agree operations, whose ordering can vary, derive the agreement patterns abiding by strict locality principles.

This section aims to illustrate how agreement patterns of regular pronouns follow trivially from the system developed above. After the inspection of local-person pronouns and the extension to third person pronouns, the section ends with a discussion on the implications of the analysis for further mismatch phenomena, such as unagreement.

Recall that in the languages discussed in this paper local-person pronouns normally control natural gender and number agreement. For instance in BCS such agreement can be found both in the singular and in the plural:

Under the analysis proposed in this paper, the plural pronouns in (30c–d) will have the structure equivalent to that of the honorific pronoun in (18). Their NumP carries the formal plural number [#:pl], the neighbouring PersP houses person features, while the natural gender and plural number are bundled together under the IND node on the nP. Even though the double encoding of plural number may seem redundant, I argue that this assumption is welcome, and in fact necessary, to explain an additional type of agreement mismatch found with regular local-person pronouns. To illustrate, both (31a), with the natural feminine plural agreement, and (31b), with the formal plural and default (failed) gender agreement can be used if the pronoun refers to an all-female group. This strongly suggests that gender agreement with local-person pronouns, just like with the honorific pronoun, is not immune to failure in the plural.

This situation can be modelled straightforwardly under the current account, and in fact, it provides further support for it. The pattern in (31a) will be derived when Gender Agree is given preference over Number Agree on the participle, parallel to (27) above. If the order of operations is reversed, Number Agree will copy the higher formal [#:pl] feature, making the copying of gender illicit due to the CAD, yielding (31b) in the manner derived in (29).

On the other hand, the only option in the singular is natural gender and number agreement. The lack of optionality can be explained by the lack of its trigger, i.e. the [#:pl] number feature. Following Nevins (2011); Pesetsky (2013); Despić (2017), I assume that being formally singular entails having no formal number features. This can be modelled by assuming that the NumP is present only if it hosts the [#:pl] feature and absent otherwise. Aside from that, the structure of local-person pronouns remains unchanged; natural gender and singular number are encoded on the n head22 and person features are specified above it. Consider (32) as a derivation of (30a). Since the pronoun denotes a single feminine speaker, its IND features on the nP include the natural feminine gender and singular number. Under any order of operations, either Gender Agree or Number Agree will be able to target this feature bundle. This ensures that, without the NumP to trigger the CAD, the end result is always strictly natural gender and number agreement, as indicated by the features copied by the Part head in (32). On the other hand, its person features will always be the Goal for finite verbs, yielding 1^st person singular agreement.

Unlike the local-person pronouns, 3^rd person pronouns morphologically distinguish between masculine (on-∅ ‘he’), feminine (on-a ‘she’) and neuter form (on-o ‘it’), each of them controlling the concomitant gender agreement. The gender that they overtly show seems to be grammatical, rather than natural, since a feminine pronoun, for instance, can be used to refer both to an animate and to an inanimate referent.23 I propose that this grammatical gender differs from the hitherto discussed natural gender in two aspects – structural and positional. As for its structure, recall from Section 4.1 that grammatical gender was argued to be simpler than the natural one in that it lacks the additional [Animate] feature. Additionally, I argue that there is a difference in their locus in the nominal spine. While natural gender belongs to the realm of nP, grammatical gender is located on the Gen(der)P, above the NumP. Combined with the assumption that PersP is present only if it carries a [Participant] feature and otherwise absent (as proposed in Section 4.1), the structure of 3^rd person pronouns can be represented as in (33).

The structure in (33) has several beneficial consequences. First, the grammatical gender feature can be taken to be responsible for the realisation of gender morphology, yielding the -∅, -a and -o suffixes for [M], [F], and [N], respectively, as well as their plural counterparts (by fusing with the Num head). Moreover, this feature is responsible for gender agreement triggered by the third person pronouns. Finally, the structure in (33) has the benefit of equating the structure of 3^rd person pronouns with the structure of other nouns in BCS. The assumption on two possible loci of gender features enables the modelling of mixed agreement with hybrid nouns that involve a mismatch between their grammatical and natural gender values. As a detailed discussion of such patterns is outside the scope of the present paper, I refer the reader to Puškar (2017; 2018) for further detail.

Putting together the possible building blocks of the DP, the question arises as to whether PersP and a GenP can co-occur, considering that GenP is absent on local-person pronouns, while PersP is absent on 3^rd person pronouns and regular nouns. One possible candidate for the co-occurrence of the two are nouns that trigger the so-called unagreement patterns (cf. Mancini et al. 2011; Ackema & Neeleman 2013; Höhn 2016). Some languages such as Spanish allow for 1^st or 2^nd person agreement with regular nouns, as in (34a) and (34b), respectively.

Under the account above, the patterns in (34) can be explained by a simultaneous presence of a GenP that carries the grammatical feminine gender and a PersP that carries local-person features. The feminine gender agreement on the adjective would result from copying the features of the GenP (or nP), while the local person agreement on the finite verb would come about as a result of copying of features from PersP (cf. Höhn 2016).

In sum, distributing φ-features across different projections on the DP-spine, as well as postulating multiple gender and number feature loci gives this account a possibility to explain regular agreement patterns as well as the potential to explain not only the mixed patterns in pronominal agreement, but also other instances of mixed agreement in different languages.

This paper has addressed patterns of formal and semantic agreement on finite verbs and participles collected by Comrie (1975); Corbett (1983); Wechsler (2011), whereby the agreement controller (the polite 2^nd person pronoun) contains conflicting grammatical and natural feature values. What these languages have in common is the uniformly formal plural agreement on finite verbs. Where they differ is the type of agreement they show on participles (grammatical masculine plural agreement or natural singular agreement).

The account proposed above derives the effects of mixed agreement as a narrow-syntactic process which involves precise loci of φ-features in the DP structure, feature-geometric structure of φ-features, separate agreement for individual features and variable ordering of Agree operations. I have argued for a unified structure of honorific pronouns across different languages, which encode both grammatical features (person and number) and natural gender and number. What derives mismatches across languages (and within a single one) is the specific functioning of Agree. Finite verbs will only probe for person and number (in that order), while participles agree for number and gender under a strict, or underspecified order, depending on a language. The formal plural agreement pattern will result from probing for number first, and thereby blocking the access to natural gender due to a locality restriction on agreement (the Condition on Agree Domains (19)), while probing for gender first will lead to semantic agreement by copying the entire geometry of natural gender and number features in one fell swoop.

The current approach enables capturing crosslinguistic variation by letting languages choose whether they will allow both orders of Agree operations on all, some, or none of their probes. Languages can choose to maintain one order of operations throughout, or perhaps to vary the order of Agree operations depending on the type of the probe. As a result, a long-standing observation such as the Predicate Hierarchy need not be defined as a primitive of grammar, but can instead be derived as a result of the opaque interactions of Agree operations.