Rethinking the Description and Typology of Cantonese Causative–Resultative Constructions: A Dynamic Constructionist Lens

Abstract

This article proposes a new description of Cantonese causative–resultative constructions (CRCs), constructions with two verbal elements relevant to the cause and the effect of an event respectively. We present a constructional schema for the CRC with three argument types and without using traditional categories—such as subject, object and pseudo-passivation, present various syntactic and semantic properties, and subsume constructions such as the comparative construction and numerous particle constructions under this banner. We then argue against traditional approaches to CRCs with two lexical verbs that treat the argument structure of the CRC as composed from argument structures of individual verbs (the decompositional approach); instead, CRC arguments belong to the entire construction and have only semantic orientation-based relationships with individual verbs (the holistic approach). We show how our account can shed light on Sinitic typology and the grammaticalisation mechanism of verbal particles within CRCs, particularly the extension of result verbs into a broader range of contexts to become particles. We also argue that Cantonese CRCs challenge many assumptions of serial verb typology, which typically ignore the existence of multiple layers of constructional abstraction and assume decompositional descriptions, and urge for methodological advancements in this field.

1. Introduction

This paper presents a novel approach to Cantonese causative–resultative constructions (CRCs) (terminology from Yue-Hashimoto (2003)), with implications for both Sinitic and worldwide typology. An example of the construction is (1):

(1)	喊	濕-咗	幾多	包	紙巾？
	haam3	sap1-zo2	gei2do1	baau1	zi2gan1
	cry	wet-PFV	how many	packet	tissue
	‘How many packets of tissue did (he) wet by crying?’ [SEA112].

This example (1) is a single clause with two predicative components: the first verb, cry, is the cause of the second verb, wet. ‘Typical’ CRC examples such as (1) put together one dynamic and one stative verb, but our definition of CRCs encompasses a myriad of other construction types, such as the following directional particle (2) and causative verb (3) constructions:

(2)	落	低	個	窗簾
	lok6	dai1	go3	coeng1lim2
	descend	low	CLF	curtain
	‘Lower the curtains.’ (Yiu 2013).

(3)	成	個	電影	業	俾	你	搞	軭	晒
	seng4	go3	din6jing2	jip6	bei2	nei5	gaau2	waang1	saai3
	whole	CLF	film	industry	AGT	2SG	make	ruined	all
	‘The entire film industry has been ruined by you.’ [SEA159].

In Chinese linguistics, the structure and semantics of these constructions have been an evergreen topic. In China, the resultative component of such constructions is typically known as a ‘complement’, and there are detailed investigations of syntacto-semantic properties of lexical items appearing in this slot in different varieties of Chinese across time and space (e.g., Yue-Hashimoto 1993, 2003; Cheung 1972; Wú 2003, 2005). In Western traditions, these constructions have been explored for their insights on argument structure (e.g., Cheng and Huang 1994; Cheng et al. 1997; Wang 2001; Lau and Lee 2015, 2021) and their role in the grammaticalisation of verbal particles (e.g., Yiu 2013; Chor 2010, 2013, 2018).

Outside Chinese linguistics, these constructions have received less attention. For example, Chappell et al. (2007) wrote that, ‘an area in which the study of Sinitic languages can afford a rich contribution to typology is that of verb complementation…These structures show great diversity across the dialect spectrum in China yet are little known outside of Chinese linguistic circles.’ Nonetheless, at least two strands of worldwide typological research have examined this construction: Talmyan verbal semantics (e.g., F. Li 1997; Lamarre 2003) and serial verb typology (Aikhenvald and Dixon 2006; Haspelmath 2016).

These traditions differ substantially in assumptions, terminology, framing and research foci. In this paper, we present and defend our own approach to CRCs which draws from many of these approaches but departs substantially from all of them. Our general approach, while eclectic, is mainly inspired by Western dynamic constructionist approaches that view syntax as an inventory of signs at various levels of abstraction (e.g., Croft 2007; Diessel 2019), and supplemented with Chinese-origin notions, particularly semantic orientation, which has received scant attention in the West.

This paper has two main goals. The first is to argue that the constructions under our banner of causative–resultative construction (CRC) form a coherent category. Most traditions do not recognise this exact constructional level; they use only terms that are either supersets (e.g., complement, secondary predicate, serial verb) or subsets (e.g., resultative compound/complement, directional complement) of the CRC. We propose a construction schema for the CRC with various semantic and syntactic properties. Dispensing with traditional notions such as ‘subject’, ‘object’ and ‘pseudo-passive’, our schema accounts for a wide variety of constructions under a single schema, including such subconstructions as resultative, directional, phase and causative constructions. Although the CRC is not without precedents in the literature, we believe our account contains the most explicit definition of and argumentation for this constructional level.

The second purpose is to argue against traditional views, commonplace in most contemporary Western descriptions and typology, that the causative–resultative is composed of two separate argument structures (the decompositional approach). After presenting a range of examples not amenable to the decompositional analysis, we argue that a single argument structure construction (Goldberg 1995; Goldberg and Jackendoff 2004) suffices to describe the Cantonese CRC (the holistic approach). We show that phenomena previously described in terms of individual argument structures can be captured by our approach with the notion of semantic orientation, and finally present a typology of Cantonese CRC argument structures that capture all constructions previously described in compositional approaches such as Lau and Lee (2021), while also successfully accommodating argument structure configurations not captured in previous typologies.

As we will argue, these two points pose substantial difficulty for typology but also open new avenues for investigation. Methodologically, the higher-level CRC construction’s existence urges typologists to rethink what constitutes a ‘construction’ in typology, which often prides itself on having developed from whole language typology to typologies of constructions (Bickel 2007). The difficulties faced by the decompositional approach also call into question some definitions and proposed universals of serial verb constructions. However, the higher-level CRC construction also provides evidence to distinguish between the two competing accounts of the relationship between SVC symmetry and grammaticalisation in Bisang (2009) and Aikhenvald and Dixon (2006). Semantic orientation analysis also sheds light on the grammaticalisation mechanism of some verbal particles in Chinese, and our work has potential implications for synchronic comparison across Sinitic varieties.

This paper is structured as follows. In Section 2, we will review several research traditions and how they have traditionally dealt with the construction that we discuss in this paper. In Section 3, we define the causative–resultative construction in Cantonese and discuss how it applies to various subconstructions described in the literature. In Section 4, we discuss the challenges that the Cantonese causative–resultative construction poses for decompositional accounts of argument structure. Section 5 discusses the consequences of the results of Section 3 and Section 4 and expands the analysis to sketch how the constructional schema discussed in Section 3 and Section 4 may apply to other varieties of Chinese and aid in the comparison of differences between varieties. Section 6 concludes.

In contrast to most previous approaches relying on elicited data, our paper uses exclusively natural discourse data to provide a more comprehensive picture, especially as some examples are difficult to understand without context.1 Examples are mainly taken from the Cantonese Universal Dependencies corpus (Wong et al. 2019) and Google Search, which allows us to locate rarer constructions, many of which are crucial to our analysis.2 Universal Dependencies sentences are accompanied by a code of the form [UDXXXX], where XXXX indicates sentence number in the corpus, while those from Google are accompanied by a code [SEAXXX] pointing to the relevant line in a spreadsheet given as Supplementary Materials. Premodern examples are taken from the Chinese Basic Ancient Texts Database (Beijing Erudition Digital Research Center 2017).

2. The Chinese Causative–Resultative Construction in Different Linguistic Traditions

In this section, we will review several research traditions that have examined the causative–resultative construction, particularly in regard to our two main points: whether they posit a CRC-like category, and whether they gravitate towards the holistic or decompositional approach to argument structure for resultatives with two lexical verbs.

2.1. The Chinese Structuralist Tradition

Most linguists in the Chinese structuralist tradition consider V1s in CRCs as the (main) verb. The V2 is known as 補語 bǔyǔ (‘complement’). Originally a translation of complement as used for post-verbal adjectival predicates in English linguistics (e.g., Quirk et al. 1985), it has since expanded to refer to any post-verbal element other than the ‘object’ 賓語 bīnyǔ (Jīn 2009). Thus, bǔyǔ is typically much wider than CRC. Cheung (1972) exemplifies this tradition in Cantonese.

Within this tradition, some proposals have proposed subsets of complements that resemble our CRC. Rejecting bǔyǔ altogether, Jīn (2009) considers ‘complements’ that depend semantically on the subject or object as ‘secondary predicates’, including most constructions under our CRC. Takahashi’s (2021) kinōsei hogo 機能性補語 ‘functional complement’ is even closer; though Takahashi only explicitly includes resultative and directional complements, the definition can easily encompass the other constructions under our CRC. In addition, many Chinese structuralists treat the potential structure (see Section 3.1) as a separate construction, e.g., Zhū’s (1982) ‘potential complex verb-complement structure’. We consider the potential resultative construction a form, i.e., subconstruction, of the CRC, so studies on these constructions are about the CRC, minus the non-potential form.

With some exceptions (e.g., Yuán 2001; Jīn 2009), the Chinese structuralist tradition typically does not regard bǔyǔ as verbs with their own argument structure, thus following the holistic approach.

2.2. The Talmyan Tradition in Worldwide and Sinitic Typology

The Talmyan tradition of verb lexicalization patterns (Talmy 1985) resembles the Chinese structuralist tradition in treating V1 as the main verb and V2 as the satellite.3 Originally focused on motion verbs, it has come to encompass other verbal semantic properties such as cause, result and phase (Talmy 2000), similar to Goldberg and Jackendoff (2004), who regard English directional phrases as resultative. Thus, Talmyan studies frequently regard directional, phase and resultative ‘satellite’ constructions in Chinese as subsets of a single construction resembling our CRC. This approach is adopted in many diachronic and typological approaches in Chinese linguistics (F. Li 1997; Lamarre 2003). Yiu (2005, 2013) represents this tradition in Cantonese, though she does not extend her discussion to non-directional verbs.

As this tradition focuses on the verb, there is no explicit position on argument structure, though most seem to adopt a holistic approach implicitly.

2.3. Contemporary Western Descriptive Traditions

Contemporary Western and Western-influenced descriptions, both generative and functional-cognitive, tend to regard Chinese resultative constructions as compositionally formed VV compounds (e.g., Thompson 1973), whereas phase and directional complements are treated as particles modifying verbs with simple argument structures. Matthews and Yip (2011) influentially adopt this approach. Therefore, there is typically no unified CRC, a position explicitly defended by Huang (2007).

The exact details of the argument structure composition differ. In the ‘morphological’ or ‘lexical’ approach, the thematic role structures of the verbs are composed (e.g., Huang 2007); in the ‘syntactic’ approach, it is grammatical relations, i.e., subject and object (e.g., Her 2007; Matthews 2006; Lau and Lee 2015).4 Some authors seem to compose on multiple levels (e.g., Chow 2011, 2012). Many authors adopt a hybrid of the holistic and decompositional approaches; they supplement the two verbs’ individual argument structures with a construction-level argument structure that is not simply a sum of individual ones. For Chang (2003) and, in Cantonese, Lau and Lee (2021), referents have thematic roles assigned by the individual verbs and event roles in the wider resultative construction. Similarly, Huang (2007) argues that referents are event participants of individual verbs and constructional participants of the wider construction. Similar positions are widespread in 21st-century accounts (e.g., C. Li 2013; Fong 2018). Cheng and Huang (1994) come closest to the holistic approach. For them, arguments of the resultative construction may be required by the composite event structure even if they are not arguments of the individual verbs.

2.4. Serial Verb Constructions in Worldwide Typology

The literature on serial verb constructions in worldwide typology and descriptions based on this tradition often discusses Chinese SVCs. The definitions of SVCs vary, ranging from very wide (e.g., Li and Thompson 2009), moderately wide (Matthews 2006), to fairly narrow (e.g., Chor 2018), but they always require that a clause denote a single event (or some formal correlate of single eventhood), and that the two verbs have no explicit marking of coordination or subordination, which CRCs satisfy. Major works include Aikhenvald and Dixon (2006) (including Matthews’ (2006) chapter on Cantonese), Durie (1997), Bisang (2009) and Haspelmath (2016).

Works in this tradition typically assume each language variety has a finite, countable number of SVCs. They first identify the SVCs in each language and then classify them using formal and semantic properties. Generally, this tradition only recognises a category far wider than the CRC, i.e., the SVC (which encompasses many other construction types in Chinese), plus much narrower constructions such as causative and cause-effect SVCs (which fall under our CRCs in Cantonese).

A recurring theme of this tradition is argument sharing—how arguments simultaneously belong to the argument structures of the two verbs involved. It is frequently noted that the vast majority of SVCs exhibit some argument sharing; in fact, some authors define SVCs to exhibit argument sharing (e.g., Foley and Van Valin 1984) or even a certain type thereof (e.g., Ameka 2005). Thus, this tradition falls squarely into the decompositional approach. Lovestrand (2021) reviews this tradition and additionally details several conceptual difficulties it faces, including some that interlock with our arguments.

3. Properties and Scope of the Causative–Resultative Construction

3.1. Definition and Terminology

Before introducing our constructional schema for the CRC, we define it as follows:

A causative–resultative construction is a construction with two verbal elements that has an affirmative and a negative potential form. In the affirmative potential form, 得 dak1 appears between the two elements, indicating the possibility of the event expressed by the clause. In the negative one, 唔 m4 appears between the two elements, indicating impossibility.5

Example (4) shows, for example, that the construction with the two verbal elements 追 zeoi1 ‘chase’ and 到 dou2 ‘accomplishment’ is a CRC:

(4)	a.	怪	唔	之	得	追	唔	到	女仔	啦！
		gwaai3	m4	zi1	dak1	zeoi1	m4	dou2	neoi5zai2	laa1
		blame	NEG	ASSOC	POT	chase	NEG	DOU	girl	SFP
		‘No wonder why (you) can’t successfully court any girls!‘ [UD605]
	b.	佢	追	得	到	女仔…
		keoi5	zeoi1	dak1	dou2	neoi5zai2
		3sg	chase	POT	DOU	girl
		‘He managed to court a girl successfully…’ [SEA122].

The term causative–resultative construction references the roles of the two verbal elements; one is relevant to the cause of the event, and the other to the result. Our label and definition sidestep two hot-button issues in the literature: wordhood and headedness. We use the word construction in the CxG sense of any conventional form-meaning pairing (Croft 2007) and remain equivocal on whether the two verbal elements form a word, unlike terms such as resultative compound. Moreover, unlike terms such as secondary predicate, complement or satellite, we do not imply that the first verbal element is the head, which is hotly debated (e.g., Tai 2003; C. Li 2009).

The term verbal element indicates only that the form exhibits verb-like properties when used in this construction; the element may not be classified as a verb in traditional word-class systems. In dynamic constructionist approaches (Croft 2007), categories such as ‘noun’ and ‘verb’ are not inherent to the forms, but emergent from patterns in which constructional slots the forms may appear. Nevertheless, the rest of the paper will use verb for brevity.

To qualify as a CRC, both positive and negative potential forms must exist, since, in Cantonese, the affirmative potential form exists outside of the CRC (cf. Chappell and Peyraube 2015).6 In the following example, (a) is not a causative–resultative construction, since it only has the affirmative potential form. Although (c) is an acceptable sentence in Cantonese, it cannot be interpreted as the negative potential form of (a):7

(5)	a.	係	咪	叫	佢	讀書	就	可以	解決	所有	問題?
		hai6	mai6	giu3	keoi5	duk6syu1	zau6	ho2ji5	gaai2kyut3	so2jau5	man6tai4
		COP	NEG.COP	tell	3sg	study	then	can	solve	all	problem
		‘Can telling them to study solve all problems?’ [SEA001]
	b.	諗	落	佢	老竇	叫	得	佢	讀書…
		nam2	lok6	keoi5	lou5dau6	giu3	dak1	keoi5	duk6syu1
		think	down	3sg	dad	tell	POT	3sg	study
		‘Thinking along these lines, if his dad felt like it was necessary to tell him to study …’ [SEA002]
	c.	叫	佢	唔	讀書	就	出	去	搵	工作
		giu3	keoi5	m4	duk6syu1	zau6	coet1	heoi3	wan2	gung1zok3
		tell	3sg	NEG	study	then	go out	go	find	work
		‘to tell him to get a job if he doesn’t study/ *If you can’t manage to tell him to study, go out and get a job.’ [SEA003].

Under this definition, the CRC includes prototypical resultatives in Cantonese, many verb-particle constructions (including resultative, directional and some quantifying particles), some causative constructions, and the comparative construction. Details of each construction will be discussed in Section 3.3. Supplementary Materials discusses constructions that apparently resemble the CRC but are excluded from it.

3.2. The General Schema and Basic Properties

The range of constructions considered CRCs under our definition in Section 3.1 exhibit a cluster of shared syntactic and semantic properties, which makes the CRC a useful category for description. This section summarises these properties, which will be elaborated on in Section 3.1. The CRC’s form may be summarised by the following constructional schema, and by Figure 1 (note that the positions of ArgA and ArgN are flexible rather than fixed in their positions within this schema, as will be described later):

(ArgC) V1 (m4) (dak1) V2 [ArgA]* (V3) (V4) (ArgN)

a. ArgC (ArgCause) is the referent most relevant to the cause of the event;

b. ArgA (ArgAffectee) is the referent most affected by the event, i.e., the affectee;

c. V2, V3 and V4 are verbs relevant to the result of V1, which is the cause of the event (cf. Y. Li 1995). Note that the numbers, such as most conventions, denote the sequence of appearance;

d. ArgN (ArgNeither-Cause-Nor-Affectee) is a referent that is non-cause, non-affectee;

e. Brackets indicate optionality (*note: [] is used around ArgA because it is rarely absent. See Section 4.4.4 for an example where no arguments are present).

For ‘typical’ resultatives, rough equivalents to V1/V2/ArgC/ArgA are widely used in Western descriptive traditions under different names, such as ‘causer’/‘causee’ and ‘initiator’/‘target of action’ (e.g., Y. Li 1995; Cheng and Huang 1994; Her 2007; C. Li 2013; Lau and Lee 2021).8 The applicability of these notions to other constructions under the CRC will be justified in Section 3.3. The following examples illustrate the notation:

(6)	其他	同事	呢，	都係	返	返	埋	去
	[kei4taa1	tung4si6	ne1],	dou1hai6	faan1	faan1	maai4	heoi3
	[ArgC=ArgA]				V1	V2	V3	V4
	other	colleague	TOP	still	return	return	approach	go
	自己	個	座位	嗰	度
	[zi6gei1	go3	zo6wai2	go2	dou6]
	[ArgN]
	REFL	CLF	seat	DEM	place
	‘Other colleagues, please do return to your seats anyway …’ [UD0673].

(7)	阿爺	大家樂	見	唔	倒	我哋	呢
	aa3je4	Daai6gaa1lok6	gin3	m4	dou2	ngo5dei6	ne1
	ArgC		V1	m	V2	ArgA
	grandpa	Café.de.Coral	see	NEG	DOU	us	SFP
	‘Grandpa could not see us at Café de Coral.’ [UD0234]

(8)	就	搵	返	嗰	橛	出	嚟
	zau6	wan2	faan1	[go2	gyut6]	ceot1	lai4
		V1	V2	[ArgA]		V3	V4
	then	search	return	DEM	CLF	exit	come
	‘then find that piece out!’ [UD0186].

In (6), the other colleagues are both the cause and main affectee of the act of returning. The fact that the colleagues have approached and returned to their own seats is an effect of their act of returning. In (7), the would-be cause is the grandfather, and the affectees ‘us’; however, the negated V2 dou2 suggests that V1 was unsuccessful, and thus its effect (i.e., ‘our’ location being known to Grandpa) is absent.

Using the above schema and terminology, some formal properties of the construction noted in the literature (e.g., Cheung 1972; Matthews 2006; Matthews and Yip 2011; Wú 2003, 2005; Yiu 2005, 2013; Yue-Hashimoto 2003; Lai 2018) are as follows:

Modification properties:

a. V1 may take adverbial modifiers or, in directional constructions, aspect markers. Occasionally, V1 may even take additional result verbs before the dak1/m4 (these are not labelled with numbers to ensure that the first result verb after dak1/m4 is always labelled identically);

b. V2 may take aspect markers if V3 and V4 are absent. However, V3/V4 do not, and none of these can be modified by adverbial modifiers.

Word order properties:9

a. ArgC is always placed before the verb;

b. ArgA may be placed in various positions in the clause. The placement of ArgA depends on information structure, pronominal vs. full noun phrase and possibly other factors. The only positions where it never appears are between V1 and dak1, and between m4 and V2. Moreover, outside of path resultatives, it is rarely between V1 and V2 in non-potential forms, and it is never between V1 and V2 when ArgN is present;10

c. ArgN may also be placed before ArgC or between ArgC and V1, but not between V1 and V2;

d. ArgC and ArgA may be the same argument; in that case, the ArgC=ArgA must be placed before V1.

In addition, CRCs also have the following semantic properties:

Relational-semantic properties:

a. The verb series is culturally construed as a coherent event (Matthews 2006; see also Section 5.1);

b. V2+ are semantically dichotomous: either affirmative or negative, with no degrees;

c. Because V1 describes a cause and V2 is relevant to the effect of V1, the CRC must describe a dynamic event rather than a state (see Section 3.3 for how certain CRCs describing states are interpreted metaphorically as dynamic). This is unless m4 or dak1 are present, i.e., the construction is in the potential form, in which case the construction is stative since it expresses potential (or lack thereof);

d. V1 semantically orients to ArgC, V2+ semantically orients to ArgA and/or V1, and all verbs semantically orient to ArgN (see Section 4.3).

Of these properties, (b) follows from the fact that V2+ cannot be modified by degree adverbs. CRC-like constructions with degree adverbs modifying V2 lack negative potential forms, in conflict with the definition in Section 3.1 (see Supplementary Materials for examples). (d) will be explained and justified in Section 4.3.

It should be noted that this schema only describes facts that hold for all subconstructions of the CRC and does not aim to account for all facts of subconstructions, such as why particular V1–V2 combinations are only attested in the potential form, i.e., with dak1 or m4, and never outside of it. This is consistent with a network-based constructionist approach (e.g., Diessel 2019; Croft 2007), where constructions of varying abstraction and schematicity are represented, and concrete lower-level constructions can have specific properties and co-occurrence restrictions not observed with the higher-level construction. We leave it to future research to examine the subconstructions and delineate what exactly leads some elements of the construction to co-occur more frequently with others.

The rest of this section will clarify and justify two aspects of our account diverging from virtually all previous ones: the lack of grammatical relations and the addition of ArgN.

3.2.1. Lack of Grammatical Relations and Voice Operations

This description departs from virtually all previous accounts in making no reference to traditional grammatical relation-related notions such as subject, object or passivation. While ArgC/ArgA-like categories are widely known and used in many Western accounts, those accounts still map them to subject/object and word order is described in subject/object terms. We reject this approach, presenting word-order properties of ArgC/ArgA directly.

In most previous accounts, the ‘subject’ is preverbal, and the ‘object’ is always described as being placed after V1 and usually after V2. However, when the ‘object’ precedes all the verbs, this is either a ‘pseudo-passive’ (sometimes also known as a ‘middle construction’ in studies on Mandarin, e.g., Xiong 2018), with the ‘object’ becoming the ‘subject’ and ‘subject’ suppressed, or ‘object fronting’ (e.g., Lau and Lee 2015). Some examples are (9–11); by Lau and Lee’s definitions, (9) is pseudo-passive while (11) and (10) are object fronting:

(9)	中國	製	玻璃	「打	唔	爛」
	zung1gwok3	zai3	bo1lei1	daa2	m4	laan6
	China	made	glass	hit	NEG	break
	‘The China-made glass could not be broken.’ [SEA125].

(10)	呢	啲	我	搞	掂	啦！
	ni1	di1	ngo5	gaau2	dim6	laa1
	DEM	CLF	1sg	do	well	SFP
	‘Let me finish these up!’ [SEA049].

(11)	佢	晚餐	食	唔	落
	keoi5	maan5caan1	sik6	m4	lok6
	3sg	dinner	eat	NEG	down
	‘She cannot manage to eat dinner (i.e., doesn’t have the appetite).’ [SEA054].

However, in topic-prominent languages such as Chinese, word order can be adequately described in topic-comment terms without grammatical relations, so there is little motivation for fronting or pseudo-passive processes couched in grammatical relations (LaPolla 1990, 1993, 2009; Li and Thompson 1976). In fact, for certain V1-V2 combinations, especially in potential forms, the ‘pseudo-passive’ can be more common than the ‘active’ form. We conducted a mini study of the negative potential form 打唔爛 daa2 m4 laan6 (hit NEG break, ‘cannot break’). Out of 51 results on Google (excluding duplicates and examples from Cantonese textbooks and linguistic research), only 11 (21.7%) turned out to have ArgCs, explicit or implicit (95% Clopper-Pearson CI: (0.113, 0.353)). It is unnecessarily complicated to posit an underlying two-argument clause from which a single-argument pseudo-passive is then derived.

Our schema and properties above can describe these phenomena without any grammatical relations or syntactic operations. ‘Pseudo-passive’11 and ‘object fronting’ simply occur when ArgA appears before V1, as opposed to between V1 and V2 (12a) and after V2 (12b), with ‘pseudo-passives’ additionally requiring that ArgC be semantically absent (not just implicit):

(12)	a.	堆	黑	煙	入面	躝-咗
		deoi1	hak1	jin1	jap6min6	laan1-zo2
		pile	black	smoke	interior	crawl-PFV
		個	成	身	白	色	嘅	人
		go3	seng4	san1	baak6	sik1	ge3	jan4
		CLF	whole	body	white	colour	ASSOC	person
		出	嚟
		ceot1	lai4
		out	come
		‘From inside the pile of black smoke crawled out a person whose entire body was white.’ [SEA060].
	b.	做	咩	嘢	都	好，
		zou6	me1	je5	dou1	hou2,
		do	what	thing	also	good
		都	需要	真心	做	好	佢
		dou1	seoi1jiu3	zan1sam1	zou6	hou2	keoi5
		all	need	sincere	do	well	3sg
		‘Whatever you do, you need to do it well sincerely.’ [SEA053].

Under our account, we may simply state that ArgA precedes the V1 when it is the topic (e.g., (9–10)) or the secondary topic (e.g., (11)), and follows the V1 otherwise (12); no “fronting” operation is needed. An additional advantage of our account is that it predicts the existence of cases where ArgC is absent, but ArgA remains post-V1, which we will show in Section 4.4.1.

One may ask why we do not simply redefine ‘subject’ and ‘object’ as ArgC and ArgA. We return to this question in Section 3.3 when discussing ‘inverted’ resultatives.

3.2.2. The ArgN

Our distinction between ArgA (affectee) and ArgN (non-cause, non-affectee) is fully novel; both, most of the time, are traditionally ‘objects’.12 The ArgN ‘your seats’ in (6) is the goal of a directional verb. More examples are as follows:

(13)	a.	當	你	學	識	釣	魚	後
		dong1	nei5	hok6	sik1	diu3	jyu2	hau6
			ArgC	V1	V2	ArgN
		when	you	learn	know	hook	fish	after
		‘once you have learnt to fish’ [SEA047].
	b.	你	中文	學	唔	好，
		nei5	zung1man2	hok6	m4	hou2
		ArgC	ArgN	V1	m4	V2
		you	Chinese	learn	NEG	good
		唔	代表	你	英文	學	唔	到
		m4	doi6biu2	nei5	jing1man4*2	hok6	m4	dou2
				ArgC	ArgN	V1	m4	V2
		NEG	represent	you	English	learn	NEG	DOU
		‘Just because you can’t learn Chinese well doesn’t mean you won’t manage to learn English.’ [SEA048].
	c.	找	返	你	三十
		zaau2	faan1	nei5	saam1sap6
		V1	V2	ArgA	ArgN
		give change	return	you	thirty
		‘give you thirty dollars back as change’ [UD0050].

Generally, when ArgN is present, ArgA does not appear between the verbs:

(14)	a.	不過	我	用	佢	醫	返	好	隻	手
		bat1gwo3	ngo5	jung6	keoi5	ji1	faan1	hou2	zek3	sau2
		but	1sg	use	3sg	cure	go back	well	CLF	arm
		‘But I used it to cure my arm.’ [SEA155].
	b.	*醫	我	唔	返	隻	手
		ji1	ngo5	m4	faan1	zek3	sau2
		cure	1sg	NEG	go.back	CLF	arm
		‘cannot cure my arm’ (unattested regardless of ArgC presence, ArgA and ArgN identity).

The only exception is gu2 dou2 (‘guess achieve’; Lai 2018):

(15)	估	佢	唔	到	係	乜水
	gu2	keoi5	m4	dou2	[hai6	mat1seoi2]
	V1	ArgA	m4	V2	[ArgN]
	guess	3sg	NEG	DOU	COP	who
	‘cannot guess who they are’ [SEA036].

V2-ArgN sequences are often conventionalised sequences with a non-referential ArgN (note that in (16), the CRC is within a relative clause relativised on the ArgA, and hence we have labelled the lexical head as ArgA):

(16)	a.	有	種	講	唔	出	口	嘅	失望
		jau5	zung2	gong2	m4	ceot1	hau2	ge3	sat1mong6
				V1	m4	V2	ArgN		ArgA
		EXST	type	say	NEG	exit	mouth	ASSOC	disappointment
		‘I have an unspeakable disappointment (lit. a kind of disappointment that cannot be verbalized out of my mouth).’ [SEA037].
	b.	聽	完	返	唔	轉	頭
		teng1	jyun4	faan1	m4	zyun3	tau4
				V1	m4	V2	ArgN
		listen	finish	return	NEG	turn	head
		‘After listening to it, you can’t turn back!’ [SEA038].

The ArgA–ArgN distinction is a matter of pragmatics and construal, and not always predictable from thematic roles. Consider the following pair:

(17)	a.	朋友	話	俾	我	知	水塘	滿	瀉	水
		pang4jau5	waa6	bei2	ngo5	zi1	seoi2tong4	mun5	se2	seoi2
		friend	tell	give	me	know	reservoir	be full	spill	water
		‘My friend informed me that the reservoir filled up, causing water to spill.’ [SEA138].
	b.	如果	你	個	浴缸	啲	水	滿	瀉
		jyu4gwo2	nei5	go3	juk6gong1	di1	seoi2	mun5	se2
		if	you	CLF	bathtub	CLF	water	full	spill
		‘If your bathtub’s water fills up (the bathtub) and spills …’ [SEA139].

(17a) is about the environmental impacts of continuous rain, so the affectee is not the water, but the reservoir that can no longer hold any excess water. However, (17b) is taken from a passage on saving water, so the affectee refers to the water that went to waste.

Sometimes, when a human argument is ArgA, the item they possess is ArgN, whereas if an item appears alone, it would be ArgA:

(18)	a.	跑	渣	馬	畀	人	撞	跌-咗	隻	airpod
		paau2	zaa1	ma5	bei2	jan4	zong6	dit3-zo2	zek3	airpod
		run	Chartered	Marathon	AGT	person	bump	fall-PFV	CLF	airpod
		‘I was running in the Hong Kong Marathon (sponsored by Standard Chartered) when someone bumped into (me), causing my airpods to drop.’ [SEA145].
	b.	但	老人家	唔	小心	撞	跌-咗	個	痰罐
		daan6	lou5jan4gaa1	m4	siu2sam1	zong6	dit3-zo2	go3	taam4gun3
		but	elder	NEG	careful	hit	fall-PFV	CLF	spittoon
		‘But the elder carelessly knocked down the spittoon.’ [SEA146].

The fact that the space between the two verbs is reserved for ArgAs and disallowed for ArgNs can be seen in the following contrast:

(19)	a.	五十	盒	食	佢	唔	死	仲	奇怪
		ng5sap6	hap6	sik6	keoi5	m4	sei2	zung6	kei4gwaai3
		fifty	box	eat	3sg	NEG	die	even more	strange
		‘It would be even stranger if fifty boxes aren’t enough to make her die from eating.’ [SEA148]
	b.	唯有	同	個	通粉	say	sorry ，
		wai4jau5	tung4	go3	tung1fan2	say	sorry
		can only	to	CLF	macaroni	say	sorry
		因為	食	佢	唔	晒
		jan1wai6	sik6	keoi5	m4	saai3
		because	eat	3sg	NEG	all
		‘All I could do was to say sorry to the macaroni, because I could not eat it all.’ [SEA147]

In (a), the affectee is the eater since the effect is about death (an ‘inverted resultative’—see Section 3.3.1). In (b), the affectee is the macaroni, since the V2 quantifies the food (see Section 3.3.4). Never do we find cases such as (19a) where the food is between the verbs, or cases such as (19b) where the eater is.

3.3. Subconstructions of the CRC

This section will discuss seven subconstructions of the CRC, particularly with respect to the properties discussed in Section 3.2. Section 3.3.1 will discuss ‘typical’ resultatives with two lexical verbs, focusing on ‘inverted’ constructions, which we account for much more readily than traditional approaches. The rest of the section examines other subconstructions typically not analysed together with CRCs and shows that they are compatible with the definitions and properties in Section 3.1 and Section 3.2, which shows the value of our CRC category; it allows us to draw generalisations about a broad class of constructions.

3.3.1. Regular and ‘Inverted’ Resultative Constructions

Prototypical resultatives with two verbs that can stand alone as verbs in monoverbal predicates, such as (1), are part of the CRC. However, one specific type of resultative is relatively challenging to describe in traditional accounts using grammatical relations. Those are traditionally called ‘causative’ (Cheng et al. 1997; Lau and Lee 2015) or ‘inverted’/‘flip-flop’ resultative constructions (e.g., Matthews and Yip 2011, p. 177; Chow 2011, 2012). In either case, the original ‘subject’ of the V1 is typically said to become the ‘object’ of the resultative construction, and the original ‘object’ of the V2 becomes the ‘subject’ of the resultative construction:

(20)	都係	貪	外國	牌子
	dou1hai6	taam1	ngoi6gwok3	paai4zi2
	also	greedy	foreign country	brand
	冇	咁	易	食	死	人
	mou5	gam3	ji6	sik6	sei2	jan4
	NEG	so	easy	smoke	die	person
	‘I’m just taking advantage of the fact that foreign brands don’t smoke people to death so easily.’ [SEA031].

Here, the ‘original subject’ is ‘foreign brands (of electronic cigarettes)’, whereas the ‘original object’ is ‘people’; their roles then flip to become object and subject.

As our account does not assume any grammatical relations associated with particular syntactic positions, we do not need an inversion process.13 This has multiple advantages over traditional inversion-based approaches, which face several empirical difficulties.

Firstly, ‘inverted’ resultatives have no actual ‘non-inverted’ equivalent. For example, (20) cannot be expressed in an un-inverted way:

(21)	*你	食	死-咗	煙
	nei5	sik6	sei2-zo2	jin1
	2sg	smoke	die-PFV	tobacco
	‘smoke tobacco to death’ [unattested regardless of ArgC].

The idea in (21) can only be expressed with a verb-copying construction, thereby distributing the ‘tobacco’ argument to a regular single-verb clause:

(22)	食	煙	食	死	你
	sik6	jin1	sik6	sei2	nei5
	smoke	tobacco	smoke	die	you
	‘smoking tobacco, smoking you to death’ [SEA032]

In the above example, the ‘you’ and ‘tobacco’ can at least be the agent and patient arguments of ‘eat’ in monoverbal contexts. There are cases where even this does not work:

(23)	鉛芯筆	寫	壞	手勢。
	jyun4sam1bat1	se2	waai6	sau2sai3
	mechanical pencil	write	bad	gesture
	‘Writing with mechanical pencils makes your writing gestures bad.’ [SEA112]

In monoverbal contexts, ‘gesture’ and ‘mechanical pencil’ cannot be the agent and patient arguments of ‘write’—one is a manner, and the other is an instrument of writing.14

In our account, (20) and (23) simply have ArgC (foreign brands, mechanical pencils) as the cause and ArgA (people, gestures) as the affectee, similar to any other causative–resultative construction. There is no requirement that ArgC be agentive or ArgA non-agentive. In addition, (21) is unattested because the tobacco is not affected by the event and hence cannot be ArgA, nor can it be ArgN since death does not semantically orient to it (Section 4.3.3).

Second, when we look at actual discourse examples, we often find cases with no ArgC, either explicit or context-implied, and thus nothing for the ArgA to ‘flip’ with:15

(24)	唔	好	咩	都	想-着	食	好	唔	好？
	m4	hou2	me1	dou1	soeng2-zoek6	sik6	hou2	m4	hou2
	NEG	good	what	no matter	think-CONT	eat	good	NEG	good
	大哥，	咩	都	食	好	易	食	死	人	㗎。
	daai6go1	me1	dou1	sik6	hou2	ji6	sik6	sei2	jan4	gaa3
	big brother	what	no matter	eat	very	easy	eat	die	person	SFP
	‘Don’t just think of eating all the time, okay? Dude, if you eat anything and everything, you’ll easily eat yourself to death.’ [SEA030].16

These are unproblematic in our account; there is simply a postverbal ArgA with no ArgC.

Finally, the traditional account fails to explain the absence of ‘inverted’ constructions such as (25), derived from (13b):

(25)	~[英文]	學	你	唔	到
	jing1man4*2	hok6	nei5	m4	dou2
	English	learn	2sg	NEG	DOU
	‘*You cannot manage to learn English.’17

Under traditional approaches, it is unclear why jing1man4*2 ‘English’, an object in sentences such as (13b), cannot be ‘inverted’ to become the subject of the clause. However, in our account, ‘English’ is not a cause in this situation, and hence cannot be ArgC.

One may argue that we may still define our ArgC and ArgA as language-specific, construction-specific ‘Resultative Subject’ and ‘Resultative Object’, thus retaining continuity to traditional categories. However, we believe these labels are misleading, given how divergent they are from the definition of ‘subject’ and ‘object’ in typology and other descriptive traditions. Generally, transitive subjects are more agentive than objects, and common criteria for relative agency (e.g., Witzlack-Makarevich and Bickel 2013) clearly show that the ArgC is less agentive in ‘inverted’ constructions. ArgC and ArgA do not have agency-related baggage.

3.3.2. Most Resultative and Phase Particle Constructions

Matthews and Yip (2011, p. 243) list a number of resultative particles in Cantonese, such as 好 hou2 ‘completion’ (glossed ‘good’ in this paper), 掂 dim6 ‘decisively’, 到 dou2 ‘accomplishment’ (glossed DOU in this paper) and 完 jyun4 ‘to the end’. For Matthews and Yip (2011, p. 245), the possibility of the potential constructions is a defining characteristic of resultative particles, so constructions involving these particles count as CRCs in our definition:

(26)	a.	睇	到	睇	唔	到	唔	係	靠	彩數
		tai2	dou2	tai2	m4	dou2	m4	hai6	kaau3	coi2sou3
		see	DOU	see	NEG	DOU	NEG	COP	depend	luck
		‘Whether you can see it or not does not depend on luck.’ [SEA006].
	b.	睇	得	到	外邊	係	金	黃	色
		tai2	dak1	dou2	ngoi6bin6	hai6	gam1	wong4	sik1
		see	POT	DOU	outside	COP	gold	yellow	colour
		‘I can see that the exterior is golden yellow.’ [SEA007].

The adversative particle 親 can1, sometimes regarded as resultative (Gu and Yip 2004) and sometimes a sui generis particle class (Matthews and Yip 2011), also participates in CRCs:18

(27)	a.	餓	佢	唔	親	就	得	啦
		ngo6	keoi5	m4	can1	zau6	dak1	laa1
		hungry	3sg	NEG	ADV	then	okay	SFP
		‘As long as he doesn’t go hungry (lit. he’s not so hungry that it hurts him), it’s fine!’ [SEA008].
	b.	點	會	餓	得	親	啊
		dim2	wui5	ngo6	dak1	can1	aa3
		how	IRR	hungry	POT	ADV	SFP
		‘How would one manage to go hungry?’ [SEA009].

Resultative particles not listed by Matthews and Yip include 著 zoek6 ‘asleep/alight/etc.,’ 通 tung1 ‘get through’, 切 cit3 ‘on time’, 真 zan1 ‘to achieve better result’, etc.

Many resultative particles fall in what Chinese structuralists such as Cheung (1972) classify as phase complements, which express the degree to which the V1 was actualized rather than the actual result and might prima facie constitute exceptions to our statement that V2 is relevant to the result. Some of these are straightforwardly result-expressing if we refine our understanding of the V1. Take 着 zoek6:

(28)	點解	我	飲	完	酒
	dim2gaai2	ngo5	jam2	jyun4	zau2
	why	1sg	drink	finish	alcohol
	會	易	啲	瞓	得	着	嘅?
	wui5	ji6	di1	fan3	dak1	zoek6	ge2
	IRR	easy	a bit	sleep	POT	ZOEK	SFP
	‘Why can I go to sleep more easily after having drunk alcohol?’ [SEA044].

At first, this seems to constitute a counterexample to our claim that V2 is always relevant to the result since zoek6 tells us that the action of sleeping is ongoing, not effects of sleep such as gaining energy. However, we have evidence that fan3 in such contexts has a meaning closer to ‘get oneself to sleep’. Consider the following example:

(29)	又	失眠，	瞓	極	都	瞓	唔	著
	jau6	sat1min4	fan3	gik6	dou1	fan3	m4	zoek6
	again	insomnia	sleep	limit	still	sleep	NEG	ZOEK
	‘I have insomnia again—no matter how hard I (try to) sleep, I cannot fall asleep.’ [SEA045].

The first fan3 is not negated even though the writer never went asleep. Thus, reaching the state of being asleep is still the result of 瞓 fan3. This is an example of coercion: a verb gains a meaning (here, lack of actualisation) by virtue of being in a constructional environment. This is similar to F. Li’s (1997) description of the word 殺 shā ‘kill’ in Mandarin, which originally implied the death of the patient in Old Chinese, but was gradually ‘leached’ of this meaning by participating in the CRC 殺死 shā sǐ.

Other phase complements denote that the event has reached a phase that produces certain results, such as 到 dou2 and 成 seng4 ‘succeed’. Recall that in the constructional schema, we only define V2 to be related to the result, not necessarily the result itself; such phase complements are good examples. For example, dou2 and seng4 indicate that V1 was successfully performed, which is a precondition for the event’s result to take place:

(30)	做	唔	成	戀人	做	唔	做	到	朋友？
	zou6	m4	seng4	lyun2jan4	zou6	m4	zou6	dou2	pang4jau5
	do	NEG	succeed	lover	do	NEG	do	DOU	friend
	‘If we cannot be lovers, can we be friends?’ [SEA161].

Quite often, V1 is static, in which cases the ‘result’ is a standard that the property denoted by the stative verb achieves or not. In these constructions, V1 and V2 satisfy the semantic requirement metaphorically. If we think of the static verb as dynamic, e.g., think of ‘red’ as ‘becoming red’, then the standard expressed by V2 can be conceptualised as the result of this becoming—another example of coercion. We see this in phase complements and even conventional resultatives:

(31)	a.	男朋友	話	我	襯	佢	唔	起
		naam4pang4jau5	waa6	ngo5	can3	keoi5	m4	hei2
		boyfriend	say	1sg	suit	3sg	NEG	HEI
		‘My boyfriend said I was not in his league.’ [SEA046].
	b.	開心	死	喇！
		hoi1sam1	sei2	laa5
		happy	die	SFP
		‘I am so happy I could die!’ [SEA082].

One may ask why dou2 is seen as a V2, but the affirmative potential form marker dak1 is not seen as a V2 but is treated as a syntactic marker in the construction. Syntactically, one cannot have a construction that consists of V1 + m4 + dak1 but no other V2 (though one could in Middle Chinese), e.g., *做唔得朋友 zou6 m4 dak1 pang4jau6 ‘cannot be friends’. Semantically, when dak1 occurs with a V2, it does not convey the success of the implied end-state of V1 but states V2 as possible. For example, 食得死 sik6 dak1 sei2 ‘eat DAK die’ means that the ArgC can kill people who eat it. Dak1 here indicates the possibility of dying, not of completing the process of eating; dou2 cannot be used this way. Finally, dak1 implies potentiality, whereas dou2 does not.

3.3.3. The Comparative Construction

The comparative construction in Cantonese, which uses the V2 過 gwo3 ‘exceed’, also satisfies our CRC definition:

(32)	a.	有	咩	開心	得	過	買	得	抵？
		jau5	me1	hoi1sam1	dak1	gwo3	maai5	dak1	dai2
		EXST	what	happy	POT	COMP	buy	DAK	good value
		‘What can be happier than buying (things) with good value?’ [SEA034].
	b.	點	食	都	開心	唔	過
		dim2	sik6	dou1	hoi1sam1	m4	gwo3
		how	eat	FOC	happy	NEG	COMP
		同	一	大	班	同事	食
		tung4	jat1	daai6	baan1	tung4si6	sik6
		with	one	big	group	colleague	eat
		‘No matter how (we) eat, it can’t be as joyous as eating with a big group of colleagues.’ [SEA035].

The comparative construction can be seen as a special case of the resultative, again with the V1 coerced from stative to dynamic. This is a metaphorical extension of gwo3 ‘pass’ as a directional particle; however, the comparative construction patterns with non-directional resultatives syntactically (V1s cannot take aspect markers, and ArgA does not fall between V1 and V2 outside potential forms).

3.3.4. Some Quantifying Particle Constructions

Two of Matthews and Yip’s (2011, p. 243) quantifying particles, 哂 saai3 ‘all’ and 埋 maai4 ‘also’, participate in CRCs. Examples with saai3 are shown below:

(33)	a.	點解	啲	tasks	做	極	都	做	唔	哂?
		dim2gaai2	di1	taas1	zou6	gik6	dou1	zou6	m4	saai3
		why	CLF	task	do	extreme	also	do	NEG	all
		‘Why can’t I finish all the tasks no matter how much I do them?’ [SEA010].
	b.	唔	會	話	一	個	人
		m4	wui5	waa6	jat1	go3	jan4
		NEG	IRR	say	one	CLF	person
		做	得	哂	所有	野	既
		zou6	dak1	saai3	so2jau5	je5	ge3
		do	POT	all	all	thing	SFP
		‘It’s not like one person can finish all the stuff.’ [SEA011]						.

Other V2s include夠 gau3 ‘enough’, 足 zuk1 ‘enough’ and 齊 cai4 ‘complete’.

These quantifying particles may quantify different elements of a sentence but are, in all cases, relevant to the result. For example, when it quantifies an argument, that argument is always the affected ArgA, such as ‘tasks’ in (33a). When it quantifies the V1, this quantification gives information about the result. An example is (34), again with the metaphorical extension mentioned for resultatives and comparatives:

(34)	兩	個	仔	都	似	哂	阿媽
	loeng5	go3	zai2	dou1	ci5	saai3	aa3maa1
	two	CLF	son	all	resemble	all	mother
	‘Both sons look completely like their mother.’ [SEA126] (cf. Wong 2008, (21)).

Thus, quantifying results are result-related either way.

3.3.5. Some Causatives

Causatives in Matthews and Yip’s (2011) sense have V1s roughly translating to ‘cause’ or ‘make’ in English.19 Some causatives involving 整 zing2 and 搞 gaau2 fall into this category:

(35)	a.	老豆	咁	高	lv	都	整	得	喊
		lou5dau6	gam3	gou1	le1fou2	dou1	zing2	dak1	haam3
		dad	so	high	level	even	CAUS	POT	cry
		‘You could even make someone as high-level (i.e., hard to touch) as your Dad cry?’ [SEA020].
	b.	無論	歌	定	戲	都	整	唔	喊	我
		mou4leon6	go1	ding6	hei3	dou1	zing2	m4	haam3	ngo5
		no matter	song	or	film	also	CAUS	NEG	cry	1sg
		‘Neither songs nor films can make me cry.’ [SEA021].

Other causatives take a full clausal complement rather than a V2 and do not fall into the CRC (Supplementary Materials).

3.3.6. Some Manner Constructions

Some manner ‘complements’ have both potential forms and qualify as CRCs, generally when the manner can be characterised as the goal of the V1:

(36)	a.	感覺	點	用	力	都	跑	唔	快
		gam2gok3	dim2	jung6	lik6	dou1	paau2	m4	faai3
		feel	how	use	force	still	run	NEG	fast
		‘I feel that no matter how much energy I use, I still can’t run quickly.’ [SEA126].
	b.	你	跳	唔	齊	跳	唔	靚
		nei5	tiu3	m4	cai4	tiu3	m4	leng3
		2sg	jump	NEG	in sync	jump	NEG	pretty
		就	無	得	出道
		zau6	mou5	dai1	ceot1dou6
		then	NEG	can	debut
		‘You can’t debut if you can’t dance in sync and beautifully’. [SEA127].

Here, ‘fast’ is the writer’s goal when running, and ‘in sync’ and ‘pretty’ are the dancers’ goals when dancing. The lack of such manners can be conceptualised as the failed realisation of the desired result. The same construction in (36a) is unattested with 煮 zyu2, presumably because speed is seldom seen as a desired outcome of cooking.

3.3.7. Most Directional Particle Constructions

Cantonese verbs may be accompanied by up to three directional particles. The inventory of directional particles is in

Table 1. Directional particle paradigms in Cantonese.

a. Return	b. Location-Oriented	c. Speaker-Oriented
返 faan1 ‘return’	上 soeng5 ‘ascend’ 落 lok6 ‘descend’ 出 ceot1 ‘exit 入 jap6 ‘enter’ 開 hoi1 ‘depart’ 埋 maai4 ‘approach’ 過 gwo3 ‘pass’ 起 hei2 ‘rise’ 到 dou3 ‘arrive’	嚟 lai4 ‘come’ 去 heoi3 ‘go’

(Cheung 1972; Yiu 2005, 2013).

Here are some examples:

(37)	a.	會	唔	會	走	得	返	埋	一齊？
		wui5	m4	wui5	zau2	dak1	faan1	maai4	jat1cai4
		IRR	NEG	IRR	walk	POT	return	approach	together
		‘Will they be able to get back together?’ (i.e., be a couple again) [SEA013].
	b.	心痛	得	喫	唔	落	去
		sam1tung3	dak1	jaak3	m4	lok6	heoi3
		heart bleed	POT	eat	NEG	descend	go
		‘My heart bled so much that I could not continue eating.’ (Context: The meal cost over $100.) [SEA014].

Though Yiu (2013) writes that the combination 返起嚟 faan1 hei2 lai4 (return rise come) has no potential form, we do find examples on the Internet:

(38)	治療	中	企	得	返	起	嚟
	zi6liu4	zung1	kei5	dak1	faan1	hei5	lai6
	therapy	duration	stand	POT	return	rise	come
	‘to be able to stand back up during therapy [SEA012].

We thus still consider constructions with faan1 hei2 lai4 to be CRCs.

Directional particles, such as those in (37a), though literally indicating direction (the celebrities metaphorically approaching each other), are strongly indicative of the resultant end state—the two celebrities becoming a couple again, and hence still highly relevant to the result. As Yiu (2013) shows, this is the case for physical directions as well. She gives the following example:

(39)	嗰	啲	嘢	放	落	檯	上
	go2	di1	je5	fong3	lok6	toi2	soeng6
	that	CL	thing	put	descend	table	on
	‘to drop those things down on the table’ (O’Melia 1941, cited by Yiu (2013, p. 171)).

Notice that although lok6 ‘descend’ is still used, the ArgN is the destination of descent.

Formally, directional particle constructions constitute a clear subclass of the CRC, as they differ in at least three respects mentioned in Section 3.2: aspect marker placement, ArgA placement and ArgN appearance. Nevertheless, there is still good reason to treat it with other CRCs under one construction. For example, under accounts that treat the directionals as modifiers to the verbal structure, it is unclear how to account for examples such as the following:

(40)	啲	香料	爆-咗	啲	味	出	嚟
	di1	hoeng1liu2	baau3-zo2	di1	mei6	ceot1	lai4
	CLF	spice	burst-PFV	CLF	aroma	go out	come
	‘Some aroma came out as the spice burst.’ [SEA144].

In the absence of directionals, 爆 baau3 ‘burst’ generally takes the thing that burst as its sole argument, so it is not clear what the spices are doing. However, if we consider this a CRC, this can be easily explained; the spice is the ArgC, and the aroma is the ArgA.

3.4. Interim Conclusion

This section gave a clear description of the sense and extension of the causative–resultative construction. The CRC label is shown to be useful because its members share numerous semantic and syntactic properties. Furthermore, directional and non-directional CRCs may be seen as two major subgroups under the CRC, since directional CRCs have clearly distinctive properties within CRCs. The next section will zero in on one aspect of our schema, the holistic approach to argument structure, which is a departure from most Western approaches, and elaborate on the semantic relationships briefly touched on in Section 4.2.

4. The Holistic Approach to Argument Structure

As mentioned above, most Western approaches approach ‘regular’ resultatives with two lexical verbs, such as those discussed in Section 3.3, decompositionally. Clauses or argument structures of individual verbs are thought to combine to form the entire construction’s argument structure:

(41)	佢	剪	短-咗	頭髮
	koei5	zin2	dyun2-zo2	tau4faat3
	3sg	cut	short-PFV	hair
	‘He cut his hair short’ (Lau and Lee 2021).

Lau and Lee (2021) break this construction down as follows:

NP1	+	V1	+	NP2
佢		剪		頭髮
3sg		cut		hair
+
NP2	+	V2
頭髮		短-咗
hair		short-PFV

This approach of decomposing the construction into two monoverbal constructions works well where the following properties coincide:

• A verb is attracted to the V1 slot of the CRC as well as the sole V slot of either a simple transitive or an intransitive construction (not both). Moreover, one or both of the arguments of the CRC are attracted to this verb in both the CRC and the (in)transitive construction, and the semantic relation between the verb and the arguments is the same in the CRC and in the (in)transitive construction;
• The above also applies to the V2 slot;
• The CRC has no arguments other than those in 1–2;
• The V1 and V2 have the same meaning in the CRC as in monoverbal clauses.

However, there are empirically many CRCs in Cantonese that do not fulfil these criteria (to be illustrated in Section 4.1.1, Section 4.1.2, Section 4.1.3 and Section 4.1.4). In contrast, we adopt a holistic approach with no composition (or decomposition). This section will first discuss various empirical difficulties with the decompositional approach (Section 4.1), explain how our approach dispels these problems (Section 4.2), and then explain how our account can explain and extend previous findings about CRCs couched in decompositional terms (Section 4.3 and Section 4.4). Figure 2a,b show a side-by-side comparison of the decompositional and holistic approaches as slotboards, and how different CRCs are respectively analysed in the two approaches.

4.1. Problems with the Decompositional Approach

4.1.1. Ambitransitive Verbs

Cantonese has many ambitransitive verbs, which may take one or two arguments. When they are involved in CRCs, it is indeterminate whether the ‘pre-composition’ clause contains one or two arguments. For example, consider 濕 sap1 ‘wet’:

(42)	a.	媽咪	同	我地	兩	個	淋	濕-咗	個	身	先
		maa1mi4	tung4	ngo5dei6	loeng5	go3	lam4	sap1-zo2	go3	san1	sin1
		mummy	BEN	1pl	two	CLF	soak	wet-PFV	CLF	body	first
		‘Mummy drenched both of our bodies first.’ [SEA074].
	b.	書包	濕-咗，	個	身	濕-咗	一	半
		syu1baau1	sap1-zo2	go3	san1	sap1-zo2	jat1	bun3
		bookbag	wet-PFV	CLF	body	wet-PFV	one	half
		‘The bookbag got wet, and my body was half wet.’ [SEA075].
	c.	無幾耐，	呢	個	傻佬	已經	濕-咗	身	喇
		mou5gei2noi1	ni1	go3	so4lou2	ji5ging2	sap1-zo2	san1	laa3
		before long	DEM	CLF	fool	already	wet-PFV	body	SFP
		‘Before long, this fool had already wet his body.’ [SEA076].

In (42a), 淋 lam4 is transitive, so in the decompositional approach, it must be that 媽咪 maa1mi4 ‘mummy’ is its ‘subject’ and 個身 go3 san1 ‘the bodies’ is its ‘object’. However, the ambitransitive sap1 is less clear. It may be intransitive with 個身 go3 san1 as its ‘subject’ (cf. 42b), or transitive with maa1mi4 ‘mum’ as its ‘subject’ and go3 san1 ‘the bodies’ as its ‘object’ (cf. 42c). Most accounts assume that sap1 is intransitive (e.g., Matthews 2006; Lau and Lee 2021), but without arguing against the alternative.20 On the other hand, the (ambi)transitivity of V2 sap1 does not interfere with our holistic account; the non-verb elements are simply arguments of the entire CRC construction. Figure 3 shows another side-by-side comparison of both approaches.

In our approach, the ambitransitive nature of this and other verbs is unproblematic because it is unnecessary to ’pick and choose’ the transitivity of the V2.

4.1.2. Idiomatic Constructions with No Corresponding Simple Clauses

In (43), two idiomatic constructions are presented with V1 睇 tai2 and ArgN 眼 ngaan5, meaning that ArgC harbours some resentment towards ArgA.21 No second clause may be separated with the V2 as the verb:

(43)

a.

有

好

多

事

我

都

睇

佢

唔

過

眼。

jau5

hou2

do1

si6

ngo5

dou1

tai2

keoi5

m4

gwo3

ngaan5

EXST

very

many

matter

1sg

all

see

3sg

NEG

pass

eye

‘there are many matters on which I could not bear to see her.’ (‘Many matters’ is a hanging topic). [SEA058].

b.

點

會

睇

得

佢

順

眼

直頭

想

即時

處決

dim2

wui5

tai2

dak1

keoi5

seon6

ngaan5

zik6tau4

soeng2

zik1si4

cyu5kyut3

how

IRR

look

POT

3sg

be along

eye

in fact

want

at once

execute

‘How would I bear looking at it? In fact, I would want to execute it at once.’ (humorous sentence on ugly fingernails). [SEA059].

c.

~*佢

過

/

順

眼

keoi5

gwo3

/

seon6

ngaan5

3sg

pass

/

be along

eye

[unattested, meaning unclear]

Again, complications arising from the treatment of V2 in the decompositional approach become irrelevant when considering the arguments at the holistic CRC level, as seen in Figure 4.

There are also some idiomatic constructions where one of the verbs has a metaphorical meaning restricted to the CRC. Similar cases have been noted in SVC typology, where a verb has a different meaning inside and outside of biverbal clauses (e.g., Enfield 2009; Lovestrand 2021, pp. 9–10). Consider the following:

(44)	嗰	度	係	私人	地方，	告	佢	唔	入	嘅	喎
	go2	dou6	hai6	si1jan4	dei6fong1	gou3	keoi5	m4	jap6	ge3	wo3
	DEM	place	COP	private	place	sue	3sg	NEG	enter	SFP	SFP
	‘That’s a private place—you can’t sue him successfully!’ [SEA061].

Though one can say 佢唔入 keoi5 m4 jap6 ‘he does not enter’, this does not mean ‘he cannot be sued successfully.’ Another example is as follows:

(45)	但係	最終	你	都	係	走	唔	甩
	daan6hai6	zeoi3zung1	nei5	dou1	hai6	zau2	m4	lat1
	but	eventually	2sg	still	COP	run	NEG	loose
	‘But eventually you can’t escape.’ [SEA095].

In monoverbal clauses, 甩 lat1 ‘loose’ either refers to an inanimate object (e.g., chips of paint) coming loose in intransitive clauses, or dumping a romantic partner in transitive ones. It cannot refer to people being free in intransitive clauses. We searched for 佢甩咗 keoi5 lat1-zo2 (3sg loose-PFV) online, and there were no examples where the ‘loose’ thing was animate, despite the third-person pronoun keoi5 skewing animate in Cantonese, especially outside of ‘objects’ (Matthews and Yip 2011, p. 95).

A particularly extreme example is (46). This special phrase means that ArgC cannot do anything to affect ArgA, with minimal semantic contribution from the lexemes in the two verbal slots:

(46)	連	阿sir	都	奈	佢	唔	何
	lin4	aa3soe4	dou1	noi6	keoi5	m4	ho4
	even	male police officer	also	NOI	3sg	NEG	HO
	‘Even the male police officers could not do anything about him.’ [SEA072].

Here, 奈 noi6 and 何 ho4 have no meaning on their own. This construction is derived from an Old Chinese construction, where奈何 on its own means ‘what can we do’, and when included in negative constructions with two arguments, the meaning is similar to the Cantonese CRC. The potential form was not yet developed in Old Chinese, and the negation would come before the 奈:

(47)	無	奈	秦	何	矣
	NEG.EXST	NOI	Qín	HO	SFP
	‘They (Hán) will not be able to do anything about Qín.’ (lit. ‘There will be nothing that Hán can do about Qín.’) (Zhàn Guó Cè 4.10).

This original construction was seemingly forced into the mould of the Cantonese CRC, with the 奈 noi6 taken to mean any generic action, and the 何 ho4 any generic desired result. This coercion can be elegantly described with our holistic CRC schema, which does not require that noi6 and ho4 be usable in monoverbal constructions.

4.1.3. Semantic Dependence between Verbs

Some more schematic and productive constructions are not idiomatic, but because the interpretation of the two verbs is very interdependent, they cannot be readily separated, even though the V1 and V2 can both normally appear in single-verb constructions. The causative construction (Section 3.3.5) is a good example, but this problem appears in ‘typical’ resultatives too.

Consider the construction with the V2 齊 cai4. It can be paired with different verbs and optionally the quantifying 哂 saai3 as V3; the resultant meaning is that the action performed by V1 was performed on all of the ArgA (a–b):

(48)	a.	今	次	仲	唔
		gam1	ci3	zung6	m4
		this	time	still	NEG
		食	齊	哂	啲	口味？
		sik6	cai4	saai3	di1	hau2mei6
		eat	complete	all	CLF	flavour
		‘Won’t you taste all the flavours this time?’ (i.e., They finally get to taste all the flavours this time round). [SEA064].
	b.	今年	先	識	自動自覺	做	齊	啲	功課
		gam1nin4*2	sin1	sik1	zi6dung6zi6gok3	zou6	cai4	di1	gung1fo3
		this year	only	know	self-consciously	do	complete	DEM	homework
		‘Only this year did he start completing all his homework self-consciously.’ [SEA062]
	c.	當	佢哋	嘅	食物	齊	哂
		dong1	keoi5dei6	ge3	sik6mat6	cai4	saai3
		when	3pl	ASSOC	food	complete	all
		‘When all of their food was there (i.e., was served).’ [SEA065].
	d.	佢	係	齊	晒	六	粒	石	喎
		keoi5	hai6	cai4	saai3	luk6	lap1	sek6	wo3
		3sg	COP	complete	all	six	CLF	stone	SFP
		‘But he (Thanos) has all six (stones).’ [SEA156].

Although 齊cai4 can be used as a V1 with no verb before it and with saai3 as V2, as in (48c), the meaning is not consistent with a story where sentences such as (48ab) contain such a clause, because cai4 only indicates that the ArgA all exists, not that an action was performed on all of ArgA. In the intransitive (48c), it is not that an action (say, eating) was completed on all of the food; rather, the sentence means that all the food exists (i.e., was served). Similarly, in the transitive (48d), it only means that Thanos has all six stones in his possession and cannot mean that he performed some action on all six stones.22 Thus, we cannot say that sentences such as (a–b) contain a clause with cai4 independent of the V1s 食 sik6 and 做 zou6.

A similar argument may be applied to 完 jyun4:

(49)	a.	食	完	生果	成日	唔	舒服？
		sik6	jyun4	saang1gwo2	seng4jat6	m4	syu1fuk6
		eat	finish	fruit	always	NEG	comfortable
		‘Always feeling unwell after eating fruit?’ [SEA066].
	b.	~*生果	完
		saang1gwo2	jyun4
		fruit	finish
		‘The fruit has been eaten (lit. the fruit is over)’ [unattested].

One may classify these V2s as particles, declare them non-verbal, and exclude them from consideration. Yet there are also cases where the interpretation of V1 depends on V2. Consider the verb 放 fong3 ‘put’:

(50)	成	程	車	都	要	屈-住
	seng4	cing4	ce1	dou1	jiu3	wat1-zyu6
	whole	CLF	car	all	need	bend-CONT
	/	放	歪	對	腳
	waak6ze2	fong3	me2	deoi3	goek3
	or	put	in unnatural position	CLF	leg
	‘(They) had to keep bending their legs or putting them in an unnatural position throughout the ride.’ [SEA077].

While 放 fong3 can be used in a single-verb construction, it is unnatural to use it in such a construction without some specification of manner or position, whether that specification comes from a V2 or not. We looked up the phrase ‘放對腳’ (put CLF foot) in Google, and in all of the cases, there is a manner or position specified.23 Thus, we cannot extract a monocausal fong3 clause from this CRC.

Another example, also observed in Igbo (Lord 1975), is with the verb 打 daa2 ‘hit’, which does not apply to inanimate objects outside of CRCs:

(51)	媽媽	唔	小心	打	爛-咗	隻	碟。
	maa4maa1	m4	siu2sam1	daa2	laan6-zo2	zek3	dip2
	mum	NEG	careful	hit	break-PFV	CLF	plate
	‘Mum carelessly hit and broke the plate.’ [SEA154].

In monoverbal constructions, daa2 ‘hit’ does not apply to plates; the only examples we found on the Internet of 打隻碟 daa2 zek3 dip2 ‘hit CLF plate’ were about burning CDs.

4.1.4. Missing and ‘Wrong’ Arguments

There are also problems with ‘missing’ arguments. Consider the following example:

(52)	食	飽	先	有	力	減	肥
	sik6	baau2	sin1	jau5	lik6	gaam2	fei4
	eat	full	only	have	strength	reduce	fat
	‘I don’t have the strength to lose weight until I have eaten myself full!’ [SEA092].

In (52), there is no specific food implied. Yet in Cantonese, 食 sik6 ‘eat’ is strictly transitive (Matthews 2006). Even if no specific patient is explicitly present, there must be a contextually inferable one. So, the V1 is ‘missing’ a patient in the decompositional approach. Again, this issue can be dealt with by taking the holistic approach, as shown in Figure 5.

Sometimes, V2s may have no arguments at all:

(53)	電話	打	唔	通
	din6waa2	daa2	m4	tung1
	phone	hit	NEG	come through
	‘The telephone (number) could not be reached.’ [SEA151].

In monoverbal clauses, 通 tung1 takes a channel (e.g., intestines, pipes) as its argument, but there is nothing channel-like in this example.

Missing arguments are not unknown in SVC literature (e.g., Lord 1975, pp. 33–34; Aikhenvald and Dixon 2006, p. 13), and decompositional approaches can allow for argument ‘suppression’ (e.g., Her 2007), so decompositional accounts are not incompatible with missing arguments. However, they still handle these phenomena less elegantly than our holistic approach with no suppression.

‘Wrong’ arguments pose the biggest problem for decompositional approaches. Consider the following example:

(54)	人生	都	無	意義	啦,
	jan4sang1	dou1	mou5	ji3ji6	laa1,
	life	FOC	NEG.EXST	meaning	SFP
	咪	食	煙	食	死	佢	囉
	mai6	sik6	jin1	sik6	sei2	keoi5	lo1
	so	smoke	tobacco	smoke	die	3sg	SFP
	‘Since life is meaningless anyway, let me smoke it to death.’ [SEA033].

Here, the ArgA refers to the speaker’s life, not the speaker themselves, as it is in the third person. However, in Cantonese, humans, not lives, are the argument of 死 sei2 ‘die’.

Alternatively, consider the following pair:

(55)	a.	夏天	洗	乾淨	塊	面	好	重要
		haa6tin1	sai2	gon1zeng6	faai3	min6	hou2	zung6jiu3
		summer	wash	clean	CLF	face	very	important
		‘In summer it is important to wash your face clean.’ [SEA067].
	b.	請教	點樣	可以	洗	乾淨	啲	橙	汁	漬
		cing2gaau3	dim2joeng2	ho2ji5	sai2	gon1zeng6	di1	caang2	zap1	zik1
		HON	how	can	wash	clean	CLF	orange	juice	stain
		‘May I be enlightened as to how to wash the orange juice stain clean?’ [SEA068].

Example (55a) is straightforward: We wash our face, and our face becomes clean, hence 面 min6 ‘face’ is simultaneously the patient of 洗 sai2 ‘wash’ and sole argument of 乾淨 gon1zeng6 ‘clean’. However, in (b), an ‘orange juice stain’ does not become clean after washing—it simply disappears! The theme of ‘clean’ is still whatever surface was washed—which is not mentioned at all. This mismatch poses another difficulty for the decompositional approach, where both verbs are expected to share at least one argument, as reflected in Figure 6a. Similarly, our approach provides an uncomplicated resolution to the decompositional approach’s conundrums without compromising the semantic integrity of the construction, as shown in Figure 6b.

In (56), the stamps are not what become ‘full’; the stamp card is:

(56)	儲	滿	3	個	印花
	cou5	mun5	saam1	go3	jan3faa1
	collect	full	three	CLF	stamp
	‘Once you’ve collected three stamps …’ [SEA158].

Perhaps the most convincing evidence is when two constructions differ minimally semantically, but the decompositional analysis works in one case but not the other. Any attempt to ‘save’ the composite approach by excluding problematic constructions from the resultative would have to argue, inelegantly, that the two sentences are different constructions. Consider (57):

(57)	a.	擰	實	個	蓋
		ning2	sat6	go3	goi3
		screw	tight	CLF	lid
		‘screw the lid tight’ [SEA069].
	b.	即係	唔	洗	綁	實	個	人
		zek1hai6	m4	sai2	bong2	sat6	go3	jan4
		that is	NEG	need	tie	tight	CLF	person
		‘that is, you don’t need to tie the person tight’ [SEA070].
	c.	唔該	睇	實	個	細路	啦！
		m4goi1	tai2	sat6	go3	sai3lou6	laa1
		please	watch	tight	CLF	kid	SFP
		‘Please watch the kid tight!’ [SEA071].

Example (57a) can be straightforwardly analysed in decompositional terms: the lid is the patient of 擰 ning2 ‘screw’ and sole argument of 實 sat6 ‘tight’. However, this analysis would be strange for (57b), since people cannot be tight (only tied tightly), and impossible for (57c), where being watched does not render the child ‘tighter’. However, the three are semantically very similar; all involve constraints on ArgA’s movement.

Similarly, in the example below, his bad stuff can be an argument of 爆 baau3 ‘expose’, but him cannot:

(58)	a.	一於	就	督	爆	佢	啲	衰	野
		jat1jyu1	zau6	duk1	baau3	keoi5	di1	seoi1	je5
		let me	then	rat out	expose	3sg	CLF	bad	thing
		同	佢	離	婚
		tung4	keoi5	lei4	fan1
		with	him	leave	marriage
		‘Then let me expose his bad stuff and divorce him.’ [SEA140].
	b.	不過	我	自己	都	唔	會	督	爆	佢
		bat1gwo3	ngo5	zi6gei1	dou1	m4	wui5	duk1	baau3	keoi5
		but	1sg	REFL	also	NEG	will	rat out	expose	3sg
		‘But I won’t rat him out myself either.’ [SEA141].

A more metaphorical version is as follows:

(59)	a.	一早	睇	穿	你	啲	手法	啦	老屈	成	性
		jat1zou2	tai2	cyun1	nei5	di1	sau2faat3	laa1	lou5wat1	sing4	sing3
		long ago	see	pierce	you	CLF	method	SFP	slander	become	nature
		‘I have seen through your methods long ago, you chronic slanderer.’ [SEA142].
	b.	瀟	姐	一早	睇	穿	你	啲	衰	嘢
		siu1	ze1	jat1zou2	tai2	cyun1	nei5	di1	seoi1	je5
		Siu	Sister	long ago	see	pierce	you	CLF	bad	thing
		‘Sister Siu has seen through you and found out about your bad stuff long ago.’ [SEA143]

One can see through the methods in (59a), but not the ‘bad things’ in (59b); they are what one sees after seeing through someone’s façade!

4.2. Interim Conclusion of the Holistic Approach

Our holistic approach offers an uncomplicated and elegant way to account for CRC while preserving the semantic structures of the construction. Of course, numerous CRCs remain amenable to the decompositional analysis, since: (1) the transitivity of the verbs is unambiguous; (2) the semantic structure of the verbs is complete and the same, whether analysed independently or integrated as a CRC; and (3) the argument structure of the individual verbs are compatible and remain the same when pieced together. However, under our holistic approach, all the aforementioned constructions that do not adhere to these properties can be straightforwardly described with the CRC schema. Arguments for our holistic approach thus far can be summarised as follows:

• The CRCs covered in Section 4.1 all have an ArgA construable as affectee, even if they are ‘arguments’ of neither verb. Quite often, affectee status can be further supported by constructions such as the disposal construction with 將 zoeng1:24

(60)

a.

用

洗潔精

就

可以

將

啲

污跡

洗

乾淨

jung6

sai2git3zing1

zau6

ho2ji5

zoeng1

di1

wu1zik1

sai2

gon1zeng6

use

washing liquid

then

can

DISP

CLF

stain

wash

clean

‘You can wash the stains clean just with dishwashing liquid.’ (cf. 48b). [SEA130].

b.

我

將

佢地

集

齊

響

屋企

再

影

相

啦。

ngo5

zoeng1

keoi5dei6

zaap6

cai4

hoeng2

uk1kei2

zoi3

jing2

soeng2

laa1.

1sg

DISP

3pl

collect

complete

be at

home

again

take

picture

SFP

‘I’ll collect them all at home and take a picture again.’ [SEA131].

2.

In all of the examples, V2 is construed as relevant to V1’s result, sometimes resulting in a verb meaning distinct from the meaning in monoverbal clauses, e.g., 齊 cai4 ‘complete’ described above.

By removing the requirement that resultatives be decomposable into individual clauses, our approach also allows particles with no independent existence in monoverbal sentences, such as the quantifying 哂 saai3 or adversative 親 can1, to be included in the CRC; the lack of a corresponding monoverbal construction is no reason to exclude them.

Nevertheless, some valid generalisations about CRC phenomena have been made in the literature before, couched in decompositional terms. The next two sections will explain how we account for them. Section 4.3 will explain semantic orientation, mentioned without explanation in Section 3.2, and Section 4.4 will give a valency typology under our approach.

4.3. Semantic Orientation: An Alternative to Argument Linking

In most contemporary Western approaches to typical Chinese resultatives assuming a decompositional or hybrid holistic-decompositional approach, verbs possess an inherent set of grammatical or thematic roles, and descriptions of resultatives focus on how these roles are mapped onto the construction-level grammatical and/or thematic relations, i.e., argument linking or argument realisation. This applies to generativists/formalists (e.g., Cheng and Huang 1994; Cheng et al. 1997; Her 2004; C. Li 2007, 2013; Lau and Lee 2015, 2021), constructionists/functionalists (e.g., Matthews 2006; Huang 2007; Fong 2018; Liu 2020), and mixed approaches (Chow 2011, 2012; Lee and Ackerman 2011).

A sceptic may claim that, by dispensing of single verb-level argument structures, our account fails to exclude logically possible examples such as (61):

(61)	~*我	撞	爛-咗	啲	水
	ngo5	zong6	laan6-zo2	di1	seoi2
	1sg	bump into	break-PFV	CLF	water
	‘Intended: I bumped into and broke the teapot) and the water (spilled).’ [Unattested regardless of ArgC, classifier presence, and aspect marker presence]

Although the water is affected by teapot breaking, it is not permissible in this construction. In the traditional argument linking approach, this example would be excluded because ‘water’ cannot be an argument of ‘break’. While we will not be adopting argument linking in our approach, we still need a mechanism to express the relationship between individual verbs (e.g., break) and arguments (e.g., water), and in particular, why certain combinations such as (61) are not permissible. To resolve this query, we suggest that semantic orientation analysis, an alternative approach developed in China, is a better fit since it can account for facts about the semantic relationship between different elements of the CRC without requiring the CRC to be decomposed into component argument structures.

Similar to argument linking, semantic orientation grew out of Fillmore’s Case Grammar (Ài 2022). It describes semantic relationships between elements of a sentence that are not necessarily directly syntactically dependent. For example, in the English sentence, “They have all gone”, “all” is semantically oriented towards “they”, despite syntactically modifying “gone.”.

An element is typically said to be semantically oriented to another element if it explains or illuminates it in some way. We define semantic orientation more explicitly as follows: if a verb is semantically oriented towards an argument or another verb, then the situation evoked by the verb must logically involve the role played by said argument or other verb. Taking (55b) as an example, V2 describes something becoming clean. This implies there must have been some dirtiness before that is now gone—in this case, the orange juice stains—even though the stains are not normally an argument of ‘clean’.

Semantic orientation analysis has been extensively applied to Mandarin CRCs (e.g., Kāng 2008; Zhāng 2008; Liú 2022), but we depart from these accounts somewhat. Since these accounts assume V1 as the head, arguments of the CRC are assumed arguments of V1, and semantic orientation analysis applies only towards V2. Descriptions of V2’s semantic orientation are thus couched in terms of which arguments (or non-arguments) of V1 they orient towards. By contrast, our approach does not assume V1 as the head. Thus, we extend semantic orientation analysis to V1 too. We make three generalisations:

• V1s must be semantically oriented towards ArgCs;
• V2s must be semantically oriented towards ArgAs when ArgA is present; otherwise, they must be semantically oriented towards V1;
• Where an ArgN is present, outside of directional constructions and certain non-referential ArgNs, all verbs must be semantically oriented towards the ArgN.

The following sections explain how these generalisations apply in different situations, including to account for patterns previously described decompositionally.

4.3.1. Generalisation 1: On V1 Orienting to ArgC

The requirement for V1 to be semantically oriented towards ArgCs explains Cheng and Huang’s (1994) observation for Mandarin, which is also valid for Cantonese, that non-‘inverted’ two-argument resultatives cannot be interpreted as having indirect causers as ArgCs. For example, in the following example, the zero ArgC cannot be interpreted as causing someone else to cry on the tissue:

(62)	喊	濕-咗	幾多	包	紙巾？
	haam3	sap1-zo2	gei2do1	baau1	zi2gan1
	cry	wet-PFV	how many	packet	tissue
	‘How many packets of tissue did (he) wet by crying? / *How many packets of tissue did he cause to be wet be crying?’ [SEA112].

The use of semantic orientation instead of clausal decomposition easily explains cases such as the following, where the ArgC ‘mechanical pencil’ is not usually an argument of V1 ‘write’ in monoverbal contexts:

(63)	鉛芯筆	寫	壞	手勢。
	jyun4sam1bat1	se2	waai6	sau2sai3
	mechanical pencil	write	bad	gesture
	‘Writing with mechanical pencils makes your writing gestures bad.’ [SEA153].

Since writing necessarily involves a writing implement, ‘write’ semantically orients to ‘mechanical pencil’. This is advantageous over traditional inversion-based accounts, where ‘gesture’ and ‘mechanical pencil’ cannot normally be the two arguments of ‘write’ (see Figure 7).

4.3.2. Generalisation 2: On V2 Orienting to ArgA (and V1)

As mentioned above, V2s of CRCs still orient towards the ArgAs, even when those ArgAs are not arguments of the V2 in monoverbal contexts. Consider (64) again:

(64)	請教	點樣	可以	洗	乾淨	啲	橙	汁	漬
	cing2gaau3	dim2joeng2	ho2ji5	sai2	gon1zeng6	di1	caang2	zap1	zik1
	HON	how	can	wash	clean	CLF	orange	juice	stain
	‘May I be enlightened as to how to wash the orange juice stains clean?’ [SEA068].

V2 describes something becoming clean. This implies there must have been some dirtiness before that is now gone—in this case, the orange juice stains—even though the stains are not normally an argument of ‘clean’ (see Figure 8).

By abandoning grammatical relations, our account is much simpler than previous semantic orientation accounts of Mandarin (e.g., Kāng 2008; Zhāng 2008; Liú 2022), which typically describe three to six types of orientation. ‘Subjects’ and ‘objects’ that V2 orient to are both ArgAs in our account, though such ‘subjects’ are also simultaneously ArgCs while ‘objects’ are not. Non-‘subject’, non-‘object’ arguments are also ArgAs:

(65)	a.	朱	智賢	出	街	跑	步	仆	損	手	腳
		zyu1	zi3jin4	ceot1	gaai1	paau2	bou6	puk1	syun2	sau2	goek3
		Chu	Chi yin	go out	street	run	step	fall	injure	arm	leg
		‘Ashley Chu went out to the streets to run and fell and injured her limbs.’ [SEA132]. [Part of subject].
	b.	大家	行	爛-咗	幾	多	對	鞋
		daai6gaa1	haang4	laan6-zo2	gei2	do1	deoi3	haai4
		everyone	walk	break-PFV	how	many	pair	show
		‘How many pairs of shoes has everyone ripped by walking?’ [SEA129] [Quasi-instrument].
	c.	喺	店	內	既	牆身	寫	滿-咗
		hai2	dim3	noi6	ge3	coeng4san1	se2	mun5-zo2
		be at	shop	inside	ASSOC	wall surface	write	full-PFV
		一	堆	奇怪	既	日文
		jat1	deoi1	kei4gwaai3	ge3	jat6man2
		one	CLF	strange	ASSOC	Japanese
		‘The wall surface in the shop was filled with a pile of strange Japanese writing.’ [SEA133] [Location].

Zhāng or Liú would classify ‘limbs’, ‘how many pairs of shoes’ and ‘the wall surface in the shop’ as non-subject, non-object arguments that V2 orient to; for us, they are all ArgAs.

Phase, quantifying and comparative V2s are oriented towards both V1 and ArgA. For example, in (28), successfully falling asleep implies that a person (ArgA) has gone (V1) to sleep, and the different examples of 嗮 saai3 ‘all’ all involve some situation (V1) applying to all of something or group of things (ArgA). Manner V2s are always oriented towards V1, and also towards ArgA when it is present (see example (66) and Figure 9):

(66)	好好地	跳	靚	隻	舞	咪	算	囉
	hou2.dei6.dei6	tiu3	leng3	zek3	mou5	mai6	syun3	lo1
	well	jump	pretty	CLF	dance	then	count	SFP
	‘If only he would just (stay in their lane and) dance nicely!’ [SEA134].

4.3.3. Generalisation 3: On All Verbs Orienting to ArgN When Present

Outside of the directional construction, ArgNs must be semantically linked to all the verbs, unlike ArgC or ArgA. So, for example, under our definition, in (65), the two verbs are also semantically oriented towards the ArgN, Japanese. This is because writing implies something being written, and if something is full, then it must be full of something—in this case, Japanese writing (even though in Cantonese, 滿 mun5 ‘full’ cannot take Japanese as an argument in monoverbal clauses).

For phase complement constructions (Section 3.3) such as (28–30), V2 simply discusses the extent to which V1 is carried out, so ArgN is semantically related to phase V2 by virtue of being strongly associated with V1.

Generalisation three can capture the following generalisation by Lau and Lee (2021), which is originally stated in decompositional terms:

For active resultative sentences with two arguments, the NP argument with the target of activity role [but not the locus of affect role] is linked to the position immediately following the second verb only if the V2 is transitive.

L&L’s locus of affect roughly corresponds to our ArgA, and target of activity roughly refers to a patientive argument. The ‘NP argument’ in this paragraph is thus a non-ArgA patientive argument, i.e., ArgN. Generalisation three states that ArgNs must be semantically related to all verbs in the construction. When V2 is ‘intransitive’ in L&L’s account, that means only ArgA is semantically related to V2—ArgN is not. Here is an example:

(67)	*佢	寫	攰-咗	小說
	keoi5	se2	gui6-zo2	siu2syut3
	3sg	write	tired-PFV	novel
	‘He got tired from writing novels.’ (=L&L’s (19b)).

For L&L, the unacceptability is because ‘tired’ is intransitive. For us, it is because ‘tired’ does not semantically orient to ‘novel’. Thus, such constructions are also ruled out by our approach (Figure 10).

In some cases, V2 only tenuously invokes ArgN. For example, 飽 baau2 ‘full’ may invoke ‘food’, but one might also feel full for other reasons without food (e.g., illness). If ArgN does appear in these cases, it involves generic, non-referential ArgNs. In the following, (a) is attested, but (b) is unattested because 啲飯 di1 faan6 ‘rice’ is referential:25

(68)	a.	食	飽-咗	飯，	我	今日	cosplay，	勝	新太郎！
		sik6	baau2-zo2	faan6	ngo5	gam1jat6	kos1plei1	sing3	san1taai3long4
		eat	full-PFV	rice	1sg	today	cosplay	Katsu	Shintarou
		‘Having eaten, I will now cosplay as Katsu Shintarou!’ [SEA102].
	b.	~*食	飽-咗	啲	飯
		sik6	baau2-zo2	di1	faan6
		eat	full-PFV	CLF	rice
		‘I ate some rice and got full’ [unattested].

These cases may suggest that semantic orientation is gradient, and weaker orientations, such as full-rice, place more restrictions on the information status of the ArgN. Note that eat-full-rice (and drink-drunk-alcohol) are well-known sources of exceptions in other varieties of Chinese such as Mandarin (Cheng and Huang 1994; Shi 2002) and Southern Min (Lin 2015).

4.3.4. Marrying Semantic Orientation to the Decompositional Approach?

A sceptic may argue that the decompositional approach can simply be modified by introducing argument structures where the relation is one of semantic orientation, rather than the usual argument-structural relationship. This allows us to retain the decompositional approach while accounting for most, perhaps all, of the examples in Section 4.1. However, we believe this account is far less elegant and plausible.

Firstly, the main advantage of the traditional decompositional account is that individual CRCs are built up by existing argument structures that are used elsewhere in the grammar (i.e., in monoverbal clauses), reducing the inventory of signs needed in the language. Yet in the modified decompositional account, the semantic orientation-based argument structures for individual verbs would be akin to cranberry morphemes, since they do not appear alone, but must be in a CRC, just as the morpheme cran- is restricted to the context __berry in English. However, this is much less justified in the CRC context than for cran-. Firstly, examples such as (63) would have to be composed by putting two cranberries together (since 壞 waai6 is usually not used predicatively with 手勢 sau2sai3 ‘gesture’). More generally, while there is no clearly plausible alternative for cran-, the holistic approach can describe CRCs without resorting to syntactic cranberries, making it a more elegant option.

Secondly, many of the component argument structures would be semantically weird in such a modified decompositional account, making it less plausible. For example, one of the component argument structures of (50) would consist of the verb ‘put’, plus a person and their legs. Such an argument structure is not clearly meaningful, since ‘put’ inherently requires a position. For these reasons, we believe the holistic account is preferable to this modified decompositional account.

4.4. Valency Patterns

Previous decompositional accounts, especially Lau and Lee (2021), established typologies of resultatives according to the argument structures of the individual verbs and entire construction. Our holistic approach can also produce an argument structure typology. It collapses some of L&L’s categories by doing away with individual argument structures, while successfully covering rarer argument structure types missed by other approaches.

Table 2. Our valency typology of Cantonese CRCs, compared to L&L’s. ✓ indicates that an argument is present. * indicates situations where ArgC and ArgA are the same argument.

Type	ArgC	ArgA	ArgN	L&L Type
I		✓		Type 1 Pseudo-passive of Type 3 [Pseudo-passive of Type 6]
II	✓*			Type 1
III	✓	✓		Type 2 Type 3 [Type 6]26
IV		✓	✓	Type 4 Pseudo-passive of Type 5
V	✓*		✓	Type 4
VI	✓	✓	✓	Type 5
VII				/
VIII	✓			/

shows the possible argument structures of CRCs: six common types and two rare types. For cross-reference, these are compared to L&L’s typology.

We now describe each of the types in detail.

4.4.1. Types I (ArgA Only) and IV (ArgA + ArgN)

This includes all cases where the sole argument is an affected party that is not the cause. This includes constructions typically described as shared-subject constructions with two intransitives (a), as well as ‘pseudo-passives’ with no ArgN (b):

(69)	a.	小編	開心	死	喇！！！
		siu2pin1	hoi1sam1	sei2	laa3
		editor.HUM	happy	die	SFP
		‘I (the editor) am so happy I could die!!!’ [SEA082]
	b.	Hall	啲	野食	食	晒	未	啊？
		ho1	di1	je5sik6	sik6	saai3	mei6	aa3
		hall	DEM	food	eat	all	NEG.PERF	SFP
		‘Has all the food in the hall been eaten yet?’ [SEA081].

The constructions in (69) have preverbal ArgAs. Unlike traditional pseudo–passive analyses, however, our Type I also accounts for cases with postverbal ArgA, which are missed in analyses such as L&L’s:

(70)	a.	開心	死	我	喇	…
		hoi1sam1	sei2	ngo5	laa3
		happy	die	me	SFP
		‘I am so happy I could die ….’ [SEA083].
	b.	落	雨	溻	濕	個	袋
		lok6	jyu5	dap6	sap1	go3	doi2
		fall	rain	hit	wet	CLF	bag
		都	唔	好	溻	濕	自己
		dou1	m4	hou2	dap6	sap1	zi6gei1
		even if	NEG	good	hit	wet	self’
		‘When raining, better your bag gets (hit) wet than you get (hit) wet.’ [SEA078].27
	c.	唔該	Ethan	爸爸	揸	機，
		m4goi1	Ethan	baa4baa1	zaa1	gei1
		ask	Ethan	dad	hold	camera
		呢	幅	相	見	唔	到	佢	嘞
		ni1	fuk1	soeng2	gin3	m4	dou2	keoi5	laa3
		DEM	CLF	photo	see	NEG	DOU	3sg	SFP
		‘I asked Ethan’s dad to hold the camera, so (one) cannot see him in this photo.’ [SEA157].

Example (24) was another example of Type I where ArgA is in a postverbal position; note that in the example, V1 is the volitional verb 食 sik6 ‘eat’, but because the listener is not construed as the intentional causer of the V2 死 sei2 ‘die’, they are simply ArgA, not ArgC.

Type IV is similar, but with an additional ArgN:

(71)	條	數	啲	錢	入-咗	落	別人	個	袋	裏面
	tiu4	sou3	di1	cin2	jap6-zo2	lok6	bit6jan4	go3	doi2	leoi5min6
	CLF	sum	DEM	money	enter-PFV	go down	other	CLF	bag	inside
	‘The money involved went down someone else’s pocket.’ [SEA085].

Some CRCs involve a preverbal argument that is the possessor of the postverbal argument. The preverbal argument is clearly not a cause. It may be analysed two ways: it can be construed as a hanging topic, in which case the postverbal argument is ArgA and the construction is Type I, or it may be construed as an ArgA, in which case the postverbal argument is ArgN and the construction is Type IV. Consider the following example:

(72)	屋企人	係	好	想	佢	醫	好	個	病
	uk1kei2jan4	hai6	hou2	soeng2	keoi5	ji1	hou2	go3	beng6
	family member	COP	very	want	3sg	cure	good	CLF	illness
	‘His family members want him to get well from the illness.’ [SEA079].

4.4.2. Types II (ArgC=ArgA Only) and V (ArgC=ArgA + ArgN)

This includes all CRCs where ArgC=ArgA, both without ArgNs (60a, Type II) and with them (60b, Type V):

(73)	a.	我	食	飽	喇，	你哋	慢慢	食。
		ngo5	sik6	baau2	laa3	nei5dei6	maan6maan1	sik6
		1sg	eat	full	SFP	2pl	slowly	eat
		‘I’m full; you guys take your time to eat!’ [SEA089].
	b.	我	學	識-咗	好	多	嘢
		ngo5	hok6	sik1-zo2	hou2	do1	je5
		1sg	learn	know-PFV	very	many	thing
		‘I have learnt very many things.’ [SEA088].

Yiu’s (2013) self-agentive directional complements may also fall into either Type II (a) or V (b):

(74)	a.	開	船	後，	我	行-咗	出	去	影	相。
		hoi1	syun4	hau6	ngo5	haang4-zo2	ceot1	heoi3	jing2	soeng2
		open	ship	after	1sg	walk-PFV	go out	go	take	picture
		‘After the ship took off, I walked out to take pictures.’ [SEA090].
	b.	然後	我	行-咗	入	去	中央	公園
		jin4hau6	ngo5	haang4-zo2	jap6	heoi3	zung1joeng1	gung1jyun2
		and then	1sg	walk-PFV	enter	go	Central	Park
		‘And then I walked into Central Park.’ [SEA091].

Other particle constructions can also fall into this category; (75) exemplifies a resultative particle construction:

(75)	想	逃走	都	逃走	唔	到
	soeng2	tou4zau2	dou1	tou4zau2	m4	dou2
	want	escape	also	escape	NEG	DOU
	‘Even if I wanted to escape, I could not.’ [SEA096].

4.4.3. Types III (ArgC + ArgA) and VI (ArgC + ArgA + ArgN)

This includes all sentences where the cause and affectee are different arguments. For example, this includes Lau and Lee’s (2015) cross-referential accusatives (a) and causatives (b), and Yiu’s (2013) agentive directional complements (c):

(76)	a.	幻覺	嚟	嘅	啫，	嚇	我	唔	到	嘅！
		waan6gok3	lai4	ge3	ze1	haak3	ngo5	m4	dou2	ge3
		illusion	come	SFP	SFP	scare	1sg	NEG	DOU	SFP
		‘It’s just an illusion—it can’t scare me!’ [SEA094].
	b.	睇	M	club，	陸	永	笑	死	我！
		tai2	em1	kab1	luk6	wing5	siu3	sei2	ngo5
		watch	M	Club	Luk	Wing	laugh	die	me
		‘Watching the M Club, Billy Luk made me laugh to death!’ [SEA096]								.
	c.	放-咗	個	袋	入	locker	度
		fong3-zo2	go3	doi2	jap6	lok1kaa2	dou6
		put-PFV	CLF	bag	enter	locker	place
		‘I put the bag in the locker.’ [SEA116].

Many cases that do not fit into the clause concatenation paradigm because ArgA cannot be construed as an argument of V2, such as (54, 55b, 57bc), also belong here, as do resultative particle constructions with affected Ps and comparative constructions:

(77)	頭先	又	搵	唔	到	門匙。
	tau4sin1	jau6	wan2	m4	dou2	mun4si4
	just now	also	find	NEG	DOU	door key
	‘I could not find my door key just now either.’ [SEA097].

Even in these types, the causer referent can be an affectee; ArgC (causer) and ArgA (affectee) can be separate, coreferential forms, e.g. ArgA being reflexive (a), or a possessee of ArgC (in which case ArgC is also affected by the situation) (b):

(78)	a.	最近	太	過於	博，	死	博	爛	博，
		zeoi3gan6	taai3	gwo3jyu1	bok3	sei2	bok3	laan6	bok3
		recently	too	excessively	work hard	die	work hard	broken	work hard
		最後	病	親	自己
		zeoi3hau6	beng6	can1	zi6gei1
		finally	sick	ADV	self
		‘Recently, I have been working too hard, working far too hard, finally making myself sick.’ [SEA084].
	b.	唔係	真係	要	食	飽	個	肚
		m4hai6	zan1hai6	jiu3	sik6	baau2	go3	tou5
		NEG.COP	really	want	eat	full	CLF	stomach
		‘I do not really want to eat my stomach full.’ [SEA093].

4.4.4. Types VII (No Arguments) and VIII (ArgC Only)

Type VII refers to constructions with no arguments at all, largely ignored in the literature:

(79)	光	返	喇
	gwong1	faan1	laa3
	bright	return	SFP
	‘it’s bright again’ [SEA029].

Type VIII also lacks ArgA and ArgN, but does have an ArgC, which is an implicit first-person:

(80)	感覺	點	用	力	都	跑	唔	快
	gam2gok3	dim2	jung6	lik6	dou1	paau2	m4	faai3
	feel	how	use	force	still	run	NEG	fast
	‘I feel that no matter how much energy I use, I still can’t run quickly.’ [SEA128].

4.5. Interim Conclusion

In Section 4, we have shown that Western descriptions of CRC argument structure as the composition of the argument structures of individual verbs are often untrue for the Cantonese CRC, motivating our holistic approach where only the entire construction, not individual verbs, has arguments. The relationship between individual verbs and arguments is instead in terms of semantic orientation, which successfully accounts for phenomena previously described in decompositional terms. Our approach also results in a typology of CRC valences encompassing structures not captured by previous typologies.

5. Typological Implications

Although our discussion so far focuses on Cantonese, we believe similar ideas can apply to other Chinese varieties. Section 5.1 extends our framework to Cantonese–Mandarin comparison and Section 5.2 describes how it might be useful for comparing with other Chinese varieties. Section 5.3 examines the methodological implications of our two main points for worldwide typology, and Section 5.4 examines theoretical implications for diachronic typology and grammaticalisation.

5.1. Cantonese-Mandarin Comparison

Recall that for L&L, in CRCs with two arguments, an argument that is the target of activity role but not the locus of affect can only be the postverbal argument (traditional ‘object’) when V2 is transitive:

(81)	*佢	寫	攰-咗	小說
	keoi5	se2	gui6-zo2	siu2 syut3
	3sg	write	tired-PFV	novel
	‘He got tired from writing novels.’ (=L&L’s (19b)).

For us, (81) is explained by ‘tired’ not orienting semantically to ‘novel’.

L&L contrast their generalisation on Cantonese with Mandarin, where one can have an identical initiator and affectee, a separate target of action argument, and intransitive V2 simultaneously:

(82)	我	寫	累-了	小説
	Wǒ	xiě	lèi-le	xiǎoshuō
	1sg	write	tired-PFV	novel
	‘I got tired writing the book.’ (Mandarin, =L&L’s (19a)).

For Mandarin, we modify our semantic orientation restriction. Only V2 needs to be semantically oriented towards ArgN; V1 does not.

L&L mention two other differences with Mandarin; however, we believe those are mistaken. They write that, unlike Mandarin, Cantonese does not allow (a) ‘inverted’ resultatives and (b) structures where both Vs are intransitive and the initiator and affectee are distinct (also noted by Chow 2012). The first generalisation is inconsistent with well-documented examples in the literature (Matthews and Yip 2011; Chow 2012) and this paper (e.g., Section 3.3.1). The second claim is invalidated by examples such as these:

(83)	你	無視	我	喊	紅-咗	雙	眼
	nei5	mou4si6	ngo5	haam3	hung4-zo2	soeng1	ngaan5
	2sg	neglect	1sg	cry	red-PFV	pair	eye
	‘You neglected my crying my eyes red.’ [SEA042].

However, this leaves open the question of why the Cantonese examples that L&L contrasted with acceptable Mandarin equivalents in support of (b) are unacceptable. We believe Lau and Lee’s earlier (2015) account, based on semantics rather than argument structure and applying it to all argument structure types, was more appropriate. L&L use Washio’s (1997) distinction between ‘strong’ and ‘weak’ resultatives to explain the unacceptability of these constructions in Cantonese (whose cognate-for-cognate equivalents are possible in Mandarin):

(84)	a.	*我	跑	跌-咗	張	車飛
		ngo5	paau2	dit3-zo2	zoeng1	ce1fei1
		1sg	run	fall-ASP	CLF	ticket
		‘I ran, dropping my ticket.’ (=L&L’s (2)).
	b.	*佢	對	眼	喊	紅-咗
		keoi5	deoi3	ngaan5	haam3	hung4-zo2
		3sg	pair	eye	cry	red-PFV
		‘He cried his eyes red.’ (=L&L’s (15c)).

They claim that Mandarin has both ‘strong’ resultatives, where the meaning of V1 and V2 are completely independent, and ‘weak’ resultatives, where the V2 is the purpose or conventional result of V1. Cantonese lacks the latter, explaining (84). Yet most ‘strong’ V1–V2 combinations that L&L deem impossible are attested on the Internet. We searched for examples of all nine strong resultatives L&L deemed impossible in Cantonese, adding the perfective marker -zo2 to ensure examples are in Cantonese, and found examples of seven of these. One example was (83), as are the following:

(85)	a.	我	反而	係	覺得	個	男	嘅	追	攰-咗
		ngo5	faan2ji4	hai6	gok3dak1	go3	naam4	ge3	zeoi1	gui6-zo2
		1sg	on the contrary	COP	feel	CLF	male	ASSOC	chase	tired-PFV
		‘I, on the contrary, feel that the male got tired of chasing women.’ [SEA040].
	b.	真係	驚	隻	碟	俾	我	睇	花-咗
		zan1hai6	geng1	zek3	dip2	bei2	ngo5	tai2	faa1-zo2
		really	fear	CLF	disc	AGT	1sg	watch	scratched-PFV
		‘I really fear I’m watching so much that the disc gets scratched.’ [SEA041].

Rather than dichotomising between strong and weak resultatives, we believe these combinatorial restrictions are gradient collocational patterns, not structural ungrammaticality. Since semantics and usage frequencies affect collocational strength, when V2 is not the purpose or conventional result of V1, verbs are more likely to be collocated. The acceptability of Mandarin CRCs, then, is simply less sensitive to such effects than Cantonese, i.e., unusual V1–V2 fit less easily into Cantonese CRCs than Mandarin ones.

As a preliminary investigation of this hypothesis, for each of L&L’s ‘strong’ resultatives, we obtained, as a proxy for semantics, FastText word vectors (Grave et al. 2018) for the translational equivalents of the two verbs in Mandarin, then calculated their Euclidean distance. Figure 11 relates the number of tokens on Google search with semantic distance.28 This very small sample shows suggestive though inconclusive evidence that frequency is negatively correlated with semantic distance (Spearman’s rho = −0.622, p = 0.0738). We leave it to later work to investigate this issue with more rigorous measurements.

5.2. Potential Contributions to Dialectological Comparison

As mentioned in Section 2, most Sinitic dialectologists implicitly follow a holistic argument structure approach, albeit slightly different from ours, and many have some notion roughly corresponding to our CRC, especially when investigating potential forms. However, our ArgC–ArgA–ArgN reframing of the CRC template still diverges significantly from current typology practice retaining notions such as subject and object, and can potentially be fruitful in dialectological research.

For example, much comparative research in Sinitic typology examines the positions of the object relative to the verbs and potential form markers (Wú 2003, 2005; Lín 2006). The traditional ‘object’ corresponds to a postverbal ArgA or an ArgN in our framework. Most typological work cites examples with arguments corresponding to our ArgC and ArgA, with few ArgNs. Nevertheless, the few examples of ArgN cited in the literature may shed light on the difference between Cantonese and other varieties.

Even within the Yuè family, there is significant variation in word order. For example, Kwok (2010) discusses the Nánníng variety, which is much more permissive of arguments between V1 and V2. It is not limited to the potential form and is frequent even with full NP objects. Strikingly, this applies to both ArgAs and ArgNs:

(86)	a.	食	飯	飽	去	啊！
		sik6	faan6	beu2	hyu3	aa1
		eat	rice	full	go	SFP
		‘Eat yourself full with rice, then go!’ (Nánníng, =Kwok’s (8)).
	b.	*你	食	佢	唔	飽
		nei5	sik6	keoi5	m4	baau2
		2sg	eat	3sg	NEG	full
		Intended: ‘You couldn’t get full by eating it.’ (Cantonese).

As shown in (14), (86b) is not possible in Cantonese, even if rice were replaced by a pronoun and the whole construction became potential. The possibility of ArgN between V1 and V2 is thus another difference between Cantonese and Nánníng.

Shèng and Zhū (2020, p. 313) cite a variety of examples with ArgNs in the Shàoxìng Wú. However, one difference with Cantonese is that even when ArgNs are present, ArgAs can be between V1 and V2 (a), which is largely unattested in Cantonese (b), as mentioned in Section 3.2:

(87)	a.	个	苹果	我	想	喫	伽	患
		koh4	bin1ku2	ngo2	shian2	chieh4	noh4	waen3
		CLF	apple	1sg	want	eat	you	COMPL
		‘I want to eat up your apple.’ (Shàoxìng, =Shèng and Zhū’s (15)).
	b.	*我	食	你	唔	到	個	蘋果
		ngo5	sik6	nei5	m4	dou2	go3	ping4guo2
		1sg	eat	2sg	NEG	achieve	CLF	apple
		‘I can’t eat your apple.’ [Unattested regardless of ArgC and ArgN].

Because most comparative dialectological work in Chinese is framed in subject-object terms and uses mainly ArgA objects in examples, the extent of diversity regarding non-cause, non-affectee elements is currently unclear. We hope that future work will shed more light on this diversity by investigating ArgN-type objects in addition to ArgA.

5.3. Methodological Implications on Worldwide Typology

5.3.1. The Need for Explicitly Defining Constructional Levels

The existence of the schematically abstract causative–resultative construction poses serious challenges to worldwide typology. As mentioned above, typologists typically assume that each variety has a finite, enumerable set of serial verb constructions. In practice, constructions such as those listed in the headers of Section 3.3—resultative, comparative, causative, etc.—are assumed to be the relevant ‘constructions’ (e.g., Aikhenvald and Dixon 2006; Matthews 2006; Luke and Bodomo 2000). However, as construction grammarians have long argued (e.g., Diessel 2019; Croft 2007), ‘constructions’ are arranged into numerous levels of abstraction, from maximally concrete to highly abstract. How do we know that the level traditionally investigated is the ‘right’ one?

This has important consequences. For example, Matthews (2006), following Aikhenvald and Dixon (2006), states that the Cantonese causative SVC is asymmetric (with one of the verbs coming from a closed class) and the cause-effect SVC is symmetric (with both of the verbs coming from an open class). Yet, if we look at the CRC level, it is just symmetric. If we look at the directional vs. non-directional CRC level, then the directionals remain asymmetric, but the CRCs are still all symmetric. In the future, we hope typology can better specify the level of abstraction desired in analysis to ensure better comparability between languages investigated.

5.3.2. Rethinking Argument-Sharing Typology

The problems with the decompositional approach pose serious problems to typology, especially in approaches that do not separate language description and comparison. For example, Aikhenvald and Dixon (2006) set up several types of SVCs, giving both semantic and syntactic (in terms of component argument structure) characterisations. Those relevant to CRCs are shown in

Table 3. Aikhenvald and Dixon’s SVC categories relevant to Cantonese CRC.

Type	Shared Argument	Component Transitivity	Corresponds to Our Subconstruction
Cause-effect	O of V1 = S/A of V2	V1-transitive, V2-intransitive or occasionally transitive	Typical resultatives
Causative			Causatives
Event-argument	/	One V transitive/intransitive, Other V intransitive	Manner constructions
Resultative	/	Both intransitive	Typical resultatives

.

Even without the results of this paper, this schema is highly problematic. For example, our Type IV CRCs (ArgA + ArgN) such as (13ab) are semantically cause-effect, but ‘share’ both A and O, rather than the O of V1 being the A of V2. However, even if the table were expanded to include every combination of individual argument structures possible, the examples in Section 4.1, which cannot be easily separated into individual argument structures, still cannot be included.

Since the language of argument structure composition and argument sharing cannot be applied to all SVCs, one possible way to improve upon the typology of serial verb constructions is to take the multivariate approach (Bickel 2010) and decompose traditional concepts into smaller features. For example, traditional statement formats such as ‘The O of V1 is shared with the S of V2’ may be decomposed step-by-step into:

• Is it possible for V1 to appear in a monoverbal clause with one of the arguments of the SVC as the O?
• Is it possible for V2 to appear in a monoverbal clause with one of the arguments of the SVC as the S?
• If 1 is possible, is there no change in verbal semantics? How about 2?
• Are the NP targeted by 1 and 2 the same?

If, in traditional terms, ‘the O of V1 is shared with the S of V2’, then the answers are ‘yes’ to all four. The exceptions discussed in Section 3.3 all have ‘no’ to at least one of these answers. These variables can describe both ‘straightforward’ SVCs capturable by the decompositional approach, and more unusual constructions.

In approaches such as Haspelmath (2016), which strictly demarcate language-internal descriptive categories and comparative concepts for typology, one may continue using the decompositional approach if the comparative concept is carefully defined to exclude examples not amenable to decomposition. Indeed, Haspelmath already excludes our causative and perhaps manner CRCs (using his no predicate-argument structure between verbs criterion), and CRCs where one of the Vs is absent from monoverbal constructions (using his independent verb criterion). However, he does not go far enough; his independent verb does not guarantee that the verb has identical semantics in monoverbal and multiverbal contexts, nor that all and only arguments that would appear with the individual verbs show up in the SVC. Thus, Haspelmath’s definition needs to be further tightened to test generalisations hinging on decomposition (Generalisations 7–10) against Cantonese. Once tightened, Generalisation 7 (all SVCs share arguments) still seems unsupported:

(88)	但	好	驚	行	爛	對	鞋
	daan6	hou2	geng1	haang4	laan6	deoi3	haai4
	but	very	afraid	walk	wear	pair	shoe
	‘But (I)’m really afraid I’ll wear out the shoes by walking.’ [SEA162].

Here, the shoe would be the sole argument of wear out in a monoverbal clause, and the unexpressed I would be the sole argument of walk. Thus, there is no argument sharing despite the decompositional approach working.

5.4. Theoretical Implications on Constructional Change

5.4.1. Constructional Levels, SVC Symmetry and Grammaticalisation

As mentioned above, the existence of multiple levels of CRCs poses a challenge to Aikhenvald and Dixon’s symmetric-asymmetric typology, since the CRC level is symmetric while specific subconstructions may be symmetric or asymmetric. This raises the question of how to reconcile with our approach to A&D’s generalisation that asymmetric SVCs are sites for grammaticalisation (while symmetric SVCs are lexicalization sites). For example, in the A&D account, in Cantonese asymmetric causative CRCs (Section 3.3.5), the cause verbs are semantically light and grammaticalising into causative markers. However, if the higher-level construction, CRCs, is symmetric, aren’t causatives simultaneously favourable and unfavourable for grammaticalisation?

In fact, our approach is in concert with Bisang’s (2009) modification of A&D’s generalisation. Adopting a definition of grammaticalisation as the process where an originally open-class form becomes the marker for a construction, Bisang restates the diachronic statement as follows: symmetric SVCs are starting points of grammaticalisation, and as certain verbs in the symmetric SVC come to be grammaticalised as markers for a specific construction, they become members of a closed class, creating asymmetric SVCs. Thus, rather than asymmetric SVCs favouring grammaticalisation, it is grammaticalisation that leads to asymmetric SVCs. The asymmetric causative CRC, then, was formed with the (symmetric) CRC as the starting point and came into being as the causer verb became a semantically light constructional marker.29

As this grammaticalisation process progresses, some verbs take on meanings much more abstract and general than their lexical sources, which the literature has typically referred to as (resultative, phase, directional, etc.) ‘particles’. The tail end of this process can result in the construction dropping out of the CRC altogether. For example, consider the directional V2 返 faan1 ‘result’, which has been grammaticalised into a stance marker (Chor 2013) and is no longer indicative of the result:

(89)	沖	返	個	靚	涼
	cung1	faan1	go3	leng3	loeng4
	flush	return	CLF	pretty	shower
	‘Let me take a nice shower.’ (Chor 2013).

Expressions using this sense of faan1 no longer have potential forms, and so are not CRCs.

5.4.2. Semantic Orientation as Potential Initiator of the Grammaticalisation of Verbal Particles

Our holistic approach to argument structure also sheds light on the mechanisms involved in some grammaticalisation pathways whereby asymmetric SVCs are created. Since the V2 only has to be semantically oriented towards ArgA and does not need to have a predicate-argument relationship with it, the V2 can appear with more types of ArgAs, widening the range of contexts the V2 appears in and therefore favouring its grammaticalisation (cf. Himmelmann 2004) as a verbal particle relevant to the result of V1. This section will focus on Mandarin, where examples where ArgA is not an argument of V2 are also easily found, and historical texts are easily available.

Consider Mandarin 光 guāng ‘bare’ (J. Wáng 2010), which has been grammaticalised into a quantifying particle meaning all of ArgA was removed. Before grammaticalisation, we see uses where guāng as V2 clearly means ‘bare’:

(90)	誠	恐	他	吃	光-了	世界
	chéng	kǒng	tā	chī	guāng-le	shìjiè
	honest	fear	3sg	eat	bare-PFV	world
	‘She sincerely feared that it would eat the world bare.’ [Sānbǎo Tàijiàn Xīyángjì 43].

Later, bridging constructions such as the following began to arise:

(91)	怎的	把	一	盤	肉	包子	通	吃	光-了
	zěnde	bǎ	yì	pán	ròu	bāozǐ	tōng	chī	guāng-le
	how	DISP	one	tray	meat	bun	all	eat	bare-PFV
	‘How did he eat a whole tray of met buns bare?’ [Sūn Páng Dòuzhì Yǎnyì 20].

Here, if the classifier (i.e., tray) is seen as the head of ArgA, then it is an argument of bare. However, in our approach, CRCs allow V2s to only semantically orient towards ArgAs, and meat satisfies this condition. So, if meat is treated as the head of ArgA, it still fits into the construction.

This process eventually gave rise to constructions with only the latter interpretation, i.e., the present quantifying use, which is now most common, and can no longer be decomposed into two clauses with guāng meaning ‘bare’:

(92)	如果	蟲子	把	樹葉	都	吃	光-了
	rúguǒ	chóngzǐ	bǎ	shùyè	dōu	chī	guāng-le
	if	bugs	DISP	leaf	all	eat	all-PFV
	‘If bugs eat up all the leaves …’ [SEA149].

Mandarin constructions with V2s such as 破 pò ‘break’, 穿 chuān ‘pierce’, or 透 tòu ‘pass through’ followed a similar path. They are often used metaphorically when the ArgC sees through some incorrect or deceptive thing, e.g., a façade or illusion. One frequent collocation is 看破 kàn pò ‘look break’, frequently used when the ArgC has seen through the empty and transient nature of worldly matters:

(93)	把	興	亡	看	破
	bǎ	xīng	wáng	kàn	pò
	DISP	prosperity	vanquishment	look	break
	‘(I) saw through prosperity and vanquishment (i.e., am no longer bothered by them).’ [Jīnzhǎnzǐ, Sì Shí Huái Gǔ Qiū Cí from Běn Táng Jí].

Such worldly matters are generally the implied affectee even when there is no affectee made explicit. In these examples, the affectee is unproblematically an argument of ‘break,’ since they were demolished in the eyes of the ArgC. Quoting nun and Buddhologist Fat Yan, ‘In Buddhism we always say kàn pò—what are we ‘breaking’? Our wrong concepts.’30

Again, as the construction developed, ArgAs began appearing that cannot be direct arguments of ‘break’ in monoverbal clauses. Sometimes, the affectee may be the person putting up a façade or act:

(94)	我	已	三	五	日	前	看	破	他	了
	wǒ	yǐ	sān	wǔ	rì	qián	kàn	pò	tā	le
	1sg	already	three	five	day	before	see	break	3sg	ASP
	‘I have already seen through him three to five days ago.’ [Sānbǎo Tàijiàn Xīyángjì 82]

Alternatively, the affectee may be the truth revealed after breaking through the illusion. Consider (95):

(95)	如	曾	點	卻	被	他	超然	看	破	這	意思
	rú	Zēng	Diǎn	què	bèi	tā	chāorán	kàn	pò	zhè	yìsi
	like	Zēng	Diǎn	but	AGT	3sg	transcendently	see	break	this	meaning
	‘Yet as for Zēng Diǎn, he transcendently realised this meaning.’ [Zhūzǐ Yǔlèi 40].

Here, the meaning is not what is ‘broken’—rather, what ‘broke’ was the reasons (discussed in the preceding context) that make the meaning difficult to understand. Z. Wáng (2016)’s example (40) resembles this, though he did not appear to notice that the ArgA was not actually broken.

Thus, by virtue of the fact that ArgAs do not have to be ‘arguments’ of V2 in monoverbal predicates, V2s such as pò seem to be grammaticalising into a resultative particle, indicating that the preceding verb results in dispelling an illusion, rather than necessarily ‘breaking’.

In the Dàjīnggǎng variety of Southwestern Mandarin, 破 [pʰo] has been further grammaticalised into a general completive marker, and this change may have involved similar processes. [pʰo] is not restricted to situations that involve breaking, literally or metaphorically, but can be used in any situation where something disappears, diminishes or is destroyed, or even changes state (Chén and Zhōng 2021). An example where the affectee is destroyed is as follows:

(96)	他	把	我	作业	烧	破	哒。
	3sg	DISP	1sg	homework	burn	break	ASP
	‘He burnt up my homework.’ (Chén and Zhōng 2021, p. 103).

A change-of-state example is as follows:

(97)	我	要	去	把	衣服	洗	破。
	1sg	want	go	DISP	clothes	wash	break
	‘I want to go wash up the clothes.’ (Chén and Zhōng 2021, p. 103).

The jump from examples such as (96) to change-of-state examples such as (97) may be motivated by the same principle. The clothes did not diminish or get damaged, but rather the stains on them disappeared. However, the clothes remain affected by the stain’s disappearance.

6. Conclusions

In this paper, we provide a new account of the Cantonese causative–resultative construction, a broadly defined notion covering a variety of constructions with a range of similarities across syntactic and semantic domains. Though our account is constructionist-inspired, we dispense with traditional notions common to most previous accounts, such as subject, object, pseudo-passives and inversion. Instead, we directly describe the order of ArgC (cause), ArgA (affectee) and ArgN (non-cause non-affectee) without recourse to these notions. Thus, our analysis falls in line with framework-free grammatical theory (Haspelmath 2009), assuming no a priori syntactic categories, in response to criticism (e.g., Stern 2019) that construction grammar still relies excessively on traditional categories.

We also show that the decompositional approach to resultative argument structure is empirically untenable, compared to our holistic account, where arguments belong only to the whole construction, and where the relationships between individual verbs and referents are expressed with semantic orientation instead of traditional predicate-argument terminology. This clearly shows that when two historically distinct clauses are combined, the resulting construction can become crystallised as a construction per se rather than derivative of the biclausal source, even when neither of the verbs is clearly grammaticalised; and that, pace Foley and Olson (1985), this process is not limited to constructions where the two verbs are contiguous: the verbs in Cantonese CRCs are separable by potential markers and ArgAs.

Finally, as a personal remark, although our paper, of course, benefited greatly from existing theoretical knowledge, we could not have arrived at our conclusions without starting our investigation by observing interesting phenomena in natural language use, which led us to document a host of phenomena that appear erratic in traditional approaches. We believe a bottom-up, observational approach can enrich traditional theoretical approaches and take us further.