A word order typology of adnominal person

2.1 Terminology

The overall phenomenon under discussion can be defined as in (2). It is understood to be distinct from the cross-referencing of person features on arguments in verbal morphology and from nominal morphology indicating the person of a possessor.

While unaccompanied/“plain” personal pronouns may also encode pers _n as part of an extended nominal projection (xnP) (Grimshaw 2005; Panagiotidis 2015; van Riemsdijk 1998), the focus here is on pers _n marking of nominal expressions containing overt lexical material (nouns, adjectives). I distinguish two types of overt pers _n marking based on its morphological realisation.^[3] The most commonly attested type are APCs as defined in (3) and exemplified by English we linguists, where a pronoun forms a close grammatical unit with the remainder of a nominal expression. The term APC is adapted from Rauh (2003), but the phenomenon is also referred to as pronoun-noun or noun-pronoun construction (Choi 2014; Comrie and Smith 1977; Patz 2002), pronominal determiner (e.g. Roehrs 2005) or, particularly with some Papuan languages, pronominal copy (Davies 1989: 107f.; Reesink 1987: 53f., 167; Roberts 1987: 162; Scott 1978: 100).

The second type of overt pers _n marking is the bound person construction (BPC) characterised in (4). BPCs are found in a smaller number of languages, with the marking typically occurring at the edge of nominal phrases as illustrated by the first person singular marker -na: in (5) from the TNG language Fore.

Bound pers _n markers are found under various designations in the literature, including pronominal prefix (Gravelle 2010), person prefix (Andrews 1975), nominal designant (Haacke 1977), pronominal apposition (Haiman 1980), appositional pronoun (Scott 1978), affixed form of the personal pronoun (Renck 1975), pronominal enclitic (Obata 2003), person/number/gender clitic (Whitehead 2006) or PGN/person-number-gender (PNG)-marker/suffix (Bruce 1984; Haacke 2013; Kilian-Hatz 2008).

Personal pronouns (exponents of pers _n in present terminology) and demonstratives have been argued to form one distributional category (Blake 2001; Choi 2014; Jenks and Konate 2022; Lyons 1999; Rauh 2003; Sommerstein 1972) because they both involve deixis (or indexation, cf. Jenks and Konate 2022) and they are in complementary distribution in many languages. However, some languages allow their co-occurrence in pers _n /pronoun-demonstrative constructions (PPDCs) as defined in (6) and illustrated in (7), suggesting cross-linguistic variation concerning whether person and demonstrative features are encoded jointly (Höhn 2015, 2017: ch. 7; Hsu and Syed 2020).

2.2 Previous work and hypotheses under investigation

Most pers _n -related work has focused on APCs in individual languages like Basque (Artiagoitia 2012), English (Delorme and Dougherty 1972; Keizer 2016; Panagiotidis 2002; Pesetsky 1978; Postal 1969; Sommerstein 1972), German (Lawrenz 1993: ch. 6; Rauh 2003, 2004; Roehrs 2005), Hoava (Palmer 2017), varieties of Italian (Cardinaletti 1994; Höhn et al. 2016), Japanese (Furuya 2008; Inokuma 2009; Noguchi 1997), Papuan Malay (Kluge 2017: ch. 6.2), Romanian (Cornilescu and Nicolae 2014) or Warlpiri (Hale 1973).

A few publications offer a wider cross-linguistic perspective. Arguing against an appositive analysis for English APCs, Pesetsky (1978) presents some data from Russian, French and German (with additional languages mentioned without examples). Himmelmann (1997: 213–219) and Lyons (1999: 141–145) address pers _n -related data from several languages (e.g. Khoekhoe/Nama) when discussing whether pers _n is (Lyons 1999) or is not (Himmelmann 1997) related to articles and definiteness, but provide no detailed description of a larger sample. Based on contrasts between standard Italian and Greek as well as some additional languages, Choi (2014) and Höhn (2016) independently argue that the availability of the unagreement phenomenon from footnote 3 is related to properties of nominal structure, specifically the occurrence of definite articles in APCs. APCs and their relation to nominal determination in several Australian languages are addressed in a handout by Stirling and Baker (2007) and in a detailed typological study focusing on third person APCs by Louagie and Verstraete (2015) (cf. also Louagie 2017: ch. 5). Based on a wider language sample Höhn (2020) observes that the lack of third person APCs in languages like standard English (*they linguists) is largely restricted to Europe and adjacent areas and typically implies the existence of definite articles.

This is expected if definite articles in such languages are allomorphs of third person pronouns as proposed by the pronominal determiner analysis, a prominent approach to APCs in languages like English or German (Abney 1987; Lyons 1999; Postal 1969; Rauh 2003; Roehrs 2005 among others). Based on the complementary distribution of adnominal pronouns and definite articles in the relevant languages, this analysis holds that both compete for realisation of the same syntactic head D in (8). Some further arguments in favour of this approach are sketched in Section 2.3.2 below.

(8)

As mentioned earlier, there are languages with obligatory definite articles in APCs (Choi 2014; Höhn 2016), so an analysis building on the complementary distribution of adnominal pronouns and definite articles cannot be universally applicable. However, I adopt from the pronominal determiner analysis the generalised working hypothesis that pers _n marking corresponds to a syntactic head in the xnP in a substantial number of languages. This admits the possibility that pers _n marking can also be marked by non-heads (e.g. specifiers, phrasal adjuncts) in some languages.^[4] The range of variation implied here is illustrated in the decision tree in (9), modelled after the parameter hierarchies of Roberts (2012, 2019) and related work. The rightmost leaf of the tree is a shorthand for various other potential syntactic positions of pers _n heads within the xnP.^[5]

(9)

Word order patterns have been argued to show systematic cross-linguistic correlations across different domains (Dryer 1992, 2009, 2011; Greenberg 1963; Hawkins 1980, 1983), among them verb-object order (VO vs. OV), adposition directionality (pre- vs. postpositions) and the order of dependent genitives (NGen vs. GenN). One theoretical perspective on such correlations is a head-directionality parameter (Huang 1982; Koopman 1984; Stowell 1981; Travis 1984) which treats these correlating configurations as instances of head-initial (VO, prepositions, NGen) or head-final (OV, postpositions, GenN) structures respectively. However, one language-wide parameter simply set to either head-inital or head-final is now widely agreed to be too strong, since these correlations are not deterministic. Nevertheless, there are still suggestive tendencies for cross-categorial harmony in word order patterns (Baker 2008; Greenberg 1963; Hawkins 1980, 1982, 1983), such that, for instance, a language with postpositions (head-final PPs) is more likely to display OV order (head-final VPs) than not. I will not attempt an explanation here, but for recent theoretical approaches to the relationship between macro- and micro-parametric variation in terms of learning mechanisms and the concept of (emergent) parameter hierarchies see, e.g., Baker (2008), Roberts and Holmberg (2010), Biberauer and Sheehan (2013), Biberauer and Roberts (2016), Roberts (2019) and van der Wal (2020).

If a substantial number of languages treat pers _n as a syntactic head in the xnP (answering “yes” to the topmost question in (9)), its directionality relative to the remaining nominal expression and specifically the lexical noun is expected to display a cross-linguistic tendency to pattern harmonically as formulated in Hypothesis 1 (10).

At a reviewer’s request, I also briefly address the relation to the directionality of definite articles. If definite articles also behave as syntactic heads, Hypothesis 1 applies, predicting a tendency to match in directionality with pers _n marking. For the subclass of languages where definite articles cannot occur in APCs the pronominal determiner analysis predicts an even stronger effect, see Hypothesis 1a (11). Adnominal pronouns and definite articles should match directionality not only due to (violable) harmony effects, but because they occupy the same position in the relevant languages.

Another issue beyond headedness concerns the aforementioned proposal that personal pronouns and demonstratives form a single distributional category, based on their similar deictic semantics and their complementary distribution in many languages (Blake 2001; Choi 2014; Lyons 1999; Rauh 2003; Sommerstein 1972). Jenks and Konate (2022) capture this in the single index constraint (SIC). The existence of PPDCs (Section 2.1) raises some challenges for a universal statement to this effect (see Section 5.3), but it seems a plausible working assumption that the cited literature is indicative of a cross-linguistic tendency for demonstratives and pers _n to be members of the same distributional category.^[6] Such co-categoriality also entails a structural parallel between pers _n and demonstrative modifiers. If this is cross-linguistically common, Hypothesis 2 in (12) should hold.

2.3 Delineating APCs

APCs can resemble other configurations involving independent personal pronouns, such as resumptive pronouns that happen to be adjacent to a co-referent noun phrase or appositions. Below, I summarise arguments that APCs cannot be subsumed under these phenomena. Section 2.3.2 specifically presents some of the arguments in favour of a pronominal determiner analysis for languages like English.

4.1 Overview

Figure 1 shows the geographical distribution of APCs and BPCs as well as their respective directionality for the languages in the sample.^[15] Colours indicate the directionality of APCs and shapes the directionality of BPCs. Circles mark languages without (evidence for) BPC and red symbols indicate languages without (evidence for) APCs. The area of Melanesia is a hotspot of variation in the type and directionality of pers _n in the current sample, cf. Figures 2 and 3. The clustering of postnominal APCs and BPCs there may suggest that they play a role as areal features, although they are not exclusive to this area and the property is clearly not universal even among the non-Austronesian languages of the area either.

Figure 1:

Map of APC and BPC directionality of languages included in the sample. Key: 1–Kinyarwanda, 2–Northern Calabrese, 3–Southern Calabrese, 4–Aromanian, 5–Calabrian Greek.

Figure 2:

Map of APC and BPC directionality around Papua New Guinea.

Figure 3:

Map of APC and BPC directionality in the Solomon Islands.

Table 1 summarises the results with the columns indicating the number of languages with pre-, postnominal and ambidirectional APCs or BPCs.

For each type of pers _n marking the lines indicate the number and relative percentage of languages with specific directionality patterns for adpositions, dependent genitives, verb-object order and demonstrative modifiers respectively.^[16] The sample contains 103 languages with APCs and 13 languages displaying BPCs. Note that three languages seem to allow both APCs and BPCs,^[17] so these two types of nominal person marking are not necessarily exclusive. Among APCs, prenominal pers _n marking is the most frequent option, while postnominal marking is more frequent among BPCs.^[18]

4.2 Tendencies

Table 2 tentatively summarises word order tendencies in Table 1 above according to type and directionality of pers _n marking. A tendency is assumed if the most numerous value within a category has double the instances of the next lower one. For prenominal APCs, a DemN tendency is indicated in brackets because it just barely fails to meet the criterion. With the exception of object-verb order, the tendencies for ambidirectional APCs are also bracketed due to the small differences in absolute numbers. The two languages with prenominal BPCs are insufficient for establishing any tendencies.

Consistent with Hypothesis 1, languages with postnominal APCs and BPCs show patterns commonly associated with head-finality in all three investigated domains (postpositions, prenominal genitives and OV order), while prenominal APCs display a tendency for head-initiality at least with respect to prepositions and VO order.

Demonstrative order also suggests an interaction with pers _n , albeit with an interesting exception. For languages with pre- and postnominal APCs demonstrative directionality patterns with pers _n directionality consistent with Hypothesis 2. However, postnominal BPCs display a tendency for prenominal demonstratives, resembling in this respect the (marginal) DemN tendency of prenominal APCs and contrasting with postnominal APCs.

Due to the unbalanced nature of the sample, these tendencies in the raw data have to be treated with care. The next section statistically evaluates the validity of the two main hypotheses for pre- and postnominal APCs. Further theoretical implications are discussed in Section 5, where I will also address the interaction of definite articles and APCs predicted by the pronominal determiner analysis (Hypothesis 1a).

4.3 Statistical analysis

The numbers of languages with BPCs and ambidirectional APCs are too small for meaningful statistical analysis, so this section exclusively focuses on pre- and postnominal APCs. A generalised linear mixed-effects model was fit using R 4.4 (R Core Team 2024) and the lme4 package (Bates et al. 2015) within the R-Studio environment (RStudio Team 2024) in order to test the significance of head-directionality (Hypothesis 1) and demonstrative order (Hypothesis 2) for the directionality of APC.^[19]

As noted in Section 2.2, the factors adposition, genitive and object-verb order tend to correlate cross-linguistically and are not independent even conceptually due to their role as stand-ins for head-directionality in Hypothesis 1. This poses a problem for the reliability of a model directly incorporating them as predictors. At the same time, the correlations are not perfect because word order is not always harmonious, so a reduction of these factors is not trivial. I address this by deriving the factor headfin.value which is set to 0 by default and increased by 1 for every property value typically associated with “head-finality” (postpositions, GenN, OV), decreased by 1 for every “head-initial” characteristic (prepositions, NGen, VO) and left unchanged for any other value of the three relevant properties. This results in a value range between −3 for languages with three head-initial properties and 3 for languages with three head-initial properties, and values in between for languages with mixed or unclear values for some or all relevant features. Demonstrative directionality (demfin.value) is coded as 1 for languages with NDem order, −1 for languages with DemN order and 0 otherwise. This approach with the use of discrete numerical values for both predictor variables allows the inclusion of languages in the overall analysis even if their head- or demonstrative directionality value is outside the scope of the hypotheses under investigation, so the analysis can include a larger subset of the database.

The logistic regression model in (31) treats APC directionality (APCDir) as dependent variable with two predictors, headfin.value representing Hypothesis 1 and the directionality of demonstrative modifiers (demfin.value) representing Hypothesis 2. Language genus is also included as random intercept to ameliorate the impact of the unbalanced representation of language genera within the dataset.^[20]

In order to allow binomial fitting, the sample was restricted to binary values for the dependent variable APCdir (prenominal/postnominal), leaving a subset of 96 languages from 51 different genera in 31 language families for the analysis. The results are presented in Table 3.

The negative estimate of the intercept reflects the higher number of prenominal APCs in the dataset. The positive estimate for headfin.value suggests that an increase in head-final characteristics is associated with an increased chance that a language displays postnominal APCs. Similarly, the positive estimate for demfin.value suggests an association between postnominal demonstratives and APCs. These results are significant at p < 0.001, but since the fixed effects for smaller datasets like the present one are unreliable, following Bates et al. (2015: 33f.) the log-likelihood for the maximal model in (31) was compared to that of two reduced models where one of the factors headfin.value or demfin.value is dropped respectively. The AIC is used to measure the relative goodness of fit of competing models, while penalising an increased number of parameters in order to avoid overfitting. A lower AIC value indicates a lower estimated loss of information compared to other models. The column AIC _diff in Table 3 lists the difference between the AIC of the full model and the smaller model without the relevant factor. The negative results indicate an increase in AIC for the reduced models compared to the maximal model in (31), suggesting that removal of either factor leads to a significant loss of informativity.

5.1 Head-directionality (Hypothesis 1)

The interaction between the directionality of pers _n marking and head-directionality observed in Section 4 is compatible with Hypothesis 1, repeated in (32).

Languages with postnominal APCs and BPCs in the sample also overwhelmingly display characteristic head-final word orders and the reverse holds for a large number of languages with prenominal APCs, as expected if pers _n tends to be encoded as a syntactic head in the xnP. This suggests that the strategy is salient enough to leave a detectable trace in the current sample. Importantly, though, it does not mean that languages universally realise pers _n as head in the extended nominal projection, as also suggested in (9) in Section 2.2. Against this background, the remainder of this section discusses two configurations that do not conform to the pattern predicted by Hypothesis 1: languages with ambidirectional APCs and languages with prenominal APCs and head-final word order characteristics.

The flexible positioning of person marking in languages with ambidirectional APCs (Supplementary Material S3, Section 1) is at odds with a pers _n -as-head analysis, since head-directionality is generally assumed to be fixed for particular types of heads. As far as the pronominal determiner analysis is concerned, such word order flexibility is also untypical for article-like determiners (Himmelmann 2001: 832). While I offer no specific analyses for the languages with ambidirectional APCs here, I sketch two possible approaches. Considering the information-structural effects tentatively observed for some of these languages (Supplementary Material S3, Section 1), those might actually involve one basic APC order with the reversed surface order of noun and pronoun resulting from (information-structurally driven) movement inside the nominal domain. Moreover, as noted above adnominal pronouns – and possibly also similarly flexible demonstratives – might simply not be heads in some languages, but rather phrasal adjuncts to the noun phrase with flexible (pre- or postnominal) linearisation, similar to the flexibility observed for some English VP-adjuncts (e.g. to [go boldly ] vs. to [ boldly go]).

Languages with prenominal APCs form a more heterogeneous group than languages with postnominal APCs or BPCs, including a sizable number of languages with head-final characteristics. The sample contains 22 languages with prenominal APCs and at least two out of the three characteristics of head-finality considered here (prenominal genitives, postpositions, OV order); for the full list see Supplementary Material S3, Section 2.1.

On the pers _n -as-head analysis, prenominal pers _n marking as indicator of head-initiality would be disharmonic with the other indicators of head-finality. This might not be too problematic, since word order harmony is widely acknowledged to be a tendency and not a universal rule after all. However, the Final-over-Final Condition (FOFC), e.g. Biberauer et al. (2014) and Sheehan et al. (2017), provides a tool to further probe structural interactions. This structural condition (33) aims to account for various structural asymmetries across languages by the assumption that head-final projections can only dominate head-final projections within the same extended projection, ruling out configurations where a head-final projection dominates a head-initial one (34).

(34)

Biberauer et al. (2014: 186–189) illustrate this with Finnish. While mostly postpositional, some Finnish adpositions like yli ‘across’ may be used pre- or postnominally. In (35b), yli ‘across’ is used as a preposition taking a head-initial NP complement and FOFC imposes no restriction. In ungrammatical (35a), yli ‘across’ is a postposition whose complement NP is itself head-initial. If adpositions are part of the extended nominal projection at least in Finnish (Biberauer et al. 2014: 187), this configuration of a head-final PP dominating a head-initial NP corresponds to (34) ruled out by FOFC.

Finnish nominal modifiers can appear prenominally using special linking morphology.^[21] In (36b), the prepositional yli ‘across’ still trivially satisfies FOFC. However, in contrast to (35a) the head-final yli-PP in (36a) also causes no FOFC violation anymore, since the complex NP in its complement is now head-final as well.

Twenty of the 22 largely head-final languages with prenominal APCs have postpositions (Supplementary Material S3, Section 2.1, Table 4; the others lack adpositions or clear information). If adpositions are part of the xnP more generally (e.g. Grimshaw 2005), a head-final PP with a head-initial (i.e. prenominal) APC complement would violate FOFC, cf. (37).^[22] If FOFC is a general property of linguistic structure, there are different ways such languages could deal with this issue.

(37)

One way is to avoid FOFC-violating configurations (Biberauer et al. 2014: 179, 186–189). While I am not aware of relevant empirical data, some of these languages might avoid/disallow using postpositions with (prenominal) APC complements, which would allow maintaining a head-analysis of pers _n for them.

Another possibility, mentioned before with reference to the decision tree in (9), is that pers _n is not a syntactic head in the xnP in some languages of this type, so configurations of prenominal APCs with postpositions are not assigned the FOFC-violating structure in (37). Instead, pers _n is a phrasal modifier, for instance an adjunct to the nominal expression or a specifier of some functional projection Y as in (38).

(38)

Analyses treating adnominal pronouns as specifiers have indeed been independently proposed for at least two of the 22 languages with prenominal APCs and indicators of head-finality: Finnish (Gröndahl 2015a,b; Höhn 2020: 24f.) and Japanese (Furuya 2008; Inokuma 2009; Noguchi 1997), see e.g. (39) where the pronoun watasi-tati ‘we’ is analysed as a phrasal DP in the specifier of a distinct DP projected over the lexical noun.

A similar approach may apply to Bilua, one of the languages allowing both prenominal APCs and postnominal BPCs (Supplementary Material S3, Section 1), see (40a) for an example where they co-occur^[23] and (40b,c) for individual examples of both types. Given the tendency for head-finality across languages with postnominal BPCs (Section 4.2), I suggest that their postnominal markers realise the syntactic pers _n head in these languages. Prenominal pronouns in such languages could then occupy a distinct phrasal position (e.g. in the specifier of the pers _n head).

The approach developed here offers a perspective on why prenominal APC seem to be more common than the other forms of pers _n marking. Prenominal APCs can be derived not only from head-initial nominal structures with pers _n as head, but also from configurations where pers _n marking is introduced as a non-head, whether as specifier, adjunct or even apposition. At least specifiers have been argued to generally merge to the left to account for cross-linguistic asymmetries (Kayne 1994). Languages opting for this strategy of encoding pers _n would then converge on a prenominal APC irrespective of whether their xnPs are generally head-initial or head-final. If postnominal APCs and BPCs are largely confined to occurring in head-final pers _n -as-head languages, while prenominal APCs can result from of a wider range of structural configurations than just head-initial xnPs encoding pers _n as head, this could at least partly explain why the latter are cross-linguistically the prevalent option.

5.2 Directionality of pronominal determiners (Hypothesis 1a)

This section comments on the interaction of APCs with definite articles. The sample contains 45 languages with articles, of which 41 have APCs. If not only pers _n , but also definite articles tend to behave as syntactic heads (Abney 1987; Szabolcsi 1994), Hypothesis 1 predicts a tendency for them to match in directionality as well. This seems to be borne out, as shown in Table 4a. In languages with prenominal APCs articles tend to be prenominal and languages with postnominal APCs display the opposite tendency.

Recall that for languages with pronominal determiners Hypothesis 1a, repeated in (41), predicts an even stronger correlation between the directionality of definite articles and adnominal pronouns, since they are argued to occupy the same syntactic position.

The sample contains 13 relevant languages that have articles, but exclude them in APCs. As shown in Table 4b, they all have prenominal articles and APCs in line with (41) and the predictions of the pronominal determiner analysis.

5.3 Demonstrative directionality (Hypothesis 2)

The findings in Section 4 concerning the interaction between the directionality of adnominal pronouns and demonstratives are consistent with Hypothesis 2 in (42).

Matching directionality is expected if pronouns and demonstratives are co-categorial across a range of languages (e.g. Blake 2001; Choi 2014; Höhn 2016; Jenks and Konate 2022; Rauh 2003). It is not a sufficient criterion for categorial identity as the discussion of pers _n /pronoun-demonstrative constructions (PPDCs) below shows, but provides a heuristic for identifying apparent or real deviations for closer inspection.

Postnominal demonstratives in most languages with postnominal APCs (Supplementary Material S3, Section 1, Table 1) are consistent with Hypothesis 2. Against the background of Section 5.1, demonstratives could occupy a final head position in the xnP (cf. also Sybesma and Sio 2008) with potential variation concerning whether its position is identical to that of the pers _n marker. The sample contains 29 languages with APCs where contrary to Hypothesis 2 the directionality of demonstratives and pers _n marking does not match (Supplementary Material S3, Section 2.1, Table 5).^[24] Two potential explanations for the deviations are listed in (43).

Starting with (43a), eight of these languages allow an alternative demonstrative order matching the directionality of adnominal pronouns, i.e. DemN for languages with prenominal APCs and NDem for postnominal APCs.^[25] The distributional facts of these languages are then still consistent with an analysis where pronouns and (certain) demonstratives are members of the same category. Adnominal pronouns may simply have properties that restrict them to the marked position.

This also explains why Hypothesis 2 does not make clear predictions for languages classified with ‘mixed’ demonstrative order. Whatever the directionality of pers _n marking, there is a matching demonstrative order. Note in this context that almost all languages with ambidirectional APCs also allow pre- and postnominal demonstratives. Kuku Yalanji and Imonda are already categorised as mixed in WALS, but Pitjantjatjara (Bowe 1990: 32), Guugu Yimidhirr (Haviland 1979: 104) and Katu (Costello 1969: 33, 35) also permit the opposite demonstrative order from that marked in WALS (and Table 3 in Supplementary Material S3) in line with the assumption of co-categoriality.^[26]

Turning to explanation (43b) for deviations from Hypothesis 2, recall from Section 2.1 the co-occurrence of demonstratives and pers _n marking in pers _n /pronoun-demonstrative constructions (PPDCs), cf. (44).

Based on PPDCs, Höhn (2015, 2017) and Hsu and Syed (2020) argue that some languages encode pers _n and demonstrative features in distinct syntactic positions and thus belong to distinct distributional categories, so Hypothesis 2 is not expected to apply. Seventeen of the 29 languages with directionality mismatches display PPDCs, see Supplementary Material S3, Section 2.2, Table 5.^[27]

Cheke Holo and Eastern Geshiza are treated as having an alternative demonstrative order matching pers _n directionality and also allowing PPDCs. In both languages different demonstratives occur in pre- and postnominal positions. The (typically) prenominal position of the marked emphatic demonstrative u in Cheke Holo (Boswell 2018: 169) might correspond to that of the prenominal pronoun in the PPDC in (44a), in line with Hypothesis 2.

Eastern Geshiza with its postnominal APCs and prenominal demonstratives also has an additional postnominal demonstrative t ^h ə. However, it is this marker that can occur in PPDCs, cf. (45), showing that while matching in directionality it nevertheless occupies a distinct position from adnominal pronouns. Honkasalo (2019: 301f., 702) suggests that this postnominal morpheme is developing into a topic marker and it is glossed as such in (45), so depending on the degree of grammaticalisation – i.e. whether the demonstrative and topicalising uses involve distinct morphemes – the status of (45) as PPDC may be debated. Importantly, even if one rejects that Eastern Geshiza has PPDCs on this basis, the prenominal position of the remaining demonstrative, contrasting with the postnominal pronoun, suggests that demonstratives and pronouns are not co-categorial in this language.

PPDCs also occur in 18 languages with matching directionality of adnominal pronouns and demonstratives (Supplementary Material S3, Section 2.2, Table 6), suggesting that directionality is indeed not a sufficient criterion for co-categoriality. PPDCs generally represent a challenge for Jenks and Konate’s (2022) single index constraint (SIC), the proposal that there is only one referential index in nominal expressions, due to the co-occurrence of two potential referential indices in their system, pers _n and demonstratives. To tackle this issue, Jenks and Konate (2022: 16f., 31) propose that in languages with prenominal pers _n and demonstratives in PPDCs both are “packaged” in the same syntactic projection, the demonstrative heading a (referentially indexed) D_xP and the pronoun occupying its specifier (46a). A reviewer notes that languages with PPDCs, prenominal APCs and postnominal demonstratives are consistent with Jenks and Konate’s (2022) proposal if the postnominal demonstrative heads D_xP and hosts the adnominal pronoun in its specifier, see (46b).

This analysis might apply to at least some of these languages, but PPDCs like those in (47) are not immediately amenable to either analysis in (46). Jenks and Konate (2022: 16) justify their analysis of pronouns as specifiers with the claim that “the pronoun occurs before the demonstrative” in PPDCs, slightly oversimplifying the Extremity Of Person Hypothesis by Höhn (2017: 261, 288) suggesting that pers _n is typically more distant from the nominal than demonstratives. Examples (47a,b) challenge that empirical claim.^[28] Following Jenks and Konate’s (2022) logic, (47a) would require treating the pronoun as head of DxP and the demonstrative as its specifier, conflicting with proposals that Japanese adnominal pronouns are specifiers (see Section 5.1). While (47b) could be derived from (46a) by movement of NP to some higher position than DxP (Abels and Neeleman 2012; Cinque 2005), the order of (46c) is the inverse of (46a) and assuming a rightward specifier-position for the pronoun contravenes the widespread assumption that specifiers generally branch to the left (but see Abels and Neeleman 2012). Moreover, for both (46b,c) any analysis building on (46) would be at odds with the proposal from Section 5.1 that postnominal APCs involve final pers _n heads. Finally, (46d) shows a PPDCs with three expressions of referential indices, a personal pronoun, a demonstrative and a definite (or familiarity) marker, requiring at least one additional index-related projection compared to (46), thus further challenging the SIC.

A last systematic deviation from the matching directionality of demonstratives and pers _n I want to address concerns the tendency for prenominal demonstratives among languages with postnominal BPCs (Section 4.2). Since those languages have postpositions and presumably a final pers _n head, the FOFC-based reasoning from Section 5.1 suggests that the prenominal demonstratives should occupy phrasal positions, e.g. the specifier of the pers _n -head (48).

This can capture the inapplicability of Hypothesis 2 in most BPC languages, since clitic pers _n markers and demonstratives are indeed not co-categorial. Correspondingly, PPDC-like co-occurrences of demonstratives and clitic pers _n marking are expected and indeed attested for six of these 13 languages (Supplementary Material S3, Section 2.2, Table 7). Another type of expression may, however, show parallels to demonstratives after all. Several languages permit additional pers _n -related expressions to co-occur in their BPCs (see Supplementary Material S3, Sections 1 and 2.2, Table 7). These may be simply pronouns (49) or, e.g. in Khoekhoe, article-like elements expressing a subset of person features (50), cf. (Haacke 1976, 1977, 2013).^[29] These prenominal pers _n -related markers seem to occupy the same position as the prenominal demonstratives; see the Khoekhoe demonstrative in (50b), suggesting co-categoriality for demonstratives with these expressions.^[30] Assuming an analysis like (48), these patterns are also compatible with Jenks and Konate’s (2022) SIC.

The analysis does not directly carry over to the four BPC languages with postnominal demonstratives: Basque, Windesi Wamesa, Menya and Moskona. Basque demonstratives occupy the same phrase-final position as – and are in complementary distribution with – the definite article, including the pers _n /proximate plural article -ok (Section 3), so this seems to be an instance of co-categoriality properly in line with Hypothesis 2. Initial pronouns are obligatory at least in western varieties of Basque (Xabier Artiagoitia, pers. comm.) in examples like (51). The suffixal markers -au and -ori are equivalent to the close (dem.1) and medium (dem.2) distance demonstratives hau and hori. Artiagoitia (2012: 67) suggests that the initial pronouns occupy the specifier of the projection headed by the final demonstrative/pers _n marking, which parallels the structure in (46b), putting these data in line with the SIC.

In the remaining three languages, demonstratives clearly realise distinct positions from pers _n , potentially challenging the SIC. Windesi Wamesa demonstratives precede the determiner (Gasser 2014: 250), while the pers _n markers follow it (52). This implies distinct positions for both as sketched in (53). However, since they seem not to co-occur (Emily Gasser, pers. comm.), the pattern may be compatible with the SIC after all.

In Menya, PPDCs are attested (54). While the prenominal pronoun probably occupies the specifier position of the final pers _n marker, the demonstrative and pers _n marking are structurally distinct, albeit part of a clitic cluster. Hence, co-categoriality is unlikely and issues remain for Jenks and Konate’s (2022) approach.

Moskona raises similar issues because it has immediately prenominal clitic pers _n marking and xnP-final demonstratives, which can co-occur (55a). The clitic pers _n marking can also be preceded by a full pronoun (55b). While I found no examples containing all three deictic expressions simultaneously, three distinct loci are involved, raising the same issues for the SIC as the Hausa example (47d) above.

While the assumption that demonstratives and pers _n form one distributional category seems justified for a substantial number of languages, this section has addressed a range of apparent or real counterexamples. Some directionality mismatches can be explained by the occurrence of pers _n in marked demonstrative positions. The discussion of PPDCs has shown that matching directionality is not a sufficient condition for co-categoriality and that a number of languages seem to indeed encode pers _n separately from demonstrativity. Jenks and Konate (2022) propose that even such distinct realisations can be structurally packaged in specifier-head configurations to align with the SIC. This constraint is conceptually appealing and the analysis is probably applicable to a range of languages, but several datapoints are not obviously compatible with the required configurations. I therefore still consider analyses with distinct projections for pers _n and demonstratives to be a viable and empirically necessary option (Giorgi and Pianesi 1997; Hsu and Syed 2020; Höhn 2017).