The semantics of embedding predicates influence the acceptability of internally headed relative clauses in Korean

List of abbreviations

A

actor

ACC

accusative

ADN

adnominalizer

ART

article

CAUS

causative

COMP

complementizer

CONTIN

continuous

COP

copula

DECL

declarative

e

exclusive

ERG

ergative

GEN

genitive

INTR

intransitive

IPFV

imperfective

LOC

locative

NMZ

nominalizer

NOM

nominative

ns

non-singular

p

plural

PFV

perfective

PROG

progressive

PRS

present

PST

past

PUT

future

REL

relative clause

s

singular

SUBJ

subject

TOP

topic

U

undergoer

1 Background

Similar to other aspects of linguistic knowledge, argument structure undergoes continuous change (Diessel 2019). This study examines how such change occurs in the type of verbs associated with a particular schema. Focusing on internally headed relative clauses (IHRCs) in Korean, we propose that IHRC construals emerge by extending verb usage. Furthermore, we examine whether item-based analogy influences the innovative use of argument-structure constructions, forming Korean IHRCs as either partially productive or verb-class-specific constructions (e.g., Boas 2003; Croft 2012; Goldberg 1995, 2019; Perek 2016; Suttle and Goldberg 2011). We argue that a usage-based approach helps explain the limited distribution of Korean IHRCs and enriches our knowledge of the interaction between two entities in the cognitive model given by Verspoor (2000).

A relative clause (RC) construction comprises a head noun and an RC. The semantic function of the head noun is to establish a set of entities (Keenan and Comrie 1977), and that of an RC is to provide information about the head noun. The information given by an RC either assists in focusing – or restricting – the reference of the head noun or provides further background information about the head noun, which is already uniquely identified (Dixon 2010, p. 314). In (1), for instance, “who was bleeding” can be identified as an RC narrowing down the denotation of the head noun “the man” by providing information about him.

In (1), the head noun appears before the RC. English is a language that uses postnominal RCs. Expanding our investigation to different languages reveals other types of RCs in terms of the position of the head noun vis-a-vis the RC. The Korean examples in (2) and (3) illustrate this.

The RC in (2) is an example of a prenominal RC, with the head noun namca ‘man’ appearing after the RC phi-lul hulli-ko iss-nun ‘who was bleeding.’ In the RC, the head noun is deleted along with its case marker. The head noun appearing outside the RC is supported by how the head noun is marked by the accusative marker -lul, which reflects its grammatical role in the matrix clause, rather than the nominative marker -ka, which would reflect its grammatical role in the RC. Prenominal RCs, as in (2), comprise the predominant RC type in Korean. The example in (3) includes an IHRC. This RC type can be distinguished from prenominal RCs by two formal features. First, unlike in (2), the head noun in (3) appears in the RC, and the head noun namca ‘man’ is marked by the nominative case marker -ka. Second, the RC in (3) is marked by the combination of a relativizer and kes ‘thing,’ which can be considered a quasi-nominalizer in Korean (Jo 2003; Kim 1984; Mun 2017). This RC type, in which the head noun appears within the RC, is found only in a few languages. In Dryer’s (2013) sample, 24 languages use IHRCs as their sole RC type, and 10 languages use IHRCs as their non-primary RC type; Korean falls into the latter group. Furthermore, the distribution of Korean IHRCs is highly restricted, making them even more unique.

To understand the restricted use of Korean IHRCs, existing studies have employed three research approaches: (i) explaining their formal restrictions by proposing a set of rules (Chung and Kim 2003; Jhang 1991, 1994; Kim 2016, 2008b, 1996, 2002; Lee 2006); (ii) investigating how they are used in naturally occurring data (Cho 2014, 2016; Lee 2020a, 2020b, 2021a, 2021b); (iii) refuting their existence as an independent construction and offering an alternative construction in which an IHRC construal can be made (Chung 1999; Kim 2013; Mun 2012, 2017; Park 1994, 2022a, 2022b, 2019; Park and Yeon 2023; Ryu 2022; Yeon and Park 2021).

The first approach concerns grammatical restrictions in the formation of IHRCs. The head noun tends to take a grammatical role as a subject or object in an RC, as in (4a) and (4b), respectively. In (4a), there is only one participant (Mary), and it is marked by the nominative case marker -ka in the RC. Similarly, in (4b), phili- ‘a flute’ is understood as the head noun marked by the accusative case marker -lul in the RC.

Additionally, the head noun can take a subject, object, or oblique role in a matrix clause, as illustrated in (5), (6), and (7), respectively.

When the head noun takes an object role in a matrix clause (hereafter object IHRC), it shows less semantic restriction on the referent of the head noun compared to non-object IHRCs. Specifically, the head noun can refer to a human without any derogative meaning. In (5a), (6a), and (7a), the head noun is emeni ‘mother,’ and only the object IHRC in (6a) turns out grammatical. Unlike in (5a) and (7a), the head noun is ai ‘child’ in (5b) and mwul ‘water’ in (7b), which are not sensitive to derogative meaning, and both sentences turn out grammatical. The semantic restriction regarding the referent of the head noun can be understood in terms of the degree of grammaticality of kes, as described in the examples below.

In (8a), kes ‘thing’ has its lexical meaning and is used as a dependent noun referring to what Mary gave to John. However, in (8b), kes forms a nominalizer (without any lexical meaning) associated with the relativizer -un. Kes in (8c), an IHRC example, also does not seem to have any lexical meaning, and it has been considered a kind of (dependent) pronoun (Chung and Kim 2003; Kim 2007, 2009, 2022; Lee 2006), nominalizer (Jo 2003; Kim 1984), complementizer (Jhang 1994; Yoon 1991), or schematic noun (Park 2022a, b; Park and Yeon 2023; Yeon and Park 2021). The ungrammaticality of the sentences with the non-object IHRC in (5a) and (7a), whose head noun refers to an honorific human being, may suggest that the grammaticality of kes in object IHRCs is different from that in non-object IHRCs (Kim and Song 2022; Lee 2021a).

Furthermore, the RC shows an aspectual restriction when the head noun takes an object role in a matrix clause such that it should represent a temporary state; specifically, an in-progress state is achieved by any sentence with a progressive aspect, regardless of the lexical aspect of a predicate, but a target state is achieved only by a sentence with a perfective form of a telic predicate (Kim 2008b). Examples of in-progress and target states are given in (9a) and (9b), respectively.

Unlike the IHRCs in (9), those in (10) do not represent a temporary state. The embedded predicate in (10a) is an adjective, whereas the one in (10b) is a stative verb; both examples turn out ungrammatical.

Notably, the aspectual restriction is required only for object IHRCs. The non-object IHRC in (11) does not represent a temporary state, with an adjective being used as an embedded predicate, but it does not turn out ungrammatical.

Furthermore, some idiosyncrasies in Korean IHRCs cannot be encapsulated in a set of rules. For example, an adjective is used as an embedded predicate in object IHRCs in (12), but these do not turn out ungrammatical. However, as Mun (2012) marks (12a) with a question mark, the IHRC in (12a) is less acceptable than the one in (12b), which cannot be clearly explained by grammatical rules.

In addition to the aspectual restriction, previous studies have argued that only a set of predicates, such as a stage-level predicate or a verb that lexically requires its object to refer to a referential individual, can take IHRCs as the object. For example, as Kim (2002) shows, in (13a) the individual-level predicate salangha- ‘love’ is used as an embedding predicate, and the embedded clause cannot be interpreted as an IHRC.^[1] Conversely, in (13b), the stage-level predicate ttayly- ‘hit’ is used as an embedding predicate, and the embedded clause can be interpreted as an IHRC.

Similarly, Chung and Kim (2003, p. 53) state that “when the matrix predicate is a type of recognition verb such as po- ‘see’, al- ‘know’, and kiekha- ‘remember,’” only an event reading is possible, which does not permit an IHRC construal. The embedding predicate in (14b) is mollu- ‘not know,’ and the embedded predicate does not yield an IHRC construal.

When a perceptive or factive verb is used as an embedding predicate, however, disagreement arises about whether the embedded clause is an IHRC, as existing studies have excluded perceptive and factive verbs (Chung and Kim 2003; Kim 2016, 2007, 2009; Lee 2021a, 2021b), included only a specific example of perceptive verbs (Jo 2003), or included perceptive and (some) factive verbs (Lee 2006; Mun 2017; Park 2022b). For instance, Jo (2003) regards (15a), not (15b), as an example of IHRCs because it is only interpretable as an IHRC construal. Conversely, Park (2022b) suggests that (15b) and (15c) are amenable to an IHRC construal. Although Lee (2006) suggests the embedded clause in (15d) as an example yielding an IHRC construal, whether the example in (15d) can be interpreted as “Mary already knew the answer to that question John explained” is doubtful; this approach is found only in Lee’s (2006) work.

In this study, we argue that only a specific example of perceptive verbs, such as (15a), can yield an IHRC construal for three reasons. First, as Park (1994, p. 46) indicates, the entity reading found in (15b) and (15c) “comes as a pragmatic consequence of that semantic interpretation.” Essentially, in a scene, we see the participants involved in it. Second, an immediate perception of the state of affairs is distinguished from an immediate perception of the individual (Dik and Hengeveld 1991). For example, although the implication is that the subject of the matrix clause perceived an individual, (16b) is not conventionally analyzed as both an immediate perception of the state of affairs and an immediate perception of the individual.

Third, IHRC examples similar to (15a) appear in other studies. Bickel (1999), for instance, investigates several Kirant languages, such as Belhare, Limbu, and Athpare, which exhibit two similarities to Korean. First, they have prenominal RCs with IHRCs as a minor type. Second, although prenominal RCs and IHRCs are marked by nominalization devices, only IHRCs have the same surface shape as nominalized clauses. Although these languages differ from Korean in that whether the nominal clause is a complement clause or an IHRC can be distinguished by the presence or absence of the agreement (Mun 2017, p. 69), they offer helpful resources for determining IHRCs in Korean. Disambiguating complement clauses and IHRCs, Bickel (1999) provides examples from Athpare (17a) and Belhare (17b) as IHRCs and an example from Belhare (18) as a complement clause.

In the IHRCs in (17), pu- ‘look’ and chitt- ‘find’ are used as a matrix predicate, and in the complement clause in (18), nis- ‘see’ is used as a matrix predicate. The embedded clauses recognized as an IHRC in (17) allow only an IHRC construal. In (17a), the goat was purchased yesterday; thus, only the goat – and not the event of purchasing it – can be presented. Similarly, in (17b), the car had already broken down by the time the subject of the matrix clause came across it, and what they observed was only the car – an entity – which was broken. Although the matrix predicates in (17) are not exactly the perceptive verb “see,” the examples in (17) being identified as IHRCs supports our argument that only a specific example of perceptive verbs, such as in (15a), can yield an IHRC construal.

Unlike the first approach, the second approach empirically shows that IHRCs appear in naturalistic spoken and written corpora of Korean, with unique pragmatic functions. Lee (2020a) found 39 IHRCs in spoken and 58 in written sentences out of 10,000 samples, comprising the form of kes. In the spoken data, 14 examples take a subject role in an RC, and 25 examples take an object role. In the written data, 26 examples take a subject role in an RC, and 32 examples take an object role. Similarly, Cho (2016) finds 99 examples of IHRCs in spoken data, with the head noun taking a subject role in an RC (N = 30) and an object role in an RC (N = 69). Examining naturally occurring IHRCs, Lee (2020a) proposes six pragmatic functions of IHRCs that are not shared by prenominal RCs: (i) quickly representing the common argument; (ii) interacting with the audience; (iii) representing information strategically; (iv) accurately representing the order of events; (v) dropping the common argument; (vi) being allowed to have the particle of the common noun in an RC.

Finally, the third approach to the restricted use of IHRCs refutes the existence of IHRCs as an independent construction, and some of the researchers adopting this approach have suggested an alternative construction in which IHRCs can be categorized as a complement clause of a null perception verb (Chung 1999; Park 1994), noun complement construction (Mun 2017; Park 2019),^[2] adnominal clause + a schematic noun kes (Park 2022a, b; Park and Yeon 2023; Yeon and Park 2021), adjunct (Kim 2013), or not an IHRC (Mun 2012; Ryu 2022).

The main reason behind refuting Korean IHRCs as independent constructions is that they do not have their own encoding strategy and exhibit idiosyncrasies that cannot be explained by a set of rules. However, we are reluctant to discharge IHRCs from Korean, instead preferring to wait for better evidence, for several reasons (see also Sun 2023). First, for IHRCs, sharing encoding strategies with other constructions is not uncommon (Bickel 1999; DeLancey 1989; Dixon 2006; Ebert 1994; Genetti 1992; Matisoff 1972; Noonan 1997). Second, although IHRCs may resemble other constructions, only IHRCs take an argument role in a matrix clause with unique grammatical restrictions and pragmatic functions. Third, the existence of idiosyncrasies may imply that the boundaries between different constructions are vague, as some typical examples show consistent grammatical features (c.f. Lee 2020a). Fourth, although the differences in grammatical constraints between object and non-object IHRCs may suggest that Korean IHRCs are merely interpreted based on context, this phenomenon may also indicate that object IHRCs are developing into a distinct construction (c.f. Kim and Song 2022; Lee 2021a).

2 A new approach to understanding object IHRCs in Korean

While these three approaches have advanced the understanding of Korean IHRCs, more attention should be given to the role of embedding predicates, especially for object IHRCs. Predicates that allow object IHRC construals do not generally take clauses as arguments, which explains their low frequency and acceptance (see Section 5.3). To understand the phenomenon, we examine the semantic type of verbs allowing an IHRC construal. As some perceptive verbs allow an object IHRC construal, whereas factive verbs do not (see Section 1), we also investigate the shared features of these predicates. We discuss that Korean object IHRCs are a suitable candidate for partially productive or verb-specific constructions.

An IHRC construction in Korean has a form analogous to that of perceptive and factive verb constructions (Chung and Kim 2003; Jo 2003; Kim 2004a, b, c, 2008a, 2009; Lee 2021b; Mun 2017; Park 2022a, b; Yeon and Park 2021, 2023).

When al- ‘know’ or po- ‘see’ is used as an embedding predicate, as in (19a) and (19b), the bracketed part is considered a part of the perceptive and factive verb constructions, respectively. In (19c), pwuthcap- ‘catch’ is used as an embedding predicate, and the bracketed part has the construal of an IHRC. Nevertheless, when pwuthcap- ‘catch’ is used as an embedding predicate, a mismatch between syntax and semantics arises. Structurally, the entire clause Mary-ka naka-lyeko ha-nun-n kes functions as the object of the embedding predicate. Nonetheless, semantically, only the head noun Mary is the object of the action pwuthcap- ‘catch.’ The construal of IHRCs comes from this mismatch, in which the head noun appears in an embedded clause and can be interpreted as the object of the embedding predicate (Shibatani 2018, p. 354). Verbs such as pwuthcap- ‘catch’ typically take a noun phrase as their object; thus, for such verbs, taking a nominal clause as their object is highly unusual, as in (19c). In other words, the construal of IHRCs appears to be engendered via the creative use of verbs.

Two things should be noted. First, the creative use of verbs appears only when embedded clauses are marked by the quasi-nominalizer – the combination of a relativizer and kes (Jo 2003; Park and Yeon 2023; Yeon and Park 2021). When other nominalizers such as -um and -ki are used, as in (20), the sentences turn out ungrammatical.

The ungrammaticality of the examples in (20) highlights the key role of kes in an IHRC construal. As shown in Section 1, the meaning of kes in a particular example is not always clear, covering a continuum of abstract meanings, such as a part of a nominalizer or concrete object (i.e., a thing). Previous studies have primarily focused on the versatility of kes to explain how an IHRC construal is triggered in Korean. Jo (2003, p. 554) attributes this ambiguity regarding kes and its generality in terms of being used in perceptive and factive verb constructions to the IHRC construal, proposing that “matrix verbs excluding a nominalized clause (i.e., proposition) forces to turn a nominalizer kes into a pronominal kes by selecting a referential NP within a clause as its binder.” Therefore, the rise of the IHRC in Korean is motivated by a forced processing, rather than syntactic legitimacy that it uniquely establishes.” Park and Yeon (2023, p. 153) suggest that “metonymy is the fundamental source of the interpretations of the IHRCs” (see also Park 2022a, b; Yeon and Park 2021). Defining kes as a schematic nominal devoid of semantic content, they show how kes “transparently mediates an entity in the adnominal clause and the matrix predicate” (Park and Yeon 2023, p. 152). They further explain that kes deals with the possible semantic conflict between the adnominal clause and an embedding predicate, such as cip- ‘pick up,’ by being connected not to the entire embedded clause but to the active zone, which refers to an entity.

Second, the perceptive verb construction in (19b) shows more similarity to IHRCs than the factive one. Perceptive verb constructions allow an entity construal of the embedded clause (described in Section 1), which can be considered an example yielding an IHRC construal. However, object IHRCs exhibit grammatical restrictions parallel to those of perceptive verb constructions. This feature is not shared by factive verb constructions, as indicated by Kim (2008a), who argues that the embedded clauses of object IHRCs and perceptive verb constructions do not allow an individual-level predicate such as yeppu- ‘pretty’ or the indicative mood marker -ta.^[3] As perceptive verb constructions share the encoding strategy and grammatical restriction, in addition to yielding event and entity construals, we can assume that perceptive verb constructions may function as an incubator for the birth of an object IHRC in Korean. Essentially, perceptive verb constructions may have stimulated or accelerated the extension of the usage of verbs that do not typically take a clause as their argument.

In sum, an object IHRC construal is created through the innovative use of verbs, and the result seems to be supported by the similarities between IHRCs and existing constructions, such as perceptive verb constructions. This engenders the question of which verbs would allow this innovative use of taking a clause as their argument, as opposed to an NP. We can explore the answer by referring to the usage-based approach, whose findings show that, when a word is used in a creative or unconventional manner, words semantically similar to it may also attain the new usage, forming partially productive or verb-class-specific constructions (Boas 2003; Croft 2003, 2012; Goldberg 1995, 2006, 2016a, b, 2019; Perek 2015, 2016; Suttle and Goldberg 2011). The examples in (21) illustrate this.

Although “sneeze” does not usually appear as a predicate in caused-motion constructions, it can be ratified as in (21b) through its semantic similarity to the verb “blow” in (21a). In other words, the use of “sneeze” is extended through its semantic overlap with “blow”: both verbs describe air emission (Boas 2003, 2005; Perek 2015; Goldberg 1995; Goldberg and Jackendoff 2004). As described in previous studies, one of the factors constraining the innovative use of words observed in (21b) is semantic similarity. For instance, not every air-emission verb is ratified in caused-motion constructions. Whether an unexpected verb can be used in a caused-motion construction can vary depending on how similar it is to the one that is conventionally used in that construction (e.g., Boas 2003). As seen in (22), “breath,” “pant,” and “wheeze” in (22b–d) do not sound as acceptable as “cough” in (22a), while “exhale” and “inhale” in (22e–f) are ungrammatical.

This varied acceptability can be explained by the strength of the air emission that each verb connotes when other conditions are the same: the more it is similar to “blow,” the more it is acceptable. As an IHRC construal can be created through the extension of the usage of verbs, examining whether an IHRC construal is more acceptable when a certain semantic type of verb is used as its embedding predicate is reasonable.

Therefore, we investigated the potential influence of embedding predicates’ semantic type on the acceptability of Korean IHRCs. To test whether IHRCs with a certain semantic type of embedding predicates show a higher acceptance rate, we conducted an acceptability judgment task. Specifically, by narrowing down IHRCs with a head noun taking an object role in the matrix clause, we predicted that predicates that imply physical contact between two entities take IHRCs as their arguments more naturally than those that do not. By focusing on object IHRCs, we tried to control the grammatical role of the head noun in the matrix clause that can affect the acceptability of IHRCs (Jhang 1991; Kim 2002; Kim and Song 2022; Lee 2020a; Mun 2012).

Within this narrowed scope, we referred to Verspoor (2000) to categorize the subtypes of embedding predicates. Based on Edelman (1989), Verspoor (2000, p. 212) suggests three types of interactions between a person and an object: direct physical, mental, and rational (Figure 1). Among these interactions, embedding predicates that render an IHRC construal indicate a direct physical interaction between the subject of the embedding clause and the object, the entity in the embedded clause. In (23), for instance, John interacts with a cookie that exists in his immediate environment.

Figure 1:

Cognitive modes and their respective objects (Verspoor 2000 p. 212).

This interaction between John and the cookie can be identified as a direct physical interaction because the cookie itself, as opposed to its image or symbol, exists in the same space and time. Unlike cip- ‘pick’ in (23), verbs such as al- ‘know’ do not render an IHRC construal. When al- ‘know’ is used as an embedding predicate that takes a clause marked by the combination of a relativizer and kes as its argument, a rational interaction is indicated between the subject of the embedding clause and the object, the proposition described in the embedded clause. For instance, in (24), John – the subject of the embedding clause – interacts with a symbol representing an event that does not exist in his immediate environment.

The perceptive verb po- ‘see’ engenders a more complicated situation. It may or may not allow an IHRC construal when it takes a clause marked by the combination of a relativizer and kes as its argument (see Section 1).

Among the examples in (25), only (25a) involves a direct physical interaction between the subject of the embedding clause and the object, which is the entity in the embedded clause. The subject of the matrix clause, John, interacts with an entity that exists in his immediate environment. Conversely, (25b) shows a direct physical interaction between the subject of the embedding clause and the object; nonetheless, this time, it is the event described in the embedded clause, not an entity. Similarly, the embedded clause in (25c) is not identified as an IHRC because it describes a situation in which an entity does not exist. One can see a situation in which a particular entity does not exist, but one cannot see an entity that does not exist. In other words, John can interact with a situation in his immediate environment, but he cannot interact with an entity that does not exist in his immediate environment.

The ungrammaticality of the example in (26) can be explained in a similar manner. While cip- ‘pick’ is used as an embedding predicate in (23) and (26), no direct physical interaction occurs between the subject and object, the entity in the embedded clause in (28), as one cannot pick up something that does not exist in one’s immediate environment.

In sum, an object IHRC construal can be rendered when the embedding predicate ensures a direct physical interaction between the subject of the embedding clause and an entity in the embedded clause. Compared to previous studies, referring to the types of interactions between the subject of the embedding clause and the object facilitates a more sophisticated understanding of when an object IHRC construal is created. For instance, a stage-level predicate (Kim 2002) or a verb that lexically requires its object to refer to a referential individual (Chung and Kim 2003) has been suggested as a requirement for an embedding predicate to render an IHRC construal. However, although po- ‘see’ is a stative predicate rendering an event reading, it allows an IHRC construal as in (25a) – when a direct physical interaction is implied between the subject of the embedding clause and an entity in the embedded clause. This adheres to our prediction in this study.

Considering these observations, we divided verbs that ensure a direct physical interaction between the subject of the embedding clause and an entity in the embedded clause into two groups based on whether they imply physical contact (Phil 2004; Shin 2017). Defining physical contact as activities that include touching using body parts or tools (c.f. Faber and Marial Usón 1999; Kim 2000, 2017; Phil 2004; Rambaud and Briones 2002; Viberg 1999), we predicted that verbs assuming physical contact between two entities are more likely to render an IHRC construal resulting in higher acceptability. This prediction was based on how, when physical contact occurs between two entities, a direct physical interaction is naturally implied, which ensures the existence of a common argument (or the head noun). In this sense, verbs assuming physical contact may serve as a core group among verbs that ensure direct physical interaction between two entities. We tested our prediction with five different types of embedding predicates. First, among verbs ensuring direct physical interaction between two entities, we used those that imply physical contact (e.g., pwuthcap- ‘catch’). Second, among verbs ensuring direct physical interaction between two entities, we used those that do not imply physical contact (e.g., pwulu- ‘call’). Third, we used perceptive verbs (e.g., po- ‘see’), followed by factive verbs (e.g., al- ‘know’). Perceptive and factive verbs were used because they take clauses marked by the combination of a relativizer and kes as their arguments, productively sharing the encoding strategy with object IHRCs. Nevertheless, although perceptive verbs share grammatical restrictions regarding embedded clauses with IHRCs, factive verbs do not, as described in this section. We did not use any examples with perceptive verbs in which only an entity construal is available, as in (25a), as perceptive verbs may allow an IHRC construal. Finally, we used adjectives (e.g., manh- ‘many’) as a baseline for ungrammatical sentences.

3 Acceptability judgment task

To examine whether embedding predicates’ semantic type affects the acceptability of IHRCs, we generated 40 sentences with five types of embedding predicates, adhering to Keating and Jegerski’s (2015) recommendation to include twice as many filler items as there are target items.

4 Results

We analyzed the data of 72 participants who completed the task, after excluding participants who did not complete the task (N = 2) or showed less than 50 % accuracy in their grammatical judgment of the filler sentences (N = 6). The results of the acceptability judgment task are shown in Figure 2. We used R (R Core Team 2024) and lme4 (Bates et al. 2015) to perform a linear mixed effects analysis of the relationship between acceptability ratings (z-scored) and predicate type. In the model, we entered predicate type and stimuli list as fixed effects, as well as the random intercepts for subjects and sentences as random effects [lmer(Resp(z) ∼ PredicateType + List + (1|Subj) + (1|item)]. Additionally, multiple comparisons among the predicate types were conducted using lsmeans (Lenth 2016) with Bonferroni correction. The results revealed that List effect was not significant (β = −0.035, SE = 0.018, t = −1.94, p = 0.057). The multiple comparisons showed significant differences between each pair of conditions, except for CON versus IHRC_NP, IHRC_NP versus IHRC_P and KNO versus PER (see Table 3). The results have two implications. First, although we observed no difference between IHRC_P and IHRC_NP, the fact that IHRC_P, not IHRC_NP, showed significantly higher acceptability than the CON condition supports our prediction: predicates implying physical contact accommodate IHRCs more naturally as their arguments than those that do not. Second, the acceptability of IHRC_P was lower than that of factive (i.e., KNO) and perceptive (i.e., PER) verb constructions but higher than that of ungrammatical sentences (i.e., CON).

Figure 2:

Means and standard deviations of acceptability judgement.

5 Discussion

We tested whether the semantics of embedding predicates influence the acceptability of object IHRCs in Korean. Specifically, we predicted that verbs that imply physical contact between two entities are more likely to render an object IHRC construal, resulting in higher acceptability, than verbs that do not. Although our prediction was supported by the acceptability judgment task, four issues should be discussed further: exemplary sentences, world knowledge, low acceptability, and limitations.

6 Conclusions

In this study, we investigated Korean IHRCs, focusing on embedding predicates, to enhance our understanding of the limited distribution of IHRCs. Specifically, we examined whether embedding predicates’ semantic type influences the acceptability of IHRCs. The results showed significantly higher acceptability, compared to the control condition, for predicates implying physical contact between entities but not for those that do not. This finding may indicate that Korean IHRCs can be considered a partially productive or verb-class-specific construction. This study contributes to understanding partial productivity or local generalization in language use and the interaction, discussed in Verspoor’s (2000) work, between a human and an object.