2.2.1 General Issues

A basic tenet of descriptive linguistics is that linguistic categories are only valid to the extent that they are empirically justified, and the view that there is no such thing as a universal grammatical category is gaining acceptance.¹ This holds for all categories, including lexical categories such as noun, verb and adjective (to name just three) as well as the corresponding endocentric syntactic categories, NP, AdjP and VP, even though these categories – or at least the lexical categories – are generally considered universal.

Although there are considerable differences of opinion with respect to the ‘correct’ approach to this topic, most researchers would probably agree with an approach which views lexical categories as language-specific categories based on feature bundles, in which ‘prototypical’ members of a particular class possess all of the features potentially associated with the particular class.² For example, prototypical nouns in English would denote persons, places or things, mark for the plural (usually with ‑s), be compatible with definite and indefinite articles, etc., while less prototypical nouns, for example, may not be compatible with plural marking or the definite article. Problematic for this approach, however, is determining which features are more important than others, if indeed a hierarchy can be established at all, as well as how many features are ‘enough’ for an item to be assigned to a particular category. We will return to this topic in the following pages.

As Himmelmann (Reference Himmelmann, Austin and Musgrave2008: 260) notes, there is little doubt that ‘the inventory of function words is highly language-specific’. Rather, ‘[w]henever there is a controversy regarding the number and kind of syntactic categories in a given language, it pertains to the linguistic classification of content words’. Himmelmann assumes two grammatical levels for the discussion of parts of speech: (i) ‘the level of terminal syntactic categories where lexical items are categorised according to their phrase-structural properties’, and (ii) ‘the level of lexical categories proper where lexical items are categorised according to those grammatical features which are not directly relevant for phrase structure’ (Himmelmann, Reference Himmelmann, Austin and Musgrave2008: 263). To illustrate these two levels,³ Himmelmann proposes a hypothetical example in a language L, in which the most basic phrasal categories always consist of an overt function word X and an overt content word Y, with virtually all content words being compatible with all function words. This can be schematically portrayed as in (1). Here the status of the function word X distinguishes the phrase from other phrase types, hence ‘XP’.

These two levels, that is, lexical and syntactic, need not correspond to one another directly, although this will often be the case, as, for example, with Ns as heads of NPs, Adjs as heads of AdjPs and Vs as heads of VPs in many languages. Himmelmann (Reference Himmelmann, Austin and Musgrave2008: 264) notes five logical possibilities with respect to these two levels and their correspondence to one another, given in Table 2.1. With respect to category I, both levels possess distinct categories but these do not directly correlate in Ia (‘≠’), while they do in Ib (‘=’).

Returning to our example above, if, following Himmelmann, we imagine that L has a further restriction such that not all content words Y behave similarly, as some Ys take plural suffixes, others take plural prefixes, while the members of a third class do not allow pluralization, then we would have case Ia, with distinct lexical categories based on the compatibility of Y with plural markers and also distinct terminal syntactic categories, but with no direct correspondence between the two levels, as the compatibility of plural affixes does not correlate with the compatibility of a particular Y with a particular X.

To these two descriptive levels we can add a third level, that of function, here primarily reference, predication and attribution. This can involve a direct correlation between terminal syntactic cateogies and functional categories, that is, XP₁ = reference, XP₂ = predication, etc., or there can be no direct correlation, so that XP₁, XP₂, etc. can both be used in reference, predication and modification. At one extreme we would then have endocentric categories specialized for certain sentence-level functions, for example N = NP = reference, V = VP = predication, etc., and at the other extreme – at least putatively – no correlations whatsoever, although it is questionable whether such a language is possible.

Such mismatches are not only a theoretical possibility; many researchers have claimed that the languages they are describing either do not in fact possess a certain category such as ‘N’ or ‘NP’, or that the respective language can at least best be described without recourse to such categories. Even in languages such as English, which possess relatively clearly defined lexical classes such as noun, verb, and adjective, it has long been known that there is no one-to-one relationship between part of speech and function. Consider the examples in (2a–g), from Van Valin (Reference Van Valin2008: 165).

1. (2)

   a.	Chris will [nuc [pred see]] the movie.	PRED in NUC = V
   b.	Chris [nuc is [pred a very good detective]].	PRED in NUC = NP
   c.	Pat [nuc is [pred exceedingly tall]].	PRED in NUC = ADJP
   d.	Pat [nuc is [pred in the house]].	PRED in NUC = PP
   e.	Chris [nuc [pred wiped]]	PRED in NUC1 = V
   	the table [nuc [pred squeaky clean]].	PRED in NUC2 = ADJP
   f.	Pat [nuc [pred pushed]]	PRED in NUC1 = V
   	the table [nuc [pred out the door]].	PRED in NUC2 = PP
   g.	Chris [nuc was [pred elected]]	PRED in NUC1 = V
   	[nuc [pred president of the club]]	PRED in NUC2 = NP

As the examples in (2) show, the predicate of a nucleus in English need not be verbal. While the predicate is verbal in (2a), in (2b) it is an NP, in (2c) an AdjP and in (2d) a PP. Figure 2.1 presents the constituent projections for (2a) (from this author), and (2b) and (2d) (adapted from Van Valin, Reference Van Valin2008: 166).⁴

Figure 2.1 Constituent projections for (2a), (2b) and (2d)

(adapted from Van Valin 2008: 166)

In English, as in many other languages, there is a general requirement that clauses in which the predicate is not verbal must contain a kind of ‘auxiliary’ such as the copula is in order for the clause to be grammatical. As a very good detective in (2b), exceedingly tall in (2c) and in the house in (2d) are not verbs, the copula is is required; however, it is important to stress two things here. First, that while is is a verb, it is not the predicate; this function is filled by the NP in (2b), by the AdjP in (2c), and by the PP in (2d). Second, this is a language-specific criterion which is not found in all languages or which is only found with certain interpretations. For example, one quite common situation is found in Russian, where a clause whose predicate is non-verbal does not require a copula when there is a present or atemporal interpretation, as in (3).

1. (3)

   Russian
   eto	stol
   this	table
   ‘This [is a] table.’

In other languages, the element which is often considered a copula may not even be verbal, as in Maltese (Semitic, Malta), where the ‘copula’ is in fact a pronoun agreeing with the subject with respect to person and number (and gender in the third-person singular) with a present or atemporal interpretation, as in (4).

1. (4)

   Maltese (Ambros Reference Ambros1998: 69)
   Dan	il-professur	huwa	Malti.
   this.m.sg	def-professor	3m.sg	Maltese.m.sg
   ‘This professor is Maltese.’

As Van Valin (Reference Van Valin2008: 166) notes, non-verbal predicates do not pose a problem in RRG, since the clause is not an endocentric category: ‘Indeed, there is no head at all. The nucleus cannot be considered the “head” of the core or the clause, because it is not a lexical category, on the one hand, and is often phrasal, on the other. The notion of “head” is of no relevance to the layered structure of the clause.’ Instead, all universal categories assumed in RRG are semantically motivated, so that in RRG a clause minimally consists of a core – which minimally consists of a nucleus which contains the predicate – and possibly one or more arguments.

Instead of positing universal lexical categories which are also the heads of corresponding phrasal categories – cross-linguistically an even more problematic assumption – mainstream RRG has assumed since Van Valin (Reference Van Valin2008) that the core contains a predicate, the functional/semantic base of the nucleus, and possibly one or more referential phrases, referred to as RPs. Just as the predicate need not necessarily be verbal, neither does an RP have to be nominal. In addition, RRG assumes that these units may be modified by modifying phrases or MPs, which can (but need not) be specialized so as to modify only an RP, as with prototypical adjectives, or only predicative elements, as with prototypical adverbs, although again this is not necessary, as there are many languages with either no or only very few adjectives and/or adverbs, where these functions are realized by other categories, such as nouns or verbs.

Not all NPs are RPs and not all adjectives or adverbs are MPs. For example, expletive or ‘dummy’ subjects like those in (5a) are not referential and hence not RPs: these are pronouns and therefore NPs, but they are not referential and hence not RPs. Similarly, while a very good detective in example (2b) is an NP, repeated here as (5b), it is not referential and therefore also not an RP.

1. (5)

   a.	It looks like it’s going to rain today.
   b.	Chris is a very good detective.

The same holds for the relationship between adjectives / adjective phrases and the MP. For example, while the AdjP very good in (5b) is a modifier in the NP a very good detective, and hence an MP, the AdjP very good in (6) is not a modifier, and therefore not an MP.

In languages which do not have the usual parts of speech such as noun, adjective or verb, these units will of course belong to different, language-specific categories. For example, as we will see in Section 2.2.3.1 on Kharia, translation equivalents of nouns, adjectives and verbal stems all belong to the open class of contentive lexemes of that language, whereas NPs and finite verbs in English correspond most closely to the Kharia-specific units ‘Case-syntagma’ and ‘TAM/Person-syntagma’, respectively. Again, these structural units must be carefully distinguished from the functionally motivated units RP, MP and predicate.

As mentioned above, modifying phrases or MPs in many languages have as their default value either adjectives or adverbs, depending on what is being modified, that is, a noun (the tall tree) or a verb (the mouse ran quickly into the closet). These appear in the peripheries of RPs and clauses (Van Valin Reference Van Valin2008: 172).⁵ As Van Valin notes, MPs also have a layered structure, for two reasons. First, many languages allow modifier phrases with an adjectival nucleus to take a core argument, such as the German der auf seinen Sohn stolze Vater (lit. ‘the of his son proud father’). Secondly, the modifiers can themselves be modified, ‘which means that they must have a periphery to house the modifying MP, e.g. the very quickly extinguished fire’ (Van Valin 2008: 172).

In addition to adjectives and adverbs, many languages, including English, also allow other, much more complex categories to serve as modifiers. Consider the following examples, from the same source.

1. (7)

   a.	The Charles and Di syndrome is no longer relevant.
   b.	The God is dead philosophers are mostly dead.
   c.	My grandson likes to give me the who’s the boss now, silly old grandpa wink frequently.

In (7a) the MP contains a conjoined RP in its nucleus, that in (7b) contains a clausal nucleus, and the nucleus of the MP in (7c) contains a constituent question as well as a vocative in its post-detached position [PoDP]. Figure 2.2 gives the constituent projection of (7c) (PrCS = pre-core slot).

Figure 2.2 Structure of RP with MP containing sentential nucleus

(adapted from Van Valin Reference Van Valin2008: 173)

Even the seemingly inconspicuous category of pre-/postpositional phrases is not without problems. As Van Valin (Reference Van Valin2008: 170) notes, these are normally thought of as endocentric projections of a (pre- or postpositional) lexical head in predicative adpositional phrases, such as the underlined element in (8), which has a prepositional nucleus consisting of the preposition in, which functions as a predicate and thus licenses the object the library.

The status of this unit in (8) is markedly different from that of the underlined units in (9), from the same source, referred to as non-predicative PPs as these units do not predicate but are licensed by the verb and therefore do not license an object.

1. (9)

   a.	Chris showed the photo to Pat.
   b.	Chris stole the photo from Pat.
   c.	Chris presented Pat with the photo.

As the underlined units in (9) are non-predicative, they do not contain a PRED and hence neither a NUC_p (nucleus of the PP) nor a CORE_p . Rather, they consist merely of a pre- or postpositon and an RP. Figure 2.3 contrasts the internal structure of these two PP types (see also Chapter 10 for adpositional phrases in RRG).

(Van Valin Reference Van Valin2008: 171)

Figure 2.3 Non-predicative (a) and predicative (b) prepositional phrases in English

With respect to these two PP types, Van Valin (Reference Van Valin2008: 171) notes that predicative PPs most closely correspond to the X-bar notion of an endocentric category, as the nucleus here is always a P and always a head. He also notes that a modifier like right in right under the table can be handled in terms of a modifier in the periphery within the predicative PP, while with a non-predicative PP there is no prepositional nucleus, the only nucleus being that of the RP. As such, they can arguably be analysed as ‘exocentric PPs, i.e. PPs without a prepositional nucleus. Hence there is no layered structure, and this predicts that the admittedly limited set of PP-internal modifiers should not occur with a non-predicative PP, which seems to be the case: *Chris showed the photo right to Pat/*Chris stole the photo right from Pat/*Chris presented Pat right with the photo.’

2.2.2 Gradience in Lexical Categories

Any definition of lexical classes in terms of feature bundles will eventually run into problems as there will always be at least some content morphemes which show many but not all of the ‘prototypical’ features associated with a certain lexical category. For example, although most nouns in English and many other languages can appear in the plural, this is problematic with others, such as peace. Similarly, while most nouns in German are compatible with definite and indefinite articles, such as der/ein Mann ‘the/a man’, others, such as Verlaub ‘leave; permission’ are compatible with neither, in this case being restricted to occurrence in the fixed (prepositional) phrase mit Verlaub with the preposition mit ‘with’ and the meaning ‘with respect; if you will pardon my saying so’. Although Verlaub is listed in many dictionaries with the definite article der (masculine, singular), it is in fact probably never found with this article in actual speech, although it is considered a noun for other reasons, for example its compatibility with the preposition mit, which presupposes an NP (which is assumed to have a nominal head!) and its perceived inherent (masculine) gender, as inherent gender is restricted to nouns in German.

Similar problems are encountered with what Ross (Reference Ross1972) refers to as ‘category squishes’ or sequences of forms whose feature bundles gradually change so that it is often difficult to decide which category they should belong to. Van Valin (Reference Van Valin2008: 174) cites the examples in (10).

1. (10)

   a.	The savant quickly proved the theorem
   b.	That the savant quickly proved the theorem	(astonished the professors)
   c.	the savant/*the quickly proving the theorem	(astonished the professors)
   c′.	the savant having quickly proved the theorem
   c′′.	(I am aware of) the theorem having quickly been proved by the savant.
   d.	the savant’s/*the quickly proving the theorem	(astonished the professors)
   d′.	the savant’s having quickly proved the theorem
   d′′.	the theorem’s quickly being proved by the savant
   e.	the savant’s/the quick proving of the theorem	(astonished the professors)
   e′.	*the savant’s quick having proved of the theorem
   e′′.	the quick proving of the theorem by the savant
   e′′′.	*the theorem’s quick proving by the savant
   f.	the savant’s/the quick proof of the theorem	(astonished the professors)
   f′.	the quick proof of the theorem by the savant
   f′′.	the savant’s/the quick proof	(astonished the professors)

Although proved in example (10a) is clearly verbal and proof in the (10f) examples is clearly nominal, the status of the forms ending in ‑ing is much less clear. While some instances of proving, such as those in (10c) and (10d), take a direct object, suggesting that this form is a verb, the forms in (10e) and (10e′′) require an ‘object’ marked by the preposition of, suggesting that this is a noun. Furthermore, the ‘subject’ NP in (10c) appears as a bare NP, while in (10d) – as in the more ‘nominal’ form in (10e) or the clearly nominal form proof in the (10f) examples – it appears in the genitive. These and other factors such as adjectival modification, optional complements, tense, aspect and voice are summarized in tabular form in Van Valin (Reference Van Valin2008: 175).⁶

As we see in (10), even in languages such as English, which otherwise seem to have quite clear lexical classes, it is often not feasible to assume that each and every word form unambiguously belongs to a particular lexical category. RRG thus assumes that it is the individual features which are relevant and not the labels themselves:

In the following two subsections, we discuss concrete examples from two other languages, Kharia (Munda, India) and Wari’ (Chapakuran, Brazil and Bolivia), where assuming clear-cut lexical classes and strictly endocentric syntactic categories is even more difficult, and we will show how these ‘problematic’ elements are in fact dealt with straightforwardly in RRG.

2.2.3 ‘Problematic’ Categories

2.2.3.1 Parts of Speech in Kharia

In the South Munda language Kharia, spoken in eastern-central India, Peterson (e.g. Reference Peterson2011a, b, 2013) argues that the familiar lexical categories of ‘noun’, ‘verb’, ‘adjective’ etc. are not only unnecessary for an adequate grammatical description of the language, assuming their presence in fact considerably complicates the discussion.

Peterson divides the Kharia lexicon into two major classes: the vast majority of content morphemes, that is, morphemes denoting referents, attributes and events/states, may be used freely in referential, attributive and predicative functions with no overt derivational marking, ‘light verbs’, etc. This class is open as new morphemes (e.g. modern technology and loanwords in general) are constantly being added to its ranks. There is also a closed class which can be further divided into two subclasses: ‘Proforms/Deictics’, which may be used freely in referential and predicative function, but may only be used attributively if marked for the genitive, and ‘grammatical morphemes’, which may never be used referentially, predicatively or attributively. This is illustrated in Figure 2.4.

Figure 2.4 The Kharia lexicon

(Peterson, 2011a: 78)

(11)–(12) present two examples demonstrating the flexibility of content morphemes.

1. (11)

   Kharia ((11a) from Peterson (Reference Peterson2011a: 76); (11b) adapted from Malhotra (Reference Malhotra1982: 136))
   a.	lebu	ɖel=ki.
   	man	come=mid.pst
   	‘The/a man came.’
   b.	bhagwan	lebu=ki	ro	ɖel=ki.
   	God	man=mid.pst	and	come=mid.pst
   	‘God became man [= Jesus] and came [to earth].’

1. (12)

   Kharia (Peterson Reference Peterson2013: 133)
   [In a play about me and you, in which both of us will be taking part.]
   ‘naʈak=te	iɲ=ga	ho=kaɽ=na=iɲ		ro	am=ga	iɲ=na=m.’	‘umboʔ.
   play=obl	1sg=foc	that=sg.hum=mid.irr=1sg		and	2sg=foc	1sg=mid.irr=2sg	no
   am=na   um=iɲ    pal=e.			ɖirekʈar	seŋ=gaʔ	iɲ=te	ho=kaɽ=oʔ.
   2sg=inf	neg=1sg	be.able=act.irr	director	early=foc	1sg=obl	that=sg.hum=act.pst
   am=ga	am=na=m.’
   2sg=foc	2sg=mid.irr=2sg
   ‘“In the play I will be him and you will be me.” “No. I can’t be you. The director already made me him. You will be you.”’

This process is entirely productive, given a proper context. See examples (13)–(22), from Peterson (Reference Peterson2011a: 75–92), which give some idea of the pervasiveness of this flexibility throughout the lexicon.

(13)

Interrogatives:  i ‘what; which; do what?’

(14)

Indefinites:  jahã ‘something; some (attribution); do something’

(15)

Quantifiers:  moɲ ‘one (referential/attributive); become one’

(16)

Properties:  rusuŋ ‘red (one); red (attributive); become red’, maha ‘big (one); big; grow, become big’

(17)

Proper names:  aʔghrom ‘Aghrom (name of a town) (referential/attributive); come to be called “Aghrom” (middle voice), name [something] “Aghrom” (active voice)’

(18)

Status and Role:  ayo ‘mother; become a mother (middle voice), accept someone as a mother (active voice)’

(19)

Deictics and proforms:  iɖaʔ ‘yesterday; become yesterday (middle), turn (e.g. today) into yesterday (active, e.g. with God as subject)’

(20)

Physical objects and animate entities:  kaɖoŋ ‘fish; become a fish (middle), turn into a fish (active)’

(21)

Locative:  tobluŋ ‘top, rise (middle), raise (active)’

(22)

Activities:  siloʔ ‘plowing (n.); plowed; plow’

In addition to this precategoriality, what appear to be entire NPs can also denote an event. Consider examples (23)–(24), both of which have the same propositional content.

1. (23)

   Kharia (Peterson Reference Peterson2013: 134)
   ho	rochoʔb=te	col=ki=ɲ
   that	side=obl	go=mid.pst=1sg
   ‘I went to that side’

1. (24)

   Kharia (Peterson 2013: 134)
   ho    rochoʔb=ki=ɲ
   that	side=mid.pst=1sg
   ‘I went to that side’ (lit.: ‘I that-side-d’)

Example (24) shows that an analysis of Kharia as possessing an N/V distinction is problematic. Even if we were to treat rocho^ʔb ‘side’ in (24) as a (presumably zero-derived) verb, we would then have a verb modified by the demonstrative ho ‘that’. As (25)–(26) show, the construction in (24), with an apparent NP as the semantic base of the predicate, is entirely productive and can also contain both quantifiers and genitive attributes, in addition to demonstratives. This is especially problematic for assuming ‘rampant zero conversion’ (Evans and Osada Reference Evans and Osada2005), as this would entail productive zero derivation of a verb (i.e. a lexical stem), from a full-fledged NP (i.e. from a complex syntactic unit).

1. (25)

   Kharia (Peterson Reference Peterson2013: 134)
   ubar   rochoʔb=ki=ɲ
   two	side=mid.pst=1sg
   ‘I moved to both sides (i.e. this way and then that).’

1. (26)

   Kharia (Peterson 2013: 135)
   a.	oʔ=yaʔ      teloŋ=ki.
   	house=gen	roof=mid.pst
   	‘The house’s roof was thatched.’
   b.	oʔ=yaʔ      teloŋ=oʔ=ki.
   	house=gen	roof=act.pst=pl
   	‘They thatched the house’s roof.’

On the basis of data such as these, Peterson (Reference Peterson2013) assumes two structurally defined categories, the Case-syntagma and the TAM/Person-syntagma, both of which have the same potential structure for the semantic base and which differ only with respect to their functional marking, namely case or TAM/voice and person/number/honorific marking, motivated by similar structures assumed for Tongan by Broschart (Reference Broschart1997). The structure of the semantic base is given in (27). None of these elements is obligatory, as long as some non-enclitic unit is present. The Kleene star (*) denotes that potentially any number of content morphemes (lex) is possible, including zero.

(27)

(lex=gen) (dem) (quant (clf)) (lex=gen) (lex*) (=poss) (=num)

The maximal potential structure of the Case-syntagma is given in (28), where X is equal to the structure of the semantic base in (27).

(28)

X=case

Case in (28) refers to postpositions such as buŋ ‘with; ins’, seŋ ‘before’, etc., as well as the two cases given in (29), so that any postposition, =te, or the lack of any overt marking signals that the unit is a Case-syntagma.

1. (29)

   Case:	Direct (zero marking)7 − the case of subjects and indefinite direct objects;
   	Oblique (marked by =te) − marks definite direct objects, ‘indirect objects’ and adverbials

The genitive is not a case in this sense: whereas the direct and oblique cases are relevant at the clause level to mark the relation of a constituent to the predicate (roughly: subject/non-subject), the genitive is only relevant within the semantic base and serves to integrate one semantic base into a larger semantic base. The oblique case also cannot appear in a TAM/Person-syntagma, as (30) shows, whereas the genitive can (31).

1. (30)

   Kharia (Peterson Reference Peterson2013: 139)

   *sahar=te=ki=ɲ.

   city=obl=mid.pst=1sg

   ‘I went to the city.’

1. (31)

   Kharia (Peterson 2013: 139)

   ayo=yaʔ=yoʔ

   mother=gen=act.pst.3sg

   ‘he or she made [it] mother’s’ (lit. ‘he or she mother’s-ed [it]’)

The structure of the TAM/Person-syntagma is given in (32), where X again refers to the semantic base given in (27). (V2) or ‘vector verbs’, as they are often referred to in South Asian studies, refers to phonological and morphosyntactic words which mark Aktionsart or the passive. voice refers to basic voice, that is, the active and middle (Klaiman Reference Klaiman1991).⁸

(32)

X (V2*) (=prf)=tam/voice=person/num/hon

Examples (33)–(36) illustrate this structure with a number of TAM/Person-syntagmas in Kharia. The semantic base is underlined.

1. (33)

   Kharia (Peterson Reference Peterson2013: 142)

   col=ki=may

   go=mid.pst=3pl

   ‘they went’

1. (34)

   Kharia (Peterson 2013: 142)
   boksel=nom	goʔɖ=ki
   sister.in.law=2poss	c:tel=mid.pst
   ‘she became your sister-in-law’ (= ‘she your sister-in-law-ed’)

1. (35)

   Kharia (Peterson 2013: 142)
   ho=jeʔ	u=jeʔ=ki	goʔɖ=ki
   that=sg.nhum	this=sg.nhum=pl	c:tel=mid.pst
   ‘that became these’ (= ‘that these-d’)

1. (36)

   Kharia (AK, 1: 69)9
   ro	tama	am=pe	u	nãw	kuʈum=te=ga
   and	now	2=2pl	this	nine	family=obl=foc
   sadi   biha   kersoŋ=na=pe …
   marry	marry	marry=mid.irr=2pl
   ‘And now you will marry in only these nine families …’

Examples (37)–(38) show a Case-syntagma and a TAM/Person-syntagma with the same semantic base (underlined), differing only with respect to their functional marking. Their basic structures are illustrated in Figure 2.5.¹⁰

1. (37)

   Kharia (elicited)
   bides=aʔ     lebu=ki=yaʔ     rupraŋ=te
   abroad=gen	person=pl=gen	appearance=obl
   ‘the appearance of foreigners’ (e.g. as the object of the predicate)

1. (38)

   Kharia (Peterson 2013: 157)
   bides=aʔ      lebu=ki=yaʔ      rupraŋ=ki=may.
   abroad=gen	person=pl=gen	appearance=mid.pst=3pl
   ‘They took on the appearance of foreigners (e.g. by living abroad so long).’

Figure 2.5 TAM/Person- and Case-syntagmas with the same semantic base

These two syntactic units are defined purely structurally, and both can appear in attributive, predicative and referential functions, as examples (39)–(42) from Peterson (Reference Peterson2013: 137) show.

Case-syntagma in attributive function

1. (39)

   Kharia
   kuda	koloŋ	daru
   millet	bread	Tree
   ‘a millet	bread	tree’ (name of a children’s story)
   (Note: kuda modifies koloŋ and the two together modify daru.)11

TAM/Person-syntagma in attributive function

1. (40)

   Kharia (Peterson Reference Peterson2011a: 74)
   yo=yoʔj	lebu	col=ki.
   see=act.pst.1sg	man	go=mid.pst
   ‘The man I saw left.’

Case-syntagma in predicative function

1. (41)

   Kharia (AK 1: 57)
   …	ro	u=ga	ho   jinis=aʔ       komaŋ.
   	and	this=foc	that	animal=gen	meat
   ‘… and this [is] that animal’s meat.’

TAM/Person-syntagma in referential function

1. (42)

   Kharia (MT, 1: 180)
   kunɖab   aw=ki		tomliŋ	khaɽiya	gam	ɖom=na
   behind	qual=mid.pst	milk	Kharia	say	pass=inf
   laʔ=ki=may	ina no
   ipfv=mid.pst=3pl	because
   u=ki	tomliŋ	uʔɖ=ga	ɖel=ki=may.
   this=pl	milk	drink=foc	come=mid.pst=3pl
   ‘[Those who] were in the rear (= literally: ‘they were behind’) were called ‘Milk Kharia’ because they came drinking milk.’

Finally, an analysis of the forms given in the preceding pages as consisting of nominals and ‘hidden verbs’ is also not tenable. As noted above, it is not possible in Kharia to combine TAM/Person marking with case marking or with postpositions, so that a construction such as the one in (43), which ends in a postposition, is not grammatical if directly followed by TAM/Person-marking, as (44) shows.¹²

1. (43)

   Kharia (Peterson Reference Peterson2013: 145)
   ikuʔɖ	jughay	duʔkho	buŋ
   very	much	sorrow	ins
   ‘very sad’, literally: ‘with very much sorrow’

1. (44)

   Kharia (Peterson 2013: 145)
   *ikuʔɖ   jughay   duʔkho   buŋ				goʔɖ=ki.
   very	much	sorrow	ins	c:tel=mid.pst
   ‘(He) became very sad’ (= ‘with very much sorrow’)

However, if the semantic part of this unit is re-ordered as in (45), with two constituents instead of just the one in (43), the unit which is not marked by a postposition may serve as the semantic base of a TAM/Person-syntagma. This is a common predicate type in Kharia (Peterson Reference Peterson2011a: 220).

1. (45)

   Kharia (RD, 1: 18)
   gupa	lebu	duʔkho   buŋ   ikuʔɖ   jughay				goʔɖ=ki.
   watch	person	sorrow	ins	very	much	c:tel=mid.pst
   ‘The shepherd (= watch person) became very sad.’

At issue here is that the unit ending in the postposition may not be directly followed by a TAM marker, whereas the unit which does not end in a postposition/case marker can. Furthermore, if a ‘light verb’ such as hoy ‘become’ (borrowed from Kharia’s Indo-Aryan neighbour Sadri) is inserted between buŋ ‘ins’ and go^ʔɖ=ki in (44), this results in a grammatical predicate, consisting of a Case-syntagma and a marker of qualitative predication (≈ copula).¹³

1. (46)

   Kharia (Peterson Reference Peterson2011a: 111)
   ikuʔɖ   jughay   duʔkho   buŋ				hoy	goʔɖ=ki.
   very	much	sorrow	ins	become	c:tel=mid.pst
   ‘(He) became very sad.’

Thus ‘hidden verbs’ in Kharia would have a very different distribution from the ‘light verb’ hoy ‘become’ or in fact any other supposed verb. They could only be assumed to occur after ‘predicate nominals’ which are not case-marked (e.g. for the structure in (34), i.e. those units we consider to be the semantic base of either a TAM/Person- or a Case-syntagma), but not in structures such as (44), so that they would also not be able to occur where the overt ‘light verb’ hoy does (compare (44) with (46)). Thus, hidden verbs do not simplify the analysis but rather complicate it considerably, as they would not only be non-overt but would also have a unique distribution.

Example (47) presents a simplified example of an intransitive sentence in Kharia and Figure 2.6 presents its constituent projection in the present analysis.

1. (47)

   Kharia (adapted from Roy and Roy Reference Roy and Roy1937: 180–181)
   u=kiyar	taŋ	el=aʔ	sori=ga	ɲog=e	uɖ=e=kiyar.
   this=du	now	1pl.excl=gen	com=foc	eat=act.irr	drink=act.irr=du
   ‘They two will now eat and drink with us.’

Figure 2.6 The constituent projection of (47)

RRG can thus easily account for analyses of individual languages which do not have supposedly universal categories such as ‘noun’, ‘adjective’ and/or ‘verb’, without needing any empty categories or hidden verbs.

Finally, there is a rather marginal type of TAM/Person-syntagma in Kharia, quotatives, in which the semantic base consists of an entire sentence, as in example (48).

1. (48)

   Kharia (Kerkeʈʈā Reference Kerkeʈʈa1990: 31)
   iɖib	tunboʔ	‘kersoŋ=e       la!   kersoŋe la!’    loʔ=na=kiyar.
   night	daytime	marry=act.irr	voc	rep		cont=mid.irr=du
   ‘Day and night they both will keep on [say]ing “Marry! Marry!”’

Although Peterson (Reference Peterson2013) offers no final analysis of this form, as it is unique in his corpus, the next section, which deals with very similar predicates in the Chapakuran language Wari’ in which these structures appear to be much more common, offers a possible solution to analysing predicates of this type in Kharia.

2.2.3.2 The ‘Intentional State Constructions’ in Wari’

A similar construction to the rather marginal Kharia quotative predicate discussed in the previous section is the much more productive ‘intentional state construction’ (ISC) in Wari’, a member of the Chapakuran family of Brazil and Bolivia, to which we now turn.

As Everett (Reference Everett2008: 383) notes, in Wari’, a predicate-initial language (VOS), in verb-initial sentences the verb is immediately followed by the VIC or ‘verbal inflectional clitic’, which marks for tense, voice, person, number and gender and agrees with both subject and object. See examples (49)–(50), where the VIC is underlined.

1. (49)

   Wari’ (Everett Reference Everett2008: 383)
   Quep	na-in	xirim	te	pane	ta.
   do	3sg.real.pst/prs-3n	house	father.1sg	rem.pst	emph
   ‘My father made a house long ago.’

1. (50)

   Wari’ (Everett 2008: 384)
   Ten	ta	wao’.
   weave	pass.3sg	type of basket
   ‘Baskets are woven.’

In sentences in which the verb is not the first sentence constituent, for example with constituent questions, this sentence-initial unit is followed by a different class of clitics which mark tense and which agree with the gender of the item in sentence-initial position, regardless of its function in the sentence, as in (51).

1. (51)

   Wari’ (Everett 2008: 385)
   Ma’	co	tomi’	na?
   that.prox.hearer	m/f.real.pst/prs	speak	3sg.real.pst/prs
   ‘Who is speaking?’

As Everett (Reference Everett2008: 386) writes, ‘Many Amazonian languages report on others’ thoughts, character, reactions, and other results of intentional states by means of quotatives, i.e. literally putting words in people’s mouths.’ Wari’ also makes use of quotatives for these purposes, but their use is apparently much more common in that language, having also spread to other uses. What is interesting is that in Wari’, similar to the Kharia example in (48), there is no morpheme denoting ‘say’. (52) provides an example of this construction in Wari’, where the indices _{i, j} and _k show the respective congruence relationships.

1. (52)

   Wari’ (Everett Reference Everett2008: 386)
   Ma’	co	mao	na	-ini
   that.prox.hearer	m/f.real.pst/prs	go.sg	3sg.real.pst/prs-3n
   Guajarái	naj	-namk	‘oro	narimak’	taramaxiconj.
   Guajará (Brazilian city)	3sg.real.pst/prs-3pl.f		collective	woman	chief
   ‘“Who went to Guajará?” (said) the chief to the women.’

Everett (Reference Everett2008: 391–392) notes a number of similarities between the ISC predicator and simple verbs, suggesting that the clausal semantic base of the ISC occupies the same functional slot as verbs otherwise do, namely the predicate, although it consists of an entire clause. These include:¹⁴

• the ISC predicate appears in the clausal position otherwise occupied exclusively by the verb

• the last syllable of the ISC predicate is stressed, as with the verb in other sentence types. This suggests that the ISC predicate is considered a single unit

• the predicate of an ISC may be modified just like a verb

• the predicate of an ISC may undergo compounding, like all verbs.

If the ISC predicate is considered the predicate, just like verbs in verb-initial sentences, the following generalizations may be made for constituent order in Wari’ (Everett Reference Everett2008: 392):

• Wari’ sentences always begin with a verb, ISC predicate, or preverbal mood marker

• the VIC appears directly after the predicate, whether this is clausal or a simple verb

• tense appears in sentence-second position.

Everett (Reference Everett2008: 399) argues that Wari’ inflectional clitics – that is, those which follow the predicate of the clause – follow the NUC in an RRG analysis. As such, they follow the ISC predicate as well, which then occupies the NUC slot. Figure 2.7, adapted from Everett (Reference Everett2008: 398),¹⁵ shows the structural analysis which he assumes for example (52), where the entire ‘quote’ is viewed as a nucleus.¹⁶ The structure in Figure 2.7 captures the fact that the nucleus of the clause, which contains the predicate, consists of a form which, despite its role as the predicate, also has many of the trappings of a clause, as it consists of a precore unit and a core, which itself contains a nucleus (predicate) and two arguments. Nevertheless, it is clearly the predicate of the larger clause. This then, together with the two RPs narima’ and taramaxicon, forms the core of the higher-level clause.

Figure 2.7 Constituent projection of (52)

(adapted from Everett Reference Everett2008: 398)

2.3.1 Proposition-Grounding Operators

We begin with a discussion of propositional grounding, which is expressed in individual languages by language-specific grammatical categories, such as present tense, past tense, hearsay, mirative, etc. The grounding information expressed by these language-specific categories is divided into three different groups according to which level of the layered structure of the clause (LSC) they ground, the nucleus, the core or the clause. RRG assumes that individual languages will differ as to which of these grounding concepts are expressed through grammatical categories, and there is no requirement that every language have all of the possible operators. In fact, most languages do not have grammatical categories expressing all of these semantic categories, although it is assumed that all languages will have grammatical means to express at least two of these categories – negation and illocutionary force (Van Valin 2005: 9). These operators are summarized in Table 2.2.

Nuclear operators such as aspect or directionals modify the state/event itself without reference to the participants, while the core-level operators modify the relation between the arguments and the event, most notably directionals, (deontic) modality (‘be able’, ‘must’, etc.), narrow scope negation, and event quantification. Finally, operators which have scope over the entire clause include ‘status’ (epistemic modality and external negation), as well as tense, evidentials and illocutionary force.

Operators are represented by an operator projection, given below the respective language example. This projection is joined to the constituent projection through the nucleus. Figures 2.9 and 2.10 illustrate this for English and Japanese.

Figure 2.9 Constituent and operator projections for two English sentences

(adapted from Van Valin 2005: 14)¹⁹

Figure 2.10 Constituent and operator projections in a Japanese sentence

(adapted from Van Valin 2005: 14)

Note that in the left-hand example in Figure 2.9 the information for the aspect operator comes from two different positions: the auxiliary be preceding leaving and the suffix ‑ing on this form. In fact, this information can potentially appear anywhere in the sentence. The only claim that RRG makes with respect to the placement of operators is that ‘the ordering of the morphemes expressing operators with respect to the verb indicates their relative scopes. That is, taking the nucleus as the reference point, the morphemes realizing nuclear operators should be closer to the nucleus than those expressing core operators, and those manifesting clausal operators should be outside of those signaling nuclear and core operators’ (Van Valin Reference Van Valin2005: 11). All evidence so far suggests that this is true of language in general. Although there may be language-specific variation of the ordering of the operators within a particular level, cross-linguistic data for operator marking appears to substantiate this claim for those markers for different levels which appear together on the same side of the nucleus: see examples (53)–(57) (from Van Valin 2005: 10).¹⁸

1. (53)

   Kewa (Papua-New Guinea) (Van Valin 2005: 10)
   a.	Íra-pa-niaa-ru.
   	cook-prf-down-1sg.pst
   	‘I burned it downward (as a hill).’	(V-ASP-DIR-TNS)
   b.	Íra-pa-saa-ru.
   	cook-prf-up-1sg.pst
   	‘I burned it upward (as a hill).’	(V-ASP-DIR-TNS)

1. (54)

   Turkish (Van Valin 2005: 10)
   a.	Gel-iyor-du-m.
   	come-prog-pst-1sg
   	‘I was coming.’	(V-ASP-TNS)
   b.	Gel-emi-yebil-ir-im.
   	come-able.neg-psbl-aor-1sg
   	‘I may be unable to come.’	(V-MOD-STA-TNS)

1. (55)

   Korean (Van Valin 2005: 10)
   An-tul-li-wu-(si)-lswuiss-cianh-ass-keyss-up-nita.
   neg-hear-caus-pass-(sh)-able-neg-pst-presum-pol-decl
   ‘(I) guess that (he) [HON] might not be heard.’	(NEG-V-MOD-NEG-TNS-EVID-IF)

1. (56)

   English (Van Valin 2005: 10)
   a.	He may be leaving soon.	(IF/TNS-STA-ASP-V)20
   b.	She was able to see them.	(IF/TNS-MOD-V)
   c.	Will they have to be leaving?	(IF/TNS-MOD-ASP-V)

1. (57)

   Tiwi (Van Valin 2005: 10)
   ŋə-ru-uṉṯiŋ-apa.
   1sg-pst-prog-eat
   ‘I was eating.’	(TNS-ASP-V)

Cross-linguistic data on the position of operators (or markers of finiteness in general)²¹ within the sentence also suggests that these operators can theoretically appear at any position in the sentence and are not restricted to any particular position (e.g. adjacent to the verb, sentence-final, or sentence-second (i.e. the Wackernagel position)), and it is only their respective order when they appear together on the same side of the predicate which is restricted. For example, in his cross-linguistic study of finiteness, Maas (Reference Maas2004) suggests the typology of morphological finiteness given in Table 2.3. Types I, II and III denote simple predicates which consist of a single predicating unit (generally a simple verb), while Types Ia, IIa, IIb and IIc are complex predicate types which consist of at least two parts (whether verbal or not) which together express a single predicate.

Types I and Ia are typical of isolating languages. Here, neither the simple predicate, consisting of a single element (usually a verb) (Type I, cf. (58)a), nor either of the two or more parts of the complex predicate (Type Ia) is morphologically marked for ‘finite’ categories (cf. (58)b).

1. (58)

   Mandarin Chinese
   a.	Type I:	tā	qù
   		3sg	go
   		‘s/he goes’
   b.	Type Ia:	tā	yào	qù
   		3sg	want/fut	go
   		‘s/he wants to / will go’

In Type II, typical of fusional and agglutinating languages, the simple predicate is marked for all finite categories (Type II), while for the complex predicates in Types IIa–c either the modified element (i.e. the lexical verb), the modifying or situating element (e.g. the auxiliary) or both is/are marked for all finiteness categories. Note that only complex predicates of Type IIb correspond to the traditional notion of auxiliaries. Also, as the Maltese examples in (59b–c) and (61a–b) show, languages can belong to more than one category.

1. (59)

   Maltese (Semitic, Neo-Arabic, Malta)
   a.	Type II	ji-kteb
   		3sg.m-write.ipfv fut
   		‘he writes’
   b.	Type IIa	sa	ji-kteb
   		fut	3sg.m-write.ipfv
   		‘he will write’
   c.		qed	ji-kteb
   		prog	3sg.m-write.ipfv
   		‘he is writing’

1. (60)

   Type IIb:	English
   a.	She has written.
   b.	I have gone.

1. (61)

   Type IIc:	Maltese
   a.	ir-rid	im-mur
   	1sg-want.ipfv	1sg-go.ipfv
   	‘I want to go’ (literally: ‘I-want I-go’)
   b.	kon-t	mur-t
   	cop.pst-1sg	go.pfv-1sg
   	‘I had gone’ (literally: ‘I-was I-went’)

Finally, in Type III at least some of the language-specific finiteness categories are marked elsewhere in the clause and not together with the other finiteness categories. For example, in (62), from Paez, a Chibcha language (possibly a language isolate) spoken in Columbia, all finiteness markers attach to the predicate in the default case, as in example (62a), with the form uʔxuetstxu ‘I went’, where the marker ‑txu ‘fac.1sg’ directly attaches to the verb form. In (62b), on the other hand, it attaches (in its phonologically determined alternative form ‑tx) to the nominal element nengasu ‘through Belcazar’ to mark this element as focused, while the verb form uʔxuets is now ‘partially finite’ (see the discussion in Maas Reference Maas2004: 378).

1. (62)

   Type III:	Paez (Maas Reference Maas2004: 378)
   a.	xuʔna	ʧamb-na	uʔx-ue-ts-txu	juʔ	nava	kar:o	suw-ku.
   	yesterday	village-to	go-ipfv-prog-fac.1sg	them	but	car	broken-fac.3sg
   	‘Yesterday I went to the village, but the car was broken.’
   b.	nenga-su-tx		uʔx-ue-ts	juʔ	saʔ	tja-xu	paʔ	jaʔ-tx.
   	Belcazar-through-fac.1sg		go-ipfv-prog	them	and	dem-from	come	already-fac.1sg
   	‘I passed Belcazar and am already back from there.’

As noted above, RRG does not posit a privileged position for such marking, and markers of these categories can appear almost anywhere in the sentence. This allows us to account for such cross-linguistic variability in a straightforward way, without having to resort to movement or other processes to derive the actual word order in a particular language. In fact, the only claim that RRG makes is that the ordering of these morphemes with respect to the predicate – when they appear on the same side of this unit – corresponds to the relative scopal properties of the individual operators. And in the case of the Paez example in (62) this is unproblematic, as the ‘mobile’ markers are clausal operators which appear in the final position when attached to the verb in (62a), after aspectual marking and therefore conforming to this constraint. The position of this marking in (62b) directly preceding the predicate is also not an exception to this rule, as here the marking of this clausal operator is not on the same side of the verbal root as the other operators and hence this constraint does not apply here.

Finally, as discussed in Chapter 13 of this volume, the scopal properties of these operators play a pivotal role in the discussion of nexus and juncture relations in RRG. Consider Figure 2.11, from Kharia (own data), which provides an example of nuclear cosubordination, as the scope of kan, a marker of aspect/Aktionsart, has obligatory scope over both nuclei, leŋ ‘fly’ and col ‘go’.

Figure 2.11 Constituent and operator projections in Kharia

Operators are represented formally in the semantic representation of the sentence in italicized caps within angled brackets indicating their scope within the logical structure. The overall pattern is given in (63), from Van Valin (Reference Van Valin2005: 50), with arbitrarily chosen values. Note that each of the individual operators refers to a grammatical category within a specific language and as such its value is subject to the system of that language, for example, with respect to tense, which could be pst/npst in one language, pst/prs/fut in another, or fut/nfut in yet another.

(64), also from Van Valin (Reference Van Valin2005: 50), provides an example of the formal representation of the operators for the English question Has Kim been crying?.

2.3.2 Reference-Grounding Operators

As noted in Section 2.3, the RP also partakes in the semantic/pragmatic grounding of the clause, for instance with respect to definiteness. Table 2.4 summarizes those operators which are assumed for the RP in RRG. Table 2.4 is still somewhat tentative as it is based on the operators which RRG assumed for the NP before the notion of the RP was introduced in Van Valin (Reference Van Valin2005: 28) and further justified in Van Valin (Reference Van Valin2008). We thus tentatively assume that the same operators hold for the RP as for the NP, as we are dealing with the same underlying function here, namely the semantic/pragmatic grounding of referential units, although this topic requires further study.

As with the sentence, RPs also possess an operator projection. Figure 2.12 provides an example of such a projection for an RP in English.

Figure 2.12 Layered structure of the RP in English

(adapted from Van Valin 2005: 25)

Similar to the position of sentence-grounding operators discussed in the previous section, there are no general requirements in RRG that the individual operators appear adjacent to any particular element(s) of the RP, only language-specific rules regulating this, although it is expected that operators appearing on the same side of the nucleus will be arranged according to their scope, with nuclear operators appearing closer to the nucleus, followed by core operators.

To give an example of the positional variability of these operators, consider the different orders in the Dyirbal examples in (65). Dyirbal is spoken in northern Australia and has exceptionally free word order. Note that all three variants in (65) are grammatical and have the same meaning. Note also the position of the determiners with respect to the nouns which these modify.

1. (65)

   Dyirbal (Van Valin 2005: 28)
   a.	Bayi	bargan	baŋul	yaɽa-ŋgu	d̨urga-ɲu	gambi-ɽa.
   	det.abs	wallaby.abs	det.erg	man-erg	spear-tns	mountains-loc
   	‘The man speared the wallaby in the mountains.’
   b.	Baŋul	gambi-ɽa	yaɽa-ŋgu	bayi	d̨urga-ɲu	bargan.
   	det.erg	mountains-loc	man.erg	det.abs	spear-TNS	wallaby.abs
   c.	Yaɽa-ŋgu	d̨urga-ɲu	gambi-ɽa	bargan	baŋul	bayi.
   	man.erg	spear-TNS	mountains-loc	wallaby.abs	det.erg	det.abs

The operator projection of the various orders in (65) is straightforward and is given in Figure 2.13, which presents the projection of the structures in (65b).

Figure 2.13 The structure of the Dyirbal sentence in (65b)

(adapted from Van Valin 2005: 29)

Similar to the proposition-grounding operators in the semantic representation of the sentence, the reference-grounding operators are represented in the semantic representation of the RP. An example is given in (66), from Van Valin (Reference Van Valin2005: 52), assuming that these are the same for the RP as those previously assumed for the NP. The conventions used here (i.e. angle brackets denoting scope, italicized small caps, etc.) are the same as those used to represent the proposition-grounding operators. ∃ is the existential operator, roughly meaning ‘there is (at least one)’.

The logical structure for a phrase like the scarf would then be represented as in (67), also from Van Valin (Reference Van Valin2005: 52).

Although in principle every RP in a sentence has such a semantic representation, for practical reasons it is seldom feasible to include all of this information in the full semantic representation of the sentence, hence it is generally omitted. An example of a full structure is given in (68) for the simple intransitive question Has the tall man been crying? (Van Valin 2005: 53). The underlining of man in the semantic representation of the RP denotes that it is the head of the RP.²²

(68)

3.2.1 Semantic Representation of Predicates

The classification of predicates in RRG is an enriched version of Vendler’s (Reference Vendler1967) theory of Aktionsart. Van Valin (Reference Van Valin2005) augments the four basic classes (state, activity, achievement and accomplishment) with one more class, namely, semelfactives. He also distinguishes causatives as a parameter that crosscuts these classes.¹ Each of these classes is represented by a logical structure (see Chapter 1). States and activities are primitives. In fact, they are the definiens or genus in the representation of achievements, semelfactives and accomplishments.

For example, a state predicate, such as see is represented by see′ (x, y). The representation is composed of a primitive and two argument variables. Activity predicates are expressed with do′,² which is a characteristic of this class of predicate, along with a primitive designating the activity carried out. Activity predicates can have one argument (do′ (x, [hiccup′ (x)])), or two arguments (do′ (x, [write′ (x, y)])). As previously mentioned, achievements, semelfactives, accomplishments and active accomplishments are based on states and activities in combination with operators.

More specifically, an achievement comprises a state or activity predicate combined with INGR,³ indicating the sudden onset of the eventuality. For instance, the vase cracked is represented as INGR cracked′ (vase). Semelfactive predicates have a SEML operator and can be either activities or states. The activity designated by these predicates is punctual but does not cause a change of state or lead to a result. Thus, the janitor hiccupped would be SEML do′ (janitor, [hiccup′ (janitor)])). A semelfactive predicate can also be a state, for example flicker (SEML shine′ (x, y)).

Accomplishments are represented by means of BECOME, which signals a change of state that reflects some sort of internal transition or transformation, for example ossify: BECOME hardened′ (x). In Van Valin (Reference Van Valin2005: 44, Reference Van Valin, Kailuweit, Künkel and Staudinger2018), BECOME is broken down into PROC, the operator which indicates the duration of the change, and INGR, the operator which indicates the change of state itself. Active accomplishments are a complex structure, composed of an activity and an achievement, which reflects the telic nature acquired by the predicate in conjunction with a noun phrase or an adpositional phrase. For example, the witch brewed a potion is represented as follows: do′ (witch [brew′ (witch, potion)]) & INGR exist′ (potion). This formulation also includes the ampersand (&), which means ‘and then’. Finally, each of these predicate classes also has a causative version.⁴

Although verbs are extremely important, language competence also involves other categories. The RRG treatment of adjectives, adverbs and nouns is explained in Van Valin (Reference Van Valin2005: 48–51) and Van Valin and LaPolla (Reference Van Valin and LaPolla1997: 154–171). When adjectives function as predicates, they are treated as states in RRG logical structures (Van Valin Reference Van Valin2005: 48).

Adjectives can also occur as part of attributive predications. Note that be′ is only found in the representation of attributive predications, not in the representation of the resulting state; moreover, be′ is not the representation of the copula, but the operator of a state, comparable to the operator do′. The operator be′ thus figures in these representations as the principal characteristic of attributive logical structures:

1. (2)

   a.	Laura is wealthy.
   	be′ (Laura, [wealthy′])
   b.	John is afraid of birds.
   	feel′ (John, [afraid′ (birds)])

In regard to adverbs, Van Valin (Reference Van Valin2005: 49) explains that they are represented as one-place predicates and can modify different parts of a logical structure. For example, the argument of a time adverb can be the entire logical structure. In (3), the entire event of Lucas walking and then beginning to be at the zoo takes place within the time frame designated by tomorrow.

Generally speaking, manner adverbs modify activity predicates, whereas pace adverbs can modify any type of durational predicate. In (4), the activity of stealing, performed by Bugsy, is both an activity and durational predicate. As a result, it can be carried out in a certain way (furtively) and at a certain speed (fast):

Aspectual adverbs modify basic state or activity predicates. In (5), there is an evident change of state since the water that was in the teapot is no longer in a liquid state because it has been transformed into vapour.

In (6), a sustained activity is performed by the juror, who does not stop moving throughout the time period of the stated event (trial).

Finally, Van Valin (Reference Van Valin2005) describes how predicative prepositions are represented and underlines the fact that they are also state predicates. As shown in (7), the logical structure be-in′ is the highest predicate since the grassland is the place where the kangaroo is jumping. As such, it has two arguments, namely, the grassland and the logical structure for jump.

This contrasts with the logical structure of the active accomplishment, the kangaroo jumped to the grassland, in which the predicative preposition (PP) indicates the final location of the kangaroo and not the location of the jumping event.

3.2.2 Semantic Representation of Nouns

Nouns as a grammatical and semantic category have never been a focus in RRG since their syntax is less complex. Up until now, RRG has proposed a semantic representation based on the qualia analysis in Pustejovsky (Reference Pustejovsky1995). Although, unlike verbs, nouns (e.g. cat and table) lack a logical structure, they have semantic properties in the form of qualia roles, which parameterize their meaning. According to Pustejovsky et al. (Reference Pustejovsky, Havasi, Littman, Rumshisky and Verhagen2006: 3), nouns can be described in terms of the following:

• formal role: the basic type distinguishing the meaning of a word

• agentive role: the factors involved in the object’s origins or ‘coming into being’

• telic role: the purpose or function of the object if there is one

• constitutive role: the relation between an object and its constituent parts.

Since the formal role refers to the type_of relation and the constitutive role refers to the part_of relation, this is still another indicator of an implicit typology of concepts underlying the semantic description of nouns. Just as important is the fact that the agentive role identifies a set of events linked to the object concept, whereas the telic role refers to an event description, which is associated with that object as its function (Pustejovsky et al. Reference Pustejovsky, Havasi, Littman, Rumshisky and Verhagen2006: 333). According to Pustejovsky (Reference Pustejovsky, Bouillon and Busa2001: 5), qualia provide functional tags for words, which are linked to the network of concepts in Figure 3.1.

Figure 3.1 Tripartite concept lattice

(Pustejovsky Reference Pustejovsky, Bouillon and Busa2001)

Within the context of this concept lattice or network, each entity is described in terms of its qualia structure. For example, Figure 3.2 shows the representation of the qualia structure of violin, which is classified as a functional entity or artefact that has been created for a purpose. The formal role (F) indicates that violin is a type of musical instrument. Its agentive role (A) shows that a violin is the result of a building process. The telic role (T) reflects that the purpose of a violin is to produce music; and the constitutive role (C) indicates the parts of the violin, such as the strings.

Figure 3.2 Qualia structure of violin

(Pustejovsky and Batiukova Reference Pustejovsky and Batiukova2019: 162)

The variables w, x, y and z refer to the different entities involved in events typically associated with a violin. The violin itself is assigned the variable x and is given a basic type in F; the agent building the violin is y (‘y builds x’), as expressed in A. The person using the violin in T to produce the musical sound on the violin is z (‘z produces sound on x’); and in C, the strings are identified as w (‘w of x’) (Pustejovsky and Batiukova Reference Pustejovsky and Batiukova2019: 162).

This overview of the RRG treatment of lexical units shows that words belonging to all grammatical categories would benefit from a system of semantic decomposition, which would make their meaning explicit and highlight the relations between words in the same category as well as with those in different categories.

For example, such a system would show that predicates describing emotional states could be further decomposed in terms of feel, or those predicates that express possession could be decomposed in terms of have. Artefact entities such as table, gun and bread would be assigned to conceptual categories of furniture, weapon and food, respectively. Humans would also have different types of social roles (aunt, neighbour, friend, etc.) and professional roles (lawyer, zookeeper, hairdresser, etc.).

Despite the fact that in RRG there is still no clearly defined system that indicates which words should be treated as primitives and which should be defined in terms of other more basic lexical units, this could be achieved by specifying the meaning representations of lexemes and linking them to an underlying ontological system.

3.2.3 Semantic Decomposition in RRG

Although the use of English words to designate potential primitives is a notational convenience (Van Valin Reference Van Valin2005: 46), the criteria used to select these primitives are somewhat less clear. Even though lexical primitives may have conceptual correlates, they still lack an ontology. There is also the need for research that will provide evidence of the scope of semantic primitives or at least give some indication of their potential cross-linguistic validity. This would doubtlessly contribute to clarifying the lexical–conceptual interface and how it works.

In the RRG lexical representation system, another issue is why certain predicates undergo semantic decomposition whereas others do not. For example, in the lexical representations for love (8) and snore (9), the same predicates (love′ and snore′) are used in their own definitions. Nonetheless, love, which is a verb of feeling, could be defined in terms of feel, whereas snore is a type of sound emission.

The same also occurs in (10) and (11) except for the fact that the BECOME operator introduces the resulting state.⁵

These examples are in direct contrast to the representations in (12–14) in which each predicate is semantically decomposed. More specifically, learn is defined as to come to know, receive is to come to have, and show is to cause to see. In this causative sense, x does something that causes y to come to see z.

According to Van Valin (Reference Van Valin1993) and Van Valin and Mairal Usón (Reference Van Valin, Mairal Usón, Gómez, de Mendoza and Gonzálvez-García2014), it is necessary to improve the lexical representation system so as to obtain more consistent semantic decompositions. However, part of the problem lies in the fact that logical structures, as they were initially conceived, only focus on capturing grammatically relevant aspects of word meaning and neglect contextual meaning.

3.3.1 Conceptual Logical Structures

Adopting this conceptual approach would significantly modify RRG in that logical structures (LSs) would become conceptual logical structures (CLSs). Although CLSs resemble LSs in that they maintain the Aktionsart distinctions, the two differ because in CLSs, predicates become concepts. This type of ontologically based representation has the format in Table 3.1.

As can be observed, in Table 1, the lexical primitives are replaced by concepts (<C>), which provide a more granular semantic decomposition and represent any conceptual unit, whether basic or terminal. One of the advantages of this approach is that each lexical unit is linked to a conceptual unit and is enriched by the inheritance of the conceptual properties (i.e. thematic frame and meaning postulate) of the related conceptual unit. As widely discussed in Van Valin and Mairal Usón (Reference Van Valin, Mairal Usón, Gómez, de Mendoza and Gonzálvez-García2014), CLSs arise from the interaction of information in the ontology and the lexicon. In order to ascertain how the ontology and the lexicon actually interact and arrive at what has come to be known as the lexical–conceptual interface, it is necessary to review what type of information is present in both the ontology and the lexicon.

The choice of English as the conceptual language for lexical representation is not a problem since FunGramKB is based on the premise that at a certain level, all languages contain a set of core concepts that lie at the heart of communication. Since the ontology contains non-language-specific information, the language used to encode concepts is irrelevant (in fact, numbers could also have been used). Each concept derives its meaning from its position in the ontology as well as from its relations with other concepts. In other words, a concept is defined by its conceptual route. This is exemplified in the representation of $STERILIZE _00 in Figure 3.3.

Figure 3.3 FunGramKB Ontological Hierarchy for $STERILIZE_00

The ontology in Figure 3.3 represents $STERILIZE_00 as a material event that is a type of transformation process, represented by the meta-concept #TRANSFORMATION. At the more subordinate levels are the basic concepts, +CHANGE_00 and +CLEAN_01. As shown, +CHANGE_00 is a type of transformation, whereas +CLEAN_01 is a type of change in that an entity that is dirty is changed, so that the dirt is removed and it becomes clean.

In Figure 3.3, $STERILIZE_00 is a daughter concept of +CLEAN_01. The cleaning process in $STERILIZE_00 involves killing bacteria (or something perceived to have bacteria). As such, $STERILIZE_00 is a terminal concept with various predicates bound to it, namely, fumigate, sanitize, disinfect, pasteurize, decontaminate and sterilize. Predicate selection is constrained by the context. World knowledge (not syntax) pertaining to the entity with bacteria, the sterilization instrument or method used, and the end result pursued guide the speaker’s choice of predicate. Although the meaning distinctions encoded in the predicates do not have an impact on syntactic form, this type of semantic–pragmatic information should be included in the RRG lexicon since it is a major component of lexical competence.

Moreover, the fact that lexical entries are linked to conceptual units and thus have different conceptual routes is a way of resolving polysemy. CLSs facilitate the selection of the most suitable concept in the case of predicates with more than one meaning. In this sense, each meaning of a predicate is linked to a different concept. The result is a robust enumerative lexicon with fine-grained meaning postulates, capable of generating a complex conceptual network. It is thus no longer necessary to condense all the meanings of a lexical unit into a single infra-specified representation as proposed by Generative Lexicon Theory.

In CLSs, differences in meaning are not marked by the syntactic behaviour of a lexeme but rather are specified in its meaning. Accordingly, nuances of meaning resulting from differences in syntactic constructions do not affect the conceptual representation or the selection of the conceptual unit. This method of dealing with polysemy is exemplified by fix. For instance, one of the various meanings of fix is to repair something that is broken or not working properly (15).

Another meaning of fix is to prepare a meal or drinks.

FunGramKB links each of these senses to a different concept: +REPAIR_00 and +COOK_00, respectively. Consequently, depending on the sense, fix has one of the following formats:

1. (17)

   a.	do′ (x, [+REPAIR_00 (x, y)])
   b.	do′ (x, [+COOK_00 (x, y) & become <+EXIST_00> (y)]

The selection of (17a) or (17b) depends on the basic conceptual event since cooking is a creation event whereas repairing is a change event. Accordingly, +REPAIR_00 and +COOK_00 differ in their conceptual routes and positions in the ontology. In the case of +COOK_00, the conceptual path is #MATERIAL >> +DO_00 >> +CREATE_00 >> +COOK. Its meaning postulate is the following:

In contrast, the conceptual path of +REPAIR_00 is: #MATERIAL >> #TRANSFORMATION >> +CHANGE_00 >> +REPAIR_00. However, the change resulting from a +REPAIR_00 event is of a different type since it involves changing an entity that was damaged in the past so that it can function again.

The existence of a previously specified ontology provides a more specific conceptual context that allows users to access concepts from a structured inventory. There is thus no need to create ad hoc primitives on demand since it is the ontology that determines the concept unit in the representation.

3.3.2 Case Studies: pasteurize, obsolete and invoice

This section provides examples of the CLSs of verbs, adjectives and nouns.⁹ As previously mentioned, each lexical unit in the lexicon is linked to a conceptual unit in the ontology.

For example, −PASTEURIZE is a subconcept linked to the terminal concept, $STERLILIZE_00 and as such, it is marked with a dash (−). Its conceptual route is #EVENT > #MATERIAL > #TRANSFORMATION > +CHANGE_00 > +CLEAN_01 > $STERILIZE_00. It does not appear in the basic hierarchical structure of the ontology since it shares the basic thematic frame and meaning of $STERILIZE_00. The difference lies in the fact that in −PASTEURIZE, there is a specification of the second argument (milk). Words in different languages can thus be linked to −PASTEURIZE, such as pasteurize (English), pasteurizar (Spanish), pastörize etmek (Turkish), 巴氏滅菌法 (Chinese). Example (20) shows a partial representation of the CLS for the predicate pasteurize.

In (20), pasteurize is represented as an activity predicate that is the lexical designation of the conceptual unit −PASTEURIZE. However, this predicate can be decomposed further. Figure 3.4 shows the conceptual entry for −PASTEURIZE, which consists of a thematic frame and a meaning postulate. Thematic frames and meaning postulates are two complementary types of conceptual schema used to encode the meaning of the concepts in the ontology.

Figure 3.4 FunGramKB conceptual entry for −PASTEURIZE

The thematic frame for −PASTEURIZE is inherited from the terminal concept $STERILIZE_00, which involves #TRANSFORMATION. In the thematic frame of #TRANSFORMATION concepts, there are two participants: (i) a Theme Entity that transforms another entity and (ii) a Referent Entity that is transformed by another entity. However, the selection restrictions for −PASTEURIZE specify that the theme is human and the referent is restricted to milk.

In Figure 3.4, the meaning postulate of pasteurize is composed of two events, the second of which is the purpose of the other. In the second event (e2), the theme (x1) kills a referent (x3), in this case, bacteria. Since −PASTEURIZE is a type of $STERILIZE_00, which entails a cleaning event as well as a killing event, there is a semantic decomposition of this predicate into more basic units. In fact, it is possible to reach a semantic representation in COREL composed of semantic primitives by applying stepwise conceptual decomposition (Periñán-Pascual and Arcas-Túnez Reference Periñán Pascual and Arcas-Túnez2005), where the defining term in a meaning postulate can become the defined term in another. This process is based upon Dik’s (Reference Dik1978) stepwise lexical decomposition as applied in Faber and Mairal Usón (Reference Faber and Mairal Usón1999). However, instead of lexical items, the results of this process are regarded as conceptual units.

This type of conceptual representation can also be used to represent qualities designated by adjectives such as obsolete. The ontological hierarchy in Figure 3.5 shows that obsolete is linked to the meta-concept of #QUALITY, and that it exists as a social perception #SOCIAL. An entity can only be perceived as +USELESS_00 (→obsolete)) within a social context in which something better has been created to replace it.

Figure 3.5 FunGramKB Ontological Hierarchy for concepts linked to $OBSOLETE_00

In the same way as with −PASTEURIZE, words from different languages can also be linked to $OBSOLETE_00 (e.g. Engl. obsolete, Sp. obsoleto, Fin. vanhentunut, Hung. elavult). The CLS for these predicates has the following format:

This CLS for obsolete consists of the conceptual unit $OBSOLETE_00, which, in turn, is composed of a thematic frame and a meaning postulate (Figure 3.6). Since it is a state, the thematic frame only has one participant (x1), which is a previously useful entity that is now socially perceived as useless.

Figure 3.6 FunGramKB conceptual entry for $OBSOLETE_00

Obsolete is a predicate that can be applied to a wide variety of objects, such as typewriter. As is well known, a typewriter is a machine with keys that are pressed to print letters of the alphabet onto paper. Since the 1990s, personal computers more efficiently perform the functions of typewriters. Thus, understanding why typewriters have become obsolete entails elements of world knowledge, such as an awareness of past and present time frames as well as of the affordances of typewriters in comparison to those of more modern machines that can be used for the same purpose. The meaning postulate in Figure 3.6 describes the three events that occur when an entity, such as a typewriter, is described as obsolete:

As reflected in (22–24), $OBSOLETE_00 is a more specific type of +USELESS_00 since it presupposes a time sequence or period during which the (once useful) entity has become useless. The representation in (25) portrays +USELESS_00 as the negation of +USEFUL_00.

The conceptual entry for +USEFUL_00 in (26) states that something is useful when it helps a human to do something.

All of this information is inherited by $OBSOLETE_00. When specifically applied to a typewriter, what is also activated is the list of affordances provided, which other more modern machines now perform more efficiently.

Finally, nouns, such as invoice, are also represented in terms of a conceptual logical structure. From an ontological perspective, +INVOICE_00 is a basic concept (see Figure 3.7). It is a member of the meta-conceptual category of #PHYSICAL_OBJECT and more specifically, that of #SELF_CONNECTED_OBJECT. Its conceptual path is +INFORMATION_OBJECT_00→ +WRITING_00→ +DOCUMENT_00. An invoice is a type of information object, which is informative because of the writing in/on it, and which has the form of a document.

Figure 3.7 FunGramKB Ontological Hierarchy for +INVOICE_00

Again, lexemes in different languages, such as invoice (English), factura (Spanish), faktura (Norwegian), فاتورة (Arabic), can be linked to this conceptual entry, which has the following CLS:

As a noun, invoice does not have a thematic frame. However, this does not mean that it lacks a relational structure. As a commercial document that itemizes a transaction between a buyer and a seller, invoice indirectly refers to a complex knowledge configuration, which speakers have to access in order to understand the term. This information should be included in the meaning postulate of the conceptual entry. Accordingly, the conceptual entry for +INVOICE_00 in Figure 3.8 is composed of a meaning postulate with two main events, which state what type of entity an invoice is (e1) and the context in which it is produced (e2).

Figure 3.8 FunGramKB conceptual entry for +INVOICE_00

This second event contains three subevents (e3, e4, e5). Examples (28–31) show each event and subevent within the corresponding frame, which encodes the purpose of an invoice (f1) and the possible reasons for issuing it (f2, f3).

As reflected in (28–31), an invoice activates knowledge that is linked to a given sociocultural context. This shows that language does not exist in a void but is closely related to the culture and society in which it is used. Evidently, invoices would have little or no meaning for a tribe in the Amazon rainforest. In contrast, invoices are crucial in a sociocultural environment that contemplates commercial transactions involving the emission of documents that notify a receiver that payment is due for goods or services of some type. When this kind of encyclopedic knowledge is included in a conceptual entry, it provides users with the information necessary to understand the concept.

As can be observed, within the context of an underlying ontology, a relatively simple representation, such as +INVOICE_00 (x), do′ (x [ −PASTEURIZE (x, y)]), or $OBSOLETE_00 (x), allows users to access a rich variety of encyclopedic information regarding objects, processes and attributes. This is possible because of the thematic frames, meaning postulates and inference mechanisms stored in the knowledge base.

In sum, the only difference from current LSs is that CLSs are made of conceptual units and not lexical words. CLSs maximize informativeness and minimize redundancy without losing the simplicity and elegance of the formalism in LSs. A lexical entry is now defined within the framework of a CLS, which is able to connect to a huge knowledge base from where it is feasible to retrieve information via a reasoning engine, which in turn uses two major functions: inheritance and inference (see Periñán-Pascual and Arcas-Túnez Reference Periñán Pascual and Arcas-Túnez2010). By using artificial intelligence techniques, an RRG lexical representation in terms of a CLS can be successively augmented by information coming from different modules of the knowledge base (i.e. the ontology, the cognicon and the onomasticon).¹⁰ Finally, a CLS is further enriched by the inclusion of the operators that provide a fully-fledged semantic representation of an input sentence (Van Valin and Mairal Usón Reference Van Valin, Mairal Usón, Gómez, de Mendoza and Gonzálvez-García2014).

4.4.1 Causativity

When Foley and Van Valin (Reference Foley and Van Valin1984) first proposed the AUH, they presupposed – in line with Dowty (Reference Dowty1979) – a logical link between causative constructions and accomplishments. In addition, they considered the argument of αCAUSE to be an argument of DO. As the argument of αCAUSE is necessarily the most active argument in the hierarchy, no problem arose with macrorole assignment.

However, later research in RRG (Van Valin and LaPolla Reference Van Valin and LaPolla1997: 97, 106–109; Van Valin Reference Van Valin, Kailuweit, Künkel and Staudinger2018) has shown that equating causative constructions with accomplishments was erroneous. On the one hand, there are accomplishments that are not causative, like the one in (14a), which contrasts with its causative counterpart in (14b).

1. (14)

   a.	The ice melted (accomplishment)
   b.	Hot water melted the ice (causative accomplishment)

On the other hand, non-causative states, activities and achievements can also have corresponding causative constructions.

1. (15)

   a.	The boy is afraid (state)
   b.	The ball bounced around the room (activity)
   c.	The balloon popped (achievement)

1. (16)

   a.	Bill’s owning a gun frightened Martha (causative state)
   b.	The girl bounced the ball around the room (causative activity)
   c.	The cat popped the balloon (causative achievement)

The representation of causativity did not pose formal macrorole assignment and linkage problems for Foley and Van Valin (Reference Foley and Van Valin1984). Perhaps for this reason, the interaction of the AUH and causative constructions has not been discussed in detail in later versions of RRG. Some putatively problematic cases for macrorole assignment that arise from the isolated application of the AUH for the individual predicates in causative constructions can easily be ruled out if macrorole assignment begins with the actor and simply occurs in the causal chain from left to right. These cases are discussed below.

We will start with causative activities. Van Valin and LaPolla (Reference Van Valin and LaPolla1997: 107) propose the LS [do′ (girl, Ø)] CAUSE [do′ (ball, [bounce′ (ball)])] for (16b). Both components of this LS consist of activities. Hence, the two arguments (girl and ball) occupy the same position in the hierarchy, but the processing rule assures that the actor is assigned to girl. Being the rightmost argument, ball in the CAUSEβ part takes the undergoer macrorole, although it is an argument of do′ (x,…). Obviously, this a coherent interpretation from the semantic point of view, since the CAUSEβ component is the effect of the causing event.

This is also true for activities embedded into causative chains (Van Valin and LaPolla Reference Van Valin and LaPolla1997: 121). The AUH explains why knife could eventually become actor in an instrument-subject construction (cf. 17b).

1. (17)

   a.	Tom is cutting the bread with a knife
   	[do′ (Tom, [use′ (Tom, knife)])] CAUSE [[do′ (knife, [cut′ (knife, bread)])] CAUSE [BECOME cut′ (bread)]]
   b.	The knife cuts the bread
   	[do′ (knife, [cut′ (knife, bread)])]

A proposal to make the possible undergoer assignment to arguments of do′ (x,…) more explicit would be to mark this position as accessible for undergoer choice in the AUH (Kailuweit Reference Kailuweit2012: 110) (see Table 4.3).

In (16a), an example of a causative state, macrorole assignment does not follow straightforwardly from the AUH. Van Valin and LaPolla (Reference Van Valin and LaPolla1997: 107) propose the LS [have′ (Bill, gun)] CAUSE [feel′ (Martha, [afraid′])] for (16a). Both compounds of CAUSE consist of a two-place state predicate (have′(x ,y) and feel′ (x, y)). For the β part the y-argument is a predicate [afraid′]. The two x-arguments of these predicates (Bill and Martha) are both instances of the same middle position of the AUH. If we follow the rule of processing from left to right, Bill will be the actor. However, this is inaccurate from a semantic point of view. It is not Bill, but his possession of a gun that is responsible for causing Martha’s fear. Hence, the actor of frighten is not Bill, but the whole αCAUSE compound. It may seem strange to think of a state of affairs as an actor, but otherwise frighten would cease to be a macrorole transitive predicate in (16a).

Figure 4.1 Actor-Undergoer Hierarchy (modified)

4.4.2 Three-Place Predicates

According to Van Valin (Reference Van Valin2007), almost all three-place predicates are causative constructions. We have noted that these predicates give rise to problems of variable undergoer choice. The overall LS of these predicates would be [do′ (x, Ø)] CAUSE [BECOME pred′ (y, z)]. The actor is represented in the αCAUSE part and the CAUSEβ part hosts the remaining two arguments that are accessible for undergoer assignment. As we have already seen, the conditions of undergoer assignment vary across languages and even across individual predicates. Nonetheless, there are a few (but highly frequent) three-place predicates that do not seem to be causative. In the remainder of this section, I will deal with talk (to somebody about something)¹⁹ as a candidate for a possible non-causative three-place activity and with envy (somebody (for) something) as a candidate for a possible a non-causative three-place state.

In Van Valin and LaPolla (Reference Van Valin and LaPolla1997: 116–118) we find a detailed description of English verbs of saying, including verbs of talking. The authors propose a general LS for non-causative verbs and an LS for causative verbs of telling.

Talk is listed as a verb that selects an α component and a β component (talk to Pat about Sandy) or a γ component (talk Cajun). Since it is the absence of causativity that distinguishes verbs of talking from verbs of telling, the causative LS in (18) cannot describe the three-place use of talk. In a later chapter concerned with coreference, Van Valin and LaPolla (Reference Van Valin and LaPolla1997: 401) propose an LS for talk, when they discuss the coreference of the reflexive in Bob talked to Susan about herself.

The binding facts seem to be the reason why the β component appears as the y-argument of the two-place state and the α component as the argument of the preposition co-referring with the y-argument. In a footnote Van Valin and LaPolla (Reference Van Valin and LaPolla1997: 668) remark that about herself is an argument PP and not an argument-adjunct, because the PP is ‘the optional realization of the α variable’. The fact that the preposition seems to be semantically empty and commutable (talk about, talk of as well as German sprechen über, sprechen von) corroborates this view. However, it is precisely for this reason that it seems problematic to use an argument-adjunct construction to represent the third argument of talk.²⁰

Staudinger et al. (Reference Kailuweit, Wiemer, Staudinger and Matasović2008) proposed an alternative analysis. RRG allows for several classes of two-place activities: do′ (x, [sing′ (x,(y))]); do′ (x, [see′ (x,(y))]); do′ (x, [tap′ (x,(y))]), etc. (Van Valin and LaPolla Reference Van Valin and LaPolla1997: 115).²¹ The second argument represents an instance of pred′ (x, y) (ibid.: 127), but the predicate embedded in activities is not a state. Hence, RRG could theoretically allow for three-place activities with an argument in each of the three medium positions of the AUH: do′ (x), pred′ (x, …), and pred′ (…, y). Staudinger et al. (Reference Kailuweit, Wiemer, Staudinger and Matasović2008) thus suggested the following LS for French parler, the counterpart of English talk.

Note that according to the formalism used in RRG, the predicate talk′ (x,(y),(z)) is not a state, similarly to sing′ (x, (y)) in do′ (x, [sing′ (x, (y))]). As indicated before, activities do not consist of states embedded under a do′ operator. Hence, in an RRG description of talk, one could associate the two arguments y and z with the positions pred′ (x, …) and pred′ (…, y) of the AUH, as RRG associates the y-argument of sing′ with the pred′ (…, y) position.

However, this solution would not be available for three-place non-causative states. Verbs of envy are possible candidates for this class (Kailuweit Reference Kailuweit2012: 118). In line with two-place stative verbs of emotion such as love or hate, envy takes an emoter argument, the person who experiences envy, and a target argument, the envied person. In addition, envy possesses a third argument expressing the object or circumstance the target argument is envied for.²² For the English verb envy, this third argument could be realized as a direct core argument (21a) or as an oblique (21b).

1. (21)

   a.	Many people envied the shareholders their large profits.
   b.	Max envied Paul for his wealth.
   c.	Max envied him.

The fact that the third argument, the object of envy, can also be marked by for (21b) may suggests that it is an adjunct, which gets incorporated into the core as a direct argument, by analogy with recipient beneficiaries, such as Max baked a cake for Paul vs. Max baked Paul a cake. Thus, a possible LS for (21b) could be because.of′ ([have′ (Paul, wealth)], [envy′ (Max, Paul)]). However, in (21a) and (21b) the target, the envied person, is the undergoer, while the for-complement becomes the undergoer in the recipient-beneficiary construction Max baked a cake for Paul → Max baked Paul a cake.

French envier (‘envy’) shows variable undergoer choice. The target can be realized either as an undergoer (22a) or as non-macrorole direct core argument in the dative case (22b). As in English (21c), the object of envy argument is optional if the target argument is realized as undergoer.

1. (22)

   a.	Beaucoup	d’hommes		vous	envient	pour	Madame	Linné
   	Many	of men		you.2pl.acc	envy.3pl	for	Mrs	Linné
   	‘Many men envy you for Mrs Linné’
   	(FRANTEXT: HERMARY-VIEILLE, C.)
   b.	…cette	belle	figure	creuse […]	il	la	lui	enviait
   	this	pretty	face	haggard	he	3f.sg.acc	3m.sg.dat	envied
   	‘It was this pretty haggard face that he envied him for.’
   	(FRANTEXT: ETCHERELLI)

The same case variation as in French can be found in German. In addition, the contrast between the two constructions is realized with the help of the applicative construction with the prefix be-. German neiden (‘envy’) realizes the target argument as a non-macrorole direct core argument in the dative case and the object of envy as the undergoer. German beneiden realizes the target as the undergoer and the object of envy as a prepositional object: jemanden um etwas beneiden (‘to envy somebody for something’). Hence, the object of envy seems to be the unmarked choice for undergoer in German.

Admitting that envy is a three-place state would raise the problem of assigning an LS that provides information about macrorole assignment. One might consider envy′ (x, y, (z)) as a possible LS for envy. However, the three argument positions in this LS cannot correspond to the three stative argument positions in the AUH. As far as the three positions referring to states are concerned – first argument of pred′ (x, y), second argument of pred′ (x, y) and argument pred′ (x) – they represent two basically different types of state. On the one hand, there are two-place states (pred′ (x, y)), for example verbs of emotion (love′ (x, y)) or possession (have′ (x, y)), with a less affected second argument (a theme in Gruber’s terms). On the other hand, there are one-place states (pred′ (x)) with an affected prototypical undergoer argument, for example resultative verbs of dying (BECOME dead′ (x)) or destruction (BECOME broken′ (x)). Hence, the structures pred′ (x) and pred′ (x, y) logically exclude each other and cannot be added to each other.

Note that verbs of envy are not the only candidates for non-causative three-place states. A more language-specific case is the French verb en vouloir à quelqu’un de quelque chose ‘to be angry with someone about something’.²³ The argument denoting the person who experiences anger is realized as the subject, and hence it takes a macrorole. In contrast to verbs of envy, en vouloir is M-intransitive. Therefore, the two other arguments, the person the anger is directed at, and the circumstance that the anger is related to, are realized as a direct core argument in the dative case and as an oblique object respectively. In comparison with verbs of envy, en vouloir raises an additional problem: the nature of the macrorole. RRG describes non-episodic verbs of anger as instances of verbs of internal experience with the LS feel′ (x, y). The x-argument corresponds to the experiencer, and the y-argument is a predicate of sensation having one internal argument that is not accessible for macrorole assignment. Therefore, the predicates of internal sensation are by default M-intransitive. As intransitive states, they select an undergoer. The undergoer macrorole is assigned to the x-argument, the experiencer.

A possible LS for three-place en vouloir would be feel′ (x, angry.at′(y).about′(z)) with two internal arguments of the sensation predicate. However, the feel′ (experiencer, sensation) analysis is neither convincing for the French three-place en vouloir nor for its two-place German or Spanish counterparts zürnen and rabiar ‘to be angry at’. While German zürnen realizes the argument denoting the person the anger is directed at as a dative, Spanish rabiar expresses this with a preposition (contra, lit. ‘against’). Independently of these syntactic differences, this argument is obligatory for French en vouloir, German zürnen and Spanish rabiar. To consider an obligatory argument inaccessible for macrorole assignment seems to be problematic. In addition, if the argument denoting the person the anger is directed at were embedded into the sensation predicate, ‘being angry at James’ and ‘being angry at Paul’ would be two different sensations. Kailuweit (Reference Kailuweit2005b, Reference Kailuweit, Nolan and Diedrichsen2013) proposes that French en vouloir, German zürnen and Spanish rabiar are atypical activities rather than states. They denote an emotional behaviour motivated by anger and directed at a person held responsible for the annoying situation.²⁴ Therefore, the macrorole is an actor and not an undergoer. An LS that parallels the one proposed for verbs of saying (Staudinger et al. Reference Kailuweit, Wiemer, Staudinger and Matasović2008) would be able to deal with macrorole assignment and linking.

5.3.1 Privileged Syntactic Argument: Controller and Pivot

As the phenomenon we are talking about is construction-specific, and there are many different types of restricted neutralizations, the term ‘subject’ is not appropriate, and instead we use the term ‘privileged syntactic argument’ (PSA) for an argument that is the controller or pivot of a restricted neutralization of semantic roles for grammatical purposes, generally referent tracking (i.e. keeping track of who is doing what in discourse). No other grammatical relations are recognized in this approach; the characteristics that have traditionally been associated with ‘direct object’, such as taking accusative case, being able to appear as the privileged syntactic argument in a passive clause or being the target of applicative constructions, are seen as properties of the macrorole Undergoer, while those associated with ‘indirect object’ are seen as properties of the non-macrorole direct core argument (see the discussion of (5) below).

The terms ‘controller’ and ‘pivot’ (first used in Heath Reference Heath1975) refer to different types of PSA, as in the examples in (1), the English conjoined clause coreference construction, where the immediately preverbal (core-initial) referring expression in the first clause is the controller, while the implied argument of the second clause is the pivot of the construction (marked by ‘(pivot)’, where the implied argument would be if it appeared in an independent clause), regardless of whether either clause is an active or passive construction.

1. (1)

   a.	Bobcontroller handed Jim the money and (pivot) left. [A,S]
   b.	Jimcontroller was handed the money by Bob and (pivot) left. [S,S]
   c.	The moneycontroller was handed to Jim but (pivot) not seen again after that. [S,S]
   d.	Jimcontroller took the money and (pivot) was seen later buying a new car. [A,S]
   e.	Jimcontroller took the money and (pivot) thanked Bob for it. [A,A]
   f.	Jimcontroller smiled and (pivot) took the money. [S,A]

In this construction, there is obligatory coreference between the controller, whether it is the Actor (A) of an active transitive clause or single direct argument of an intransitive clause (S)⁴ or the Undergoer of a passive clause (S), and the pivot, again whether it is A or S. The construction then aids in the inference of who is doing what. The restricted neutralization we find in this particular construction is [A,S], but as shown in the examples, the correspondence can be [A,A] or [S,S] as well, as what is important is that the arguments participating in the coreference are grammatically either A or S; the Undergoer of an active transitive clause does not participate in this coreference. The Undergoer argument cannot simply be left to implication when using this structure, and so a different construction must be used to allow the Undergoer to participate in the coreference of the conjoined clause coreference construction. The construction used is the English passive construction, as in (1b, c ,d), as it is an intransitive clause in which the Undergoer is the PSA. Using the passive construction in this conjoined clause coreference construction allows the Undergoer argument (regardless of what specific semantic role it has) to participate in the coreference construction by casting it as an S.

The English passive construction allows limited variable access to the syntactic controller and syntactic pivot positions. Without that construction the Undergoer would not be able to appear as the PSA. Some languages manifest constructions with a particular restricted neutralization but do not have constructions that allow variable access to the PSA (i.e. they do not have alternative voice constructions such as passive; e.g. Enga (Papua New Guinea), Warlpiri (Australia), Lakhota (North America); see Van Valin and LaPolla Reference Van Valin and LaPolla1997, pp. 274–285 for details). These constructions are said to have an invariable syntactic controller and invariable syntactic pivot. Some languages have one or a few constructions (e.g. the English passive construction or the Dyirbal antipassive construction – see the next section) that allow limited variable access to the PSA. These constructions are said to have a variable syntactic controller and a variable syntactic pivot. And some languages, such as Tagalog (see Section 5.3.2), have many constructions allowing a range of semantic roles access to the PSA. These constructions have a semantic controller with no neutralization. As we will see, in Tagalog there is no neutralization we can call S, nor even neutralization of a single grammatical category of Actor or Undergoer in terms of the marking on the predicate, as different types of actor and undergoer (e.g. with different degrees of intention, agentivity, transitivity, affectedness) are marked differently on the predicate depending on the nature of the action and the Topic of the clause.⁵ In other constructions (e.g. the Reflexive Construction) there is a generalized Actor, and it is an invariable semantic controller. For ease of discussion we will generalize across the different types of actor, patient and location in Tagalog by glossing the various relevant forms as ‘Actor-Topic’, ‘Patient-Topic’ or ‘Location-Topic’, respectively.

5.3.2 Alignment

The particular PSAs in the conjoined clause coreference construction discussed here are found in English, but many other languages, even closely related ones, do not manifest PSAs in conjoined clause coreference constructions, and so the determination of the relevant argument of the second clause would be left to inference unconstrained by the grammar, and undergoers in active clauses could be left to inference as easily as A and S. It is also the case that some languages manifest constructions for conjoined clause coreference with particular PSAs, but the restricted neutralization is [S,U] rather than [S,A].⁶ Dyirbal, a Pama-Nyungan language spoken in north-eastern Australia (Dixon Reference Dixon1972, 1980), is a well-known example of a language that has grammaticalized an [S,U] restricted neutralization for conjoined clause coreference. That is, the arguments participating in the coreference must be U and/or S, but A does not participate in the coreference unless it is cast as an S in the Dyirbal antipassive construction. Consider the examples in (2) (adapted from Dixon Reference Dixon1980: 462).

1. (2)

   a.	balan	guda	buŋa-n	baŋgul	yara-ŋgu	bura-n
   	3sgf.abs	dog.abs	descend-pst	3sgm.erg	man-erg	see-pst
   	The dogcontroller went downhill and the man saw (pivot).
   b.	bayi	yara	buŋa-n	buralŋanyu	bagun	guda-gu
   	3sgm.abs	man.abs	descend-pst	see:pst:antip	3sgm.abs	dog-dat
   	The mancontroller went downhill and (pivot) saw the dog.
   c.	balan	guda	baŋgul	yara-ŋgu	bura-n	buŋa-n.
   	3sgf.abs	dog.abs	3sgm.erg	man-erg	see-pst	descend-pst
   	The man saw the dogcontroller and (pivot) went downhill.

These examples parallel the English examples in (1), but the interpretation of the implied argument in the second clause of (2a) is obligatorily coreferential with the Undergoer in the first clause, that is, it must be that the dog is the controller and pivot (i.e. is the one that went downhill). In order to have coreference that involves an A argument, the Dyirbal antipassive construction must be used. This construction is an intransitive construction with the verb marked with the antipassive marker ŋa, and has the Actor as the single direct argument in the absolutive case and the Undergoer in the dative case. Because Dyirbal has this antipassive construction, there is variable access to the controller and pivot positions, and so the controller is a variable syntactic controller and the PSA is a variable syntactic PSA, though they manifest a different restricted neutralization from the corresponding English construction. These grammaticalized constraints on interpretation we have been looking at force a particular interpretation of an utterance in both English and Dyirbal, but as the restricted neutralizations are different, the interpretations are different. For example, if The man saw the dog and went downhill is said in English, the interpretation has to be that the man went downhill; but if the corresponding structure is used in Dyirbal, as in (2c), the meaning has to be that the dog went downhill. We see here that the construction must be taken as a whole, as it is the total construction that influences the interpretation, and is not simply the sum of the individual words.

Where there is a choice of argument for PSA, the RRG theory of PSA selection posits a default choice, depending on the Privileged Syntactic Argument Hierarchy (3) and the privileged syntactic argument selection principles (4): actor for PSA in so-called nominative-accusative alignment, and Undergoer in ergative-absolutive alignment.

Yet it isn’t the case that all languages necessarily have such a default choice. As Foley and Van Valin argued (Reference Foley and Van Valin1984, §4.3) Tagalog has many different constructions for having different semantic roles as PSA, yet none is a default choice. In all these languages where we have seen a choice of PSA, the choice of pivot is determined by pragmatic factors, such as the identifiability or topicality of the referent involved (the default is used when there is no difference in terms of the pragmatic factors), and so we refer to these as pragmatic PSAs, as opposed to those based strictly on semantic factors, which we call semantic PSAs.

While the conjoined cross-clause coreference construction in English manifests an [S,A] restricted neutralization, it is not the case that all constructions in English manifest a restricted neutralization, and it is not the case that all of the constructions that do manifest a PSA in English manifest the same restricted neutralization. For example, in the following example of the English purposive construction the controller and pivot of the construction are the Undergoers of the two clauses, not the Actor or S of the clauses:

As in the construction in (1), the controller determines the reference of the implied argument in the second clause, but in this case the referent referred to appears as the traditional direct object (the Undergoer in RRG) of the first clause, and is understood as the traditional direct object (the Undergoer in RRG) of the second clause as well. This shows that the concept of PSA is not the same as the traditional concept of ‘subject’.

Aside from the possibility of different restricted neutralizations, there is also the possibility of unrestricted neutralization. Some languages, such as Riau Indonesian (Gil Reference Gil1994) and Mandarin Chinese (LaPolla 1993), have not conventionalized any constraints on referent role identification of the type associated with grammatical relations in any constructions in the language (though they may have conventionalized other types of constraints on interpretation). This means that in all the constructions we have looked at interpretation is based entirely on inference from context and is not constrained by the structure. In other languages, only some constructions will have unrestricted neutralization. It is entirely language-specific and construction-specific. In English, for example, the relative clause construction manifests neutralization, but no restriction, that is, any argument of the modifying clause can appear as the head of this construction, as shown in (6):

1. (6)

   a.	the girl who[A] sang the song
   b.	the girl who[U] the police saved
   c.	the girl who[S] just came in
   d.	the girl to whom[non-macrorole core argument] the award was given
   e.	the car in which[peripheral argument: location] the man was held up
   f.	the car out of which[peripheral argument: source] the radio was stolen

In Tagalog, a Malayo-Polynesian language spoken in the Philippines, on the other hand, there is a restricted neutralization for the Tagalog relative clause construction: the head of the construction must be the grammatical Topic of the modifying clause, and so it is a syntactic pivot for that construction. To explain this we will first give some background on the structure of the Tagalog main clause.

Tagalog is a consistently focus-initial language and so in general the predicate appears in initial position and the Topic appears (when it is not a pronoun and not in focus) at the end of the clause, and the information structure is Comment-Topic rather than Topic-Comment. The predicate in most cases marks aspect, realis/irrealis, and the semantic role of the Topic of the clause, and so the Topic is the controller of the semantic role marking (it is a semantic and not syntactic controller because the marking differs with each semantic role, i.e. there is semantic restriction but no neutralization relative to each type of marking). Topic here is actually a grammatical status, as it is an argument singled out for special morphological treatment, as well as a pragmatic status, as it is what the clause is about. Generally, almost any referent associated with the situation in some way, whether core or peripheral argument semantically or even a very indirectly affected referent, can be the Topic of the clause, though usually (but not always) it is one that is identifiable to the hearer, and in many cases the predicate takes a form to reflect the semantic role of that referent, and the marking of the other references in the clause is usually also different.⁷ These different constructions are not passive or antipassive, but simply different ways of profiling the event (Foley and Van Valin Reference Foley and Van Valin1984, §4.3), similar to the choice of the A Construction vs. the O Construction in Jawarawa (Dixon Reference Dixon2000, Reference Dixon2004), depending on what is considered to be what the clause or discourse is about, but with more choices for Topic in Tagalog.

The representation of the Topic argument (if it appears in the clause and is not a pronoun) generally takes either a marker of specificity (ang or si with proper names of single referents) or a topic form of demonstrative pronoun (most commonly ’yung) to mark it as the Topic. The set of pronouns, which appear as second-position clitics, also distinguishes Topical from non-Topical referents, with the latter appearing as possessive forms or dative forms. In (7) there is a short natural conversation to exemplify this feature (from my own fieldwork; see also Schachter (Reference Schachter, Reid and Comrie2008: 337–338) for sets of constructed parallel examples with the same arguments but with different choices of topic, and Schachter and Otanes (Reference Schachter and Otanes1972, Ch. 5) for many of the major constructions used for marking different semantic roles).⁸

1. (7)

   1.	Jirehel: Madali lang kasi’ng gumawa ng salsa eh.
   	ma-dali	lang	kasi	ang	g<um>awa	ng	salsa	eh
   	stat-easy	just	because	spec	<at>make	poss	sauce	emphatic
   	‘Making sauce is easy.’
   2.	Wendy: Oo, madali lang.
   	oo	ma-dali	lang
   	yes	stat-easy	just
   	‘Yes, it’s easy.
   3.	Gawin mo lang ketsap,
   	gawa-in	mo	lang	ketsap
   	do-pt:irr	2sg.poss	just	ketchup
   	You just make it with ketchup,
   4.	tapos lagyan mo ng tomatoes,
   	tapos		lagay-an	mo	ng		tomatoes
   	finish		put-lt	2sg.poss	poss		tomatoes
   	then you add tomatoes,
   5.	lagyan mo ng salt and pepper to taste, tapos na.
   	lagay-an	mo	ng	salt and pepper to taste			tapos	na
   	put-lt	2sg.poss	poss	salt and pepper to taste			finish	cs
   	add salt and pepper to taste; then, it’s done.’

In (7), the first speaker, Jirehel, refers to the making of sauce using a form (gumawa) where the infix ‑um- marks it as Actor-Topic (it happens that in this utterance the speaker has made the relevant clause the Topic of a higher clause, but the phenomenon is the same), then the second speaker, Wendy, uses the same root in line 3 of the example, but in the irrealis Patient-Topic construction, to profile the event from the point of view with the sauce as the Topic. She then follows this in lines 4 and 5 with two tokens of the root lagay ‘put’ in the Location-Topic construction, to keep the sauce as the Topic, but with the sauce now having the semantic role of the location where the tomatoes and salt and pepper are to be added. Notice how there is no overt reference to the Topic in any of Wendy’s utterances (e.g .’yung salsa ‘the sauce’ could have been added to the end of each of Wendy’s utterances in lines 3–5, but it wasn’t), yet we can tell what is being talked about (what is the pragmatic and grammatical Topic) because of the marking on the predicate.

We now can return to the issue of grammatical relations, that is, restricted neutralizations. The choice of different roles as Tagalog Topic affects the particular form of the construction and the behaviour of the Topic in this and other constructions, such as the Tagalog relative clause construction, where the head of the construction generally must be the Topic of the modifying predicate (i.e. the form of the predicate must be the same as if it were a main clause with that referent as Topic). For example, if we were to recast the clauses in (7) as relative clause constructions (and with realis predicates), we would get the forms in (8):

1. (8)

   a.	lalaking gumawa ng salsa		‘man who made salsa’
   	lalake	=ng	g<um>awa	ng	salsa
   	man	lnk	<at.realis>make	poss	sauce
   b.	salsang ginawa ng ketsap		‘salsa made with ketchup’
   	salsa	=ng	g<in>awa	ng	ketsap
   	sauce	lnk	<pt.realis>	poss	ketchup
   c.	salsang nilagyan ng tomatoes		‘salsa in which tomatoes were put’
   	salsa	=ng	in-lagay-an	ng	tomatoes
   	sauce	lnk	realis-put-lt	poss	tomatoes

In Tagalog, then, for this construction there is a clear restricted neutralization: the head of the construction must be the grammatical Topic of the predicate that modifies it, unlike in English, where there is neutralization but no restriction on what argument can be the head of an English relative clause construction.

We saw that in Tagalog the controller of the semantic role marking on the predicate (the Topic) is a semantic controller. English agreement, on the other hand, manifests a syntactic controller. That is, the agreement is not with a particular semantic role like in Tagalog, but manifests a restricted neutralization, where the agreement is with the Actor or S of the clause, regardless of its semantic or pragmatic role. We can see this from the examples in (9).

1. (9)

   a.	I am helping Bill with his homework	[agreement with A]
   b.	Bill is being held up by the activities	[agreement with Undergoer as passive S]
   c.	Bill is smiling	[agreement with S Actor]
   d.	Bill is falling into a trap	[agreement with S Undergoer]
   e.	There are many people in the park	[agreement with non-topical S]
   f.	There are people grilling meat there	[agreement with non-topical A]

Acehnese (Durie Reference Durie1985, Reference Durie1987), an Austronesian language spoken in Indonesia, manifests a system of pronominal agreement on the verb with Actor-marking prefixes and Undergoer-marking suffixes on the verb. The marking is restricted to Actor and Undergoer, respectively, and there is no neutralization of Actor and Undergoer, even with intransitive constructions. See, for example, the following (from Durie Reference Durie1987: 366, 369):

1. (10)

   a.	(Gopnyan)	geu-mat	lôn	b.	(Lôn)	lôn-mat-geuh
   	(3sg)	3-hold	1sg		(1sg)	1sg -hold-3
   	‘(S)he holds me.’				‘I hold him/her.’
   c.	Geu-jak	(gopnyan)		d.	Lôn	rhët(-lôn).
   	3-go	(3sg)			1sg	fall(-1sg)
   	‘(S)he goes.’				‘I fall.’

From these examples we can see that the prefixes are used for the Actors of transitive constructions (10a, b) and the single arguments of intransitive constructions where the action is voluntary and so the argument is an Actor (so-called S_A), as in (10c), and the suffixes are used for the Undergoers of transitive constructions and the single argument of intransitive constructions where the action is involuntary (an Undergoer, so-called S_P or S_O), as in (10d).⁹

5.3.3 Referent Tracking

The different choices of grammatical Topic we saw in Tagalog have the same function in discourse as the variable syntactic pivots in English and Dyirbal in terms of allowing for topic chains where an (often unmentioned) referent can remain the topic across clauses even if its semantic role changes, and the structure helps constrain the inference of the identity and role of the relevant referent, as in (7). This type of referent-tracking mechanism is known as a ‘switch function’ (Foley and Van Valin Reference Foley and Van Valin1984: 321–374; Van Valin Reference Van Valin, Verschueren and Bertuccelli-Papi1987; Van Valin and LaPolla Reference Van Valin and LaPolla1997, §6.4), as the structure identifies the difference in function of the referent while maintaining it as topic. There are differences among languages, though, not only in the type of restricted neutralization, such as in English vs. Dyirbal, but also in terms of what referents can be an unmarked topic of the clause: if we contrast English, Dyirbal, Tagalog and Chinese, we can see that English and Dyirbal severely limit what can be topic in the unmarked clause structure ([S,A] or [S,U] respectively); Tagalog allows just about anything relevant to be topic, but marks it morphosyntactically, thereby constraining the inference of the referent and its role; Chinese does not manifest any restricted neutralizations (LaPolla Reference LaPolla1993) and so there is no grammatical restriction on what can be topic in the unmarked clause structure, though there is no marking of it as topic other than initial word order, and no marking of its role or identity, and so identification of the referent and its role is not aided by the structure (see, for example, the discussion of (12) below). This forms a separate though related typological cline, from most restricted to least restricted in terms of access to topic, aside from the typological cline related to the degree to which the structure constrains the interpretation of the role and identity of the referent being tracked.

Yet another type of restricted neutralization with variable PSAs, known as the ‘switch-reference’ pattern, is found in Barai, a language of Papua New Guinea (Olsen 1978, 1981; Van Valin Reference Van Valin2009, §4), and in Choctaw, a Muskogean language of North America (Heath Reference Heath1975, Reference Heath1977), among others. In this type of system, when clauses are coordinated, the PSAs of the individual clauses (defined differently in each construction and in each language) can be coreferential or not, and there is marking on the predicate to show this (i.e. whether the PSA is the same as in the following clause or different from the following clause), as in the following Choctaw examples (adapted from Heath Reference Heath1977: 212):

1. (11)

   a.	(0i)-0-pi:sa-ča:,	0i-iya-h	‘Hei sees himj and hei goes’
   	(3A)-3P-see-same	3A-go-present
   b.	0-0i-pi:sa-na:,	0i-iya-h	‘Hei sees himj and hej/k goes’
   	3A-3P-see-different	3A-go-present

As can be seen from these examples, even without overt arguments, the affix on the predicate marking whether the PSA in the following clause is the same or different from that of the marked clause constrains the interpretation of who is doing what. The pivots in this sort of system are generally invariable syntactic pivots.

We have seen above that there are different kinds of constructions languages can have for constraining the inference of who is doing what, if they have any at all. They may have some constructions that have invariable semantic pivots (i.e. restriction with no neutralization), such as the verbal marking in Tagalog and Acehnese, or invariable syntactic pivots, such as reference across conjoined clauses in Warlpiri and Enga, or variable syntactic pivots, such as reference across clauses in Dyirbal and English, and the relative clause construction in Tagalog. The restricted neutralization found in a construction could treat [S,A] the same in opposition to [U], or it may treat [S,U] the same in opposition to [A]. As this is a construction-specific phenomenon, the same language may have different constructions with different PSAs, as we saw with Tagalog (see also Van Valin 1981 on Jakaltek and Van Valin and LaPolla Reference Van Valin and LaPolla1997: 282ff. on Tzutujil), and constructions with different PSAs can sometimes be combined into a single complex structure, such as when a Tagalog reflexive construction (invariable semantic pivot) and any type of Tagalog clause structure (variable syntactic pivot) are combined to form a complex structure. And of course a language may not have any constructions that manifest PSAs, such as Mandarin Chinese and Riau Indonesian.

6.3.1 Noun Incorporation

A common feature of head-marking languages is noun incorporation. In some languages, such as West Greenlandic, noun incorporation is so productive that it is reported ‘there are infinitely many possible forms involving noun incorporation’ (Sadock Reference Sadock1991: 84).

We can start with an example from Nahuatl (Uto-Aztecan) given by Sapir (Reference Sapir1911: 260) in one of the first discussions of noun incorporation.

1. (4)

   a.	ni-c-qua	in	nacatl
   	1sbj-3obj-eat	the	flesh
   	‘I eat the flesh.’
   b.	ni-nica-qua
   	1sbj-flesh-eat
   	‘I flesh-eat.’

In (4a) the verb is clearly transitive, requiring the direct object pronominal form. The verb in (4b), however, is intransitive and so lacks the direct object marking. Now consider the following examples from Tongan (Polynesian) (5a–b) and Yucatec Maya (6a–b) (from Mithun Reference Mithun1984: 851, 857).

1. (5)

   a.	Na'e	inu	'a	e	kavá	-'é	Sione.
   	pst	drink	abs	conn	kava	erg	John
   	‘John drank the kava.’
   b.	Na'e	inu	kava	'a	Sione
   	pst	drink	kava	abs	John
   	‘John kava-drank.’

1. (6)

   a.	t-in-č’ak-Ø-ah	če'.
   	comp-I-chop-it-prf	tree
   	‘I chopped a tree.’
   b.	č'ak-če'-n-ah-en.
   	chop-tree-antip-prf-I(abs)
   	‘I wood-chopped.’

In both of these alternations, the verb is intransitive when the noun is incorporated, as can be seen by the absolutive case for the actor in both (5b) and (6b). (The intransitivity of the Yucatec Maya form in (6b) is also marked by the antipassive suffix, which is typically present in the derivation of an intransitive verb from a transitive.)

The key difference regarding semantic arguments in (4), (5) and (6) is that the unincorporated patient is referential and maps onto the undergoer position in a transitive construction. The incorporated nouns, however, are non-referential and therefore do not correspond to any specific argument in the LS, resulting in intransitive sentences. This kind of noun incorporation, therefore, has a detransitivizing effect.

Another type of noun incorporation – perhaps the most common cross-linguistically (McGregor Reference McGregor, Simon-Vandenbergen, Davidse and Noël1997) – is the incorporation of nouns referring to body parts. These constructions typically do not change a transitive verb into an intransitive. Rather than deleting the undergoer macrorole, they display an alternation in the linking to undergoer: in the non-incorporated forms, the body part is the undergoer; in the incorporated forms, the undergoer is the person, the possessor of the body part. In some languages the only nouns that can be incorporated into the verb are body parts.

Body part incorporation

The person affected can assume ‘a primary case role’ in these constructions because the transitivity of the verb is unchanged, with the possessor of the body part linking to the undergoer macrorole, as the ʻaffected personʼ.² Mithun (Reference Mithun1984: 857) provides the following example from Tupinamba (Tupí).

1. (7)

   a.	s-oβá	a-yos-éy
   	his-face	I-it-wash
   	‘I washed his face.’
   b.	a-s-oβá-éy
   	I-him-face-wash
   	‘I face-washed him.’

Unlike the previous examples of noun incorporation, the verb continues to be transitive – both forms in (7) have an actor and undergoer. Both sentences have the same LS but differ in regard to undergoer selection.

1. (8)

   a.	wash′ (a-, [have.as.part′ (s-, oβá)]) = (7a)
   b.	wash′ (a-, [have.as.part′ (s-, oβá)]) = (7b)

In both examples, the first person singular prefix a- maps onto the actor macrorole. In (8a) ‘his face’, as patient, is the unmarked choice for undergoer. The incorporated form in (8b) requires the marked choice for undergoer and the possessor maps onto the undergoer macrorole.

One other kind of noun incorporation needs to be briefly discussed here. At the beginning of this section, it was mentioned that Greenlandic Eskimo is known to have extremely productive noun incorporation (Sadock Reference Sadock1991). A language with similar constructions is Southern Tiwa (Kiowa-Tanoan) and both languages are discussed in Rosen (Reference Rosen1989). Consider the following examples from Allen et al. (Reference Allen, Gardiner and Frantz1984: 297).

1. (9)

   a.	Wisi	bi-musa-tuwi-ban
   	two	1sg:B-cat-buy-pst3
   	‘I bought two cats.’
   b.	Yedi ibi-musa-tuwi-ban
   	those cats-B:B-buy-pst
   	‘They bought those cats.’

These examples, like many in Allen et al. (Reference Allen, Gardiner and Frantz1984) and Sadock (Reference Sadock1991), are notably different from the noun-incorporation examples discussed previously, in at least two ways. First, the incorporated nouns are modified by a quantifier (9a) and a deictic (9b), and, second, they are clearly referential. Van Valin and LaPolla (Reference Van Valin and LaPolla1997: 66–68) present an analysis of similar constructions in Greenlandic Eskimo, showing how RRG accommodates such structures with distinct operator and constituent projections in the layered structure of the noun phrase. The significant conclusion is that, unlike the earlier examples discussed in this section, these do not involve the incorporation of a bare noun, but rather full RPs, as shown by the fact that they are referential and by the presence of modifiers.

6.3.2 Ditransitives

For our purposes here, I will assume the definition of ditransitives offered by Malchukov et al. (Reference Malchukov, Haspelmath, Comrie, Malchukov, Haspelmath and Comrie2010: 1):

As they point out, this excludes some verbs of transfer which do not involve a recipient (such as put) as well as benefactives (discussed in Section 6.4); but it includes verbs with a ‘recipient’ in both a literal and an extended sense. These are verbs that in English manifest the well-known dative alternation.

In RRG, ditransitives and other verbs with three arguments require a complex representation with more than one predicate. The common LS for a ditransitive involves two states of affairs joined by CAUSE.

1. (10)

   [do′ (x, Ø)] CAUSE [BECOME have′ (y, z)]

Following the AUH, the first argument of do′, x, is the actor, while the theme, the second argument of have′, z, is the undergoer.

Over the last fifty years, there have been many studies of the English dative alternation that try to account for the semantic similarities and differences between pairs of sentences such as those in (2) and (11).

1. (11)

   a.	Chris sent the money to Sam.
   b.	Chris sent Sam the money.

It should be noted that dative alternation is absent from many languages. Consider the following examples from Alacatlazala Mixtec, an Otomanguean language of Mexico (from Zylstra Reference Zylstra, Bradley and Hollenbach1991: 13) and from French.

1. (12)

   táshī	i	tūtū	ndāha	sīhí ñá
   (con)give	I	paper	hand	mother her
   ‘I’m giving the paper to her mother.’

1. (13)

   sīkō	ñá	noní	noo	i
   (compl)sell	she	corn	face	my
   ‘She’s selling corn to me.’

1. (14)

   Jean	a	donné	le	livre	à	Marie
   John	has	given	the	book	to	Mary
   ‘John gave the book to Mary.’

1. (15)

   *Jean	a	donné	Marie	le	livre.
   John	has	given	Mary	the	book
   ‘John gave Mary the book.’

Unlike English, Mixtec and French do not have dative alternation. In Mixtec languages the indirect object requires a body-part relational noun, functioning as a preposition (p.c. Carol Zylstra). If the indirect object follows the verb the result is ungrammatical. Likewise in French, a construction similar to the English ‘dative movement’ is ungrammatical. In both languages, the direct object behaves like the direct object in a simple transitive clause and the indirect object must be preceded by a preposition (or a relational noun, in the case of Mixtec). These exemplify what Dryer (Reference Dryer1986: 815) calls a ‘direct object’ language: the undergoer of the simple transitive continues to manifest the same morphosyntactic properties in the ditransitive construction (the ‘direct object’ is still the ‘direct object’). Such ‘direct object’ languages require the unmarked undergoer choice predicted by the AUH in ditransitives, choosing theme or patient as undergoer rather than the recipient.

However, some languages require a marked undergoer choice in ditransitives: the recipient, not the theme, is consistently linked to the undergoer macrorole. Dryer (Reference Dryer1986: 815) calls these ‘primary object languages’. Peterson (Reference Peterson2007: 144) provides the example of Hakha Lai, a Tibeto-Burman language.

1. (16)

   a.	vok	na-Ø-hmuʔ
   	pig	2sg.sbj-3sg.obj-see
   	‘You saw the pig.’
   b.	na-ka-hmuʔ
   	2sg.sbj-1sg.obj-see
   	‘You saw me.’
   c.	vok	na-ka-peek
   	pig	2sg.sbj-1sg.obj-give
   	‘you gave me the pig.’

The verb agrees with the actor and the undergoer in the simple transitive clauses (16a) and (16b) and with the recipient in the ditransitive clause (16c). This agreement pattern in the ditransitive is obligatory – the alternation found in English dative movement is absent. (For further discussion of Tibeto-Burman examples, see Van Valin and LaPolla Reference Van Valin and LaPolla1997: 272–273.) In ‘primary object languages’, unlike the Mixtec and French examples, the recipient is obligatorily linked to the undergoer macrorole.

In Van Valin and LaPolla (Reference Van Valin and LaPolla1997: 387) this marked choice for undergoer is attributed to animacy – the animate indirect object is chosen over the inanimate direct object for undergoer position (see, however, Guerrero and Van Valin Reference Guerrero and Van Valin2004 for further discussion).

Some languages, such as Mixtec and Tepehua, have no syntactically ditransitive verb roots. Although Tepehua has no verb roots with three direct core arguments, the LS for predicates such as ma:laqatʃa: ‘send’ (17a) and stʼa: ‘sell’ (18), like their counterparts in other languages, require three semantic arguments. Consider the examples in (17) and (18).

1. (17)

   a.	ma:laqatʃa:-ɬ
   	send-pfv
   	‘s/he sent it’
   b.	ma:laqatʃa:-ni-ɬ
   	send-dat-pfv
   	‘s/he sent it to him/her’

1. (18)

   a.	ʃtaq-ɬi
   	give-pfv
   	‘s/he gave it.’
   b.	ʃtaq-ni-ɬ
   	give-dat-pfv
   	‘s/he gave it to him/her’

Totonac-Tepehua languages have only one (or two) adpositions, with applicatives fulfilling a preposition-like function (Watters Reference Watters and Guerrero2019). Rather than employing a preposition to refer to the recipient, Totonac-Tepehua languages use the suffix ‑ni, signalling a marked undergoer choice. The applicative ‑ni licenses a third argument.

In (17) and (18), both the transitive and the ditransitive forms have the basic LS given in (10). The only difference is that the transitive constructions in (17a) and (18a) do not specify the y-argument (the recipient) in the LS.

1. (19)

   a.	[do′ (x, Ø)] CAUSE [BECOME have′ (Ø, z)] (= (17a), (18a))
   b.	[do′ (x, Ø)] CAUSE [BECOME have′ (y, z)] (= (17b), (18b))

Following the Completeness Constraint (1) the unspecified recipient in (19a) is absent from the syntax; in (19b) the recipient is specified in the LS, so must also be syntactically present. The recipient argument cannot appear in a clause unless the applicative suffix, ‑ni, occurs on the verb.

In the case of the verbs in (17a) and (18a), the theme is linked to the undergoer position and what would correspond to the indirect object is unexpressed. In both cases the simple transitive forms of such verbs are typically used in contexts where the recipient must be invoked from the context, as a definite null complement (Fillmore Reference Fillmore, Nikiforidou, VanClay, Niepokuj and Feder1986). Semantically, the recipient is required to complete the predication. This is one way in which an argument – in this case the unexpressed recipient – contrasts with adjuncts. Adjuncts modify the predication rather than complete it, and, when syntactically absent, the interpretation of the utterance does not require they be invoked by the listener.

The key evidence that the argument of the applicative ‑ni (the recipient) is linked to the undergoer macrorole is the effect of the antipassive suffix, ‑nVn (20b) (see Section 6.5.2), which deletes the undergoer.⁴ With a simple transitive verb in Tepehua, the direct object is syntactically obligatory. Consider the example sentences in (20). If no object noun phrase appears in the clause, it is still transitive, requiring a translation with a definite third singular pronoun (20a).

1. (20)

   a.	kin-kuku	st’a:-y
   	1poss-uncle	sell-ipfv
   	‘My uncle sells it.’
   b.	kin-kuku	st’a:-nan
   	1poss-uncle	sell-antip
   	‘My uncle sells.’
   c.	kin-kuku	ki-st’a:-ni-y	laqtʃ’iti
   	1pos-uncle	1obj-sell-dat-ipfv	clothes
   	‘My uncle sells me clothes.’
   d.	kin-kuku	st’a:-ni-nin	laqtʃ’iti
   	1poss-uncle	sell-dat-antip	clothes
   	‘My uncle sells clothes’

The translation of (20b) shows that the antipassive suffix ‑nVn in Tlachichilco Tepehua results in deletion of the patient undergoer. In (20c), the presence of the applicative suffix ‑ni, marks the recipient as undergoer. This is evidenced by the fact that it is the recipient that is deleted in (20d), when the antipassive suffix is present.

In sum, RRG posits only two macroroles even though predicates may have more than two arguments in their logical structure. Ditransitive constructions display different approaches to the syntactic position of the recipient argument. In some languages, the theme is the undergoer, following the default choice on the AUH. In other languages, the recipient is regularly the marked choice for undergoer, reflecting its prominence as the typically more animate non-actor argument. Finally, other languages allow variable linking to the undergoer macrorole.

6.3.3 Other Three-Place Predicates and Lexical Applicatives

While cross-linguistic studies such as Malchukov et al. (Reference Malchukov, Haspelmath, Comrie, Malchukov, Haspelmath and Comrie2010) limit the notion of ditransitive to three-place predicates in which the third argument is some kind of recipient, there are, of course, other kinds of three-place predicates. Common examples in English include the following.

1. (21)

   a.	Chris put the jar on the table.
   b.	Kim sprayed insect repellent under the bed.

The constructions in (21) involve a predicate with an indirect core argument of location. The predicates put and spray both have the LS [do′ (x, Ø)] CAUSE [BECOME be-loc′ (y, z)] and ‘in the actual semantic representation of a sentence be-loc′ would be replaced by the LS of a preposition’ that corresponds to the specific example (e.g. on, under, in, etc.) (Van Valin and LaPolla Reference Van Valin and LaPolla1997: 160).

There are at least two ways that the English dative constructions discussed in 6.3.2 are different from three-place predicates like these. First, these three-place predicates do not display similar argument alternations that affect undergoer assignment. Second, they involve predicative prepositions rather than the non-predicative prepositions found in ditransitives.

There are other classes of three-place verbs in English and other European languages that do not involve a dative argument but do display alternating argument structure. One well-known set of such verbs is the spray/load class (Fillmore Reference Fillmore, Bach and Harms1968: 48; Levin Reference Levin1993: 50–51).

1. (22)

   a.	He sprayed the paint on the wall.
   b.	He sprayed the wall with the paint.

It has often been noted that this alternation reflects a semantic distinction in that the NP or RP immediately following the verb shows a higher level of affectedness: as Fillmore notes (1968: 48, fn.49), the sentence in (22b) implies the entire wall got painted but (22a) does not. Furthermore, in the corresponding passive sentences, the semantic distinction is maintained, showing that the difference in affectedness is due to undergoer status, not to the syntactic position of direct object.

1. (23)

   a.	The paint was sprayed on the wall.
   b.	The wall was sprayed with the paint.

This alternation is a modulation of argument structure and, like the dative alternation, reflects different linkings to the undergoer macrorole. The semantic macrorole of undergoer, not the syntactic position of direct object, ‘represents the non-instigating, affected participant in a state of affairs’ (Van Valin Reference Van Valin2005: 61–62). Thus, considering the two non-actor arguments in each of the sentences in (22) and (23), the undergoer is the more affected one (see Van Valin Reference Van Valin2005: 113–114).

6.3.4 Causative

Lexical causative constructions built on intransitive verbs are generally straightforward when it comes to argument structure: the causee is linked to undergoer in both the intransitive and transitive constructions; in the transitive construction, the causer links to actor.

1. (24)

   a.	The door opened.
   b.	[BECOME open′ (door)]

1. (25)

   a.	Chris opened the door.
   b.	[do′ (Chris, Ø)] CAUSE [BECOME open′ (door)]

However, in languages that allow a causative built on a transitive verb base, the result is a predicate with three core arguments. The causer will be linked to the actor macrorole, but, as in the case of ditransitives, the question arises about which of the remaining two arguments will be linked to the undergoer position.

West Coast Bajau, one of the Sama-Bajaw languages of the Western Malayo-Polynesian branch of Austronesian, as reported by Miller (Reference Miller2007: 303), has a productive morphological causative prefix pe-.

1. (26)

   a.	Togor	bana	tiang	pagar	e.
   	upright	very	post	fence	dem
   	‘The fence post stands very straight.’
   b.	Boi	pe-togor	Mali	tiang	pagar	e.
   	comp	caus-upright	Mali	post	fence	dem
   	‘Mali erected the fence post.’

Miller (Reference Miller2007: 302–308) includes various examples of causatives with pe- on statives, inchoatives, manner, and activity verbs. He reports that few transitive verb roots take the pe- causative, but when they do, ‘the causee is the new undergoer’ (2007: 306). This is also the case in Tepehua, but in a more indirect fashion. The causee appears in the clause due to the presence of the same applicative ‑ni, presented in 6.3.2, which links an indirect object to the undergoer (the vowel of ‑ni is lengthened as part of the causative formation rule).

1. (27)

   transitive		causative of transitive
   ʔah-ya	‘s/he digs it’	maːʔah-ni:-y	‘s/he makes her/him/it dig it’
   ʃʔoq-ya	‘s/he unties it’	ma:ʃʔoq-niː-y	‘s/he makes him/her/it untie it’

In clear contrast to the pattern in the West Coast Bajau and Tepehua examples is causative formation on a transitive verb in which the undergoer of the base verb retains the syntactic marking and behaviour of undergoer. The causee then appears as a peripheral or indirect core argument. An example of this can be seen in the Turkish sentences in (28) (from Underhill Reference Underhill1976: 346):

1. (28)

   a.	Yusuf	diș-in-i	çek-ti
   	Yusuf	tooth-3poss-acc	pull-pst
   	‘Yusuf pulled his tooth.’
   b.	Yusuf	doktor-a	diș-in-i	c̣ek-tir-di
   	Yusuf	doctor-dat	tooth-3poss-acc	pull-caus-pst
   	‘Yusuf had the doctor pull his tooth.’

In both the basic transitive as well as the causative construction, the direct object, diṣ, ‘tooth’ is marked as accusative and, in the causative, the causee must be marked as dative.

Like other lexical constructions in this section, morphological causatives affect the level of mapping from predicate argument structure onto the actor and undergoer macroroles. For causatives formed on a transitive verb base, two non-actor core arguments are present in the LS, and languages differ in regard to which of the two links to the undergoer macrorole.

6.3.5 Anticausative

The anticausative construction has been described as the inverse of the causative: instead of adding a causer and causing event to the LS, the anticausative is a derived intransitive in which the undergoer is the PSA. Haspelmath and Müller-Bardey (Reference Haspelmath and Müller-Bardey2001) call the anticausative ‘the most radical agent-removing category’ and include examples such as the following from Hungarian and Turkish.

1. (29)

   a.	András-t	három	tárgy-ból	elvág-t-ák
   	András-acc	three	subject-elat	fail-pst-3pl
   	‘They failed András in three subjects.’
   b.	András	három	tárgy-ból	elvág-ód-ott.
   	András	three	subject-elat	fail-antic-pst(3sg)
   	‘András failed in three subjects.’

1. (30)

   a.	Anne-m	kapı-yı	aç-tı.
   	mother-1sg	door-acc	open-pst(3sg)
   	‘My mother opened the door.’
   b.	Kapı	aç-ıl-dı.
   	door	open-antic-pst(3sg)
   	‘The door opened.’

In each of these examples, the verb root is causative with actor and undergoer and the intransitive is derived by a morphological operation.

The RRG literature discusses two kinds of anticausatives. The alternations in (29) and (30) display an anticausative that is similar to the ‘middle’ construction in traditional grammar. In these constructions, ‘the function of the morphological markers is to cancel part of the logical structure’ (Van Valin Reference Van Valin2005: 46) – the anticausative removes the causer and causing event from an otherwise causative verb. Typically, the anticausative alternation applies to a minor subclass of verbs. This is the case in Tepehua languages, which have a small group of transitive verb roots that have derived intransitive forms marked by the inchoative prefix, ta-, as in the following examples.