Referentiality and modifiability of incorporated nouns

2.1 Referentiality

The term referentiality, or reference, has a long history in the linguistic literature and is used in highly different ways (Chen 2009: 1657). A primary distinction can be made between a semantic and a pragmatic notion of referentiality (Abbott 2017: 240; see also Keizer 2015: 83). Very generally, a linguistic expression is semantically referential if its semantics inherently point at an entity in the world (Abbott 2010: 3, 2017: 240; Chen 2009: 1658). Pragmatic referentiality, by contrast, concerns the way a speaker uses a linguistic expression in context: a linguistic expression is pragmatically referential if a speaker uses it, in a particular discourse, to point at an entity (Abbott 2010: 2, 2017: 240; Chen 2009: 1659).

In FDG, the term referentiality is used for the pragmatic notion that pertains to the way in which nominal expressions are used in context (Hannay and Hengeveld 2009: 105, referring to Keizer 2015: 83; Lyons 1977: 177; Hengeveld and Mackenzie 2008: 107, referring to Dik 1978: 55, 128). The framework makes a pragmatic distinction between nouns that are used by a speaker to evoke an entity as a referent and nouns that a speaker employs to ascribe a property or entity (Hengeveld and Mackenzie 2008: 108–109, 113, 192–193; Keizer 2015: 83, 90–91). Because the pragmatic referentiality of a noun is determined by the way a speaker uses it, one and the same noun can both have a referential and a non-referential usage, in different contexts (Hannay and Hengeveld 2009: 105). Thus, whereas girl in example (2) is a referentially used noun, as it, together with the article the and the adjective intelligent, is used to evoke an entity as a referent, the noun girl in example (3) is a non-referentially used noun, as the noun phrase an intelligent girl is here only used to ascribe the entity ‘an intelligent girl’ to Hannah.

Referentially used nouns typically represent arguments, as in (2), or adjuncts, while non-referentially used nouns often function as predicates, as in (3). Nevertheless, there is no direct relation between the pragmatic referentiality of a noun and its status as argument or adjunct, on the one hand, or predicate, on the other hand. Although referentially used nouns never predicate by themselves, they may be part of larger, relational predicative expressions.^[2] Thus, in (4), the noun Shakespeare has a referential function but is at the same time part of the predicate by Shakespeare (Hengeveld and Mackenzie 2008: 190). Moreover, non-referentially used nouns may also represent arguments or adjuncts, as exemplified by piano and bike in (5) and (6) respectively.^[3]

Importantly, in FDG not only nouns that are used to refer to specific entities but also nouns that are used to evoke non-specific entities are considered to be pragmatically referential (Hannay and Hengeveld 2009: 112; Hengeveld and Mackenzie 2008: 113, 122; Keizer 2015: 95–96). Thus, both the noun cottage in (7), which is used to refer to a particular cottage that the speaker can identify, and the noun cottage in (8), which is used to evoke “any entity that fits the description” (Rijkhoff 2002: 235), are used referentially.^[4]

In this respect, the FDG notion of referentiality differs from some other pragmatic notions of referentiality in which only nouns used to refer to specific entities are considered to function referentially (Chen 2009: 1659; Lyons 1999: 165; Payne 1997: 264). Payne (1997: 264) argues, for instance, that a noun is only used pragmatically referentially “if it exists as a bounded, individuated entity in the message world”. When a noun is used non-specifically, it is not clear if a corresponding entity really exists.

The reason why FDG nevertheless regards both nouns like cottage as used in (7) and nouns like cottage as used in (8) as referentially used nouns is that they are both available for anaphoric reference (Hengeveld and Mackenzie 2008: 114). Thus, not only the specific entities evoked by the nouns girl in (9a) and cottage in (9b) but also the non-specific entity referred to by the noun cottage in (10) can function as the antecedent for an anaphoric pronoun.

By contrast, non-referentially used nouns do not evoke entities that can function as antecedents for regular anaphoric reference, whether they are used as predicates (11a–b), arguments (11c) or adjuncts (11d) (Hengeveld and Mackenzie 2008: 114; Keizer 2015: 91). Correspondingly, anaphoric reference is either inappropriate, as in (11c–d), or a special type of anaphor is required that does not refer back to a referentially evoked entity but only to an entity or property that is ascribed, as in (11a–b) (Hengeveld and Mackenzie 2008: 121, 193; see also Chen 2009: 1663; de Swart and Zwarts 2009: 289; Doron 1988: 284).^[5]

Similarly, referentially used nouns can be co-referential with other referentially used nouns (Rijkhoff 2008: 798; see also Dik 1997: 130), while this is not possible for nouns that are used non-referentially. In example (12), the noun cat in the second clause is clearly used to refer to the same cat as the noun cat in the first clause, as emphasized by the demonstrative modifier that. By contrast, in example (13), the noun girl in the first and second clause cannot be co-referential, because girl in the first clause is not used to refer.

FDG thus uses anaphoric reference and co-reference as tests for referentiality. In addition, referentially used nouns can be recognized based on their ability to combine with elements that mark an entity as identifiable or non-identifiable for the addressee.^[6] A speaker may use a noun referentially in order to re-identify an entity that is already identifiable for the addressee or to introduce an entity into the discourse that, at that moment, is not identifiable for the speaker. Definite articles, demonstrative modifiers and interrogative modifiers indicate that a speaker assumes that the entity he or she refers to by means of a particular noun is identifiable for the addressee (Hengeveld and Mackenzie 2008: 122; Keizer 2015: 95; Rijkhoff 2008: 797, 807–808; see also Dik 1997: 180; Lyons 1999: 18; Payne 1997: 102, 263). Interrogative pronouns, deictic personal pronouns and proper nouns heading a noun phrase also present the evoked entity as identifiable for the addressee (Hengeveld and Mackenzie 2008: 117–119, 122–123; Keizer 2015: 93–94; see also Lyons 1999: 21; Payne 1997: 39, 263).^[7] Indefinite articles and indefinite pronouns, by contrast, can signal that the entity referred to is assumed to be non-identifiable for the addressee (Hengeveld and Mackenzie 2008: 122–123; Keizer 2015: 46, 91).^[8]

Finally, referentially used nouns can be identified based on their ability to form a noun phrase with a possessive noun or pronoun that is used to refer to a possessor entity that is assumed to be identifiable for the addressee.^[9] Crucially, based on the possessive relation between the possessor entity that is presented as being identifiable for the addressee and the possessed entity, the addressee is assumed to be able to identify the possessed entity as well (Rijkhoff 2008: 808–809; see also Lyons 1999: 23–24; Payne 1997: 263–264). The assumed identifiability of the possessor entity is often indicated in the same way as for other entities that are expected to be identifiable for the addressee: the possessor entity may be expressed by a noun combined with a definite article, demonstrative modifier or interrogative modifier or by a pronoun or proper noun (Rijkhoff 2008: 808–809; see also Lyons 1999: 24; Payne 1997: 263–264). An example of a possessive noun that is used to refer to an entity that is identifiable for the addressee is included in (14).

In this example, the possessor entity is referred to by means of the proper noun Leila, which shows that it is assumed to be identifiable for the addressee. By expressing a possessive relation between Leila and the possessed noun fiancé, the speaker presents the entity evoked by fiancé as identifiable for the addressee as well. The speaker thus uses this noun referentially.

2.2 Modifiability

In FDG, the issue of modifiability is closely related to the semantic distinction between what can be called property-denoting nouns and entity-designating nouns. A property-denoting noun is a noun that heads a noun phrase that only denotes a property, which has no independent existence and can only be evaluated in terms of its applicability (Hengeveld and Mackenzie 2008: 131; Keizer 2015: 105). By contrast, an entity-designating noun functions as the head of a noun phrase that designates an entity to which the property that the noun expresses applies and which may be evaluated in terms of its existence (Hengeveld and Mackenzie 2008: 131, 215; Keizer 2015: 105).^[10] Thus, the noun president in example (15) is a property-denoting noun, as it is not used to designate an entity but only denotes a property that happens to apply to that man, whereas president in example (16) is used to designate a particular entity by expressing that the property ‘president’ applies to it.

As exemplified for president in (15) and (16), nouns can in principle be used both as property-denoting and as entity-designating nouns, in different contexts.^[11] Note also that property-denoting nouns may not only be nominal predicates, as in (15), but may also function as arguments or adjuncts, as shown by town and prison in (17) and (18) respectively. At the same time, entity-designating nouns can be used predicatively, as exemplified by doctors in (19), or represent adjuncts, like knife in (20), just as well as they can be arguments, as in (16) above.

Crucially, property-denoting nouns contrast with entity-designating nouns in terms of modifiability. Property-denoting nouns can generally not be modified (Hengeveld and Mackenzie 2008: 230–231; see also de Swart et al. 2005: 452; Stowell 1991: 50–51), as shown in example (21).^[12]

By contrast, entity-designating nouns can combine with both grammatical and lexical modifiers.^[13] Grammatical modifiers may take the form of grammatical number, possessive and diminutive marking, grammatical quantifiers and demonstratives (Hengeveld and Mackenzie 2008: 245–247; Keizer 2015: 158–159; Rijkhoff 2008: 795; Smit 2005: 103). Lexical modifiers include adjectives, restrictive relative clauses, participial clauses, possessive modifiers, adpositional phrases and lexical numerals (Hengeveld and Mackenzie 2008: 241–245; Keizer 2015: 156–158; Rijkhoff 2008: 794–797; Smit 2005: 103). Note that the expression of these modifiers as bound morphemes forming a single word with the noun they modify, as independent words within the noun phrase of the modified noun or as separate appositive noun phrases is a morphosyntactic issue that is not relevant for the semantic modifiability of nouns (Hengeveld and Mackenzie 2008: 115, 297–298).

With respect to grammatical possessive marking and lexical possessive modifiers, a difference has to be made between alienable possessors and inalienable possessors (Hengeveld and Mackenzie 2008: 243, 306–307; see also Nikolaeva and Spencer 2012: 207; Rijkhoff 2002: 87; von Prince 2016: 71). The alienable possessor student in example (22) is an optional addition to the entity-designating noun teacher and can as such be called a possessive modifier.

By contrast, the possessive relationship between the inalienably possessed noun brother and its possessor king in (23) is inherent to the property of being a brother: the king’s brother is only a ‘brother’ because of the specific family relationship with the king.

The possessor king is therefore not an optional modifier of the entity-designating noun brother but rather an obligatory argument of the property brother that is here used to designate the entity. Correspondingly, property-denoting nouns may express inalienable but not alienable possession, as shown in (24) and (25), in which the nouns brother and teacher are used as non-modifiable property-denoting nouns.

Because inalienable possessors are considered arguments of properties rather than modifiers of entities in FDG, in this study they are not taken as evidence for the modifiability of nouns.

2.3 Pragmatic-semantic typology of common nouns in FDG

Based on the FDG approach to referentiality and modifiability, a three-way typology of pragmatic-semantic usages of common nouns with different combinations of referentiality and modifiability characteristics can be proposed (Smit 2005: 102–103; see also Genee 2018: 258–259).^[14] This typology is presented in Table 1.

Prototypically, nouns are pragmatically referential and semantically modifiable, i.e. entity-designating (Hengeveld 2008: 46). Such nouns, abbreviated here as +R/+M nouns, are exemplified in (26).

The possibility to refer back to the noun cottage as used in (26a) by means of the anaphor it shows that this noun is used referentially. In addition, the presence of the adjectival modifier lovely indicates the modifiability of this noun cottage. The noun president in (26b) is also a referentially used modifiable noun, as demonstrated by the definite article the and the adjectival modifier tall.

A second pragmatic-semantic possibility for nouns is to be used as non-referential but modifiable entity-designating nouns, i.e. −R/+M nouns, as demonstrated in example (27).

The nouns girl and doctors in example (27) are used non-referentially: they can only be referred back to by using special anaphors such as so and that, which relate to ascribed properties and entities rather than to entities that are evoked as referents. The adjectival modifiers intelligent and fine indicate that girl and doctors are modifiable. Moreover, the noun doctors includes plural marking, which also shows the modifiability of that noun.

Thirdly, nouns may be non-referentially used non-modifiable nouns. For these property-denoting nouns the abbreviation −R/−M can be used. Two examples are shown in (28).

In example (28a), the noun president cannot be combined with the demonstrative modifier that without changing its pragmatic referentiality: without the demonstrative, president is used to ascribe the property ‘president’ to Roosevelt, i.e. it is used non-referentially, whereas the addition of the demonstrative gives the noun president a referential function and makes it co-referential with Roosevelt. The use of the adjectival modifier good is not possible either, i.e. president in (28a) is also non-modifiable. In (28b), the noun prison cannot be referred back to by the anaphoric pronoun it and cannot be modified by the adjectival modifier crowded, which shows that it is also a non-referentially used non-modifiable noun.

The typology presented in Table 1 and the examples in (26)–(28) show that there is no one-to-one relation between the pragmatic referentiality and semantic modifiability of nouns. On the one hand, referentially used nouns are always modifiable because non-modifiable, property-denoting nouns necessarily function non-referentially (Smit 2005: 102): they do not represent entities that can be referred to. On the other hand, non-referentially used nouns may either be modifiable or non-modifiable.

3.1 Sampling procedure

The variation in the referentiality and modifiability of incorporated nouns is examined in this paper on the basis of a sample of 21 languages. As the study focuses on noun incorporation, the sampling procedure was designed to include incorporating languages only. On the basis of previous research a list of incorporating languages was set up, from which the sample could be drawn. The sources used to compile this list were Velupillai’s (2012b) survey of languages with noun incorporation, which is based primarily on Mithun (1984), Gerdts (1998) and Aikhenvald (2007); several theoretical studies on incorporation (Anderson 2000; Baker 1988, 1996; Rosen 1989; Sadock 1980, 1985, 1986; Sapir 1911); overview articles on incorporation including Mithun (1994, 2010, Iturrioz Leza (2001), Anderson (2007) and Massam (2009); and the typological incorporation studies by Caballero et al. (2008), Štekauer et al. (2012) and Barrie and Mathieu (2016). In addition, languages were added to the list on the basis of a search of the Linguistic bibliography (Bobyleva et al. n.d.) and the Modern Language Association international bibliography . This procedure resulted in a list of 248 languages that were described as showing incorporation. This list is included in Appendix 1.

Subsequently, a 30-language variety sample was drawn from this list. Variety samples are aimed to include as much of the existing linguistic variation regarding a particular linguistic phenomenon as possible in order to enable a cross-linguistic exploration of that phenomenon (Bakker 2011: 104; Croft 2003: 21; Rijkhoff and Bakker 1998: 265; Velupillai 2012a: 50). A variety sample generally includes genealogically, geographically and typologically diverse languages, as it is assumed that this way of sampling yields the highest chance of capturing all existing variation when little is known about the variation regarding the phenomenon under consideration (Croft 2003: 21; Hengeveld 2006: 46–47; Moravcsik 2013: 18; Velupillai 2012a: 50). Correspondingly, the sampling procedure in the current research also took into account the genealogical background, geographical distribution and typological properties of incorporating languages.

For both the genealogical and the geographical diversity the language classification from Glottolog (Hammarström et al. 2017) was used. The 248 languages in the list of incorporating languages belong to 79 different language families, including 10 isolates. The sample languages were selected in such a way that the sample contains languages from 30 different families and thus shows genealogical variation. Regarding the geographical diversity, the spread of the languages over the six macro-areas identified in Glottolog (Africa, Australia, Eurasia, North America, Papunesia and South America) was considered. Importantly, incorporating languages are not distributed evenly over the different macro-areas. As the geographical distribution of the sample languages was aimed to reflect the spread of the 79 incorporating language families over the world, the six macro-areas are not all represented by an equal number of languages. The calculation of the proportion of language families in the list of incorporating languages in each macro-area and the corresponding number of languages from each area in the sample is presented in Appendix 2.

The typological feature taken into consideration concerns the variation in the forms of incorporated nouns. Although some assume that only nominal stems can be incorporated (Baker 1988: 71–72; Gerdts 1998: 85; Mithun 2000: 917), currently incorporated derived or compounded nouns, nominal inflected words and noun phrases also receive attention in cross-linguistic studies (Aikhenvald 2007; Barrie and Mathieu 2016; Iturrioz Leza 2001; Muro 2009; Olthof 2020). As more complex nominal forms seem to be more rarely involved in incorporation (Aikhenvald 2007: 12–13; Smit 2005: 94), four languages known to allow highly complex incorporated nominal elements, i.e. Bininj Kun-Wok (Hengeveld and Mackenzie 2008: 415), Chukchi (Hengeveld and Mackenzie 2008: 415–416), Crow (Barrie and Mathieu 2016: 33–34) and Eastern Ojibwa (Barrie and Mathieu 2016: 17–18), were deliberately included to make the sample typologically varied.

In addition to the genealogical background, geographical distribution and typological properties of the languages, the definition of noun incorporation used in the present study, to be introduced in the next subsection, was important in the sampling procedure. Authors use highly different definitions of noun incorporation (Haugen 2015: 414; Johns 2017; Massam 2009: 1077; Murasugi 2014: 284), such that the list of 248 incorporating languages includes languages with greatly varying incorporation-like constructions. For the sample, however, only languages were selected that show constructions that can be considered to involve noun incorporation according to the present study’s definition.

Finally, the amount of available data was taken into account in the sampling procedure. 30 languages were selected for which a reasonable set of data sources, including reference grammars and articles on incorporation, could be used. Additionally, experts on several of the relevant languages were consulted during the data collection. However, the process of data collection showed that for 9 of the 30 sample languages the available data on the referentiality and modifiability of incorporated nouns was ultimately insufficient for the data analysis. Consequently, the results presented in this paper only concern the remaining 21 languages. Table 2 shows the sample and additionally indicates for which languages sufficient data were available to include them in the study. The consulted data sources for these 21 languages are listed in Appendix 3.

Although three of the four languages that were included because of their formally complex incorporated elements, i.e. Chukchi, Crow and Eastern Ojibwa, had to be excluded from the final sample due to insufficient data, the sample languages still show varied forms of incorporated nouns, including simple stems, derived or compounded stems, inflected words and morphosyntactic phrases (Olthof 2020).

3.2 Definition of noun incorporation

As studies on noun incorporation do not always target the same set of constructions, a precise definition of the phenomenon of noun incorporation must be given that can be used to select the relevant constructions from the different sample languages in a systematic way. The present study defines noun incorporation on the basis of both semantic and morphosyntactic characteristics. Semantically, incorporated nouns and the verbs in which they are incorporated are in a dependency relation of the form head-modifier or predicate-argument (see also Haugen 2015: 414–415; Mithun 2000: 917). A head-modifier relation between an incorporated noun and an incorporating verb is shown in example (29) from Bininj Kun-Wok, in which yaw ‘child’ modifies ni ‘sit’.

A predicate-argument relation is exemplified in (30) from Mapudungun. Here, waka ‘cow’ functions semantically as an argument of the predicate kintu ‘seek’.

Morphosyntactically, an incorporated noun forms a single word with its incorporating verb (see also Caballero et al. 2008: 385). There is no general agreement about the criteria for morphosyntactic or grammatical words (Dixon and Aikhenvald 2002: 18–25; Haspelmath 2011: 38–59, 2018: 313–314, 317), but the present study uses three types of evidence for the morphosyntactic word status of incorporation constructions. Firstly, verbal inflectional affixes and verbal pro- and enclitics may show that a noun is incorporated, as a noun appearing between a verbal affix or clitic and the stem of a verb must be incorporated into this verb (see also Caballero et al. 2008: 385).^[15] Thus, in example (31) from Yucatec Maya the position of the noun che’ ‘tree’ relative to that of the verbal stem ch’ak ‘cut’ and the verbal suffixes -nah and -en indicates that it is part of the morphosyntactic verbal word, i.e. is incorporated.

Similarly, the Nadëb verbal inflectional proclitic ta= demonstrates that, in example (32), tʉí ‘food’ is morphosyntactically incorporated into the verb with the stem tɨɨ ‘fish’ (Weir 1990: 331).

Secondly, some languages make use of morphosyntactically bound verbs that obligatorily attach to a noun that functions as their argument or modifier. When a noun is directly preceded or followed by such a verb, it is here considered to be incorporated. For instance, many Kalaallisut verbs, including -si ‘receive’, are bound (Fortescue 1980, 1984: 320–324), such that allagar ‘letter’ in example (33) must be incorporated.

Some authors claim that verbs like -si should be analyzed as derivational affixes rather than as verbs and, correspondingly, call constructions with these affixes denominal verbs (Gerdts 1998: 97–98; Kurebito 2001; Mithun 1986: 32; Sapir 1911: 254; Stonham 2008: 513–514). However, the present study recognizes both unbound and bound verbal morphemes (Hengeveld and Mackenzie 2008: 404, see also Delahunty and Garvey 2010: 132), such that bound elements with a highly lexical, verbal meaning, like -si in (33), are considered incorporating verbs. In this respect, the present study also follows Sadock (1980, 1985, Caballero et al. (2008: 409), Barrie and Mathieu (2016: 10), Johns (2017) and several others (see Massam 2017).

Thirdly, in several languages, incorporated nouns have a special form that is only used in the context of noun incorporation (Caballero et al. 2008: 387–388; Mithun 1984: 875–876). This form can then show that a noun is morphosyntactically incorporated into the verb. Thus, in Halkomelem the noun qeq ‘baby’ takes the form -əyeɫ when it is incorporated, which means that -əyeɫ in example (34) is an incorporated noun.

Elements like -əyeɫ in (34) are sometimes argued to be derivational affixes rather than suppletive forms of independent nouns (Bischoff 2011: 15; Gerdts 1998: 94–97; Sapir 1911: 251–252). However, because these affixes have lexical meanings and are large in number, they are nevertheless counted as incorporated nouns in the present study, in the same way as in Aikhenvald (2007: 13), Caballero et al. (2008: 387–388) and other studies (see Massam 2017).

The definition of noun incorporation employed in this study does not specify any phonological requirements. It is therefore assumed here that a combination of incorporated noun and incorporating verb does not have to form a single phonological word (see also Aikhenvald 2007: 14; Massam 2017; Mithun 1984: 849–845). Thus, the Nadëb example in (32) above is regarded as an incorporation construction because the noun tʉí ‘food’ and the verb tɨɨ ‘fish’ form a predicate-argument relation and constitute a single morphosyntactic word, even though the noun and verb remain independent phonological words as evidenced by stress placement (Weir 1990: 323). Constructions like (32) have also been called juxtaposition (Mithun 1984: 849), loose incorporation (Miner 1986: 252) and pseudo-incorporation (Massam 2009: 1087), but are included within the domain of noun incorporation in the present study (see also Aikhenvald 2007: 14–15; Caballero et al. 2008: 385–386; Massam 2017).

Finally, incorporation is here considered to be a grammatical process that is distinct from lexical compounding (Hengeveld and Mackenzie 2016: 1150–1153; see also Baker 1988: 78–80, 1996: 307–308; Barrie and Mathieu 2016: 4–5, 23). Incorporation is as such restricted to productive processes that yield verbs with semantically predictable meanings, whereas lexical compounding has limited productivity and creates verbs with possibly non-transparent semantics. In addition, the nominal components of lexical compounds necessarily have a non-referential function and cannot take modification, in contrast to incorporated nouns, of which the potential pragmatic referentiality and semantic modifiability is examined in this paper. Constructions that classify as lexical compounds are thus excluded from the study.

3.3 Data analysis

In this study, the referential potential and modification possibilities of incorporated nouns in the sample languages were examined on the basis of the available examples of incorporation constructions in these languages and descriptive information about their characteristics. For each language it was investigated to which noun types its incorporated nouns belong: +R/+M nouns, −R/+M nouns and/or −R/−M nouns. As the present study does not only focus on cross-linguistic variation but also on intra-linguistic variation, for each language it was necessary to find either positive or negative evidence for the existence of incorporated nouns of each of the three noun types.

Evidence for the occurrence of referentially used incorporated nouns in a particular language consisted of at least one grammatical example of an incorporated noun showing at least one of the characteristics of referentially used nouns described in Section 2.1. By contrast, a language was considered to show incorporated nouns with a non-referential function if at least one ungrammatical example of an incorporated noun showing at least one of these characteristics was found. The relevant characteristics include availability for anaphoric reference, ability to appear in a co-referential relation with another noun, ability to combine with an element marking the entity evoked by the noun as identifiable for the addressee, and ability to combine with a referential possessor.^[16] The possibility vs. impossibility for incorporated nouns to show one or more of these characteristics could also be verified on the basis of descriptive information given by the language expert. In addition, if descriptive information was available stating that incorporated nouns in the language could or could not have a specific or definite interpretation, this information was also used to determine the pragmatic referentiality of incorporated nouns in the language.^[17] Regarding modifiability, the presentation of at least one grammatical example of an incorporated noun combined with a grammatical or lexical modifier was used as evidence for the occurrence of modifiable incorporated nouns, while an ungrammatical example of an incorporated noun with such a modifier was taken to show that a language makes use of non-modifiable incorporated nouns. The grammaticality or ungrammaticality of the use of such modifiers could also be determined based on descriptive information.

It should be emphasized that in this study, modifiability is considered a semantic issue, such that the morphosyntactic characteristics of apparent modifiers of incorporated nouns are not relevant for the issue of modifiability. Modifiers usually appear either morphologically attached to the noun they modify or syntactically adjacent to it in a noun phrase. Correspondingly, some languages allow modifiers to incorporate together with the noun they modify and to appear next to it in the incorporation construction, as shown by the adjectival modifier kurü ‘black’ in example (35) from Mapudungun, the plural infix in example (36) from Nuu-chah-nulth, and the possessive suffix -mi in example (37) from Kalaallisut.^[18]

Such incorporated modifiers are clear evidence for the modifiability of the incorporated noun.

In many other languages, by contrast, incorporation appears to be restricted to nominal stems, such that modifiers may not be incorporated. However, in some of these languages, lexical modifiers that appear to relate to the incorporated noun can occur external to the incorporation construction. Thus, in example (38) from Bininj Kun-Wok the demonstrative na-mekke looks like a modifier of the incorporated noun murrng ‘bone’, and in example (39) from Mohawk the relative clause nehneh a-ak-ahninuʔ ‘that she would buy’ seems to modify the incorporated noun nuhs ‘house’.

Importantly, there is disagreement in the literature on noun incorporation about the interpretation of such external modifiers. Whereas some claim that these modifiers indeed modify the incorporated nouns (Baker 1988: 92–105, 1996: 308; Barrie and Mathieu 2016: 4; Sadock 1980: 307–310, 1991: 91–99; Van Geenhoven 1998: 17–22), others maintain that such modifiers constitute separate noun phrases without a nominal head that are completely independent of the presence of an incorporated noun (Barrie 2010: 293–294; Di Sciullo and Williams 1987: 65–66; Mithun 1984: 865–866, 870; Rosen 1989: 298). As the present study separates the morphosyntactic position and expression of modifiers from the issue of modifiability, which concerns semantics, it is considered irrelevant whether a modifier is incorporated together with the noun it modifies or appears external to the incorporation construction: both types of modifiers are regarded as evidence for the modifiability of incorporated nouns.^[19] Interestingly, however, only for two sample languages the inclusion of external modifiers turned out to affect the results (see Section 4).

One special type of external modifier is exemplified by the second person pronoun õm from Nadëb in (40b).

The unincorporated noun mooh ‘hand’ in (40a) is combined with the possessive pronoun a. By contrast, in example (40b), in which mooh is incorporated into the verb, this possessive pronoun is replaced by the regular pronoun õm, which is morphosyntactically the absolutive argument of the incorporating verb (Weir 1990: 323). Nouns and pronouns designating apparent possessors that are expressed as arguments of main verbs, like õm in (40b), are known as external possessors (Aikhenvald 2013: 36; Herslund and Baron 2001: 14–15; Payne and Barshi 1999: 3).

External possessors that appear in the context of noun incorporation can be analyzed in different ways. Some researchers claim that, despite their special morphosyntactic expression, these external possessors constitute semantic units together with the nouns that designate the entities that they are assumed to possess (Gerdts 2003: 352–355; Van de Velde 2013: 172–173, see also Allen et al. 1984: 306–307; Baker 1988: 96–105, who argue that the external possessors and possessed nouns form a single unit underlyingly). Such an analysis entails that an external possessor can be considered to show that an incorporated noun is modifiable. Alternatively, however, external possessors like õm in (40b) may simply be analyzed as arguments of incorporating verbs that are independent of the incorporated nouns involved (Mithun 1984: 856, 859), in which case external possessors do not form evidence for the modifiability of incorporated nouns.

Importantly, external possession is typically limited to inalienable possessive relations (Aikhenvald 2013: 36; Herslund and Baron 2001: 15). As nouns designating inalienable possessors are possessive arguments of properties rather than possessive modifiers of entities in FDG (see Section 2.2), inalienable external possessors used in the context of noun incorporation do not provide evidence for the modifiability of incorporated nouns, irrespective of the choice between the two possible analyses of external possessors just described. Because the data from the sample languages indeed only show inalienable external possessors, for the question of modifiability the present study can simply leave external possessors aside. However, whether or not external possessors can present evidence for the referentiality of incorporated nouns depends on the pragmatic analysis of these possessors. Because external possessors are often argued to express a special affectedness or empathy on the side of the speaker (Van de Velde 2013: 167, referring to O’Connor 2007), they may be considered to differ from regular attributive possessive nouns and pronouns in terms of their pragmatics. Regular attributive possessive nouns and pronouns form a single pragmatic unit with their possessed entities. Correspondingly, they have a shared referential status, in that a noun corresponding to an entity possessed by a referentially used attributive possessive noun or pronoun necessarily also has a referential function (see Section 2.1). By contrast, it seems suitable to analyze external possessors as independent pragmatic units, as they can be specified separately for pragmatic affectedness or empathy (Van de Velde 2013: 172–173). Their possible referential pragmatic status is then also independent of the referentiality of the incorporated possessed nouns, and external possessors are therefore not taken to provide evidence for the referentiality of incorporated nouns either, even if these possessors are used referentially.