1. Introduction: Realisation events, fulfilment types, and manner/result complementarity

According to Talmy’s (Reference Talmy2000) event integration theory, the human mind often packs events into ‘macro-events’ by condensing multiple simplex events into the same clause. In the Motion domain, for example, ‘I ran into the room’ is a combination of the manner of motion (running) and the path traversed by the agent (into). In this macro-event, entering is the ‘framing event’ outlining the conceptual frame for how this event unfolds, and running is the ‘co-event’ fleshing out the frame with extra information.

While myriad studies have been devoted to Motion, little academic attention has been paid to other types of macro-events (Li, Reference Li2013). For instance, I hunted the fugitive down is a combination between an intention (to capture) and its realisation (successful capture). This type of macro-event is a Realisation event (Talmy, Reference Talmy2000, pp. 261–278), with intention being the co-event and its realisation being the framing event (Kou & Hohenstein, Reference Kou and Hohenstein2022).

Within the Realisation event framework, verbs differ in their ‘fulfilment type’ (Talmy, Reference Talmy2000, pp. 262–268), a property indicating how likely an intended goal denoted by the verb is realised (Kou & Hohenstein, Reference Kou and Hohenstein2022). This property divides verbs into four types: 1) intrinsic-fulfilment verbs do not indicate specific intended results without context (e.g., kick); 2) moot-fulfilment verbs independently encode intended results but leave moot whether they are realised (e.g., hunt); 3) implied-fulfilment verbs encode intended results and imply their realisation (e.g., wash); 4) attained-fulfilment verbs encode intended results and entail their realisation (e.g., kill). Figure 1 is a schematic illustration of fulfilment types arranged in terms of their certainty of intention realisation. A question mark next to ‘intrinsic-fulfilment’ shows that these verbs encode no specific intentions.

Figure 1. Schematic cline of certainty of intention realisation in verbs with different fulfilment types

The concept of fulfilment types boils down to three keywords: manner, result,Footnote ¹ and intention.Footnote ² The first two elements constitute a widely-attested semantic rule for verbs: manner/result complementarity (Rappaport Hovav & Levin, Reference Rappaport Hovav, Levin, Hovav, Doron and Sichel2010), namely, a verb (root) can only lexicalise either manner or result, not both. While manner and result are fundamental to verb meaning, intention could link manner and result and thus blur the manner/result dichotomy in the mind. It is argued that semantic manner/result verb statuses are mainly ‘extensional or referential definitions that characterize a verb in terms of the set of events it can label […] and make no specific claims about […] the mental representations Footnote ³ of these verbs’ meanings’ (Behrend, Reference Behrend1990, p. 685, note 1, our emphasis). Thus, in this study, we take a cognitive approach to verb meaning from the perspective of fulfilment types and investigate whether verbs mentally fall into dichotomous manner/result verb mental representations.

In Section 2, we briefly review relevant literature on manner, result, and intention, discuss how they interact in verb meaning, and specify the research question. Next, we empirically examine how verbs of different fulfilment types cluster and vary in different task situations (Section 3), before discussing the blur between manner/result verb representations in the mind and its implications on Talmyan event typology in the Realisation domain (Section 4).

2. Manner, result, intention, and their interplay in verb meaning

Although manner and result are argued to be mutually exclusive in verb semantics, they are often intertwined in cognition. Manner verbs (e.g., hit) often denote intended results and sometimes imply their realisation (Levin & Rappaport Hovav, Reference Levin, Rappaport Hovav, Pensalini, Turpin and Guillemin2014, p. 339; Talmy, Reference Talmy2000, pp. 262–267), and result verbs can often be associated with certain manners (e.g., clean is often achieved by wiping/scrubbing/washing/etc., Levin & Rappaport Hovav, Reference Levin, Rappaport Hovav, Pensalini, Turpin and Guillemin2014). The conceptual link between manner and result might lead to the co-existence of manner and result in a verb’s denotation in the mind, though presumably with different cognitive salience. Of note, we are not arguing for semantic co-existence of manner and result in a verb; rather, our discussion pertains to the cognitive representation of these elements.

Salience of event components is well-discussed in Motion events, focussing particularly on manner and path. Cognitively more salient components tend to be mentioned more frequently in speech (e.g., Gennari et al., Reference Gennari, Sloman, Malt and Fitch2002; Hohenstein et al., Reference Hohenstein, Eisenberg and Naigles2006; Ji et al., Reference Ji, Hendriks and Hickmann2011), allocated more visual attention (e.g., Ji & Hohenstein, Reference Ji and Hohenstein2017; Soroli & Hickmann, Reference Soroli, Hickmann, Marotta, Lenci, Meini and Rovai2011), processed faster (e.g., Ünal et al., Reference Ünal, Richards, Trueswell and Papafragou2021), considered dominant in similarity judgments (e.g., Hickmann et al., Reference Hickmann, Engemann, Soroli, Hendriks, Vincent and Ibarretxe-Antuñano2017; Pourcel, Reference Pourcel2003) and novel verb interpretation (e.g., Hohenstein et al., Reference Hohenstein, Naigles, Eisenberg, Halls and Waxman2004), and remembered better (e.g., Billman et al., Reference Billman, Swilley and Krych2000; Filipović, Reference Filipović2011, Reference Filipović2022; Flecken et al., Reference Flecken, von Stutterheim and Carroll2014; but see Engemann et al., Reference Engemann, Hendriks, Hickmann, Soroli and Vincent2015 and Papafragou et al., Reference Papafragou, Massey and Gleitman2002 for opposite findings). A general bias towards path over manner is often reported (e.g., see Santin et al., Reference Santin, van Hout and Flecken2021 and Ünal & Papafragou, Reference Ünal and Papafragou2019 for reviews). Of note, the relative salience of manner and path in these studies has been generally discussed as a property of the language (i.e., if participants mention manner more in language A than in language B, then manner is more salient in language A than in language B), rather than a property of the verb (i.e., how salient manner and path are in a verb’s meaning).

Relatively few studies have explored the salience of different event components as a property of the verb (e.g., Behrend, Reference Behrend1990; Forbes & Farrar, Reference Forbes and Farrar1993). For example, in Behrend (Reference Behrend1990), native English speakers viewed videos depicting novel scenes labelled with novel verbs and alternate videos with manner, result, agent, or instrument altered, judging whether the alternate videos could be described by the same novel verbs. It was found that adults tended to interpret novel verbs as encoding result rather than manner – or as result verbs in the view of manner/result complementarity.

However, although the studies above were targeted at verb meaning, it is difficult to pin down the factors underlying verb meaning interpretation. Were the novel verbs interpreted in such a way because that is how the verbs should be? Or could it stem from other factors? Indeed, Behrend (Reference Behrend, Tomasello and Mcrriman1995) pointed out three groups of factors influencing children’s interpretation of novel verbs: child-driven factors (e.g., perceptual and cognitive abilities), language-driven factors (e.g., syntactic framing, verb inflection), and environment-driven factors (e.g., adult verbal input). Although this remark was specifically made for children’s verb acquisition, it nevertheless showcases the diversity of factors that can contribute to one’s interpretation of verb meaning. Thus, although previous research does illuminate the relative salience of manner and result in verb meaning, it only offers ‘snapshots’ of how verbs are interpreted in specific experimental settings with specific combinations of factors. To ensure that the salience of manner and result is indeed examined as a property of the verb, one needs to examine how manner/result salience varies as a response to variation generated by verbs. Fulfilment types provide an ideal testbed for this. Given that verbs of the four fulfilment types encode intention and intention realisation to different degrees of specificity and certainty, if change in fulfilment types leads to change in manner/result salience, we can be confident that manner/result are examined as a verb property.

Apart from manner and result, the fulfilment type concept introduces intention into verb meaning representation. The idea of intention influencing verb semantics is present in Behrend and Wittek (Reference Behrend and Wittek2003) and Behrend and Scofield (Reference Behrend, Scofield, Hirsh-Pasek and Golinkoff2006), who showed that when intention is unknown (i.e., unfamiliar event scenes), the act of labelling an action with a (novel) verb encourages the belief that the action is intended (although in reality, the action might be a failed attempt to fulfil other goals), which in turn shapes novel verb interpretation. For example, in one scene the actor tried to lift a pretend piece of bread with a magnet on top using a spatula with a magnet on its underside; the attempt failed as the bread stayed put. When the actor’s action was not labelled with a novel verb, children tended to see through the failed attempt and imitate the real intention (i.e., to lift the bread). However, when the same activity came with a novel verb (e.g., ‘Look, I’m meeking!’), children would interpret the intention as not to lift the bread. Behrend and colleagues were primarily concerned with intention inference through language when intention is unclear. In the context of fulfilment types, we wish to explore how clear intention would shape the mental representation of verb meaning. The lack of literature in this regard calls for a new evaluation of the contribution of intention in the mental representation of verbs.

Though not as directly visible as manner and result, intention is present in almost all human behaviour (Pourcel, Reference Pourcel2004; Zwaan & Radvansky, Reference Zwaan and Radvansky1998) and has ‘causal effects in the physical world’ (Kim, Reference Kim2000, p. 31). Understanding intention facilitates event apprehension by activating the knowledge of how relevant events typically develop (Zwaan & Radvansky, Reference Zwaan and Radvansky1998), and the fulfilment of intention marks event boundaries (e.g., Carroll & von Stutterheim, Reference Carroll, von Stutterheim, Bohnemeyer and Pederson2010, p. 69; Gerwien & von Stutterheim, Reference Gerwien and von Stutterheim2018). Importantly, intention can boost the perceived smoothness in motion, making fragmented actions with short pauses appear more continuous (Peng et al., Reference Peng, Ichien and Lu2020) and shortening the perceived temporal lapse between the causing and the caused events (Haggard et al., Reference Haggard, Clark and Kalogeras2002; Moore et al., Reference Moore, Teufel, Subramaniam, Davis and Fletcher2013). Focussing on French motion verbs, Nakamura et al. (Reference Nakamura, Marque-Pucheu, Leclère and Laporte2021) noted that when manner-of-motion verbs such as courir (‘run’) co-occur with goal/purpose information (e.g., à sa chambre, ‘to his/her room’), they show characteristics of path verbs (e.g., no longer compatible with low-velocity adverbs such as doucement, ‘mildly’); this also suggests that intention can bring manner(-of-motion) verbs closer to result (path) verbs. The above findings suggest that intention can shorten the cognitive distance between action and result. Would intention also draw manner and result closer in the mental lexicon?

In light of manner/result complementarity (Rappaport Hovav & Levin, Reference Rappaport Hovav, Levin, Hovav, Doron and Sichel2010), intrinsic-/moot-/implied-fulfilment verbs would fall into the category of manner verbs (as they do not semantically entail intention realisation); attained-fulfilment verbs would be result verbs (as they semantically entail intention realisation). However, given the potential influence of intention-binding manner and result, would interpretation of verbs in the Realisation domain show nuanced verb representations beyond the manner/result split?

The idea of more nuanced categories within manner/result verbs is not new (e.g., Aurnague, Reference Aurnague2011; Sarda, Reference Sarda, Aurnague and Stosic2019; Stosic, Reference Stosic2009). For motion verbs, manner and path have been conceptualised as sets of independent features (e.g., directionality, intentional opposition to a force, and so forth for path, Aurnague, Reference Aurnague2011; speed, force, means, and so forth for manner, Stosic, Reference Stosic2009, p. 111). These features are organised in family resemblance, and a verb can denote one or more of these features (Sarda, Reference Sarda, Aurnague and Stosic2019); the more relevant factors a verb contains, the more typical manner/result verb representation it has. Thus, motion verbs tend to simultaneously contain elements of both manner and path, contrary to manner/result complementarity (e.g., Rappaport Hovav & Levin, Reference Rappaport Hovav, Levin, Hovav, Doron and Sichel2010). However, these studies differ from the current discussion in two aspects. First, previous research concerns verb semantics, while the present discussion concerns the cognitive representation of verbs. Second, previous studies deconstructed manner/result into independent features; motion verbs fall into sub-classes because they contain different (numbers of) semantic components, so the fine-grained classification is multi-dimensional in nature. In the present study, manner and result are conceptualised as a unified cline, and variation among verbs is regarded as differences in a single dimension.

Specifically, we expect intrinsic-fulfilment verbs to be the closest to having a manner verb representation, while attained-fulfilment verbs are the closest to having a result verb representation, with moot- and implied-fulfilment verbs located increasingly further away from manner interpretation and closer to result interpretation.

3. The present study

This study contains two experiments. Experiment 1 is a novel verb learning task exploring how fulfilment types would influence the salience of manner and result in verb meaning interpretation. Experiment 2 measures the mental relatedness between intrinsic-, moot-, implied-, and attained-fulfilment verbs and their semantically corresponding result (i.e., attained-fulfilment) verbs in English, in order to examine whether verbs with the four fulfilment types differ in cognitive proximity to result verb counterparts. Together, the experiments investigate whether verbs of different degrees of fulfilment are represented in a binary (as the manner/result split) or in a more nuanced way.

Participants in both experiments were adult monolingual British English speakers (all British nationals resident in the UK) with normal or corrected-to-normal vision and without neurological or psychiatric disorders or literacy difficulties recruited via Prolific (www.prolific.co). The data in this study was collected from 2021 to Spring 2022, and thus participation was remote via the online experiment builder Gorilla (www.gorilla.sc) due to COVID-19 restrictions. Each participant received a small monetary reward upon full completion. All statistical analyses were performed on R (version 4.1.2, R Core Team, 2021) with the initial alpha set to 0.05 and Bonferroni-adjusted p values in post hoc analyses. The experiments were piloted to ensure the validity of materials, experiment design, and procedure. All data, R codes, pilot studies, stimuli, additional analyses, and extra tables are available as appendices at https://osf.io/jpaqu/?view_only=8880dc4deac84e9988070e455e1c56aa.

3.1 Experiment 1: Novel verb learning task

Given the imbalanced cognitive salience of manner and result (Section 2), it is unlikely that these features carry equal weight in verbs with different fulfilment types. We hypothesised that the four fulfilment types are represented along a cline between manner and result verbs. We tested this hypothesis through a novel verb learning task. Specifically, novel verbs with intrinsic-, moot-, implied-, and attained-fulfilment properties should yield progressively higher result salience and lower manner salience in verb meaning interpretation.

3.1.1 Forbes and Farrar ( Reference Forbes and Farrar 1993 ) and the challenge of the present experiment

Before explaining the current experiment, we first review a particularly relevant study. Forbes and Farrar (Reference Forbes and Farrar1993) examined how manipulation of the salience of event components would influence novel verb interpretation. Focussing on manner, result (‘outcome’ in their terminology), agent, and instrument, they trained each novel verb with three videos that were said to exemplify the verb. The authors manipulated the frequency of the training videos displaying the relevant event components: in the same condition, the event components remained the same across the training videos; in the different condition, one event component changed in each video while all other components held constant (thereby informing participants that the changing component was unimportant for the novel verb’s meaning); in the mixed condition, one event component remained constant while all other components varied in each video. After training for each novel verb, participants were shown a test video featuring a changed event component and judged whether it was an acceptable instance of the novel verb. Adult participants were reluctant to generalise novel verbs to videos containing manner and result changes compared to agent and instrument changes, suggesting that manner and result were considered as more salient components of verb meaning, but the salience of manner and result did not significantly differ. Importantly, the frequency of adults’ extensions for each event component did not significantly vary across the same/different/mixed training conditions (Forbes & Farrar, Reference Forbes and Farrar1993, p. 12), indicating stable perceptions of the salience of the event components.

Having reviewed Forbes and Farrar (Reference Forbes and Farrar1993), let us now discuss its relevance to the present study. A challenge in our experiment is to conjure up fulfilment type readings for novel verbs. It might be useful to generate fulfilment type readings in novel verbs using frequency-based methods (akin to Forbes & Farrar, Reference Forbes and Farrar1993). For example, showing several teaching videos varying in results to conjure up an intrinsic-fulfilment verb reading; showing teaching videos with identical results to encourage an implied-fulfilment verb reading. There are two reasons why this paradigm might not suit our current purpose. First, as mentioned above, the three training conditions in Forbes and Farrar (Reference Forbes and Farrar1993) did not elicit significantly different degrees of salience associated with the event components (for adults), suggesting that frequency-based training might not effectively assign different fulfilment type readings to novel verbs. Second, even if the frequency-based strategy was effective in distinguishing verbs in different training conditions, it would be difficult to decide what frequency of result match would generate which fulfilment type reading – how many different-result teaching videos would be necessary to generate intrinsic-/moot-/implied-/attained-fulfilment readings? Thus, we chose to alternatively generate fulfilment type readings for novel verbs with definitions (Section 3.1.3.2).

3.1.2 Participants

A priori calculation on G*Power (version 3.1.9.4) for this experiment yielded a sample size of 100. In total, 172 participants took part in Experiment 1. Among them, 51 (29.65%) dropped out, 2 (1.16%) were timed out by Prolific, and 16 (9.30%) participants were rejected due to lack of attention. The remaining 103 participants were monolingual English speakers (mean age: 36.30 years; age range: 20–56 years; 49 females; ethnicity: 2 Asian, 2 Black, 4 mixed, 95 White; student status: 68 not student, 5 students, 30 unknown; employment status: 46 full-time, 15 part-time, 9 unemployed, 33 unknownFootnote ⁴).

3.1.3 Materials

3.1.3.1 Video stimuli

This experiment used 10 sets of silent videos showing novel activities 3 to 8 s long. In each set of 4 videos, the teaching video depicted a manner and a result. One alternate video was identical to the teaching video except that the result was not successfully fulfilled (‘Same-Manner-Unfulfilled-Result’, or SMUR). Another alternate video also had the same manner as the teaching video but showed a different result (‘Same-Manner-Different-Result’, or SMDR). A third alternate video depicted the same result as the teaching video but showed a different manner (‘Different-Manner-Same-Result’, or DMSR). While the SMDR and DMSR videos are respectively analogous to the alternate same-manner-different-path and same-path-different-manner videos used in classic similarity judgment tasks in Motion research, the present experiment further involved SMUR videos. This setup allows a more thorough representation of event scenarios. While in SMDR videos manner obligatorily comes with some result, SMUR videos acknowledge that manner can exist without (conventionally received) results obtaining, thereby allowing a more nuanced examination of how one interprets the verb. To illustrate, in the ‘Walnut’ video set, the teaching video depicts a situation in which the actor cracks a walnut by stapling it with a stapler. In the SMDR video, the actor staples a walnut and the walnut slips off. In the SMUR video, the actor staples the walnut, but the walnut stays still and remains closed. In the DMSR video, the actor cracks a walnut by hitting it with a stapler.

Videos within each set had the same duration. To ensure novelty, the objects involved in the teaching videos were not used in ways that they would typically be used to reduce ‘object affordances’Footnote ⁵ (i.e., common uses for objects, Huang et al., Reference Huang, Heyes and Charman2002). The teaching video of each video set was rated in terms of familiarity with a different group of 19 native British English speakers recruited through institution platforms. Participants were asked to rate how familiar the activity in each video looked to them on a Likert scale from 1 (very uncommon) to 5 (very common). All teaching videos were rated between 1.6 and 1.9, suggesting that the scenes were fairly novel (video descriptions and ratings available in Appendix 1). The videos were designed in such a way that the actions were highlighted, showing only the actor’s hands, to preclude potential stereotypes about the actor’s appearance, gender, ethnicity, and so forth.

3.1.3.2 Novel verbs and their definitions

Each set of videos was paired with a novel verb, making a total of 10 novel verbs. The novel verbs were mono-syllabic 3–5 letter long words borrowed from previous studies involving English speakers to ensure that they satisfy phonotactic restrictions of English. The novel verbs were: dotch (Chan et al., Reference Chan, Tardif, Chen, Pulverman, Zhu and Meng2011), krad, zub (Hohenstein, Reference Hohenstein2005), smick (Buccola et al., Reference Buccola, Dautriche and Chemla2018), jav, glorp (Imai et al., Reference Imai, Li, Haryu, Okada, Hirsh-Pasek and Golinkoff2008), bock, stife (Behrend, Reference Behrend1990), spog (Kersten et al., Reference Kersten, Meissner, Lechuga, Schwartz, Albrechtsen and Iglesias2010), and hirsh (Maguire et al., Reference Maguire, Hirsh-Pasek, Golinkoff, Imai, Haryu, Vanegas, Okada, Pulverman and Sanchez-Davis2010).

The novel verbs in this experiment were defined to take on different fulfilment types. Having shown that a frequency-based approach might not effectively serve this purpose (Section 3.1.1), we present here a semantically-based approach. In the intrinsic-fulfilment condition, novel verbs were given manner-only definitions due to their lack of goal information. For example, in one scene, someone crushed a walnut placed in a stapler by stapling it using their right hand; the novel verb for this scene, hirsh, was defined as ‘staple using right hand’ in the intrinsic-fulfilment condition. In the moot-fulfilment condition, the uncertainty in intention realisation was highlighted through the try [result] by [manner] construction, and hirsh here was defined as ‘try crushing walnut by stapling using right hand’. In the implied-fulfilment condition, the infinitive was used to create the implication of intention realisation and to downplay the sense of uncertainty with the formula of [manner] in order to [result], and hirsh was defined as ‘staple using the right hand to crush walnut’. The attained-fulfilment condition had result-only definitions, as attained-fulfilment verbs semantically entail result and no manner; hirsh was thus defined as ‘crush walnut’. Apart from these defined conditions, a no-definition condition served as control.

Considering that the novel verbs were defined, a possible criticism is that the acceptability judgments would simply reflect participants following these definitions, rather than participants’ own biases for manner and result. However, let us clarify that Experiment 1 is not a conventional novel verb learning task – this paradigm is only adopted to legitimately present the definitions. Our aim is not to explore the tendency to attend to manner or result – this can only happen when participants freely interpret the novel verbs. When a verb has a specific fulfilment type, its meaning is no longer subject to free interpretation in terms of manner and result. Rather, we are only concerned with manner and result salience under the influence of the definitions. Thus, we would need participants to follow the definitions instead of showing their own manner/result biases.

During the display of teaching videos, the novel verbs were introduced in written form in the transitive and the progressive (e.g., Someone was hirshing the walnut), each with a definition underneath (‘[novel verb] = [definition]’), except for the no-definition condition. Motion research shows that syntactic framing could influence novel verb interpretation (e.g., Naigles & Terrazas, Reference Naigles and Terrazas1998). The perfective and the transitive tend to trigger path verb (and possibly result verb) interpretations, while the progressive and the intransitive encourage manner-of-motion verb (and potentially manner verb) readings. For the present experiment, however, syntactic framing might not be much of a concern, because the novel verbs were explicitly defined apart from the no-definition condition. Nevertheless, effort was made to neutralise the impact of syntax. As Realisation events depict agentive events (Jia & Li, Reference Jia and Li2015), the transitive frame is necessary. To downplay the influence of the transitive frame, the progressive aspect was used, as these two frames have opposing influences on meaning interpretation and might thus offset one another.

3.1.4 Procedure

Participants were randomly assigned to one of the five fulfilment type conditions. In the novel verb learning phase, participants viewed 10 sets of novel videos, randomised for order. For each set, the teaching video was shown first with a sentence introducing the novel verb, with or without a definition depending on the condition. Next, participants viewed three alternate videos one at a time randomised for order. Each video was played manually by clicking on the video, and participants could view each video up to two times by clicking on it again after the first display (clicks during the video display were invalid). Following each alternate video, participants were asked the following questions: 1) Is this [novel verb]-ing? (e.g., Is this hirshing?) 2) Does the actor have the same manner as in the teaching video? 3) Does the actor have the same result as in the teaching video? Answers were recorded by Yes/No buttons under each question. Prior to the task, there was a practice trial to familiarise participants with the procedure. The entire task lasted 10 to 15 min.

As the classic novel verb learning task often only involves acceptability questions, it is necessary to clarify why manner and result judgments were added. Manner and result judgments were conducted to ensure that participants perceived manner and result in the way the videos were configured (e.g., in DMSR videos, participants should see different manners and identical results to the teaching videos). Our data suggested that this was not always the case, so we only analysed the observations containing the expected manner and result judgments to avoid contamination of data.

This setup not only examines whether novel verbs in each fulfilment type condition are interpreted as manner/result verbs, but also reveals whether the verb meaning interpretations across conditions fall into two or more groups. If only two clusters emerge, the manner/result complementarity would be cognitively confirmed; if more groups emerge, manner/result complementarity could be questioned as the only representation of verbs in the mental lexicon.

3.1.5 Data analysis

For each of the three judgment tasks, 3090 observations were collected. These included 9 (0.29%)Footnote ⁶ blank data points where no responses were given (8 acceptability judgments, 1 result judgment), 2 (0.06%) acceptability judgments made before the relevant teaching videos were displayed, and 3 (0.10%) impossibly fast responses made within 200 ms (1 acceptability judgment, 1 manner judgment, 1 result judgment). To maintain completeness of observation, the problematic data were removed along with the corresponding data in the other judgments. The remaining 3076 (99.55%) observations were then sifted according to whether they yielded the expected interpretation of manner and result in the videos.Footnote ⁷ This left 2522 (81.62%) acceptability judgments. Positive and negative acceptability judgments were coded as ‘1’ and ‘0’, respectively.

To examine how novel verbs in different fulfilment type conditions vary in terms of manner and result salience, we performed a binomial mixed-effect logistic regression (glmerlme4 package, Bates et al., Reference Bates, Mächler, Bolker and Walker2015), which predicted acceptability judgments (yes/no) based on fulfilment type condition (intrinsic/moot/implied/attained/no-definition) and video type (SMUR/SMDR/DMSR). A random intercept of participants was added to minimise non-independence issues (e.g., Winter & Grice, Reference Winter and Grice2021) and enhance the generalisability of findings (e.g., Roettger, Reference Roettger2021; Yarkoni, Reference Yarkoni2020). Post hoc estimated marginal squares analysis (emmeansemmeans package, Lenth, Reference Lenth2022) was conducted with Bonferroni-adjusted p values and odds ratios (ORs)Footnote ⁸ for effect sizes.

3.1.6 Results

Table 1 presents the raw counts and percentages of positive acceptability judgments. It is clear that the intrinsic-/moot-/implied-fulfilment and no-definition conditions all elicited over 60% acceptability on the manner-matching SMUR and SMDR videos and lower than 35% acceptability on the result-matching DMSR videos, suggesting that the novel verbs in these conditions were interpreted overall as manner verbs in the traditional sense. For novel verbs in the attained-fulfilment condition, acceptability was low for SMUR (8.74%) and SMDR (12.40%) videos but high for DMSR videos (99.40%), indicating a stronger result verb interpretation.

Table 1. Raw counts and percentages of positive acceptability judgments

Abbreviations: SMUR, same-manner-unfulfilled-result; SMDR, same-manner-different-result; DMSR, different-manner-same-result.

The best-fit model with the lowest Akaike Information Criterion (AIC) returned both fulfilment type condition and video type as significant fixed effects and an interaction between the two (Table 2).

Table 2. Generalised linear mixed model fit by maximum likelihood for acceptability

Abbreviations: AIC, Akaike Information Criterion; BIC, Bayesian Information Criterion; SD, standard deviation; SE, standard error; SMUR, same-manner-unfulfilled-result; SMDR, same-manner-different-result; DMSR, different-manner-same-result.

Post hoc analysis (Table 3) was conducted comparing different fulfilment type conditions by video type to explore the number of groupings that would appear; more than two groups would call into question the accuracy of manner/result complementarity in the cognitive interpretation of verbs.

Table 3. Post hoc analysis for acceptability judgments

Abbreviations: SMUR, same-manner-unfulfilled-result; SMDR, same-manner-different-result; DMSR, different-manner-same-result; OR, odds ratio; 95% CI, 95% confidence interval.

In manner-matching SMUR videos, the attained-fulfilment condition yielded the lowest acceptability. Further, acceptability for the implied-fulfilment condition was significantly lower than that for the intrinsic- and moot-fulfilment conditions.

For manner-matching SMDR videos, the attained-fulfilment condition again yielded the lowest acceptability. The intrinsic-fulfilment condition led to significantly higher acceptability than the moot-fulfilment, implied-fulfilment, and no-definition conditions.

Turning to result-matching DMSR videos, the attained-fulfilment condition this time elicited the highest acceptability. The no-definition condition took the second highest place. The intrinsic-, moot-, and implied-fulfilment conditions clustered together at the lower end.

In summary, novel verbs assigned intrinsic-/moot-/implied-fulfilment properties yielded higher acceptability in manner-matching videos (i.e., mentally represented as manner verbs), while novel verbs in the attained-fulfilment condition yielded higher acceptability in result-matching videos (i.e., mentally represented as result verbs). The clustering patterns across the fulfilment type conditions show that they often fell into more than two groups, confirming the prediction that verbs (in the Realisation domain) cannot be neatly subsumed under the manner/result dichotomy in the mental lexicon. What deserves attention is that participants in the non-defined condition behaved more similarly to participants in the intrinsic-, moot-, and/or implied-fulfilment conditions than to those in the attained-fulfilment condition (see Table 1), suggesting that these verbs are naturally interpreted as having manner verb representation, but not so pure as that of the intrinsic-fulfilment condition. Finally, the fact that manner-matching SMUR and SMDR videos yielded more groupings than result-matching DMSR videos suggests that the variation among different fulfilment type conditions was driven by novel verbs with manner verb status.

3.1.7 Discussion

Experiment 1 offered two findings. First, although verbs in the four defined fulfilment type conditions could be categorised into two general groups (i.e., intrinsic-/moot-/implied-fulfilment vs. attained-fulfilment), echoing manner/result complementarity (Rappaport Hovav & Levin, Reference Rappaport Hovav, Levin, Hovav, Doron and Sichel2010), within-group variation was observed for manner verbs (i.e., novel verbs whose meaning interpretation hinged more on manner sameness than on result sameness). Second, non-defined verbs were generally interpreted as manner verbs, in particular resembling verbs in the moot-/implied-fulfilment conditions.

For the first point, the intrinsic-/moot-/implied-fulfilment conditions yielded higher acceptability in manner-matching videos (indicating manner verb representation), and the attained-fulfilment condition elicited higher acceptability in result-matching videos (indicating result verb representation). This is hardly surprising given how the novel verbs were defined. What deserves more attention is the variation among the intrinsic-/moot-/implied-fulfilment conditions, which display increasingly less dependence on manner sameness and more dependence on result sameness in acceptability judgments. This could be due to the different degrees of specificity in the intention and implications of intention realisation encoded in these verbs, showcasing how intention bonds manner and result and blurs verbs’s categorical boundaries.

Having mapped out the clustering pattern of the four defined conditions, let us move on to the no-definition control. A closer look at Table 1 shows that participants in the non-defined condition tended to base acceptability primarily on manner matches, but they rated these verbs somewhere between verbs in the attained-fulfilment condition and the other three conditions. The non-defined verbs were mainly interpreted as manner verbs deviated from the well-observed tendency to interpret novel verbs as path verbs rather than manner verbs in previous Motion studies (e.g., Behrend, Reference Behrend1990; Ünal et al., Reference Ünal, Richards, Trueswell and Papafragou2021). One possible explanation is that the activities in our video stimuli were all facilitated by instruments, which has been argued to highlight manner salience (Ji & Hohenstein, Reference Ji and Hohenstein2017; Pourcel, Reference Pourcel2004). A second account concerns the ‘efficiency’ of verb meaning. According to Behrend (Reference Behrend1990, p. 688), compared to manner and result verbs, instrument verbs (e.g., hammer, saw) are more ‘efficient labels’ for event description, as they often combine instrumental information with manner (e.g., hammer conjures up a pounding manner) or result (e.g., saw implies that something is cut). Similarly, moot-/implied-fulfilment verbs might be more efficient labels than intrinsic-/attained-fulfilment verbs, as the former cognitively denote both manner and result, thereby maximising information. The pursuit of efficiency might have also underlain the tendency to interpret the non-defined verbs as encoding both manner and result. A third reason might be the influence of syntactic framing on meaning interpretation (e.g., Gillette et al., Reference Gillette, Gleitman, Gleitman and Lederer1999; Naigles & Terrazas, Reference Naigles and Terrazas1998). Recall that the gerundive intransitive frame was used for the current acceptability judgment prompt (i.e., ‘Is this v.-ing?’), which might have encouraged a manner verb reading. Although the novel verbs were introduced in the relatively neutral syntactic frame of ‘Someone is v.-ing something’ in the definitions, the prompt might have been more recent to – and thus had a greater impact on – acceptability judgments. Notwithstanding, syntax might not impact verb meaning interpretation in Realisation as strongly as in Motion research. The syntactic influence on meaning interpretation of Motion verbs partly stems from the fact that manner-of-motion verbs can seldom be used transitively with locations (e.g., *I ran the room) but path verbs can (e.g., I entered the room). Conversely, both manner and result verbs in Realisation can be used transitively with affected objects (e.g., I washed/cleaned the shirt), so compatibility with the transitive frame no longer distinguishes between these two verb categories.

Method-wise, future research can trial the frequency-based training method mentioned in Section 3.1.1 and consider the rates of result match in training videos needed to generate a specific fulfilment type reading in novel verb meaning interpretation.

Summarising, Experiment 1 attested more than two patterns of manner and result salience in the interpretation of novel verbs assigned the four fulfilment types, suggesting that these verbs do not neatly conform to manner/result complementarity in cognition. It is worth noting that the effect sizes in Experiment 1 tend to be very large. This might be because the definitions served as strict distinctions for novel verbs in different fulfilment type conditions. In verbs encountered in real language, the distinctions might not be as clear-cut, which makes it important to examine genuine verbs with different fulfilment types. This was done in Experiment 2.

3.2 Experiment 2: Mental relatedness task

Having shown that novel verbs of different fulfilment types correspond to distinct verb representations, we explored whether the same would apply to genuine English verbs. The novel verb learning paradigm was understandably no longer suitable, so we examined the clustering patterns of verbs by measuring the semantic relatedness between the verbs through a recall task in the paired-associate learning paradigm. The rationale is that semantically related words have a recall advantage compared to unrelated words, that is, the ‘semantic relatedness effect’ (e.g., Kowialiewski et al., Reference Kowialiewski, Krasnoff, Mizrak and Oberauer2022 and references therein).

The next step is to decide which verbs’ semantic relatedness to measure. Often, manner verbs have semantically compatible result verbs that can describe the same scenes (cf. Behrend, Reference Behrend1990). For example, I hunted (manner verb) vs. arrested (result verb) the fugitive and I washed (manner verb) vs. cleaned (result verb) the shirt can respectively describe the same scenarios and only differ in terms of certainty of intention realisation. Given the different extents to which intrinsic-/moot-/implied-fulfilment verbs theoretically encode intention and imply its realisation, these verbs would arguably show varying cognitive distances with semantically compatible attained-fulfilment verbs. To illustrate, the distance between hunt (moot-fulfilment) and arrest (attained-fulfilment) would be greater than that between wash (implied-fulfilment) and clean (attained-fulfilment), because hunt encodes less certainty of intention realisation than wash does.

Revisiting the semantic relatedness effect, if a manner verb and its semantically compatible result verb can be recalled well based on one another, they are mentally close. Therefore, if manner verbs of different fulfilment types show a range of relatedness to their compatible result verbs, it could show that they do not constitute a homogeneous manner verb group, thereby evidencing the existence of a mental cline between manner/result verb representations.

3.2.1 Participants

A priori calculation on G*Power (version 3.1.9.4) for this experiment yielded a sample size of 48. A new group of 101 adult monolingual English speakers were recruited, among whom 3 (2.97%) were rejected for breaching instructions, 27 (26.73%) dropped out, and 3 (2.97%) were timed out by Prolific. Data from the remaining 68 (67.33%) participants (mean age = 33.81 years; age range: 20–50 years; 34 females; ethnicity: 2 Black, 2 mixed, 64 White; student status: 44 no student, 6 students, 18 unknown; employment status: 34 full-time, 7 part-time, 5 unemployed, 22 unknown) was used.

3.2.2 Materials

This experiment used 42 pairs of verbs. Each verb pair contained a target verb that was intrinsic-, moot-, implied-, or attained-fulfilment and a semantically compatible attained-fulfilment verb (e.g., wash [implied-fulfilment] – clean [attained-fulfilment]). The 42 target verbs were 12 intrinsic-fulfilment, 12 moot-fulfilment, 12 implied-fulfilment, and 6 attained-fulfilment verbs. This means that 36 pairs featured verbs with different fulfilment types, and 6 pairs contained attained-fulfilment verbs only. Attained-fulfilment verbs were compared between themselves to provide a thorough comparison for the fulfilment type hierarchy – as both verbs in such pairs are result verbs, they should reveal the shortest semantic distance compared to pairs featuring different fulfilment types. For simplicity, verb pairs containing the four types of target verbs are hereafter abbreviated as intrinsic_T/moot_T/implied_T/attained_T pairs. The verbs in each pair were presented side by side on the screen, with target verbs appearing on the left in half of the trials and on the right in the other half for each participant.

As word frequency is a well-known factor influencing recall performance (e.g., Gregg, Reference Gregg and Brown1976), we collected the frequency of each verb via Google Books Ngram Viewer (https://books.google.com/ngrams/, British English corpus 2019) for the year 2018. A univariate ANOVA was performed on R (version 4.1.2, R Core Team) to explore whether word frequencies of verbs varied across the four fulfilment type groups. No significant difference was found (F(3) = 0.34, p = .798), suggesting that word frequency could not have significantly influenced recall performance. Full frequency data and analysis are available in Appendix 3.

The verb pairs were also given context sentences outlining intention in half of the trials for each participant. For example, the pair pull (intrinsic-fulfilment) – pluck (attained-fulfilment) was given the following context: A fruit was hanging from the branch. I wanted to have it. This factor was included to examine whether verb-external cues could strengthen the relatedness between verbs.

3.2.3 Experiment design

The present experiment had three within-subjects variables: a) context (present/absent), b) fulfilment type of the target verb (intrinsic/moot/implied/attained), and c) presentation side of the target verb (left/right). The side-of-presentation factor was considered because the recall task required responses for ‘verbs on the right of the screen’ based on ‘verbs on the left of the screen’ in the instructions. This factor is apparently inapplicable to attained_T pairs, but to ensure consistency, one verb in each attained_T pair was nevertheless designated as the target verb and also shifted in side of presentation. Among the 42 verb pairs, the total number of attained_T pairs (i.e., 6 pairs) is half that of any other type of verb pairs (i.e., 12 pairs) because shifting in side of presentation in attained_T pairs would not alter the fulfilment types of the verbs on each side.

To avoid high cognitive load and floor effects in recall performance, each participant completed only 28 trials, instead of the full list of 42 pairs. The 28 pairs comprised 2 pairs per each side of presentation and per context condition for intrinsic_T, moot_T, and implied_T pairs, as well as 2 attained_T pairs per context condition. For each participant, if the target verbs in two attained_T pairs appeared on the left with context, the target verbs of the other two attained_T pairs would appear on the right without context, and vice versa. The experiment design from the perspective of a single participant is given in Table 4.

Table 4. Design of Experiment 2 from the perspective of a single participant

3.2.4 Procedure

This experiment consisted of two phases connected by a distractor task. In the encoding phase, the verbs in each pair were displayed side by side in the upper middle of the screen, one pair per trial and randomised for order. For half of the trials, context sentences appeared above the verbs. For each verb pair, participants were asked Does the verb on the right have the same intended result as the verb on the left? After all pairs were completed, a 2-min sound-detecting task served as a distractor (‘How many times did you hear sea waves in this music audio?’). Next, participants were shown the verb previously presented on the left for each pair and recalled the verb on the right by typing it (or ‘?’ when they could not remember a verb) into a designated textbox. No contextual sentences were shown in the recall phase, but the side of presentation within each pair was kept identical to the encoding phase. The verbs in each pair were displayed in infinitive to do form rather than the base form, as bare verbs can sometimes serve as nouns (e.g., the bare word hammer is ambiguous between a tool and an action, see Vigliocco et al., Reference Vigliocco, Vinson and Siri2005). The entire task lasted about 10 min.

A potential criticism for asking for intention judgments is that it encourages participants to think in intention/result terms. This encouragement is necessary because only when one thinks in intention/result terms can one discuss fulfilment types,Footnote ⁹ which by definition are based on the recognition of intended results. Moreover, intention judgments can tell us whether and how easy it is to recognise the same intention in a pair of verbs, which provides further evidence that the target verbs in the pairs truly differ in their denotation of intention (recall that intrinsic-fulfilment verbs are theorised to denote no intention).

Due to space limitations, we only present below the analysis for the more relevant recall task; analysis and discussion for intention judgments are presented in Appendix 5. For recall accuracy, we expected intrinsic-, moot-, implied-, and attained-fulfilment verbs to be increasingly related to their corresponding result verbs, manifesting as better recall. It was further expected that the presence of context would facilitate recall.

3.2.5 Data analysis

For recall performance, 1904 data points were collected, which were coded as ‘1’ (correct) or ‘0’ (incorrect/non-attempts). We performed a binomial mixed-effect logistic regression predicting recall accuracy (1/0) based on target verbs’ fulfilment types (intrinsic-/moot-/implied-/attained-fulfilment), their side of presentation (left/right), their interactions, as well as a random intercept of Participant. Regression was performed on all possible combinations of variables and their interactions, and we present below the best-fit model with the lowest AIC, which removed context from the analysis. Indeed, context did not prove to significantly predict recall performance when included in any model. Post hoc analysis followed the procedure outlined in Section 3.1.5.

3.2.6 Results

Table 5 presents the raw counts and percentages of correct recalls in all types of verb pairs.

Table 5. Raw counts and percentages of correct recalls

Table 6 presents the best-fit logistic regression model. Fulfilment types and side of presentation of target verbs were both returned as significant fixed effects; there was further an interaction between the two.

Table 6. Generalised linear mixed model fit by maximum likelihood for correct recalls

Abbreviations: AIC, Akaike Information Criterion; BIC, Bayesian Information Criterion; SD, standard deviation; SE, standard error.

Pairwise comparisons (Table 7) were performed to follow up on the fixed effect of fulfilment type and the interaction. For the main effect of side of presentation, as this factor is binary (left/right), it is clear from the regression model that target verbs appearing on the right elicited poorer recall accuracy than target verbs appearing on the left (β = −0.64, p = .030). Thus, the calculation in the regression model already includes the only possible pairwise comparison for this effect, so it is not included in the post hoc analysis.

Table 7. Post hoc analysis for recall accuracy

Abbreviations: OR, odds ratio; 95% CI, 95% confidence interval.

Following up the fixed effect of fulfilment type, significant distinctions were found between all fulfilment types except between intrinsic_T and moot_T pairs, with attained_T pairs eliciting the best recall performance and intrinsic_T/moot_T the poorest. Turning to the interaction between fulfilment type and side of presentation, we found the same pattern, no matter whether target verbs appeared on the left or right.

In summary, recall performance in Experiment 2 showed increasing accuracy for intrinsic_T, moot_T, implied_T, and attained_T pairs, although the difference between intrinsic_T and moot_T did not reach significance.

3.2.7 Discussion

The recall task measured the cognitive relatedness between verbs with different fulfilment types to their semantically compatible result verbs. Understandably, attained-fulfilment verbs, being result verbs, were the closest to their corresponding result verbs and yielded significantly better recall performance than the other verb pairs. What deserves more attention is that intrinsic-, moot-, and implied-fulfilment verbs did not form one homogeneous group (as would be expected of them due to their semantic status as manner verbs) but split into two groups, with intrinsic- and moot-fulfilment verbs clustering together with poorer recall performance and implied-fulfilment verbs independently forming another group with better memory accuracy. The former cluster suggests that although moot-fulfilment verbs, compared to intrinsic-fulfilment ones, denote intended results, this does not necessarily draw them mentally closer to result verbs; rather, it is the implication of intention realisation (as in implied-fulfilment verbs) that mentally demarcates manner verbs in terms of their proximity to result.

Summarising, while interpreted distinctively from attained-fulfilment (i.e., result) verbs, intrinsic-/moot-/implied-fulfilment (i.e., manner) verbs are interpreted with significant within-group variation. The better recall performance elicited by implied-fulfilment verbs than by intrinsic-/moot-fulfilment verbs supports that the former are mentally represented as closer to result verbs than the latter are. This suggests that intrinsic-, moot-, and implied-fulfilment verbs cannot be subsumed under one homogeneous group, inviting the conclusion that there are more than two types of verbs in the mental lexicon.

4. General discussion

In this study, we investigated whether verbs of the four fulfilment types (Talmy, Reference Talmy2000, pp. 262–268) fall into more than two verb representations in the mental lexicon. Through a novel verb learning task, Experiment 1 showed that novel verbs whose definitions entail no result (i.e., having a semantic status of manner verbs) vary in manner and result salience. Experiment 2 revealed through a recall task that manner verbs with different fulfilment types differ in terms of their mental distance with relevant results. This variation suggests that verbs of different fulfilment types have more nuanced representations in the mind than the dichotomous manner/result verb split.

Of note, although Experiments 1 and 2 offer similar results, they are not comparable due to their disparate experimental paradigms. Future studies might seek to use a single task paradigm to assess both novel and genuine words with respect to intention/fulfilment types and manner/result. Relatedly, recall that the definitions in Experiment 1 might have heightened the distinctions between the four fulfilment types, so future studies are necessary to see whether fulfilment type readings can be generated without definitions. Another limitation is that intention was presumed to be the sole explanation for the manner/result blur. Although this derives from Talmy’s fulfilment type framework, there might be other factors modulating verbs’ mental representations as manner/result verbs. A further limitation is that this study examines English only, and we aim to expand this line of research to cross-linguistic settings in future studies. In addition, although both experiments reflect activities involved in general learning scenarios, external validity might be less relevant for Experiment 1 due to the involvement of novel scenes and nonce verbs, compared to Experiment 2. Moreover, in-person data collection, had it been feasible for this study, might have provided better data quality than remote participation. In-person participation allows participants to ask questions about the experimental procedures to better follow the instructions. Face-to-face participation can additionally control the practical factors involved in experimentation (e.g., screen size/lightness/resolution; internet speed), which could minimise variation in the quality of video display and thus ensure that unexpected manner/result judgments were not due to unsatisfactory display quality. Additionally, the face-to-face mode makes it easier to verify whether participants meet the sampling criteria (e.g., language status, literacy).Footnote ¹⁰ Future empirical studies would benefit from mixed modes of participation to cross-examine the validity of experimental procedures and accessibility to participants in both settings. Further, our participants were predominantly white British nationals resident in the UK; the generalisability of the current findings might increase should a broader range of demographic factors be considered. Despite these limitations, let us see what insights the present study can offer.