Affectedness and Differential Object Marking in Turkish and Uzbek

2.1 DOM in Turkish and Uzbek

There are many language families that show a quite coherent behavior with respect to DOM, including the Turkic family. We will focus on two closely related Turkic languages, namely Turkish and Uzbek.

Turkish is the largest language in the Turkic language family, with the number of first language speakers (approx. 70 million) covering over a third of the Turkic language community (Göksel and Kerslake 2005; Kornfilt 1997; Lewis 1965). The Differential Object Marking patterns of Turkish have received much attention in the literature on DOM. Traditionally, Turkish is seen to exemplify differential case marking along the referentiality scale (Aissen 2003; Bossong 1985; Enç 1991; von Heusinger and Kornfilt 2005; Kamali 2015; Kornfilt 2003; Öztürk 2005).

Uzbek is the second largest Turkic language in terms of native speakers (approx. 27 million), with a rich variety of dialects spoken across the Central Asian region.^[4] In this study we focus on the Standardized Uzbek, as spoken in Uzbekistan. The role of the accusative case and the different properties of direct objects in Uzbek have been discussed in earlier works (G’ulom 1941; Kamilova 1947; Kononov 1948, 1960; Reshetov 1951; Sjoberg 1963; Tursunov et al. 1975), as well as in more recent grammars (Bodrogligeti 2003; Nurmonov et al. 1992; Rahmatullayev 2010; Sapayev 2009). These works uniformly link the semantic contrast between marked and unmarked direct objects to definiteness and indefiniteness, respectively. The investigation of Uzbek DOM within the generative framework is an emerging area. Factors that have been proposed as enhancing DOM include referentiality (Kagan 2020; Levy-Forsythe 2018; Niyazmetova 2009; Türker 2019), as well as animacy and verbal transitivity (Guntsetseg et al. 2008).

Like most other Turkic languages, Turkish and Uzbek are SOV suffixing languages. They do not have definite articles, but they have indefinite articles. We assume that Turkish has the indefinite article bir, which is derived from the numeral bir ‘one’, but differs from the numeral in various aspects: it can be deaccented and takes a different position in the nominal phrase (see Göksel and Kerslake 2005: 184–186; Kornfilt 1997: 98; Schroeder 1999: 42). Uzbek has two indefinite articles, bir and bitta. Bir is derived from the numeral bir ‘one’, and bitta is derived from bir and the classifying suffix –ta (Beckwith 1998: 127; Bodrogligeti 2002: 96; Boeschoten 1998: 363; Kononov 1948: 120; Sapayev 2009: 91; Sjoberg 1963: 25; Tursunov et al. 1975: 162). Von Heusinger and Klein (2009) provide evidence from a corpus search that both forms are indefinite articles and that bitta is becoming more frequent in the last 30 years. They also provide evidence from a questionnaire study that bitta seems to be slightly preferred with human referents. In our study, we choose the indefinite article bitta as it seemed the most adequate choice for our test items.

2.2 Nominal parameters

3.1 Affectedness

Affectedness refers to the change an argument undergoes through the event, and is a linguistic notion that figures centrally in determining transitivity and defining direct objecthood (Dowty 1991; Hopper and Thompson 1980; Tenny 1994). For Naess (2004), it is the central notion behind DOM. In her account, what is marked by accusative case is not definiteness or animacy but a high degree of affectedness that correlates with the two parameters. Following Hopper and Thompson (1980), she argues that the affectedness of an object is dependent on definiteness and animacy, as definite and animate objects are more completely affected by an action than indefinite or inanimate ones. One language for which affectedness has been linked to the synchronic and diachronic distribution of DOM is Spanish. In Spanish, DOM with indefinite human direct objects is optional and depends on specificity and on the potential to introduce a discourse referent (Leonetti 2004). On the basis of examples, as in (10) with the verb golpear ‘beat up’, Torrego (1998, 1999 argues that for Modern Spanish, DOM is obligatory with affected objects. Building on work by Hale and Keyser (1992) and Marantz (1993), she follows a structural account to DOM and associates affected direct objects with a certain syntactic position, namely the specifier of VP.

The role of affectedness in the diachrony of Spanish DOM is investigated in von Heusinger and Kaiser (2011). The authors provide extensive corpus data that shows a correlation between verb classes high on Tsunoda’s (1985) Affectedness scale and the diachronic spread of Spanish DOM.

However, the emerging literature on event structure and DOM is also accompanied by controversies. Sceptics of a DOM approach based on affectedness have criticized that affectedness is a hard notion to measure and often rather intuitively defined (Fábregas 2013: 27; de Hoop 2015). In the following, we will sketch the notion of affectedness in more detail and ultimately classify our verb classes along Beavers’ (2011) Affectedness Hierarchy that provides a semantic characterization of degrees of affectedness based on monotonically weakening truth conditions.

Affectedness is rooted in the lexical semantics of a verb and refers to the entailment of a changing property in the theme argument due to the event, i.e., a change of location, change of state, change of extent etc. Change can be conceived as a new (target) state ψ that obtains for some entity x due to the event (Kratzer 2000) and semantically modelled along a path or scale that measures the progress of the event. In functional typological literature, change in the affected argument has been defined in maximal semantic distinction to, and as a direct result of, a volitional agent instigating the event (Naess 2004). In (11a) but not (11b) the direct object undergoes a change of state or location.

Well-established diagnostics for affectedness are the What happened to X is Y test (Cruse 1973; Jackendoff 1990), cf. (12a)–(12b), the entailment of change test, cf. (12c)–(12d), and resultative predication as shown in (12e)–(12f) (see Beavers 2011 for discussion of affectedness tests).

Affectedness has been modelled as a graded phenomenon with four degrees by Beavers (2011) (Figure 1). His hierarchy starts at the lowest degree with see x type predicates that do not encode any change in x through the event taking place. In contrast, contact verbs like hit x encode that the theme x may potentially change, followed by a group of widen x type verbs, whose meaning requires that x necessarily changes in some property even though it does not encode a result state (widen x ↛ x is wide). The highest degree is encoded in break x type verbs which encode change in the theme argument that ends with a result state (break x → x is broken), and as such are inherently telic.

Figure 1:

Affectedness hierarchy following Beavers (2011).

We investigate affected telic and affected non-telic predicates, but set aside potentially affected ones, as, in the languages investigated, they do not uniformly select for the accusative but often select the dative case frame. Furthermore, we differentiate for telicity in the predicates underspecified for change by contrasting find x or win x type telic verbs with wait for x or see x type atelic verbs.

Within a certain group of contact by impact verbs, Lundquist and Ramchand (2012) link affectedness to animacy. According to the authors, animacy can mediate the ability of an argument to be interpreted as affected. As mentioned above, Naess (2004) also correlates animacy and affectedness. We take this as motivation to investigate animacy and control for this parameter alongside affectedness. Note that our verb classes differ from Lundquist and Ramchand’s (2012), as we exclude contact by impact verbs which fall under the potentially affected class.

3.2 Telicity

Telicity refers to a natural end point of an event and has also been associated with DOM in Spanish, but has not been taken as a unified principle underlying DOM like affectedness. Torrego’s (1998: 17–21) argumentation with respect to the role of telicity for Spanish DOM is two-fold: First, for telic verbs like accomplishments and achievements in the sense of Vendler (1957), she argues that the DOM marker is obligatory. Second, for atelic verbs, case marking is optional and the DOM marker shifts an event from an atelic to a telic interpretation. Consider her example, which we report in (13). (13a) allows for the interpretation of a single, resultative event with the result state lasting for two years. Such an interpretation is ruled out in the (13b) version that can only be understood as involving repeated events of hiding a prisoner in the scope of two years.

Torrego’s claims have given rise to a debate on telicity and DOM; see Fábregas (2013: 25–26) and Delbecque (2002: 95–97) for conflicting speaker judgements and further critical discussion. Based on a diachronic corpus search, Carbajal (2008) and Romero Heredero (2021) argue against telicity as an independent factor involved in the evolution of DOM in Spanish. Romero Heredero (2021) suggests that earlier claims regarding the role of telicity and DOM may stem from an interaction with affectedness. As affectedness and telicity are interrelated, we will first characterize telicity, and then briefly discuss the interdependency claimed between the two notions.

Telicity or resultativity is related to the boundedness of events, and is found in predicates involving an endpoint or a final result state at which the event culminates. It can be probed for by in/for adverbial modifiers. Telic predicates favor in x time over for x time adverbials, highlighting results over processes and vice versa for atelic predicates. The telicity of a transitive predicate is conditioned by various factors such as the lexical semantics of the verb (Beavers 2011), cf. (14a)–(14b), real-world knowledge (Hay et al. 1999) and the quantizedness of the theme argument’s reference (Krifka 1989). Specific, definite arguments yield telic predicates (14a), whereas mass nouns and bare plurals yield atelic predicates (14c).

In more formal accounts, telicity is determined by the subinterval property (Krifka 1989), which states that whenever a telic event is true at a time interval, it is not true at any part of that interval. For a telic event, as in (14a), a subpart of the predicate ruin the picture does not require the picture to be ruined. Conversely, for the atelic events in (14b)–(14c), a subinterval of the predicate push the cart does require the cart to be pushed, and likewise, a subpart of the predicate ruin pictures can also be described as ruin pictures.

While affectedness has been argued to comprise of the defining property of patienthood, telicity has also been linked to the accusative case frame. Building on Kiparsky’s (1998) work on the connection between boundedness and accusative case in Finnish, Kratzer (2004) argues for an agreement relationship between telicity and accusative case. In her account, telicity is a verbal feature that requires the event to culminate with respect to a measure provided by the direct object referent. It is not a mere property of events but a relation between individuals and events. In dynamic verbs, properties of the direct object that undergoes change and marks the temporal end point of the event figure centrally in determining telicity.

In Tenny’s (1994) seminal work, argument linking at the interface of lexical semantics and syntax is aspectually constrained. Affectedness and telicity figure together, as an affected argument is defined by two properties: undergoing change and marking the temporal end of the event (Tenny 1994: 158). Under the semantic analysis of affectedness by Beavers (2011), change in the theme argument that ends with a result state contributes to a higher degree of its affectedness. An interdependence between affectedness and telicity is also found in Ramchand (2008), where the addition of a result is caused by a dynamic event expressing a changing property predication for the affected argument.

While in general there is little controversy that affectedness and telicity are both distinguished notions prominently related to events and their arguments, the open question that remains is whether grammatical phenomena that concern the direct object are driven by both notions independently or whether there exists a dependency relation amongst them. To disentangle the two notions and at the same time explore their interaction, we work with four subclasses of verbs that exhibit the four logical value combinations of affectedness and telicity, cf. (15α)–(15δ).

3.3 Affectedness and telicity in Turkish and Uzbek

In Turkish, the accusative case frame has been argued to encode higher degrees of affectedness than the dative or ablative case frames (Çetinoğlu and Butt 2008; Dede 1981). Nakipoğlu (2009) suggests a correlation between the accusative case marker and boundedness in definite DPs. With the exception of a short discussion on accusative case and aspectual verb class in Travis (2010: 148–151), there exists no work investigating the relation between DOM marking of indefinite direct objects and affectedness or telicity in either Turkish or in Uzbek.

Motivation to investigate verbal properties and DOM of indefinite direct objects in Turkic comes from the data in (16) which instantiates distinct case marking patterns of the direct object with different types of verbs. (16a)–(16b) illustrates example predicates with affected direct objects. The actions of ruining x or pushing x necessarily denote a change of state or location respectively in the theme argument, unlike finding x or waiting for x in (15g)–(15d). In turn, (16a) and (16c) differ from (16b) and (16d) in telicity. No proper subpart of ruining a picture (16a) or finding a map (16c) entails that the picture was ruined or the map was found, meaning the events are telic. In contrast, any proper subpart of pushing a cart (16b) or waiting for a parcel (16d) entails that the cart was pushed and the parcel waited for, meaning the events are atelic. Our informal impression suggested a strong bias for overt accusative case marking on affected objects irrespective of telicity. To our knowledge, this correlation had not been noted before, and was confirmed by the results of our forced choice studies.

Furthermore, adverbials like completely that are related to the object seem to enhance DOM, as compared to more subject-oriented adverbials like carefully that are related to the manner in which the event takes place, in this case to the intentions of the agent.

In syntactic approaches to DOM on referentiality, marked direct objects have been argued to move out of the VP to escape existential closure, while unmarked direct objects stay VP internal (Diesing 1992; Kornfilt 2008). It is interesting to note that a VP external position has also been claimed for affected arguments (Hale and Keyser 1992; Torrego 1998; Travis 2010). In this article, we do not discuss the syntax of affected objects any further but refer the interested reader to Kizilkaya (2021).

4.1 Set-up and predictions

To investigate how verbal semantic properties like affectedness and telicity shape the distribution of DOM for indefinite direct objects in Turkish and Uzbek, we conducted a forced choice study in each language with cognate examples. As we were interested in the question of whether affectedness and telicity interact amongst each other and/or with the direct object’s animacy, we tested DOM [±acc] depending on the nominal predictor (i) animacy of the direct object referent and the verbal predictors (ii) affectedness of the direct object, and (iii) telicity of the predicate. We hypothesized that animacy and affectedness would enhance morphological accusative case on the direct object. For telicity, we did not hypothesize such a correlation independent of affectedness. As the transition that the direct internal argument undergoes marks the temporal end of the event when change is involved on its behalf, we expected telicity to raise overt acc case marking in combination with affected predicates but not with predicates underspecified for change in the direct object.

4.2 Method

4.2.2 Materials

Critical items were organized in eight conditions by a 2 × 2 × 2 factorial design, manipulating animacy (human vs. inanimate), affectedness (±affected), and telicity (±telic).^[7] We had eight verbs for each of the conditions manipulating the variables affectedness and telicity, see α–δ in Table 1. The variable animacy was manipulated within individual items, see indexes 1 and 2 in Table 1. Each of the 64 critical items consisted of a critical sentence with an indefinite noun phrase preceded by the indefinite article bir in Turkish and bitta in Uzbek. As the same verb was used across the animacy conditions, items were distributed on two lists balanced for animacy, such that each participant would see each verb only once. Entailments of affectedness and telicity were pretested in a prior experiment (see footnote 7). Conditions and selected example items are illustrated below:

Table 1:

Overview of experimental conditions with example items.

Condition	Example (Turkish/Uzbek)
α1:	Ayşe	bir bina(-yı)	yık-tı.
[inanimate]	Oysha	bitta bino(-ni)	yiq-di.
[+affected]	pn	a building-acc	topple-pst
[+telic]	‘Ayşe/Oysha toppled a building.’
α2:	Ayşe	bir lider(-i)	yık-tı.
[human]	Oysha	bitta lider(-ni)	yiq-di.
[+affected]	pn	a leader-acc	topple-pst
[+telic]	‘Ayşe/Oysha toppled a leader.’
β1:	Yusuf	bir arac(-ı)	it-ti.
[inanimate]	Yusuf	bitta aravacha(-ni)	itar-di.
[+affected]	pn	a cart-acc	push-pst
[−telic]	‘Yusuf pushed a cart.’
β2:	Yusuf	bir adam(-ı)	it-ti.
[human]	Yusuf	bitta odam(-ni)	itar-di.
[+affected]	pn	a man-acc	push-pst
[−telic]	‘Yusuf pushed a man.’
γ1:	Selim	bir harita(-yı)	bul-du.
[inanimate]	Salim	bitta xarita(-ni)	top-di.
[−affected]	pn	a map-acc	find-pst
[+telic]	‘Selim/Salim found a map.’
γ2:	Selim	bir sekreter(-i)	bul-du.
[human]	Salim	bitta kotib(-ni)	top-di.
[−affected]	pn	a secretary-acc	find-pst
[+telic]	‘Selim/Salim found a secretary.’
δ1:	Fatih	bir paket(-i)	bekle-di.
[inanimate]	Fotih	bitta paket(-ni)	kut-di.
[−affected]	pn	a parcel-acc	wait-pst
[−telic]	‘Fatih/Fotih waited for a parcel.’
δ2:	Fatih	bir şoför(-ü)	bekle-di.
[human]	Fotih	bitta haydovchi(-ni)	kut-di.
[−affected]	pn	a driver-acc	wait-pst
[−telic]	‘Fatih/Fotih waited for a driver.’

In addition to the set of 64 critical items, a set of 32 filler items was designed, including 16 control fillers, 8 incremental theme fillers and 8 neutral fillers. The control and incremental filler items required a choice between marked [+acc] and unmarked [−acc] direct objects. With the control fillers, only one of the options was grammatical or strongly biased towards [+acc] (circumambulate the Kaaba) or towards [−acc] (pick flowers). The item in (18a) is strongly biased to the unmarked version of the bare noun resulting in a number neutral, activity reading of flower picking. A case marked version should be dispreferred, as it would imply a definite reading with the same flower being picked repeatedly, which is pragmatically strange. Conversely, with the object denoting a unique entity, (18b) is only grammatical with overt accusative case marking.

(18)

a.

Turkish/Uzbek, [−acc] control filler
Tufan	sabahları	çiçeğ( ?? - i )	topla-dı.
Tohir	ertalablari	gul( ?? - ni )	ter-di.
pn	mornings	flower-acc	pick-pst
Intended: ‘In the mornings, Tufan/Tohir picked flowers (did flower-picking).’

b.

Turkish/Uzbek, [+acc] control filler
Meltem	Mekke’de	kabe*(- yi )	tavaf et-ti.
Muhayyo	Makkada	Kaʼba*(- ni )	tavof et-di.
pn	Mekka-loc	kaaba-acc	circumambulate-pst
Intended: ‘In Mekka, Meltem/Muhayyo circumambulated the Kaaba.’

The incremental theme fillers included object-oriented adverbials (eat an apple completely) or neutral/subject-oriented adverbials (eat an apple carefully), cf. (17) above, and required a choice between marked [+acc] and unmarked [−acc] direct objects. The neutral fillers required a choice between accusative [+acc] and dative [+dat] marked objects, both options being grammatical but with slightly different interpretations, as illustrated in (19).

(19)

Turkish/Uzbek, neutral filler
Şeyda	bir yol- u /	bir yol- a	yürü-dü.
Shodiya	bitta yoʻl- ni /	bitta yoʻl- ga	oʻt-di.
pn	a path-acc	a path-dat	walk-pst
acc: ‘Şeyda/Shodiya walked a path.’
dat: ‘Şeyda/Shodiya walked towards a path.’

The order of marked and unmarked [+acc vs. −acc] direct objects, or accusative and dative marked forms [+acc vs. +dat] of the object shown on top or bottom (see Figure 2 below), was balanced throughout the experiment. Critical items and filler items were distributed between two lists in a Latin Square Design. Each list contained 60 items in total, consisting of half of the critical items (32 critical items) and 28 filler items (16 control + 4, i.e. half of the 8 incremental theme + 8 neutral fillers). Items were presented in randomized order.

Figure 2a:

Turkish critical item, cf. α₂.

Figure 2b:

Uzbek critical item, cf. α₂.

4.3 Results and discussion

Looking at the control filler items shows that participants understood the given task and were attentive to sentences with obligatory accusative case marking and sentences with ungrammatical markings. The proportion of DOM with the control fillers for which only [+acc] on the direct object was grammatical (i.e., definites, proper names) lies much higher than with the control fillers that were strongly biased to [−acc] on the direct object (i.e., bare nouns in stereotypical events), as is illustrated in Table 2.

For the critical items, we hypothesized that a ranking high on the animacy scale would enhance morphological accusative case on the direct object (H1). Inspection of the data shows that human direct objects were more often chosen with DOM than inanimate ones in both languages (Figure 3). The generalized linear mixed effects model reveals a significant main effect of the predictor animacy on the outcome in Turkish (β = 1.20, SE = 0.46, p < 0.01)^[9] and in Uzbek (β = 1.76, SE = 0.44, p < 0.001). Thus, in favor of H1, our results make clear that there is a correlation between the animacy of the direct object referent and the likelihood for it to be [+acc] marked in both languages. This observation replicates prior results in the literature for Turkish (Krause and von Heusinger 2019) and provides the first empirical evidence of this kind for Uzbek.

Figure 3:

Proportion of DOM [+acc] on the direct object along animacy.

Furthermore, we hypothesized that affectedness would enhance morphological accusative case on the direct object (H2). Inspection of the data reveals that the affectedness of the direct object evokes clear contrasts in the proportion of DOM in Turkish. Participants more often chose [+acc] on affected direct objects than on non-affected ones. The statistical model shows a significant main effect of the predictor affectedness on the distribution (β = 1.92, SE = 0.46, p < 0.001). In Turkish, this impact is stable across the other predictors and even stronger than the observations along animacy. In comparison, affectedness evokes only a slight overall contrast in the proportion of DOM in Uzbek (Figure 4). This may be attributed to a ceiling effect with human direct object referents, as the contrast gets stronger within the subset of inanimate direct object referents (Figure 5). The statistical model does not confirm the main effect of the predictor affectedness in Uzbek, but it reveals an effect of affectedness within inanimate direct objects which is at the margin of statistical significance (β = −1.04, SE = 0.63, p = 0.10). Our observations support H2, indicating that the distribution of DOM is sensitive to affectedness and that the influence of this verbal parameter may interact with other parameters. In both languages, these results are the first of their kind with respect to verbal parameters and their interrelation with DOM.

Figure 4:

Proportion of DOM [+acc] on the direct object along affectedness.

Figure 5:

Proportion of DOM [+acc] on the direct object along affectedness and animacy in Uzbek.

For the parameter telicity, we hypothesized that it would enhance the likelihood of DOM only in combination with affectedness, but not in isolation of it (H3). As predicted, the data reveal no general increase in the proportion of DOM along telicity (Figure 6). However, telic predicates constituted the most likely case for choosing [+acc] on the direct object if combined with affectedness, while in the absence of affectedness telicity even reduced the proportion of accusative marking (Figure 7). This pattern is stable across the various animacy levels in both languages. The generalized linear mixed effects model reveals no significant main effect of telicity in either language, but a statistically significant interaction of telicity and affectedness in Uzbek (β = −3.17, SE = 0.86, p < 0.001) and an interaction of telicity and affectedness in Turkish which is at the margin of statistical significance (β = −1.67, SE = 0.91, p = 0.07). In support of H3, the observations discount the possibility of an independent influence of telicity on the distribution of DOM. They suggest that the affectedness of the direct object is crucial in mediating the role of telicity.

Figure 6:

Proportion of DOM [+acc] on the direct object along telicity.

Figure 7:

Proportion of DOM [+acc] on the direct object along telicity and affectedness.

A closer inspection of individual items reveals that those in the affected conditions behave homogeneously with respect to the proportions of DOM observed, while in the non-affected condition, individual items pattern rather diverse. On non-affected direct objects in punctually telic events like find x, discover x, or choose x, i.e., achievements in the terminology of Vendler (1957), DOM is relatively dispreferred. In Uzbek, this is only visible with inanimate direct objects, as in bitta uzuk tanlamoq ‘choose a ring’ (45% DOM), bitta xarita topmoq ‘find a map’ (45% DOM) or bitta qalam yutmoq ‘win a pen’ (28% DOM). With these verbs, there is no necessarily implied change of location or state, etc., for the direct object. Note that some authors may argue for a more abstract change on behalf of direct objects of find x type verbs. However, the affectedness diagnostics introduced in (12) show clear contrasts between find x type verbs, as compared to break x type verbs. Neither the entailment of change test nor resultative predication are felicitous with find x type verbs while they are with break x type verbs. This is empirically confirmed by a pretest we undertook that investigated the acceptability of the entailment of change test (see footnote 7). Comparing the proportions of DOM on affected and non-affected direct objects in telic events suggests that the involvement of the direct object in determining telicity is crucial for DOM. When the telicity of the predicate arises from the change to a specific result state that the direct object undergoes, as in break x, ruin x, tear apart x, it enhances DOM. Conversely, in predicates that are underspecified for change in the direct object, telicity does not enhance DOM. Thus, we conclude that telicity is not an independent factor driving overt accusative marking in the languages investigated, but its influence on DOM rather stems from interaction with affectedness. Surprisingly, across the non-affected conditions, the proportion of DOM is higher in atelic predicates than in telic ones. In Turkish, this is due to strong biases for DOM with individual verbs in the atelic condition, among them the psych-verb sevmek ‘love’ (92% DOM), the verb of contact okşamak ‘caress’ (91% DOM) and the negative judgment verb yalanlamak, ‘refute/deny’ (95% DOM). Their DOM patterns contrast with other verbs in this condition, like duymak ‘hear’ (63% DOM), beklemek ‘wait for’ (51% DOM), or görmek ‘see’ (16% DOM). The lexical semantics of all of these verbs does not entail a change in some property for the direct object, which is why they were classified as non-affected. Interestingly, in the pretest on affectedness that we undertook for the Turkish items, the verbs sevmek ‘love’, okşamak ‘caress’, and yalanlamak, ‘refute/deny’ constituted the outliers in the non-affected condition. They were evaluated by around half of the participants (n = 31) to invoke change on the direct object (see footnote 7). This could explain why they show a similar pattern to affected direct objects. In Uzbek, there is a stronger overall tendency for DOM and the lower proportion for DOM in the non-affected condition can be attributed to the punctually telic predicates with inanimate direct objects, as mentioned above. For a detailed discussion of individual items and further possible factors influencing the distribution see Kizilkaya (2021).