2.1 Alienable possession constructions

Alienable possessors in Jarawara precede the possessed entity and are formally identified by the possessive marker kaa. The basic structure of inalienable possession phrases is shown in (8).

Dixon (2004: 35) describes kaa as a postposition that constitutes a distinct phonological word.⁹ A series of singular possessive pronouns involve vestiges of kaa (historically *ka); the first and second person singular possessive pronouns are proposed to have arisen through univerbation of the pronouns o- and ti- with *ka, resulting in oko jobe ‘my house’, tika jobe ‘your house’ (Dixon 2004: 47). There is also a third person singular pronoun hinaka, but no independent form hina (Dixon 2004: 77, 296).

Predicate agreement with alienably possessed phrases follows a pattern consistent with principles of ‘canonical’ headedness within the NP. As first illustrated in (2), the gender of the head of the subject NP determines the gender on the predicate: the verb agrees with the gender value of the possessed noun, not the possessor. Similarly, in (9a), where the possessor Manira is feminine and the head of the phrase jomee ‘dog’ is masculine, the auxiliary verb agrees with the features of the head of the S argument, not those of the possessor. The gender of a possessive NP is determined in exactly the same way when the possessor of an alienable entity is pronominal, as in (9b).

When an alienably possessed noun is a controller, grammatical properties of the possessor play no role in predicate agreement. If it did, one would expect verbal agreement to take a feminine form in (9). Rather, the pattern of agreement we find here is the expected one based on general assumptions about canonical headedness and ‘feature percolation’ (Lieber 1980; Gazdar & Pullum 1982; Pollard & Sag 1987; 1994; Norris 2014; Myler 2016).

2.2 Inalienable possession constructions

A different pattern prevails with inalienably possessed nouns. Jarawara has around 175 inanimate nouns which are inherently possessed, and therefore always understood to be relational (Dixon 2004: 310).¹⁰ Dixon (2000; 2004) refers to this class as possessed nouns (PNs). Items belonging to this class include body parts, parts of plants, indicators of part-whole relations, spatial orientations and a range of other meanings related to sensory input and physical containment (Dixon 2004: 311).¹¹ The possessor always precedes the PN, as in (10).

Based on Dixon’s (2004) description, I have been able to determine that PNs belong to at least seven inflectional classes (see §3) showing different patterns of syncretism. They vary in form based on the person, number and (in some cases) gender features of their possessor.¹² Around 65 PNs (belonging to four of the seven inflection classes) have distinct masculine and feminine agreement forms in the third person singular (Dixon 2004: 310). These include: mano/mani ‘arm’, teme/tame ‘foot’, and maho/mahi ‘smell’. The remaining 110 or so PNs have syncretic masculine and feminine forms (see §3.1). An example paradigm for mano/mani ‘arm’, which belongs to what we will call inflection Class I, is provided in Table 1.

Table 1

PN paradigm for mano ‘arm’ (based on data in Dixon 2004: 315).

In Class I, the first and second person singular forms are distinguished internal to the phonological word. Independent pronominal possessors (in parentheses) are required if the possessor is non-singular. The third person singular forms typically co-occur with a nominal possessor (as shown below), but this is not necessarily so if the antecedent is retrievable from discourse (see §4). When they do co-occur, the nominal possessor and possessed entity must share the same gender value, as in (10). For instance, the inalienably possessed noun mano/mani ‘arm’ has third person singular masculine and feminine agreement forms that co-occur with masculine and feminine nominal possessors respectively, as first shown in (6). Around 35 PNs have associated ‘free nouns’ (i.e. non-relational, etymologically related forms) that do have inherent gender, but these belong to a separate nominal category and exhibit some idiosyncratic relationships in form and meaning to the PNs (see Dixon 2004: 310–4, Adamson 2024: 239–40). While the attested free nouns associated with PNs nearly always control feminine agreement, there is one exception: the masculine free noun atiwa (m) ‘thorn’. This corresponds to the PN atine/atine ‘thorn’, which has syncretic masculine and feminine agreement forms (Dixon 2004: 323). If we assume (as Dixon 2004 and Adamson 2024 do) that the gender of the PN matches the gender of an associated free noun, PNs can be masculine or feminine. But most PNs do not have an associated free noun. There is no independent way of verifying what the gender of the remaining 140 or so PNs would be. We could indeed extrapolate that they are likely to be feminine, based on the predominant pattern observed where a free noun exists, but it could equally be the case that PNs without ‘free nouns’ are all masculine. Given that the inherent gender value of a PN is never manifested through agreement, I propose that PNs simply lack an inherent gender value. See Adamson (2024) for an analysis in which they have a featurally unmarked gender value, namely feminine.¹³

Most inalienably possessed nouns in Jarawara do not have distinct masculine and feminine agreement stems, but rather have a single syncretic form, e.g. tati ‘head’ and kone ‘hair’ (see §3). In examples like (11) the masculine possessor Kimi appears to control agreement on the predicate (the position argued for by Dixon 2000; 2004 and Adamson 2024) because it is the only element within the noun phrase that clearly has a masculine gender value.

In §4 I propose that examples like (11) merely resemble predicate-possessor agreement. Rather, the NP internal agreement between possessors and relational nouns involves feature sharing, which results in the whole NP having the features of the possessor.

Turning now to pronominal inalienable possessors, Dixon’s (2000; 2004) analyses propose that the third person non-singular pronoun mee occurs with a feminine form of the PN, as in (12a), while first and second person pronouns take masculine PNs, as in (12b–c). Rather than simply being formally syncretic with gender-valued forms, each of the PNs in (12) is analysed by Dixon as having a gender value; in (12a) the PN is considered to be feminine, while in (12b–c) it is masculine.

This is a crucial part of Dixon’s logic: since predicate agreement in these structures is always feminine despite the presence of ‘masculine PNs’, the pronouns, which always control feminine gender when occurring as predicate arguments (Dixon 2000: 488), must be the controllers of agreement on the verb.¹⁴ Similar arguments are developed based on modifier agreement. When an inalienably possessed noun has a first or second person non-singular possessor, adjective agreement is feminine, even though the PN is in the same form as when it has a masculine controller.

Dixon concludes that since it is normally the head of an NP which determines its gender, the inalienable possessor must be the head of the NP (Dixon 2000: 490) and therefore determines the category of its phrase as a whole (Dixon 2000: 487, 489).

2.3 Unresolved issues in Dixon’s analysis

A close inspection of Dixon’s arguments reveals that his analysis of possessive agreement in Jarawara is partly dependent on the view that there is a one-to-one correspondence of form and meaning throughout a morphological system. That is to say, form meaning correspondences are consistent and reflect ‘natural’ classes. This view of morphology leads to several unwarranted assumptions and unresolved issues which indicate further scrutiny of Jarawara possession constructions is justified. I call these: the feature problem, the adjective problem, and the recursion problem.

The feature problem: Dixon’s (2000; 2004) analysis of PNs rests on the assumption that a single form is associated with a single feature profile. For instance, that the stem mano is masculine and that mani is feminine. He assumes that PNs that are identical in form to those used in agreement with masculine or feminine nominal possessors always have those feature values, regardless of their distribution. No justification other than formal similarity is given by Dixon to support the view that all PNs have a masculine or feminine gender value. I argue throughout the paper, and especially in §3 and §5.3, that a morphomic analysis is essential for understanding Jarawara possessor agreement.

The adjective problem: In Dixon’s analysis, agreeing adjectives that form a constituent with the noun they modify are said to be controlled by a possessor rather than the possessed entity, as in (13). This is highly unusual from a cross-linguistic perspective (but see Nichols 2017: 393 for other possible counter examples). His account of the data rests on the assumption that pronouns have a feminine gender value in Jarawara. An alternative proposal consistent with morphological defaults and the feature sharing approach adopted here is presented in §5.1.

The recursion problem: Although Dixon proposes that the possessor is the head of inalienable possession phrases for the purposes of clause-level agreement and adjectival agreement, he is vague about the principles that account for the properties of gender agreement in possessive phrases containing more than one PN. This, in combination with his assumptions about features (the feature problem), leads to issues in accounting for recursive structures like those in (14). If omano is masculine (as claimed by Dixon 2000: 489), and the PN determines the gender of the possessive phrase for NP internal agreement, why does baki occur in its feminine form in (14a)? Similarly, if otaa controls feminine agreement, and PNs exhibit agreement with their possessors, why does mano ‘arm’ occur in a masculine form in (14b)?

This problem is explicitly identified as unresolved by Dixon (2000: 505; 2004: 316), but a solution is proposed in §5.3. I argue that all of these apparent problems can be explained through a careful reanalysis of the morphology and syntax of possessive constructions that considers (i) key principles of morphology concerning paradigmatic structure, morphomic distributions and morphological defaults, and (ii) feature sharing through agreement in inalienable possession constructions.

3.1 Inflectional classes of possessed nouns

Inalienably possessed nouns in Jarawara are inflected for the person (1 vs. 2 vs. 3) and number (sg vs. nsg) of their possessor. Third person singular forms are further distinguished by the gender value of their possessor (masculine vs feminine), but gender is irrelevant elsewhere. Both the current account, and Adamson (2024) concur that PNs agree with their possessors in person, number and gender. This contrasts with Dixon’s description, which casts possessor agreement only in terms of gender.

Jarawara PNs belong to seven main inflectional classes. There are also a number of nouns in flux, characterised by overabundant paradigms which exhibit intermediary stages between these main types (see Dixon 2004: 343–360 for a detailed and insightful description). Although inflectional classes are not explicitly identified by Dixon (2004), I have been able to determine the number of classes based on his careful cataloguing of the properties of possessed nouns in the language. The paradigms that follow were constructed from the shorthand listings of the forms of PNs with singular possessors (Dixon 2004: 343–360). The non-singular part of the paradigm rests on two important observations (described in Dixon 2004: 314–5) that are true of all attested PNs, namely that first and second person non-singular forms are always the same as the third person singular masculine agreement form, while the third plural form is always the same as the third person singular feminine agreement form.

3.1.1 Class I

Class I inalienably possessed nouns have four different morphological forms, built using two distinct stems. This can be observed by considering the paradigm for mano/mani ‘arm’ in Table 1 and teme/tame ‘foot’ in Table 2. In this class, the third person singular forms make a distinction between masculine and feminine gender. In Class I, the stem used with third person singular masculine possessors (here teme) is used as the inflectional base for formation of the first and second synthetic person forms (here oteme and titeme), as seen in Table 2.

Table 2: PN paradigm for teme ‘foot’ (based on Dixon 2004: 350).

Syncretism exists between the third singular forms and those used in the non-singular cells of the paradigm. The third person singular masculine form teme ‘his foot’ is syncretic with the first and second person non-singular form, while the third person singular feminine form tame ‘her foot’ is syncretic with the third person non-singular form. As with all classes, pronouns are obligatory controllers of agreement in the non-singular part of the paradigm. The non-singular possessors do not form part of the inflectional paradigm of PNs, but are shown in parentheses to help illustrate the featural distinctions made by controllers. The shape of the paradigm indicates that person, number and gender of the possessor are all relevant for determining which form to use even though values for these features are not expressed directly. If the stem is vowel initial, epenthetic /w/ occurs between the stem and the possessor prefixes e.g. o-w-ete ‘my forehead’, ti-w-ete ‘your forehead’ (Dixon 2004: 66).

3.1.2 Class II

Class II nouns have three different base stems. This can be observed by considering the paradigm for fanako/fanaki ‘thigh’ in Table 3. They differ from Class I, in that the base stem used in the formation of the first and second person singular forms is distinct from the third person singular forms. Note that the first person singular form ofanoko ‘my thigh’ and second person singular form tifanoko ‘your (SG) thigh’ make use of the base stem fanoko. One possible explanation for this pattern is that the ‘masculine’ stem underwent vowel assimilation in the presence of the prefix o- and spread by analogy to the second singular cell of the paradigm (Dixon 2004: 48, 65, 343–360).

Table 3: PN paradigm for fanako ‘thigh’ (based on Dixon 2004: 350).

3.1.3 Class III

As with Class I nouns, the stems of Class III nouns are distributed according to the person, number and gender features of their possessor. They differ from Class I nouns only in that the first person singular and second person singular stems are built from forms syncretic with their feminine stem, as exemplified in Table 4.

Table 4: PN paradigm for ifo ‘lower lip’ (based on Dixon 2004: 348).

This is a marginal class with only three members: ifo/ifi ‘lower lip’, iso/isi ‘leg’ and ino/ini ‘tooth, name’, all of which have bimoraic roots that are i-initial. As with vowel initial stems in Class I, an epenthetic /w/ occurs between the stem and the possessive prefix.¹⁵ In Class III the base stems used with first and second singular possessors are identical, as observed for every inflection class of PNs.

3.1.4 Class IV

In Class IV, the loss in the formal distinction between masculine and feminine forms in the singular has resulted in the collapse of a formal distinction between person in the non-singular part of the paradigm. In this class, the same form is used as a stem throughout the whole paradigm, as illustrated in Table 5.

Table 5: PN paradigm for tati ‘head’ (based on Dixon 2004: 347).

Evidence for the loss of formal marking of masculine and feminine on stems in this class comes from comparative Arawan data, discussed in detail by Dixon (2004: 316–328).

3.1.5 Class V

In Class V a distinction between masculine and feminine forms is retained in the singular part of the paradigm. Unlike in Classes I–IV, first person singular and second person singular cells are expressed periphrastically by combining featurally distinct possessive pronouns with a stem that is syncretic with the masculine stem, as in Table 6. The forms of the first person singular and second person singular possessive pronouns in Class V (as well as in Classes VI and VII) are identical to those used in alienable possession phrases (discussed in §2.1). This similarity across alienable and inalienable paradigms is discussed further in §4.1 and §4.2.1. Therein, arguments are presented that the PN stems in these cells are non-agreeing and have a different syntactic structure from the non-singular formations. Note that the possessors oka and tika are not in parentheses in Table 6 because, there occurrence is determined by the inflectional class of the PN.

Table 6: PN paradigm for emene ‘blood’ (based on Dixon 2004: 347).

As in Class I, in Class V a stem syncretic with the masculine stem (in this case emene) is used widely throughout the paradigm. However, it only realises a gender feature when indexing a third person singular possessor, as argued in §3.2.

A comparison of the distribution of forms across each of the paradigms presented so far indicates that their structure is not uniform across inflection classes. In Classes I–IV, the possessor and possessed can be expressed by a single word when the possessor is singular. But in Class V, a separate pronoun is required if the possessor is first person or second person singular. Despite surface parallels between the cells, the singular and non-singular first and second person cells of the Class V paradigm are not predicted by a uniform syntactic analysis. The non-singular cells make use of a ‘nominative’ pronoun series that otherwise function as fully fledged NPs, while the singular cells are formed using possessive pronouns, which form DPs and arguably occupy a different structural position (see §4 for discussion of the NP/DP distinction adopted). Since the form and morphological or syntactic status of a possessor is determined by the inflection class it belongs to, selection principles sensitive to purely morphological features are at work. The cells for first and second person singular are expressed synthetically or periphrastically, depending on the inflection class of the PN. In §4, I demonstrate that the structures used for expressing (i) first and second person singular inalienable possession, and (ii) non-singular inalienable possession are different. In §5.3, I lay out the predictions this makes for agreement in recursive possessive structures.

3.1.6 Class VI

In Class VI the loss in the formal distinction between masculine and feminine forms in the singular has resulted in the collapse of a distinction between person in the non-singular part of the paradigm. The first and second person singular cells are expressed by combining featurally distinct possessive pronouns with a form syncretic with the third person singular form of the PN, as in Table 7.

Table 7: PN paradigm for boti ‘inside of, deep part’ (based on Dixon 2004: 344).

3.2 Stem distribution

The shape of these seven inflectional classes allows the following generalisations to be made about stem distribution across all (inalienably) possessed nouns. In what follows, (base) stem refers to the form of the PN to which prefixes attach, or which combine with a possessive pronoun to form a possessive phrase:

1. The base stems of second person singular PN forms are always syncretic with the base stems of first person singular forms.

2. The first and second person non-singular PN forms are always syncretic with the third person masculine singular form.

3. The third person non-singular form of a possessed noun is always syncretic with the form used for the feminine singular form.

4. The base stem in first person singular and second person singular cells is nearly always syncretic with the third person masculine singular form (as in Classes I, IV, V and VI), but it can be different from this form (as in Classes II, III and VII).

5. PNs in Classes I, II, III and VI have distinct masculine and feminine forms; elsewhere the feminine is syncretic with the masculine.

PNs that occur with third person singular possessors or non-singular possessors inflect for the properties of their possessor, which occupies a distinct syntactic node. The form of the PN reflects a morphomic pattern of exponence. That is, the paradigm exhibits a ‘systematic morphological syncretism which does not define a (syntactically or semantically) natural class’ (Trommer 2016: 60). The use of the term morphome to refer to such patterns originated in the work of Aronoff (1994), and the concept has been discussed and adopted in a number of works (Maiden 1992; 2005; 2018; Round 2011; 2015; Smith 2013; Bermúdez-Otero & Luís 2016; Koontz-Garboden 2016; Spencer 2016; Trommer 2016; Herce 2020; 2023; among others). For recent overviews of the literature on morphomes see Maiden (2021) and Herce (2023: 1–12). For discussion of what constitutes a natural class see Herce (2023: 25–33).

To illustrate which aspect of the PN paradigms in Jarawara is morphomic, consider the abstraction over the (base) stems used in inflection classes in Table 9.

Table 8: PN paradigm for sowiri ‘penis’ (based on Dixon 2004: 351).

Table 9: Stem distribution.

As detailed in §3.1, all inflectional paradigms of PNs adhere to the basic patterns of syncretism in Table 9, but they differ as to whether there is identity between the three stems found in the distributions labelled as A, B and C. The distribution of Stem A arises as part of a natural syncretism between first and second person forms. This syncretism is motivated by the participant feature in Adamson’s (2024) analysis. Stem C arises as part of an intersection between third person feminine and third person plural. This syncretism is considered ‘natural’ because each syncretic cell has at least one shared feature value (namely, third person). Conversely, the patten of shared exponence with Stem B is not natural because there is no shared value between the third person singular masculine cell on the left side of the paradigm, and the first and second person (i.e. participant, or non-third person) non-singular cells on the right. This is a metamorphome in the sense of Round (2015), namely a collection of ‘cells that, within the inflectional paradigm of a given lexeme, share exponents or morphological traits despite not constituting a natural class’ (Herce 2023: 121).¹⁶

An alternative analysis of the Stem B pattern in which the masculine cell and first and second person non-singular cells are simply accidentally homophonous rather than the result of a morphologically specified relationship obscures the fact that this pattern is systematic across all PN paradigms regardless of inflection class. Furthermore, metamorphomes themselves may exhibit “syncretism”. For instance, where there is no formal distinction in the exponence of masculine and feminine agreement forms this leads to mass syncretism in the paradigm, as in Class VII. This is even more extreme in Classes IV and VI, where all the singular forms are syncretic, leading to no distinctions in the form of the stems. We certainly would not expect so much regularity in an accidental homophony approach, especially given the various ways that the gender distinction is expressed by stems, as illustrated in §3.1.

Differences in the identity between the three stem distributions in Table 9 gives rise to the inflectional patterns summarised in Table 10. All of the logically possible patterns of identity are attested.

Table 10: Possible stem distribution syncretisms.

4.1 Syntactic constituency of alienably possessed phrases

Alienable possessors are contained within DPs that occupy the specifier position of an NP headed by the possessed noun. Arguments to support this analysis come in various guises:

Using this evidence, the basic structure of phrases formed with an alienably possessed noun is provided in (16) for a nominal possessor and in (17) for a pronominal possessor. In each case the c(onstituent)-structure of the phrase is illustrated by the tree. The accompanying attribute-value matrix (AVM) details the associated f-(unctional) structure.

The c-structure demonstrates the phrase containing the alienable possessor – a DP – is a sister of an intermediate projection (i.e. it is a specifier), rather than a complement of the head. Each of the phrases is head final. Here kaa is associated with providing a genitive case feature. An alternative analysis of (17), in which the possessor DP is branching, is possible in an approach that permits lexical sharing (e.g. Lowe 2016) or the decomposition of lexical items into binary branching trees (e.g. Adamson 2024), but nothing hinges on this here. I follow Sadler’s (2000) proposal for Welsh common nouns in treating possessors as being a non-argument function of Jarawara alienable nouns. This reflects the fact that alienable possessors are optional and is indicated by positioning the poss function outside of the argument structure in the lexical entries for jomee in (16) and maone in (17).

Within LFG, index is the term used to refer to the set of features relevant to indexation of grammatical and discourse functions, namely gender, number and person (cf. phi-features in derivational approaches). These contrast with concord features, which are generally assumed to be relevant in the nominal domain (see Wechsler & Zlatić 2003 and Haug 2023 in the LFG tradition, and Norris 2014 for concord in the Minimalist tradition) and sometimes differ from the index features of a noun. In the absence of evidence to the contrary, I assume that index and concord features of a noun or pronoun are always identical in Jarawara, and therefore only provide index features in lexical entries, f-structures and c-structure annotations. Note that there is no agreement within the phrases in (16) and (17) and only the index features of ‘dog’ and ‘tapir’, not the possessor, are part of the index features of their NPs’ f-structures.

4.2 Syntactic constituency of inalienably possessed phrases

One of the central lines of argumentation advanced here, is that inalienably possessed phrases in Jarawara do not all have the same structure. The periphrastic first person singular and second person singular cells of the paradigm pattern like alienably possessed phrases – they are NPs with DP specifiers and do not involve agreement between possessor and possessed. Others occupy a single node in c-structure and likewise do not involve agreement. The remaining possessors (§4.2.2–4.2.4) exhibit a different structure in which a pronominal or nominal possessor is a sister to a PN. Such structures permit both agreement and feature sharing (§5).

5.1 Adamson’s (2024) Gender Locality Hypothesis

Adamson (2024) uses Jarawara data in support of his Gender Locality Hypothesis (GLH), which attempts to account for cross-linguistic differences in the agreement behaviour of alienable and inalienably possessed nouns.

The GLH states that ‘gender features on n must be valued within nP’ (Adamson 2024: 225). Because both gender and inalienable possession are introduced locally to the nP as a property of nouns, they have the potential to interact. However, the gender value on the nominalizing head n is fixed before the nP combines with other external syntactic material, such as a PossP (Myler 2016; Adamson 2024: 220). This results in an asymmetry in the gender assignment potential of local inalienable possessors on one hand, and non-local alienable possessors on the other, as indicated by the dotted line in (36). His central proposal is that a noun’s gender assignment can be affected by virtue of being inalienably possessed (ipossession), but it cannot be affected by being alienably possessed (apossession).

With respect to Jarawara, Adamson (2024: 239) proposes that inalienably possessed nouns are licensed only by ns with unmarked feminine gender. He calls it an instance of possessee gender, whereby ‘ipossessability is associated specifically with a particular gender value’. He also argues explicitly that possessor agreement does not feed gender assignment in Jarawara.

The theoretical mechanisms used in each of our accounts are rather different. In Minimalism, agreement is modelled in constituent structure, while in LFG agreement is modelled in f-structure. However, it is nevertheless possible to compare them with respect to the data they can handle, the predictions they make and similarities they share. As with the current proposal, Adamson (2024: 239) distinguishes between alienable possessors introduced as a specifier (in his analysis, this is the specifier of PossP, with the Poss head realised as kaa) and inalienable possessors that are structurally closer to the possessed (in his analysis, introduced internally to nP), as represented in (36).

Both the current account, and Adamson (2024) concur that PNs agree with their possessors in person number and gender, known as phi-features in Adamson’s account, and index features in the current proposal. This contrasts with Dixon’s description, which casts agreement between the possessor and a PN only in terms of gender.

The most important difference between our proposals concerns our views about how PNs acquire their features. Adamson (2024) proposes that all PNs are inherently feminine, and that any other features that they realise are simply agreement features. In his approach to the Jarawara data, PNs do not ‘inherit’ their features from the possessor. In the analysis proposed here, PNs do not have an inherent gender value independently of their possessor. They acquire their gender value through feature sharing. As I show in §5.2 and §5.3, a feature sharing analysis accounts for the agreement patterns observed on agreeing attributive modifiers and recursive PNs.

5.2 Feature sharing and adjective agreement

In Jarawara inalienably possessed phrases with third person singular possessors, agreeing adjectival modifiers always share the same gender feature as the possessor. This holds true when the adjective forms a constituent with the possessor, or the PN, as in (37), where underlining indicates which head is semantically modified in each case.

In (37a), the adjective agrees with the noun it modifies, jomee ‘dog’, while in (37b) it agrees with the features of the possessive phrase headed by the PN. Assuming that PNs do not have either index or concord gender values of their own, the PN acquires the index features of the possessor through agreement with the ncomp. The f-structures for the phrases in (37a) and (37b) are shown in (38) and (39) respectively. Note that in (38) the adj is within the f-structure of the ncomp, while in (39) it is part of the containing f-structure.

Consider next the behaviour of agreeing adjectives in possessive structures with pronominal possessors, as in (40). Whenever there is a pronominal inalienable possessor, masculine agreement on agreeing adjectives is ungrammatical. But this introduces what I have called the adjective problem (§2.3), namely that adjectives (which form a constituent with the noun) end up having have a non-local controller of head-modifier agreement as in (40).

The fact that there can be a mismatch between the gender of an agreeing modifier and the apparent gender of a PN has been employed as a central argument in support of pronominal possessors being feminine in Dixon’s analysis. Following the line of argumentation in Dixon (2000; 2004), since the adjective in (40a) does not agree with the superficial surface gender (i.e. masculine) of the PN teme ‘foot’ (see Table 2), it must be agreeing with the feminine pronoun ee. However, under the analysis presented here, neither the pronoun nor the PN have an inherent gender value. Rather, the form of the PN is determined by the person and number features of the possessor, as reflected in the glosses in (40), and default (i.e. ‘feminine’) agreement on the adjective is inevitable, because neither the pronoun nor PN have gender values. As we will see, the distribution of stems in recursive structures also points to an analysis in which phrases headed by a PN involve feature sharing.

A central claim that Adamson (2024: 239–40) makes is that all PNs have inherent feminine gender, regardless of the agreement behaviour they show. This differs from my account in that I propose that PNs lack a gender value of their own but may acquire one through feature sharing. Nothing in my analysis rests on this assumption (it could be that PNs are feminine in general), however, it would mean that feature sharing can replace existing index features of a noun where they already exist. My proposal recognises that stem alternations that manifest agreement with possessors may be morphomic and differ across inflection classes. This accounts for identities in the form of PNs within the agreement paradigm, without similarities in index feature structure. To account for agreement on predicates, as well as agreement within the NP, I propose that independent and prefixal pronouns lack a gender feature and cannot control agreement on the predicate or on attributive adjectives. Pronouns, but not nouns or PNs, have an inherent number feature, specified in their lexical entry, as seen in (23) and (24). PNs can acquire a number value through feature sharing, as seen in (32). Noun phrases can be specified for number only through an anaphoric or co-referring relationship with a non-singular pronoun (i.e. with mee). This neatly accounts for the observed facts without having to specify additional postsyntactic impoverishment rules (needed by Adamson 2024: 237–8).

With respect to the adjective problem, Adamson’s (2024: 244) proposal predicts that if an attributive adjective agrees with a PN, it will occur in its feminine form, because regardless of the agreement form it has, the PN itself is always feminine. Examples like (37b), (and a similar example cited by Adamson 2024: 244, fn 27) where a masculine adjective modifies a (supposedly feminine) PN, illustrate that this is not the case; without a work around, Adamson’s (2024) proposal about adjective agreement doesn’t adequately capture the empirical facts.

In my proposal adjectives agree with the head they modify only if it has a masculine gender value. If it lacks this value, the default feminine form of the adjective occurs. Data from Dixon (2000) indicates that attributive adjectives can agree in gender with the possessor if it is masculine singular. I argue that they can also agree with a phrase headed by a PN that acquires its features from a possessor through feature sharing.

5.3 Feature sharing in recursive possessive phrases

Building on the evidence presented so far, I argue that the behaviour of possessed nouns in recursive structures falls out naturally from a feature sharing approach if we recognise that PN stems follow a morphomic agreement pattern (§3). This accounts for the feature problem and the recursion problem identified in §2.3.

To understand how the morphological patterns seen in recursive structures support a morphomic analysis, consider those in (41), in which the PN ‘inside’ varies in form based on the person, number and gender features of its possessor, in this case an inalienable possession phrase headed by a PN. The patterns observed here remain unresolved by Dixon (2000; 2004), who has no explanation for why there is so much variation in the formation of recursive inalienable possessives.

If we assume, as Dixon does, that the PN forms that are syncretic with the masculine or feminine stem always bear a masculine or feminine gender value respectively, the forms in (41a–b) should not be possible since baki should agree with masculine mano. Similarly, if we follow Dixon’s claims that inalienable possessors are the head of a phrase formed with a PN, and that all pronouns have feminine gender, we expect to find a pattern in which all possessors of PNs that are themselves composed of a possessed PN would control feminine agreement on any subsequent PNs except where the most embedded possessor is masculine. Under such an analysis, the examples in (41e–g) would be unaccounted for.

Neither of these descriptions work for explaining the forms of PNs in recursive structures because they do not incorporate three important properties of the system:

1. Feature sharing: Most PNs acquire their index features through feature sharing, a dimension of agreement observed in a subset of agreement relations.

2. Lexical splits: Synthetic and periphrastic first person and second person singular PNs do not involve feature sharing. The synthetic forms (which may have involved agreement at an earlier diachronic stage) are synchronically fully specified for their features in the lexicon. The periphrastic forms follow a syntactic pattern that is identical to the alienable possessive phrase, and do not involve agreement.

3. Morphomic patterns of agreement: PNs exhibit a morphomic pattern of possessor agreement. That is to say, the forms of PNs are distributed according to a pattern of exponence that cannot be explained by reference to natural classes alone.

To put these claims into perspective, consider first the following noun phrases with recursive third person possessors. In each case below, the most embedded possessor, Okomobi in (42a) and Jane in (42b), determines the gender value for the first possessed noun (‘arm’) through feature sharing. This possessive phrase, in turn, determines the gender of the second possessed noun (‘inside’).²²

Building on the constituent structure analysis put forward in §4, the recursive structure of (42a) is shown in (43).

As with the feature sharing structures illustrated in §4.2 and §5.1, feature sharing in (44) is indicated by a line between the most embedded ncomp and the index features of the pred of which it is an argument. The index features of that PN are in turn shared with the pred of which it is an argument.

Where the most embedded possessor is first person singular or second person singular, as in (41a–b), there is no feature sharing between the pronominal possessor and the most local PN. Therefore, only the features of the PN and not the possessor can be shared with any subsequent PN, as in (45).

In my approach, there is no feature sharing at the lexical level; feature sharing is a syntactic phenomenon that arises through agreement. Consequently, forms like omano and tiwisi are treated as third person NPs – the index features of the possessors are not shared with the PN. The resulting head lacks any gender feature (since PNs lack an inherent gender feature). This accounts for the fact that a default feminine form is used in recursive structures like (45).

Adamson (2024) proposes that first person singular and second person singular forms behave distinctively in recursive structures because the possessor and possessed are expressed by a single word (in contrast to those expressed syntactically). In his approach, agreement occurs between o- and ti- and the stems to which they attach, just as with other participant pronouns such as ee and otaa. Once the possessor and possessed elements form a single word, the features of the possessor are no longer visible to subsequent agreement processes. Consequently forms like omano and tiwisi cannot control agreement related to the phi-features of o- and ti- when they occur in recursive structures. Although my analysis and Adamson’s(2024) account approach the problem from different perspectives, they each use the same property – wordhood – to account for problems for which Dixon (2000; 2004) has no explanation.

However, Adamson’s (2024) argument only attempts to account for the stems of Class I PNs. There are a number of stem alternations that remain unexplained in other classes. These include the use of ‘feminine’ stems for agreement with some participants (i.e. first person singular and second person singular) in Class III forms like tiwisi ‘your (sg) leg’. It remains to be seen if Adamson’s (2024) approach is capable of handling such differences.

In general, it is unclear what happens in recursive structures where the most embedded possessor is in Classes V–VII. There are no examples of recursive structures involving non-synthetic first person and second person singular possessors in Dixon (2004), and Adamson (2024) does not make any explicit predictions about such structures. I predict that the pronouns in periphrastic cells do not control agreement in gender on subsequent PN stems because the stems heading the phrase do not have a gender value (regardless of any formal similarity they have with stems that do).

5.4. Possible prerequisites of feature sharing

Feature sharing through agreement has been proposed to be important in accounting for agreement phenomena of different types in different frameworks. Haug & Nikitina (2016) provide a detailed overview of proposals that involve feature sharing including Kathol (1999) in HPSG, and Frampton & Gutmann (2000); Legate (2005); Pesetsky & Torrego (2007); Bobaljik (2008); and Ackema & Neeleman (2013) in derivational frameworks (see Keine, forthcoming, for an instructive and informative overview). Recent LFG analyses of agreement with feature sharing are found in Haug & Nikitina (2012; 2016) and Nikitina & Haug (2015); Alsina & Vigo (2014; 2017) and Sadler (2019).

Yet feature sharing is not a property of all agreement relations. This leads to the question of why feature sharing is possible in Jarawara inalienable possessive constructions, when it is not seen more generally in possession constructions.

One particularly important observation in this regard is that feature sharing through agreement is only possible when an agreement relation between controller and target already exists. Consequently, we only expect feature sharing from possessor to possessed to be attested in languages where the possessor controls agreement on the possessed noun. A related factor is that all of the Jarawara examples in which feature sharing is proposed to exist involve relational nouns which obligatorily require a possessor. Feature sharing in Jarawara is restricted to possessive relationships which are broadly characterised as ‘inalienable’. The majority of these encode relational concepts such as body parts (e.g. teme/tame ‘foot’), plant parts (e.g. mowe/mowe ‘flower’), relational spatial concepts (e.g. bako/baki ‘inside’), or properties of and entity (e.g. maho/mahi ‘smell’ (see Dixon 2004: 311 for further details). It is possible that feature sharing in possessive constructions is restricted to relations of this kind, perhaps first originating in meronymic relations where the possessive phrases identify a component (e.g. a body part) and the integral entity (i.e. its possessor) at the same time. Another factor that may play an important role is the underspecification of the gender value of PNs, with feature sharing developing as an alternative to the default agreement strategy. Structural locality (a central issue for Adamson 2024) and topicality of the possessor may play an important role, but for the time being the relative importance of these factors remains fairly speculative.