The Discovery of Aspect: A Heuristic Parallel Corpus Study of Ingressive, Continuative and Resumptive Viewpoint Aspect

Abstract

Languages differ in how systematically and obligatorily they encode conceptual categories such as tense and aspect. By drawing on large parallel corpora, these differences can be exploited heuristically: expressive obligatoriness and the systematicity of a conceptual category in one language can function as a probe for other languages that do not (evidently) encode it. This study applies this method—called heuristic translation mining (HTM)—to viewpoint aspect in Mandarin (an aspect-oriented language) and Dutch (a non-aspect-oriented language). Specifically, it takes the Mandarin aspect markers 起来-qilai (“ingressive”) and 下去-xiaqu (“continuative”) and collects translation strategies for these markers from a corpus of five Mandarin novels and their Dutch translations. The outcomes are methodological, descriptive and theoretical in nature. Methodologically, it is shown how conceptual templates consisting of temporal boundaries and phases facilitate annotating specific types of viewpoint aspect consistently. Descriptively, the exercise indicates at which linguistic levels viewpoint aspect may be encoded in a non-aspect-oriented language. Theoretically, conducting an HTM analysis with several aspect markers at once makes it possible to quantify (non-)marking of conceptual content; it turns out that the types of viewpoint under study correspond to varying marking frequencies, which may correlate with conceptual complexity.

1. Introduction

Languages vary synchronically in the ways they encode conceptual categories such as tense, aspect and modality (i.a., Comrie 1976; Bybee et al. 1994; Bhat 1999; Binnick 2012). Major axes of variation are the degrees of obligatoriness and systematicity of expression. A well-known construction illustrating this point is the English Progressive, which is systematic in its composition (i.e., be V-ing), cf. (1a), and obligatory for ongoing situations vis-à-vis the simple form of the verb, cf. (1b). Lower on these axes—at least when it comes to progressivity—is Dutch, which allows a simple present verb for an ongoing situation, cf. (2a), and has at least two progressive constructions: the prepositional progressive (aan het V_INF zijn) in (2b), and the posture progressive (zitten te V_INF) in (2c). Since (2a–c) may all translate to (1a), Dutch encodes progressive aspect in a less obligatory and systematic way than English does (cf. Bertrand et al. 2022 for a similar juxtaposition of the English and Dutch present perfect constructions).

(1)	a.	Rosie is eating a sandwich.
	b.	*Rosie eats a sandwich.
(2)	a.	Rosie eet een broodje.
	b.	Rosie is een broodje aan het eten.
	c.	Rosie zit een broodje te eten.
		‘Rosie is eating a sandwich.’

Advances in the accessibility and searchability of large-scale parallel corpora (cf. Barlow 2008; Cysouw and Wälchli 2007; Dahl 2007, 2014) have made it possible to exploit this type of crosslinguistic variation heuristically, remedying a basic methodological problem: identifying the means of expression for a conceptual category in a language that does not (evidently) encode that category obligatorily or systematically. By working with parallel translation corpora of languages on opposite ends of these axes, the obligatoriness and systematicity of expression in one language can help to discover the equivalent means of expression in the other—using the source language, in other words, as a probe for the target language.

This particular application of parallel corpus methods—which builds upon corpus-based contrastive linguistics (CCL: i.a., Aijmer et al. 1996; Johansson 2007; Granger and Lefer 2020), Translation Mining (TM: i.a., van der Klis et al. 2017, 2020, 2021; de Swart et al. 2022) and multiple parallel text analysis (MParT: i.a., Lu and Verhagen 2016; Lu et al. 2018; Knotková and Lu 2020)—has been called heuristic translation mining (HTM: Bogaards 2019a).1 HTM is a useful instrument in the linguist’s toolkit for both descriptive and theoretical questions. Descriptively, HTM may point the researcher in the direction of means of expression they were not aware of. And theoretically, HTM may contribute to understanding more precisely how languages vary in how they encode grammatical concepts, which may in turn inform general theories about those concepts by broadening their empirical basis.

The probing perspective HTM capitalizes on is not new; the idea has some precedent in CCL in particular. What distinguishes HTM from CCL, though, is that HTM specifically seeks out languages that are maximally divergent in terms of optionality and systematicity of expression of a given grammatical concept. This provides empirical input for theories working from the assumption that crosslinguistic variation is constrained by universal ordering principles, for instance the cartographic framework (Cinque 1999 et seq.) and generative approaches to the syntax–semantics interface (i.a., Travis 2010; Ramchand 2018). The connection between parallel corpus methods and this type of theoretical work is underdeveloped, and HTM can be seen as a subtype of CCL particularly suited to bridging this gap. This paper is part of a larger research project aiming to make this connection.

The aim of this paper is thus to expand the scope of HTM. It does so by applying it to viewpoint aspect in Mandarin Chinese and Dutch. As an aspect-oriented language, Mandarin has a set of verbal particles encoding particular types of viewpoint aspect (i.a., Chao 1968; Li and Thompson 1981; Yang 1995; Dai 1997; Xiao and McEnery 2004; Wiedenhof 2016; Zhao 2022). Dutch is on the other end of the spectrum, with periphrastic aspectual expressions—like those in (2b,c)—characterized by optionality and limited systematicity (cf. Boogaart and Bogaards forthcoming).

The present study takes two Mandarin viewpoint aspect markers and examines how they are translated to Dutch in a specially assembled parallel corpus drawing on five Mandarin novels and their Dutch translations. The two selected aspect markers are 起来 -qilai and 下去 -xiaqu, which encode “ingressive” (also called “inchoative” and “inceptive”) and “continuative” (also called “successive” and “continuous”) aspect, respectively.2 These particular types of viewpoint aspect were chosen because they vary in conceptual complexity yet are similar enough to be compared and operationalized simultaneously—that is, using the same conceptual building blocks.

This study innovates in three ways vis-à-vis previous studies applying (H)TM or MParT. First, two distinct conceptual categories are involved at once rather than just one; examples of the latter include parallel studies into motion events (Verkerk 2014), speech representation (Verhagen 2012), indefinite pronouns (Beekhuizen et al. 2017) and definiteness (Bremmers et al. 2021). Working with two types of aspect makes it possible to compare the (non-)marking of each type as a proxy for obligatoriness in the target language.

Second, the types of viewpoint aspect under study differ conceptually from previous (H)TM studies into viewpoint aspect, which operate at the level of the perfective–perfect and perfective–imperfective distinctions (Dahl 2014; van der Klis et al. 2017, 2020, 2021; Bogaards 2019a). I will try to show that ingressivity and continuativity can be incorporated in a parallel corpus study by defining internal-temporal templates based on selection theory (Bickel 1997).

Third, the aspect markers under study are low-frequency expressions, so rather than extracting a set number of sentences featuring a given aspect marker from one novel (van der Klis et al. 2020, 2021) or a number of novels (Bogaards 2019a), this study is based on a corpus of five full novels, amounting to almost 30,000 translation pairs. Working with a corpus of this size is an attempt to work out the scale necessary to apply HTM to forms with lower frequency.

The paper is structured as follows. Section 2 goes over the basic assumptions of HTM, its relation to other types of parallel corpus research, and two issues raised by the design of this study. Then, Section 3 reviews previous research on viewpoint aspect in Dutch, aiming to give an overview of the expressions already known to be tied to ingressivity and continuativity. Next up are the Mandarin aspect markers: in Section 4, the aspectual values encoded by 起来 -qilai and 下去 -xiaqu are described on the basis of Mandarin reference grammars and previous studies, and operationalized such that (i) it is clear how they carve up the aspectual domain within their ‘native’ system and (ii) they can be uniformly applied to the target language. Section 5 reports on corpus assembly and annotation and Section 6 presents the results: a discussion of the types of ingressive and continuative aspect marking found in a non-aspect-oriented language like Dutch, and a comparison of (non-)marking frequency of these types of viewpoint aspect. Section 7 concludes by giving an overview of the ‘discovered’ means of expression and reflecting on methodological and theoretical points raised by the study.

2. Heuristic Translation Mining

At its core, heuristic translation mining (HTM) involves systematically studying parallel corpus items—i.e., original and translated language material aligned sentence-by-sentence—with the aim of identifying means of expression in the language of translation (Lu and Verhagen 2016, p. 12; Bogaards 2019a, p. 161). This section starts out by reviewing HTM’s basic assumptions and relating the method to similar parallel corpus approaches. Then, I raise two additional issues: selecting conceptual content and discourse type (discourse vs. narrative), and the relation between parallel and non-parallel corpora.

2.1. Basic Assumptions

HTM takes several basic assumptions from multiple parallel text analysis (MParT), specifically regarding sensibility, heterogeneity and asymmetry. These principles both overlap with and deviate from corpus-based contrastive linguistics (CCL) and translation mining (TM). The sensibility assumption bears on the text producers of the parallel corpus material: the translator, faced with the task of verbalizing the conceptual content from the original text in the language of translation, is assumed to come up with sensible strategies to complete this task (Lu and Verhagen 2016, p. 170). In this sense, HTM treats translations as large surveys with a limited number of respondents.3

An intervening factor is crosslinguistic heterogeneity: languages vary in the means of expression they offer their speakers; they may “provide different conventional construals” (Verhagen 2012, p. 1). Especially at the level of grammatical notions such as viewpoint aspect, the consequence is that (aspect) marking in the original need not entail marking or even equivalent construal in the translation. Corpus annotation for HTM, therefore, requires labels for ‘unmarked’ and ‘untranslated’ items (Bogaards 2019a, pp. 170–73).4 These labels have direct counterparts in corpus-based contrastive linguistics (CCL), namely, ‘zero correspondence’ and ‘divergent correspondence’ (Johansson 2007, pp. 25–26).

The third basic principle drawn from MParT pertains to the methodological status of the languages under study, which is asymmetrical: the source language (here: Mandarin) is used to gather new insights about the target language (here: Dutch), not the other way around. This is because conceptual content (here: ingressive and continuative aspect) is defined in terms of particular formal manifestations in one language. Naturally, the analysis can be reversed afterwards, by assembling a new corpus drawing on, for instance, Dutch original texts and Mandarin translations. But this requires a separate, extra step; in its basic application, HTM is a “one-way street” (Bogaards 2019a, p. 162). In this sense, HTM has a narrower focus than CCL, which treats translations with caution but does generally pursue bidirectional comparison (e.g., Xiao and McEnery 2004; Johansson 2007). HTM’s monodirectionality is in line with its focus on maximally divergent expression—in the present case, aspect-oriented versus non-aspect-oriented languages.

HTM shares with MParT its emphasis on the detailed qualitative analysis of parallel sentence pairs as a way of getting a grip on crosslinguistic variation. However, it aligns with TM in that it also seeks to quantify the corpus data, contrary to MParT. The main difference between HTM and TM is the monodirectionality discussed previously, which TM does not apply as strictly. TM starts out by identifying a crosslinguistic formal category (or a set of them), for instance, the have-Perfect in Germanic and Romance languages (Dahl 2014; van der Klis et al. 2017, 2020, 2021). The goal is then to study the crosslinguistic variation of this category based on differences in distribution, visualized by means of multidimensional scaling (van der Klis and Tellings 2022). Conceptual content is thus not defined in terms of one or more forms in one language, but by forms shared by all languages under study—based on the initial observation that the same form exists across these languages. By extension, TM does not pursue HTM’s heuristic goals: rather than ‘discovering’ new means of expression in one language, TM aims to map the distribution of known ones in several languages. Hence, TM and HTM are complementary approaches: they map different things.

2.2. Conceptual Content and Discourse Type

It is well known that aspect may interact with discourse type (de Swart 2007). In the case of Dutch, use of the Perfect (vis-à-vis the Simple Past) depends on whether one is dealing with narrative or non-narrative discourse (Boogaart 1999; Le Bruyn et al. 2019; van der Klis et al. 2020, 2021).5 In the English and Dutch data collected by Le Bruyn et al. (2019, p. 168), the Perfect only shows up in non-narrative discourse, never in narrative discourse. The question is whether this restrictive interaction is a property of viewpoint aspect in general, or of subtypes like (im)perfectivity in particular. In the present case: do ingressivity (i.e., construing the start of a situation) and continuativity (i.e., construing the continuation of a situation) depend on this discourse-level distinction for their realization?

One known way of realizing ingressivity and continuativity in Dutch is with the (semi-)auxiliaries beginnen ‘start’ and blijven ‘stay’, respectively (cf. Section 3). They can be used in both narrative sequencing, as in (3), and non-narrative dialogue, as in (4).

(3)	a.	Toen begon Kailyn te zingen.
		‘Then Kailyn started to sing.’
	b.	Vervolgens bleef ze maar zingen.
		‘After that she just kept on singing.’

(4)	a.	“Wat doe je in de keuken?’’ “Ik ben begonnen met lunch maken.”
		‘”What are you doing in the kitchen?” “I’ve started making lunch.”’
	b.	“Schilder je nog?” “Ja, ik ben blijven schilderen tijdens de pandemie.”
		‘“Are you still painting?” “Yes, I kept on painting during the pandemic.”’

The difference between (3) and (4) lies not in the expression of ingressivity/continuativity (by beginnen and blijven in both cases), but in the tense marking on these verbs: simple past in (3), present perfect in (4). If one abstracts away from the viewpoint contributed by the tense markings in these clauses, then the construal and formal realization of ingressivity and continuativity in (3) and (4) are stable: beginnen and blijven mark initial and continuative points regardless and are therefore not sensitive to discourse type.

The more general point is: interaction with discourse-level distinctions such as ‘narrative/non-narrative’ seems to depend on the specific aspectual categories under study. While the Dutch Perfect is restricted to non-narrative discourse, there is no reason to suspect that ingressivity and continuativity are as well. For this reason, this study will not make this distinction in the corpus annotation, contrary to previous (H)TM studies, which deal with (im)perfectivity (e.g., Le Bruyn et al. 2019; Bogaards 2019a).

2.3. Parallel and Non-Parallel Corpora

Parallel corpora have their limitations: they are largely comprised of translated discourse, which is different from non-translated discourse (e.g., Vandevoorde et al. 2016), and they are restricted in size by the limited availability of high-quality translations. To make this more tangible: the (large) corpus assembled for the present study, consisting of five full novels, contains 413,094 Dutch tokens. By comparison, the SoNaR corpus of written Dutch (Oostdijk et al. 2013) contains more than 500 million—a ratio of roughly 1 to 1200. SoNaR also covers a variety of genres, whereas the corpus assembled here draws only on novels. For these reasons, HTM is said to produce results with strictly hypothetical status (Bogaards 2019a, p. 188), that is, hypotheses of the type “beginnen is a frequent ingressive marker in Dutch”, which are to be tested on a larger amount of non-translated linguistic data. A related point is that these hypotheses are likely not exhaustive; due to the limited size and composition of parallel corpora, there is always the possibility that certain means of expression were missed by the heuristic.

In HTM, parallel and non-parallel corpora should thus be used side-by-side. Two interrelated studies illustrating this point are Bogaards (2019a, 2019b): the former investigated the imperfective domain in Dutch by using the Mandarin imperfective marker 着 -zhe in a parallel corpus, which among other things yielded the construction ‘posture verb + past participle’ (e.g., begraven liggen ‘lie buried’); the latter then examined the construction’s meaning and restrictions on its productivity by examining about 8000 attestations from the SoNaR corpus. In the same vein, this study aims to produce testable hypotheses concerning ingressivity and continuativity in Dutch.

3. Ingressive and Continuative Aspect in Dutch

What is already known about ingressive and continuative aspect in Dutch? After presenting a very general definition of these types of viewpoint aspect, this section will survey the means of expression associated with them in the literature.

According to Dik and Hengeveld (1997, pp. 221–25), ingressivity and continuativity belong to a particular subtype of viewpoint aspect, namely, phasal aspect distinctions. Phasal aspect operates on situation aspect (e.g., States, Activities in the sense of Vendler 1967) in that phasal distinctions divide situations up into “phase[s] of development […] in terms of beginning-continuation-end” (Dik and Hengeveld 1997, p. 221).6 The ingressive viewpoint focuses on the first of these, i.e., the initial temporal boundary; and the continuative viewpoint focuses on a temporal point somewhere in between the initial and terminal boundaries.7 Dik and Hengeveld (1997, p. 225) provide the simple English examples in (5) to illustrate these general definitions.8

(5)	a.	Ingressive	‘John started crying.’
	b.	Continuative	‘John continued crying.’

Having formulated a working definition of these types of viewpoint aspect, we will turn to previous treatments of these notions in Dutch grammar.

3.1. Ingressivity in Dutch

The Syntax of Dutch (Broekhuis et al. 2015; Broekhuis and Corver 2015) discusses the verb beginnen ‘start’ from (3) and (4) as an “inchoative”—which is interchangeable in this context with “ingressive”—verb, as well as the “aspectual verbs” gaan ‘go’ and komen ‘come’ combined with a bare infinitive (Broekhuis et al. 2015, pp. 150–56). These verbs take either a VP-complement (e.g., gaan lopen ‘begin walking’) or TP-complement (e.g., beginnen te lopen ‘begin walking’). They also mention verbs that do not take a VP- or TP-complement, namely, aanvangen met ‘commence with’ and overgaan tot ‘take to’. These other verbal expressions (except overgaan tot) are mentioned by Vandevoorde et al. (2016, p. 136) as well, who also add een aanvang nemen ‘commence’, starten ‘start’ and aanvatten ‘commence’.

A more elaborate category of ingressive markers in Dutch is described in the general reference grammar Algemene Nederlandse Spraakkunst (ANS: Haeseryn et al. 1997): the prepositional pattern shown previously in (2b), which combines with the main verb zijn ‘be’ to encode progressive aspect, may also be licensed by other verbs to encode ingressivity—simply called “starting” in the ANS (Haeseryn et al. 1997, pp. 1048–54). This pattern is called the ‘aan het-construction’. The ingressive verbs licensed by the construction subdivide into causative and non-causative forms; (6) illustrates both: krijgen ‘obtain’ in (6a) is causative, slaan ‘hit’ in (6b) non-causative.

(6)	a.	De docent krijgt zijn leerlingen aan het lezen.
		‘The teacher gets his students to start reading.’
	b.	De leerlingen slaan enthousiast aan het lezen.
		‘The students enthusiastically start reading.’

Van Pottelberge (2004, pp. 27–51), Booij (2010, pp. 146–68) and Bogaards (2020a, pp. 59–92) further developed the ANS classification. Van Pottelberge noted that certain modal verbs get an ingressive reading when used in this construction.9 

Table 1. Ingressive verbs licensed by the ‘aan het-construction’.

Ingressive Verb	[±causative]	[±modal]
gaan ‘go’, komen ‘come’, (ge)raken ‘get’, slaan ‘hit’, zich zetten ‘put oneself’	−	−
krijgen ‘obtain’, maken ‘make’, zetten ‘put’, brengen ‘bring’	+	−
kunnen ‘can’, moeten ‘must’, mogen ‘may’, willen ‘want’	−	+

—which covers part of the full classification in Bogaards (2020a, p. 88)—lists the 14 verbs with ingressive meaning in this construction, subcategorized by causativity and modality.

A final construction linked to ingressivity is the verb worden ‘become’ in its function as a “change-of-state copula” (Hanegreefs 2004). Worden selects an adjectival or nominal complement (e.g., boos worden ‘become angry’, vader worden ‘become a dad’) and construes the beginning of the state denoted by the adjective or noun.10

Summing up, the literature mentions five strategies for ingressivity in Dutch:

• Beginnen ‘start’ + te + V_INF;
• Gaan ‘go’ and komen ‘come’ + V_INF;
• Ingressive verb set (Table 1) + aan-PP;
• Change-of-state copula worden ‘become’ + A/N;
• Other verbal strategies: aanvangen met ‘commence with’, overgaan tot ‘take to’, een aanvang nemen ‘commence’, starten ‘start’, aanvatten ‘commence’…

3.2. Continuativity in Dutch

The verb set of which the ingressive verbs in Table 1 are a subset also includes two continuative verbs: blijven ‘stay’ and its causative counterpart houden ‘hold’ (Haeseryn et al. 1997, p. 1050). They are illustrated in (7).

(7)	a.	De docent houdt zijn leerlingen aan het lezen.
		‘The teacher keeps his students reading.’
	b.	De leerlingen blijven enthousiast aan het lezen.
		‘The students keep on reading enthusiastically.’

The verb blijven ‘stay’ also combines with a VP (rather than a PP), e.g., blijven lezen ‘keep on reading’ (Broekhuis and Corver 2015, pp. 1019–20), and as a copula similar to worden, e.g., boos blijven ‘stay angry’ (Broekhuis and Corver 2015, p. 796). Another verb that Broekhuis and Corver mention, again in passing, is vervolgen ‘continue’ (Broekhuis and Corver 2015, p. 695). Finally, there is the ‘expressive en maar-construction’, which encodes continuativity while evaluating the continuation negatively (Broekhuis and Corver 2017; Anthonissen et al. 2019, pp. 1148–50). An example from Broekhuis and Corver (2017, p. 311) is given in (8).

(8)	En maar leugens verspreiden! Houd daar toch mee op!
	‘You just keep on spreading lies! Stop doing that!’

In sum, we can find five types of strategies in the literature for encoding continuativity in Dutch, three of which involve blijven ‘stay’:

• Blijven + VINF;
• Continuative copula blijven + A/N;
• Continuative verb (blijven or houden ‘hold’) + aan-PP;
• En maar + V_INF;
• Other verbal strategies: vervolgen ‘continue’…

4. Operationalizing Viewpoint Aspect

This section aims to operationalize the viewpoint-aspectual values encoded by Mandarin 起来 -qilai and 下去 -xiaqu for HTM corpus annotation. This is no trivial task for two reasons. First, both heuristic elements are polysemous and have several uses besides their aspectual ones. These meanings need to be teased apart and accounted for during annotation. Second, corpus annotation based on the markers’ aspectual meaning must be transparent and consistent. This is only possible if the conceptual content under study is made explicit and applicable. At the end of this section, I will propose that one way of doing so is by working with viewpoint-aspectual templates defined by temporal boundaries and phases. Before that, I will go into the non-aspectual and aspectual semantics of -qilai and -xiaqu. Items from the corpus will be used as examples. (See Supplementery Matertials).

4.1. Non-Aspectual Uses

As aspect markers, both -qilai and -xiaqu derive from lexical verbs denoting movement (Wu 2000). Synchronically, they still exist as lexical verbs with opposite meanings in terms of spatial direction: qilai ‘stand up’ and xiaqu ‘go down’, see (9) and (10) below.11

(9)	哦，小夕，你刚起来吗？
	O,	Xiaoxi,	ni	gang	qilai-ma?
	oh	Xiaoxi	you	Just now	qilai-qst
	‘Oh	Xiaoxi	did you only get up just now?’

(10)	可是我哭着下去找他。
	Keshi	wo	ku-zhe	xiaqu	zhao	ta.
	but	I	cry-imp	xiaqu	search	him
	‘But I went downstairs crying to look for him.’

A closely related use is as directional “Resultative Verb Compounds” (Li and Thompson 1981, pp. 58–65; Cherici 2019), i.e., directional RVCs, where -qilai and -xiaqu attach to a main verb and specify the situation denoted by that verb to involve completed upward or downward movement (see (11) and (12)). It is easy to see how these types of spatial movement relate to ‘standing up’ and ‘going down’.

(11)	小染想跳起来。
	Xiaoran	xiang	tiao-qilai
	Xiaoran	want	jump-qilai
	‘Xiaoran wanted to jump upright.’

(12)	吴敏把头低了下去，没有答腔。
	Wu	Min	ba	tou	di-le-xiaqu,	meiyou	daqiang
	Wu	Min	take	head	bow-prf-xiaqu	not.have	answer
	‘Wu Min bowed his head down, and didn’t answer.’

Abstracting away from ‘up’ and ‘down’, -qilai and -xiaqu may also function as RVCs without directional meaning (Xiao and McEnery 2004, pp. 227–28; Xiong 2017, p. 51). In these cases, they indicate not completed movement but completed event (see (13) and (14)).

(13)	我的两只手被捆绑起来吊了两天。
	Wo	de	liang	zhi	shou	bei	kungbang-qilai	diao-le	liang	tian.
	I	sub	two	clf	hand	by	bind-qilai	hang-prf	two	days
	‘They had bound my hands together and left me hanging like that for two days.’

(14)	只是看也不看地咽下去。
	Zhishi	kan	ye	bu	kan	de	yan-xiaqu.
	only	look	also	not	look	sub	swallow-xiaqu
	‘She swallowed everything without looking.’

For -xiaqu, these are all non-aspectual types. But for -qilai there is one more category, which Xiong (2017) terms “semantically opaque qilai”, covering cases where “it is hard to flesh out the semantic contribution of qilai” (Xiong 2017, p. 51). This category covers multiple types, including impersonal and evaluative uses (Wang and Wu 2018), as well as the qilai-middle, which is illustrated in (15). Since the present study is concerned with ingressive -qilai, I will adopt Xiong’s ‘opaque’ category as a cover term here for all non-aspectual uses of qilai other than lexical, directional and resultative ones.

(15)	行走起来，非常不方便，鞋子里渗进雪水，冻得两只脚又僵又痛。
	Xingzou-qilai	feichang	bu	fangbian	xiezi-li	shen-jin	xue	shui
	walk-qilai	very	not	convenient	shoe-inside	seep-in	snow	water
	dong	de	liang-zhi		jiao	you	jiang	you	tong.
	freeze	res	two-clf		foot	also	stiff	also	ache
	‘It walks very unpleasantly, the snow water gets into your shoes and your feet get so cold that they become both stiff and painful.’

There are thus three non-aspectual uses of -xiaqu (lexical, directional, resultative) and four non-aspectual uses of -qilai (lexical, directional, resultative, opaque) to be accounted for in the corpus annotation. Next up is the aspectual semantics of both markers.

4.2. Ingressive -qilai

While -qilai synchronically retains its lexical meaning of ‘standing up’ as well as directly extended uses such as directionality, it is also widely analyzed as a viewpoint aspect marker encoding ingressivity (i.a., Chao 1968; Dai 1997; Xiao and McEnery 2004; Shyu et al. 2013; Xiong 2017; Wang and Wu 2018). As an ingressive marker, -qilai attaches to a main verb and contributes an aspectual viewpoint to the situation denoted by that verb. Xiao and McEnery define its aspectual contribution as “a situation has started and will continue for some time” (Xiao and McEnery 2004, p. 219), which aligns with the general definition given earlier in terms of phases of development: -qilai selects the beginning phase. In terms of situation aspect, ingressive -qilai is sensitive to durativity, being felicitous with all situation types except Achievements.

As for its syntax, -qilai may be split if it attaches to a verb that consists of two syllables or licenses some sort of object. Several elements may come between qi and lai. For example, in (16), -qilai attaches to the bisyllabic verb 跳舞 tiaowu ‘dance’, producing the structure 跳起舞来 tiao-qi wu-lai. In (17), -qilai is selected by the verb–object constituent 开玩笑 kai wanxiao ‘crack jokes’, in which case qi comes in between the verb and its object: 开起玩笑来 kai-qi wanxiao-lai. These split patterns have to be accounted for in the corpus.

(16)	几个年轻人跟着彼头的“黑鸟”歌跳起舞来了。
	Ji-ge	nianqingren	genzhe	Bitou	de	“Heiniao”	ge
	Several-clf	young.people	follow	Beatles	sub	Blackbird	song
	tiao-qi	wu-lai-le
	jump-qi	dance-lai-prf
	‘A couple of young people started dancing to Blackbird by the Beatles.’

(17)	她竟给男人开起玩笑来，
	Ta	jing	gei	nanren	kai-qi	wanxiao-lai,
	she	unexpectedly	give	man	open-qi	joke-lai
	‘She even started cracking jokes to the man.’

Finally, there is some discussion in the literature on whether the ingressive viewpoint manifested by -qilai is a subtype of perfective or imperfective aspect. Xiao and McEnery (2004) consider it an imperfective viewpoint because it leaves a situation’s terminal boundary unfocused. Smith (1997) argues that it is neither, and introduces an additional category at the level of (im)perfectivity—‘neutral viewpoint aspect’—to account for their relationship. A third way out is to extend one’s definition of (im)perfectivity to include not just totality but also transition sensitivity (Koss et al. 2022), which results in 2 × 2 = 4 categories (in which case ingressivity would be non-total but transition-sensitive). For a heuristic study like the present one, however, the question whether -qilai’s ingressivity abstracts to perfectivity or imperfectivity is not relevant. We are interested in the specific means of expression translators come up with for this conceptual content, not its theoretical status in relation to similar concepts.

For -qilai, we now have input for both the operationalization (initial boundary and subsequent phase) and the corpus annotation (split structures). The aspectual use of -xiaqu is up next.

4.3. Continuative -xiaqu

-Xiaqu runs parallel to -qilai in its development as an aspect marker with its lexical source still intact. Besides its lexical meaning of ‘going down’ and immediate directional and resultative extensions, it has also been analyzed as a “successive” or “continuative” aspect marker (Chao 1968; Yang 1995; Dai 1997; Kang 1999; Xiao and McEnery 2004). In terms of situation type, -xiaqu is sensitive to the same parameter as -qilai, namely, durativity (Xiao and McEnery 2004:228). It is thus likewise incompatible with Achievements.

-xiaqu is both more complex and less complex than -qilai. Syntactically, it is less complex, as it cannot be split.12 Semantically, however, -xiaqu’s “continuativity” covers two types of viewpoint aspect. The general characterization given by Xiao and McEnery for the viewpoint encoded by -xiaqu is twofold: it “presupposes the prior initiation of a situation” and “focuses on the resumptive point and its continuative stage” (Xiao and McEnery 2004, p. 227). The conceptual complexity is manifested in the space between the prior initiation and resumptive point: these may be contiguous, as in (18), where the ‘staring’ is followed directly by ‘keeping on staring’. Crucially, though, -xiaqu also allows the prior initiation and resumptive point to be separated by a period in which the ‘continued’ situation does not hold. In this case, -xiaqu construes a special type of ingressive point: one preceded by prior initiation. This is illustrated by (19), a request to resume talking, which implies both that the other party talked before and that they stopped talking.

(18)	姜子牙再那么盯下去我就变成尼姑了。
	Jiang	Ziya	zai	name	ding-xiaqu	wo	jiu	biancheng	nigu	le.
	Jiang	Ziya	again	like.that	gaze-xiaqu	I	just	become	nun	prf
	‘Jiang Ziya, if you keep on staring at me like that, I will turn into a nun.’

(19)	“你说下去吧！妈，不要停。”
	“Ni	shuo-xiaqu-ba!	Ma,	bu	yao	ting.”
	you	speak-xiaqu-sug	mother	not	will	stop
	‘Resume talking, mother, don’t stop.’

Xiao and McEnery term cases like (18) and (19) “uninterrupted” and “interrupted” continuativity, respectively (Xiao and McEnery 2004, p. 233). Indeed, the (lack of) interruption produces the two kinds of continuative points in (18) and (19): due to interruption, -xiaqu in (19) construes a special kind of ingressive point, whereas there is no interruption and hence no ingression in (18); the situation is already ongoing. Instead, cases like (18) are more closely related to ‘stopping’ than ‘starting’—that is, they boil down to the negation of ‘stopping’ (Jackendoff 1990, p. 44; Lieber and Baayen 1997, p. 804). This is reflected by the fact that the semantic contribution of -xiaqu in cases like (18) can be paraphrased as ‘not stop’ (‘if you don’t stop staring at me like that…’). Moreover, when non-interrupted continuative -xiaqu is negated, it gets the meaning ‘stop’, i.e., ‘not not stop’, cf. (20). Conversely, if a sentence like (19) is negated, it would mean “not resume” (the same goes for -qilai, which would mean “not begin”).

(20)	小纲，别说下去了！
	Xiaogang,	bie	shuo-xiaqu-le!
	Xiaogang	not	speak-xiaqu-prf
	‘Xiaogang, stop talking now!’

This means that Xiao and McEnery’s (un)interrupted distinction should play a central role in the present corpus study. To highlight the distinction, I will use two more distinct labels, reserving “continuative” for cases like (18) and (20), where the situation is ongoing and was not terminated; and using “resumptive” for cases like (19), where the situation was terminated and -xiaqu profiles a special type of initial (i.e., resumptive) point. A related point for annotation is that negated instances of continuative -xiaqu (cf. (20)) should not be part of the heuristic, as they do not translate to a negated version of a continuative marker, but to a notion that is not itself continuative (i.e., ‘stopping’).

We now have all the input we need for -xiaqu in order to operationalize (prior initiation and either not stopping or starting again after stopping) and annotate (no split structures and exclusion of negated instances of the continuative marker).

4.4. Operationalization

As the previous two sections made clear, the aspect markers -qilai and -xiaqu both deal with boundaries and phases in time, albeit different kinds of temporal points in varying constellations. In these terms, they interact with the situations (lexicalized by verbs or adjectives) that they project their aspectual viewpoint onto. By that I mean: situations are likewise made up of temporal phases (e.g., ‘eating a sandwich’) and boundaries (e.g., the first and last bites of that sandwich), and what aspectual viewpoint does is to impose or focus those boundaries and/or phases. This view corresponds to what are called “(Radical) Selection Theories of Aspect” (i.a., Breu 1994; Bickel 1997; Sasse 2002, pp. 222–25; Croft 2012, pp. 49–52; Koss et al. 2022; Bogaards forthcoming), which assume aspectual bidimensionality (i.e., situation and viewpoint) and claim that viewpoint markers (e.g., ingressive and continuative) map phasal structure already (partly) present in the aspectual structure of situations. They thus have selectional preferences.13 The aspectual building blocks shared by both situation and viewpoint aspect are temporal boundaries (notated as τ) and middle phases (notated as ϕ) (Bickel 1997). Boundaries τ are punctual and involve change (e.g., telicity); phases ϕ are durative and may be either dynamic or stative.

My claim is that these building blocks can be used to construct templates that make the viewpoint-aspectual value of the heuristic elements explicit and, at a more practical level, lay out a way to annotate the corpus data consistently and transparently.

So, what do ingressive -qilai and continuative/resumptive -xiaqu look like in terms of these building blocks? Although Xiao and McEnery (2004) do not explicitly subscribe to the type of aspectual theory discussed above, they do work with temporal boundaries to visualize -qilai’s and -xiaqu’s aspectual values. I will therefore base my templates on their work (cf. Xiao and McEnery 2004, pp. 223–33), but attach more weight to (i) the phases construed by boundaries, (ii) the dissimilarities between -xiaqu’s continuative and resumptive meanings and (iii) the similarities between resumptivity and ingressivity.

Starting with -qilai, the ingressive viewpoint encoded by this marker is quite simple, consisting of just one initial boundary τ and one middle phase ϕ in which the situation denoted by the main verb holds. This is visualized in Figure 1 across a temporal axis t; the ‘initial’ status of τ is captured by including an additional phase preceding it during which the same situation does not hold (¬ϕ). Put more generally, initial boundaries τ are those in between two phases ¬ϕ and ϕ. The braces map these phases onto the temporal axis, and their (half-)dottedness is meant to indicate that they are not (fully) in focus; the emphasis lies on the initial point.

Schematizing continuative -xiaqu using the same building blocks and conventions, the viewpoint in Figure 2 is somewhat more complex; in addition to the focused boundary τ₂ there is an unfocused preceding boundary τ_1, which is initial (between ¬ϕ and ϕ), contrary to τ₂ (between ϕ₁ and ϕ₂). Normally, a boundary following a phase would itself constitute a terminal boundary (in between ϕ and ¬ϕ), but here the second phase is not negated. The analysis in Section 4.3 would account for this by claiming that ϕ₂ in Figure 2 corresponds to ¬(¬ϕ).

Resumptive -xiaqu is the most complex (see Figure 3), featuring three boundaries and four phases: two where the situation holds, two where it does not. The focused boundary τ₃ is sandwiched between ¬ϕ and ϕ, making it initial. However, there is an additional unfocused complete phase preceding τ₃ defined by an initial boundary τ₁ and a terminal boundary τ₂. In this sense, resumptivity constitutes a focused initial boundary embedded in a more complex viewpoint. I would thus argue that in its focused region, resumptivity is in fact more similar to ingressivity than to continuativity.14

How do the templates in Figure 1, Figure 2 and Figure 3 help the annotator? In two ways: first, the annotator can test whether the viewpoint is present in the translation by checking whether the situation in question also holds without a given marker. If so, they are dealing with continuativity; if not, with either ingressivity or resumptivity. Second, they can draw on the varying complexity of the viewpoints to motivate their annotation. The ingressive viewpoint construes only one temporal boundary, the continuative viewpoint two, the resumptive viewpoint three. Only if these complex aspectual structures are encoded by a given means of expression, can the translation be said to be marked for that type of viewpoint aspect. Taken together, working with these templates gives the annotator explicit and transparent schemas to fall back on when determining whether a specific type of viewpoint aspect is explicitly marked in a translation item in the corpus.

5. Corpus Assembly and Annotation

This study is part of a larger project aiming to develop and apply HTM in the domain of viewpoint aspect, with special attention for Mandarin and Dutch as aspect-oriented and non-aspect-oriented languages. The parallel corpus that the present study is based on is being assembled as part of this project. Corpus assembly is currently in an intermediate stage: five full novels and their translations have been digitized (via scanning and OCR) and aligned sentence-by-sentence, out of a planned twelve novels total. This study is the first application of HTM to a corpus of this size.

Table 2. Novels and translations in the parallel corpus.

Original	Author	Translation	Translator(s)
桑青与桃红 Sangqing yu Taohong ‘Mulberry and peach’	聂华 Nie Hualing	Twee Chinese vrouwen ‘Two Chinese women’	Anne Sytske Keijser
孽子 Niezi ‘Unworthy sons’	白先勇 Pai Hsien-Yung	Jongens van glas ‘Boys made of glass’	Mark Leenhouts
只有一个太阳 Zhi you yige taiyang ‘There’s only one sun’	张洁 Zhang Jie	Er is maar één zon ‘There’s only one sun’	Koos Kuiper
十爱 Shi ai ‘Ten loves’	张悦然 Zhang Yueran	Tien liefdes ‘Ten loves’	Various, edited by Annelous Stiggelbout
第七天 Di qi tian ‘The seventh day’	余华 Yu Hua	De zevende dag ‘The seventh day’	Jan De Meyer

lists the Mandarin novels and Dutch translations with their authors and translators.

In its current state, the parallel corpus comprises 29,668 aligned sentence pairs: 611,232 characters on the Mandarin side, 413,094 words on the Dutch one. The basis for determining sentence pairs was orthographic: the full stop (“。” in Mandarin orthography) was taken as a boundary to distinguish corpus items. But as became clear during the alignment process, sentence boundaries often shift in translation; shorter Mandarin sentences are sometimes sewn together by Dutch translators, and long-winded Mandarin sentences are generally broken up into two or more separate clauses in Dutch. To ensure the quality of alignment, all sentence pairs in the corpus were therefore aligned manually.

With the parallel corpus for this study in place, the next step was to extract all instances of 起来 -qilai and 下去 -xiaqu. For -xiaqu this was simple enough, as it cannot be split. However, for split cases of -qilai a number of tokens had to be allowed between qi and lai. To extract as many instances as possible while also limiting noise, a maximum of three tokens was allowed. This yielded a total of 1074 items featuring qi[]{0,3}lai and 230 items featuring xiaqu. Noise—that is, items that did not correspond to either the non-aspectual or aspectual uses of xiaqu or qilai discussed in Section 4—was very rare: none for -xiaqu, only seven for -qilai. Sentence (21) is an example of noise: in this sentence, qi is part of the word 一起yiqi ‘together’, and lai is part of the verb 过来guolai ‘come over’.

(21)	“我们是一起过来的。”男声回答。
	“Women	shi	yiqi	guolai	de.”	Nan	sheng	huida.
	we	be	together	come.over	sub	man	voice	answer
	‘ “We came here together,” the male voice answered.’

The first annotation round concerned coding all 1304 items for the usage types laid out in Section 4: lexical (qilai ‘stand up’, xiaqu ‘go down’), directional (‘up’ and ‘down’) and aspectual. For -qilai there was also the leftover category ‘opaque’ for items corresponding to none of these uses (e.g., middles). Finally, for -xiaqu the specific type of aspect had to be determined: continuative or resumptive. Sometimes this became clear from the immediate co-text (e.g., (18)–(20)), sometimes the wider co-text had to be consulted. Sentence (22a) is an example of a co-textually determined ‘continuative’ annotation: the direct co-text was not entirely conclusive, so the sentence directly before it—(22b)—had to be checked in order to pin down the aspectual type. In this way, all items could be annotated with certainty.

(22)	a.	她仍然用中文不停地说下去。
		Ta	rengran	yong	Zhongwen	bu	ting	de	shuo-xiaqu.
		she	still	use	Chinese	not	stop	sub	speak-xiaqu
		‘She continues/resumes speaking in Chinese without stopping.’
	b.	他打了几次手势叫桃红住嘴。
		Ta	da-le	ji-ci	shoushi	jiao	Tao	Hong	zhuzui.
		he	hit-prf	several-clf	gesture	call	Tao	Hong	shut.up
		‘He gestures to Tao Hong to stop talking.’

Table 3. Types of -qilai in the corpus, absolute and relative frequencies.

Type	n	%
Ingressive	452	42.1%
Directional	400	37.2%
Opaque	115	10.7%
Resultative	85	7.9%
Lexical	15	1.4%
Noise	7	0.7%
Total	1074	100%

and

Table 4. Types of -xiaqu in the corpus, absolute and relative frequencies.

Type	n	%
Directional	104	45.2%
Continuative	56	24.3%
Resultative	40	17.4%
Resumptive	26	11.3%
Lexical	4	1.7%
Noise	0	0%
Total	230	100%

show the annotation results from the first round for both markers. Everything except the ingressive, continuative and resumptive categories will be left out of consideration from this point on.

An additional concern for the annotation of continuative -xiaqu raised in Section 4 involves negation, which—because it results in the interpretation ‘stopping’—is not itself continuative. Within the continuative subtype, negation at the clause-level was therefore also annotated. There were 10 negated continuative items total, which were not included in the second round of annotation. The total number of tokens to be involved in the analysis is listed in

Table 5. Total number of tokens of aspectual -qilai and -xiaqu to be involved in HTM analysis.

Aspect Marker	n
-qilai	452
-xiaqu	72
Total	524

.

Then came the second annotation round—the core business of HTM: determining the expression(s) that the aspectual content (viz., ingressive, continuative, resumptive) was translated with. Following from the basic assumptions laid out in Section 2, not all content is necessarily translated or marked. The difference between the two is the following. Untranslated items either leave out the part of the clause with the aspect marker in the translation or deviate so strongly from the original structure that the locus of viewpoint aspect cannot be pinpointed. Unmarked items follow the original structure but do not mark the viewpoint aspect explicitly.

Sentence (23) exemplifies an untranslated item, (24) an unmarked one. From here on out, the Dutch translation from the parallel corpus will be given under (a), and my own English translation of the Dutch translated corpus item under (b). In (23), the reporting clause containing -qilai was simply left out in the translation; in (24), the verb marked by -qilai (尖叫jianjiao ‘shriek’) was translated to riep uit ‘exclaimed’, without any marking.

(23)		“是匈奴吧？”吴每笑了起来。
		“Shi	Xiongnu	ba?”	Wu	Min	xiao-le-qilai.
		be	Huns	sug	Wu	Min	laugh-prf-qilai
	a.	‘Ah, Hunnen dan.’
	b.	‘Ah, the Huns then.’

(24)		他看着就尖叫起来: […]
		Ta	zhe	jiu	jianjao-qilai:
		he	look-imp	just	shriek-qilai
	a.	‘Al lezende riep hij uit: […].’
	b.	‘While reading he exclaimed: […].’

All remaining corpus items (i.e., neither untranslated nor unmarked) were marked for ingressivity, continuativity or resumptivity in some way. The labels for their aspect marking strategies were designed and assigned in two ways. First, if the marking was the same as one of the strategies mentioned in previous work (Section 3.1 and Section 3.2), a label was assigned immediately. If not, the second way to work cyclically: first a preliminary assessment was made and later on—if the same strategy showed up more frequently—a definitive label was assigned to all of the corpus items featuring that strategy. In this way, all 524 relevant corpus items were ‘mined’ heuristically for viewpoint-aspectual means of expression. The results of this exercise will be presented in the next section.

6. Results

This section reports on the results from the HTM annotation: 452 tokens of aspectual -qilai and 72 tokens of aspectual -xiaqu were coded for the locus of viewpoint aspect in the Dutch translations. For each marker, I will present a quantitative overview of the ‘discovered’ categories, and then go over each of the categories, providing a more qualitative analysis based on specific examples. The idea is that these characterizations serve as hypotheses of the type “X encodes ingressive/continuative/resumptive aspect”, which may then be tested in future research on non-parallel corpora with a view to establishing the precise (non-)aspectual semantics and restrictions on productivity. At the end of this section, I will compare (non-)marking frequency of these types of viewpoint aspect and offer a tentative explanation for marking differences in terms of selection theory.

6.1. Ingressive -qilai

Table 6. Strategies for translating ingressive -qilai to Dutch.

Supercategories			Subcategories
Translation	n	%	Translation	n	%
unmarked	172	38.1%	unmarked	172	38.1%
beginnen ‘begin’	146	32.3%	beginnen te VINF	141	31.2%
			beginnen	5	1.1%
worden ‘become’	41	9.1%	worden A	37	8.2%
			worden N	4	0.9%
barsten ‘burst’	20	4.4%	uit-barsten PP	19	4.2%
			los-barsten	1	0.2%
gaan ‘go’	11	2.4%	gaan VINF	7	1.6%
			gaan PP	4	0.9%
komen ‘come’	10	2.2%	komen PP	4	0.9%
			komen VINF	2	0.4%
			komen A	2	0.4%
			komen	2	0.4%
krijgen ‘obtain’	10	2.2%	krijgen N	7	1.6%
			krijgen A	2	0.4%
			krijgen PP	1	0.2%
adverbial strategy	9	2%	adverbial past participle (e.g., beschaamd ‘embarrassed’)	4	0.9%
			other adverb (e.g., schor ‘hoarse’)	2	0.4%
			nu ‘now’	1	0.2%
			voor het eerst ‘for the first time’	1	0.2%
			nog nooit zo ‘never like that’	1	0.2%
other verbal strategy	9	2%	een gesprek aanknopen ‘strike up a conversation’	1	0.2%
			de oren spitsen ‘prick up one’s ears’	1	0.2%
			verdriet doen ‘give (sb.) pain’	1	0.2%
			ontvouwen ‘unfold’	1	0.2%
			oplaaien ‘flare up’	1	0.2%
			overgaan tot ‘take to’	1	0.2%
			plaatsmaken voor ‘make way for’	1	0.2%
			volgen ‘follow’	1	0.2%
			zich mengen in ‘join in’	1	0.2%
untranslated	6	1.3%	untranslated	6	1.3%
moeten ‘must’	5	1.1%	moeten VINF	5	1.1%
raken ‘get’	5	1.1%	raken PP	5	1.1%
maken ‘maken’	4	0.9%	maken A	3	0.7%
			maken PP	1	0.2%
zetten ‘put’	2	0.4%	op-zetten	1	0.2%
			in-zetten	1	0.2%
schieten ‘shoot’	2	0.4%	schieten PP	2	0.4%
Total	452	100%	total	452	100%

gives a quantitative overview of all translation strategies (including unmarked and untranslated items) for ingressive -qilai. The labels are presented at two levels of abstraction: supercategories, which constitute either the head of a syntagmatically variable construction (e.g., gaan ‘go’) or a categorial abstraction over individual translations (i.e., adverbial and other verbal strategies); and subcategories, which specify either the different types of complement selected by a head (e.g., [gaan PP] and [gaan V_INF] under gaan) or the specific strategies corresponding to an abstracted category (e.g., nu ‘now’ and voor het eerst ‘for the first time’ under adverbial strategies). Adverbial strategies are cases where the ingressive value of -qilai is encoded by means of an extra AdvP, that is, one that was not present in the original; other verbal strategies are verbs that do not select a VP- or TP-complement.

What is immediately apparent from the overview in Table 6 is that ‘unmarked’ is the largest supercategory (38.1%), although all marked cases taken together do constitute the majority (60.6%). Untranslated cases are very infrequent (1.3%). Among marked cases, strategies from previous work (Section 3.1) are well represented: beginnen (32.3%), copular worden (9.1%) and verbs from the ingressive set (gaan, komen, krijgen, moeten, raken, maken, zetten: cumulatively 7.5%) together make up nearly half of all items (48.9%). Other specific verbs taking a complement are less frequent (barsten, schieten, laten: cumulatively 5%), and least frequent are adverbial and other verbal strategies (both 2%). Overall, it seems that the Dutch linguistics literature has a decent grip on ingressivity: all mentioned strategies occur in the corpus.

I will first discuss the marked categories in order of frequency: beginnen and worden (Section 6.1.1), the ingressive set (Section 6.1.2), specific verbs selecting a complement (Section 6.1.3), and other verbal and adverbial strategies (Section 6.1.4). Last up are unmarked cases (Section 6.1.5).

6.1.1. Known Strategies from Previous Work: Beginnen and Worden

The prototypical Dutch ingressive verb, beginnen ‘begin’, is the most frequent marking strategy. It occurs both with a clausal complement headed by te ‘to’ and without (see (25) and (26)).

(25)		杨教头唰地一下，打开摺扇，扇了起来。
		Yang	jiaotou	shua	de	yixia,	dakai	zheshan,	shan-le-qilai.
		Yang	master	rustle	sub	once	open	fan	fan-prf-qilai
	a.	‘Meester Yang sloeg zijn waaier open en begon ermee te wapperen.’
	b.	‘Master Yang opened his fan and started fanning with it.’

(26)		阁楼屋顶的声音又响起来了。
		Gelou	wuding	de	shengyin	you	xiang-qilai-le.
		attic	roof	sub	sound	again	make.sound-qilai-prf
	a.	‘De geluiden op zolder beginnen weer.’
	b.	‘The sounds in the attic start again.’

Copular worden ‘become’ is also present, with both adjectival and nominal complements (see (27) and (28)). As a copula, worden attributes the property denoted by the adjective or noun to its subject (in (27) and (28): boos ‘angry’ and een cultuurland ‘a cultured country’); specifically, it signals the start of that attribution.

(27)		我气起来就打了他一记耳光，
		Wo	qi-qilai	jiu	da-le	ta	yi-ji	erguang,
		I	angry-qilai	just	hit-prf	him	a-clf	slap.on.the.face
	a.	‘Ik werd kwaad en gaf hem een draai om zijn oren...’
	b.	‘I got upset and gave him a whack around the ears…’

(28)		幸亏文化故国至少在一百年内还不会很快地文化起来，
		Xinkui	wenhua	guguo		zhishao		zai	yibai-nian-nei
		fortunately	culture	ancient country		at least		at	one.hundred-years-within
		hai	bu	hui	hen		kuai		de	wenhua-qilai,
		still	not	will	very		quick		sub	culture-qilai
	a.	‘Gelukkig zou dit land met zijn oude cultuur tenminste de eerstkomende honderd jaar niet gauw weer een cultuurland worden.’
	b.	‘Fortunately, this country with its ancient culture wouldn’t become a cultured country for at least the next one hundred years.’

These findings underscore previous work on ingressivity in Dutch, but do not add anything new. The next section will offer findings that are (partially) novel.

6.1.2. Verbs from the Ingressive Set: Gaan, Komen, Krijgen, Moeten, Raken, Maken, Zetten

Seven out of the fourteen verbs in the Dutch ingressive verb set (Section 3.1) are represented in the corpus. Crucial about previous work on the ingressive verb set is that these verbs are said to be in a syntagm with a situational PP headed by the preposition aan (i.a., Van Pottelberge 2004; Bogaards et al. forthcoming; Bogaards under review). However, these verbs also take other types of complements in the corpus: PPs headed by other prepositions, verbs, nouns and adjectives.

Table 7. Types of complements selected by verbs from the ’ingressive set’.

Verb	Type of Complement
	PP	VINF	N	A
gaan ‘go’	aan het werk gaan ‘start working’ op de vuist gaan ‘start a fight’	gaan vechten ‘start fighting’ gaan zitten ‘sit down’	-	-
komen ‘come’	op gang komen ‘get going’ in beweging komen ‘start moving’	-	-	los komen ‘get loose’
krijgen ‘obtain’	het aan zijn hart krijgen ‘start having problems with one’s heart’	-	ruzie krijgen ‘get into a fight’ moed krijgen ‘become brave’	het druk krijgen ‘become busy’
moeten ‘must’	-	moeten lachen ‘start laughing’	-	-
raken ‘get’	aan de praat raken ‘end up chatting’ in de war raken ‘become confused’	-	-	-
maken ‘make’	aan het hoesten maken ‘make (sb.) cough’	-		lui maken‘ make (sb.) lazy’ bedroefd maken ‘make (sb.) sad’

gives an overview of the specific combinations, providing one or two examples for each one. I will not go into each individual combination here, but the upshot is that the verbs from the ingressive set license a wider set of complements than just aan-PPs. Some of these complements may in fact be more frequent than PPs headed by aan, given that they are attested here contrary to many combinations with aan. As for the verbs themselves, it is striking that only one causative and one modal were found in the corpus (moeten and maken).15

Finally, zetten also occurs in the corpus, but prefixed with in- or op- (i.e., inzetten ‘set in’ and opzetten ‘put on’). These have ingressive meaning by themselves. One of them is illustrated in (29).

(29)		他的声音也变得幼稚温柔起来。
		Ta	de	shengyin	ye	biande	youzhi	wenrou-qilai.
		he	sub	voice	also	change	childish	tender-qilai
	a.	‘Hij zette er ook een lief kinderstemmetje bij op.’
	b.	‘He also put on a sweet childish voice.’

In sum, the parallel corpus reveals that the ‘ingressive verb set’ in Dutch is more heterogeneous than previously thought.

6.1.3. Other Verbs Selecting a Complement: Barsten, Schieten

The corpus also features three complement-selecting verbs not in the ingressive verb set: barsten ‘burst’, schieten ‘shoot’ and laten ‘let’. The first two of these combine with PPs denoting laughter, crying or related actions, such as in de lach ‘lit. in the laugh’, in huilen ‘lit. in crying’, and in gillen ‘lit. in shouting’. Barsten is prefixed with uit- ‘out’ in these cases (i.e., uitbarsten ‘burst out’). The punctual semantics of ‘bursting’ and ‘shooting’ appears to map to the initial boundary of these situations. This might be facilitated by the fact that the initiation of laughter and crying is usually accompanied by some vehemence. This is likewise lexicalized in English by ‘bursting into laughter/tears’. Corpus items (30) and (31) illustrate schieten and uitbarsten.

(30)		他们两个互相看看后又嘻嘻笑起来。
		Tamen	liang-ge	huxiang	kankan-hou	you	xixi	xiao-qilai.
		they	two-clf	each.other	look-after	again	happy	laugh-qilai
	a.	‘De twee keken elkaar aan en schoten opnieuw in de lach.’
	b.	‘The two of them looked at each other and burst into laughter again.’

(31)		听到这里，司马南江忍不住大笑起来。
		Tingdao	zheli,	Sima	Nanjiang	renbuzhu	daxiao-qilai.
		hear	here	Sima	Nanjiang	cannot.help	big.laugh-qilai
	a.	‘Toen Sima Nanjiang dat hoorde, kon hij zich niet inhouden en barstte in hard lachen uit.’
	b.	‘When Sima Nanjiang heard that, he couldn’t help himself and burst out in loud laughter.’

Barsten and schieten appear to impose selectional restrictions with regard to their PP-complement (e.g., one can in lachen uitbarsten ‘burst out in laughter’ but not in eten uitbarsten ‘burst out in eating’).

6.1.4. Adverbial and Other Verbal Strategies

Certain adverbs and verbs in the corpus appear to lexicalize focus on a situation’s initial boundary. Examples of such adverbs are nu ‘now’ and voor het eerst ‘for the first time’. A different adverbial strategy is to relegate the Mandarin main verb marked by -qilai to the status of adverbial modifier—for instance, as a past participle—and attribute a property to the former verb’s subject that way. An ingressive interpretation is still possible in these cases, though, as (32) illustrates: by functioning as an adverb to the main verb liet zakken ‘lowered’, beschaamd ‘embarrassed’ allows for the interpretation that the state of ‘being embarrassed’ started at this point, rather than holding already.

(32)		吴敏不好意思起来，头一直俯着。
		Wu	Min	buhaoyisi-qilai,	tou	yizhi	fu-zhe.
		Wu	Min	feel embarrassed-qilai	head	straight	look.down-imp
	a.	‘Wu Min liet beschaamd zijn hoofd zakken.’
	b.	‘Wu min lowered his head in shame.’

Verbs or verbal expressions lexicalizing ingressivity in the corpus include de oren spitsen ‘prick up one’s ears’, ontvouwen ‘unfold’, plaatsmaken voor ‘make way for’ and een gesprek aanknopen ‘strike up a conversation’, which is illustrated in (33).

(33)		并且不经他人介绍就和司马南江交谈起来。
		Bingqie	bu	jing	taren	jieshao	jiu	he	Sima	Nanjiang	jiaotan-qilai.
		and	not	through	another	introduce	just	with	Sima	Nanjiang	chat-qilai
	a.	‘En zelfs zonder aan hem voorgesteld te zijn een gesprek met hem aanknoopten.’
	b.	‘And even without having been introduced to him, struck up a conversation with him.’

Means of expression of this type, where ingressivity is subtly lexicalized, are the most difficult to identify using introspection because they are not very frequent. Indeed, all of them are hapaxes in the parallel corpus. This is where HTM is of clear added value. With a larger parallel corpus, there would likely be more lexicalized expressions ‘discovered’.

6.1.5. Unmarked Cases

The largest category of translations for ingressive -qilai is unmarked. Sentences (34)–(36) are examples of these unmarked corpus items.

(34)		屋顶的声音又响起来了。
		Wuding	de	shengyin	you	xiang-qilai-le.
		roof	sub	sound	again	make.sound-qilai-prf
	a.	‘Het geluid op het dak is weer terug.’
	b.	‘The sound on the roof is back again.’

(35)		郭老摇头笑了起来。
		Guo	lao	yaotou	xiao-le-qilai.
		Guo	old	shake head	laugh-prf-qilai
	a.	‘Hij lachte hoofdschuddend.’
	b.	‘He laughed, shaking his head.’

(36)		他突然沉默起来，
		Ta	turan	chenmo-qilai.
		he	suddenly	silent-qilai
	a.	‘Opeens zweeg hij.’
	b.	‘Suddenly he became silent.’

In (34), the locus of ingressivity is the stative predicate is terug ‘is back’, in (35) the durative and atelic predicate lachte ‘laughed’, and in (36), the punctual and telic predicate zweeg ‘became silent’. Their unmarked ingressive interpretations in Dutch appear to be licensed given the right context, and provided that it is clear what the initial boundary should roughly look like. However, with the exception of (36), which independently features an adverb suggesting ingressivity (i.e., 突然 turan ‘suddenly’), these items could also receive non-ingressive readings where ‘being back’ and ‘laughing’ are simply ongoing. It thus seems that, at least in the parallel corpus, Dutch is more often underspecified for ingressivity than Mandarin.

6.2. Continuative/Resumptive -xiaqu

The two types of viewpoint aspect encoded by -xiaqu—continuative and resumptive—were treated separately for the purposes of annotation (cf. Section 4 and Section 5).

Table 8. Strategies for translating continuative -xiaqu to Dutch.

Supercategories			Subcategories
Translation	n	%	Translation	n	%
door- ‘through’	14	30.4%	doorgaan (te/met VINF)	9	19.6%
			door-V	5	10.9%
verder- ‘further’	11	23.9%	verder V	10	21.7%
			verdergaan met VINF	1	2.2%
blijven ‘stay’	10	21.7%	blijven VINF	8	17.4%
			blijven door-VINF	1	2.2%
			aan-blijven	1	2.2%
adverbial strategy	5	10.9%	aan een stuk door ‘continuously’	2	4.3%
			steeds ‘continually’	1	2.2%
			nog langer ‘even longer’	1	2.2%
			in één adem ‘in one breath’	1	2.2%
untranslated	3	6.5%	untranslated	3	6.5%
unmarked	1	2.2%	unmarked	1	2.2%
houden ‘keep’	1	2.2%	uit-houden	1	2.2%
voort- ‘forward’	1	2.2%	voort-V	1	2.2%
Total	46	100%	Total	46	100%

lists the translation strategies for its continuative sense, according to the same format as -qilai: abstracted supercategories subdividing into subcategories. Contrary to the ingressive overview, the most frequent strategies are in fact not known from previous work (Section 3.2): the verbal prefixes door- ‘through’ and verder- ‘further’, which make up more than half of all items (54.3%), are thus an aspectual ‘discovery’ in this sense. The verbs from the continuative verb set (blijven and houden: cumulatively, 23.9%) are represented, but not with the types of PP-complement discussed in the literature. There are a few adverbial strategies (10.9%), and just one unmarked case (2.2%). Another prefix not known from the literature—voort- ‘forward’—also occurs, but only once (2.2%). Conspicuously absent from the corpus with previous work in mind is the en maar-construction (cf. Section 3.2); this is likely because it is a colloquial construction not or rarely found in the literary genre (although it could have popped up in embedded dialogue).

For resumptive -xiaqu, the quantitative picture in

Table 9. Strategies for translating resumptive -xiaqu to Dutch.

Supercategories			Subcategories
Translation	n	%	Translation	n	%
verder- ‘further’	17	65.4%	verder V	14	53.9%
			verdergaan met VINF	3	11.5%
ingressive verb + adverbial strategy	5	19.2%	weer gaan ‘go again’	3	11.5%
			weer beginnen ‘begin again’	2	7.7%
other verbal strategy	2	7.7%	vervolgen ‘proceed’	2	7.7%
door- ‘through’	1	3.8%	doorgaan met VINF	1	3.8%
voort- ‘forward’	1	3.8%	voort-V	1	3.8%
N	26	100%	N	26	100%

is largely but not entirely different. The prefixes verder- and door- are both represented, but in a rather different distribution: verder- (65.4%) is almost three times as frequent (65.4/23.9 = 2.7), and there is only one instance of resumptive door- (3.8%). A chi-squared test (with Yates correction for small sample sizes) shows that the difference in the distribution of continuative and resumptive interpretations over door- and verder- is significant (χ² = 9.61, df = 1, p = 0.002).16 I will return to this finding in Section 6.2.5. Voort- ‘further’ occurs once here too, and there is one instance of a verbal strategy: vervolgen ‘proceed’ (both 3.8%). A notable strategy for resumptivity is the use of an ingressive verb (gaan, beginnen) combined with the adverb weer ‘again’ (19.2%). This aligns with the idea voiced in Section 4.4 that resumptivity basically constitutes a special type of ingressivity embedded in a more complex viewpoint. A final observation is that there are zero cases of untranslated or unmarked resumptivity.

In what follows, the markers from Table 8 and Table 9 will be discussed in order of frequency: prefixes (Section 6.2.1), verbs from the continuative set (Section 6.2.2) and adverbial and other verbal strategies (Section 6.2.3). The section concludes with a discussion of the most frequent markers from Table 8 and Table 9 in terms of their overlap and specialization with regard to continuativity and resumptivity.

6.2.1. Prefixes: Verder-, Door-, Voort-

The most striking ‘discovery’ of the present HTM exercise is the prevalence of prefixes not mentioned in the literature on Dutch viewpoint aspect: verder- ‘further’, door- ‘through’ and voort- ‘forward’. Verder- and voort- occur in both continuative and resumptive interpretations. Sentence (37) exemplifies a continuative interpretation of verder-, (38) a resumptive one. Formally, verder- may either select a main verb (e.g., verder leven ‘keep living’ in (37)) or occur in the combination verdergaan met (e.g., verdergaan met praten ‘resume talking’ in (38)). Voort- seems to be more limited in what it attaches to, but the precise restrictions are not clear from the parallel corpus.

(37)		有一次我听到女的一边哭泣一边说，再也不愿意和他这个穷鬼过下去了，
		Youyici	wo	tingdao	nüde	yibian	kuqi	yibian	shuo,	zaiyebu
		once	I	hear	woman	one.side	weep	one.side	speak	not.any.more
		yuanyi	he	ta	zhe-ge	qiong	gui	guo-xiaqu-le.
		wish	with	him	this-clf	poor	ghost	live-xiaqu-prf
	a.	‘Op een keer hoorde ik haar huilend zeggen dat ze niet meer verder wilde leven met een armoedzaaier zoals hij.
	b.	‘One time I heard her cry and say that she didn’t want to keep on living with a bum like him.’

(38)		我上了炕还得说下去。
		Wo	shang-le	kang	hai	dei	shuo-xiaqu.
		I	upon-prf	kang	still	must	speak-xiaqu
	a.	‘Als ik op de kang lig wil ik verdergaan met praten.’
	b.	‘When I lay down on the kang I want to resume talking.’

Formally, the prefix door- mirrors verder-, occurring attached to a main verb or in the combination doorgaan met. Door- has a clear preference for continuativity, with only one attestation being resumptive. Both interpretations are illustrated below: (39) continuative, (40) resumptive.17

(39)		他们再那样子谈下去，我就要下车了。
		Tamen	zai	na	yangzi	tan-xiaqu,	wo	jiu	yao	xiache-le.
		they	again	that	way	chat-xiaqu	I	just	want	get off-prf
	a.	‘Als zij in deze trant doorpraten, wil ik eruit.’
	b.	‘If they keep on talking in this way, I want to get off.’

(40)		她继续唱下去。
		Ta	jixu	chang-xiaqu.
		she	proceed	sing-xiaqu.
	a.	‘Ze gaat door met zingen.’
	b.	‘She resumes singing.’

Further research into these prefixes is desirable, since they have received little treatment in previous work—much in favor of the verb blijven, which is up next.

6.2.2. Verbs from the Continuative Set: Blijven, Houden

In previous work (Section 3.2), blijven ‘stay’ was associated with continuativity, and in various syntagms: with an infinitive, a PP headed by aan ‘on’, and as a copula with a nominal or adjectival complement. It is, however, only attested in the corpus with an infinitive, and it is overall less frequent than two of the prefixes discussed before (door- and verder-). More strikingly, blijven is restricted to continuative (i.e., non-resumptive) interpretations, which is not made explicit in previous work. All of this suggests that blijven occupies a more central position in the Dutch viewpoint aspect literature than is empirically justified.

Blijven’s causative counterpart houden ‘keep’ is likewise restricted to non-resumptive continuativity. Houden is attested once, also not with the type of complement mentioned in the literature (aan-PP), but with a prefix that allows it to be used non-causatively (i.e., uithouden ‘endure’). This attestation is shown in (41).

(41)		“我们撑船人可以这样子活下去，你们舱客就不能吗？”
		“Women	chengchuan	ren	keyi	zheyangzi	huo-xiaqu,	nimen
		we	punt.boat	people	can	like.this	live-xiaqu	you.pl
		cang	ke	jiu	bu	neng	ma?”
		cabin	customer	just	not	can	qst
	a.	‘Als wij vaargasten het zo kunnen uithouden, kunnen jullie passagiers dat toch ook?’
	b.	‘If us sailors can endure it like this, you passengers should be able to as well, right?’

6.2.3. Adverbial and Other Verbal Strategies

Continuative and resumptive viewpoint aspect are realized in the translations by distinct adverbial and/or other verbal strategies. For continuativity, there are four adverbs in the corpus, all of which boil down to ‘not stopping during the situation’: aan een stuk door ‘continously’, steeds ‘continually’, in één adem ‘in one breath’, and nog langer ‘even longer’, which is illustrated in (42).

(42)		再盯下去我就变成姨太太了。
		Zai	ding-xiaqu	wo	jiu	biancheng	yitaitai-le.
		again	stare-xiaqu	I	just	change	concubine-prf
	a.	‘Als je me nog langer aanstaart zal ik in de concubine veranderen.’
	b.	‘If you stare at me me any longer I’ll transform into the concubine.’

For resumptivity there is one case of this type of strategy, but a notable one: the use of an ingressive verb (beginnen or gaan—cf. Section 6.1.1) with an adverb indicating the return of a former situation (i.e., weer ‘again’). This speaks for the analysis of resumptivity as a special type of ingressivity with additional complexity; precisely this complexity (i.e., the situation held previously) is contributed by weer. Corpus item (43) illustrates this.

(43)		她将钱收到裙子口袋里，推起她的车子，顶着风，吃力地行走下去，
		Ta	jiang	qian	shoudao	qunzi	koudai-li,	tui	qi	ta
		she	take	money	receive	skirt	pocket-in	push	rise	she
		de	Chezi,	ding-zhe	Feng,	chili	de	xingzou-xiaqu.
		sub	cart	go.against-imp	wind	strain	sub	walk-xiaqu
	a.	‘Ze stak het geld weg in haar rok en begon weer moeizaam haar kar tegen de wind in te duwen.’
	b.	‘She put the money away in her skirt and started to push her cart again, with difficulty, against the wind.’

6.2.4. Unmarked Case

There is one unmarked continuative case in the corpus, and zero unmarked resumptive ones. How can continuativity go unmarked? Given the conceptual structure of continuativity schematized in Section 4.4, there are at least two co(n)textual prerequisites: (i) the situation may not have been interrupted and (ii) the continuative (i.e., ‘not stopping’) boundary must somehow be salient, otherwise it would be necessary to mark it. The one unmarked case, shown in (44), indeed meets these criteria: the situation—i.e., 活huo ‘live’—clearly cannot be interrupted, and the sentences preceding it specify that anaphoric er ‘it’ refers to afflictions of old age.18 Hence, a kind of abstract ‘not stopping’ boundary can be understood from the transition from not suffering to suffering from these afflictions.

(44)		要不他们怎么活下去。
		Yaobu	tamen	zenme	huo-xiaqu.
		otherwise	they	how	live-xiaqu
	a.	‘Hoe viel er anders mee te leven?’
	b.	‘How else would they be able to live with it?’

Generally, though, continuativity is basically always marked in the Dutch translations—contrary to ingressivity. Section 6.3 will speculate on why this might be.

6.2.5. Specialization and Overlap in Continuativity and Resumptivity

Aspectual -xiaqu covers both continuativity and resumptivity. The most frequent Dutch translation strategies (verder-, door-, blijven) differ in this respect. To get a better grip on this variation,

Table 10. Frequency of verder-, door- and blijven encoding continuative/resumptive aspect in the corpus.

Marker	Continuative		Resumptive		Total
	n	%	n	%	n	%
verder- ‘further’	11	39.3%	17	60.7%	28	100%
door- ‘through’	14	93.3%	1	6.7%	15	100%
blijven ‘stay’	10	100%	0	0%	10	100%

shows the frequency of the Dutch strategie encoding continuativity or resumptivity.

Table 10 makes it clear that Dutch verder- covers both types, although it prefers resumptivity. As shown at the start of Section 6.2, this association is statistically significant. The other two markers, on the other hand, have a clear preference for continuativity: door- gets a resumptive interpretation only once, as opposed to 14 continuative interpretations; blijven is restricted to continuativity.

So contrary to Mandarin -qilai and the Dutch ingressive strategies, which differed chiefly in (non-)marking (the Dutch sentences often being underspecified for ingressivity), the Mandarin and Dutch continuative strategies vary in a different way: where -xiaqu takes care of both continuativity and resumptivity, Dutch has a specialized marker for continuativity and two markers with a strong preference for just one of the two (door- for continuativity, verder- for resumptivity).

6.3. Marking Frequency and Aspectual Complexity

HTM with at least two heuristic elements makes it possible to compare marking frequency. As the previous sections showed, there are considerable differences in the distribution of (un)marked translations over the viewpoint aspect markers: -qilai was marked less frequently than -xiaqu, and there was also a slight difference between -xiaqu’s continuative and resumptive readings. This comparison can be enhanced with marking frequencies from previous HTM research. Bogaards (2019a) applied HTM using the Mandarin “imperfective” verbal marker 着-zhe, using a similar novel/translation-based corpus and equivalent criteria for ‘unmarkedness’ and ‘untranslatedness’.

Table 11. Distribution of marked/unmarked translations over translated aspect markers19.

Type of Aspect	Marked		Unmarked		Total
	n	%	n	%	n	%
Ingressive	274	61.4%	172	38.6%	446	100%
Continuative	42	97.7%	1	2.3%	43	100%
Resumptive	26	100%	0	0%	26	100%
Imperfective	127	26.4%	354	73.6%	481	100%

shows marking frequency for not only the markers used in the present study, but also imperfective -zhe, taken from Bogaards (2019a, p. 174).

Now there is a threefold difference: in the Dutch translations, -zhe’s imperfectivity goes unmarked most of the time (73.6%), -qilai’s ingressivity is marked in the majority of items (61.4%) while -xiaqu’s continuativity is marked basically always (97.7%) and its resumptivity in all cases. Three separate chi-squared tests (with Yates correction) show that (i) the difference in marking between continuative and resumptive translations is not significant (χ² = 0.07, df = 1, p = 0.798), (ii) the difference between continuative/resumptive and ingressive and ingressive translations is significant (χ² = 35.26, df = 1, p < 0.001), (iii) as is the difference between ingressive and imperfective translations (χ² = 115.7, df = 1, p < 0.001). Thus, the marking hierarchy is imperfective > ingressive > continuative/resumptive. How can these differences in marking be accounted for? A straightforward explanation would be that Dutch grammaticalizes these notions to different degrees. But there may be a more general force at play here. I will put forward a tentative explanation for the marking differences in Table 11.

As was claimed in Section 4.4, the ingressive, continuative and resumptive viewpoints vary in aspectual complexity. That is, the basic building blocks (boundaries τ and phases ϕ) required to model each viewpoint differ in both number and organization. As the conceptual structures in Figure 1, Figure 2 and Figure 3 make clear, ingressivity is made up out of three building blocks (¬ϕ, τ, ϕ), continuativity out of five (¬ϕ, τ₁, ϕ₁, τ₂, ϕ₂) and resumptivity out of seven (¬ϕ, τ₁, ϕ₁, τ₂, ¬ϕ, τ₃, ϕ₂). The imperfective viewpoint, on the other hand, is simpler; its representation in these terms is ‘ϕ’ along a temporal axis t. In this sense, the number of building blocks required for each viewpoint mirrors the degree to which they are marked in the translations: more complexity means more marking. I would therefore like to suggest that, in the domain of viewpoint aspect, complexity correlates with obligatory marking. Clearly, this claim requires more empirical evidence to be at all convincing; I am positing it here merely as a general hypothesis. HTM offers a way of testing this type of claim, as working with translations in this manner facilitates quantifying (non-)marking of conceptual content.

7. Conclusions

This study conducted a heuristic translation mining (HTM) analysis in the viewpoint-aspectual domain—specifically: ingressive, continuative and resumptive aspect—in Mandarin Chinese and Dutch. The aims were threefold: (i) methodological, to expand HTM by working with several and more specific heuristic elements; (ii) descriptive, to heuristically map out ingressivity, continuativity and resumptivity in a non-aspect-oriented language like Dutch; and (iii) theoretical, to contribute to general research into crosslinguistic variation in encoding specific conceptual (aspectual) content.

Methodologically, the study drew on selection theories of aspect (e.g., Bickel 1997) to construct viewpoint-aspectual templates in terms of phases ϕ and boundaries τ, which were used to annotate the corpus data consistently and analyze them in a transparent way. A second finding having to do with method is that aspect markers at this level of specificity are in fact quite rare. In its current state of 29,668 parallel pairs and 611,232 characters, the corpus yielded 452 ingressive tokens, 56 continuative tokens and only 26 resumptive ones. The larger parallel corpus that is being assembled for the research project that this study is a part of is therefore by no means an unnecessary luxury. Third and last, the use of several distinct markers made it possible to compare marking frequency, which is not possible without some kind of quantifiable comparative method such as HTM.

Descriptively, the HTM analysis showed that the literature on Dutch viewpoint aspect has quite a solid grip on ingressivity, but not so much on continuativity (and not at all on resumptivity). Newly ‘discovered’ means of expression for ingressivity are:

• Other complements for the verbs from the ingressive set: various prepositional, nominal and adjectival complements;
• Other verbs taking a complement: barsten ‘burst’ and schieten ‘shoot’ selecting a PP with specific restrictions (e.g., in lachen or in de lach ‘in laughter’);
• Adverbial and other verbal strategies;
• Unmarked verbs.

For continuativity, the focus in previous work was on the verb blijven ‘stay’ and its causative counterpart houden ‘hold’. Because of that, the more frequent markers (in the present corpus) door- and verder- were overlooked, and resumptivity largely ignored. The new findings for these types of aspect in Dutch are:

• Prefixes door- ‘through’, verder- ‘further’ and voort- ‘forward’, with varying preferences for continuativity and resumptivity;
• Adverbial strategies for continuativity;
• Ingressive verb + ‘return to former situation’-adverb (e.g., weer ‘again’) for resumptivity.

These lists constitute testable hypotheses for the Dutch viewpoint-aspectual domain and as such provide direct input for further research in this field. That is to say, the hypotheses that these constructions encode ingressive/continuative/resumptive aspect may be specified by examining their (non-)aspectual semantics and restrictions on productivity. At a more general level, these lists also indicate where specific types of viewpoint aspect may be encoded in a non-aspect-oriented language: rather than in grammatical particles, they may be articulated by specific verbs, lexical and adverbial strategies, prefixes, and verb–adverb combinations. In the long run, this may also feed into linguistic theories on the crosslinguistic ordering of functional projections (e.g., cartographic approaches in the spirit of Cinque 1999); if one postulates that continuative viewpoint aspect occupies a fixed syntactic position crosslinguistically, for example, then Mandarin -xiaqu and Dutch door-, verder- and voort- present a specific case to support or challenge such an assumption. In this sense, these hypotheses are theoretically relevant for future research.

Theoretically, this study found a (non-)marking difference between Mandarin and Dutch regarding ingressivity (but not for continuativity/resumptivity). In the parallel corpus, Dutch left its verbs underspecified for ingressive aspect in 38.6% of cases. Of course, this finding is not very telling if the analysis is not reversed: the question is how frequently Mandarin translations of Dutch originals with, say, the marker beginnen are marked for ingressive aspect (and whether there are other ways of doing so than with -qilai). A more tangible finding is the difference in marking across viewpoint aspects within Dutch, especially in comparison to earlier (H)TM research into imperfectivity in Mandarin and Dutch: whereas imperfectivity was marked in only 26.4% of cases, ingressivity was marked 61.4% of the time and continuativity/resumptivity basically always. Based on this marking hierarchy, the tentative hypothesis was formulated that aspectual complexity (in terms of selection theory) correlates with obligatory marking. HTM provides the opportunity to test this type of hypothesis quantitatively and without needing to rely on intuitions.

New (aspectual) discoveries can be made by extending HTM; doing so requires assembling larger parallel corpora and being specific about the conceptual content under study. Ironing out these methodological issues makes HTM a useful empirical tool for investigating the expression of viewpoint aspect across languages.