Word Classes in Australian Languages (2024)

26.1 Introduction

At the time of European settlement in the 18th century, approximately 250 distinct indigenous languages or 700 to 800 language varieties were spoken in Australia (Dixon 2002: 5–7; Koch & Nordlinger 2014). Today, only 12 traditional languages are still being learned by children (Commonwealth of Australia 2020) and another 100 or so are being spoken by the elderly generation (Koch & Nordlinger 2014: 4). Documentation of these languages varies considerably, going from wordlists compiled in the early days of colonization to detailed grammatical descriptions and more extensive text collections. The degree of documentation is partly related to historical factors: languages from the south and south-east are for instance less well documented, as these areas were first affected by colonization (see Koch & Nordlinger 2014 for an overview).

Australian languages belong to roughly 25 language families. The largest family is Pama–Nyungan (PN), which includes roughly two-thirds of all Australian languages and covers about 90% of the continent (Bowern & Atkinson 2012: 817). Internally, there is no consensus on higher-level subgroups, but many lower-level groups are well supported; see Bowern & Atkinson (2012) for a recent account. The other 24 families and isolates, found in the north and north-east of the continent, are sometimes collectively labelled ‘non-Pama–Nyungan’ but are not related to each other (at least as far as evidence can tell us). These families are much smaller: the biggest is probably Gunwinyguan, with about a dozen languages, while there are also several isolates, like Tiwi and Limilngan. See Evans (2003b) for the most recent classification; note that Anindilyakwa has since been reclassified as Gunwinyguan (Van Egmond 2012). See Koch (2014) for a recent overview of issues related to genetic classification in Australian languages.

The typical part-of-speech system of Australian languages includes, according to Dixon’s (2002: 66) survey of these languages, two major sets of word classes, viz. nominal and verbal ones, and a residue set of smaller classes. These sets of word classes include the following:

The sets are grouped together mostly based on morphological criteria: all nominal parts of speech have the same or similar case marking at least for peripheral roles, while verbal parts of speech take inflections for tense, aspect or mood; coverbs are usually uninflected, however (Dixon 2002: 67–71) (hence the alternative label ‘uninflecting verbs’ (UVs)). Particles and other classes do not inflect at all (Dixon 2002: 66). The word classes within each set are argued to be morphologically distinct from each other as well.

There are a few issues with Dixon’s description. First, the morphological criteria discussed in the previous paragraph are not watertight, as a nominal word may occur with verbal affixes and vice versa (e.g. McGregor 2004: 102–103, on languages spoken in the Kimberley area). For instance, in Gooniyandi the word yoowooloo ‘man’ usually occurs with case markers, but may also occur with a verbal classifier as in yoowooloo-windi ‘he became a man’; the word ward- ‘go’ usually occurs with verbal affixes but may also occur with a case marker (McGregor 2004: 102; see also McGregor 1990: 560–563). The morphological criteria do work with some refinement, however: nominal words occur with nominal morphology most of the time and can only take a small subset of verbal morphology, and the other way around (McGregor 2004: 102). This is further discussed in section 26.2.

Second, the list above includes separate word classes for nouns and adjectives, but at the same time, Dixon also argues that nouns and adjectives ‘generally show the same morphological and syntactic possibilities, so that it can be difficult to give criteria for recognizing them as distinct classes’ (2002: 67). In fact, much of the Australianist literature agrees with the latter position, and argues that Australian languages typically lack a clearly distinct word class of adjectives and instead have a single, flexible class of elements which can be used both referentially and attributively, usually labelled ‘nominals’1 (e.g. Hale 1983: 33–36; McGregor 2004: 102; Nordlinger 2014: 237–238). However, a more recent study has shown that, while a substantial number of languages indeed has such a single flexible class, many languages also have distinct (sub)classes of nouns and adjectives based on morphological and/or distributional criteria (Louagie 2020: 66–83). This issue is taken up in section 26.3.

Third, inflecting verbs (IVs) and uninflecting verbs (UVs) are included as distinct verbal parts of speech, but it is not clear how they relate to the function of predication. Schultze-Berndt (2017) argues that IVs are the only elements that can be used for both independent and dependent predication while UVs are restricted to dependent predication (defined by Schultze-Berndt to include use as part of a complex predicate). Moreover, McGregor (2002: 252–266) argues that in many languages an IV, when used in combination with an UV, does not in fact function as a predicate but rather as a marker of verb classification. In this way, IVs are problematic for part-of-speech typologies where verbs are defined as being used for predication only (Hengeveld 1992a, 1992b). In addition, it is clear that the word class of simple or IVs is different in languages that also have a class of UVs compared to languages that do not (e.g. in terms of the concepts that are covered or the open/closed character). Note also that Dixon’s classification of (manner) adverbials as one of the verbal parts of speech is only relevant for a small number of languages; in many, it seems that manner adverbials are a separate class of non-inflecting elements, or are more like nominals in that they may show case marking (e.g. Dench & Evans 1988: 14–16; Dixon 2002: 181–183). These issues are further discussed in sections 26.4 and 26.5.

Finally, Dixon’s (2002: 66–71) discussion of parts of speech does not really refer to distribution over syntactic functions like predication or head of a referential phrase, while this is an important parameter for many theories of word classes (e.g. Hengeveld 1992a, 1992b; Croft 2000). Syntactic and functional distribution is especially interesting where the first set of word classes, viz. the nominal ones, are concerned: the mapping of elements to functional roles like entity, qualifier, classifier and determiner is relatively flexible in a number of Australian languages (e.g. McGregor 1990; Wilkins 2000; Louagie & Verstraete 2016; Louagie 2017), and this may play a role in the characterization of word classes.

The rest of this chapter further surveys how lexemes are generally organized in word classes in Australian languages, specifically focusing on the issues mentioned in the previous paragraphs. The discussion centres around lexical elements that may be used as heads of referential phrases (referential use), as modifiers within referential phrases (attributive use), as predicates, and as modifiers of predicates (manner adverbial use), since these are also the main focus of this handbook. I will largely ignore functional word classes, but give some references where interested readers can find good overviews of what these classes typically look like in Australian languages. I will also not further discuss particles, ideophones, or interjections. Section 26.2 discusses the distinction of nominal from verbal parts of speech (specifically nouns from verbs) and how they may show rather minimal morphological and/or distributional overlap. Sections 26.3 and 26.4 then discuss each of these sets of word classes in more detail. Section 26.3 focuses on the different nominal parts of speech as listed by Dixon (2002), discussing whether Australian languages have a separate word class of adjectives or not (section 26.3.1) and whether a separate word class of classifiers or one of determiners can be distinguished (section 26.3.2). Section 26.4 turns to verbal parts of speech, specifically focusing on the distinction between IVs and UVs. Manner adverbials are briefly commented on in section 26.5. Section 26.6, finally, concludes the chapter with a general overview of the main parts of speech and issues discussed in this chapter. Not each part of speech is discussed equally thoroughly, for reasons of space. Most attention is devoted to the question whether Australian languages have distinct noun and adjective parts of speech, because this is a frequently discussed issue in individual descriptions and in continent-wide surveys, and because there are some misconceptions (or at least overgeneralizations) in the wider typological literature. I am also personally more familiar with nominal parts of speech than with verbal ones.

26.2 Distinguishing verbal from nominal parts of speech

Dixon (2002: 66) distinguishes two major sets of word classes in Australian languages, nominal and verbal ones, following earlier suggestions in the literature (e.g. Hale 1982 on Warlpiri). In this section, I summarize the motivations for distinguishing these sets, and for ease of discussion focus only on the distinction between the two main word classes of each set, viz. verbs on the one hand and nominals on the other hand (or nouns, depending on the language; see section 26.3.1).

In terms of syntactic distribution, nominals are mostly used for functions related to reference (e.g. as head or modifier within a referential phrase, see section 26.3.1), while verbs are used for predication. This is not a sharp division of labour, as nominals2 can generally also be used for predication (see e.g. McGregor 2005, 2013; Simpson 2005; Schultze-Berndt 2006; Nordlinger 2014: 239–240). The predicative use of nominals is illustrated in (1a) and use as secondary predicate in (1b). The same flexibility is not found the other way around, i.e. verbs cannot be used referentially, unless they undergo derivation. There are a few examples where verbs do seem to serve a referential function, as in (1c); these are analysed in grammars as embedded clauses which may or may not take nominal inflections (e.g. Evans 2003a: 122–124 on Bininj Kunwok; Dench 1994: 198 on Martuthunira).3 Note that verbs may have yet other functions in some languages; see section 26.4.

Verbs and nominals can be distinguished morphologically as well, although there is some overlap. Thus, only verbs can inflect for TAM and/or person (see section 26.4 on UVs), while nominals normally cannot. In some languages, however, nominals which are used predicatively may show some verbal inflection, like pronominal affixes or TAM marking. They are still morphologically distinct from the word class of verbs because they cannot take all types of verbal inflectional marking. In Bininj Kunwok, for example, a nominal used predicatively can take the past imperfective suffix -ni (2a), or pronominal prefixes (2b), but not the past perfective suffix; in addition, the form of the third person pronoun prefix is different when attached to a nominal or to a verb. See section 26.3.1 for examples of inchoative suffixes on nominals, deriving them into verbs.

Conversely, nominals in many languages regularly inflect for case,4 while verbs normally do not. However, verbs can also take nominal case marking in some languages, but only in contexts of subordination and insubordination (e.g. Dench & Evans 1988; Blake 1999; Evans 2007). Thus, case may appear on the verb and/or other elements of a subordinate clause (depending on the language), to specify the relation between the subordinate and the main clause. This function is labelled ‘complementising’ in Dench & Evans (1988). A straightforward example is given in (3a), where the dative case on the (non-finite) verb dupay- ‘sit’ marks a purposive relation; compare with (3b) where a dative on a nominal also indicates purpose.5^,6^,7

In fact, Dench & Evans (1988) distinguish two specific functions of complementizing case, which they label ‘t-complementiser’ and ‘c-complementiser’. T-complementizer functions mark temporal, spatial or logical relations between subordinate and main clauses, as with the purposive in (3a). Another example is found in (4a) where the locative marks a simultaneous temporal relation between the subordinate and the main clause. C-complementizer functions link a subordinate clause to its antecedent in the main clause through case agreement, thus marking coreference. This is illustrated in (4b), where each element of the subordinate clause, including the verb, is marked for ergative case in agreement with the antecedent ‘man’ (it is the man who intends to cook the wallaby). Nordlinger (2014: 244) hypothesizes that c-complementizer functions are only found in Australia.

One Australian language that presents a well-known and rather far-reaching case of morphological overlap between verbs and nominals is Kayardild (Evans 1995: 89–91). In Kayardild, nouns can inflect for ‘verbal case’: these verbalizing suffixes turn a word into a morphological verb that agrees in tense, mood, and polarity with the main verb, but at the same time they function as oblique cases, in the sense that they are attached to every word of the NP and code case-like meanings like beneficiary or purpose. This is illustrated in (5a), where a verbal dative -maru is attached to each word of the noun phrase to express a role of beneficiary. Because this is a verbalizing suffix, each word is also inflected for potential mood, in agreement with the main verb. Note that the reverse situation also occurs: verbs can be nominalized and then take nominal case, while still functioning as verbs syntactically, as illustrated in (5b). What is interesting here is that both verbal case and ‘ongoing nominalization’ operate at phrase level and thus are inflectional rather than derivational.

Finally, some languages have a small word class whose elements show features associated with nominal word classes and features associated with verbal ones. It concerns a class of predicate nominals or adjectives, which often have (some) nominal morphology but function predicatively only. This word class often only has a smallish number of members, expressing states like ‘know’, ‘be ignorant’, ‘be jealous’, ‘asleep’, or ‘pregnant’. For example, Kuuk Thaayorre has a category of predicate adjectives, which are like elements from nominal word classes in the sense that they cannot take TAM inflections and can function as complements for copula constructions, but like verbs in the sense that they cannot take case inflections, can only function predicatively, and take one or two arguments, as illustrated in (6) (Gaby 2017: 87–88, 377–379). The class only has two members which are used commonly: walmeerem ‘knowledgeable of’ and pamngongkom ‘ignorant of’ (Gaby 2017: 87–88). Similarly, in Kayardild, predicate nominals have nominal morphology (e.g. derivational possibilities), but can only be used predicatively and can sometimes take direct or indirect objects (Evans 1995: 86, 231–232). The class has more than 30 members (listed in the dictionary in Evans 1995: 638–800). Kayardild also has a small class of ‘manner nominals’ which similarly have nominal morphology but can only function as secondary predicates (Evans 1995: 86, 227–229), as kantharrk ‘alone’ in (7). A similar category is found for instance in Martuthunira (Dench 1994: 53) and Yankunytjatjara (‘active adjectives’; Goddard 1985: 17); see also section 26.5 on manner adverbs.

26.3 Nominal parts of speech

The previous section discussed how nominal parts of speech can generally be distinguished from verbal ones. This section focuses on nominal parts of speech, such as nouns and adjectives (if distinguished), demonstratives, personal pronouns, locationals, and temporals. These are sometimes grouped together as such (e.g. by Dixon 2002 but also in individual grammatical descriptions) because they share general inflectional possibilities such as case marking and regularly occur in referential expressions, as discussed in sections 26.1 and 26.2. Ignoring for now the question whether nouns and adjectives are separate parts of speech, the other categories listed above are usually analysed as separate (sub)classes, the argumentation for which I briefly outline in the following paragraphs.

Free8 personal pronouns are most clearly distinct from other nominal parts of speech such as nouns. They form a closed class of deictic elements,9 which obligatorily distinguish person and number (while nouns often only optionally mark number or collectivity (e.g. Dixon [1980] 2010: 275; Louagie 2020: 91–99)). Many languages show split case alignment between pronouns and nouns, or between first/second person pronouns and third person pronouns/nouns (Dixon [1980] 2010: 285–291; Nordlinger 2014: 224–227). In a substantial number of Australian languages, third person pronouns are multifunctional: they not only occur pronominally but also adnominally (Blake 2001; Stirling & Baker 2007; Louagie & Verstraete 2015; Louagie 2017). As modifiers, they almost always occur in a determiner role, and only rarely in another role, viz. as number marker (in Djapu (Morphy 1983: 47–48; Louagie & Verstraete 2015: 177–178)) or as emphatic focus marker (in Gooniyandi (McGregor 1990: 270) and Dalabon (Cutfield 2011: 54)). Modifying pronouns sometimes show signs of grammaticalization (in all known cases with number marking or focus marking functions, and in some cases with determiner function), and could perhaps even be analysed as belonging to a different part of speech in some cases (see Louagie & Verstraete 2015: 178–183 for discussion). The examples in (8) illustrate the three different distributions for personal pronouns in Dalabon (Cutfield 2011: 54, 96–99, examples): as head of a referential expression in (8a), as determiner in (8b), and as emphatic marker in (8c). When yibung is used as emphatic marker, its inherent number and person values do not apply, in the sense that the form is able to modify elements with other referential values, such as a first plural pronoun in (8c) (Cutfield 2011: 54). For surveys of the make-up of personal pronoun systems, see e.g. Dixon ([1980] 2010: 275–277; 2002: 243–319) and McGregor (2004: 110–118) on Kimberley languages.

Demonstratives may share morphological features with pronouns and/or nouns, or have their own, specific morphology, for example relating to case and number marking (Dixon [1980] 2010: 275–277; Louagie 2020: 91–97; Louagie 2023). In terms of distribution, Australian languages often have a set of demonstratives that can function both pronominally and adnominally (sometimes also adverbially); some have separate sets for these functions. See section 26.3.2 for some examples of adnominal demonstratives and the question whether a category of determiners can be distinguished. See Louagie (2023) for a survey of demonstratives in Australian languages; see also Dixon (2002: 335–336), and McGregor on Kimberley languages (2004: 125–126).

Locationals and temporals are two, often rather small, parts of speech including lexemes related to location and time. Although they mostly function adverbially, they are considered nominal parts of speech in many languages because they take nominal case marking and can sometimes occur as part of the nominal expression. They only take a subset of nominal cases, and sometimes have different allomorphs for the cases they do take (e.g. Dixon [1980] 2010: 282–283). Note that in some languages, temporal and locational adverbs do not inflect at all and are thus considered particles, and not nominal parts of speech (e.g. Bardi (Bowern 2012: 160)).

Having briefly introduced the parts of speech of personal pronouns, demonstratives, locationals, and temporals, I now turn to two questions which I will discuss in more detail. The first, discussed quite extensively in section 26.3.1, is whether Australian languages have separate noun and adjective parts of speech, or a single flexible part of speech whose members can function both referentially and attributively. The second, discussed more briefly in section 26.3.2, is whether separate word classes of classifiers or determiners may be distinguished among the nominal parts of speech. Both questions are specifically relevant in light of the flexibility a number of Australian languages show, especially in noun phrases but also beyond (e.g. McGregor 1990, 2013; Nordlinger 2014: 237–241; Louagie 2017, 2020).

26.3.1 Nouns and adjectives, or a flexible class of nominals?

There has been some discussion in the Australianist literature whether nouns and adjectives can be distinguished as separate parts of speech. For instance, in his first survey of Australian languages, Dixon ([1980] 2010: 274) argues that a distinct word class of adjectives may be identified, whereas in his second survey (Dixon 2002: 67), he argues there are few differences between putative nouns and adjectives so that they should be considered a single part of speech. The question is also discussed in many grammatical descriptions, using a variety of notional, morphological, syntactic, and/or functional criteria, and arriving at different conclusions based on similar criteria, or reverse, similar conclusions based on different criteria. The overall image that has found its way into the Australianist and typological literature is that Australian languages generally lack a noun–adjective distinction (e.g. Hale 1983: 83; Dixon 2002: 67; Nordlinger 2014: 237–238; Van Lier & Rijkhoff 2013: 7; see e.g. Hale 1982, 1983; McGregor 2013 for accounts of individual languages). However, a recent continent-wide survey based on a genetically diverse 100-language sample10 has shown that the picture is not uniform across Australian languages, and a substantial number of languages in fact have separate word classes for nouns and adjectives (Louagie 2020: 66–83); the rest of this section heavily relies on this study.

The rest of this section will first discuss some of the criteria used in the typological literature and in descriptions of individual languages for (not) distinguishing nouns from adjectives, surveying notional, morphological, and distributional criteria in turn (sections 26.3.1.1–26.3.1.3). This is followed by a brief discussion of how the values for these criteria pattern across the languages of Australia, based on the results of the survey in Louagie (2020: 66–83) (section 26.3.1.4).

26.3.1.1 Notional criteria

A number of grammatical descriptions broadly state that nouns in general refer to concepts and adjectives to qualities, or conversely, that a flexible class of nominals includes items of both types. If a distinction between nouns and adjectives is claimed for a particular language, it may be hard to make generalizations about which types of semantic notions are encoded as nouns or as adjectives. This is remarked on by several authors for Australian languages specifically (e.g. Harvey 2002: 130 on Gaagudju; Austin 2013: 43–44 on Diyari), but this issue has also been noted in wider cross-linguistic research (e.g. Dixon 1982: 1–62). It is clear we cannot use notional criteria as distinguishing factor for word classes alone. I will, however, briefly refer to issues pertaining to semantics at a few places in the following discussion where relevant. Note that I use the shorthand ‘property-denoting’ and ‘entity-denoting’ lexemes as general terms when I do not want to be specific about word-class status, fully realizing that this glosses over the nuances just discussed.

26.3.1.2 Morphological criteria

Morphological differences between putative nouns and adjectives are only found in approximately one-third of Australian languages, and have a fairly limited scope in most of these languages (Louagie 2020: 73–74; based on the 100-language sample mentioned above). The morphological criteria discussed in individual grammars, on which the previous statement is based, are quite diverse. One criterion is the need for derivation when an element is used in another syntactic function: some languages derive adjectives from nouns when they are used as modifier, or conversely, nouns from adjectives when they are used as head of a noun phrase. An example is found in Arabana/Wangkangurru, where reduplication of a noun may derive an adjective for attributive use (Hercus 1994: 96–99) as in (9a); compare with (9b) which shows the same noun as head of a noun phrase. Like many languages, Arabana/Wangkangurru also has a ‘having’ suffix which may be used among other things to derive adjectives from nouns (Hercus 1994: 90–92),11 as in (9c). Nouns without any such morphological adaptation are not used as modifiers in this language, with a few specific exceptions (e.g. close apposition of a noun and proper name where the former semantically modifies, or perhaps classifies, the latter, as in (9d) (Castel 2020: 22, 25)). The use of adjectives as head of the noun phrase is not attested (based on Hercus 1994: examples). The availability of means to derive nouns from adjectives and vice versa seems good evidence for distinct classes, since a flexible class of nominals would not require any morphological adaptation for use in different functions (following e.g. Hengeveld 1992a, 1992b). This type of derivation only occurs in a small number of languages, and is always restricted to a small part of the stock of entity- and property-denoting lexemes, which perhaps makes the evidence less strong (Louagie 2020: 75–76).

Other morphological criteria used in grammars relate to morphological potential. Many mention differences in derivational possibilities for either nouns or adjectives into other categories such as verbs. For example, for several languages it is mentioned that only adjectives, not nouns, can be used to form an inchoative or causative predicate, e.g. through affixation or as part of a complex verb construction. This is illustrated in (10) from Bardi, where only adjectives may occur in a complex predicate to form a stative (10a), inchoative (10b) or causative predicate; this is ungrammatical for nouns (10c)–(10d), and instead a structure with an allative case marker is needed to express a change of state, as in (10e) (Bowern 2012: 264–265). The question may be raised whether such differences are truly due to morphological restrictions, or rather an epiphenomenon of semantic differences, in which case they would not be a good argument for distinct part-of-speech status. For Bardi, alternative constructions are needed to express a change in state involving a noun or an adjective, so the morphological evidence seems solid. A similar example is found in Wambaya, where the causative suffix -mi can only be added to adjectives (e.g. guriny-mi ‘make good’) and not to nouns (e.g. *juwa-mi ‘make into a man’), not even when licensed by the context (e.g. in the context of Dreamtime stories) (Nordlinger 1998: 47–48). In other languages, causative derivation is possible both with entity-denoting and property-denoting lexemes, though perhaps more likely with the latter for semantic reasons (e.g. Garrwa (Mushin 2012: 170–175); Gumbaynggir (Eades 1979: 271)).

Other examples of morphological differences (all mentioned in Louagie 2020: 74–75) are that adjectives can derive adverbs while nouns cannot in Arabana/Wangkangurru (Hercus 1994: 60; see section 26.5 for an example), and coverbs can derive adjectives but not nouns in Bilinarra (Meakins & Nordlinger 2014: 81). Adjectives have an oblique stem in Warray, while nouns do not (Harvey 1986: 70). Finally, nouns and adjectives reduplicate to different effects in a number of languages, e.g. plurality and derivation into an adverb respectively in Kuku Yalanji (Patz 2002: 55), or plurality and intensification in Emmi (Ford 1998: 140), as illustrated in (11a)–(11b).

26.3.1.3 Distributional criteria

Distributional differences between nouns and adjectives are more common than morphological differences: they are found in over half of Australian languages, at least (Louagie 2020: 76–77; based on the 100-language sample mentioned earlier). Distributional criteria include both the distribution of elements with respect to the functions of reference and attribution (i.e. head or modifier in a referential phrase),12 and more specific (morpho)syntactic differences that may exist between nouns and adjectives.

Starting with the former, many grammars do not actually discuss distribution across functional roles in detail—in fact, many do not discuss functional roles at all (as one reviewer rightly pointed out). The evidence is often inconclusive and/or insufficient to draw any firm conclusions, but I will try to formulate some general tendencies and issues based on the available information. Firstly, many descriptions (often briefly) mention that property-denoting lexemes may occur as sole element of a referential expression, as in (12). Some descriptions analyse such elements as modifiers of ellipsed nouns, because these structures usually occur in contexts where the entity referred to is understood or established (e.g. Bowern 2012: 158 on Bardi). This type of analysis implies that property-denoting lexemes are not used referentially in themselves and are thus part of a distinct word class of adjectives (see also Chapter 18). Other descriptions, by contrast, analyse property-denoting lexemes in instances as (12) as head of a referential phrase (regardless of whether the referent is established). They consequently argue that both property-denoting and entity-denoting lexemes may function referentially, thus forming a single, flexible class of nominals (e.g. Eades 1979: 272 on Gumbaynggir). Their case is perhaps most convincing when any element can occur in head and modifier functions in contexts where the referent is not yet established, i.e. when there is no question of elliptical NPs (see Chapter 18). This is illustrated in (13) from Gooniyandi, where jiginya functions as head of the noun phrase in (13a) and as qualifying modifier in (13b); both NPs are not elliptical (McGregor 1990: 264).

When property-denoting lexemes occur in structures such as (12) and (13a), their regular meaning is ‘an X one’ (e.g. ‘a small one’, ‘a tall one’) across Australian languages.13 Sometimes the meaning may be more specific; for instance, while an element like Gooniyandi jiginya may be used to refer to any type of small entity, it is most frequently used in reference to children, as in (13a) (McGregor 1990: 272). Some authors even do not analyse such lexemes as inherently property-denoting. McGregor (2013) for instance argues that flexible lexemes in Gooniyandi have vague semantics: they do not have entity or property specifications. Their meaning in a phrase (or clause), he argues, is derived compositionally from their abstract coded meaning (e.g. ‘(the one who is) little’) in combination with the grammatical relation they are in (e.g. entity in a noun phrase in (13a)). (See Hengeveld & Rijkhoff 2005 for a slightly different approach to semantic vagueness with flexible lexemes.) Others may argue that examples such as (13a) involve conversion rather than flexibility, also in light of the fact that many nominals in Gooniyandi are actually preferred in one role, or are even restricted to a single role (see further below; see e.g. Croft 2005, Evans & Osada 2005, Don & van Lier 2013 for discussions of semantic compositionality, semantic shifts, and conversion).

Secondly, if a flexible word class of nominals is claimed, we not only expect property-denoting lexemes to be used referentially, but also entity-denoting lexemes to be used attributively (see e.g. Evans & Osada 2005 on bidirectionality). Unfortunately, very little information is available on this. For a few languages, it is mentioned that ‘nouns’ may function as modifiers, but most examples encountered in descriptions have only human stage-of-life terms or words for ‘man’ or ‘woman’ as modifiers. This was illustrated in (9d) from Arabana/Wangkangurru. Another example is found in (14) from Nyulnyul, where waringkil ‘girl’ is used as head in (14a) and as modifier in (14b) (McGregor 2011: 403). The limited possibilities for entity-denoting lexemes to occur as modifiers indicates that flexibility is rather limited, which implies that an analysis in terms of a single fully flexible class may be too strong in many languages (see also Evans & Osada’s 2005 criterion of exhaustivity; see Chapter 18 in this volume for suggestions on unidirectional flexibility).

Third, it should be noted that even in languages for which flexibility is claimed, elements show preferences for, or even restrictions to, certain roles. For instance, in Gooniyandi, nominals can generally occur as head or attributive modifier in the noun phrase, but many individual lexemes are in fact restricted in the functions they can have. For instance, numerals (which are considered nominals for morphological and functional reasons) can only be used attributively, and conversely, several nominals referring to plant or animal species can only be used as head of the noun phrase (McGregor 2013: 236). Similarly, Hale (1983: 33–35) posits a single nominal word class for Warlpiri, but suggests a rough subclassification as presented in (15), where nominals of type (a) are most likely (or sometimes exclusively) used referentially, and nominals of type (f) predicatively, with types in the middle having an equal likelihood for either. The labels refer to ‘semantic functions’ (Hale 1983: 34); the label ‘attributives’ thus seems to indicate that elements in this group are more likely to be used attributively, although Hale does not state this explicitly.

Apart from distribution across the functions of reference and attribution, we find other distributional criteria, which are morphosyntactic in nature and more widely discussed in grammatical descriptions. I briefly discuss the three most commonly mentioned ones. A first morphosyntactic criterion relates to different behaviour of nouns and adjectives in nominal classification: while nouns normally belong to a single class (or gender), adjectives do not belong to any class, but agree with the noun they modify. It is sometimes argued that this is no solid evidence for separate parts of speech, since there are examples of nouns belonging to different classes as well (see e.g. Nordlinger 2014: 238; Louagie 2020: 55–59 for discussion and examples). This criterion is obviously only at play in languages that have well-established systems of nominal classification (see Louagie 2020: 26–65, 79 on the distribution of different classification systems across Australia); see also section 26.3.2.

A second morphosyntactic criterion to distinguish nouns from adjectives is when each part of speech is tied to its own position in the noun phrase, i.e. nouns and adjectives occur in a fixed linear order. This is true for about one-third of Australian languages (Louagie 2020: 78, based on the 100-language sample mentioned above); note that flexible order in itself obviously is no evidence against such a distinction. An example is found in Arrernte (Wilkins 1989: 104), where adjectives always follow nouns in noun phrases, as illustrated in (16a)–(16b). Arrernte has a few forms with both a referential and an attributive sense, like iperte ‘hole’ (N) or ‘deep’ (Adj).14 It is argued that such cases do not involve a single flexible lexeme, but rather two distinct lexemes linked by conversion, one belonging to the word class of nouns and the other to the word class of adjectives. This is reflected in their distribution in the noun phrase: iperte ‘hole’ occurs in initial position, which is reserved for nouns (16a), while iperte ‘deep’ occurs in the position for adjectives (16b). When they are used together in an NP, as in (16c), speakers always identify the first element as noun and the second as adjective (Wilkins 1989: 104).15

A third morphosyntactic criterion, which is mentioned in grammars of only a small number of languages (Louagie 2020: 79), is the different behaviour of nouns and adjectives with degree modifiers. The differences are of two types: only adjectives can be used in a construction with a particular degree modifier while nouns cannot (as illustrated in (17)), or they can occur with the same modifier but to a different semantic effect (as illustrated in (18)). The main issue with these cases is whether such differences are really syntactically determined, or rather related to semantic factors, such as the compatibility of the degree modifier with the lexical semantics of the head for (17) or the combinatorial semantics of a degree modifier with a quality-denoting item or an entity-denoting one for (18).16

26.3.1.4 Patterns across Australia

Having surveyed the main morphological and distributional criteria used in grammatical descriptions, I now discuss how these criteria pattern in and across languages of Australia. The results of the continent-wide survey based on a 100-language sample mentioned earlier (Louagie 2020: 65–83) indicate that three main types of systems can be distinguished, based on the evidence provided in grammatical descriptions. Australian languages are roughly equally divided over these three types, but it is hard to see specific areal or genetic patterns (see Louagie 2020: 83 for an overview map, which shows that each of the three types is spread across the continent and occurs both in Pama–Nyungan and non-Pama–Nyungan languages).

In the first type of system, nouns and adjectives are clearly distinct on both morphological and distributional grounds. Almost a quarter of Australian languages have this system (Louagie 2020: 81–82). An example is found in Arrernte. In this language, nouns and adjectives are morphologically distinct in that one part of speech may derive into the other: reduplication of a noun derives an adjective (e.g. atnerte ‘stomach’ > atnerte-atnerte ‘pregnant’ Wilkins 1989: 149) but not vice versa, and some adjectives may derive a noun by other means (e.g. kngerre ‘big’ > kngerrepenhe ‘big one; something that is big’ Wilkins 1989: 105). Nouns and adjectives are also distributionally distinct, in three ways. First, whereas nouns are in principle obligatory in the noun phrase, adjectives are optional, i.e. noun phrases without a noun are the result of ellipsis (only in subsequent mentions) while noun phrases without an adjective are not (Wilkins 1989: 102–103). Second, they have their own distribution in the NP; see example (16) and discussion there. Third, both nouns and adjectives can occur in a construction with an intensifier, but to different semantic effects; note that this last criterion is less strong (see example (18) and discussion there).

In the second type of system, nouns and adjectives are distinct on distributional grounds, in that elements from each part of speech are restricted to specific distributions, but there are no morphological differences between them. Over one-third of Australian languages has this system (Louagie 2020: 82). An example is found in Yankunytjatjara (discussion based on Goddard 1985: 17; see also Bowe 1990: 8 on close variety Pitjantjatjara). Only nouns can function as the head of a noun phrase, except in the case of ellipsis, and nouns and adjectives have their own position in the noun phrase. Morphologically, nouns and adjectives are not distinct: they can take the same inflectional and derivational markers, and have similar effects in stem reduplication (viz. weakening of some sort, e.g. pika ‘angry’—pikapika ‘annoyed’; ngura ‘camp’—ngurangura ‘a sort of camp; a temporary camp’ Goddard 1985: 73).

In the third type of system, there is one flexible word class of nominals, i.e. there are no morphological or distributional grounds for distinguishing nouns from adjectives (see Chapter 1 on flexible word classes in general). A bit less than one-third of Australian languages have this system (Louagie 2020: 81), although this number may turn out to be lower on further scrutiny: a language was included in this count whenever there was no clearly described distributional distinction, but as discussed in section 26.3.1.3, information on distribution across functional roles was actually often limited or inconclusive. Still, there are good candidates for this type. One is Gooniyandi (Bunuban; McGregor 1990: 141–143; 2013); see section 26.3.1.3 for discussion of the lack of distributional distinctions. Another is Martuthunira, for which Dench (1994: 51–55) proposes a single class of nominals which can occur in several functions: as head of the noun phrase, as modifier in a noun phrase, in an ascriptive clause without or with copula, and as secondary predicate. Some lexemes are restricted to the first function (e.g. wirra ‘boomerang’), others are flexible between the first three (e.g. ngapala ‘mud(dy)’) or between all functions (e.g. pinkarranyu ‘dry’). Distributional flexibility between head and modifier roles in the noun phrase is illustrated in (19a)–(19c), with pinkarranyu as head in (19a) and as modifier in (19b)–(19c), in pre-head and post-head position respectively. Although some nominals are restricted in the roles they may have, Dench (1994: 55) suggests that there is no neat classification into subclasses, and that ‘these possibilities of occurrence cannot be predicted by general syntactic or semantic rule’.

For some languages for which a flexible class of nominals is proposed,17 the internal structure of the noun phrase has been meaningfully described in terms of functional roles (capitalized) and not in terms of word classes. The functional roles have a fixed position in the noun phrase, where each role may be filled by different types of elements and elements may generally occur in more than one role. Such descriptions obviously do not only include nominals but also other parts of speech like demonstratives and personal pronouns, as well as more complex elements like embedded phrases or elements. An analysis in terms of functional roles may offer insight in noun phrase structure in languages with high levels of flexibility in particular, as it takes into account semantic and functional differences with different word orders. This type of analysis was introduced in Australianist linguistics by McGregor (1990: 253–276) for Gooniyandi, and taken up by several other authors, including Dench (1994: 189–193) for Martuthunira. The template for the Martuthunira noun phrase is given in (20); nominals may occur in at least the Entity, Classifier, and Qualifier roles, as illustrated in (19a)–(19c) respectively. Some may also occur in a Quantifier role (e.g. kupuyu ‘little, small’; Dench 1994: 190).18

It is clear from the preceding discussion that not all Australian languages have similar part-of-speech systems for elements involved in reference and attribution, despite the generalizations in some of the typological literature. About one-third of Australian languages have a single, flexible class of nominals, while the others have separate word classes for nouns and adjectives, which are distributionally, and less frequently also morphologically, distinct.

26.3.2 Do classifiers and determiners form separate word classes?

Australian languages have received some attention in the typological literature for two further features which are also related to word class: their systems of nominal classification, and their lack of obligatory determiners while at the same time allowing multiple determiner-like elements in a single noun phrase. Both are interesting for the current chapter because they invite the question whether a class of classifiers and one of (non-obligatory) determiners can be distinguished, and if so, in what way. I will discuss these questions briefly, first focusing on classifiers and then on determiners.

About two-thirds of Australian languages have at least one system of nominal classification, whereby nouns/nominals are overtly classified into subclasses (Louagie 2020: 27). The two main types found in Australia are noun class systems and noun classifier systems (Dixon 1982). Noun classes are marked in agreement patterns, as in (21), where the nouns diban and yibi are marked as belonging to the fourth and second noun class respectively by the agreement shown on the adnominal demonstratives. Noun class systems are not further considered here, but see Dixon (1982), Sands (1995), Dixon (2002: 460–514), McGregor (2004: 146–150), Louagie (2020: 23–65), and Skilton (2023) for surveys. In contrast to noun classes, noun classifiers are free forms that occur in juxtaposition to the noun they classify, as in (22), where the generic nouns mayi and bama mark the specific nouns jimirr and yaburu as belonging to the categories of vegetables and persons respectively. The use of noun classifiers is optional in all Australian languages that have them (Louagie 2020: 34), and in most languages the order is fixed to generic noun-specific noun as in (22) (Louagie 2020: 29). See Dixon (1982: 159–206), Harvey & Reid (1997), Wilkins (2000), Dixon (2002: 449–460), and Louagie (2020: 23–65) for surveys and more detailed studies of noun classifiers.

If a language has a system of noun classifiers as illustrated in (22), the question may be raised whether there is consequently also a separate word class of classifiers. If this question asks about the presence of a (potentially closed) word class of elements which are dedicated to this classifying function, the answer is negative:19 elements which can serve as classifiers are not dedicated to this role. This is illustrated in (23) from Arrernte: in (23a) the generic noun kere marks the category to which the referent of the specific noun aherre belongs, while in (23b) the same element is used independently. It is, however, still possible to distinguish a distributional class of noun classifiers, which could be defined as the set of elements which can occur in initial position in a ‘classifying construction’. The classifying role that elements like kere in (23a) have precisely arises from their use in the construction as a whole, and is not inherent in the individual elements (Wilkins 2000). This analysis was proposed by Wilkins (2000) for Arrernte (and preliminarily for Yankunytjatjara and Warlpiri), but has the potential to be applied more broadly in Australia (Louagie 2020: 33–35).20

Turning to determiners, Australian languages generally have no word class of articles or obligatory determiners (e.g. Blake 2001; Dixon 2002: 66–67; Stirling & Baker 2007). All languages have elements that may have determining functions, like demonstratives, but these are optional in the sense that a bare noun may for instance have a definite or indefinite interpretation. There are only a few exceptions (Louagie 2020: 182–183), such as the adnominal personal pronoun in Arrernte, which is obligatory in definite contexts, while a bare noun has indefinite reference (Wilkins 1989: 165); see examples (16) and (23) for illustration.

Nonetheless, several authors identify a determiner position in the noun phrase in which elements with determining function coalesce (e.g. McGregor 1990: 253–276 on Gooniyandi; Dench 1994: 189–193 on Martuthunira; Louagie 2017 for a continent-wide study). In other words, they identify a distributional class of determiners: those elements which may (but must not) appear in this particular slot in the noun phrase. This is very similar to the approach to classifiers as discussed in the previous paragraphs. An example is found in Harvey’s analysis of Gaagudju (Harvey 2002: 316–320), who identifies a determiner slot in initial position in the noun phrase (24a). Demonstratives, pronouns, the numeral ‘one’, interrogative-indefinites and ‘other’ may all appear in this slot and may thus be argued to form a distributional category of determiners. Multiple determiners may occur at the same time, as in (24b). However, none of these elements seem to be specialized, as they may also occur in post-head position with a qualifier function. This is illustrated in (24c), where the demonstrative does not so much contextualize the phrase, but rather has a pointing function, behaving as an attribute (Harvey 2002: 317; see also McGregor 1990: 267–268 for a detailed discussion of a similar analysis in Gooniyandi). The mapping between the determiner slot and the elements which may fill it is very flexible in Gaagudju, but it may also be relatively fixed, as in Mawng (Forrester 2015: 45) or Umpila (Hill 2018: 126–140). Some elements may also show more flexibility than others: while adnominal pronouns are restricted to the determiner slot in almost all languages, as discussed earlier, possessive pronouns and numerals tend to show much more flexibility within and across languages; see Louagie (2017) and Louagie (2020: 164–207) for extensive discussion.

In sum, while Australian languages generally do not have dedicated word classes of classifiers or determiners, relevant categories may be identified distributionally, as a set of elements which may occur in a particular position in a larger structure, with a particular functional role tied to that position. Classifiers are those elements that may occur in initial position in a classifying construction; determiners are those elements that may occur in a determiner slot in the noun phrase. Such elements may be more or less strongly tied to these positions in different languages.

26.4 Verbal parts of speech: Inflecting and Uninflecting Verbs

A number of northern Australian languages, mainly non-Pama–Nyungan ones but also some Pama–Nyungan ones, have two distinct word classes that may be characterized as verbs: IVs and UVs (e.g. McGregor 2002; Schultze-Berndt 2003, 2017; Bowern 2014).21 Their most typical uses are illustrated in (25a)–(25b) from Bardi. Example (25a) shows how an IV and an UV are combined into a complex construction, which Bowern (2014: 264) characterizes as a complex predicate, in which ‘the information normally associated with the head of a verbal predicate is spread over several parts of the predicate’. The UV wajim determines the semantic roles of the participants in the event, while the IV inmanirr provides information about tense, aspect, person and number, as well as the event type (Bowern 2014: 264–265).22 Note that the IV cannot be analysed as an auxiliary, as it also has lexical content (Schultze-Berndt 2017: 245). Example (25b) shows how an IV is used as independent verb, i.e. in a simple predicate construction.

There is good morphological and distributional evidence for distinguishing IVs and UVs as separate word classes. Morphologically, IVs inflect for TAM categories, while UVs do not usually inflect (although in some languages they occasionally do, see Bowern 2014: 279–280 for examples).23 Distributionally, IVs and UVs occupy different positions and functions in complex constructions as in (25a), and UVs cannot occur as independent predicates in simple verb constructions while IVs can. The nature of these classes is also different. IVs often form a closed class, which can consist of anything between one and hundreds of elements; in most languages IVs cannot be derived from any other lexical category (Bowern 2014: 276; Schultze-Berndt 2017: 247). UVs, by contrast, form an open class allowing borrowings, as illustrated in (25a), and can also be derived from other parts of speech such as nouns or adverbs (Bowern 2014 274). Semantically, it appears that elements from the class of IVs typically encode more generic concepts, while UVs are semantically more specific (Schultze-Berndt 2017: 264; McGregor 2002). The class of UVs may include translational equivalents of verbs, manner adverbs and spatial adverbs in English, e.g. Jaminjung waya ‘call’, gaabardag ‘quick(ly), walthub ‘inside’ (Schultze-Berndt 2017: 246–247).

It makes sense to analyse both IVs and UVs as verbal parts of speech, for different reasons: IVs show inflection that is cross-linguistically typically associated with verbs, such as marking for tense, aspect, and person/number of the subject and/or object, while UVs are functionally closer to verbs in other languages, in that they designate an event (McGregor 2002: 25).24 The question is, however, how these classes are precisely related to the function of predication. In the next paragraphs, I survey the different distributions and functions elements from each class may have in languages that have both, and discuss different analyses for such constructions. I also particularly focus on issues that have repercussions for part-of-speech theories in general.

The most common use of UVs is in combination with an IV in a complex construction (also called compound verb construction), as illustrated in (25a) and (26a)–(26b). Many authors analyse such constructions as complex predicates (or compound verbs), where ‘two or more predicative constituents jointly contribute to the argument structure of the clause’ (Butt 1997: 108; cited in Schultze-Berndt 2017: 252). In other words, both UVs and IVs are analysed as having a predicate function. In a number of languages, however, constructions as in (26) may be analysed as verb classifying constructions, in which the IV overtly classifies the UV, thus ‘indicating the category into which the stem is assigned in a particular instance of use, and correspondingly, the category to which the referent event is considered to belong’ (McGregor 2002: 44). Thus, in example (26a) the UV jarrbard is assigned by the IV -k ‘carry’ to a category of atelic activities where something is moved by constantly applied force to a new location, while in combination with the IV -m ‘put’, as in (26b), jarrbard is classified as a telic activity which induces a thing to enter a new state, condition or location (McGregor 2002: 112–114).25 For languages where such constructions may be analysed as involving verb classification, McGregor (2002) convincingly argues that the IV in fact does not have a predicate function, but rather a category marking function.26

IVs can also occur alone, as independent predicate in main clauses, as illustrated in (25b), and as dependent predicate in finite subordinate clauses, as in (27). UVs, by contrast, cannot occur as independent predicates in main clauses, but they may occur as only predicate in a subordinate clause, as in (28a) (Schultze-Berndt 2017: 253). UVs may also function as pragmatically dependent predicates, where the verbal or nonverbal context provides the necessary contextualization and where ‘the use of the UV itself is the only formal correlate of dependent status’ (Schultze-Berndt 2017: 254). An example is when the UV has the illocutionary force of a command, in a so-called ‘condensed directive’, as in (28b); such uses are stylistically marked, for example in comparison with the use of an imperative IV (Schultze-Berndt 2017: 254).

The functional–distributional differences between IVs and UVs are summarized as follows by Schultze-Berndt (2017): IVs can be used for dependent and independent predication, while UVs are restricted to dependent predication. Dependency may be syntactic (as in (25a), (26), and (28a)) or pragmatic (as in (28b)). Thus, Schultze-Berndt specifically includes the use of UVs in complex constructions in the function of dependent predication, arguing that their lack of inflectional morphology is evidence for syntactic dependency.

Both as part of a complex construction and on their own, UVs may be used expressively in some languages, i.e. with performative foregrounding (Schultze-Berndt 2001: 367–368; McGregor 2002: 330–331; Schultze-Berndt 2017: 256). Such instances show expressive prosody such as interruption of rhythmic flow, lengthening, higher intensity, larger pitch range or marked voice register (Schultze-Berndt 2001: 367; McGregor 2002: 330). Expressive use is illustrated in (29) from Jaminjung and (30) from Gooniyandi. Although Schultze-Berndt (2017: 255) labels UVs in such uses in Jaminjung as mimetic predicates, arguably not all instances may be analysed as involving predication rather being syntactically independent.

The word classes of IVs and UVs in Australia raise some issues that are relevant for general theories of parts of speech. First, on the basis of the Australian data, Schultze-Berndt (2017) argues that the functional basis used for cross-linguistic research on word classes should take into account both independent and dependent predication. In this sense, she argues, other languages could be analysed as having one flexible category of verbs that can serve both independent and dependent predication. Second, as pointed out by an anonymous reviewer, the use of IVs as markers of verb classification in complex constructions as in (26) and the expressive use of UVs as in (29)–(30) pose a more serious problem to theories of parts of speech like Hengeveld’s (1992a, 1992b), because they show that ‘verbs’ may have functions other than predication (while restriction to predicational use only is crucial to such theories). Both points show that more grammatical functions need to be recognized in part-of-speech typologies if we want to capture the relevant cross-linguistic diversity. Third, Australian languages show an interesting link between UVs and ideophones: a subset of UVs in Australian languages have their origin in ideophones, and UVs and ideophones also share a number of characteristics synchronically, such as distinctive phonological and phonotactic features, sound symbolism and minimal morphological possibilities, as well as expressive uses as in (29)–(30) (McGregor 2001; Schultze-Berndt 2001; McGregor 2002: 324–339). See Chapter 22 in this volume on ideophones as a word class.

Finally, note that a number of Australian languages exhibit related phenomena, but ones that do not imply the presence of two distinct classes of verbs. They for example have compound predicates consisting of two elements which may originally have been an IV and an UV, but that are synchronically indivisible and distributionally behave as a single unit, i.e. that have grammaticalized (Schultze-Berndt 2017: 247–248; McGregor 2002: 149–152). In fact, it is argued by McGregor (2002: 28–29, 351–354) that markers of conjugation classes in many Pama–Nyungan languages too are remnants of IVs in compound verb constructions. For more discussion of compound verb constructions and the categories of IVs and UVs, see for example Nicolas (1998), Wilson (1999), Schultze-Berndt (2000; 2001; 2003; 2017), McGregor (2002; 2004: 174–186), and Bowern (2008; 2010; 2014), among many others. For discussion of inflectional categories on verbs, see for instance Dixon ([1980] 2010: 378–437; 2002: 209–236; mostly on Pama–Nyungan languages) and McGregor (2004: 159–174) on languages from the Kimberley.

26.5 Manner adverbs

Locational and temporal adverbs were discussed in the introduction to section 26.3; this section briefly touches on manner adverbs. Unfortunately, no detailed surveys are available about manner adverbs as far as I am aware (though there are some notes in Dixon [1980] 2010: 281–282; 2002: 181–183 and McGregor 2004: 104), nor have I collected data on adverbs systematically for the languages of my sample, so discussion here remains largely anecdotal. Depending on the language, manner adverbs may have more in common with nominal parts of speech, with verbal parts of speech, or with neither, which is the reason I discuss them in a separate section.

In some languages, manner adverbs show morphological similarities to nominal parts of speech such as nouns and adjectives (or nominals, if not distinguished), in that they may inflect for case marking. In some languages, manner adverbs optionally exhibit instrumental or locative marking. This is for instance the case in Nyulnyul, where a small set of manner adverbs may show instrumental marking, as in (31) (McGregor 2011: 171). In other languages, a manner adverb may agree for nominal case with one of the participants in the clause, for instance in Martuthunira as in (32) (Dench & Evans 1988: 14–16) and in Yankunytjatjara (Goddard 1985: 29–30). In such cases, adverbs behave much like depictive secondary predicates morphosyntactically, although they are semantically subtly different (see Dench & Evans 1988: 14–16 and Himmelmann & Schultze-Berndt 2005 for discussion).

Despite these morphological similarities, it seems that the word class of manner adverbs is distinct from other word classes in most languages (Dixon [1980] 2010: 282). Several languages allow derivation from adjectives or nouns into adverbs (Dixon 2002: 181), usually with an adverbializing suffix, as illustrated in (33). At least one language, Kuku Yalanji, can also derive adverbs from adjectives by reduplication (Patz 2002: 110).

Flexibility of (at least some) lexemes between adjectival and adverbial functions is found for instance in Bardi (Bowern 2012: 562), Diyari (Austin 2013: 112–114; though no examples are provided) and Martuthunira (Dench 1994: 52–55).28 An example from Martuthunira is given in (34a)–(34b), showing the element panyu ‘good’ used in adjectival and manner adverbial functions respectively.

In other languages, manner adverbs have more in common with verbal parts of speech. For instance, in at least a few languages, adverbs must agree with the verb they modify in transitivity and/or tense, aspect or mood marking (Dixon [1980] 2010: 281–282; 2002: 181–183). In Yukulta, adverbs are derived from adjectives and then take either an intransitive or a transitive suffix, in agreement with the main verb (Keen 1983: 226; cited in Dixon 2002: 183). In Gumbaynggir, manner adverbs likewise must agree with the verb in transitivity and in other inflections like tense (Eades 1979: 304–308). A final example of manner adverbs resembling verbs is found in Dyirbal (Dixon 2002: 181–182), where ‘adverbs’ are purely verbal in that they have inherent transitivity values and may inflect for TAM as well. They are unlike other verbs, however, in that they generally do not occur alone, but with a main verb, modifying it. Adverbs must agree with the main verb in transitivity and final verbal inflection. If an inherently transitive adverb modifies an intransitive main verb, the adverb is suffixed with a reflexive marker which basically functions as an intransitivizer, as illustrated in (35) with the adverb gudi ‘do too much’.

Apart from languages where adverbs may resemble IVs, there are also languages where adverbs are very close to UVs, e.g. because they also do not inflect and may have similar types of denotations, although there is usually some way of distinguishing them, for instance because they are optional while UVs are not; for some discussion see for example Merlan (1994: 59) on Wardaman and Schultze-Berndt (2000: 72–73) on Jaminjung. This brings us to languages where adverbs do not resemble nominal or verbal parts of speech: they do not inflect at all. Some examples are Bardi (Bowern 2012: 160, 569–570), Dhuwal (Wilkinson 1991: 678–680), and Kuuk Thaayorre (Gaby 2017: 359–361).

26.6 Conclusion

This chapter discussed the major lexical parts of speech in Australian languages, both from the large Pama–Nyungan family and from the smaller families in the north and northwest of the continent. Australian languages clearly distinguish verbs from nouns (or nominals): verbs are used for predication and not for reference, at least not without formal adaptation, while nouns are used for both, though mainly for reference. There is some morphological overlap between nouns and verbs in some languages, but it is usually rather minimal and restricted to specific contexts.

Australian languages are more heterogeneous with respect to nouns and adjectives. About one quarter of the languages have morphologically and distributionally distinct word classes of nouns and adjectives; roughly one-third have nouns and adjectives that are only distributionally distinct, but not morphologically; and another third have a single flexible class of ‘nominals’ that can function both referentially and attributively. In relation to the latter, it was also pointed out that a number of languages more generally show flexibility between elements and functional roles in the noun phrase, for instance where demonstratives or possessive pronouns may be flexible between determiner, qualifier, and entity roles. It was argued that classes of determiners and nominal classifiers may only be identified on a distributional basis, in terms of the elements that may take up the positions in the noun phrase associated with these roles. Overall, the discussion showed the importance of teasing apart morphological and distributional characteristics of lexemes, and showed that more detailed research is needed on the relation between (types of) elements and the functional roles they can fill in a noun phrase, including entity and qualifier, but also roles such as classifier, quantifier, and determiner. This is a question that not only deserves more attention in research on Australian languages, but also outside Australia, as it can offer a more fine-grained perspective on parts of speech.

A number of Australian languages, especially in the north, have two distinct parts of speech associated with predication: IVs and UVs. The latter are, according to Schultze-Berndt (2017), used for dependent predication only, e.g. as part of a complex predicate, or as main predicate in a dependent clause or insubordinated clause. Schultze-Berndt (2017) thus also advocates using more fine-grained functional distinctions as baseline for typological research, including both independent and dependent predication, to capture relevant cross-linguistic variation. At the same time, McGregor (2002) shows that IVs may have another function than predication, viz. as markers of verb classification, thus also problematizing the restriction some part-of-speech theories place on verbs serving predicational functions only.

Manner adverbs, finally, are often a distinct part of speech in Australian languages that may show morphological similarities with nominals (such as case marking) or verbs (such as marking for transitivity or tense). In some languages, manner adverbs do not show any inflections. A few grammars note some flexibility of lexemes between adjectival and adverbial functions.

Acknowledgements

I would like to thank David Beck, Eva van Lier, Jean-Christophe Verstraete, and an anonymous reviewer for their valuable feedback on earlier versions of this chapter, which pushed me to discuss some issues more carefully and much helped improve the general structure and flow of the chapter. Work on this chapter was supported by a postdoctoral grant from the Research Foundation-Flanders (FWO), and carried out at the University of Leuven.