The Hulle and Goed Constructions in Afrikaans

Abstract

Over the past more than 100 years, Afrikaans associative plural constructions – especially constructions with hulle (‘they’) and goed (‘things/stuff; good’) as right-hand components – have been studied from both diachronic and synchronic perspectives, but with the main interest in their origins, and what they could tell us about the genesis of Afrikaans. One school of thought claims that they both have Germanic roots, while the other school maintains that both are creole constructions. No definitive conclusions have been reached. Moreover, there is no consensus on whether these constructions should be regarded as noun phrases, compounds, or derived words. The most recent synchronic description of the hulle construction was published in 1969, and the last synchronic description of the goed construction in 1989. In the absence of corpus data, unsubstantiated claims about these constructions abound in the literature. This article presents a synchronic, corpus-based, constructionist description of these two Afrikaans constructions. They are characterised as hybrid constructions on a scale between compounds and derivations, while some remarks on their productivity are made. Based on detailed analyses of their right- and left-hand components, the article concludes with a categorisation network of the schemas and subschemas of these constructions.

1 Introduction

Afrikaans is generally categorised typologically as a West Germanic, Low Franconian language, originating from seventeenth century colloquial Dutch. Regarding its genesis, two main schools of thought persist: those that claim that Afrikaans can be traced back mainly to seventeenth century varieties of Dutch (the Germanists), and those that claim that pidgins or creoles once spoken in the Cape Colony later developed into Afrikaans and its dialects (the Creolists) (De Kleine 1997). Following from this, Den Besten (1989: 239) proposes a convergence model: “… two types of Dutch, European and Pidgin Dutch, gradually coalesced to yield Afrikaans and its dialects”, with the bases for Pidgin Dutch mainly Khoekhoe, Creole Portuguese and Pasar Malay.

Associative plural (APL) constructions (Daniel and Moravcsik 2013; Moravcsik 2003) in Afrikaans regularly feature in the debates between the Germanists and Creolists. High-level schemas for, plus prototypical examples of the two Afrikaans APL constructions and two related coordinate constructions found in the literature (Kempen 1969) are presented in (1) to (4); the generic semantic interpretation of Daniel and Moravcsik (2013) is used here as a point of departure.¹

(1)	Schema 1: hulle construction (APL)
	[[x]Ni -hulle PN.3PL]N.APLj ↔ [sem i and other people associated
	with sem i]j
	pa-hulle
	dad-they
	‘dad and mom; dad, mom and my other siblings; dad and his friends, etc.’
(2)	Schema 2: goed 1 construction (APL)
	[[x]Ni (-)goed PN.INDF.PL]N.APLj ↔ [sem i and other people
	associated with sem i]j
	pa-goed (or pa·goed)
	dad-they (or dad·they)
	‘dad and mom; dad, mom and my other siblings; dad and his friends, etc.’
(3)	Schema 3: x en hulle construction (coordinate)
	[[x]Ni en CNJ hulle PN.3PL]NP.COORDj ↔ [sem i and other people
	associated with sem i]j
	pa en hulle
	dad and they
	‘dad and mom; dad, mom and my other siblings; dad and his friends, etc.’
(4)	Schema 4: x en dié construction (coordinate)
	[[x]Ni en CNJ dié PN.DEM]NP.CCOORDj ↔ [sem i and other people
	associated with sem i]j
	pa en dié
	dad and these
	‘dad and mom; dad, mom and my other siblings; dad and his friends, etc.’

Regarding Schema 2 (henceforth the goed ₁ construction), we need to distinguish two other, non-APL constructions with goed, viz. the goed ₂ and goed ₃ constructions:

(5)	Schema 5: goed 2 construction
	[ [x]Ni goed ADJ]Nj ↔ [sem i who is important/dear to me]j
	pa·goed
	dad·good
	‘my dear/good dad’
(6)	Schema 6: goed 3 construction
	[ [x]Ni goed N.MASS]Nj ↔ [things/stuff related to sem i]j
	kooi·goed
	bed·things or bed·stuff
	‘bedding (like sheets, duvets, etc.)’

In its general usage:

• hulle functions as a third-person plural pronoun that can be translated with ‘they’ (as subject), or ‘them’ (as object) (glossed with 3PL);

• goed functions as:

  • (a) an indefinite plural pronoun (Ponelis 1979: 103, but elsewhere also referred to as a pronominal), translated with ‘things’ or ‘stuff’ (glossed with APL);

  • (b) a mass noun, translated and glossed mostly with ‘things’, but also sometimes ‘stuff’; or

  • (c) an adjective, translated and glossed with ‘good’.

To illustrate just one of the complexities regarding these constructions, a brief introductory note on goed ₂ is in order. The goed ₂ construction is used to refer hypocoristically or emphatically to referent [x], and has a singular interpretation (unlike the goed ₁ construction). Compare the example in (7) about a legend that was retold from one female to the next in the lineage of the family; note the anaphoric usage of sy (‘she.3SG’) with the antecedent ouma·goed, clearly signaling a singular interpretation of the antecedent.

(7)	Ek het dit die eerste by my ouma·goed ge·hoor, en sy het ge·sê dit kom van háár ouma·goed (Lombard 2014) 2
	I have it the first from my grandma·good PST·hear, and she have PST·say it come from her grandma·good
	‘I have heard it first from my dear grandma, and she said that it came from her dear grandma’

In this regard, Den Besten (2001: 52) states: “I do not regard the hypocoristic use of -goed … as being part of the associative phenomenon … This usage probably came about through secondary reinterpretation of the associative -goed on the basis of Afr. [adjective] goed ‘good’ …”. Van Rensburg (p.c.), who was the project leader of a large-scale project that described Orange River Afrikaans (the geolect of Afrikaans that originates from speakers of Cape Khoekhoe and Nama, and which is today used mainly in its spoken form), additionally states that the goed ₂ construction is not only used hypocoristically, but also (and especially) reverently. He mentions the case of kaptein·goed (‘captain·good’), which is used to refer to, or even to address the leader of a socio-economic group. In the remainder of this chapter, I will assume that these two studies on the genesis of Afrikaans are correct, and will not regard goed ₂ synchronically as an APL construction.

Over the past more than 100 years, these constructions have been studied³ from both diachronic and synchronic perspectives, but with the main interest in what they (and the similarities and differences between them) can tell us about the genesis of Afrikaans. The Germanists claim that they both have Germanic roots, either in Dutch compounds with goed ‘goods/things’ as right-hand member (like (6) above), or Frisian coordinate constructions like heit-en-hjar (‘dad-and-them’), and heit-en-dy (‘dad-and-these’) (Sipma 1913). The Creolists maintain that both are creole constructions, with roots either in Cape Khoekhoe, Nama,⁴ Malayan, or African languages. No definitive conclusions have been reached, and this article does not aim to contribute directly to this debate.

The most recent synchronic description of the goed ₁ and goed ₂ constructions was done by Links (1989), while the last synchronic description of the hulle construction was published in 1969 by Kempen (although Den Besten’s (1996) study could also be considered a synchronic description, albeit more theoretical in its aims). Kempen (1969) states that pa-hulle (in (1) above) and pa-goed (in (2) above) are fully equivalent in meaning, but that the latter is regarded “socially lower”, and that it could be “ignored as untranslated Khoekhoe” (Kempen 1969). In some of the other literature similar claims about these constructions are often made in passing, but not substantiated with corpus-based data. Moreover, there is not consensus on whether the hulle and goed ₁ constructions should be regarded as noun phrases (Den Besten 1996; Smith 1940),⁵ compounds (Booij 2010: 66; Kempen 1969), derived words (Deumert 2004), or indeed as “an oddity” (Moravcsik 2003).

From the literature on associative constructions in general, two main views on the semantics of these constructions have emerged. On the one side, Daniel and Moravcsik (2013) postulates an asyndetic coordinate interpretation as in (8) below. Moravcsik (2003) identifies the hulle (and per implication goed ₁) construction as an associative plural construction, which she defines as “constructions whose meaning is ‘X and X’s associate(s)’, where all members are individuals, X is the focal referent, and the associate(s) form a group centering around X” [my emphasis – GBVH]. She points out that “associative plurals fall between ordinary morphological plurals and conjoined nominals” (2003: 472), and that “both collective and distributive readings are possible” (2003: 488).

On the other side, Vassilieva (2008) proposes a subordinate interpretation: “An associative plural is a nominal expression that refers to a group by naming its most salient member. The construction is used to introduce a new group into discourse, a group that is understood to be inherently (or contextually) associated with its named protagonist.” [my emphasis – GBVH]. This view was already introduced by Den Besten (1996), and is formalised in (9). Although the difference between these two interpretations is subtle, it is pivotal for a proper understanding of the hulle and goed ₁ constructions, as will be argued in this chapter.

(8)	[x and other people associated with x] or [x and x’s associates]
(9)	[the group surrounding and including x]

The main aim of this chapter is to present a synchronic, corpus-based, constructionist description of these Afrikaans constructions. From a construction morphology (Booij 2010) and cognitive grammar (Langacker 2008) perspective, various schemas and subschemas are identified, clearly indicating where the constructions overlap but also diverge. It is illustrated that there are many misconceptions about these two constructions, especially regarding their meaning in actual, modern usage.

In Sect. 2, an overview of the corpus data is provided. Section 3 presents information on the frequency and productivity of these constructions, as well as possible answers to why the hulle construction seems to be “winning” over the competing goed ₁ construction. For a proper characterisation of the hulle and goed constructions, it is necessary to understand the differences between hulle as a plural pronoun, and the various senses of goed, as well as the component structure they combine with. In Sect. 4 hulle and goed as right-hand components of these constructions are analysed in detail, while a description of the left-hand components is provided in Sect. 5. Section 6 aims to give an answer on whether these constructions should be analysed as subschemas of noun phrases, compounds, or derived words, or perhaps rather as new nodes in a construction network. Based on these detailed analyses, the article concludes with a categorisation network of the schemas and subschemas of these constructions.

2 Data

The primary source of data for this research is the collection of corpora available on the online corpus portal of the Virtual Institute for Afrikaans (VivA 2017). This collection is made up of seven different corpora, comprising in total more than 86 million words (see Table 1). The majority of texts in these corpora can be considered edited texts (e.g. from publishers), and represents contemporary written standard Afrikaans (SAfr). A small portion of the Taalkommissie corpus is explicitly categorised as fiction (∼5,8 million words), while two other corpora (NWU/Lapa and PUK/Protea, together ~14,3 million words) also contain fictional texts. We can assume that we might find some examples, but by far not a proper representation of (spoken) dialectical Afrikaans (DAfr) in these subcorpora of the VivA corpus collection.

Table 1 Primary data sources (VivA 2017)

The usage of written data, even for DAfr that occurs most often in spoken form, bears commenting on. When using written data, and especially edited texts, one should tread carefully. For instance, we cannot say that ma-hulle is a coordinate compound because it is styled like other coordinate compounds in Afrikaans (i.e. conjunctively with a hyphen). However, we could say that ma-hulle is conceptualised or interpreted by language users as a coordinate compound, and it is therefore styled analogously to other coordinate compounds.

Of course, the orthographic tradition of a specific word, or more generally of a language, also comes into play. Ma-hulle might be styled conjunctively with a hyphen because that is just the way it has been written arbitrarily over a long period. Similarly, Afrikaans (like Dutch) has a long-standing tradition to write words in word groups and phrases as separate words (see 15.1 and 15.25 in AWS¹¹), while compounds are written conjunctively with or without a hyphen (see 15.2 in AWS¹¹). While these rules are in themselves also arbitrary, it does hold true that words that are interpreted as compounds in edited texts (like most of our corpora, but also in orthographic transcriptions of spoken corpora), are written conjunctively. The styling of words in such corpora therefore also provides information on how the structuring of these words was interpreted by authors, text editors, transcribers, etc.

In synchronic linguistics, the orthographic (or graphemic) realisation of language is often shunned as unimportant and of no concern when discussing language processing. Berg (2013: 387) points out that “[w]riting … plays a relatively minor to non-existent role in morphological theories”, perhaps because the orthography is often considered “less natural or even artificial … [as it] is regulated at will by a regulating committee” (Neef 2012: 5). However, Langacker (2008: 15) sets the framework for an alternative perspective when he argues that sounds, gestures and orthographic representations should all be seen as overtly manifested features of constructions, which play a crucial symbolising role in such form-meaning pairings. Regarding morphology, Berg (2013: 388) takes a strong stance when he says that “any description of morphology is incomplete without reference to the morpho-graphemic level”. This viewpoint is slowly becoming more popular in morphological circles: Berg’s article was published in the journal Morphology, and Bauer et al. 2013 dedicate a whole chapter to orthography in their book on English morphology. As part of the data collection process, aspects related to the various orthographical variants have been kept in mind, and will be commented on throughout the chapter.

The primary data are words ending in hulle (n = 2319) or goed (n = 5327, of which 76 could be regarded as goed ₁ or goed ₂ constructions; see Table 2 in Sect. 3). Irrelevant material was removed (e.g. data with vergoed ‘remunerate’), while obvious spelling errors were normalised (e.g. *briegoed > breigoed; *segued > sêgoed). The data were manually analysed and annotated by myself; however, to minimise subjective interpretations of the meaning of the constructions (see below), an experienced postgraduate student did the semantic annotations, which were subsequently verified by myself. The following levels of annotation were used:

• Form: All strings were split in constituents (e.g. wasgoed > was + goed), and all left-hand constituents were tagged with part-of speech categories (e.g. verb, proper noun, mass noun, etc.), with more specific categories for person names (i.e. first name, surname, title name, nickname, and kinship name, as well as combinations of these). In cases where the left-hand constituents could be unambiguously interpreted as multiword units, these were conjoined and annotated as such (e.g. oom Phil-hulle > oomPhil-hulle; Kyle Brown-hulle > KyleBrown-hulle). Hyphens were annotated as linkers (LK).

• Meaning: Where the referents of these constructions were people, they were annotated on more specific levels, viz. gender (male/female); generic relation (parent/grandparent/sibling/spouse/child/extended family⁶); and specific relation (father/mother, grandfather/grandmother, brother/sister, husband/wife, son/daughter, uncle/aunt, nephew/niece, brother-in-law/sister-in-law). In addition, all strings were considered a priori as compounds, and were manually annotated per the categories of Ó Séaghdha (2008), as operationalised in Verhoeven et al. (2014).

• Entrenchment: Based on all the sublemmas under the lemma -goed in the Woordeboek van die Afrikaanse Taal (WAT 2017), all examples that could be deemed entrenched to some degree, were identified. In addition, all hapax legomena were identified.

Table 2 Comparison of construction frequencies (VivA 2017)

Since we can assume that the VivA corpora are skewed towards more formal SAfr, other sources that were used to verify or supplement the main data set include:

• LAC: Leipzig Afrikaans Corpus (Projekt Deutscher Wortschatz 2017), comprising 165,594,102 words in texts mined from the internet.

• PCSA: Ponelis Corpus of Spoken Afrikaans (Ponelis 1976), containing 502,420 words of spoken data from face-to-face dialogue, lectures and radio interviews, sampled in the 1970s.

• HCSA: Historical Corpus of Standard Afrikaans (Kirsten 2015), a stratified corpus of non-fiction texts written in SAfr, comprising 1,032,180 words in total, divided into four periods: 1911–1920 (242,686 words); 1941–1950 (263,838 words); 1971–1980 (262,386 words); and 2001–2010 (263,270 words) (Kirsten 2016: 67).

• JLAFC: Jana Luther’s Afrikaans Fiction Corpus (Luther 2017) is a personal corpus (not available for distribution) of Afrikaans literary and popular novels and short stories, edited and published between 1996 and 2017. The corpus contains 17,903,824 words, and comprises texts written in both SAfr and DAfr.

• Google: Searches using Google have been used for what Fletcher (2007) calls “web hunting”, i.e. to find examples of constructions that might have been mentioned in scholarly literature, but that don’t occur (or occur with a very low frequency) in any of the other available corpora. Since “[t]he query, search and ranking optimization techniques [search engines] have adopted can either assist or sabotage a scholar’s quest” (Fletcher 2007), frequency counts from Google results are never used, unless the data have been carefully curated, following the protocol outlined by Van Huyssteen (2017).

The complete, annotated dataset is available at gerhard.pro/software.

3 Competing Constructions: Productivity and Frequency

The question that will be addressed in this section is whether the two APL and two coordinate constructions are – synchronically speaking – competing constructions, and if not, why not.

Firstly, it is of significance that no corpus evidence could be found for the two coordinate constructions (see (3) and (4) above) and their variants (see (10) and (11) below) observed by Kempen (1969) in the Swartland area in South Africa. Informal enquiries on Facebook, and personal communications with inhabitants of the Swartland area also yield no evidence of the existence of these constructions. Why this observation is of significance, is that Nienaber (1994) and Den Besten (1996, 2001) build many of their arguments about the diachronic development of the hulle construction on these observations of Kempen (1969). If no evidence of these constructions can be found, bar one observation by only one linguist, we might need to also reconsider the arguments of Kempen (1969), Nienaber (1994) and Den Besten (1996, 2001) about the diachrony of these constructions (and their subsequent claims about the genesis of Afrikaans, based on these constructions). However, such an endeavour falls outside the scope of this synchronic study. For purposes of this article, we can conclude that these coordinate constructions – and especially their variants – are extinct in modern SAfr (and even DAfr), and are therefore not considered further.

(10)	Development and variants of schema 3: en hulle construction
	pa en hulle (and variants such as Jakob en hulle ‘Jakob and they’) >
	pa-en-hulle > pa-n-hulle > paanhulle > paanulle
(11)	Development and variants of schema 4: en dié construction
	pa en dié (and variants such as Jakob en dié ‘Jakob and these’) >
	pa-en-dié > pa-en-doe(n) > pa-n-doe(n) > paandoe(n) > paando(n)

Do we notice other similar shifts in the usage and productivity of the two APL constructions? At the beginning of the twentieth century, Du Toit (1905) observes that the hulle and goed ₁ constructions were well established in the so-called coloured community (i.e. the main speakers of Khoekhoe Afrikaans at the time), but that only the hulle construction could be heard in the so-called white community, and importantly, that it could only be heard “sporadically” (Du Toit 1905: 86). If we ignore the aspects related to different speech communities, does this fact still holds true more than a 100 years later?

If we compare the frequencies of the hulle and goed constructions (see Table 2), we notice that the hulle construction (n = 2319) occurs almost ten times more in the corpora than its competitor, the goed ₁ construction (n = 25). Comparing goed ₁ and goed ₂ constructions with the goed ₃ construction, the latter (n = 5251) occurs much more frequently than its former two counterparts.

However, if we compare the productivity measures of these constructions (see Table 3), we see that the goed ₃ construction is much less productive than the others. Productivity is measured here in the narrow sense (Baayen and Lieber 1991) as P = n ₁/N, where P is the productivity measure; n ₁ the total number of hapaxes that contain the component; and N the token frequency of all words with that component. P is smaller for unproductive processes, and larger for productive ones. If we compare these productivity measures with that of the assumedly highly productive nominalising suffix -heid (Kempen 1969: 481), we see that -heid and goed ₃ constructions are comparable, in contrast with the much more productive hulle and goed _1/2 constructions. Therefore, although goed ₃ is more promiscuous (see Sect. 5) than the other constructions, it is less productive.

Table 3 Productivity measures (Baayen and Lieber 1991)

The question is therefore: Why do we find – more than a century later – that the hulle construction occurs almost a hundred times more in our corpora of written Afrikaans than the competing, probably much older – according to Nienaber (1994: 62) – goed ₁ construction? In her analysis of a relatively small, balanced corpus of historical texts of SAfr, all written between 1911 and 2010 (i.e. HCSA, comprising circa 1 million words in total; see Sect. 2), Kirsten (2016: 184–185) makes two observations:

• There is no evidence of the goed ₁ and goed ₂ constructions in HCSA, which she ascribes to the fact that these two constructions are used mainly in Orange River Afrikaans (therefore in DAfr, and not in SAfr).

• It seems as if the hulle construction sees a steady growth in SAfr: from only one instance in the period 1911–1920, to eight instances in the period 1941–1950, to 63 instances in the period 1971–1980 (but with only 16 instances in the period 2001–2010). She concludes that her corpus might be too small to reach reliable conclusions about the hulle construction.

Two more pieces of evidence can contribute to support Kirsten’s preliminary observations:

• It is a well-known thesis that the translation of the Christian Bible often plays an important role in the codification process of languages, and this was especially true for the codification and standardisation of Afrikaans (Naudé 2005). A search in the online 1933/-53 translation of the Bible⁷ produces no hits for the hulle or goed _1/2 constructions; in the 1983 Afrikaans translation though, 36 hits for the hulle construction occur, and still none for the goed ₁ and goed ₂ constructions. Like Kirsten’s evidence, this evidence is also meagre, but it resonates well with her observation of a growing trend for the hulle construction, while goed _1/2 constructions remain absent in formal, written SAfr.

• One of the most influential factors in the codification process of SAfr as we know it today, was the establishment of a spelling committee for Afrikaans in 1914, which is today known as the “Taalkommissie” (henceforth TK). The first TK was tasked with the compilation of spelling rules for Afrikaans, supplemented with a list of words. The first edition of this orthography, today still known as the Afrikaanse woordelys en spelreëls (‘Afrikaans word-list and spelling rules’; henceforth AWS), was published in 1917 (AWS¹), and the eleventh edition was published in 2017 (AWS¹¹).

  In the first five editions of the AWS, there are no traces to be found of either the hulle or goed _1/2 constructions. This is not completely surprising, since the early codification process of Afrikaans relied heavily on Dutch – to such an extent that scholars refer to a period of Dutchification (Uys 1983; Van Rensburg to appear), and Dutch-centrism (Nienaber 1994). Since none of these three constructions were known in Dutch, one can assume that the TKs at the time steered clear of officiating such expressions. It is only in the 1953 edition (AWS⁶) that we find the first lemma with hulle, viz. ma-hulle (mom-3PL), and in the 2009 edition (AWS¹⁰) pa-hulle (dad-3PL) additionally. The goed _1/2 constructions are officially recognised with two lemmas as SAfr in the 2017 edition (AWS¹¹), each with two styling variants: ma-goed/magoed; and pa-goed/pagoed (alongside ma-hulle and pa-hulle).

It seems therefore that the period of Dutchification and Dutch-centrism in the first half of the twentieth century had an important influence on the growth of the hulle construction compared to that of the goed _1/2 constructions, since the former was experienced as more “Dutch-like” (Nienaber 1994: 65). Influential linguists like W. Kempen, T.H. le Roux, J.J. Smith and H.J.J.M. van der Merwe served during this period on the TK, and their views of the goed _1/2 constructions as “perversions” (Van der Merwe 1964), or Khoekhoe-isms that could be ignored as “untranslated Khoekhoe” (Kempen 1969: 294),⁸ must have had an influence on the normative work of those TKs. In the twenty-first century, the methodology and underlying philosophy of the TK have changed to become more corpus-driven, and more inclusive of all the varieties of Afrikaans (see frontmatter of AWS¹⁰ and AWS¹¹). However, only time will tell if the recognition of the goed _1/2 constructions in AWS¹¹ will have an impact on the frequency of these constructions in SAfr.

4 Component Structures: Right-Hand Components

Since we have four remaining constructions (i.e., excluding the coordinate constructions in (3) and (4)) that are related to each other in different ways, and since the right-hand components of three of these constructions are identical in orthographical form (i.e. goed), it will be useful to distinguish semantically between these different right-hand components. They are the following (with examples of their independent usage):

(12)	hulle
	POS: third-person plural pronoun; Translation: ‘they’ (subject), ‘them’ (object)
	… nadat skap·e in die nabygeleë begraafplaas opgemerk is waar hulle kunsblomme eet.
	‘… since sheep·PL have been spotted in the nearby cemetery where they were eating artificial flowers.’
(13)	goed 1
	POS: indefinite plural pronoun; Translation: ‘they’ (subject), ‘them’ (object)
	Die jakkalse naai nie ons skap·e nie, hulle eet die fokken goed!
	The jackals screw not our sheep PART.NEG, they eat the fucking things!
	‘The jackals don’t screw our sheep, they fuckin’ eat them!’
(14)	goed 2
	POS: adjective (postnominal, or predicative); Translation: ‘good’
	Alle lewensstyle, goed of sleg, word dus bo kritiek verhef.
	All lifestyles, good or bad, are thus elevated above any criticism.
(15)	goed 3 (I)
	POS: non-plural mass noun; Translation: ‘stuff’
	… goed soos ros·e en skap·e en sampioen·e …
	‘… stuff like rose·PL and sheep·PL and mushroom·PL …’
(16)	goed 3 (II)
	POS: plural mass noun; Translation: ‘things/goods’
	… geen wonder jy kon nie behoorlik loop met die goed nie, die hakke is ʼn myl hoog …
	‘… no wonder you couldn’t walk on these things, their heels are a mile high

The difference between the various right-hand components of these constructions could be explained in terms of specificity (i.e. the level of instantiation that is foregrounded), focusing (i.e. the inherent boundedness, internal homogeneity, and salience of subparts within the scope of the structure), and perspective (i.e. grounding within the current discourse space) (Langacker 2008; Taylor 2002). Once we understand their respective conceptualisations, we could have a better understanding of the constructions that they favour.

To illustrate these general constructs, consider the following examples:

(17)	Hoe vang mens ʼn skaap ?
		‘How does one catch a sheep?’
(18)	… terwyl hy besig was om ʼn skaap te red wat in ʼn watergat beland het.
	‘… while he was saving a sheep that fell in a waterhole.’
(19)	Minstens 600 skap·e het doodgebrand …
	‘At least 600 sheep·PL burned to death …
(20)	Hy sê vee is in kwarantyn geplaas …
	‘He said livestock was put in quarantine …’
(21)	Die siekte veroorsaak letsels en sere aan diere soos beeste en skap·e …
	‘The illness causes lesions and sores on animals like cattle and sheep·PL …’

The meanings of lexical items are construed at different levels of specificity: the higher the level of specificity, the more schematic its construal is, and conversely, the lower the level of specificity, the more granular it is. In this regard, Langacker (2008: 264–272) distinguishes between type and instance conceptions within a lexical item’s domain of instantiation: type conceptions profile entities at a higher, more schematic level (the type plane; Langacker 2000: 270), while instance conceptions foreground distinguishing locations in the domain of instantiation (the instance plane; Langacker 2000: 270). In example (17) ʼn skaap ‘a sheep’ is construed as a type conception in its domain of instantiation – it refers to any sheep, whatever its size or gender, wherever in the world. In contrast, ʼn skaap ‘a sheep’ in example (18) refers to a specific sheep in a distinct location, in a specific waterhole; this is the prototypical construal of a singular count noun. In Fig. 1 this difference in construal is illustrated by the solid line around the activated domain of instantiation (DI): in the case of skaap as a type (marked by “t” in Fig. 1a), the construal is less specified, unlike the case where skaap refers to a specific instance (marked by a dot in Fig. 1b). In its type conception, skaap is therefore more schematic, since it abstracts away from the specifics of different sheep.

Fig. 1

(a) Count noun, singular, type (ʼn skaap); (b) Count noun, singular, instance (ʼn skaap); (c) Count noun, plural, instance (skape); (d) Mass noun, singular, instance (vee); (e) Mass noun; plural; type (skape)

The difference between singular count nouns, plural count nouns, and mass nouns centres around the inherent boundedness of the profiled entity. A singular count noun (like skaap in (18)) profiles a thing (used here in a technical sense – Langacker 2008: 98) that is prototypically discretely bounded, and hence replicable and countable. The plural of a count noun (see skape in (19)) profiles more than one of the same discrete, salient objects as a gestalt, which in its entirety is not discretely bounded, but rather amorphous and not inherently limited (Langacker 2008: 131). In Fig. 1c this amorphous boundedness of plurals is indicated with a dashed line, enclosing an unspecified number (indicated by ellipses) of instances (indicated by circles with dots). The enclosed instances are heterogeneous to some degree, since they are still discernible from each other (and hence countable).

Like count noun plurals, a non-plural mass noun (vee in (20)) also profiles an amorphous region in the domain of instantiation, and is therefore not countable and replicable. However, unlike plurals, the enclosed instances are homogenous and undiscernible (indicated with dashed lines in Fig. 1d). Another prototypical example to illustrate this, is the non-plural mass noun water, which profiles a homogenous mass with undiscernible parts. It is only perhaps for the chemist who looks at a sample of water under a microscope, that water consists of distinct particles.

Based on this similarity between count noun plurals and non-plural mass nouns, it is not surprising that plurals can function as mass nouns (like skape in (21)). In this case, the plural profiles an amorphous region in the type plane, while the instances are still discernible (i.e. heterogenous) but not salient (Fig. 1e). The word skape profiles a type of animal, similar to diamonds profiling a type of mineral in gold is forever, diamonds not.

The difference between a count noun and proper name (or any other named entity for that matter) is that the latter incorporates grounding in its conceptualisation, and thus singles out a discourse referent (Langacker 2008: 310). The participants (speaker(s) and hearer(s)) in a particular speech event that takes place at a specific time in a specific place, share the current discourse space (CDS), which is defined as “everything presumed to be shared by the speaker and hearer as the basis for discourse at a given moment” (Langacker 2008: 281). Consider the following examples:

(22)	In die geval van Dolly is gebruik gemaak van ʼn Skotse Swartkop-ooi se oösiet.
	In the case of Dolly, the oocyte of a Scottish blackface ewe was used.
(23)	Die skaap is ’n herkouer, wat beteken dat hy sy kos opbring …
	‘The sheep is a ruminant, which means that he regurgitates his food …’

The use of a proper name (like Dolly in (22)) assumes that the speaker and hearer shares a CDS where the topic under discussion (in the immediate preceding discourse frame) is cloning of animals, and not, for example, American female country music singers. In Fig. 2a the grounding elements (speaker S and hearer H) are included in the construed space, indicating with dashed arrow that they have a shared view of the thing that is being named. In addition to this thing being discretely bounded in a specific location, it also has a (unique) name (indicated by the symbol ™). Of course, other things can also have this “trade mark”, but in the CDS this proper name refers to a specific referent. Note that the type plane is in principle unspecified, since the name Dolly itself singles out the only instance in the CDS (Langacker 2008: 317).

Fig. 2

(a) Proper name (Dolly); (b) Pronoun, singular (hy)

Similarly, pronouns also rely for their conceptualisation on the incorporation of the CDS, more specifically the immediate preceding discourse frame. To understand the singular personal pronoun hy in (23), we need to understand that a singular sheep (skaap) has already been profiled in the previous discourse frame (the first part of the sentence, indicated by the left-hand block in Fig. 2b). The grounding elements share not only the identification of this thing in the previous discourse frame, but also know that it is identical to the thing in the CDS (indicated with a dotted correspondence line). While the skaap in the previous discourse frame is an instance of the type skaap (indicated with a “t” in the type plane), the type specification of hy remains schematic (indicated with ellipses in the type plane). Figure 2b represents the prototypical interpretation of a singular pronoun.

With this background knowledge, we can now give a more precise semantic characterisation of hulle and the different senses of goed. As a third-person plural personal pronoun, hulle (as in (12)) profiles a grounded, amorphous region in the CDS, where the enclosed entities are heterogenous, discernible and still salient with reference to the previous discourse frame. The pronoun hulle is used to refer to people, animals, plants, or inanimate things, although the third-person neuter pronoun dit can also be used to refer to animals, plants and inanimate things, but not people (Ponelis 1979: 591–593). Figure 3a gives a depiction of hulle as the right-hand component of the hulle construction (i.e. not as an independent word in a sentence). The entities that hulle refers to, are linked to the previous discourse frame with a dotted correspondence line. Here hulle is shown as an subject pronoun (nominative): hulle is the trajector (tr; the element being focused on, or the actor) in a simplex relationship, while it is the landmark (lm; the patient) when used as object pronoun (accusative) (Langacker 2008: 73).⁹ Note that when hulle is used as a right-hand component in the hulle construction, it makes schematic reference to the left-hand component (e.g. ma), which serves to elaborate (or characterise) the right-hand component in finer detail (Langacker 2008: 198). This schematic element is called an elaboration site (henceforth e-site), and is indicated by hatching in these diagrams; the line arrow points to the element that specifies, or “fills” the e-site. In the case of hulle, the e-site refers to any nominal – its left-hand component needs to be a grounded noun(phrase), since the referents in the previous discourse space are also grounded by the grounding elements, and are part of the construal of hulle.

Fig. 3

(a) hulle or goed ₁; (b) goed ₂; (c) goed ₃

In example (13), we see that goed ₁ can be used as an indefinite plural pronoun (or at least in a manner that resembles indefinite pronouns; also referred to as a pronominal), and especially most often as dié goed ‘these things’ and sulke goed ‘such things’. Ponelis (1979: 103) points out that goed has a plural reading in this usage, as opposed to the singular interpretation of iets ‘something’. In this sense, goed ₁ is translated with they (as subject) and them (as object), illustrating the overlap between goed ₁ and hulle. Example (13) could be reformulated just as well as in (24), where die goed is replaced with hulle. These two can therefore be seen as near synonyms, mostly only differing in terms of sociolinguistic dimensions. Hence, goed ₁ is also depicted by Fig. 3a.

(24)	Die jakkalse naai nie ons skap·e nie, hulle fokken eet hulle !
	The jackals screw not our sheep PART.NEG, they fucking eat them!
	‘The jackals don’t screw our sheep, they fuckin’ eat them!’

The diachronic reanalysis process of goed that was responsible for the development of goed ₂ (i.e. as a postnominal adjective in a compound), as well as a detailed semantic characterisation thereof, falls outside the scope of this article (see Sect. 1). Figure 3b will suffice to illustrate its adjectival nature (Langacker 2008: 102), specifically as a gradable, scalar, bounded adjective (Paradis 2001). As a postnominal modifier (i.e. as right-hand component), goed ₂ in this construction is similar to general in attorney-general, or emeritus in archbishop emeritus. It is therefore also not completely surprising that only the goed ₂ construction allows further morphological processes (also like attorney· s -general), viz. compounding with hulle (e.g. pa·goed-hulle dad·good-3PL; Den Besten 2001). However, such compounds are very rare in our data: There are two instances in the primary data, one in JLAFC (which happens to be the same example as one found in the primary data), and nothing else in any of the other corpora. A Google search with pagoed-hulle and magoed-hulle (and their orthographic variants) resulted in four unique hits. However, Van Rensburg (p.c.) points out that, from his personal observations, expressions like pagoed-hulle and kaptein·goed-hulle (captain·good-3PL) occur frequently in spoken Orange River Afrikaans. Based on the available corpus data though, I am inclined to conclude that even the hypocoristic goed ₂ construction does not really allow further morphological processes in written language.¹⁰

In addition, goed also functions as a non-plural mass noun, profiling an amorphous region (cf. Fig. 1d) that encloses any number of unspecified and even unrelated entities, including people, animals, plants, and inanimate objects, just like hulle. In this non-plural mass noun sense (see (15)), goed ₃ is best translated with stuff (notwithstanding subtle differences in formality), while in its plural mass noun sense (as in (16)), it is mostly translated with things (cf. Fig. 1e).

In compounds, goed ₃ functions mostly (but not exclusively) as a plural mass noun, since it profiles an amorphous region in the type plane: mess·e·goed knife·LK·things ‘cutlery’ profiles a kind of equipment, not a collection of knives. But unlike its singular counterpart, as a plural mass noun goed ₃ makes reference to identifiable, discrete and heterogeneous instances, thereby having an e-site that can be elaborated by any entity (the square block in Fig. 3c), including things (nouns), temporal processes (verbs), atemporal relations (adjectives and prepositions), etc. (Langacker 2008: 98–99).

Given this plural mass noun sense, it is somewhat surprising that goed also has plural forms, viz. goedere, goete, and goeters, which are almost always interchangeable with the singular form. HAT (2015) indicates that the plural form goedere has been lexicalised to such an extent that it now only refers to commercial products, while goete and goeters are used in informal contexts to refer to people, animals or things “which you can’t or don’t want to name precisely”. WAT (2017) specifies that these two forms are often used with some degree of contempt, and that goete could even be perceived as coarse. Whether this still rings true for its usage in modern day Afrikaans (the volume of the WAT covering the letter G was published in 1957) remains to be investigated, but what is true is that it conveys some emotive value, whether ameliorative or pejorative. We can safely assume that the meaning and usage of goedere, goete, and goeters have become specialised, and that none of them should be considered additive plurals of goed.¹¹

From this general characterisation of the lexical items hulle and goed, we can summarise some similarities and differences:

• Both hulle and goed profile an amorphous (unbounded) region in either the type plane or instance plane. They can therefore both refer to specific instances in the CDS, or more generically to types of those instances.

• In the case of hulle and goed ₃, the entities enclosed by the unbounded region are discernible, salient, and heterogeneous. In the non-plural mass noun sense of goed, the enclosed entities are not discernible, not salient, and homogenous.

• While hulle functions only as a pronoun (and grounding elements are therefore part of its conceptualisation), goed can also function as a (generic) indefinite plural pronoun.

5 Component Structures: Left-Hand Components

Any synchronic description of the hulle and goed _1/2 constructions should at least account for (or provide counter-evidence for) the prototypical subschemas summarised in Table 4. These subschemas are based by and large on the examples provided by Kempen (1969), but also supplemented with categories from other literature. All categories that were found in any of the literature are marked with a black dot next to the corpus counts.¹² The noteworthy cases are therefore the ones with black dots but without corpus evidence, or the ones with corpus counts but without black dots.

Table 4 Token frequency counts of left-hand components in the hulle, goed ₁ and goed ₂ constructions (VivA 2017)

Table 5 summarises the types of left-hand components that combine with goed ₃ as right-hand component. While goed ₁ and goed ₂ combine only with grounded nominals (e.g. person names, kinship names and title names), goed ₃ is much more promiscuous (Taylor 2002; Van Huyssteen 2010): It combines with words in many of the major part-of-speech (sub-)categories, especially count nouns (n = 2,177), verbs (n = 2,011), and adjectives (n = 555). However, if we look at the type:token ratio (Plag 2003: 52) of these three categories in the last column of Table 5, it is particularly low: 0,05, 0,06 and 0,05 respectively. This suggests that many of these words might have been lexicalised.

Table 5 Token and type frequency counts of left-hand components in the goed ₃ construction (VivA 2017)

If we consider the data in Table 4, most of the emerging constructions (i.e. constructions that have not been attested in the literature) are not surprising, as they are subschemas of previously identified, more general schemas. For example, oom Jakob-hulle (uncle Jakob-3PL) is merely a combination of the two well-known schemas [[x]_N.KINSHIP -hulle]_APL and [[x]_N.FIRST -hulle]_APL. The same principle applies to NPs consisting of title names with first names or surnames (e.g. professor Richards-hulle).

The cases that are mentioned in previous literature but that do not occur in our corpus data, are more interesting. About these cases, we can make a few general remarks.

As we have already noticed from Table 2 in Sect. 3, the goed ₂ construction’s hypocoristic/reverential interpretation is clearly more prevalent than the goed ₁ construction’s plural interpretation: 51 cases of goed ₂, versus 25 of goed ₁ (a ratio of 70:30). In the literature, cases like meester-goed (master-APL) are often mentioned to illustrate the construction’s associative meaning, but from the corpus data it is clear that the construction as a whole profiles a singular referent more prototypically. The corpus data suggest that this reinterpretation process might have developed even further in recent years, or perhaps previous researchers overestimated the associative interpretation of the goed _1/2 construction. Van Rensburg (p.c.) is even of opinion that especially associative pa-goed and ma-goed have been cited (and recycled) by linguists who have not necessarily had access to first-hand (corpus) data, and it therefore resulted in an overestimation of pa-goed and ma-goed as associative plurals (instead of singular hypocoristic/reverential expressions).

This possibility that the goed _1/2 construction should most often be interpreted hypocoristically/reverentially (rather than associatively), might be linked to the fact that it doesn’t seem to combine with the hypocoristic particle ou ‘old’ – contra to what is often stated in the literature. Perhaps hypocoristic goed is a strong enough marker of a hypocoristic meaning, so that the combination with ou becomes unnecessary.

Another subschema that is discussed by Donaldson (1993: 126) and mentioned by others, is u-hulle (2SG/PL-3PL), where u is a reverential pronoun, unmarked for number (similar to you). This construction could not be found in any of the corpora, except for two cases in LAC. Even a Google search could only provide two examples in the first 130 hits. We should therefore be careful to use u-hulle as a prototypical example of the hulle construction, since it seems to be rather rare.

It will be pointed out in Sect. 6 that, contrary to Den Besten’s (1996) claims, there is no evidence in our data that plural surnames can function as left-hand components of the hulle construction (e.g. die Steenkamp·s-hulle the Steenkamp·PL-3PL ‘the Steenkamp family’). Although never mentioned in the literature, singular surnames occur frequently (n = 110) in the data. Similarly, in contrast with the literature, no instances could be found of plural kinship names and plural common nouns as left-hand constituents (e.g. die oom·s-hulle the uncle·PL-3PL; die hond·e-hulle the dog·PL-3PL).

From the available corpus data, it emerges that hulle and goed _1/2 almost exclusively combine with human referents. Contrary to Kempen’s (1969) claims, there are no examples where these components combine with animal names (e.g. Boel-hulle, where Boel is the name of an animal). Links (1989: 32) mentions Saterdag·aand-goed Saturday·evening-APL ‘roundabout Saturday evening’, but no similar examples could be found in our data. Only two examples (both identical, and both in the same document) have a locative referent, viz. Kheis-goed se mens Kheis-APL PART.GEN person ‘person from the Kheis region’. Kempen (1969: 291) mentions Tuine-hulle Gardens-3PL ‘the team from the area/school Gardens’ as a possible (metonymic) extension of the general hulle schema; however, no similar examples occur in the primary corpus data, as well as in any of the secondary corpora. Again, despite Van Rensburg’s (p.c.) observation that goed ₁ often combines with place names in spoken Orange River Afrikaans, I am inclined to conclude that the hulle and goed _1/2 constructions seems to be choosy regarding their left-hand components, which are mostly human referents (at least in written language).

It should also be noted that the hulle and goed _1/2 constructions are generally considered in the literature to be informal (e.g. Webb 1989). Kempen (1969: 292) uses a biblical context to claim that one would not find examples like Christus-hulle Christ-3PL ‘Christ and his disciples’, thereby illustrating the colloquial nature of this construction. However, there are a total of 30 examples in our data where hulle combines with a Biblical first name in religious contexts; see examples (25) and (26). This evidence confirms that the hulle construction have grown in its range of usage contexts to include formal genres, as was argued in Sect. 3.

(25)	Joganan-hulle het die Here nie gehoorsaam nie …
	Johanan-3PL have the Lord not obey PART.NEG …
	‘So they [Johanan and his companions] entered … in disobedience to the Lord …’ (Jer 43:7; The Holy Bible, New International Version)
(26)	Josafat-hulle is die volgende môre vroeg uit …
	Jehoshaphat-3PL is the next morning early out …
	‘Early in the morning they [Jehoshaphat and his companion] left …’ (2 Chron 20:20; The Holy Bible, New International Version)

Other observations that can be made based on a semantic characterisation of the input categories, are about the profiled referents in these constructions. Table 6 gives an overview of those constructions where the kinship name [x] refers to a family member of the speaker. The constructions with parents as referents (e.g. pa-hulle or ma-goed) account for more than half of the cases, while oom ‘uncle’ and tante ‘aunt’ (cf. footnote 6) for more than a quarter of the data. This is not surprising, since the use of integrated appellatives (which are based on title names and kinship names) is a well-known phenomenon in Afrikaans (Jenkinson 1982). Our data confirms that roughly the same distributions can be observed for the hulle and goed constructions, although relative to constructions with oom, cases like tante-goed (and variations thereof) seems to be rarer than their counterpart tante-hulle.

Table 6 Profiled kinship referent’s relation to speaker

Table 7 gives a summary of all the gender roles that could be identified from the data; this include not only kinship names, but also cases with first names, or cases where it was clear from the context whether the referent is male or female (e.g. Pistorius-hulle refers to the Paralympic athlete Oscar Pistorius and his legal team, and is therefore assigned a male interpretation). In the hulle and goed _1/2 constructions the profiled referents are predominantly male in two-thirds of all cases. This will make for a strong argument to explicitly include a node pertaining to male referents in the final categorisation network (see Sect. 7).

Table 7 Gender of profiled referents

6 Schemas and Subschemas

It was pointed out in Sect. 1 that there is still no consensus on whether the -hulle and goed ₁ constructions should be regarded as noun phrases, compounds, derived words, or new nodes (“an oddity”) in a construction network. The aim of this section is to answer the question what these constructions are subschemas of. Are they compounds? If so, what kind of compound? If not, are they suffixal constructions? Or are they indeed new nodes in a construction network?

Den Besten (1996) provides three reasons why the hulle and goed ₁ constructions should not be analysed as compounds, but rather as pronominals consisting of double NPs. He argues that the hulle construction is a “syntactic collocation” of an NP and the plural pronoun hulle, because:

(27)	Reason 1: The x in [ [x] -hulle] can be a coordinated NP, as in Brian en Jakob-hulle ‘Brian and Jakob-3PL’, rendering the analysis [ [Brian en Jakob]NP -hulle]APL;
(28)	Reason 2: Such coordinated NPs can contain determiners as in die Van der Merwe·s en die Steenkamp·s-hulle the Van der Merwe·PL and the Steenkamp·PL-3PL ‘both of the families, also together with others’ (Den Besten’s translation), rendering the analysis [ [die Van der Merwes en die Steenkamps]NP -hulle]APL.
(29)	Reason 3: Hulle can be added to a simple NP of the type DET + N, as in die kind·ers-hulle the child·PL-them, rendering the analysis [ [die kinders]NP -hulle]APL; and

All three these reasons for discarding a potential morphological analysis could be refuted based on alternative bracketing, careful analysis of real-world data, and taking facts about Afrikaans morphology into consideration. For example, Brian en Jakob-hulle is inherently ambiguous: It could be analysed as either [ [Brian en Jakob]_NP -hulle]_APL (as Den Besten (1996) postulates), or just as well as [ Brian en [Jakob-hulle]_APL]_NP. Compare the following examples where the latter structure is more obvious (if not natural) than the previous, and there are therefore no grounds for rejecting Jakob-hulle as a potential compound:

(30)	Die Uil·e en Piet-hulle staan buite …
	The Owl·PL and Piet-3PL stand outside …
	‘The group of boys who call themselves the Owls, together with Piet and his crowd stand outside …’

(31)	Jy en Tom-hulle is natuurlik ook ge·nooi.
	You and Tom-3PL is naturally also PST·invite.
	‘You, Tom, and his crowd are of course also invited.’

Nonetheless, it is not at all rare for a complex NP (such as a coordinated NP) to function as a left-hand constituent in Afrikaans compounds and derivations. Compare examples (32) to (35) with the structure [ [x]_NPi [y]_Nj]_Nk ↔ [sem _j with relation r to sem _i]_k (so-called compounding compounds), or examples (36) to (39) with the structure [ [x]_NPi [y]_suffj]_Nk ↔ [sem _j with relation r to sem _i]_k (so-called parasynthetic compounds):

(32)	doring-in-die-vlees-goed
	thorn-in-the-flesh-stuff
	‘stuff that bothers me’
(33)	bek-en-klou·seer
	mouth-and-hoof·sore
	‘hoof-and-mouth disease’
(34)	Waarheid-en-Versoeningskommissie
	Truth-and-Reconcilliation-commission
	‘Truth and Reconcilliation Commission’
(35)	Kuns en Kultuur-uitkoms·te
	Arts and Culture-outcome·PL
	‘outcomes of the subject Arts and Culture’
(36)	heen-en-weer·tjie
	backward(s)-and-forward(s)·DIM
	‘short visit’
(37)	oor-en-weer-prat·ery
	to-and-fro-talk·NMLZ
	‘chatting’
(38)	traak-my-nie·agtig
	touch-me-not·ADJZ
	‘heedless, negligent, inattentive’
(39)	laag-by-die-grond·s
	close-to-the-ground·ADJZ
	‘banal’

With hulle we also find forms that are orthographically and structurally similarly to these examples, thus not excluding a potential morphological analysis:

(40)	ou-Melitie-hulle
	old-Melitie-3PL
	‘ol’ Melitie and her family’

(41)	Ver-Oupa-hulle
	Far-Granddad-3PL
	‘granddad and grandma who live far away’
(42)	wat-se-naam-hulle
	what-PART.GEN-name-3PL
	‘what’s-his-name’s crowd’

Pertaining to Den Besten’s (1996) second and third argument against a morphological analysis, we should note firstly that neither cases like die Steenkamp·s-hulle (i.e. [die [ [x]_N.SURNAME [y]_SUF.PL] -hulle]_APL), nor cases like die kinders-hulle (i.e. [die [ [x]_N [y]_SUF.PL] -hulle]_APL) occur in the corpus data. Both Den Besten (1996: 15) and Donaldson (1993: 136) mention the possibility of a plural surname as a left-hand constituent in the hulle construction, but no evidence for this subschema could be found in our primary data.¹³ Similarly, no instances could be found of plural kinship names and plural common nouns as left-hand constituents (e.g. die oom·s-hulle the uncle·PL-3PL; die hond·e-hulle the dog·PL-3PL), contrary to what is mentioned by, inter alia, Den Besten (1996) and Kempen (1969). Although Den Besten’s (1996) argument doesn’t focus on the plural marking of the left-hand constituent, the fact that we don’t find left-hand constituents with plural marking in the data, opens up a stronger argument for a morphological analysis, since word-formation processes like compounding and derivation in Afrikaans mostly only allow singular forms as left-hand constituents.

Nonetheless, according to Den Besten’s (1996) argument in reason 2 and 3, determiners like possessive pronouns (e.g. my in (43)), and definite articles (e.g. die in (44)) should be analysed as part of the hulle construction. Such an analysis is necessitated by his view that the hulle construction is pronominal, as illustrated by the fact that my pa-hulle or die dominee-hulle as a whole can be substituted by the single pronoun hulle.

(43)	My pa-hulle behoort tevrede te wees. > Hulle behoort tevrede te wees.
	my dad-3PL should content to be. > they should content be.
	‘My dad and mom / my dad and his friends should be content.’
(44)	… die dominee-hulle het die oggend daar aangekom … > hulle het die oggend daar aangekom
	… the reverent-3PL have the morning there arrived … > they have the morning there arrived
	‘… the reverent and his associates arrived there that morning …’

If we look at the part-of-speech categories of the left-hand collocates of the hulle construction (see Table 8, where collocate frequency ≥ 20) we notice that such NPs indeed very often include possessive pronouns (PN.POSS; n = 447), the hypocoristic particle ou (PART; n = 21), and the definite determiner die (DET; n = 20) as grounding elements in the NP. As I have argued from a semantic point of view in Sect. 4, the presence of these grounding elements in the vicinity of the hulle and goed constructions is not surprising at all, but that doesn’t mean that they have to be analysed as part of these constructions (like Den Besten (1996) does), as I will argue below.

Table 8 Left-hand collocates of the hulle construction

Table 9 Overview of literature on the associative plural in Afrikaans

From a morphological point of view, there is no need to analyse these left-hand collocates as part of the hulle and goed constructions. While Den Besten’s (1996) bracketing renders the schema in (45), another analysis could just as well render a nonpronominal analysis as in (46) (where the NP as a whole can be replaced with a pronoun, but not the APL construction alone). Den Besten’s argument against a morphological analysis hinges on his conclusion that an “asyndetic coordination analysis does not work” (1996: 17), and although I agree with him that an asyndetic coordination interpretation is not appropriate for these constructions, I don’t agree that a morphological analysis is not possible, as I will argue below.

(45)	[ [ [a]DET [x]N ]NP -hulle]APL
(46)	[ [a]DET [ [x]N -hulle]APL ]NP

Literature on associative plurals most often follow some form of coordinate interpretation of these constructions (cf. (8) above). This definition might lead us to consider the hulle and goed ₁ constructions as coordinate compounds, similar to (47). Moreover, coordinate nominal compounds are usually styled in Afrikaans with a hyphen between the two constituents, similar to ma-hulle.

(47)	Schema 7: coordinate compound (Van Huyssteen and Verhoeven 2014)
	[ [x]Zi [y]Zj]Zk ↔ [sem i and/or sem j]k, where Z = N/V/ADJ/ADV/PREP
	digter-skilder
	poet-painter
	‘poet (and) painter’

There are several reasons why such an analysis would not be appropriate for the hulle and goed ₁ constructions:

• Like in English (Plag 2003: 138–139), Afrikaans compounds usually have leftward stress, while nominal coordinate compounds have rightward stress (i.e. digter- skil der). Den Besten (1996) states about the hulle construction (but also applicable to goed _1/2 constructions) that “main stress does not fall on hulle but on X” – similar therefore to, for example, subordinate compounds.

• Coordinate compounds require that both constituents should have the same part-of-speech subcategory (e.g. person name + person name; transitive verb + transitive verb), and that the resulting compound also has the same part-of-speech subcategory. As is evident from Sects. 4 and 5, this is not the case in the hulle and goed _1/2 constructions. On a very high level of abstraction, one could say that the hulle construction is an elaboration of a nominal + nominal compound (where a nominal is defined as a grounded NP; Langacker 2008: 310), but the more precise, lower-level part-of-speech categorisation reveals that the hulle and goed ₁ constructions should not be regarded as coordinate compounds.

• Afrikaans coordinate compounds, like their Dutch equivalents (Booij and Van der Wouden 2016), usually have final plural marking (e.g. digter-skilder·s poet-painter·PL ‘poet-painters’), although double plural marking is also possible (e.g. digter·s-skilder·s poet·PL-painter·PL ‘poets-painters’). Despite claims in the literature that the left-hand component in the hulle and goed ₁ constructions can be a plural, we haven’t found any evidence in our data of such an extension of the schema (see Sect. 5).

• Moreover, coordinate compounds without plural marking (digter-skilder) always have a singular interpretation (‘s/he is a poet and painter), unlike the hulle and goed ₁ constructions that always have plural interpretations. In this sense, these constructions are more like true Sanskrit itaretara dvandva compounds (Egenes 2003: 211–212) of the kind mātā-pitarau mother-father ‘parents’. However, the constituents in these dvanda compounds are always singular, unlike the hulle and goed ₁ constructions.

• Most importantly, coordinate compounds are in essence asyndetic, expressing an AND relationship between the two constituent. Den Besten (1996: 16) points out that the hulle construction is usually “translated as ‘X and his/her/their folks’, although that is somewhat imprecise”. He continues to argue that “Pa-hulle in the reading ‘Dad and his folks’ does not mean ‘Dad and THEM’ (or ‘Dad and STUFF’ for the goed ₁ construction) (i.e. asyndetic coordination) because there is no independent reference for hulle ‘them’ or goed ‘stuff’. In order to make the referential properties of hulle and goed explicit we should rather rephrase ‘Dad and his folks’ as something like ‘the group surrounding and including Dad’” (Den Besten 1996; cf. (9) above). In supporting this interpretation, we can then conclude that the hulle and goed ₁ constructions should not be analysed as coordinate compounds, but perhaps rather as subordinate compounds, similar to (48).

(48)	Schema 8: subordinate compound (based on Van Huyssteen and Verhoeven 2014)
	[ [x]Zi [y]Zj]Zk ↔ [sem j with relation r to sem i]k, where the Z of [x] =
	N/V/ADJ/ADV/NUM/PREP/P/Sw; and the Z of [y] =
	N/ADJ/V/V-NMLZ/V-ADJZ/Sw14
	sjokolade·koek
	chocolate·cake
	‘chocolate cake’ = [cake containing chocolate]

Since subordinate compounds usually carry left-hand stress, this analysis corresponds to the stress patterns of the hulle and all the goed constructions. The goed ₃ construction can be analysed readily and unproblematically as a subordinate compound: kooi·goed bed·things ‘bedding’ is simply paraphrased as [things related to/used on bed] (cf. schema in (6) above). However, it would be futile to attempt to analyse the goed ₂ construction as a noun + adjective compound, unless one regards it as a left-headed compound, which would be rare (though not exceptional) in Afrikaans. Such an analysis, however, does not account for the fact that goed doesn’t mean ‘good’ anymore, but rather ‘important’ or ‘dear’.

A subordinate compound analysis of the hulle and goed ₁ constructions also brings us closer to Den Besten’s (1996) and Vassilieva’s (2008) semantic interpretation of associative constructions, as postulated in (9) above. A small-scale, informal pole¹⁵ on a Facebook group for Afrikaans language practitioners also confirms this interpretation: 87% of the 52 respondents considered the subordinate interpretation [the group individuals which x is part of] most natural, while 13% chose the coordinate interpretation [x and others associated with him/her]. Nobody chose the asyndetic option [x and they/them].

However, two problems remain with such a subordinate compound analysis. There is still no independent reference (Den Besten 1996) for hulle or goed ₁. If we respect the inheritance principle, the analysis would imply the subschemas:

• [ [x]_Ni -hulle _PN.3PL]_N.APL ↔ [they with salient member sem _i]_k,

• [ [x]_Ni (-)goed _PN.INDF.PL]_N.APL ↔ [stuff / they with salient member sem _i]_k.

In addition, the problem regarding the possessive pronouns, the hypocoristic particle ou, and the definite determiner die is still not resolved by such an analysis (cf. the schema in (46) above).

The solution proposed here is that hulle and goed _1/2 should be analysed as suffixoids on the continuum between syntax and morphology, and compounding and derivation (Van Goethem 2008). Booij (2005: 114) defines affixoids as “morphemes which look like parts of compounds, and do occur as lexemes, but have a specific and more restricted meaning when used as part of a compound”. Additionally, Booij and Hüning (2014) characterise affixoids “as the lexically specified parts of constructional idioms … [i.e.] schemas for subsets of compounds in which one of the slots is lexically fixed.” The notion holds that the independent lexeme goed (‘they; things/stuff; good’) occurs regularly in the goed _1/2 construction, which is seen as a constructional idiom where the right-hand slot is lexically fixed (with goed). Goed then develops a specialised, dependent meaning (i.e. [group with salient member sem _i] and [sem _i, who is important/dear to me]), so that it might eventually grow into a bound morpheme (mostly derivational first, then inflectional). Given the idea of a hierarchical lexicon, these constructions could then be considered primarily as subschemas of the subordinate compound construction (schema 8; following Booij’s (2010) interpretation), and simultaneously of the more general category-preserving suffixal construction (schema 9).

(49)	Schema 9: category-preserving suffixal construction
	[ [x]Zi [y]SUF.CN]Zk ↔ [sem SUF related to sem i]k
	huis·ie
	house·DIM
	‘small house’

Importantly, when hulle and goed _1/2 are characterised as affixoids in constructional idioms, we should note that this doesn’t imply a new category of morphemes or word-formation processes. Rather, within the framework of construction morphology, it affords us the opportunity to acknowledge and describe phenomena that lie in-between affixes and lexical words, and in-between compounding (e.g. schema 8) and derivation (e.g. schema 9). As Booij and Hüning (2014) put it: “The term ‘affixoid’ is a useful descriptive term to denote the phenomenon of bound meanings for words when embedded in complex words”, and that it “does not force us to make an absolute distinction between compounding and derivation”.

Such an analysis provided a solution for the independent referential problem, since the specialised meaning of hulle and goed ₁ in these constructions developed into [group], while goed ₂ developed into [who is important/dear to me]. The group or person is then specified by the referent [x], to render the respective meanings [group with salient member x] and [x, who is important/dear to me].

Also, if viewed as affixoids, it opens the potential to analyse them as phrasal affixes (Anderson 1992) that also attach to NPs, similar to the English genitive (e.g. King George’s birthday, or my dad’s book), or the Dutch genitive (Koning Willem-Alexander·s verjaardag ‘King Willem-Alexander’s birthday’, or mijn vader·s boek ‘my dad’s book). Although much rarer in Afrikaans, phrasal affixes are not exceptional; compare for instance examples (36) to (42), as well as (50) to (52).

(50)	Ho Tsji Minh·stad·t·er
	Ho Chi Minh·city·LK·NMLZ
	‘inhabitant of Ho Chi Minh City’

(51)	ter tafel ge·leg·d·e (mosie) < (mosie) ter tafel lê
	on table PTCP·lay·PTCP·ATTR (motion) < (motion) on table lay
	‘(motion) has been tabled/submitted’ < ‘to table/submit (motion)’
(52)	voor die hand ligg·end < voor die hand lê
	before the hand lie·PTCP < before the hand lie
	‘obvious’ < ‘to make obvious’

Furthermore, Stevens (2005) argues that the notion of affixoids is especially useful for describing and understanding morphological change, while Booij (to appear) also points out that “morphology often derives historically from syntax,” resulting in transitional cases like the hulle and goed _1/2 constructions. Some traces in our data suggest that hulle might already be more grammaticalized than what was previously not even considered by other scholars. Compare the reduced (enclitic) forms ’le and -le of hulle, which combines with NPs containing kinship names and person names ending in [a:] in the following examples:

(53)	Sy ma· ’le het ge·skei…
	His mom·3PL have PST·divorce …
	‘His parents got divorced …’
(54)	… in haar ma -’le se kamer … (JLAFK)
	… in her mom-3PL PART.GEN room …
	‘… in her parent’s room …’
(55)	… ek en An Trooi sit by Sanna -’le … (JLAFK)
	… I and Aunt Trooi sit with Sanna-3PL …
	‘… Aunt Trooi aan I sit with Sanna and her crowd …’
(56)	Waa·’s Kanna -le dan?
	Where·is Kanna-3PL then?
	‘Where’s Kanna and her crowd then?’

There are only five such examples in all the available corpora of edited texts, but a Google search revealed that it is seemingly a productive form in unedited texts, like on social media. Forty-six examples with ma ‘mom’, and 29 with pa ‘dad’ were found, as in (57) below (by a famous Afrikaans pop singer, posting a photo of her and her family):

(57)	Ek en @bobbyvjaarsveld en @derickhougaard Anna, Sion, en my pa· ’le !
	I and @bobbyvjaarsveld and @derickhougaard Anna, Sion, and my dad·3PL!
	‘Me, Bobby, Derick, Anna, Sion, and my parents!’

This development is not surprising. Vassilieva (2008: 345) points out that Daniel (2000: 47–48) “observed that group expressions used as associative markers tend to show signs of phonetic reduction when compared to their independent lexical counterparts, which is a tell-tale sign that we are dealing with lexicalization of functional heads.” As is clear from the examples above, Daniel’s observation also rings true for Afrikaans, although in the view held in this chapter, it is a tell-tale sign of the grammaticalization (rather than lexicalisation) of the hulle construction.

Revised schemas of the initial schemas (Sect. 1) are presented in (58) to (61). Note that the hulle and goed _1/2 constructions are represented as constructional idioms, where hulle and goed _1/2 are fixed as right-hand constituents. The respective schemas also represent the specialised meanings that have developed in these constructions. Schemas 3 and 4 are not repeated here, since no evidence for the existence of those constructions could be found.

(58)	Schema 1 (revised): hulle construction (APL)
	[ [x]Ni -hulle]N.APLj ↔ [group with salient member sem i]j
	pa-hulle
	dad-they
	‘dad and mom; dad, mom and my other siblings; dad and his friends, etc.’
(59)	Schema 2 (revised): goed 1 construction (APL)
	[ [x]Ni (-)goed]N.APLj ↔ [group with salient member sem i]j
	pa-goed (or pa·goed)
	dad-they (or dad·they)
	‘dad and mom; dad, mom and my other siblings; dad and his friends, etc.’
(60)	Schema 5 (revised): goed 2 construction
	[ [x]Ni goed]Nj ↔ [sem i who is important/dear to me]j
	pa·goed
	dad·good
	‘my dear/good dad’
(61)	Schema 6: goed 3 construction
	[ [x]Ni goed N(mass)]Nj ↔ [things/stuff related to sem i]j
	kooi·goed
	bed·things or bed·stuff
	‘bedding (like sheets, duvets, etc.)’

7 Summary

Based on the general formal and semantic characterisation of the hulle and goed constructions in Afrikaans, I conclude that the hulle and goed _1/2 constructions should be analysed as constructional idioms, in-between subordinate compounds and category-preserving suffixal constructions, while the goed ₃ construction is a subschema of subordinate compounds. The words of Lowe (2013) sums it up: “Linguistic categorization is rarely neat, insofar as linguistic phenomena rarely fit absolutely into the boxes we construct for them. While there is something theoretically elegant about being able to say that the [hulle and goed constructions are] purely [compounds], or purely [affixal], the evidence does not support either absolute analysis.” [My adaptations – GBVH.].

Since “… the lexicon consists of a network of constructions on different levels of abstraction, ranging from very abstract schemas to individual words” (Hüning and Booij 2014), we can postulate a categorisation network that not only includes high-level, schematic nodes, but also specific linguistic expressions (e.g. highly entrenched instances like pagoed and magoed). In Fig. 4, when a subschema is fully compatible with a higher-level schema (i.e. an elaboration of that schema), it is indicated with a solid arrow; when it conflicts to some degree (i.e. an extension of that schema), it is indicated with a dashed arrow (Langacker 2008: 17–18). Perceived degrees of prototypicality (also based on frequency counts) are indicated with the thickness of lines: the thicker a line, the more prototypical the subschema. Similarities between schemas are indicated with dotted lines. For convenience of arrangement and surveyability, not each and every lexicalised item, or even finer details of some schemas, are included here.

Fig. 4

Categorisation network

In future, this categorisation network could be expanded to include diachronic information about the various linguistic sources of these constructions. As Booij (to appear) points out: “… it is important to understand the differences and similarities between phrasal and morphological constructions, and it may not always be easy to make this distinction due to this historical source of compounds.” In my opinion, the constructionalization approach of, amongst others, Hilpert (2013), and Traugott and Trousdale (2013), holds much promise to better our understanding of the constructional changes involved in the genesis of these constructions.

Appendix

Table 9 (continued)