2.1 Adjectival resultatives

Resultatives such as (2a) are single clauses that consist of a primary predicate, i.e., the main verb (e.g., wipe), and a secondary predicate, i.e., the adjective (e.g., clean). The adjective denotes a state that holds of a participant in the event as a result of the action denoted by the verb (e.g., Rappaport Hovav & Levin 2001; Wechsler 2005; Beavers 2012b; Williams 2015). The participant in the event is referred to by the object dp (e.g., the table). Consequently, adjectival resultatives can be regarded as change of state expressions: they describe the change of a property applied to an affected argument from one state to a different state.

Resultatives differ from so-called depictive adjectival complex predicates, illustrated in (3). While the adjective in resultatives describes the result of the event, the adjective in depictives describes a state that holds of the affected argument during the event. The affected argument may be the subject dp, as shown in (3a), or the object dp, as shown in (3b), or it may be ambiguous between the subject and object, as shown in (3c) (see also Burkhardt et al. 2019).

Resultatives and depictives also differ in that the set of adjectives participating in resultative constructions is small compared to the set of adjectives that can be used depictively (e.g., Boas 2000). Moreover, cross-linguistically the occurrence of adjectival resultatives is much more restricted: they are found in Germanic languages such as English and German, as well as in Hungarian, Latvian, Mandarin, and Persian, but not in Romance or Slavic languages (for a recent discussion see Haider 2018). Haider (2016; 2018) argues for a tight relation between the presence of adjectival resultatives and resultative particle verbs or serial verbs. More precisely, he predicts that “if a language makes available resultative complex verbal predicates (i.e., particle verbs or serial verbs), it thereby also makes available resultative adjective constructions, that is, complex verbal predicates with adjectives that receive a resultative interpretation” (Haider 2016: 2). Accordingly, languages like Mandarin that have serial verbs are predicted to allow resultative adjectives, just like Germanic languages that have particle verbs (see Maché 2022, for serial verb constructions in Benue-Kwa languages). Important for our analysis of adjectival resultatives (see Section 4), Haider (2016; 2018) proposes that the grammatical structure of resultative particle verbs and of resultative adjective constructions is the same. He argues for a complex predicate analysis (see also Neeleman & Van de Koot 2002, a.o.), in which an adjective is adjoined to the verb in a head-to-head adjunction structure, as illustrated in (4):²

In the present paper we focus on resultative constructions with an adjective; note that results can also be expressed by a pp (e.g., unter den Tisch trinken ‘drink under the table’, example from Haider 2018) or by a dp (e.g., paint the wall a shady blue, see Beavers 2012b for an overview). Adjectival resultatives are typically classified into two subtypes, depending on whether the object undergoing the change is selected by the verb. If the object is not selected by the verb, the corresponding structure without the adjective is ungrammatical. This is shown by the contrast between (5a) and (5a’) and between (5b) and (5b’).

The verb bellen ‘bark’ in (5a/a’) is intransitive, more precisely unergative, and does not select a second argument. Verbs such as bellen can be combined with an adjective only if a non-subcategorized argument is present, i.e., a reflexive pronoun that is coreferential with the subject. The verb essen ‘eat’ in (5b/b’) can select a direct object, but the argument den Teller ‘the plate’ refers to the entity being emptied during the event of eating what is on the plate and not to the entity being eaten.

The predicates freeze solid (6a) and wipe clean, repeated here in its German version (6b), illustrate the other resultative subtype, referred to by Wechsler (2005) as control resultatives. The object is selected by the verb, and, unlike the resultatives in (5) (henceforth referred to as non-control resultatives), control resultatives allow omission of the adjective, as shown in (6a’) and (6b’).

In (6a/a’), the verb is intransitive, more precisely unaccusative, and the adjective modifies the subject dp. In (6b/b’), the verb is transitive and the adjective modifies the direct object dp. Our experimental study investigates transitive adjectival control resultatives such as (6b) compared to simple transitives such as (6b’). In Section 2.2, we address the issue of telicity and of how the combination of verb and adjective conspire to achieve telicity, also in comparison to the variant without the adjective.

2.2 Telicity

Event structure constitutes one important aspect for the analysis of change of state expressions, in particular regarding their telicity readings (Demonte & McNally 2012; Rappaport Hovav 2014). Predicates like find are telic, because its meaning includes a specific moment toward which the event it describes naturally develops, referred to as a natural endpoint, while predicates such as wipe, which describe events that can continue indefinitely or can stop at any time, are atelic (Vendler 1957; Dowty 1979). In German, like other Germanic languages, the telicity of a predicate can arise from different sources: it is determined by the lexical semantics of individual verbs, sometimes in concert with other elements in the verb phrase. Inherently telic verbs such as find, open, and arrive lexicalize the natural endpoint as part of their meaning. In these cases, telicity is argued to be construed semantically, i.e., event culmination arises via entailment (Vendler 1957; Dowty 1979). This entailment is not cancellable, as illustrated in (7a). The traditional temporal adverbial modification test (Dowty 1979), illustrated in (1a), repeated here as (7b), yields the same result as the entailment test.³

Besides the lexical semantics of the verb, resultative verb particles, directional pps, and the quantificational semantics of the object undergoing a change may contribute to the construal of telicity (van Hout 1998; Krifka 1998).⁴ Following Schulz (2018), we treat all these cases as instances of compositional telicity.⁵ For example, verbs such as essen ‘eat’ do not lexicalize a natural endpoint; they are atelic. A telic reading of the predicate is achieved if these verbs combine with overt telicity markers. With resultative particles, this ensuing telic reading is obligatory, as exemplified in (8a). The particle verb construction (8a) entails the event culmination, whereas the counterpart without the verb particle does not carry any culmination entailment (8b). Note that, strictly speaking, the term entailment applies at the sentence level, i.e., in our case to sentences with a telic predicate with perfective marking (see footnote 3). In the remainder of the paper, verbs are always assumed to be in the perfective form. Accordingly, we use a shorthand to refer to telicity construal, stating that telicity may arise via entailment, either through the lexical properties of the verb or through overt markers such as resultative particles.

Following Pustejovsky (1991; 1995), the event structure of a word constitutes one level of its semantic specification. In general, mereotopological approaches focus on eventualities and their parts by assuming basic event types and proposing mechanisms for how these basic event types can be combined to complex event types with an internal subevent structure (see also van Hout 1998; Williams 2015). Adopting a subeventual analysis for predicates, Pustejovsky (1991; 1995) postulates three event types: states (S), processes (P), and transitions (T) from one state to its opposition, including a process and a state subevent. The telicity of an event is related to the existence of a complex subevent structure consisting of a process and a state (see also Demonte & McNally 2012). Telic predicates like eat up or open the door designate transitions. For these predicates, the state subevent is marked as the head-of-event, indicating that it is semantically the most prominent subevent. Therefore, they are also referred to as endstate-oriented transitions (see Schulz et al. 2002).

Finally, the quantificational semantics of the direct object dp can affect the telicity of the predicate. Mereological approaches (Krifka 1998) model events and participants taking part in the event in terms of their part structure and a homomorphism between both. Combined with verbs of consumption (e.g., eat, drink), verbs of destruction (e.g., burn, consume) and verbs of creation (e.g., build, compose), a quantized object triggers a telic reading (9a/a’), whereas a non-quantized object (a mass term or bare plural) allows an atelic reading (9b/b’). Verbs like sweep, iron, wash, and comb behave similarly regarding this difference (Filip 2020).

Whereas telicity in resultative particles arises through entailment, i.e., a hard, non-cancellable inference, the contribution of quantized dps to telicity construal seems more flexible. The culmination inference can be cancelled, as illustrated with a specific definite dp and a verb of consumption in (10a) (from Hay et al. 1999: 139) and with a specific definite dp and a surface contact verb in (10b) (from Schulz 2018: 128).⁶

Therefore, it has been argued that in combination with certain verbs, specific definite dps construe telicity pragmatically via a pragmatic inference (Talmy 1991; Brisson 1994; Jeschull 2007; Schulz 2018). Again, this is shorthand for stating that a vp with perfective marking triggers event culmination via a pragmatic inference. In line with this analysis, both an atelic and a telic reading are possible with wipe, as illustrated in (11):

As noted by Rothstein (2001), the minimal pair in (11) indicates a problem of the mereological approaches to telicity: “Unless we assume that the verb assigns two different thematic roles in each example […], we need to ask why in the one case the quantized direct object determines a culmination point and in the other it doesn’t” (Rothstein 2001: 142). This issue of ambiguity points to a general shortcoming of the mereological approach: telicity is regarded as a fixed concept, defined by quantization and the notions of homogeneity and cumulativity, without allowing for variation across different verb classes (van Hout 2018). Accordingly, this approach cannot easily capture the observation that so-called weak-endstate verbs such as wipe, wash, and sweep, combined with quantized direct objects, ’pragmatically favor‘ (Brisson 1994) a telic interpretation, but that there is no natural endpoint. Pustejovsky (1991; 1995) addresses this issue by assuming that predicates such as wipe the table designate a complex event of the type transition, with the most prominent subevent being the process and not the state. In our view, this event structure still leaves open how the telic and the atelic readings are achieved. The question is how we can capture the intuition that predicates of the type wipe the table seem to be ambiguous between atelic processes and telic transitions. We will return to this point in the discussion.

In addition to specific definite dps, a specific verb class has been assumed to construe telicity via a pragmatic inference as well: degree achievement verbs such as widen, lengthen, cool, straighten, dry, and clean (Hay et al. 1999); this is illustrated in (12).

According to Hay et al. (1999: 130), degree achievements “describe events that measure out the change an object undergoes with respect to the gradable property introduced by the base adjective”. The bound of this measure of change can, for instance, be inferred from the scalar structure of the adjectival base. If a scale has a maximal value as is the case with straight, clean and dry (see Kennedy & McNally 2005), this value can serve as the result of the change, i.e., in our terms the endpoint. Although the most likely interpretation of the degree achievement predicate clean the table, for instance, may be telic, i.e., that the table is cleaned completely, the endpoint does not need to be reached, as illustrated in (12a) with the entailment test and by the outcome of the adverbial modification test in (12b).

Let us now turn to the question of how the telicity reading is achieved in adjectival resultatives. For the purposes of the present paper, we follow the assumption that adjectival resultatives share the syntax and semantics with resultative verb particles. Accordingly, telicity in resultative adjective constructions should arise via entailment and should not be cancellable. This seems to be borne out, as illustrated in (13a); an in-adverbial is permitted (13b), but a durational reading is excluded as predicted, see (13c).

For now, we leave open which event-semantic properties the structures wipe the table (11) and wipe the table clean (13) may share, given the presence of a specific definite dp in both constructions and given the apparent ambiguity of the former but not the latter variant. We take up this question in the final discussion (Section 4).

2.3 Empirical studies on telicity

Telicity has been investigated experimentally across a number of languages. These studies aimed at shedding light on whether adults and children confirm the theoretical assumptions regarding the interaction of the verb and other elements in the sentence (e.g., Arunachalam & Kothari 2010; see van Hout 2018, for a cross-linguistic overview; Martin et al. 2020, for a cross-linguistic comparison). Several experimental studies on German probed the interpretation of inherently telic and atelic verbs as well as different telicity markers including resultative verb particles and quantized dps (see Schulz 2018, for an overview). In these controlled experimental settings, studies typically used variants of the Truth-Value Judgement Task. In the following, we describe these studies in more detail and report the adult findings, because the design of our study used the same basic set-up. The test paradigms varied event type (incomplete vs. complete) and tested the acceptance of different event descriptions. The crucial condition in this paradigm is the incomplete event. Incompletion always involved a partial result, i.e., the event progressed up to a certain point and then stopped; the object was partially affected and had undergone some change (e.g., a box the lid of which was moved a little, a piece of cheese that was half-eaten). The rationale is the following: acceptance of a sentence as a description of an incomplete event points to an atelic interpretation, whereas its rejection points to a telic interpretation. That this experimental paradigm can reliably assess participants’ interpretation of different verb types has been shown by Schulz et al. (2001) and Penner et al. (2003), who found that the inherently telic verb aufmachen ‘open’ was accepted in only 5% of the cases for incomplete events. In a follow-up study, Schulz & Wittek (2003) tested a wider range of inherently telic particle verbs (aufmachen ‘open’, zumachen ‘close’, abmachen ‘take off’, anmachen ‘turn on’) as well as atelic verbs in intransitive constructions (fegen ‘sweep’, wischen ‘wipe’, bürsten ‘brush’, pusten ‘blow, malen ‘draw’, bauen ‘build’, puzzeln ‘do a puzzle’, schneiden ‘cut’). As predicted, adults accepted the inherently telic verbs in only 3% of the cases for incomplete events, while the atelic verbs of surface contact, creation, and destruction were always accepted in this condition.

The studies by Schulz & Penner (2002) and Schulz & Ose (2008) are most important for our study (see Section 3). The authors examined the interpretation of verbs of consumption (essen ‘eat’, trinken ‘drink’) in transitive structures with a specific definite dp to probe the theoretical assumption that quantized objects trigger a telic reading. Adopting the design for Dutch and English by van Hout (1998) to German, Schulz & Penner (2002) compared quantized dp structures (e.g., Hat das Mädchen den Apfel gegessen? ‘Did the girl eat the apple?’) to intransitive structures (e.g., Hat das Mädchen gegessen? ‘Did the girl eat?’), and to resultative verb particle structures (e.g., Hat das Mädchen aufgegessen? ‘Did the girl eat up?’). Intransitive verbs were accepted for incomplete events in 97% of the cases, whereas resultative particle verbs were nearly always rejected (4% acceptance). Contrary to mereological approaches to telicity, transitive structures with a quantized dp were accepted in 52% of the incomplete-event conditions. A follow-up study by Schulz & Ose (2008) replicated this pattern: intransitives were always accepted for incomplete events, while resultative particle verbs were always rejected; again, quantized dps were accepted in 73% of the cases. In a further condition, Schulz & Ose (2008) tested verbs of consumption with non-quantized dps (bare nouns, e.g., Käse ‘cheese’); in line with mereological approaches, these constructions were always accepted for incomplete events.

Taken together, these experimental findings on German indicate that adding a resultative verb particle to atelic verbs of consumption results in a telic interpretation. Presence of a quantized dp triggers a telic interpretation only in some cases, in contrast to non-quantized dps, which never result in a telic interpretation. The results across different predicate types (i.e., inherently telic verbs, atelic verbs, verbs of consumption with and without particle, with (non-)quantized dp) provide empirical evidence that construal of telicity may arise via entailment or via pragmatic inference. Notably, cross-linguistic findings support the observation that the interpretation of quantized dps fluctuates between telic and atelic for adults (Arunachalam & Kothari 2010; van Hout 2018).

Experimental research on the interpretation of adjectival resultatives has mainly assessed whether speakers of German (adults and children) are sensitive to the selectional restrictions of adjectival resultatives (see Richter & van Hout 2013).⁷ Richter & van Hout (2013) created a metalinguistic task, in which participants heard sentences with resultative adjectives and had to indicate via word cards which noun the adjective belongs to. All children and all adults correctly opted for the object noun. Infelicitous structures (e.g., The children watched the grandma red) were rejected by the adults, whereas children took the adjective to belong to the object noun. This finding indicates that adults’ interpretation shows sensitivity to the selectional properties of verbs. Due to the nature of the task, it leaves open whether participants assigned a telic reading to the test sentence.

To our knowledge, no study has tested the contrast between adjectival control resultatives such as He wiped the table clean and the corresponding simple transitive He wiped the table regarding their respective telicity interpretation. The only empirical evidence available concerns the related intransitive structures such as Er fegte ‘He swept’, suggesting an atelic interpretation (Schulz & Wittek 2003). Accordingly, at least two empirical issues are open: first, do German-speaking adults interpret adjectival resultatives indeed as telic, as expected if the parallel with resultative particles is on the right track? Second, are the simple transitive variants interpreted as fluctuating between telic and atelic? Our study assessed the telicity readings of both constructions.

3.1 Participants

Twenty-one monolingual native speakers of German (age range = 20–35 years, mean age = 25 years, SD = 4 years) participated in the experiment. They had no background in linguistics and were naive regarding the topic of the study. All participants gave informed written consent for their participation in the study.

3.2 Method

3.2.1 Materials

Our 2 × 2 design comprised the factors Structure (resultative/transitive) and Event type (complete/incomplete), resulting in four test conditions with four items each. In line with previous telicity studies (see Section 2.3), we employed the yes/no-question variant of the Truth-Value Judgment Task. The four weak-endstate verbs were wischen ‘wipe’, bügeln ‘iron’, föhnen ‘blow-dry’, and trinken ‘drink’ (i.e., verbs of surface contact and a verb of consumption). These verbs were chosen because they are easily depictable and because they license the addition of a respective gradable adjective, whose scale has a maximal value. The latter restriction was necessary to construct the adjectival resultative structures. This set-up resulted in the following complex predicates: sauber wischen ‘wipe clean’, glatt bügeln ‘iron smooth’, leer trinken ‘drink empty’,⁸ and trocken föhnen ‘blow-dry dry’. The main verb was always marked for present perfect.⁹ Five different object nouns, all realized as singular specific definite dps, were used per verb (e.g., den Tisch/den Boden/das Fenster/die Tafel/die Badewanne wischen ‘wipe the table/the floor/the window/the blackboard/the bathtub’). For each adjectival resultative test sentence, a second structure was created without this adjective; see the example item set in (14).

1. (14)
1. a.

1. Hat
2. Has
1. er
2. he
1. den
2. the.acc
1. Tisch
2. table
1. sauber
2. clean
1. ge-wisch-t?
2. ptcp-wipe-ptcp
1. ‘Did he wipe the table clean?’

1.  
1. b.

1. Hat
2. Has
1. er
2. he
1. den
2. the.acc
1. Tisch
2. table
1. ge-wisch-t?
2. ptcp-wipe-ptcp
1. ‘Did he wipe the table?’

Half of the actions were performed by a male character and half by a female character; personal pronouns (sie ‘she’ or er ‘he’) were used to refer to the subject. In total, there were five experimental lists, which were latin-squared and pseudo-randomized.

The test scenarios were presented in form of 15-second animated cartoons combined with a prototypical sound, e.g., the sound of wiping a cloth across a surface, to underscore the time course of the event. In complete events (Figure 1a), the affected object was shown as completely clean, smooth, empty, or dry, respectively. In incomplete events (Figure 1b), the affected object had undergone a partial change; for the item wipe, for instance, some spots of dirt were still visible when the event stopped. A full list of experimental items including the test sentences and stills of the animations is provided in Supplementary Material 1.

Figure 1

Example scenario as stills for complete events (a.) and for incomplete events (b.).

Six filler trials, which did not contain the target structures (e.g., Hat der Mann ein Bügeleisen? ‘Does the man have an iron?’), were added to the experimental trials. Recall that we expected a semantically telic interpretation for resultatives and a pragmatically telic interpretation for the simple transitives. These interpretations would be reflected in the behavioral data as follows: if resultatives (see 14a) are interpreted as entailing event culmination, they should be rejected (no-answer) when the depicted event is incomplete (see Figure 1b). If the corresponding simple transitives (see 14b) are interpreted as ‘pragmatically favoring’ event culmination, they should be rejected (no-answer) or accepted (yes-answer) for incomplete events for some participants and/or for some items (see Schulz & Penner 2002). Complete events (see Figure 1a) should yield yes-responses for both structures. To counterbalance the uneven distribution of expected yes- and no- responses across the experiment, the filler questions were constructed to elicit four no-answers and two yes-answers.

An adjective pretest and two practice trials were administered before the main test. The practice items comprised two items from a different experimental list and served to familiarize the participants with the set-up of the test items using animation and sound. In the pretest, participants saw eight unanimated drawings and answered questions about the object properties described by the adjectives used in the main test (see Figure 2).

Figure 2

Example pretest item: Ist das Geschirr sauber? ‘Are the dishes clean?’.

The pretest assessed whether participants agreed with the depiction of the properties described by the adjectives. The objects used in the pretest were not used as object nouns in the main test and the participants did not receive any response-contingent feedback.

3.3.2 Experimental items

Participants’ answers were classified as yes- or no-responses. For the following analyses, we used the number of yes-answers as the dependent variable. For each of the four conditions, a total of 84 responses was analyzed. The proportion of yes-responses per condition is shown in Figure 3; the width of the violin plots varies by the density of data points in a specific region. For complete events, the violins look the same because the number of yes-answers is identical for both constructions (Mean = 98.8%, SD = 10.9%). For incomplete events, the number of yes-answers is higher for transitive structures (Mean = 71.4%, SD = 45.4%) than for adjectival resultatives (Mean = 9.5%, SD = 29.5%). The wide part of the violin for the resultatives indicates that yes-answers were rarely given. In contrast, the violin for the transitive structures points to a more balanced distribution of yes-answers, but with more yes- than no-answers.

The results were analyzed with generalized mixed effects logistic regression (1 = yes, 0 = no). We used the glmer function in the lme4 package (Bates et al. 2015) in R (R Core Team 2022). The response data were fitted to a model using maximum-likelihood estimation (Laplace Approximation). Event type (complete, incomplete) and Structure (resultative, transitive) were entered as fixed effects, using contrast coding and ‘incomplete’ and ‘resultative’ as the reference levels, with a random intercept for Participant and Item. The model revealed a significant effect of structure (β = 1.0597, Z = –2.112, p = .03), a significant effect of event type (β = 1.2149, Z = –5.188, p < .001), and a significant interaction of structure and event type (β = 2.1193, Z = –2.112, p = .03). The interaction of structure and event type was further inspected via pairwise comparisons with Tukey adjustment (emmeans package, Lenth 2022), revealing a significant difference between the structures for incomplete events (p = .001), but not for complete events (p = 1.000).

Figure 3

Violin plots depicting yes-responses (in %) by sentence structure and by event type (left and right panel). The black diamond indicates the mean.

To inspect whether there was variation across the verbs in the incomplete event condition, we analyzed the number of yes-responses for each verb separately. Given our design, the proportion of yes-responses equals the number of participants giving this answer. As can be seen in Table 1, the 9.5% cases of acceptance for adjectival resultatives were basically due to one verb, trocken föhnen ‘blow-dry dry’, which was accepted in a third of the cases. Variation across verbs was much higher for simple transitives, with acceptance rates ranging between 52% and 95%.

In a next step, we examined the source of variation for the transitives in the incomplete condition, analyzing the responses to each verb within and across participants. Nine of the participants interpreted all transitives as atelic and a further five participants interpreted three transitives as atelic. Three participants interpreted two transitives as atelic and three interpreted one transitive as atelic. Notably, only one participant interpreted all transitives as telic. Interpretations as telic or atelic were not tied to specific verbs (see Supplementary Material 2 for a detailed list).

In summary, the participants predominantly rejected the adjectival resultatives with incomplete events (90.5% no-answers). Transitive sentences with the same verb but without the adjective were often accepted for incomplete events (71.4% yes-answers), with variation across and within participants.

Table 1

Number of yes-responses for incomplete events per verb and structure.

Verb	Structure
	Resultative		Transitive
	Raw number	%	Raw number	%
Trinken ‘drink’	0	0	11	52
Bügeln ‘iron’	0	0	14	67
Wischen ‘wipe’	1	5	15	71
Föhnen ‘blow-dry’	7	33	20	95

4.1 Construal of telicity in adjectival resultatives

Recall that incomplete events constituted the crucial condition in our experiment. In this condition, the participants predominantly rejected the adjectival resultatives (90.5% no-answers) indicating a telic interpretation. A by-item and by-participant analysis revealed that the rare cases of acceptance were mostly due to a single predicate: trocken föhnen ‘blow-dry dry’. It could be that this response results from an imprecise interpretation of dry: that is, in everyday life, it may be sufficient to blow-dry something until it is nearly dry (see Syrett et al. 2010, for imprecise interpretations of full). As pointed out by an anonymous reviewer, however, the same reasoning could also be applied to the other predicates. Alternatively, it may be that the animated cartoons, although carefully designed, did not depict the remaining wetness of the objects clearly enough. For now, we leave this issue open and abstract away from this exception.

In the following, we model the telic interpretation of adjectival resultatives within a mereotopological approach (Pustejovsky 1995; Williams 2015; McNally 2022), assuming that telicity results from the existence of a complex subevent structure, which consists of a process and a result state. Consider the resultative den Tisch sauber wischen ‘wipe the table clean’. The two subevents are realized overtly: the verb describes the process and the adjective specifies the event description by making the dimension and, in particular, the result of the change explicit (e.g., Baglini & Kennedy 2019).¹⁰ This is the same analysis as has been suggested for particle verbs such as aufessen ‘eat up’ (Schulz et al. 2001). According to Pustejovsky’s analysis, the verb is underlyingly specified with an event type of process, which does not entail culmination. The adjective clean, a stage-level predicate, is underlyingly specified with an event type of state; more precisely, it is a function from processes to transitions (Pustejovsky 1991; 1995). The adjective induces an event-type shift from a process to a transition, whose most prominent subevent (head-of-event) is the state, illustrated in (15a) with the asterisk marking the head; this endstate-oriented transition entails culmination (see also van Hout 1998). We will address the role of the quantized dp in Section 4.2.

According to this analysis, wipe and wipe clean differ in that the predicates designate different event types: wipe designates a process (15b), i.e., a sequence of events identifying the same semantic expression, whereas wipe clean designates a transition (15a), i.e., an event identifying a semantic expression, which is evaluated relative to its opposition (Pustejovsky 1991). Accordingly, the “systematic ambiguity is the result of principles of semantic composition, rather than an underlying lexical ambiguity in the verb involved” (Pustejovsky 1991: 64).

A further argument for the complex subevent structure of wipe clean, comprising a process and a state, comes from adverbial modification. According to Williams (2015), adverbs such as rapidly (e.g., He very rapidly wiped the table clean) can modify the process event as well as the transition from a process to a state: the result may be reached rapidly, even when the wiping process is slow, and the result may be reached slowly, even when the wiping process is rapid. This shows that the event designated by the resultative construction cannot be reduced to a single process event.

The different subevent structures for wipe and wipe clean are analogous to the contrast between simple verbs and their particle verb variant (e.g., wipe – wipe off); the culmination inference arises via entailment and cannot be cancelled.¹¹ Moreover, unlike wipe, predicates such as wipe off and wipe clean designate transitions from one state to its opposition. Accordingly, the utterance He wiped the table clean/off is infelicitous if the affected object was clean already; wipe clean requires a change of state, while wipe does not carry such requirement (Levin & Rappaport Hovav 1991). Put differently, we cannot wipe a clean table clean.¹²

4.2 Construal of telicity in simple transitives

In contrast to the findings for adjectival resultatives, the findings for the simple transitives suggest an ambiguity: while simple transitives were frequently accepted (71.4% yes-answers) as descriptions of incomplete events indicating an atelic interpretation, in some cases simple transitives were rejected (28.6% no-answers), indicating a telic interpretation. This group data is mirrored by the inter- and intraindividual variation across participants, which we address below. In our view, these results are compatible with at least two accounts. According to the first, simple transitive structures are ambiguous between atelic processes and telic endstate-oriented transitions. The latter reading comes about by an event-type shift from a process event designated by wipe to a telic transition. The assumption that surface contact verbs like wipe and sweep in intransitive frames designate an atelic process event has been supported by empirical studies showing that they are always accepted for incomplete events (Schulz & Wittek 2003, see Section 2.3). This would mean that the ambiguity of predicates of the type wipe the table is located at the level of the semantic composition (see Schulz & Penner 2002, for a similar proposal for eat the apple). Notably, the event structure signaling telicity, derived via event-type shift, was chosen less frequently in the current study. The second account assumes that simple transitive structures designate a process event; event culmination is derived via a pragmatic inference, which is realized at a level different from the event-semantic representation. Accordingly, the apparent ambiguity found in our study exists because the telicity inference is computed only in some cases. A point in favor of the latter account concerns the predictions regarding the initial state of the respective events. Recall that adjectival resultatives like wipe the table clean are infelicitous if the affected object exhibits the endpoint at the outset of the event, i.e., if a change of state (regarding the relevant dimension, here: cleanliness) is not possible. If in their telic reading, adjectival resultatives and the corresponding simple transitive had the same event structure, i.e., an endstate-oriented transition, wipe the table should be as infelicitous as wipe the table clean if the affected object was clean at the outset of the event. This is contrary to our intuition and also rejected by previous analyses of verbs of surface contact (Levin & Rappaport Hovav 1991). In short, it seems that while we can wipe a clean table, we cannot wipe a clean table clean. Future empirical research is needed to investigate this assumption more systematically.

What does our analysis imply for the role of the quantized dp in simple transitives containing weak-endstate verbs? As mentioned in Section 2.2, telicity of simple transitives has been related to the structural properties of the direct object dp (Krifka 1998). While this mereological approach regards telicity as a fixed notion defined via the quantization and the notion of homogeneity, empirical research has shown that specific definite dps selected by verbs of surface contact or consumption may or may not trigger a telic reading (Section 2.3). Our findings provide further evidence for the fluctuation of the telicity reading across verbs and across participants, which we argue indicates that the telic reading is derived via pragmatic inference. In our study, transitive trinken ‘drink’ (a verb of consumption) was treated as atelic in 52% of the cases, confirming earlier findings by Schulz & Penner (2002) and Schulz & Ose (2008) for the verb of consumption essen ‘eat’ (52 to 73% atelic readings). Within the class of surface contact verbs, we found variation: two verbs (wischen ‘wipe’, bügeln ‘iron’), in combination with a specific definite dp, were interpreted as atelic in roughly two thirds of the cases, while the verb föhnen ‘blow-dry’ was interpreted as atelic in almost all instances. Almost half of the adults treated all four verbs in the transitive structure as atelic, while only one adult interpreted all four verbs as telic.

In summary, adjectival resultatives and the corresponding simple transitives have a different event structure. The adjective, on a par with verb particles, builds a complex predicate with the main weak-endstate verb. The resulting event structure is represented as a transition, and construal of telicity arises via entailment. In contrast, the corresponding simple transitives are represented as a process event; construal of telicity arises via pragmatic inference and is subject to variation across participants and verbs.

4.3 Next steps

In the present paper we have taken a subeventual rather than a mereological approach to telicity. The subeventual account acknowledges the presence of lexical restrictions (Pustejovsky 1991; 1995; Williams 2015), but it does not incorporate the properties of the adjectives licensed in resultatives (i.e., the adjective must be gradable and must have a closed scale with a maximal value, see Section 2.1). Mereological approaches, in contrast, can account for the role of the scale structure in the construal of telicity (e.g., Wechsler 2005; Beavers 2012a), as they model telicity as connected to a change along a path or a scale that is homomorphic to the event. As noted by Wechsler (2005), the two approaches are not mutually exclusive (see also Demonte & McNally 2012). We leave for future research how the two approaches could be integrated.

Second, it is unclear how telicity construal and change of state are related in case of degree achievements such as clean the table. Do they pattern with wipe the table clean or with wipe the table? There seem to be differences between the three structures regarding the initial state and regarding acceptance for incomplete events. Suppose the table in question is clean: in this case, we can wipe the table, but we cannot clean the table, just as we cannot wipe the table clean. Regarding the outcome, let us imagine a table with some dirt: after we have wiped the table, the degree of dirt may remain unchanged, but after we have cleaned the table, the degree of dirt must have decreased, although the table may still have some spots of dirt. That would mean that the verb clean shares with transitive wipe telicity construal via pragmatic inference but differs from transitive wipe regarding the state at the outset of the event (i.e., the change of state property). Empirical research is needed to see whether our intuitions are borne out.

Third, our study focused on control resultatives, where the verb selects the object dp. Control resultatives have a grammatical transitive counterpart without the adjective, which allowed us to examine the contribution of the adjective. But it is open whether adults assign an obligatorily telic reading also to non-control resultatives, which do not license a simple transitive structure (e.g., den Teller leer essen ‘eat the plate empty’ vs. *den Teller essen ‘eat the plate’). Fourth, we turn to the parallel we argue holds for verb particles and resultative adjectives. Cross-linguistic research is needed to examine whether all languages that have verb particles (or serial verbs) and resultative adjectives pattern like German and whether they can be analyzed along the lines proposed here. Fifth, we left open where exactly in the grammar the event composition of wischen ‘wipe’ as a process and sauber ‘clean’ as an endstate is reflected, in the lexicon, the syntax or the morphology (see Ramchand 2008, for a syntactic proposal). Finally, we argued that telicity in the corresponding simple transitives is construed pragmatically. Future research is necessary to investigate whether this holds across different experimental settings.